From a7ae218e89859b5a5b8433f57029590b8877fc18 Mon Sep 17 00:00:00 2001 From: Mengchi Zhang Date: Sun, 21 Apr 2019 15:55:21 -0400 Subject: Move cuobjdump parser to reentrant Signed-off-by: Mengchi Zhang --- libcuda/cuda_runtime_api.cc | 15 ++++++++++++--- 1 file changed, 12 insertions(+), 3 deletions(-) (limited to 'libcuda/cuda_runtime_api.cc') diff --git a/libcuda/cuda_runtime_api.cc b/libcuda/cuda_runtime_api.cc index 27644b3..ef5abce 100644 --- a/libcuda/cuda_runtime_api.cc +++ b/libcuda/cuda_runtime_api.cc @@ -1935,7 +1935,10 @@ void setCuobjdumpsassfilename(const char* filename){ } (dynamic_cast(cuobjdumpSectionList.front()))->setSASSfilename(filename); } -extern int cuobjdump_parse(); +typedef void * yyscan_t; +extern int cuobjdump_lex_init(yyscan_t* scanner); +extern int cuobjdump_parse(yyscan_t scanner); +extern int cuobjdump_lex_destroy(yyscan_t scanner); extern FILE *cuobjdump_in; //! Return the executable file of the process containing the PTX/SASS code @@ -2147,7 +2150,10 @@ void extract_code_using_cuobjdump(){ printf("Parsing file %s\n", fname); cuobjdump_in = fopen(fname, "r"); - cuobjdump_parse(); + yyscan_t scanner; + cuobjdump_lex_init(&scanner); + cuobjdump_parse(scanner); + cuobjdump_lex_destroy(scanner); fclose(cuobjdump_in); printf("Done parsing!!!\n"); } else { @@ -2196,7 +2202,10 @@ void extract_code_using_cuobjdump(){ std::cout << "Trying to parse " << libcodfn.str() << std::endl; cuobjdump_in = fopen(libcodfn.str().c_str(), "r"); - cuobjdump_parse(); + yyscan_t scanner; + cuobjdump_lex_init(&scanner); + cuobjdump_parse(scanner); + cuobjdump_lex_destroy(scanner); fclose(cuobjdump_in); std::getline(libsf, line); } -- cgit v1.3 From fe15e8714972667342e650e8c348e7d6b68e5fbc Mon Sep 17 00:00:00 2001 From: Mengchi Zhang Date: Sun, 21 Apr 2019 17:11:07 -0400 Subject: Fix cuobjdump_in Signed-off-by: Mengchi Zhang --- libcuda/cuda_runtime_api.cc | 6 +++++- 1 file changed, 5 insertions(+), 1 deletion(-) (limited to 'libcuda/cuda_runtime_api.cc') diff --git a/libcuda/cuda_runtime_api.cc b/libcuda/cuda_runtime_api.cc index ef5abce..1d1b950 100644 --- a/libcuda/cuda_runtime_api.cc +++ b/libcuda/cuda_runtime_api.cc @@ -1937,9 +1937,9 @@ void setCuobjdumpsassfilename(const char* filename){ } typedef void * yyscan_t; extern int cuobjdump_lex_init(yyscan_t* scanner); +extern void cuobjdump_set_in (FILE * _in_str ,yyscan_t yyscanner ); extern int cuobjdump_parse(yyscan_t scanner); extern int cuobjdump_lex_destroy(yyscan_t scanner); -extern FILE *cuobjdump_in; //! Return the executable file of the process containing the PTX/SASS code //! @@ -2148,11 +2148,13 @@ void extract_code_using_cuobjdump(){ if (parse_output) { printf("Parsing file %s\n", fname); + FILE *cuobjdump_in; cuobjdump_in = fopen(fname, "r"); yyscan_t scanner; cuobjdump_lex_init(&scanner); cuobjdump_parse(scanner); + cuobjdump_set_in(cuobjdump_in, scanner); cuobjdump_lex_destroy(scanner); fclose(cuobjdump_in); printf("Done parsing!!!\n"); @@ -2201,9 +2203,11 @@ void extract_code_using_cuobjdump(){ std::cout << "Done" << std::endl; std::cout << "Trying to parse " << libcodfn.str() << std::endl; + FILE *cuobjdump_in; cuobjdump_in = fopen(libcodfn.str().c_str(), "r"); yyscan_t scanner; cuobjdump_lex_init(&scanner); + cuobjdump_set_in(cuobjdump_in, scanner); cuobjdump_parse(scanner); cuobjdump_lex_destroy(scanner); fclose(cuobjdump_in); -- cgit v1.3 From c4fe0159232673b0c75bacd33b2e96d6a68dfbaf Mon Sep 17 00:00:00 2001 From: Mengchi Zhang Date: Sun, 21 Apr 2019 22:53:00 -0400 Subject: Fix order Signed-off-by: Mengchi Zhang --- libcuda/cuda_runtime_api.cc | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) (limited to 'libcuda/cuda_runtime_api.cc') diff --git a/libcuda/cuda_runtime_api.cc b/libcuda/cuda_runtime_api.cc index 1d1b950..225c402 100644 --- a/libcuda/cuda_runtime_api.cc +++ b/libcuda/cuda_runtime_api.cc @@ -2153,8 +2153,8 @@ void extract_code_using_cuobjdump(){ yyscan_t scanner; cuobjdump_lex_init(&scanner); - cuobjdump_parse(scanner); cuobjdump_set_in(cuobjdump_in, scanner); + cuobjdump_parse(scanner); cuobjdump_lex_destroy(scanner); fclose(cuobjdump_in); printf("Done parsing!!!\n"); -- cgit v1.3 From 10156f6728f10fdc33ca8796e9cc6ade306f06ec Mon Sep 17 00:00:00 2001 From: Mengchi Zhang Date: Thu, 16 May 2019 21:08:39 -0400 Subject: Remove unused extern Signed-off-by: Mengchi Zhang --- libcuda/cuda_runtime_api.cc | 1 - 1 file changed, 1 deletion(-) (limited to 'libcuda/cuda_runtime_api.cc') diff --git a/libcuda/cuda_runtime_api.cc b/libcuda/cuda_runtime_api.cc index 11dcd5a..fc152c2 100644 --- a/libcuda/cuda_runtime_api.cc +++ b/libcuda/cuda_runtime_api.cc @@ -3261,7 +3261,6 @@ extern int ptx__scan_string(const char*); extern FILE *ptx_in; extern int ptxinfo_parse(); -extern int ptxinfo_debug; extern FILE *ptxinfo_in; /// static functions -- cgit v1.3 From 913865e6ac7d0f7c12faeb8430dc5724fdc4be80 Mon Sep 17 00:00:00 2001 From: Mengchi Zhang Date: Mon, 27 May 2019 17:27:58 -0400 Subject: Move some cuobjdump parser variables Signed-off-by: Mengchi Zhang --- libcuda/Makefile | 4 ++-- libcuda/cuda_runtime_api.cc | 23 ++++++++++++----------- libcuda/cuobjdump.h | 18 ++++++++++++++++++ libcuda/cuobjdump.l | 9 ++++++--- libcuda/cuobjdump.y | 43 ++++++++++++++++++++----------------------- 5 files changed, 58 insertions(+), 39 deletions(-) create mode 100644 libcuda/cuobjdump.h (limited to 'libcuda/cuda_runtime_api.cc') diff --git a/libcuda/Makefile b/libcuda/Makefile index 13932e2..c8ff2e3 100644 --- a/libcuda/Makefile +++ b/libcuda/Makefile @@ -111,10 +111,10 @@ $(OUTPUT_DIR)/%.o: %.cc $(CPP) $(CXXFLAGS) -I./ -I$(OUTPUT_DIR) -I$(CUDA_INSTALL_PATH)/include -c $< -o $@ $(OUTPUT_DIR)/%.o: %.c - $(CC) $(CCFLAGS) -I./ -I$(OUTPUT_DIR) -I$(CUDA_INSTALL_PATH)/include -c $< -o $@ + $(CPP) $(CCFLAGS) -I./ -I$(OUTPUT_DIR) -I$(CUDA_INSTALL_PATH)/include -c $< -o $@ $(OUTPUT_DIR)/%.o: $(OUTPUT_DIR)/%.c - $(CC) $(CCFLAGS) -I./ -I$(OUTPUT_DIR) -I$(CUDA_INSTALL_PATH)/include -c $< -o $@ + $(CPP) $(CCFLAGS) -I./ -I$(OUTPUT_DIR) -I$(CUDA_INSTALL_PATH)/include -c $< -o $@ $(OUTPUT_DIR)/cuobjdump_parser.c: cuobjdump.y $(YACC) $(YFLAGS) -p cuobjdump_ -o$@ $< --file-prefix=$(OUTPUT_DIR)/cuobjdump diff --git a/libcuda/cuda_runtime_api.cc b/libcuda/cuda_runtime_api.cc index fc152c2..1d4e870 100644 --- a/libcuda/cuda_runtime_api.cc +++ b/libcuda/cuda_runtime_api.cc @@ -1972,9 +1972,10 @@ void setCuobjdumpsassfilename(const char* filename){ (dynamic_cast(cuobjdumpSectionList.front()))->setSASSfilename(filename); } typedef void * yyscan_t; +#include "cuobjdump.h" extern int cuobjdump_lex_init(yyscan_t* scanner); extern void cuobjdump_set_in (FILE * _in_str ,yyscan_t yyscanner ); -extern int cuobjdump_parse(yyscan_t scanner); +extern int cuobjdump_parse(yyscan_t scanner, cuobjdump_parser* parser); extern int cuobjdump_lex_destroy(yyscan_t scanner); //! Return the executable file of the process containing the PTX/SASS code @@ -2187,11 +2188,11 @@ void extract_code_using_cuobjdump(){ FILE *cuobjdump_in; cuobjdump_in = fopen(fname, "r"); - yyscan_t scanner; - cuobjdump_lex_init(&scanner); - cuobjdump_set_in(cuobjdump_in, scanner); - cuobjdump_parse(scanner); - cuobjdump_lex_destroy(scanner); + cuobjdump_parser parser; + cuobjdump_lex_init(&(parser.scanner)); + cuobjdump_set_in(cuobjdump_in, (parser.scanner)); + cuobjdump_parse(parser.scanner, &parser); + cuobjdump_lex_destroy(parser.scanner); fclose(cuobjdump_in); printf("Done parsing!!!\n"); } else { @@ -2241,11 +2242,11 @@ void extract_code_using_cuobjdump(){ std::cout << "Trying to parse " << libcodfn.str() << std::endl; FILE *cuobjdump_in; cuobjdump_in = fopen(libcodfn.str().c_str(), "r"); - yyscan_t scanner; - cuobjdump_lex_init(&scanner); - cuobjdump_set_in(cuobjdump_in, scanner); - cuobjdump_parse(scanner); - cuobjdump_lex_destroy(scanner); + cuobjdump_parser parser; + cuobjdump_lex_init(&(parser.scanner)); + cuobjdump_set_in(cuobjdump_in, (parser.scanner)); + cuobjdump_parse(parser.scanner, &parser); + cuobjdump_lex_destroy(parser.scanner); fclose(cuobjdump_in); std::getline(libsf, line); } diff --git a/libcuda/cuobjdump.h b/libcuda/cuobjdump.h new file mode 100644 index 0000000..61bf806 --- /dev/null +++ b/libcuda/cuobjdump.h @@ -0,0 +1,18 @@ +#ifndef __cuobjdump_h__ +#define __cuobjdump_h__ +class cuobjdump_parser { + + public: + yyscan_t scanner; + int elfserial; + int ptxserial; + FILE *ptxfile; + FILE *elffile; + FILE *sassfile; + char filename [1024]; + cuobjdump_parser() { + int elfserial = 1; + int ptxserial = 1; + } +}; +#endif /* __cuobjdump_h__ */ diff --git a/libcuda/cuobjdump.l b/libcuda/cuobjdump.l index 9359281..26fbb55 100644 --- a/libcuda/cuobjdump.l +++ b/libcuda/cuobjdump.l @@ -30,6 +30,7 @@ %{ #include #include +#include "cuobjdump.h" #include "cuobjdump_parser.h" #define YY_NEVER_INTERACTIVE 1 @@ -38,7 +39,9 @@ #define YYDEBUG 1 -void cuobjdump_error(yyscan_t yyscanner, const char* msg); +void cuobjdump_error(yyscan_t yyscanner, cuobjdump_parser* parser, const char* msg); +#define YY_DECL int cuobjdump_lex \ + (YYSTYPE * yylval_param , yyscan_t yyscanner, cuobjdump_parser* parser) %} %option stack @@ -153,10 +156,10 @@ newlines {newline}+ <> BEGIN(INITIAL);return 0; /*No other rule matched. Throw an error*/ -. cuobjdump_error(yyscanner, "Invalid token"); +. cuobjdump_error(yyscanner, parser, "Invalid token"); %% -void cuobjdump_error(yyscan_t yyscanner, const char* msg) +void cuobjdump_error(yyscan_t yyscanner, cuobjdump_parser* parser, const char* msg) { struct yyguts_t * yyg = (struct yyguts_t*)yyscanner; printf(" %s near \"%s\"",msg, yytext); diff --git a/libcuda/cuobjdump.y b/libcuda/cuobjdump.y index 66cbace..9c0c28d 100644 --- a/libcuda/cuobjdump.y +++ b/libcuda/cuobjdump.y @@ -30,6 +30,7 @@ #include typedef void * yyscan_t; +#include "cuobjdump.h" extern void addCuobjdumpSection(int sectiontype); void setCuobjdumparch(const char* arch); @@ -37,23 +38,19 @@ void setCuobjdumpidentifier(const char* identifier); void setCuobjdumpptxfilename(const char* filename); void setCuobjdumpelffilename(const char* filename); void setCuobjdumpsassfilename(const char* filename); -int elfserial = 1; -int ptxserial = 1; -FILE *ptxfile; -FILE *elffile; -FILE *sassfile; -char filename [1024]; %} %define api.pure full %parse-param {yyscan_t scanner} +%parse-param {cuobjdump_parser* parser} %lex-param {yyscan_t scanner} +%lex-param {cuobjdump_parser* parser} %union { char* string_value; } %{ -int yylex(YYSTYPE * yylval_param, yyscan_t yyscanner); -void yyerror(yyscan_t yyscanner, const char* msg); +int yylex(YYSTYPE * yylval_param, yyscan_t yyscanner, cuobjdump_parser* parser); +void yyerror(yyscan_t yyscanner, cuobjdump_parser* parser, const char* msg); %} %token H_SEPARATOR H_ARCH H_CODEVERSION H_PRODUCER H_HOST H_COMPILESIZE H_IDENTIFIER H_UNKNOWN H_COMPRESSED %token CODEVERSION @@ -79,24 +76,24 @@ emptylines : emptylines NEWLINE section : PTXHEADER { addCuobjdumpSection(0); - snprintf(filename, 1024, "_cuobjdump_%d.ptx", ptxserial++); - ptxfile = fopen(filename, "w"); - setCuobjdumpptxfilename(filename); + snprintf(parser->filename, 1024, "_cuobjdump_%d.ptx", parser->ptxserial++); + parser->ptxfile = fopen(parser->filename, "w"); + setCuobjdumpptxfilename(parser->filename); } headerinfo compressedkeyword identifier ptxcode { - fclose(ptxfile); + fclose(parser->ptxfile); } | ELFHEADER { addCuobjdumpSection(1); - snprintf(filename, 1024, "_cuobjdump_%d.elf", elfserial); - elffile = fopen(filename, "w"); - setCuobjdumpelffilename(filename); + snprintf(parser->filename, 1024, "_cuobjdump_%d.elf", parser->elfserial); + parser->elffile = fopen(parser->filename, "w"); + setCuobjdumpelffilename(parser->filename); } headerinfo compressedkeyword identifier elfcode { - fclose(elffile); - snprintf(filename, 1024, "_cuobjdump_%d.sass", elfserial++); - sassfile = fopen(filename, "w"); - setCuobjdumpsassfilename(filename); + fclose(parser->elffile); + snprintf(parser->filename, 1024, "_cuobjdump_%d.sass", parser->elfserial++); + parser->sassfile = fopen(parser->filename, "w"); + setCuobjdumpsassfilename(parser->filename); } sasscode { - fclose(sassfile); + fclose(parser->sassfile); }; headerinfo : H_SEPARATOR NEWLINE @@ -118,13 +115,13 @@ identifier : H_IDENTIFIER FILENAME emptylines {setCuobjdumpidentifier($2);} compressedkeyword : H_COMPRESSED emptylines | ; -ptxcode : ptxcode PTXLINE {fprintf(ptxfile, "%s", $2);} +ptxcode : ptxcode PTXLINE {fprintf(parser->ptxfile, "%s", $2);} | ; -elfcode : elfcode ELFLINE {fprintf(elffile, "%s", $2);} +elfcode : elfcode ELFLINE {fprintf(parser->elffile, "%s", $2);} | ; -sasscode : sasscode SASSLINE {fprintf(sassfile, "%s", $2);} +sasscode : sasscode SASSLINE {fprintf(parser->sassfile, "%s", $2);} | ; -- cgit v1.3 From 568e3185c58b07ac31fdd59f6c2aa7af7533939e Mon Sep 17 00:00:00 2001 From: Mengchi Zhang Date: Tue, 28 May 2019 15:14:52 -0400 Subject: Move some functions to enable C++ Signed-off-by: Mengchi Zhang --- libcuda/Makefile | 4 +- libcuda/cuda_runtime_api.cc | 246 ++++++++++++++++++++++---------------------- 2 files changed, 126 insertions(+), 124 deletions(-) (limited to 'libcuda/cuda_runtime_api.cc') diff --git a/libcuda/Makefile b/libcuda/Makefile index 13932e2..c8ff2e3 100644 --- a/libcuda/Makefile +++ b/libcuda/Makefile @@ -111,10 +111,10 @@ $(OUTPUT_DIR)/%.o: %.cc $(CPP) $(CXXFLAGS) -I./ -I$(OUTPUT_DIR) -I$(CUDA_INSTALL_PATH)/include -c $< -o $@ $(OUTPUT_DIR)/%.o: %.c - $(CC) $(CCFLAGS) -I./ -I$(OUTPUT_DIR) -I$(CUDA_INSTALL_PATH)/include -c $< -o $@ + $(CPP) $(CCFLAGS) -I./ -I$(OUTPUT_DIR) -I$(CUDA_INSTALL_PATH)/include -c $< -o $@ $(OUTPUT_DIR)/%.o: $(OUTPUT_DIR)/%.c - $(CC) $(CCFLAGS) -I./ -I$(OUTPUT_DIR) -I$(CUDA_INSTALL_PATH)/include -c $< -o $@ + $(CPP) $(CCFLAGS) -I./ -I$(OUTPUT_DIR) -I$(CUDA_INSTALL_PATH)/include -c $< -o $@ $(OUTPUT_DIR)/cuobjdump_parser.c: cuobjdump.y $(YACC) $(YFLAGS) -p cuobjdump_ -o$@ $< --file-prefix=$(OUTPUT_DIR)/cuobjdump diff --git a/libcuda/cuda_runtime_api.cc b/libcuda/cuda_runtime_api.cc index 168c80f..afe47f8 100644 --- a/libcuda/cuda_runtime_api.cc +++ b/libcuda/cuda_runtime_api.cc @@ -483,6 +483,130 @@ event_tracker_t g_timer_events; int g_active_device = 0; //active gpu that runs the code std::list g_cuda_launch_stack; +typedef void * yyscan_t; +#include "cuobjdump.h" +extern int cuobjdump_lex_init(yyscan_t* scanner); +extern void cuobjdump_set_in (FILE * _in_str ,yyscan_t yyscanner ); +extern int cuobjdump_parse(yyscan_t scanner, struct cuobjdump_parser* parser); +extern int cuobjdump_lex_destroy(yyscan_t scanner); + +enum cuobjdumpSectionType { + PTXSECTION=0, + ELFSECTION +}; + + +class cuobjdumpSection { +public: + //Constructor + cuobjdumpSection() { + arch = 0; + identifier = ""; + } + virtual ~cuobjdumpSection() {} + unsigned getArch() {return arch;} + void setArch(unsigned a) {arch = a;} + std::string getIdentifier() {return identifier;} + void setIdentifier(std::string i) {identifier = i;} + virtual void print(){std::cout << "cuobjdump Section: unknown type" << std::endl;} +private: + unsigned arch; + std::string identifier; +}; + +class cuobjdumpELFSection : public cuobjdumpSection +{ +public: + cuobjdumpELFSection() {} + virtual ~cuobjdumpELFSection() { + elffilename = ""; + sassfilename = ""; + } + std::string getELFfilename() {return elffilename;} + void setELFfilename(std::string f) {elffilename = f;} + std::string getSASSfilename() {return sassfilename;} + void setSASSfilename(std::string f) {sassfilename = f;} + virtual void print() { + std::cout << "ELF Section:" << std::endl; + std::cout << "arch: sm_" << getArch() << std::endl; + std::cout << "identifier: " << getIdentifier() << std::endl; + std::cout << "elf filename: " << getELFfilename() << std::endl; + std::cout << "sass filename: " << getSASSfilename() << std::endl; + std::cout << std::endl; + } +private: + std::string elffilename; + std::string sassfilename; +}; + +class cuobjdumpPTXSection : public cuobjdumpSection +{ +public: + cuobjdumpPTXSection(){ + ptxfilename = ""; + } + std::string getPTXfilename() {return ptxfilename;} + void setPTXfilename(std::string f) {ptxfilename = f;} + virtual void print() { + std::cout << "PTX Section:" << std::endl; + std::cout << "arch: sm_" << getArch() << std::endl; + std::cout << "identifier: " << getIdentifier() << std::endl; + std::cout << "ptx filename: " << getPTXfilename() << std::endl; + std::cout << std::endl; + } +private: + std::string ptxfilename; +}; + + +std::list cuobjdumpSectionList; +std::list libSectionList; + +// sectiontype: 0 for ptx, 1 for elf +void addCuobjdumpSection(int sectiontype){ + if (sectiontype) + cuobjdumpSectionList.push_front(new cuobjdumpELFSection()); + else + cuobjdumpSectionList.push_front(new cuobjdumpPTXSection()); + printf("## Adding new section %s\n", sectiontype?"ELF":"PTX"); +} + +void setCuobjdumparch(const char* arch){ + unsigned archnum; + sscanf(arch, "sm_%u", &archnum); + assert (archnum && "cannot have sm_0"); + printf("Adding arch: %s\n", arch); + cuobjdumpSectionList.front()->setArch(archnum); +} + +void setCuobjdumpidentifier(const char* identifier){ + printf("Adding identifier: %s\n", identifier); + cuobjdumpSectionList.front()->setIdentifier(identifier); +} + +void setCuobjdumpptxfilename(const char* filename){ + printf("Adding ptx filename: %s\n", filename); + cuobjdumpSection* x = cuobjdumpSectionList.front(); + if (dynamic_cast(x) == NULL){ + assert (0 && "You shouldn't be trying to add a ptxfilename to an elf section"); + } + (dynamic_cast(x))->setPTXfilename(filename); +} + +void setCuobjdumpelffilename(const char* filename){ + if (dynamic_cast(cuobjdumpSectionList.front()) == NULL){ + assert (0 && "You shouldn't be trying to add a elffilename to an ptx section"); + } + (dynamic_cast(cuobjdumpSectionList.front()))->setELFfilename(filename); +} + +void setCuobjdumpsassfilename(const char* filename){ + if (dynamic_cast(cuobjdumpSectionList.front()) == NULL){ + assert (0 && "You shouldn't be trying to add a sassfilename to an ptx section"); + } + (dynamic_cast(cuobjdumpSectionList.front()))->setSASSfilename(filename); +} + /******************************************************************************* * * * * @@ -1856,128 +1980,6 @@ __host__ cudaError_t CUDARTAPI cudaGetExportTable(const void **ppExportTable, co //#include "../../cuobjdump_to_ptxplus/cuobjdump_parser.h" -enum cuobjdumpSectionType { - PTXSECTION=0, - ELFSECTION -}; - - -class cuobjdumpSection { -public: - //Constructor - cuobjdumpSection() { - arch = 0; - identifier = ""; - } - virtual ~cuobjdumpSection() {} - unsigned getArch() {return arch;} - void setArch(unsigned a) {arch = a;} - std::string getIdentifier() {return identifier;} - void setIdentifier(std::string i) {identifier = i;} - virtual void print(){std::cout << "cuobjdump Section: unknown type" << std::endl;} -private: - unsigned arch; - std::string identifier; -}; - -class cuobjdumpELFSection : public cuobjdumpSection -{ -public: - cuobjdumpELFSection() {} - virtual ~cuobjdumpELFSection() { - elffilename = ""; - sassfilename = ""; - } - std::string getELFfilename() {return elffilename;} - void setELFfilename(std::string f) {elffilename = f;} - std::string getSASSfilename() {return sassfilename;} - void setSASSfilename(std::string f) {sassfilename = f;} - virtual void print() { - std::cout << "ELF Section:" << std::endl; - std::cout << "arch: sm_" << getArch() << std::endl; - std::cout << "identifier: " << getIdentifier() << std::endl; - std::cout << "elf filename: " << getELFfilename() << std::endl; - std::cout << "sass filename: " << getSASSfilename() << std::endl; - std::cout << std::endl; - } -private: - std::string elffilename; - std::string sassfilename; -}; - -class cuobjdumpPTXSection : public cuobjdumpSection -{ -public: - cuobjdumpPTXSection(){ - ptxfilename = ""; - } - std::string getPTXfilename() {return ptxfilename;} - void setPTXfilename(std::string f) {ptxfilename = f;} - virtual void print() { - std::cout << "PTX Section:" << std::endl; - std::cout << "arch: sm_" << getArch() << std::endl; - std::cout << "identifier: " << getIdentifier() << std::endl; - std::cout << "ptx filename: " << getPTXfilename() << std::endl; - std::cout << std::endl; - } -private: - std::string ptxfilename; -}; - -std::list cuobjdumpSectionList; -std::list libSectionList; - -// sectiontype: 0 for ptx, 1 for elf -void addCuobjdumpSection(int sectiontype){ - if (sectiontype) - cuobjdumpSectionList.push_front(new cuobjdumpELFSection()); - else - cuobjdumpSectionList.push_front(new cuobjdumpPTXSection()); - printf("## Adding new section %s\n", sectiontype?"ELF":"PTX"); -} - -void setCuobjdumparch(const char* arch){ - unsigned archnum; - sscanf(arch, "sm_%u", &archnum); - assert (archnum && "cannot have sm_0"); - printf("Adding arch: %s\n", arch); - cuobjdumpSectionList.front()->setArch(archnum); -} - -void setCuobjdumpidentifier(const char* identifier){ - printf("Adding identifier: %s\n", identifier); - cuobjdumpSectionList.front()->setIdentifier(identifier); -} - -void setCuobjdumpptxfilename(const char* filename){ - printf("Adding ptx filename: %s\n", filename); - cuobjdumpSection* x = cuobjdumpSectionList.front(); - if (dynamic_cast(x) == NULL){ - assert (0 && "You shouldn't be trying to add a ptxfilename to an elf section"); - } - (dynamic_cast(x))->setPTXfilename(filename); -} - -void setCuobjdumpelffilename(const char* filename){ - if (dynamic_cast(cuobjdumpSectionList.front()) == NULL){ - assert (0 && "You shouldn't be trying to add a elffilename to an ptx section"); - } - (dynamic_cast(cuobjdumpSectionList.front()))->setELFfilename(filename); -} - -void setCuobjdumpsassfilename(const char* filename){ - if (dynamic_cast(cuobjdumpSectionList.front()) == NULL){ - assert (0 && "You shouldn't be trying to add a sassfilename to an ptx section"); - } - (dynamic_cast(cuobjdumpSectionList.front()))->setSASSfilename(filename); -} -typedef void * yyscan_t; -#include "cuobjdump.h" -extern int cuobjdump_lex_init(yyscan_t* scanner); -extern void cuobjdump_set_in (FILE * _in_str ,yyscan_t yyscanner ); -extern int cuobjdump_parse(yyscan_t scanner, struct cuobjdump_parser* parser); -extern int cuobjdump_lex_destroy(yyscan_t scanner); - //! Return the executable file of the process containing the PTX/SASS code //! //! This Function returns the executable file ran by the process. This -- cgit v1.3 From 25bdc0dd89932f95ace0fc617649a4e041aaadd9 Mon Sep 17 00:00:00 2001 From: Mengchi Zhang Date: Wed, 29 May 2019 01:28:29 -0400 Subject: Move SectionList to context Signed-off-by: Mengchi Zhang --- libcuda/cuda_runtime_api.cc | 109 ++++++++++---------------------------------- libcuda/cuobjdump.h | 67 +++++++++++++++++++++++++++ libcuda/cuobjdump.l | 8 ++-- libcuda/cuobjdump.y | 36 ++++++++------- 4 files changed, 113 insertions(+), 107 deletions(-) (limited to 'libcuda/cuda_runtime_api.cc') diff --git a/libcuda/cuda_runtime_api.cc b/libcuda/cuda_runtime_api.cc index afe47f8..df7ddc7 100644 --- a/libcuda/cuda_runtime_api.cc +++ b/libcuda/cuda_runtime_api.cc @@ -139,6 +139,8 @@ #include "../src/gpgpusim_entrypoint.h" #include "../src/stream_manager.h" #include "../src/abstract_hardware_model.h" +typedef void * yyscan_t; +#include "cuobjdump.h" #include #include @@ -302,6 +304,7 @@ struct CUctx_st { return i->second; } + std::list cuobjdumpSectionList; private: _cuda_device_id *m_gpu; // selected gpu std::map m_code; // fat binary handle => global symbol table @@ -483,11 +486,9 @@ event_tracker_t g_timer_events; int g_active_device = 0; //active gpu that runs the code std::list g_cuda_launch_stack; -typedef void * yyscan_t; -#include "cuobjdump.h" extern int cuobjdump_lex_init(yyscan_t* scanner); extern void cuobjdump_set_in (FILE * _in_str ,yyscan_t yyscanner ); -extern int cuobjdump_parse(yyscan_t scanner, struct cuobjdump_parser* parser); +extern int cuobjdump_parse(yyscan_t scanner, struct cuobjdump_parser* parser, std::list &cuobjdumpSectionList); extern int cuobjdump_lex_destroy(yyscan_t scanner); enum cuobjdumpSectionType { @@ -496,74 +497,10 @@ enum cuobjdumpSectionType { }; -class cuobjdumpSection { -public: - //Constructor - cuobjdumpSection() { - arch = 0; - identifier = ""; - } - virtual ~cuobjdumpSection() {} - unsigned getArch() {return arch;} - void setArch(unsigned a) {arch = a;} - std::string getIdentifier() {return identifier;} - void setIdentifier(std::string i) {identifier = i;} - virtual void print(){std::cout << "cuobjdump Section: unknown type" << std::endl;} -private: - unsigned arch; - std::string identifier; -}; - -class cuobjdumpELFSection : public cuobjdumpSection -{ -public: - cuobjdumpELFSection() {} - virtual ~cuobjdumpELFSection() { - elffilename = ""; - sassfilename = ""; - } - std::string getELFfilename() {return elffilename;} - void setELFfilename(std::string f) {elffilename = f;} - std::string getSASSfilename() {return sassfilename;} - void setSASSfilename(std::string f) {sassfilename = f;} - virtual void print() { - std::cout << "ELF Section:" << std::endl; - std::cout << "arch: sm_" << getArch() << std::endl; - std::cout << "identifier: " << getIdentifier() << std::endl; - std::cout << "elf filename: " << getELFfilename() << std::endl; - std::cout << "sass filename: " << getSASSfilename() << std::endl; - std::cout << std::endl; - } -private: - std::string elffilename; - std::string sassfilename; -}; - -class cuobjdumpPTXSection : public cuobjdumpSection -{ -public: - cuobjdumpPTXSection(){ - ptxfilename = ""; - } - std::string getPTXfilename() {return ptxfilename;} - void setPTXfilename(std::string f) {ptxfilename = f;} - virtual void print() { - std::cout << "PTX Section:" << std::endl; - std::cout << "arch: sm_" << getArch() << std::endl; - std::cout << "identifier: " << getIdentifier() << std::endl; - std::cout << "ptx filename: " << getPTXfilename() << std::endl; - std::cout << std::endl; - } -private: - std::string ptxfilename; -}; - - -std::list cuobjdumpSectionList; std::list libSectionList; // sectiontype: 0 for ptx, 1 for elf -void addCuobjdumpSection(int sectiontype){ +void addCuobjdumpSection(int sectiontype, std::list &cuobjdumpSectionList){ if (sectiontype) cuobjdumpSectionList.push_front(new cuobjdumpELFSection()); else @@ -571,7 +508,7 @@ void addCuobjdumpSection(int sectiontype){ printf("## Adding new section %s\n", sectiontype?"ELF":"PTX"); } -void setCuobjdumparch(const char* arch){ +void setCuobjdumparch(const char* arch, std::list &cuobjdumpSectionList){ unsigned archnum; sscanf(arch, "sm_%u", &archnum); assert (archnum && "cannot have sm_0"); @@ -579,12 +516,12 @@ void setCuobjdumparch(const char* arch){ cuobjdumpSectionList.front()->setArch(archnum); } -void setCuobjdumpidentifier(const char* identifier){ +void setCuobjdumpidentifier(const char* identifier, std::list &cuobjdumpSectionList){ printf("Adding identifier: %s\n", identifier); cuobjdumpSectionList.front()->setIdentifier(identifier); } -void setCuobjdumpptxfilename(const char* filename){ +void setCuobjdumpptxfilename(const char* filename, std::list &cuobjdumpSectionList){ printf("Adding ptx filename: %s\n", filename); cuobjdumpSection* x = cuobjdumpSectionList.front(); if (dynamic_cast(x) == NULL){ @@ -593,14 +530,14 @@ void setCuobjdumpptxfilename(const char* filename){ (dynamic_cast(x))->setPTXfilename(filename); } -void setCuobjdumpelffilename(const char* filename){ +void setCuobjdumpelffilename(const char* filename, std::list &cuobjdumpSectionList){ if (dynamic_cast(cuobjdumpSectionList.front()) == NULL){ assert (0 && "You shouldn't be trying to add a elffilename to an ptx section"); } (dynamic_cast(cuobjdumpSectionList.front()))->setELFfilename(filename); } -void setCuobjdumpsassfilename(const char* filename){ +void setCuobjdumpsassfilename(const char* filename, std::list &cuobjdumpSectionList){ if (dynamic_cast(cuobjdumpSectionList.front()) == NULL){ assert (0 && "You shouldn't be trying to add a sassfilename to an ptx section"); } @@ -2132,7 +2069,7 @@ static int get_app_cuda_version() { * It is also responsible for extracting the libraries linked to the binary if the option is * enabled * */ -void extract_code_using_cuobjdump(){ +void extract_code_using_cuobjdump(std::list &cuobjdumpSectionList){ CUctx_st *context = GPGPUSim_Context(); unsigned forced_max_capability = context->get_device()->get_gpgpu()->get_config().get_forced_max_capability(); @@ -2195,7 +2132,7 @@ void extract_code_using_cuobjdump(){ parser.ptxserial = 1; cuobjdump_lex_init(&(parser.scanner)); cuobjdump_set_in(cuobjdump_in, (parser.scanner)); - cuobjdump_parse(parser.scanner, &parser); + cuobjdump_parse(parser.scanner, &parser, cuobjdumpSectionList); cuobjdump_lex_destroy(parser.scanner); fclose(cuobjdump_in); printf("Done parsing!!!\n"); @@ -2251,7 +2188,7 @@ void extract_code_using_cuobjdump(){ parser.ptxserial = 1; cuobjdump_lex_init(&(parser.scanner)); cuobjdump_set_in(cuobjdump_in, (parser.scanner)); - cuobjdump_parse(parser.scanner, &parser); + cuobjdump_parse(parser.scanner, &parser, cuobjdumpSectionList); cuobjdump_lex_destroy(parser.scanner); fclose(cuobjdump_in); std::getline(libsf, line); @@ -2445,7 +2382,7 @@ cuobjdumpELFSection* findELFSectionInList(std::list sectionli } //! Find an ELF section in all the known lists -cuobjdumpELFSection* findELFSection(const std::string identifier){ +cuobjdumpELFSection* findELFSection(const std::string identifier, std::list cuobjdumpSectionList){ cuobjdumpELFSection* sec = findELFSectionInList(cuobjdumpSectionList, identifier); if (sec!=NULL)return sec; sec = findELFSectionInList(libSectionList, identifier); @@ -2480,7 +2417,7 @@ cuobjdumpPTXSection* findPTXSectionInList(std::list sectionli } //! Find an PTX section in all the known lists -cuobjdumpPTXSection* findPTXSection(const std::string identifier){ +cuobjdumpPTXSection* findPTXSection(const std::string identifier, std::list cuobjdumpSectionList){ cuobjdumpPTXSection* sec = findPTXSectionInList(cuobjdumpSectionList, identifier); if (sec!=NULL)return sec; sec = findPTXSectionInList(libSectionList, identifier); @@ -2493,9 +2430,9 @@ cuobjdumpPTXSection* findPTXSection(const std::string identifier){ //! Extract the code using cuobjdump and remove unnecessary sections -void cuobjdumpInit(){ +void cuobjdumpInit(std::list &cuobjdumpSectionList){ CUctx_st *context = GPGPUSim_Context(); - extract_code_using_cuobjdump(); //extract all the output of cuobjdump to _cuobjdump_*.* + extract_code_using_cuobjdump(cuobjdumpSectionList); //extract all the output of cuobjdump to _cuobjdump_*.* const char* pre_load = getenv("CUOBJDUMP_SIM_FILE"); if (pre_load ==NULL || strlen(pre_load)==0){ cuobjdumpSectionList = pruneSectionList(cuobjdumpSectionList, context); @@ -2513,7 +2450,7 @@ void cuobjdumpRegisterFatBinary(unsigned int handle, const char* filename){ } //! Either submit PTX for simulation or convert SASS to PTXPlus and submit it -void cuobjdumpParseBinary(unsigned int handle){ +void cuobjdumpParseBinary(unsigned int handle, std::list &cuobjdumpSectionList){ if(fatbin_registered[handle]) return; fatbin_registered[handle] = true; @@ -2566,7 +2503,7 @@ void cuobjdumpParseBinary(unsigned int handle){ cuobjdumpPTXSection* ptx = NULL; const char* pre_load = getenv("CUOBJDUMP_SIM_FILE"); if(pre_load==NULL || strlen(pre_load)==0) - ptx = findPTXSection(fname); + ptx = findPTXSection(fname, context->cuobjdumpSectionList); char *ptxcode; const char *override_ptx_name = getenv("PTX_SIM_KERNELFILE"); if (override_ptx_name == NULL or getenv("PTX_SIM_USE_PTX_FILE") == NULL or strlen(getenv("PTX_SIM_USE_PTX_FILE"))==0) { @@ -2576,7 +2513,7 @@ void cuobjdumpParseBinary(unsigned int handle){ ptxcode = readfile(override_ptx_name); } if(context->get_device()->get_gpgpu()->get_config().convert_to_ptxplus() ) { - cuobjdumpELFSection* elfsection = findELFSection(ptx->getIdentifier()); + cuobjdumpELFSection* elfsection = findELFSection(ptx->getIdentifier(), context->cuobjdumpSectionList); assert (elfsection!= NULL); char *ptxplus_str = gpgpu_ptx_sim_convert_ptx_and_sass_to_ptxplus( ptx->getPTXfilename(), @@ -2664,7 +2601,7 @@ void** CUDARTAPI __cudaRegisterFatBinary( void *fatCubin ) * then for next calls, only returns the appropriate number */ assert(fat_cubin_handle >= 1); - if (fat_cubin_handle==1) cuobjdumpInit(); + if (fat_cubin_handle==1) cuobjdumpInit(context->cuobjdumpSectionList); cuobjdumpRegisterFatBinary(fat_cubin_handle, filename); return (void**)fat_cubin_handle; @@ -2779,7 +2716,7 @@ void CUDARTAPI __cudaRegisterFunction( printf("GPGPU-Sim PTX: __cudaRegisterFunction %s : hostFun 0x%p, fat_cubin_handle = %u\n", deviceFun, hostFun, fat_cubin_handle); if(context->get_device()->get_gpgpu()->get_config().use_cuobjdump()) - cuobjdumpParseBinary(fat_cubin_handle); + cuobjdumpParseBinary(fat_cubin_handle, context->cuobjdumpSectionList); context->register_function( fat_cubin_handle, hostFun, deviceFun ); } @@ -2799,7 +2736,7 @@ extern void __cudaRegisterVar( printf("GPGPU-Sim PTX: __cudaRegisterVar: hostVar = %p; deviceAddress = %s; deviceName = %s\n", hostVar, deviceAddress, deviceName); printf("GPGPU-Sim PTX: __cudaRegisterVar: Registering const memory space of %d bytes\n", size); if(GPGPUSim_Context()->get_device()->get_gpgpu()->get_config().use_cuobjdump()) - cuobjdumpParseBinary((unsigned)(unsigned long long)fatCubinHandle); + cuobjdumpParseBinary((unsigned)(unsigned long long)fatCubinHandle, GPGPUSim_Context()->cuobjdumpSectionList ); fflush(stdout); if ( constant && !global && !ext ) { gpgpu_ptx_sim_register_const_variable(hostVar,deviceName,size); diff --git a/libcuda/cuobjdump.h b/libcuda/cuobjdump.h index 66cd736..49af3e2 100644 --- a/libcuda/cuobjdump.h +++ b/libcuda/cuobjdump.h @@ -1,5 +1,9 @@ #ifndef __cuobjdump_h__ #define __cuobjdump_h__ +#include +#include +#include + struct cuobjdump_parser { yyscan_t scanner; int elfserial; @@ -9,4 +13,67 @@ struct cuobjdump_parser { FILE *sassfile; char filename [1024]; }; + +class cuobjdumpSection { +public: + //Constructor + cuobjdumpSection() { + arch = 0; + identifier = ""; + } + virtual ~cuobjdumpSection() {} + unsigned getArch() {return arch;} + void setArch(unsigned a) {arch = a;} + std::string getIdentifier() {return identifier;} + void setIdentifier(std::string i) {identifier = i;} + virtual void print(){std::cout << "cuobjdump Section: unknown type" << std::endl;} +private: + unsigned arch; + std::string identifier; +}; + +class cuobjdumpELFSection : public cuobjdumpSection +{ +public: + cuobjdumpELFSection() {} + virtual ~cuobjdumpELFSection() { + elffilename = ""; + sassfilename = ""; + } + std::string getELFfilename() {return elffilename;} + void setELFfilename(std::string f) {elffilename = f;} + std::string getSASSfilename() {return sassfilename;} + void setSASSfilename(std::string f) {sassfilename = f;} + virtual void print() { + std::cout << "ELF Section:" << std::endl; + std::cout << "arch: sm_" << getArch() << std::endl; + std::cout << "identifier: " << getIdentifier() << std::endl; + std::cout << "elf filename: " << getELFfilename() << std::endl; + std::cout << "sass filename: " << getSASSfilename() << std::endl; + std::cout << std::endl; + } +private: + std::string elffilename; + std::string sassfilename; +}; + +class cuobjdumpPTXSection : public cuobjdumpSection +{ +public: + cuobjdumpPTXSection(){ + ptxfilename = ""; + } + std::string getPTXfilename() {return ptxfilename;} + void setPTXfilename(std::string f) {ptxfilename = f;} + virtual void print() { + std::cout << "PTX Section:" << std::endl; + std::cout << "arch: sm_" << getArch() << std::endl; + std::cout << "identifier: " << getIdentifier() << std::endl; + std::cout << "ptx filename: " << getPTXfilename() << std::endl; + std::cout << std::endl; + } +private: + std::string ptxfilename; +}; + #endif /* __cuobjdump_h__ */ diff --git a/libcuda/cuobjdump.l b/libcuda/cuobjdump.l index eccc1f2..5a19d65 100644 --- a/libcuda/cuobjdump.l +++ b/libcuda/cuobjdump.l @@ -39,9 +39,9 @@ #define YYDEBUG 1 -void cuobjdump_error(yyscan_t yyscanner, struct cuobjdump_parser* parser, const char* msg); +void cuobjdump_error(yyscan_t yyscanner, struct cuobjdump_parser* parser, std::list &cuobjdumpSectionList, const char* msg); #define YY_DECL int cuobjdump_lex \ - (YYSTYPE * yylval_param , yyscan_t yyscanner, struct cuobjdump_parser* parser) + (YYSTYPE * yylval_param , yyscan_t yyscanner, struct cuobjdump_parser* parser, std::list &cuobjdumpSectionList) %} %option stack @@ -156,10 +156,10 @@ newlines {newline}+ <> BEGIN(INITIAL);return 0; /*No other rule matched. Throw an error*/ -. cuobjdump_error(yyscanner, parser, "Invalid token"); +. cuobjdump_error(yyscanner, parser, cuobjdumpSectionList, "Invalid token"); %% -void cuobjdump_error(yyscan_t yyscanner, struct cuobjdump_parser* parser, const char* msg) +void cuobjdump_error(yyscan_t yyscanner, struct cuobjdump_parser* parser, std::list &cuobjdumpSectionList, const char* msg) { struct yyguts_t * yyg = (struct yyguts_t*)yyscanner; printf(" %s near \"%s\"",msg, yytext); diff --git a/libcuda/cuobjdump.y b/libcuda/cuobjdump.y index fcc863e..8d1bca6 100644 --- a/libcuda/cuobjdump.y +++ b/libcuda/cuobjdump.y @@ -32,25 +32,27 @@ typedef void * yyscan_t; #include "cuobjdump.h" -extern void addCuobjdumpSection(int sectiontype); -void setCuobjdumparch(const char* arch); -void setCuobjdumpidentifier(const char* identifier); -void setCuobjdumpptxfilename(const char* filename); -void setCuobjdumpelffilename(const char* filename); -void setCuobjdumpsassfilename(const char* filename); +extern void addCuobjdumpSection(int sectiontype, std::list &cuobjdumpSectionList); +void setCuobjdumparch(const char* arch, std::list &cuobjdumpSectionList); +void setCuobjdumpidentifier(const char* identifier, std::list &cuobjdumpSectionList); +void setCuobjdumpptxfilename(const char* filename, std::list &cuobjdumpSectionList); +void setCuobjdumpelffilename(const char* filename, std::list &cuobjdumpSectionList); +void setCuobjdumpsassfilename(const char* filename, std::list &cuobjdumpSectionList); %} %define api.pure full %parse-param {yyscan_t scanner} %parse-param {struct cuobjdump_parser* parser} +%parse-param {std::list &cuobjdumpSectionList} %lex-param {yyscan_t scanner} %lex-param {struct cuobjdump_parser* parser} +%lex-param {std::list &cuobjdumpSectionList} %union { char* string_value; } %{ -int yylex(YYSTYPE * yylval_param, yyscan_t yyscanner, struct cuobjdump_parser* parser); -void yyerror(yyscan_t yyscanner, struct cuobjdump_parser* parser, const char* msg); +int yylex(YYSTYPE * yylval_param, yyscan_t yyscanner, struct cuobjdump_parser* parser, std::list &cuobjdumpSectionList); +void yyerror(yyscan_t yyscanner, struct cuobjdump_parser* parser, std::list &cuobjdumpSectionList, const char* msg); %} %token H_SEPARATOR H_ARCH H_CODEVERSION H_PRODUCER H_HOST H_COMPILESIZE H_IDENTIFIER H_UNKNOWN H_COMPRESSED %token CODEVERSION @@ -75,23 +77,23 @@ emptylines : emptylines NEWLINE | ; section : PTXHEADER { - addCuobjdumpSection(0); + addCuobjdumpSection(0, cuobjdumpSectionList); snprintf(parser->filename, 1024, "_cuobjdump_%d.ptx", parser->ptxserial++); parser->ptxfile = fopen(parser->filename, "w"); - setCuobjdumpptxfilename(parser->filename); + setCuobjdumpptxfilename(parser->filename, cuobjdumpSectionList); } headerinfo compressedkeyword identifier ptxcode { fclose(parser->ptxfile); } | ELFHEADER { - addCuobjdumpSection(1); + addCuobjdumpSection(1, cuobjdumpSectionList); snprintf(parser->filename, 1024, "_cuobjdump_%d.elf", parser->elfserial); parser->elffile = fopen(parser->filename, "w"); - setCuobjdumpelffilename(parser->filename); + setCuobjdumpelffilename(parser->filename, cuobjdumpSectionList); } headerinfo compressedkeyword identifier elfcode { fclose(parser->elffile); snprintf(parser->filename, 1024, "_cuobjdump_%d.sass", parser->elfserial++); parser->sassfile = fopen(parser->filename, "w"); - setCuobjdumpsassfilename(parser->filename); + setCuobjdumpsassfilename(parser->filename, cuobjdumpSectionList); } sasscode { fclose(parser->sassfile); }; @@ -101,16 +103,16 @@ headerinfo : H_SEPARATOR NEWLINE H_CODEVERSION CODEVERSION NEWLINE H_PRODUCER H_UNKNOWN NEWLINE H_HOST IDENTIFIER NEWLINE - H_COMPILESIZE IDENTIFIER {setCuobjdumparch($4);}; + H_COMPILESIZE IDENTIFIER {setCuobjdumparch($4, cuobjdumpSectionList);}; | H_SEPARATOR NEWLINE H_ARCH IDENTIFIER NEWLINE H_CODEVERSION CODEVERSION NEWLINE H_PRODUCER IDENTIFIER NEWLINE H_HOST IDENTIFIER NEWLINE - H_COMPILESIZE IDENTIFIER {setCuobjdumparch($4);}; + H_COMPILESIZE IDENTIFIER {setCuobjdumparch($4, cuobjdumpSectionList);}; -identifier : H_IDENTIFIER FILENAME emptylines {setCuobjdumpidentifier($2);} - | {setCuobjdumpidentifier("default");}; +identifier : H_IDENTIFIER FILENAME emptylines {setCuobjdumpidentifier($2, cuobjdumpSectionList);} + | {setCuobjdumpidentifier("default", cuobjdumpSectionList);}; compressedkeyword : H_COMPRESSED emptylines | ; -- cgit v1.3 From 7864d663823cb5f6af7d9d4eb25c4527a8b4e80f Mon Sep 17 00:00:00 2001 From: Mengchi Zhang Date: Wed, 29 May 2019 12:25:48 -0400 Subject: Move libSectionList Signed-off-by: Mengchi Zhang --- libcuda/cuda_runtime_api.cc | 14 +++++++------- 1 file changed, 7 insertions(+), 7 deletions(-) (limited to 'libcuda/cuda_runtime_api.cc') diff --git a/libcuda/cuda_runtime_api.cc b/libcuda/cuda_runtime_api.cc index df7ddc7..225e93b 100644 --- a/libcuda/cuda_runtime_api.cc +++ b/libcuda/cuda_runtime_api.cc @@ -305,6 +305,8 @@ struct CUctx_st { } std::list cuobjdumpSectionList; + std::list libSectionList; + private: _cuda_device_id *m_gpu; // selected gpu std::map m_code; // fat binary handle => global symbol table @@ -497,8 +499,6 @@ enum cuobjdumpSectionType { }; -std::list libSectionList; - // sectiontype: 0 for ptx, 1 for elf void addCuobjdumpSection(int sectiontype, std::list &cuobjdumpSectionList){ if (sectiontype) @@ -2193,7 +2193,7 @@ void extract_code_using_cuobjdump(std::list &cuobjdumpSection fclose(cuobjdump_in); std::getline(libsf, line); } - libSectionList = cuobjdumpSectionList; + context->libSectionList = cuobjdumpSectionList; //Restore the original section list cuobjdumpSectionList = tmpsl; @@ -2382,7 +2382,7 @@ cuobjdumpELFSection* findELFSectionInList(std::list sectionli } //! Find an ELF section in all the known lists -cuobjdumpELFSection* findELFSection(const std::string identifier, std::list cuobjdumpSectionList){ +cuobjdumpELFSection* findELFSection(const std::string identifier, std::list cuobjdumpSectionList, std::list &libSectionList){ cuobjdumpELFSection* sec = findELFSectionInList(cuobjdumpSectionList, identifier); if (sec!=NULL)return sec; sec = findELFSectionInList(libSectionList, identifier); @@ -2417,7 +2417,7 @@ cuobjdumpPTXSection* findPTXSectionInList(std::list sectionli } //! Find an PTX section in all the known lists -cuobjdumpPTXSection* findPTXSection(const std::string identifier, std::list cuobjdumpSectionList){ +cuobjdumpPTXSection* findPTXSection(const std::string identifier, std::list cuobjdumpSectionList, std::list &libSectionList){ cuobjdumpPTXSection* sec = findPTXSectionInList(cuobjdumpSectionList, identifier); if (sec!=NULL)return sec; sec = findPTXSectionInList(libSectionList, identifier); @@ -2503,7 +2503,7 @@ void cuobjdumpParseBinary(unsigned int handle, std::list &cuo cuobjdumpPTXSection* ptx = NULL; const char* pre_load = getenv("CUOBJDUMP_SIM_FILE"); if(pre_load==NULL || strlen(pre_load)==0) - ptx = findPTXSection(fname, context->cuobjdumpSectionList); + ptx = findPTXSection(fname, context->cuobjdumpSectionList, context->libSectionList); char *ptxcode; const char *override_ptx_name = getenv("PTX_SIM_KERNELFILE"); if (override_ptx_name == NULL or getenv("PTX_SIM_USE_PTX_FILE") == NULL or strlen(getenv("PTX_SIM_USE_PTX_FILE"))==0) { @@ -2513,7 +2513,7 @@ void cuobjdumpParseBinary(unsigned int handle, std::list &cuo ptxcode = readfile(override_ptx_name); } if(context->get_device()->get_gpgpu()->get_config().convert_to_ptxplus() ) { - cuobjdumpELFSection* elfsection = findELFSection(ptx->getIdentifier(), context->cuobjdumpSectionList); + cuobjdumpELFSection* elfsection = findELFSection(ptx->getIdentifier(), context->cuobjdumpSectionList, context->libSectionList); assert (elfsection!= NULL); char *ptxplus_str = gpgpu_ptx_sim_convert_ptx_and_sass_to_ptxplus( ptx->getPTXfilename(), -- cgit v1.3 From d9ca3558c774b8b86bb18024bbbf330df53722f7 Mon Sep 17 00:00:00 2001 From: Mengchi Zhang Date: Wed, 29 May 2019 14:48:14 -0400 Subject: Move version_filename and g_cuda_launch_stack Signed-off-by: Mengchi Zhang --- libcuda/cuda_runtime_api.cc | 50 ++++++++++++++++++++++++--------------------- 1 file changed, 27 insertions(+), 23 deletions(-) (limited to 'libcuda/cuda_runtime_api.cc') diff --git a/libcuda/cuda_runtime_api.cc b/libcuda/cuda_runtime_api.cc index 225e93b..f3c827c 100644 --- a/libcuda/cuda_runtime_api.cc +++ b/libcuda/cuda_runtime_api.cc @@ -153,8 +153,6 @@ std::map pinned_memory; //support for pinned memories added std::map pinned_memory_size; std::map g_mallocPtr_Size; int no_of_ptx=0; -//maps sm version number to set of filenames -std::map > version_filename; extern void synchronize(); extern void exit_simulation(); @@ -241,6 +239,8 @@ private: struct _cuda_device_id *m_next; }; +class kernel_config; + struct CUctx_st { CUctx_st( _cuda_device_id *gpu ) { @@ -306,6 +306,9 @@ struct CUctx_st { std::list cuobjdumpSectionList; std::list libSectionList; + //maps sm version number to set of filenames + std::map > version_filename; + std::list g_cuda_launch_stack; private: _cuda_device_id *m_gpu; // selected gpu @@ -486,7 +489,6 @@ typedef std::map event_tracker_t; int CUevent_st::m_next_event_uid; event_tracker_t g_timer_events; int g_active_device = 0; //active gpu that runs the code -std::list g_cuda_launch_stack; extern int cuobjdump_lex_init(yyscan_t* scanner); extern void cuobjdump_set_in (FILE * _in_str ,yyscan_t yyscanner ); @@ -1505,7 +1507,8 @@ __host__ cudaError_t CUDARTAPI cudaConfigureCall(dim3 gridDim, dim3 blockDim, si announce_call(__my_func__); } struct CUstream_st *s = (struct CUstream_st *)stream; - g_cuda_launch_stack.push_back( kernel_config(gridDim,blockDim,sharedMem,s) ); + CUctx_st *context = GPGPUSim_Context(); + context->g_cuda_launch_stack.push_back( kernel_config(gridDim,blockDim,sharedMem,s) ); return g_last_cudaError = cudaSuccess; } @@ -1514,8 +1517,9 @@ __host__ cudaError_t CUDARTAPI cudaSetupArgument(const void *arg, size_t size, s if(g_debug_execution >= 3){ announce_call(__my_func__); } - gpgpusim_ptx_assert( !g_cuda_launch_stack.empty(), "empty launch stack" ); - kernel_config &config = g_cuda_launch_stack.back(); + CUctx_st *context = GPGPUSim_Context(); + gpgpusim_ptx_assert( !context->g_cuda_launch_stack.empty(), "empty launch stack" ); + kernel_config &config = context->g_cuda_launch_stack.back(); config.set_arg(arg,size,offset); printf("GPGPU-Sim PTX: Setting up arguments for %zu bytes starting at 0x%llx..\n",size, (unsigned long long) arg); @@ -1532,8 +1536,8 @@ __host__ cudaError_t CUDARTAPI cudaLaunch( const char *hostFun ) char *mode = getenv("PTX_SIM_MODE_FUNC"); if( mode ) sscanf(mode,"%u", &g_ptx_sim_mode); - gpgpusim_ptx_assert( !g_cuda_launch_stack.empty(), "empty launch stack" ); - kernel_config config = g_cuda_launch_stack.back(); + gpgpusim_ptx_assert( !context->g_cuda_launch_stack.empty(), "empty launch stack" ); + kernel_config config = context->g_cuda_launch_stack.back(); struct CUstream_st *stream = config.get_stream(); printf("\nGPGPU-Sim PTX: cudaLaunch for 0x%p (mode=%s) on stream %u\n", hostFun, g_ptx_sim_mode?"functional simulation":"performance simulation", stream?stream->get_uid():0 ); @@ -1574,14 +1578,14 @@ __host__ cudaError_t CUDARTAPI cudaLaunch( const char *hostFun ) g_checkpoint->load_global_mem(global_mem, f1name); printf("Skipping kernel %d as resuming from kernel %d\n",grid->get_uid(),gpu->resume_kernel ); - g_cuda_launch_stack.pop_back(); + context->g_cuda_launch_stack.pop_back(); return g_last_cudaError = cudaSuccess; } if(gpu->checkpoint_option==1 && (grid->get_uid()>gpu->checkpoint_kernel)) { printf("Skipping kernel %d as checkpoint from kernel %d\n",grid->get_uid(),gpu->checkpoint_kernel ); - g_cuda_launch_stack.pop_back(); + context->g_cuda_launch_stack.pop_back(); return g_last_cudaError = cudaSuccess; } @@ -1589,7 +1593,7 @@ __host__ cudaError_t CUDARTAPI cudaLaunch( const char *hostFun ) kname.c_str(), stream?stream->get_uid():0, gridDim.x,gridDim.y,gridDim.z,blockDim.x,blockDim.y,blockDim.z ); stream_operation op(grid,g_ptx_sim_mode,stream); g_stream_manager->push(op); - g_cuda_launch_stack.pop_back(); + context->g_cuda_launch_stack.pop_back(); return g_last_cudaError = cudaSuccess; } @@ -1969,7 +1973,7 @@ char* get_app_binary_name(std::string abs_path){ } //extracts all ptx files from binary and dumps into prog_name.unique_no.sm_<>.ptx files -void extract_ptx_files_using_cuobjdump(){ +void extract_ptx_files_using_cuobjdump(CUctx_st *context){ extern bool g_cdp_enabled; char command[1000]; char *pytorch_bin = getenv("PYTORCH_BIN"); @@ -2030,10 +2034,10 @@ void extract_ptx_files_using_cuobjdump(){ } std::string vstr = line.substr(pos1+3,pos2-pos1-3); int version = atoi(vstr.c_str()); - if (version_filename.find(version)==version_filename.end()){ - version_filename[version] = std::set(); + if (context->version_filename.find(version)==context->version_filename.end()){ + context->version_filename[version] = std::set(); } - version_filename[version].insert(line); + context->version_filename[version].insert(line); } } @@ -2091,7 +2095,7 @@ void extract_code_using_cuobjdump(std::list &cuobjdumpSection //dump ptx for all individial ptx files into sepearte files which is later used by ptxas. int result=0; #if (CUDART_VERSION >= 6000) - extract_ptx_files_using_cuobjdump(); + extract_ptx_files_using_cuobjdump(context); return; #endif //TODO: redundant to dump twice. how can it be prevented? @@ -2450,7 +2454,7 @@ void cuobjdumpRegisterFatBinary(unsigned int handle, const char* filename){ } //! Either submit PTX for simulation or convert SASS to PTXPlus and submit it -void cuobjdumpParseBinary(unsigned int handle, std::list &cuobjdumpSectionList){ +void cuobjdumpParseBinary(unsigned int handle){ if(fatbin_registered[handle]) return; fatbin_registered[handle] = true; @@ -2467,7 +2471,7 @@ void cuobjdumpParseBinary(unsigned int handle, std::list &cuo #if (CUDART_VERSION >= 6000) //loops through all ptx files from smallest sm version to largest std::map >::iterator itr_m; - for (itr_m = version_filename.begin(); itr_m!=version_filename.end(); itr_m++){ + for (itr_m = context->version_filename.begin(); itr_m!=context->version_filename.end(); itr_m++){ std::set::iterator itr_s; for (itr_s = itr_m->second.begin(); itr_s!=itr_m->second.end(); itr_s++){ std::string ptx_filename = *itr_s; @@ -2479,7 +2483,7 @@ void cuobjdumpParseBinary(unsigned int handle, std::list &cuo context->add_binary(symtab, handle); load_static_globals(symtab,STATIC_ALLOC_LIMIT,0xFFFFFFFF,context->get_device()->get_gpgpu()); load_constants(symtab,STATIC_ALLOC_LIMIT,context->get_device()->get_gpgpu()); - for (itr_m = version_filename.begin(); itr_m!=version_filename.end(); itr_m++){ + for (itr_m = context->version_filename.begin(); itr_m!=context->version_filename.end(); itr_m++){ std::set::iterator itr_s; for (itr_s = itr_m->second.begin(); itr_s!=itr_m->second.end(); itr_s++){ std::string ptx_filename = *itr_s; @@ -2491,8 +2495,8 @@ void cuobjdumpParseBinary(unsigned int handle, std::list &cuo #endif unsigned max_capability = 0; - for ( std::list::iterator iter = cuobjdumpSectionList.begin(); - iter != cuobjdumpSectionList.end(); + for ( std::list::iterator iter = context->cuobjdumpSectionList.begin(); + iter != context->cuobjdumpSectionList.end(); iter++){ unsigned capability = (*iter)->getArch(); if (capability > max_capability) max_capability = capability; @@ -2716,7 +2720,7 @@ void CUDARTAPI __cudaRegisterFunction( printf("GPGPU-Sim PTX: __cudaRegisterFunction %s : hostFun 0x%p, fat_cubin_handle = %u\n", deviceFun, hostFun, fat_cubin_handle); if(context->get_device()->get_gpgpu()->get_config().use_cuobjdump()) - cuobjdumpParseBinary(fat_cubin_handle, context->cuobjdumpSectionList); + cuobjdumpParseBinary(fat_cubin_handle); context->register_function( fat_cubin_handle, hostFun, deviceFun ); } @@ -2736,7 +2740,7 @@ extern void __cudaRegisterVar( printf("GPGPU-Sim PTX: __cudaRegisterVar: hostVar = %p; deviceAddress = %s; deviceName = %s\n", hostVar, deviceAddress, deviceName); printf("GPGPU-Sim PTX: __cudaRegisterVar: Registering const memory space of %d bytes\n", size); if(GPGPUSim_Context()->get_device()->get_gpgpu()->get_config().use_cuobjdump()) - cuobjdumpParseBinary((unsigned)(unsigned long long)fatCubinHandle, GPGPUSim_Context()->cuobjdumpSectionList ); + cuobjdumpParseBinary((unsigned)(unsigned long long)fatCubinHandle); fflush(stdout); if ( constant && !global && !ext ) { gpgpu_ptx_sim_register_const_variable(hostVar,deviceName,size); -- cgit v1.3 From 9e3a9ac5ed0a70ec9b048bd3cb4df781687e85f8 Mon Sep 17 00:00:00 2001 From: Mengchi Zhang Date: Wed, 29 May 2019 19:23:40 -0400 Subject: Move fatbin etc globals Signed-off-by: Mengchi Zhang --- libcuda/cuda_runtime_api.cc | 111 +++++++++++++++++++++----------------------- src/cuda-sim/ptx_loader.cc | 2 +- src/cuda-sim/ptx_loader.h | 2 +- 3 files changed, 56 insertions(+), 59 deletions(-) (limited to 'libcuda/cuda_runtime_api.cc') diff --git a/libcuda/cuda_runtime_api.cc b/libcuda/cuda_runtime_api.cc index f3c827c..fcd5b07 100644 --- a/libcuda/cuda_runtime_api.cc +++ b/libcuda/cuda_runtime_api.cc @@ -149,10 +149,6 @@ typedef void * yyscan_t; #include #endif -std::map pinned_memory; //support for pinned memories added -std::map pinned_memory_size; -std::map g_mallocPtr_Size; -int no_of_ptx=0; extern void synchronize(); extern void exit_simulation(); @@ -241,12 +237,25 @@ private: class kernel_config; +#ifndef OPENGL_SUPPORT +typedef unsigned long GLuint; +#endif + +struct glbmap_entry { + GLuint m_bufferObj; + void *m_devPtr; + size_t m_size; + struct glbmap_entry *m_next; +}; + struct CUctx_st { CUctx_st( _cuda_device_id *gpu ) { m_gpu = gpu; m_binary_info.cmem = 0; m_binary_info.gmem = 0; + no_of_ptx=0; + g_glbmap = NULL; } _cuda_device_id *get_device() { return m_gpu; } @@ -309,6 +318,16 @@ struct CUctx_st { //maps sm version number to set of filenames std::map > version_filename; std::list g_cuda_launch_stack; + std::mapfatbin_registered; + std::map fatbinmap; + std::map g_mallocPtr_Size; + std::map name_symtab; + std::map pinned_memory; //support for pinned memories added + std::map pinned_memory_size; + int no_of_ptx; + typedef struct glbmap_entry glbmap_entry_t; + + glbmap_entry_t* g_glbmap; private: _cuda_device_id *m_gpu; // selected gpu @@ -573,7 +592,7 @@ __host__ cudaError_t CUDARTAPI cudaMalloc(void **devPtr, size_t size) *devPtr = context->get_device()->get_gpgpu()->gpu_malloc(size); if(g_debug_execution >= 3){ printf("GPGPU-Sim PTX: cudaMallocing %zu bytes starting at 0x%llx..\n",size, (unsigned long long) *devPtr); - g_mallocPtr_Size[(unsigned long long)*devPtr] = size; + context->g_mallocPtr_Size[(unsigned long long)*devPtr] = size; } if ( *devPtr ) { return g_last_cudaError = cudaSuccess; @@ -587,11 +606,11 @@ __host__ cudaError_t CUDARTAPI cudaMallocHost(void **ptr, size_t size) if(g_debug_execution >= 3){ announce_call(__my_func__); } - GPGPUSim_Context(); + CUctx_st* context = GPGPUSim_Context(); *ptr = malloc(size); if ( *ptr ) { //track pinned memory size allocated in the host so that same amount of memory is also allocated in GPU. - pinned_memory_size[*ptr]=size; + context->pinned_memory_size[*ptr]=size; return g_last_cudaError = cudaSuccess; } else { return g_last_cudaError = cudaErrorMemoryAllocation; @@ -2010,11 +2029,11 @@ void extract_ptx_files_using_cuobjdump(CUctx_st *context){ printf("ERROR: command: %s failed \n",command); exit(0); } - no_of_ptx++; + context->no_of_ptx++; } } - if(!no_of_ptx){ + if(!context->no_of_ptx){ printf("WARNING: Number of ptx in the executable file are 0. One of the reasons might be\n"); printf("\t1. CDP is enabled\n"); printf("\t2. When using PyTorch, PYTORCH_BIN is not set correctly\n"); @@ -2444,25 +2463,22 @@ void cuobjdumpInit(std::list &cuobjdumpSectionList){ } } -std::map fatbinmap; -std::mapfatbin_registered; -std::map name_symtab; //! Keep track of the association between filename and cubin handle -void cuobjdumpRegisterFatBinary(unsigned int handle, const char* filename){ - fatbinmap[handle] = filename; +void cuobjdumpRegisterFatBinary(unsigned int handle, const char* filename, CUctx_st *context){ + context->fatbinmap[handle] = filename; } //! Either submit PTX for simulation or convert SASS to PTXPlus and submit it void cuobjdumpParseBinary(unsigned int handle){ - if(fatbin_registered[handle]) return; - fatbin_registered[handle] = true; CUctx_st *context = GPGPUSim_Context(); - std::string fname = fatbinmap[handle]; + if(context->fatbin_registered[handle]) return; + context->fatbin_registered[handle] = true; + std::string fname = context->fatbinmap[handle]; - if (name_symtab.find(fname) != name_symtab.end()) { - symbol_table *symtab = name_symtab[fname]; + if (context->name_symtab.find(fname) != context->name_symtab.end()) { + symbol_table *symtab = context->name_symtab[fname]; context->add_binary(symtab, handle); return; } @@ -2479,7 +2495,7 @@ void cuobjdumpParseBinary(unsigned int handle){ symtab = gpgpu_ptx_sim_load_ptx_from_filename( ptx_filename.c_str() ); } } - name_symtab[fname] = symtab; + context->name_symtab[fname] = symtab; context->add_binary(symtab, handle); load_static_globals(symtab,STATIC_ALLOC_LIMIT,0xFFFFFFFF,context->get_device()->get_gpgpu()); load_constants(symtab,STATIC_ALLOC_LIMIT,context->get_device()->get_gpgpu()); @@ -2526,18 +2542,18 @@ void cuobjdumpParseBinary(unsigned int handle){ symtab=gpgpu_ptx_sim_load_ptx_from_string(ptxplus_str, handle); printf("Adding %s with cubin handle %u\n", ptx->getPTXfilename().c_str(), handle); context->add_binary(symtab, handle); - gpgpu_ptxinfo_load_from_string( ptxcode, handle, max_capability ); + gpgpu_ptxinfo_load_from_string( ptxcode, handle, max_capability, context->no_of_ptx ); delete[] ptxplus_str; } else { symtab=gpgpu_ptx_sim_load_ptx_from_string(ptxcode, handle); //if CUOBJDUMP_SIM_FILE is not set, ptx is NULL. So comment below. //printf("Adding %s with cubin handle %u\n", ptx->getPTXfilename().c_str(), handle); context->add_binary(symtab, handle); - gpgpu_ptxinfo_load_from_string( ptxcode, handle, max_capability ); + gpgpu_ptxinfo_load_from_string( ptxcode, handle, max_capability, context->no_of_ptx ); } load_static_globals(symtab,STATIC_ALLOC_LIMIT,0xFFFFFFFF,context->get_device()->get_gpgpu()); load_constants(symtab,STATIC_ALLOC_LIMIT,context->get_device()->get_gpgpu()); - name_symtab[fname] = symtab; + context->name_symtab[fname] = symtab; //TODO: Remove temporarily files as per configurations } @@ -2606,7 +2622,7 @@ void** CUDARTAPI __cudaRegisterFatBinary( void *fatCubin ) */ assert(fat_cubin_handle >= 1); if (fat_cubin_handle==1) cuobjdumpInit(context->cuobjdumpSectionList); - cuobjdumpRegisterFatBinary(fat_cubin_handle, filename); + cuobjdumpRegisterFatBinary(fat_cubin_handle, filename, context); return (void**)fat_cubin_handle; } @@ -2658,7 +2674,7 @@ void** CUDARTAPI __cudaRegisterFatBinary( void *fatCubin ) } else { symtab=gpgpu_ptx_sim_load_ptx_from_string(ptx,source_num); context->add_binary(symtab,fat_cubin_handle); - gpgpu_ptxinfo_load_from_string( ptx, source_num, max_capability ); + gpgpu_ptxinfo_load_from_string( ptx, source_num, max_capability, context->no_of_ptx ); } source_num++; load_static_globals(symtab,STATIC_ALLOC_LIMIT,0xFFFFFFFF,context->get_device()->get_gpgpu()); @@ -2818,10 +2834,6 @@ char __cudaInitModule( } -#ifndef OPENGL_SUPPORT -typedef unsigned long GLuint; -#endif - cudaError_t cudaGLRegisterBufferObject(GLuint bufferObj) { if(g_debug_execution >= 3){ @@ -2831,16 +2843,6 @@ cudaError_t cudaGLRegisterBufferObject(GLuint bufferObj) return g_last_cudaError = cudaSuccess; } -struct glbmap_entry { - GLuint m_bufferObj; - void *m_devPtr; - size_t m_size; - struct glbmap_entry *m_next; -}; -typedef struct glbmap_entry glbmap_entry_t; - -glbmap_entry_t* g_glbmap = NULL; - cudaError_t cudaGLMapBufferObject(void** devPtr, GLuint bufferObj) { if(g_debug_execution >= 3){ @@ -2850,7 +2852,7 @@ cudaError_t cudaGLMapBufferObject(void** devPtr, GLuint bufferObj) GLint buffer_size=0; CUctx_st* ctx = GPGPUSim_Context(); - glbmap_entry_t *p = g_glbmap; + glbmap_entry_t *p = ctx->g_glbmap; while ( p && p->m_bufferObj != bufferObj ) p = p->m_next; if ( p == NULL ) { @@ -2861,8 +2863,8 @@ cudaError_t cudaGLMapBufferObject(void** devPtr, GLuint bufferObj) // create entry and insert to front of list glbmap_entry_t *n = (glbmap_entry_t *) calloc(1,sizeof(glbmap_entry_t)); - n->m_next = g_glbmap; - g_glbmap = n; + n->m_next = ctx->g_glbmap; + ctx->g_glbmap = n; // initialize entry n->m_bufferObj = bufferObj; @@ -2903,7 +2905,8 @@ cudaError_t cudaGLUnmapBufferObject(GLuint bufferObj) announce_call(__my_func__); } #ifdef OPENGL_SUPPORT - glbmap_entry_t *p = g_glbmap; + CUctx_st* ctx = GPGPUSim_Context(); + glbmap_entry_t *p = ctx->g_glbmap; while ( p && p->m_bufferObj != bufferObj ) p = p->m_next; if ( p == NULL ) @@ -2943,7 +2946,8 @@ cudaError_t CUDARTAPI cudaHostAlloc(void **pHost, size_t bytes, unsigned int fl *pHost = malloc(bytes); //need to track the size allocated so that cudaHostGetDevicePointer() can function properly. //TODO: vary this function behavior based on flags value (following nvidia documentation) - pinned_memory_size[*pHost]=bytes; + CUctx_st* context = GPGPUSim_Context(); + context->pinned_memory_size[*pHost]=bytes; if( *pHost ) return g_last_cudaError = cudaSuccess; else @@ -2960,16 +2964,16 @@ cudaError_t CUDARTAPI cudaHostGetDevicePointer(void **pDevice, void *pHost, unsi flags=0; CUctx_st* context = GPGPUSim_Context(); gpgpu_t *gpu = context->get_device()->get_gpgpu(); - std::map::const_iterator i = pinned_memory_size.find(pHost); - assert(i != pinned_memory_size.end()); + std::map::const_iterator i = context->pinned_memory_size.find(pHost); + assert(i != context->pinned_memory_size.end()); size_t size = i->second; *pDevice = gpu->gpu_malloc(size); if(g_debug_execution >= 3){ printf("GPGPU-Sim PTX: cudaMallocing %zu bytes starting at 0x%llx..\n",size, (unsigned long long) *pDevice); - g_mallocPtr_Size[(unsigned long long)*pDevice] = size; + context->g_mallocPtr_Size[(unsigned long long)*pDevice] = size; } if ( *pDevice ) { - pinned_memory[pHost]=pDevice; + context->pinned_memory[pHost]=pDevice; //Copy contents in cpu to gpu gpu->memcpy_to_gpu((size_t)*pDevice,pHost,size); return g_last_cudaError = cudaSuccess; @@ -3204,13 +3208,6 @@ int CUDARTAPI __cudaSynchronizeThreads(void**, void*) //////// -extern int ptx_parse(); -extern int ptx__scan_string(const char*); -extern FILE *ptx_in; - -extern int ptxinfo_parse(); -extern FILE *ptxinfo_in; - /// static functions static int load_static_globals( symbol_table *symtab, unsigned min_gaddr, unsigned max_gaddr, gpgpu_t *gpu ) @@ -3330,7 +3327,7 @@ kernel_info_t *gpgpu_cuda_ptx_sim_init_grid( const char *hostFun, fflush(stdout); if(g_debug_execution >= 4){ - entry->ptx_jit_config(g_mallocPtr_Size, result->get_param_memory(), (gpgpu_t *) context->get_device()->get_gpgpu(), gridDim, blockDim); + entry->ptx_jit_config(context->g_mallocPtr_Size, result->get_param_memory(), (gpgpu_t *) context->get_device()->get_gpgpu(), gridDim, blockDim); } return result; @@ -3815,7 +3812,7 @@ cuLinkAddFile(CUlinkState state, CUjitInputType type, const char *path, strcat(file,path); symbol_table *symtab = gpgpu_ptx_sim_load_ptx_from_filename( file ); std::string fname(path); - name_symtab[fname] = symtab; + context->name_symtab[fname] = symtab; context->add_binary(symtab, 1); load_static_globals(symtab,STATIC_ALLOC_LIMIT,0xFFFFFFFF,context->get_device()->get_gpgpu()); load_constants(symtab,STATIC_ALLOC_LIMIT,context->get_device()->get_gpgpu()); diff --git a/src/cuda-sim/ptx_loader.cc b/src/cuda-sim/ptx_loader.cc index 735ff84..f037c34 100644 --- a/src/cuda-sim/ptx_loader.cc +++ b/src/cuda-sim/ptx_loader.cc @@ -367,7 +367,7 @@ void gpgpu_ptx_info_load_from_filename( const char *filename, unsigned sm_versio fclose(ptxinfo_in); } -void gpgpu_ptxinfo_load_from_string( const char *p_for_info, unsigned source_num, unsigned sm_version ) +void gpgpu_ptxinfo_load_from_string( const char *p_for_info, unsigned source_num, unsigned sm_version, int no_of_ptx ) { //do ptxas for individual files instead of one big embedded ptx. This prevents the duplicate defs and declarations. char ptx_file[1000]; diff --git a/src/cuda-sim/ptx_loader.h b/src/cuda-sim/ptx_loader.h index 36e439e..c4d8292 100644 --- a/src/cuda-sim/ptx_loader.h +++ b/src/cuda-sim/ptx_loader.h @@ -44,7 +44,7 @@ extern int no_of_ptx; //counter to track number of ptx files to be extracted in class symbol_table *gpgpu_ptx_sim_load_ptx_from_string( const char *p, unsigned source_num ); class symbol_table *gpgpu_ptx_sim_load_ptx_from_filename( const char *filename ); -void gpgpu_ptxinfo_load_from_string( const char *p_for_info, unsigned source_num, unsigned sm_version=20 ); +void gpgpu_ptxinfo_load_from_string( const char *p_for_info, unsigned source_num, unsigned sm_version=20, int no_of_ptx=0 ); void gpgpu_ptx_info_load_from_filename( const char *filename, unsigned sm_version ); char* gpgpu_ptx_sim_convert_ptx_and_sass_to_ptxplus(const std::string ptx_str, const std::string sass_str, const std::string elf_str); bool keep_intermediate_files(); -- cgit v1.3 From ed9f0e6b2a99840e9649551825a40a04e236dcd9 Mon Sep 17 00:00:00 2001 From: Mahmoud Date: Thu, 30 May 2019 18:28:15 -0400 Subject: adding new values to gpu context --- libcuda/cuda_runtime_api.cc | 42 +++++++++++++++++++------------------ src/abstract_hardware_model.cc | 9 ++++---- src/abstract_hardware_model.h | 2 +- src/cuda-sim/cuda-sim.cc | 36 +++++++++++++++---------------- src/cuda-sim/cuda_device_runtime.cc | 5 +++-- src/gpgpusim_entrypoint.cc | 12 +++++++++-- src/gpgpusim_entrypoint.h | 24 ++++++++++++++++----- 7 files changed, 78 insertions(+), 52 deletions(-) (limited to 'libcuda/cuda_runtime_api.cc') diff --git a/libcuda/cuda_runtime_api.cc b/libcuda/cuda_runtime_api.cc index fcd5b07..4990edf 100644 --- a/libcuda/cuda_runtime_api.cc +++ b/libcuda/cuda_runtime_api.cc @@ -185,7 +185,7 @@ struct cudaArray cudaError_t g_last_cudaError = cudaSuccess; -extern stream_manager *g_stream_manager; +//extern stream_manager *g_stream_manager(); void register_ptx_function( const char *name, function_info *impl ) { @@ -375,7 +375,8 @@ private: struct _cuda_device_id *GPGPUSim_Init() { - static _cuda_device_id *the_device = NULL; + //static _cuda_device_id *the_device = NULL; + _cuda_device_id *the_device = GPGPUsim_ctx_ptr()->the_cude_device; if( !the_device ) { gpgpu_sim *the_gpu = gpgpu_ptx_sim_init_perf(); @@ -428,7 +429,8 @@ struct _cuda_device_id *GPGPUSim_Init() static CUctx_st* GPGPUSim_Context() { - static CUctx_st *the_context = NULL; + //static CUctx_st *the_context = NULL; + CUctx_st *the_context = GPGPUsim_ctx_ptr()->the_context; if( the_context == NULL ) { _cuda_device_id *the_gpu = GPGPUSim_Init(); the_context = new CUctx_st(the_gpu); @@ -699,21 +701,21 @@ __host__ cudaError_t CUDARTAPI cudaMemcpy(void *dst, const void *src, size_t cou if(g_debug_execution >= 3) printf("GPGPU-Sim PTX: cudaMemcpy(): devPtr = %p\n", dst); if( kind == cudaMemcpyHostToDevice ) - g_stream_manager->push( stream_operation(src,(size_t)dst,count,0) ); + g_stream_manager()->push( stream_operation(src,(size_t)dst,count,0) ); else if( kind == cudaMemcpyDeviceToHost ) - g_stream_manager->push( stream_operation((size_t)src,dst,count,0) ); + g_stream_manager()->push( stream_operation((size_t)src,dst,count,0) ); else if( kind == cudaMemcpyDeviceToDevice ) - g_stream_manager->push( stream_operation((size_t)src,(size_t)dst,count,0) ); + g_stream_manager()->push( stream_operation((size_t)src,(size_t)dst,count,0) ); else if ( kind == cudaMemcpyDefault ) { if ((size_t)src >= GLOBAL_HEAP_START) { if ((size_t)dst >= GLOBAL_HEAP_START) - g_stream_manager->push( stream_operation((size_t)src,(size_t)dst,count,0) ); // device to device + g_stream_manager()->push( stream_operation((size_t)src,(size_t)dst,count,0) ); // device to device else - g_stream_manager->push( stream_operation((size_t)src,dst,count,0) ); // device to host + g_stream_manager()->push( stream_operation((size_t)src,dst,count,0) ); // device to host } else { if ((size_t)dst >= GLOBAL_HEAP_START) - g_stream_manager->push( stream_operation(src,(size_t)dst,count,0) ); + g_stream_manager()->push( stream_operation(src,(size_t)dst,count,0) ); else { printf("GPGPU-Sim PTX: cudaMemcpy - ERROR : unsupported transfer: host to host\n"); abort(); @@ -855,7 +857,7 @@ __host__ cudaError_t CUDARTAPI cudaMemcpyToSymbol(const char *symbol, const void assert(kind == cudaMemcpyHostToDevice); printf("GPGPU-Sim PTX: cudaMemcpyToSymbol: symbol = %p\n", symbol); //stream_operation( const char *symbol, const void *src, size_t count, size_t offset ) - g_stream_manager->push( stream_operation(src,symbol,count,offset,0) ); + g_stream_manager()->push( stream_operation(src,symbol,count,offset,0) ); //gpgpu_ptx_sim_memcpy_symbol(symbol,src,count,offset,1,context->get_device()->get_gpgpu()); return g_last_cudaError = cudaSuccess; } @@ -869,7 +871,7 @@ __host__ cudaError_t CUDARTAPI cudaMemcpyFromSymbol(void *dst, const char *symbo //CUctx_st *context = GPGPUSim_Context(); assert(kind == cudaMemcpyDeviceToHost); printf("GPGPU-Sim PTX: cudaMemcpyFromSymbol: symbol = %p\n", symbol); - g_stream_manager->push( stream_operation(symbol,dst,count,offset,0) ); + g_stream_manager()->push( stream_operation(symbol,dst,count,offset,0) ); //gpgpu_ptx_sim_memcpy_symbol(symbol,dst,count,offset,0,context->get_device()->get_gpgpu()); return g_last_cudaError = cudaSuccess; } @@ -898,9 +900,9 @@ __host__ cudaError_t CUDARTAPI cudaMemcpyAsync(void *dst, const void *src, size_ } struct CUstream_st *s = (struct CUstream_st *)stream; switch( kind ) { - case cudaMemcpyHostToDevice: g_stream_manager->push( stream_operation(src,(size_t)dst,count,s) ); break; - case cudaMemcpyDeviceToHost: g_stream_manager->push( stream_operation((size_t)src,dst,count,s) ); break; - case cudaMemcpyDeviceToDevice: g_stream_manager->push( stream_operation((size_t)src,(size_t)dst,count,s) ); break; + case cudaMemcpyHostToDevice: g_stream_manager()->push( stream_operation(src,(size_t)dst,count,s) ); break; + case cudaMemcpyDeviceToHost: g_stream_manager()->push( stream_operation((size_t)src,dst,count,s) ); break; + case cudaMemcpyDeviceToDevice: g_stream_manager()->push( stream_operation((size_t)src,(size_t)dst,count,s) ); break; default: abort(); } @@ -1611,7 +1613,7 @@ __host__ cudaError_t CUDARTAPI cudaLaunch( const char *hostFun ) printf("GPGPU-Sim PTX: pushing kernel \'%s\' to stream %u, gridDim= (%u,%u,%u) blockDim = (%u,%u,%u) \n", kname.c_str(), stream?stream->get_uid():0, gridDim.x,gridDim.y,gridDim.z,blockDim.x,blockDim.y,blockDim.z ); stream_operation op(grid,g_ptx_sim_mode,stream); - g_stream_manager->push(op); + g_stream_manager()->push(op); context->g_cuda_launch_stack.pop_back(); return g_last_cudaError = cudaSuccess; } @@ -1650,7 +1652,7 @@ __host__ cudaError_t CUDARTAPI cudaStreamCreate(cudaStream_t *stream) printf("GPGPU-Sim PTX: cudaStreamCreate\n"); #if (CUDART_VERSION >= 3000) *stream = new struct CUstream_st(); - g_stream_manager->add_stream(*stream); + g_stream_manager()->add_stream(*stream); #else *stream = 0; printf("GPGPU-Sim PTX: WARNING: Asynchronous kernel execution not supported (%s)\n", __my_func__); @@ -1689,7 +1691,7 @@ __host__ cudaError_t CUDARTAPI cudaStreamDestroy(cudaStream_t stream) //per-stream synchronization required for application using external libraries without explicit synchronization in the code to //avoid the stream_manager from spinning forever to destroy non-empty streams without making any forward progress. stream->synchronize(); - g_stream_manager->destroy_stream(stream); + g_stream_manager()->destroy_stream(stream); #endif return g_last_cudaError = cudaSuccess; } @@ -1769,7 +1771,7 @@ __host__ cudaError_t CUDARTAPI cudaEventRecord(cudaEvent_t event, cudaStream_t s if( !e ) return g_last_cudaError = cudaErrorUnknown; struct CUstream_st *s = (struct CUstream_st *)stream; stream_operation op(e,s); - g_stream_manager->push(op); + g_stream_manager()->push(op); return g_last_cudaError = cudaSuccess; } @@ -1785,11 +1787,11 @@ __host__ cudaError_t CUDARTAPI cudaStreamWaitEvent(cudaStream_t stream, cudaEven return g_last_cudaError = cudaSuccess; } if (!stream){ - g_stream_manager->pushCudaStreamWaitEventToAllStreams(e, flags); + g_stream_manager()->pushCudaStreamWaitEventToAllStreams(e, flags); } else { struct CUstream_st *s = (struct CUstream_st *)stream; stream_operation op(s,e,flags); - g_stream_manager->push(op); + g_stream_manager()->push(op); } return g_last_cudaError = cudaSuccess; } diff --git a/src/abstract_hardware_model.cc b/src/abstract_hardware_model.cc index 63b139e..7755477 100644 --- a/src/abstract_hardware_model.cc +++ b/src/abstract_hardware_model.cc @@ -34,6 +34,7 @@ #include "cuda-sim/cuda-sim.h" #include "gpgpu-sim/gpu-sim.h" #include "option_parser.h" +#include "gpgpusim_entrypoint.h" #include #include #include @@ -771,14 +772,14 @@ void kernel_info_t::notify_parent_finished() { extern unsigned long long g_total_param_size; g_total_param_size -= ((m_kernel_entry->get_args_aligned_size() + 255)/256*256); m_parent_kernel->remove_child(this); - g_stream_manager->register_finished_kernel(m_parent_kernel->get_uid()); + g_stream_manager()->register_finished_kernel(m_parent_kernel->get_uid()); } } CUstream_st * kernel_info_t::create_stream_cta(dim3 ctaid) { assert(get_default_stream_cta(ctaid)); CUstream_st * stream = new CUstream_st(); - g_stream_manager->add_stream(stream); + g_stream_manager()->add_stream(stream); assert(m_cta_streams.find(ctaid) != m_cta_streams.end()); assert(m_cta_streams[ctaid].size() >= 1); //must have default stream m_cta_streams[ctaid].push_back(stream); @@ -794,7 +795,7 @@ CUstream_st * kernel_info_t::get_default_stream_cta(dim3 ctaid) { else { m_cta_streams[ctaid] = std::list(); CUstream_st * stream = new CUstream_st(); - g_stream_manager->add_stream(stream); + g_stream_manager()->add_stream(stream); m_cta_streams[ctaid].push_back(stream); return stream; } @@ -826,7 +827,7 @@ void kernel_info_t::destroy_cta_streams() { for(auto s = m_cta_streams.begin(); s != m_cta_streams.end(); s++) { stream_size += s->second.size(); for(auto ss = s->second.begin(); ss != s->second.end(); ss++) - g_stream_manager->destroy_stream(*ss); + g_stream_manager()->destroy_stream(*ss); s->second.clear(); } printf("size %lu\n", stream_size); diff --git a/src/abstract_hardware_model.h b/src/abstract_hardware_model.h index 1735c2f..77d5f58 100644 --- a/src/abstract_hardware_model.h +++ b/src/abstract_hardware_model.h @@ -197,7 +197,7 @@ void increment_x_then_y_then_z( dim3 &i, const dim3 &bound); #include "stream_manager.h" class stream_manager; struct CUstream_st; -extern stream_manager * g_stream_manager; +//extern stream_manager * g_stream_manager; //support for pinned memories added extern std::map pinned_memory; extern std::map pinned_memory_size; diff --git a/src/cuda-sim/cuda-sim.cc b/src/cuda-sim/cuda-sim.cc index e733b7f..a456978 100644 --- a/src/cuda-sim/cuda-sim.cc +++ b/src/cuda-sim/cuda-sim.cc @@ -448,8 +448,8 @@ void gpgpu_t::memcpy_to_gpu( size_t dst_start_addr, const void *src, size_t coun m_global_mem->write(dst_start_addr+n,1, src_data+n,NULL,NULL); // Copy into the performance model. - extern gpgpu_sim* g_the_gpu; - g_the_gpu->perf_memcpy_to_gpu(dst_start_addr, count); + //extern gpgpu_sim* g_the_gpu; + g_the_gpu()->perf_memcpy_to_gpu(dst_start_addr, count); if(g_debug_execution >= 3) { printf( " done.\n"); fflush(stdout); @@ -467,8 +467,8 @@ void gpgpu_t::memcpy_from_gpu( void *dst, size_t src_start_addr, size_t count ) m_global_mem->read(src_start_addr+n,1,dst_data+n); // Copy into the performance model. - extern gpgpu_sim* g_the_gpu; - g_the_gpu->perf_memcpy_to_gpu(src_start_addr, count); + //extern gpgpu_sim* g_the_gpu; + g_the_gpu()->perf_memcpy_to_gpu(src_start_addr, count); if(g_debug_execution >= 3) { printf( " done.\n"); fflush(stdout); @@ -1270,8 +1270,8 @@ void function_info::finalize( memory_space *param_mem ) void function_info::param_to_shared( memory_space *shared_mem, symbol_table *symtab ) { // TODO: call this only for PTXPlus with GT200 models - extern gpgpu_sim* g_the_gpu; - if (not g_the_gpu->get_config().convert_to_ptxplus()) return; + //extern gpgpu_sim* g_the_gpu; + if (not g_the_gpu()->get_config().convert_to_ptxplus()) return; // copies parameters into simulated shared memory for( std::map::iterator i=m_ptx_kernel_param_info.begin(); i!=m_ptx_kernel_param_info.end(); i++ ) { @@ -2150,7 +2150,7 @@ void gpgpu_cuda_ptx_sim_main_func( kernel_info_t &kernel, bool openCL ) printf("GPGPU-Sim: Performing Functional Simulation, executing kernel %s...\n",kernel.name().c_str()); //using a shader core object for book keeping, it is not needed but as most function built for performance simulation need it we use it here - extern gpgpu_sim *g_the_gpu; + //extern gpgpu_sim *g_the_gpu; //before we execute, we should do PDOM analysis for functional simulation scenario. function_info *kernel_func_info = kernel.entry(); const struct gpgpu_ptx_sim_info *kernel_info = ptx_sim_kernel_info(kernel_func_info); @@ -2165,7 +2165,7 @@ void gpgpu_cuda_ptx_sim_main_func( kernel_info_t &kernel, bool openCL ) kernel_func_info->set_pdom(); } - unsigned max_cta_tot = max_cta(kernel_info,kernel.threads_per_cta(), g_the_gpu->getShaderCoreConfig()->warp_size, g_the_gpu->getShaderCoreConfig()->n_thread_per_shader, g_the_gpu->getShaderCoreConfig()->gpgpu_shmem_size, g_the_gpu->getShaderCoreConfig()->gpgpu_shader_registers, g_the_gpu->getShaderCoreConfig()->max_cta_per_core); + unsigned max_cta_tot = max_cta(kernel_info,kernel.threads_per_cta(), g_the_gpu()->getShaderCoreConfig()->warp_size, g_the_gpu()->getShaderCoreConfig()->n_thread_per_shader, g_the_gpu()->getShaderCoreConfig()->gpgpu_shmem_size, g_the_gpu()->getShaderCoreConfig()->gpgpu_shader_registers, g_the_gpu()->getShaderCoreConfig()->max_cta_per_core); printf("Max CTA : %d\n",max_cta_tot); @@ -2173,11 +2173,11 @@ void gpgpu_cuda_ptx_sim_main_func( kernel_info_t &kernel, bool openCL ) int inst_count=50; - int cp_op= g_the_gpu->checkpoint_option; - int cp_CTA = g_the_gpu->checkpoint_CTA; - int cp_kernel= g_the_gpu->checkpoint_kernel; - cp_count= g_the_gpu->checkpoint_insn_Y; - cp_cta_resume= g_the_gpu->checkpoint_CTA_t; + int cp_op= g_the_gpu()->checkpoint_option; + int cp_CTA = g_the_gpu()->checkpoint_CTA; + int cp_kernel= g_the_gpu()->checkpoint_kernel; + cp_count= g_the_gpu()->checkpoint_insn_Y; + cp_cta_resume= g_the_gpu()->checkpoint_CTA_t; int cta_launched =0; //we excute the kernel one CTA (Block) at the time, as synchronization functions work block wise @@ -2189,8 +2189,8 @@ void gpgpu_cuda_ptx_sim_main_func( kernel_info_t &kernel, bool openCL ) { functionalCoreSim cta( &kernel, - g_the_gpu, - g_the_gpu->getShaderCoreConfig()->warp_size + g_the_gpu(), + g_the_gpu()->getShaderCoreConfig()->warp_size ); cta.execute(cp_count,temp); @@ -2211,7 +2211,7 @@ void gpgpu_cuda_ptx_sim_main_func( kernel_info_t &kernel, bool openCL ) { char f1name[2048]; snprintf(f1name,2048,"checkpoint_files/global_mem_%d.txt", kernel.get_uid() ); - g_checkpoint->store_global_mem(g_the_gpu->get_global_memory(), f1name , "%08x"); + g_checkpoint->store_global_mem(g_the_gpu()->get_global_memory(), f1name , "%08x"); } @@ -2221,8 +2221,8 @@ void gpgpu_cuda_ptx_sim_main_func( kernel_info_t &kernel, bool openCL ) //openCL kernel simulation calls don't register the kernel so we don't register its exit if(!openCL) { - extern stream_manager *g_stream_manager; - g_stream_manager->register_finished_kernel(kernel.get_uid()); + //extern stream_manager *g_stream_manager; + g_stream_manager()->register_finished_kernel(kernel.get_uid()); } //******PRINTING******* diff --git a/src/cuda-sim/cuda_device_runtime.cc b/src/cuda-sim/cuda_device_runtime.cc index 86e8147..be8369f 100644 --- a/src/cuda-sim/cuda_device_runtime.cc +++ b/src/cuda-sim/cuda_device_runtime.cc @@ -18,6 +18,7 @@ unsigned long long g_max_total_param_size = 0; #include "cuda-sim.h" #include "ptx_ir.h" #include "../stream_manager.h" +#include "../gpgpusim_entrypoint.h" #include "cuda_device_runtime.h" #define DEV_RUNTIME_REPORT(a) \ @@ -64,7 +65,7 @@ public: std::map g_cuda_device_launch_param_map; std::list g_cuda_device_launch_op; -extern stream_manager *g_stream_manager; +//extern stream_manager *g_stream_manager(); //Handling device runtime api: //void * cudaGetParameterBufferV2(void *func, dim3 gridDimension, dim3 blockDimension, unsigned int sharedMemSize) @@ -322,7 +323,7 @@ void launch_one_device_kernel() { device_launch_operation_t &op = g_cuda_device_launch_op.front(); stream_operation stream_op = stream_operation(op.grid, g_ptx_sim_mode, op.stream); - g_stream_manager->push(stream_op); + g_stream_manager()->push(stream_op); g_cuda_device_launch_op.pop_front(); } } diff --git a/src/gpgpusim_entrypoint.cc b/src/gpgpusim_entrypoint.cc index c8770e2..de937b0 100644 --- a/src/gpgpusim_entrypoint.cc +++ b/src/gpgpusim_entrypoint.cc @@ -41,8 +41,6 @@ struct GPGPUsim_ctx* the_gpgpusim = NULL; -static void print_simulation_time(); - struct GPGPUsim_ctx* GPGPUsim_ctx_ptr(){ if(the_gpgpusim == NULL) the_gpgpusim = new GPGPUsim_ctx(); @@ -50,6 +48,16 @@ struct GPGPUsim_ctx* GPGPUsim_ctx_ptr(){ return the_gpgpusim; } +class gpgpu_sim* g_the_gpu() { + return GPGPUsim_ctx_ptr()->g_the_gpu; +} + +class stream_manager* g_stream_manager() { + return GPGPUsim_ctx_ptr()->g_stream_manager; +} + +static void print_simulation_time(); + void *gpgpu_sim_thread_sequential(void*) { // at most one kernel running at a time diff --git a/src/gpgpusim_entrypoint.h b/src/gpgpusim_entrypoint.h index e29159b..2ad0fdf 100644 --- a/src/gpgpusim_entrypoint.h +++ b/src/gpgpusim_entrypoint.h @@ -46,9 +46,15 @@ struct GPGPUsim_ctx { sg_argc = 3; sg_argv = {"", "-config","gpgpusim.config"}; + + g_the_gpu_config=NULL; + g_the_gpu=NULL; + g_stream_manager=NULL; + the_cude_device=NULL; + the_context=NULL; } - struct gpgpu_ptx_sim_arg *grid_params; + //struct gpgpu_ptx_sim_arg *grid_params; sem_t g_sim_signal_start; sem_t g_sim_signal_finish; @@ -60,19 +66,27 @@ struct GPGPUsim_ctx { class gpgpu_sim *g_the_gpu; class stream_manager *g_stream_manager; + struct _cuda_device_id *the_cude_device; + struct CUctx_st* the_context; + + int sg_argc; const char *sg_argv[3]; - pthread_mutex_t g_sim_lock; - bool g_sim_active; - bool g_sim_done; - bool break_limit; + pthread_mutex_t g_sim_lock; + bool g_sim_active; + bool g_sim_done; + bool break_limit; }; class gpgpu_sim *gpgpu_ptx_sim_init_perf(); void start_sim_thread(int api); +class gpgpu_sim* g_the_gpu(); +struct GPGPUsim_ctx* GPGPUsim_ctx_ptr(); +class stream_manager* g_stream_manager(); + int gpgpu_opencl_ptx_sim_main_perf( kernel_info_t *grid ); int gpgpu_opencl_ptx_sim_main_func( kernel_info_t *grid ); -- cgit v1.3 From 5d55ca32e2d81ee2c9187566050c5d399a142373 Mon Sep 17 00:00:00 2001 From: Mahmoud Date: Fri, 31 May 2019 13:22:29 -0400 Subject: add new params to the global stryct and fixing some bugs --- libcuda/cuda_runtime_api.cc | 6 ++++-- src/gpgpu-sim/gpu-sim.cc | 4 ++-- src/gpgpusim_entrypoint.h | 2 +- 3 files changed, 7 insertions(+), 5 deletions(-) (limited to 'libcuda/cuda_runtime_api.cc') diff --git a/libcuda/cuda_runtime_api.cc b/libcuda/cuda_runtime_api.cc index 4990edf..3d85b62 100644 --- a/libcuda/cuda_runtime_api.cc +++ b/libcuda/cuda_runtime_api.cc @@ -421,7 +421,8 @@ struct _cuda_device_id *GPGPUSim_Init() prop->maxThreadsPerMultiProcessor = the_gpu->threads_per_core(); #endif the_gpu->set_prop(prop); - the_device = new _cuda_device_id(the_gpu); + GPGPUsim_ctx_ptr()->the_cude_device = new _cuda_device_id(the_gpu); + the_device = GPGPUsim_ctx_ptr()->the_cude_device; } start_sim_thread(1); return the_device; @@ -433,7 +434,8 @@ static CUctx_st* GPGPUSim_Context() CUctx_st *the_context = GPGPUsim_ctx_ptr()->the_context; if( the_context == NULL ) { _cuda_device_id *the_gpu = GPGPUSim_Init(); - the_context = new CUctx_st(the_gpu); + GPGPUsim_ctx_ptr()->the_context = new CUctx_st(the_gpu); + the_context = GPGPUsim_ctx_ptr()->the_context; } return the_context; } diff --git a/src/gpgpu-sim/gpu-sim.cc b/src/gpgpu-sim/gpu-sim.cc index 6f19640..a557d6f 100644 --- a/src/gpgpu-sim/gpu-sim.cc +++ b/src/gpgpu-sim/gpu-sim.cc @@ -1131,7 +1131,7 @@ void gpgpu_sim::gpu_print_stat() time_t curr_time; time(&curr_time); - unsigned long long elapsed_time = MAX( curr_time - g_simulation_starttime, 1 ); + unsigned long long elapsed_time = MAX( curr_time - GPGPUsim_ctx_ptr()->g_simulation_starttime, 1 ); printf( "gpu_total_sim_rate=%u\n", (unsigned)( ( gpu_tot_sim_insn + gpu_sim_insn ) / elapsed_time ) ); //shader_print_l1_miss_stat( stdout ); @@ -1701,7 +1701,7 @@ void gpgpu_sim::cycle() time_t days, hrs, minutes, sec; time_t curr_time; time(&curr_time); - unsigned long long elapsed_time = MAX(curr_time - g_simulation_starttime, 1); + unsigned long long elapsed_time = MAX(curr_time - GPGPUsim_ctx_ptr()->g_simulation_starttime, 1); if ( (elapsed_time - last_liveness_message_time) >= m_config.liveness_message_freq && DTRACE(LIVENESS) ) { days = elapsed_time/(3600*24); hrs = elapsed_time/3600 - 24*days; diff --git a/src/gpgpusim_entrypoint.h b/src/gpgpusim_entrypoint.h index 2ad0fdf..406dd00 100644 --- a/src/gpgpusim_entrypoint.h +++ b/src/gpgpusim_entrypoint.h @@ -33,7 +33,7 @@ #include #include -extern time_t g_simulation_starttime; +//extern time_t g_simulation_starttime; struct GPGPUsim_ctx { -- cgit v1.3 From 8ad9dbc77947b4abd51ebb55ef4bbe80be01caaa Mon Sep 17 00:00:00 2001 From: Mengchi Zhang Date: Tue, 4 Jun 2019 14:26:42 -0400 Subject: Add gpgpu_context as the top level class Signed-off-by: Mengchi Zhang --- libcuda/cuda_runtime_api.cc | 182 ++++++++++++++++++++++++++++++++------------ libcuda/gpgpu_context.h | 18 +++++ 2 files changed, 150 insertions(+), 50 deletions(-) create mode 100644 libcuda/gpgpu_context.h (limited to 'libcuda/cuda_runtime_api.cc') diff --git a/libcuda/cuda_runtime_api.cc b/libcuda/cuda_runtime_api.cc index 3d85b62..c39571c 100644 --- a/libcuda/cuda_runtime_api.cc +++ b/libcuda/cuda_runtime_api.cc @@ -131,6 +131,7 @@ #if (CUDART_VERSION < 8000) #include "__cudaFatFormat.h" #endif +#include "gpgpu_context.h" #include "../src/gpgpu-sim/gpu-sim.h" #include "../src/cuda-sim/ptx_loader.h" #include "../src/cuda-sim/cuda-sim.h" @@ -313,7 +314,6 @@ struct CUctx_st { return i->second; } - std::list cuobjdumpSectionList; std::list libSectionList; //maps sm version number to set of filenames std::map > version_filename; @@ -440,6 +440,15 @@ static CUctx_st* GPGPUSim_Context() return the_context; } +static gpgpu_context* GPGPU_Context() +{ + static gpgpu_context *gpgpu_ctx = NULL; + if( gpgpu_ctx == NULL ) { + gpgpu_ctx = new gpgpu_context(); + } + return gpgpu_ctx; +} + void ptxinfo_addinfo() { if(!get_ptxinfo_kname()){ @@ -2096,7 +2105,7 @@ static int get_app_cuda_version() { * It is also responsible for extracting the libraries linked to the binary if the option is * enabled * */ -void extract_code_using_cuobjdump(std::list &cuobjdumpSectionList){ +void gpgpu_context::extract_code_using_cuobjdump(){ CUctx_st *context = GPGPUSim_Context(); unsigned forced_max_capability = context->get_device()->get_gpgpu()->get_config().get_forced_max_capability(); @@ -2266,7 +2275,7 @@ void printSectionList(std::list sl) { } //! Remove unecessary sm versions from the section list -std::list pruneSectionList(std::list cuobjdumpSectionList, CUctx_st *context) { +std::list gpgpu_context::pruneSectionList(CUctx_st *context) { unsigned forced_max_capability = context->get_device()->get_gpgpu()->get_config().get_forced_max_capability(); //For ptxplus, force the max capability to 19 if it's higher or unspecified(0) @@ -2319,7 +2328,7 @@ std::list pruneSectionList(std::list cuobj } //! Merge all PTX sections that have a specific identifier into one file -std::list mergeMatchingSections(std::list cuobjdumpSectionList, std::string identifier){ +std::list gpgpu_context::mergeMatchingSections(std::string identifier){ const char *ptxcode = ""; std::list::iterator old_iter; cuobjdumpPTXSection* old_ptxsection = NULL; @@ -2362,7 +2371,7 @@ std::list mergeMatchingSections(std::list } //! Merge any PTX sections with matching identifiers -std::list mergeSections(std::list cuobjdumpSectionList){ +std::list gpgpu_context::mergeSections(){ std::vector identifier; cuobjdumpPTXSection* ptxsection; @@ -2384,7 +2393,7 @@ std::list mergeSections(std::list cuobjdum for ( std::vector::iterator iter = identifier.begin(); iter != identifier.end(); iter++) { - cuobjdumpSectionList = mergeMatchingSections(cuobjdumpSectionList, *iter); + cuobjdumpSectionList = mergeMatchingSections(*iter); } return cuobjdumpSectionList; @@ -2409,7 +2418,7 @@ cuobjdumpELFSection* findELFSectionInList(std::list sectionli } //! Find an ELF section in all the known lists -cuobjdumpELFSection* findELFSection(const std::string identifier, std::list cuobjdumpSectionList, std::list &libSectionList){ +cuobjdumpELFSection* gpgpu_context::findELFSection(const std::string identifier, std::list &libSectionList){ cuobjdumpELFSection* sec = findELFSectionInList(cuobjdumpSectionList, identifier); if (sec!=NULL)return sec; sec = findELFSectionInList(libSectionList, identifier); @@ -2420,7 +2429,7 @@ cuobjdumpELFSection* findELFSection(const std::string identifier, std::list sectionlist, const std::string identifier){ +cuobjdumpPTXSection* findPTXSectionInList(std::list §ionlist, const std::string identifier){ std::list::iterator iter; for ( iter = sectionlist.begin(); iter != sectionlist.end(); @@ -2444,7 +2453,7 @@ cuobjdumpPTXSection* findPTXSectionInList(std::list sectionli } //! Find an PTX section in all the known lists -cuobjdumpPTXSection* findPTXSection(const std::string identifier, std::list cuobjdumpSectionList, std::list &libSectionList){ +cuobjdumpPTXSection* gpgpu_context::findPTXSection(const std::string identifier, std::list &libSectionList){ cuobjdumpPTXSection* sec = findPTXSectionInList(cuobjdumpSectionList, identifier); if (sec!=NULL)return sec; sec = findPTXSectionInList(libSectionList, identifier); @@ -2457,13 +2466,13 @@ cuobjdumpPTXSection* findPTXSection(const std::string identifier, std::list &cuobjdumpSectionList){ +void gpgpu_context::cuobjdumpInit(){ CUctx_st *context = GPGPUSim_Context(); - extract_code_using_cuobjdump(cuobjdumpSectionList); //extract all the output of cuobjdump to _cuobjdump_*.* + extract_code_using_cuobjdump(); //extract all the output of cuobjdump to _cuobjdump_*.* const char* pre_load = getenv("CUOBJDUMP_SIM_FILE"); if (pre_load ==NULL || strlen(pre_load)==0){ - cuobjdumpSectionList = pruneSectionList(cuobjdumpSectionList, context); - cuobjdumpSectionList = mergeSections(cuobjdumpSectionList); + cuobjdumpSectionList = pruneSectionList(context); + cuobjdumpSectionList = mergeSections(); } } @@ -2474,7 +2483,7 @@ void cuobjdumpRegisterFatBinary(unsigned int handle, const char* filename, CUctx } //! Either submit PTX for simulation or convert SASS to PTXPlus and submit it -void cuobjdumpParseBinary(unsigned int handle){ +void gpgpu_context::cuobjdumpParseBinary(unsigned int handle){ CUctx_st *context = GPGPUSim_Context(); if(context->fatbin_registered[handle]) return; @@ -2515,8 +2524,8 @@ void cuobjdumpParseBinary(unsigned int handle){ #endif unsigned max_capability = 0; - for ( std::list::iterator iter = context->cuobjdumpSectionList.begin(); - iter != context->cuobjdumpSectionList.end(); + for ( std::list::iterator iter = cuobjdumpSectionList.begin(); + iter != cuobjdumpSectionList.end(); iter++){ unsigned capability = (*iter)->getArch(); if (capability > max_capability) max_capability = capability; @@ -2527,7 +2536,7 @@ void cuobjdumpParseBinary(unsigned int handle){ cuobjdumpPTXSection* ptx = NULL; const char* pre_load = getenv("CUOBJDUMP_SIM_FILE"); if(pre_load==NULL || strlen(pre_load)==0) - ptx = findPTXSection(fname, context->cuobjdumpSectionList, context->libSectionList); + ptx = findPTXSection(fname, context->libSectionList); char *ptxcode; const char *override_ptx_name = getenv("PTX_SIM_KERNELFILE"); if (override_ptx_name == NULL or getenv("PTX_SIM_USE_PTX_FILE") == NULL or strlen(getenv("PTX_SIM_USE_PTX_FILE"))==0) { @@ -2537,7 +2546,7 @@ void cuobjdumpParseBinary(unsigned int handle){ ptxcode = readfile(override_ptx_name); } if(context->get_device()->get_gpgpu()->get_config().convert_to_ptxplus() ) { - cuobjdumpELFSection* elfsection = findELFSection(ptx->getIdentifier(), context->cuobjdumpSectionList, context->libSectionList); + cuobjdumpELFSection* elfsection = findELFSection(ptx->getIdentifier(), context->libSectionList); assert (elfsection!= NULL); char *ptxplus_str = gpgpu_ptx_sim_convert_ptx_and_sass_to_ptxplus( ptx->getPTXfilename(), @@ -2561,9 +2570,16 @@ void cuobjdumpParseBinary(unsigned int handle){ //TODO: Remove temporarily files as per configurations } +} -void** CUDARTAPI __cudaRegisterFatBinary( void *fatCubin ) +void** cudaRegisterFatBinary( void *fatCubin, gpgpu_context* gpgpu_ctx = NULL) { + gpgpu_context *ctx; + if (gpgpu_ctx){ + ctx = gpgpu_ctx; + } else { + ctx = GPGPU_Context(); + } if(g_debug_execution >= 3){ announce_call(__my_func__); } @@ -2625,7 +2641,7 @@ void** CUDARTAPI __cudaRegisterFatBinary( void *fatCubin ) * then for next calls, only returns the appropriate number */ assert(fat_cubin_handle >= 1); - if (fat_cubin_handle==1) cuobjdumpInit(context->cuobjdumpSectionList); + if (fat_cubin_handle==1) ctx->cuobjdumpInit(); cuobjdumpRegisterFatBinary(fat_cubin_handle, filename, context); return (void**)fat_cubin_handle; @@ -2696,31 +2712,7 @@ void** CUDARTAPI __cudaRegisterFatBinary( void *fatCubin ) #endif } -void __cudaUnregisterFatBinary(void **fatCubinHandle) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - ; -} - -cudaError_t cudaDeviceReset ( void ) { - // Should reset the simulated GPU - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - return g_last_cudaError = cudaSuccess; -} -cudaError_t CUDARTAPI cudaDeviceSynchronize(void){ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - //Blocks until the device has completed all preceding requested tasks - synchronize(); - return g_last_cudaError = cudaSuccess; -} - -void CUDARTAPI __cudaRegisterFunction( +void cudaRegisterFunction( void **fatCubinHandle, const char *hostFun, char *deviceFun, @@ -2729,9 +2721,16 @@ void CUDARTAPI __cudaRegisterFunction( uint3 *tid, uint3 *bid, dim3 *bDim, - dim3 *gDim + dim3 *gDim, + gpgpu_context *gpgpu_ctx = NULL ) { + gpgpu_context *ctx; + if (gpgpu_ctx){ + ctx = gpgpu_ctx; + } else { + ctx = GPGPU_Context(); + } if(g_debug_execution >= 3){ announce_call(__my_func__); } @@ -2740,11 +2739,11 @@ void CUDARTAPI __cudaRegisterFunction( printf("GPGPU-Sim PTX: __cudaRegisterFunction %s : hostFun 0x%p, fat_cubin_handle = %u\n", deviceFun, hostFun, fat_cubin_handle); if(context->get_device()->get_gpgpu()->get_config().use_cuobjdump()) - cuobjdumpParseBinary(fat_cubin_handle); + ctx->cuobjdumpParseBinary(fat_cubin_handle); context->register_function( fat_cubin_handle, hostFun, deviceFun ); } -extern void __cudaRegisterVar( +void cudaRegisterVar( void **fatCubinHandle, char *hostVar, //pointer to...something char *deviceAddress, //name of variable @@ -2752,15 +2751,22 @@ extern void __cudaRegisterVar( int ext, int size, int constant, - int global ) + int global, + gpgpu_context *gpgpu_ctx = NULL) { + gpgpu_context *ctx; + if (gpgpu_ctx){ + ctx = gpgpu_ctx; + } else { + ctx = GPGPU_Context(); + } if(g_debug_execution >= 3){ announce_call(__my_func__); } printf("GPGPU-Sim PTX: __cudaRegisterVar: hostVar = %p; deviceAddress = %s; deviceName = %s\n", hostVar, deviceAddress, deviceName); printf("GPGPU-Sim PTX: __cudaRegisterVar: Registering const memory space of %d bytes\n", size); if(GPGPUSim_Context()->get_device()->get_gpgpu()->get_config().use_cuobjdump()) - cuobjdumpParseBinary((unsigned)(unsigned long long)fatCubinHandle); + ctx->cuobjdumpParseBinary((unsigned)(unsigned long long)fatCubinHandle); fflush(stdout); if ( constant && !global && !ext ) { gpgpu_ptx_sim_register_const_variable(hostVar,deviceName,size); @@ -2769,6 +2775,82 @@ extern void __cudaRegisterVar( } else cuda_not_implemented(__my_func__,__LINE__); } +extern "C" { + +void** CUDARTAPI __cudaRegisterFatBinary( void *fatCubin ) +{ + return cudaRegisterFatBinary(fatCubin); +} + +void CUDARTAPI __cudaRegisterFunction( + void **fatCubinHandle, + const char *hostFun, + char *deviceFun, + const char *deviceName, + int thread_limit, + uint3 *tid, + uint3 *bid, + dim3 *bDim, + dim3 *gDim +) { + cudaRegisterFunction( + fatCubinHandle, + hostFun, + deviceFun, + deviceName, + thread_limit, + tid, + bid, + bDim, + gDim + ); + +} + +extern void __cudaRegisterVar( + void **fatCubinHandle, + char *hostVar, //pointer to...something + char *deviceAddress, //name of variable + const char *deviceName, //name of variable (same as above) + int ext, + int size, + int constant, + int global ) +{ + cudaRegisterVar( + fatCubinHandle, + hostVar, + deviceAddress, + deviceName, + ext, + size, + constant, + global ); +} +void __cudaUnregisterFatBinary(void **fatCubinHandle) +{ + if(g_debug_execution >= 3){ + announce_call(__my_func__); + } + ; +} + +cudaError_t cudaDeviceReset ( void ) { + // Should reset the simulated GPU + if(g_debug_execution >= 3){ + announce_call(__my_func__); + } + return g_last_cudaError = cudaSuccess; +} +cudaError_t CUDARTAPI cudaDeviceSynchronize(void){ + if(g_debug_execution >= 3){ + announce_call(__my_func__); + } + //Blocks until the device has completed all preceding requested tasks + synchronize(); + return g_last_cudaError = cudaSuccess; +} + void __cudaRegisterShared( void **fatCubinHandle, diff --git a/libcuda/gpgpu_context.h b/libcuda/gpgpu_context.h new file mode 100644 index 0000000..d6e564b --- /dev/null +++ b/libcuda/gpgpu_context.h @@ -0,0 +1,18 @@ +#include + +class cuobjdumpSection; +class cuobjdumpELFSection; +class cuobjdumpPTXSection; + +class gpgpu_context { + public: + std::list cuobjdumpSectionList; + void cuobjdumpInit(); + void cuobjdumpParseBinary(unsigned int handle); + void extract_code_using_cuobjdump(); + std::list pruneSectionList(CUctx_st *context); + std::list mergeMatchingSections(std::string identifier); + std::list mergeSections(); + cuobjdumpELFSection* findELFSection(const std::string identifier, std::list &libSectionList); + cuobjdumpPTXSection* findPTXSection(const std::string identifier, std::list &libSectionList); +}; -- cgit v1.3 From 37a709bb69fcd9f2b0fe53a189e92e548164cc4b Mon Sep 17 00:00:00 2001 From: Mahmoud Date: Wed, 5 Jun 2019 19:18:29 -0400 Subject: adding new cuda 9 APIs to run the deepbench workloads --- Makefile | 2 ++ libcuda/cuda_runtime_api.cc | 70 +++++++++++++++++++++++++++++++++++++++++- linux-so-version.txt | 4 +++ src/abstract_hardware_model.cc | 22 +++++++++++++ src/abstract_hardware_model.h | 1 + src/gpgpu-sim/gpu-sim.cc | 10 ++++++ src/gpgpu-sim/gpu-sim.h | 2 ++ 7 files changed, 110 insertions(+), 1 deletion(-) (limited to 'libcuda/cuda_runtime_api.cc') diff --git a/Makefile b/Makefile index 2d0466e..e1e9aaa 100644 --- a/Makefile +++ b/Makefile @@ -159,10 +159,12 @@ $(SIM_LIB_DIR)/libcudart.so: makedirs $(LIBS) cudalib if [ ! -f $(SIM_LIB_DIR)/libcudart.so.5.5 ]; then ln -s libcudart.so $(SIM_LIB_DIR)/libcudart.so.5.5; fi if [ ! -f $(SIM_LIB_DIR)/libcudart.so.6.0 ]; then ln -s libcudart.so $(SIM_LIB_DIR)/libcudart.so.6.0; fi if [ ! -f $(SIM_LIB_DIR)/libcudart.so.6.5 ]; then ln -s libcudart.so $(SIM_LIB_DIR)/libcudart.so.6.5; fi + if [ ! -f $(SIM_LIB_DIR)/libcudart.so.7.0 ]; then ln -s libcudart.so $(SIM_LIB_DIR)/libcudart.so.7.0; fi if [ ! -f $(SIM_LIB_DIR)/libcudart.so.7.5 ]; then ln -s libcudart.so $(SIM_LIB_DIR)/libcudart.so.7.5; fi if [ ! -f $(SIM_LIB_DIR)/libcudart.so.8.0 ]; then ln -s libcudart.so $(SIM_LIB_DIR)/libcudart.so.8.0; fi if [ ! -f $(SIM_LIB_DIR)/libcudart.so.9.0 ]; then ln -s libcudart.so $(SIM_LIB_DIR)/libcudart.so.9.0; fi if [ ! -f $(SIM_LIB_DIR)/libcudart.so.9.1 ]; then ln -s libcudart.so $(SIM_LIB_DIR)/libcudart.so.9.1; fi + if [ ! -f $(SIM_LIB_DIR)/libcudart.so.9.2 ]; then ln -s libcudart.so $(SIM_LIB_DIR)/libcudart.so.9.2; fi $(SIM_LIB_DIR)/libcudart.dylib: makedirs $(LIBS) cudalib g++ -dynamiclib -Wl,-headerpad_max_install_names,-undefined,dynamic_lookup,-compatibility_version,1.1,-current_version,1.1\ diff --git a/libcuda/cuda_runtime_api.cc b/libcuda/cuda_runtime_api.cc index 3a9d613..2cc84aa 100644 --- a/libcuda/cuda_runtime_api.cc +++ b/libcuda/cuda_runtime_api.cc @@ -974,6 +974,30 @@ __host__ cudaError_t CUDARTAPI cudaGetDeviceProperties(struct cudaDeviceProp *pr } } +#if (CUDART_VERSION >= 8000) +cudaError_t CUDARTAPI cudaOccupancyMaxActiveBlocksPerMultiprocessorWithFlags(int* numBlocks, const char *hostFunc, int blockSize, size_t dynamicSMemSize, unsigned int flags) +{ + printf("GPGPU-Sim PTX: cudaOccupancyMaxActiveBlocksPerMultiprocessorWithFlags %p\n", hostFunc); + CUctx_st *context = GPGPUSim_Context(); + function_info *entry = context->get_kernel(hostFunc); + printf("Calculate Maxium Active Block with function ptr=%p, blockSize=%d, SMemSize=%d\n", hostFunc, blockSize, dynamicSMemSize); + if (flags == cudaOccupancyDefault) { + //create kernel_info based on entry + dim3 gridDim(context->get_device()->get_gpgpu()->max_cta_per_core() + * context->get_device()->get_gpgpu()->get_config().num_shader()); + dim3 blockDim(blockSize); + kernel_info_t result(gridDim, blockDim, entry); + *numBlocks = context->get_device()->get_gpgpu()->get_max_cta(result); + printf("Maximum size is %d with gridDim %d and blockDim %d\n", *numBlocks, gridDim.x, blockDim.x); + return g_last_cudaError = cudaSuccess; + } else { + cuda_not_implemented(__my_func__,__LINE__); + return g_last_cudaError = cudaErrorUnknown; + } +} + +#endif + #if (CUDART_VERSION > 5000) __host__ cudaError_t CUDARTAPI cudaDeviceGetAttribute(int *value, enum cudaDeviceAttr attr, int device) { @@ -3169,6 +3193,23 @@ __host__ cudaError_t CUDARTAPI cudaDeviceSetLimit(enum cudaLimit limit, size_t v return g_last_cudaError = cudaSuccess; } +#if CUDART_VERSION >= 9000 +__host__ cudaError_t cudaFuncSetAttribute ( const void* func, enum cudaFuncAttribute attr, int value ) { + + //ignore this Attribute for now, and the default is that carveout = cudaSharedmemCarveoutDefault; // (-1) + return g_last_cudaError = cudaSuccess; +} + +#if CUDART_VERSION >= 9020 +__host__ __device__ unsigned __cudaPushCallConfiguration(dim3 gridDim, dim3 blockDim, + size_t sharedMem = 0, + void *stream = 0) { + + return 1; +} +#endif + +#endif #endif @@ -4071,6 +4112,33 @@ CUresult CUDAAPI cuMemHostRegister(void *p, size_t bytesize, unsigned int Flags) printf("WARNING: this function has not been implemented yet."); return CUDA_SUCCESS; } +__host__ cudaError_t cudaHostRegister(void* ptr, size_t size, unsigned int flags) +{ + if(g_debug_execution >= 3){ + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return g_last_cudaError = cudaSuccess; +} + +__host__ cudaError_t cudaProfilerStart ( ) +{ + if(g_debug_execution >= 3){ + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return g_last_cudaError = cudaSuccess; +} + +__host__ cudaError_t cudaProfilerStop ( ) +{ + if(g_debug_execution >= 3){ + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return g_last_cudaError = cudaSuccess; +} + #endif #if CUDART_VERSION >= 4000 @@ -4952,7 +5020,7 @@ CUresult CUDAAPI cuOccupancyMaxActiveBlocksPerMultiprocessorWithFlags(int *numBl printf("WARNING: this function has not been implemented yet."); return CUDA_SUCCESS; } - + CUresult CUDAAPI cuOccupancyMaxPotentialBlockSize(int *minGridSize, int *blockSize, CUfunction func, CUoccupancyB2DSize blockSizeToDynamicSMemSize, size_t dynamicSMemSize, int blockSizeLimit) { if(g_debug_execution >= 3){ diff --git a/linux-so-version.txt b/linux-so-version.txt index c6e03e9..a7c2d3c 100644 --- a/linux-so-version.txt +++ b/linux-so-version.txt @@ -1,4 +1,8 @@ +libcudart.so.9.0{ +}; libcudart.so.9.1{ }; +libcudart.so.9.2{ +}; libcuda.so.1{ }; diff --git a/src/abstract_hardware_model.cc b/src/abstract_hardware_model.cc index cebdb25..023f51b 100644 --- a/src/abstract_hardware_model.cc +++ b/src/abstract_hardware_model.cc @@ -691,6 +691,28 @@ unsigned g_kernel_launch_latency; unsigned kernel_info_t::m_next_uid = 1; +kernel_info_t::kernel_info_t( dim3 gridDim, dim3 blockDim, class function_info *entry) +{ + m_kernel_entry=entry; + m_grid_dim=gridDim; + m_block_dim=blockDim; + m_next_cta.x=0; + m_next_cta.y=0; + m_next_cta.z=0; + m_next_tid=m_next_cta; + m_num_cores_running=0; + m_uid = m_next_uid++; + m_param_mem = new memory_space_impl<8192>("param",64*1024); + + //Jin: parent and child kernel management for CDP + m_parent_kernel = NULL; + + //Jin: launch latency management + m_launch_latency = g_kernel_launch_latency; + + volta_cache_config_set=false; +} + /*A snapshot of the texture mappings needs to be stored in the kernel's info as kernels should use the texture bindings seen at the time of launch and textures can be bound/unbound asynchronously with respect to streams. */ diff --git a/src/abstract_hardware_model.h b/src/abstract_hardware_model.h index 22ef509..64bbaa2 100644 --- a/src/abstract_hardware_model.h +++ b/src/abstract_hardware_model.h @@ -212,6 +212,7 @@ public: // m_num_cores_running=0; // m_param_mem=NULL; // } + kernel_info_t( dim3 gridDim, dim3 blockDim, class function_info *entry); kernel_info_t( dim3 gridDim, dim3 blockDim, class function_info *entry, std::map nameToCudaArray, std::map nameToTextureInfo); ~kernel_info_t(); diff --git a/src/gpgpu-sim/gpu-sim.cc b/src/gpgpu-sim/gpu-sim.cc index c1ba934..6de5845 100644 --- a/src/gpgpu-sim/gpu-sim.cc +++ b/src/gpgpu-sim/gpu-sim.cc @@ -806,6 +806,16 @@ int gpgpu_sim::shader_clock() const return m_config.core_freq/1000; } +int gpgpu_sim::max_cta_per_core() const +{ + return m_shader_config->max_cta_per_core; +} + +int gpgpu_sim::get_max_cta( const kernel_info_t &k ) const +{ + return m_shader_config->max_cta(k); +} + void gpgpu_sim::set_prop( cudaDeviceProp *prop ) { m_cuda_properties = prop; diff --git a/src/gpgpu-sim/gpu-sim.h b/src/gpgpu-sim/gpu-sim.h index c8dad89..7336cac 100644 --- a/src/gpgpu-sim/gpu-sim.h +++ b/src/gpgpu-sim/gpu-sim.h @@ -455,6 +455,8 @@ public: int num_registers_per_block() const; int wrp_size() const; int shader_clock() const; + int max_cta_per_core() const; + int get_max_cta( const kernel_info_t &k ) const; const struct cudaDeviceProp *get_prop() const; enum divergence_support_t simd_model() const; -- cgit v1.3 From d553124832aca461dda4dd7d503748d22b44bbe2 Mon Sep 17 00:00:00 2001 From: Mengchi Zhang Date: Wed, 5 Jun 2019 23:28:12 -0400 Subject: Add cuda_api_object and move some var Signed-off-by: Mengchi Zhang --- libcuda/cuda_api_object.h | 11 +++++++++++ libcuda/cuda_runtime_api.cc | 30 ++++++++++++++---------------- libcuda/gpgpu_context.h | 19 +++++++++++++++++-- 3 files changed, 42 insertions(+), 18 deletions(-) create mode 100644 libcuda/cuda_api_object.h (limited to 'libcuda/cuda_runtime_api.cc') diff --git a/libcuda/cuda_api_object.h b/libcuda/cuda_api_object.h new file mode 100644 index 0000000..86ffa98 --- /dev/null +++ b/libcuda/cuda_api_object.h @@ -0,0 +1,11 @@ +#ifndef __cuda_api_object_h__ +#define __cuda_api_object_h__ +class cuobjdumpSection; + +class cuda_runtime_api { + public: + // global list + std::list libSectionList; + // member function list +}; +#endif /* __cuda_api_object_h__ */ diff --git a/libcuda/cuda_runtime_api.cc b/libcuda/cuda_runtime_api.cc index c39571c..4221634 100644 --- a/libcuda/cuda_runtime_api.cc +++ b/libcuda/cuda_runtime_api.cc @@ -132,6 +132,7 @@ #include "__cudaFatFormat.h" #endif #include "gpgpu_context.h" +#include "cuda_api_object.h" #include "../src/gpgpu-sim/gpu-sim.h" #include "../src/cuda-sim/ptx_loader.h" #include "../src/cuda-sim/cuda-sim.h" @@ -314,9 +315,6 @@ struct CUctx_st { return i->second; } - std::list libSectionList; - //maps sm version number to set of filenames - std::map > version_filename; std::list g_cuda_launch_stack; std::mapfatbin_registered; std::map fatbinmap; @@ -2005,7 +2003,7 @@ char* get_app_binary_name(std::string abs_path){ } //extracts all ptx files from binary and dumps into prog_name.unique_no.sm_<>.ptx files -void extract_ptx_files_using_cuobjdump(CUctx_st *context){ +void gpgpu_context::extract_ptx_files_using_cuobjdump(CUctx_st *context){ extern bool g_cdp_enabled; char command[1000]; char *pytorch_bin = getenv("PYTORCH_BIN"); @@ -2066,10 +2064,10 @@ void extract_ptx_files_using_cuobjdump(CUctx_st *context){ } std::string vstr = line.substr(pos1+3,pos2-pos1-3); int version = atoi(vstr.c_str()); - if (context->version_filename.find(version)==context->version_filename.end()){ - context->version_filename[version] = std::set(); + if (version_filename.find(version)==version_filename.end()){ + version_filename[version] = std::set(); } - context->version_filename[version].insert(line); + version_filename[version].insert(line); } } @@ -2229,7 +2227,7 @@ void gpgpu_context::extract_code_using_cuobjdump(){ fclose(cuobjdump_in); std::getline(libsf, line); } - context->libSectionList = cuobjdumpSectionList; + api->libSectionList = cuobjdumpSectionList; //Restore the original section list cuobjdumpSectionList = tmpsl; @@ -2418,10 +2416,10 @@ cuobjdumpELFSection* findELFSectionInList(std::list sectionli } //! Find an ELF section in all the known lists -cuobjdumpELFSection* gpgpu_context::findELFSection(const std::string identifier, std::list &libSectionList){ +cuobjdumpELFSection* gpgpu_context::findELFSection(const std::string identifier){ cuobjdumpELFSection* sec = findELFSectionInList(cuobjdumpSectionList, identifier); if (sec!=NULL)return sec; - sec = findELFSectionInList(libSectionList, identifier); + sec = findELFSectionInList(api->libSectionList, identifier); if (sec!=NULL)return sec; std::cout << "Could not find " << identifier << std::endl; assert(0 && "Could not find the required ELF section"); @@ -2453,10 +2451,10 @@ cuobjdumpPTXSection* findPTXSectionInList(std::list §ionl } //! Find an PTX section in all the known lists -cuobjdumpPTXSection* gpgpu_context::findPTXSection(const std::string identifier, std::list &libSectionList){ +cuobjdumpPTXSection* gpgpu_context::findPTXSection(const std::string identifier){ cuobjdumpPTXSection* sec = findPTXSectionInList(cuobjdumpSectionList, identifier); if (sec!=NULL)return sec; - sec = findPTXSectionInList(libSectionList, identifier); + sec = findPTXSectionInList(api->libSectionList, identifier); if (sec!=NULL)return sec; std::cout << "Could not find " << identifier << std::endl; assert(0 && "Could not find the required PTX section"); @@ -2500,7 +2498,7 @@ void gpgpu_context::cuobjdumpParseBinary(unsigned int handle){ #if (CUDART_VERSION >= 6000) //loops through all ptx files from smallest sm version to largest std::map >::iterator itr_m; - for (itr_m = context->version_filename.begin(); itr_m!=context->version_filename.end(); itr_m++){ + for (itr_m = version_filename.begin(); itr_m!=version_filename.end(); itr_m++){ std::set::iterator itr_s; for (itr_s = itr_m->second.begin(); itr_s!=itr_m->second.end(); itr_s++){ std::string ptx_filename = *itr_s; @@ -2512,7 +2510,7 @@ void gpgpu_context::cuobjdumpParseBinary(unsigned int handle){ context->add_binary(symtab, handle); load_static_globals(symtab,STATIC_ALLOC_LIMIT,0xFFFFFFFF,context->get_device()->get_gpgpu()); load_constants(symtab,STATIC_ALLOC_LIMIT,context->get_device()->get_gpgpu()); - for (itr_m = context->version_filename.begin(); itr_m!=context->version_filename.end(); itr_m++){ + for (itr_m = version_filename.begin(); itr_m!=version_filename.end(); itr_m++){ std::set::iterator itr_s; for (itr_s = itr_m->second.begin(); itr_s!=itr_m->second.end(); itr_s++){ std::string ptx_filename = *itr_s; @@ -2536,7 +2534,7 @@ void gpgpu_context::cuobjdumpParseBinary(unsigned int handle){ cuobjdumpPTXSection* ptx = NULL; const char* pre_load = getenv("CUOBJDUMP_SIM_FILE"); if(pre_load==NULL || strlen(pre_load)==0) - ptx = findPTXSection(fname, context->libSectionList); + ptx = findPTXSection(fname); char *ptxcode; const char *override_ptx_name = getenv("PTX_SIM_KERNELFILE"); if (override_ptx_name == NULL or getenv("PTX_SIM_USE_PTX_FILE") == NULL or strlen(getenv("PTX_SIM_USE_PTX_FILE"))==0) { @@ -2546,7 +2544,7 @@ void gpgpu_context::cuobjdumpParseBinary(unsigned int handle){ ptxcode = readfile(override_ptx_name); } if(context->get_device()->get_gpgpu()->get_config().convert_to_ptxplus() ) { - cuobjdumpELFSection* elfsection = findELFSection(ptx->getIdentifier(), context->libSectionList); + cuobjdumpELFSection* elfsection = findELFSection(ptx->getIdentifier()); assert (elfsection!= NULL); char *ptxplus_str = gpgpu_ptx_sim_convert_ptx_and_sass_to_ptxplus( ptx->getPTXfilename(), diff --git a/libcuda/gpgpu_context.h b/libcuda/gpgpu_context.h index d6e564b..576ec67 100644 --- a/libcuda/gpgpu_context.h +++ b/libcuda/gpgpu_context.h @@ -1,4 +1,9 @@ +#ifndef __gpgpu_context_h__ +#define __gpgpu_context_h__ #include +#include +#include +#include "cuda_api_object.h" class cuobjdumpSection; class cuobjdumpELFSection; @@ -6,13 +11,23 @@ class cuobjdumpPTXSection; class gpgpu_context { public: + gpgpu_context() { + api = new cuda_runtime_api(); + } + // global list std::list cuobjdumpSectionList; + //maps sm version number to set of filenames + std::map > version_filename; + cuda_runtime_api* api; + // member function list void cuobjdumpInit(); void cuobjdumpParseBinary(unsigned int handle); void extract_code_using_cuobjdump(); std::list pruneSectionList(CUctx_st *context); std::list mergeMatchingSections(std::string identifier); std::list mergeSections(); - cuobjdumpELFSection* findELFSection(const std::string identifier, std::list &libSectionList); - cuobjdumpPTXSection* findPTXSection(const std::string identifier, std::list &libSectionList); + cuobjdumpELFSection* findELFSection(const std::string identifier); + cuobjdumpPTXSection* findPTXSection(const std::string identifier); + void extract_ptx_files_using_cuobjdump(CUctx_st *context); }; +#endif /* __gpgpu_context_h__ */ -- cgit v1.3 From bd5bbc6b1d56436dbcc0cfd84e96c2d514ab4ccc Mon Sep 17 00:00:00 2001 From: Mengchi Zhang Date: Thu, 6 Jun 2019 01:07:21 -0400 Subject: Move g_cuda_launch_stack Signed-off-by: Mengchi Zhang --- libcuda/cuda_api_object.h | 3 + libcuda/cuda_runtime_api.cc | 910 +++++++++++++++++++++++--------------------- 2 files changed, 475 insertions(+), 438 deletions(-) (limited to 'libcuda/cuda_runtime_api.cc') diff --git a/libcuda/cuda_api_object.h b/libcuda/cuda_api_object.h index 86ffa98..2001f91 100644 --- a/libcuda/cuda_api_object.h +++ b/libcuda/cuda_api_object.h @@ -1,11 +1,14 @@ #ifndef __cuda_api_object_h__ #define __cuda_api_object_h__ class cuobjdumpSection; +class kernel_config; + class cuda_runtime_api { public: // global list std::list libSectionList; + std::list g_cuda_launch_stack; // member function list }; #endif /* __cuda_api_object_h__ */ diff --git a/libcuda/cuda_runtime_api.cc b/libcuda/cuda_runtime_api.cc index 4221634..ea96570 100644 --- a/libcuda/cuda_runtime_api.cc +++ b/libcuda/cuda_runtime_api.cc @@ -237,8 +237,6 @@ private: struct _cuda_device_id *m_next; }; -class kernel_config; - #ifndef OPENGL_SUPPORT typedef unsigned long GLuint; #endif @@ -315,7 +313,6 @@ struct CUctx_st { return i->second; } - std::list g_cuda_launch_stack; std::mapfatbin_registered; std::map fatbinmap; std::map g_mallocPtr_Size; @@ -531,51 +528,448 @@ enum cuobjdumpSectionType { }; -// sectiontype: 0 for ptx, 1 for elf -void addCuobjdumpSection(int sectiontype, std::list &cuobjdumpSectionList){ - if (sectiontype) - cuobjdumpSectionList.push_front(new cuobjdumpELFSection()); - else - cuobjdumpSectionList.push_front(new cuobjdumpPTXSection()); - printf("## Adding new section %s\n", sectiontype?"ELF":"PTX"); +// sectiontype: 0 for ptx, 1 for elf +void addCuobjdumpSection(int sectiontype, std::list &cuobjdumpSectionList){ + if (sectiontype) + cuobjdumpSectionList.push_front(new cuobjdumpELFSection()); + else + cuobjdumpSectionList.push_front(new cuobjdumpPTXSection()); + printf("## Adding new section %s\n", sectiontype?"ELF":"PTX"); +} + +void setCuobjdumparch(const char* arch, std::list &cuobjdumpSectionList){ + unsigned archnum; + sscanf(arch, "sm_%u", &archnum); + assert (archnum && "cannot have sm_0"); + printf("Adding arch: %s\n", arch); + cuobjdumpSectionList.front()->setArch(archnum); +} + +void setCuobjdumpidentifier(const char* identifier, std::list &cuobjdumpSectionList){ + printf("Adding identifier: %s\n", identifier); + cuobjdumpSectionList.front()->setIdentifier(identifier); +} + +void setCuobjdumpptxfilename(const char* filename, std::list &cuobjdumpSectionList){ + printf("Adding ptx filename: %s\n", filename); + cuobjdumpSection* x = cuobjdumpSectionList.front(); + if (dynamic_cast(x) == NULL){ + assert (0 && "You shouldn't be trying to add a ptxfilename to an elf section"); + } + (dynamic_cast(x))->setPTXfilename(filename); +} + +void setCuobjdumpelffilename(const char* filename, std::list &cuobjdumpSectionList){ + if (dynamic_cast(cuobjdumpSectionList.front()) == NULL){ + assert (0 && "You shouldn't be trying to add a elffilename to an ptx section"); + } + (dynamic_cast(cuobjdumpSectionList.front()))->setELFfilename(filename); +} + +void setCuobjdumpsassfilename(const char* filename, std::list &cuobjdumpSectionList){ + if (dynamic_cast(cuobjdumpSectionList.front()) == NULL){ + assert (0 && "You shouldn't be trying to add a sassfilename to an ptx section"); + } + (dynamic_cast(cuobjdumpSectionList.front()))->setSASSfilename(filename); +} + +//! Return the executable file of the process containing the PTX/SASS code +//! +//! This Function returns the executable file ran by the process. This +//! executable is supposed to contain the PTX/SASS code. It provides workaround +//! for processes running on valgrind by dereferencing /proc//exe within the +//! GPGPU-Sim process before calling cuobjdump to extract PTX/SASS. This is +//! needed because valgrind uses x86 emulation to detect memory leak. Other +//! processes (e.g. cuobjdump) reading /proc//exe will see the emulator +//! executable instead of the application binary. +//! +std::string get_app_binary(){ + char self_exe_path[1025]; +#ifdef __APPLE__ + uint32_t size = sizeof(self_exe_path); + if( _NSGetExecutablePath(self_exe_path,&size) != 0 ) { + printf("GPGPU-Sim ** ERROR: _NSGetExecutablePath input buffer too small\n"); + exit(1); + } +#else + std::stringstream exec_link; + exec_link << "/proc/self/exe"; + + ssize_t path_length = readlink(exec_link.str().c_str(), self_exe_path, 1024); + assert(path_length != -1); + self_exe_path[path_length] = '\0'; +#endif + + printf("self exe links to: %s\n", self_exe_path); + return self_exe_path; +} + +//above func gives abs path whereas this give just the name of application. +char* get_app_binary_name(std::string abs_path){ + char *self_exe_path; +#ifdef __APPLE__ + //TODO: get apple device and check the result. + printf("WARNING: not tested for Apple-mac devices \n"); + abort(); +#else + char* buf = strdup(abs_path.c_str()); + char *token = strtok(buf, "/"); + while(token !=NULL){ + self_exe_path = token; + token = strtok(NULL,"/"); + } +#endif + self_exe_path = strtok(self_exe_path, "."); + printf("self exe links to: %s\n", self_exe_path); + return self_exe_path; +} + +static int get_app_cuda_version() { + int app_cuda_version = 0; + char fname[1024]; + snprintf(fname,1024,"_app_cuda_version_XXXXXX"); + int fd=mkstemp(fname); + close(fd); + std::string app_cuda_version_command = "ldd " + get_app_binary() + " | grep libcudart.so | sed 's/.*libcudart.so.\\(.*\\) =>.*/\\1/' > " + fname; + system(app_cuda_version_command.c_str()); + FILE * cmd = fopen(fname, "r"); + char buf[256]; + while (fgets(buf, sizeof(buf), cmd) != 0) { + std::cout << buf; + app_cuda_version = atoi(buf); + } + fclose(cmd); + if ( app_cuda_version == 0 ) { + printf( "Error - Cannot detect the app's CUDA version.\n" ); + exit(1); + } + return app_cuda_version; +} + +//! Keep track of the association between filename and cubin handle +void cuobjdumpRegisterFatBinary(unsigned int handle, const char* filename, CUctx_st *context){ + context->fatbinmap[handle] = filename; +} + +/******************************************************************************* + * Add internal cuda runtime API call to accept gpgpu_context * + *******************************************************************************/ + +void** cudaRegisterFatBinaryInternal( void *fatCubin, gpgpu_context* gpgpu_ctx = NULL) +{ + gpgpu_context *ctx; + if (gpgpu_ctx){ + ctx = gpgpu_ctx; + } else { + ctx = GPGPU_Context(); + } + if(g_debug_execution >= 3){ + announce_call(__my_func__); + } +#if (CUDART_VERSION < 2010) + printf("GPGPU-Sim PTX: ERROR ** this version of GPGPU-Sim requires CUDA 2.1 or higher\n"); + exit(1); +#endif + CUctx_st *context = GPGPUSim_Context(); + static unsigned next_fat_bin_handle = 1; + if(context->get_device()->get_gpgpu()->get_config().use_cuobjdump()) { + // The following workaround has only been verified on 64-bit systems. + if (sizeof(void*) == 4) + printf("GPGPU-Sim PTX: FatBin file name extraction has not been tested on 32-bit system.\n"); + + // This code will get the CUDA version the app was compiled with. + // We need this to determine how to handle the parsing of the binary. + // Making this a runtime variable based on the app, enables GPGPU-Sim compiled + // with a newer version of CUDA to run apps compiled with older versions of + // CUDA. This is especially useful for PTXPLUS execution. + //Skip cuda version check for pytorch application + std::string app_binary_path = get_app_binary(); + int pos = app_binary_path.find("python"); + if (pos==std::string::npos){ + // Not pytorch app : checking cuda version + int app_cuda_version = get_app_cuda_version(); + assert( app_cuda_version == CUDART_VERSION / 1000 && "The app must be compiled with same major version as the simulator." ); + } + + //int app_cuda_version = get_app_cuda_version(); + //assert( app_cuda_version == CUDART_VERSION / 1000 && "The app must be compiled with same major version as the simulator." ); + const char* filename; +#if CUDART_VERSION < 6000 + // FatBin handle from the .fatbin.c file (one of the intermediate files generated by NVCC) + typedef struct {int m; int v; const unsigned long long* d; char* f;} __fatDeviceText __attribute__ ((aligned (8))); + __fatDeviceText * fatDeviceText = (__fatDeviceText *) fatCubin; + + // Extract the source code file name that generate the given FatBin. + // - Obtains the pointer to the actual fatbin structure from the FatBin handle (fatCubin). + // - An integer inside the fatbin structure contains the relative offset to the source code file name. + // - This offset differs among different CUDA and GCC versions. + char * pfatbin = (char*) fatDeviceText->d; + int offset = *((int*)(pfatbin+48)); + filename = (pfatbin+16+offset); +#else + filename = "default"; +#endif + + // The extracted file name is associated with a fat_cubin_handle passed + // into cudaLaunch(). Inside cudaLaunch(), the associated file name is + // used to find the PTX/SASS section from cuobjdump, which contains the + // PTX/SASS code for the launched kernel function. + // This allows us to work around the fact that cuobjdump only outputs the + // file name associated with each section. + unsigned long long fat_cubin_handle = next_fat_bin_handle; + next_fat_bin_handle++; + printf("GPGPU-Sim PTX: __cudaRegisterFatBinary, fat_cubin_handle = %llu, filename=%s\n", fat_cubin_handle, filename); + /*! + * This function extracts all data from all files in first call + * then for next calls, only returns the appropriate number + */ + assert(fat_cubin_handle >= 1); + if (fat_cubin_handle==1) ctx->cuobjdumpInit(); + cuobjdumpRegisterFatBinary(fat_cubin_handle, filename, context); + + return (void**)fat_cubin_handle; + } +#if (CUDART_VERSION < 8000) + else { + static unsigned source_num=1; + unsigned long long fat_cubin_handle = next_fat_bin_handle++; + __cudaFatCudaBinary *info = (__cudaFatCudaBinary *)fatCubin; + assert( info->version >= 3 ); + unsigned num_ptx_versions=0; + unsigned max_capability=0; + unsigned selected_capability=0; + bool found=false; + unsigned forced_max_capability = context->get_device()->get_gpgpu()->get_config().get_forced_max_capability(); + if (!info->ptx){ + printf("ERROR: Cannot find ptx code in cubin file\n" + "\tIf you are using CUDA 4.0 or higher, please enable -gpgpu_ptx_use_cuobjdump or downgrade to CUDA 3.1\n"); + exit(1); + } + while( info->ptx[num_ptx_versions].gpuProfileName != NULL ) { + unsigned capability=0; + sscanf(info->ptx[num_ptx_versions].gpuProfileName,"compute_%u",&capability); + printf("GPGPU-Sim PTX: __cudaRegisterFatBinary found PTX versions for '%s', ", info->ident); + printf("capability = %s\n", info->ptx[num_ptx_versions].gpuProfileName ); + if( forced_max_capability ) { + if( capability > max_capability && capability <= forced_max_capability ) { + found = true; + max_capability=capability; + selected_capability = num_ptx_versions; + } + } else { + if( capability > max_capability ) { + found = true; + max_capability=capability; + selected_capability = num_ptx_versions; + } + } + num_ptx_versions++; + } + if( found ) { + printf("GPGPU-Sim PTX: Loading PTX for %s, capability = %s\n", + info->ident, info->ptx[selected_capability].gpuProfileName ); + symbol_table *symtab; + const char *ptx = info->ptx[selected_capability].ptx; + if(context->get_device()->get_gpgpu()->get_config().convert_to_ptxplus() ) { + printf("GPGPU-Sim PTX: ERROR ** PTXPlus is only supported through cuobjdump\n" + "\tEither enable cuobjdump or disable PTXPlus in your configuration file\n"); + exit(1); + } else { + symtab=gpgpu_ptx_sim_load_ptx_from_string(ptx,source_num); + context->add_binary(symtab,fat_cubin_handle); + gpgpu_ptxinfo_load_from_string( ptx, source_num, max_capability, context->no_of_ptx ); + } + source_num++; + load_static_globals(symtab,STATIC_ALLOC_LIMIT,0xFFFFFFFF,context->get_device()->get_gpgpu()); + load_constants(symtab,STATIC_ALLOC_LIMIT,context->get_device()->get_gpgpu()); + } else { + printf("GPGPU-Sim PTX: warning -- did not find an appropriate PTX in cubin\n"); + } + return (void**)fat_cubin_handle; + } +#else + else { + printf("ERROR ** __cudaRegisterFatBinary() needs to be updated\n"); + abort(); + } +#endif +} + +void cudaRegisterFunctionInternal( + void **fatCubinHandle, + const char *hostFun, + char *deviceFun, + const char *deviceName, + int thread_limit, + uint3 *tid, + uint3 *bid, + dim3 *bDim, + dim3 *gDim, + gpgpu_context *gpgpu_ctx = NULL +) +{ + gpgpu_context *ctx; + if (gpgpu_ctx){ + ctx = gpgpu_ctx; + } else { + ctx = GPGPU_Context(); + } + if(g_debug_execution >= 3){ + announce_call(__my_func__); + } + CUctx_st *context = GPGPUSim_Context(); + unsigned fat_cubin_handle = (unsigned)(unsigned long long)fatCubinHandle; + printf("GPGPU-Sim PTX: __cudaRegisterFunction %s : hostFun 0x%p, fat_cubin_handle = %u\n", + deviceFun, hostFun, fat_cubin_handle); + if(context->get_device()->get_gpgpu()->get_config().use_cuobjdump()) + ctx->cuobjdumpParseBinary(fat_cubin_handle); + context->register_function( fat_cubin_handle, hostFun, deviceFun ); +} + +void cudaRegisterVarInternal( + void **fatCubinHandle, + char *hostVar, //pointer to...something + char *deviceAddress, //name of variable + const char *deviceName, //name of variable (same as above) + int ext, + int size, + int constant, + int global, + gpgpu_context *gpgpu_ctx = NULL) +{ + gpgpu_context *ctx; + if (gpgpu_ctx){ + ctx = gpgpu_ctx; + } else { + ctx = GPGPU_Context(); + } + if(g_debug_execution >= 3){ + announce_call(__my_func__); + } + printf("GPGPU-Sim PTX: __cudaRegisterVar: hostVar = %p; deviceAddress = %s; deviceName = %s\n", hostVar, deviceAddress, deviceName); + printf("GPGPU-Sim PTX: __cudaRegisterVar: Registering const memory space of %d bytes\n", size); + if(GPGPUSim_Context()->get_device()->get_gpgpu()->get_config().use_cuobjdump()) + ctx->cuobjdumpParseBinary((unsigned)(unsigned long long)fatCubinHandle); + fflush(stdout); + if ( constant && !global && !ext ) { + gpgpu_ptx_sim_register_const_variable(hostVar,deviceName,size); + } else if ( !constant && !global && !ext ) { + gpgpu_ptx_sim_register_global_variable(hostVar,deviceName,size); + } else cuda_not_implemented(__my_func__,__LINE__); +} + +cudaError_t cudaConfigureCallInternal(dim3 gridDim, dim3 blockDim, size_t sharedMem, cudaStream_t stream, gpgpu_context* gpgpu_ctx = NULL) +{ + gpgpu_context *ctx; + if (gpgpu_ctx){ + ctx = gpgpu_ctx; + } else { + ctx = GPGPU_Context(); + } + if(g_debug_execution >= 3){ + announce_call(__my_func__); + } + struct CUstream_st *s = (struct CUstream_st *)stream; + ctx->api->g_cuda_launch_stack.push_back( kernel_config(gridDim,blockDim,sharedMem,s) ); + return g_last_cudaError = cudaSuccess; +} + +cudaError_t cudaSetupArgumentInternal(const void *arg, size_t size, size_t offset, gpgpu_context* gpgpu_ctx = NULL) +{ + gpgpu_context *ctx; + if (gpgpu_ctx){ + ctx = gpgpu_ctx; + } else { + ctx = GPGPU_Context(); + } + if(g_debug_execution >= 3){ + announce_call(__my_func__); + } + gpgpusim_ptx_assert( !ctx->api->g_cuda_launch_stack.empty(), "empty launch stack" ); + kernel_config &config = ctx->api->g_cuda_launch_stack.back(); + config.set_arg(arg,size,offset); + printf("GPGPU-Sim PTX: Setting up arguments for %zu bytes starting at 0x%llx..\n",size, (unsigned long long) arg); + + return g_last_cudaError = cudaSuccess; } -void setCuobjdumparch(const char* arch, std::list &cuobjdumpSectionList){ - unsigned archnum; - sscanf(arch, "sm_%u", &archnum); - assert (archnum && "cannot have sm_0"); - printf("Adding arch: %s\n", arch); - cuobjdumpSectionList.front()->setArch(archnum); -} +cudaError_t cudaLaunchInternal( const char *hostFun, gpgpu_context* gpgpu_ctx = NULL ) +{ + gpgpu_context *ctx; + if (gpgpu_ctx){ + ctx = gpgpu_ctx; + } else { + ctx = GPGPU_Context(); + } + if(g_debug_execution >= 3){ + announce_call(__my_func__); + } + CUctx_st* context = GPGPUSim_Context(); + char *mode = getenv("PTX_SIM_MODE_FUNC"); + if( mode ) + sscanf(mode,"%u", &g_ptx_sim_mode); + gpgpusim_ptx_assert( !ctx->api->g_cuda_launch_stack.empty(), "empty launch stack" ); + kernel_config config = ctx->api->g_cuda_launch_stack.back(); + struct CUstream_st *stream = config.get_stream(); + printf("\nGPGPU-Sim PTX: cudaLaunch for 0x%p (mode=%s) on stream %u\n", hostFun, + g_ptx_sim_mode?"functional simulation":"performance simulation", stream?stream->get_uid():0 ); + kernel_info_t *grid = gpgpu_cuda_ptx_sim_init_grid(hostFun,config.get_args(),config.grid_dim(),config.block_dim(),context); + //do dynamic PDOM analysis for performance simulation scenario + std::string kname = grid->name(); + function_info *kernel_func_info = grid->entry(); + if (kernel_func_info->is_pdom_set()) { + printf("GPGPU-Sim PTX: PDOM analysis already done for %s \n", kname.c_str() ); + } else { + printf("GPGPU-Sim PTX: finding reconvergence points for \'%s\'...\n", kname.c_str() ); + kernel_func_info->do_pdom(); + kernel_func_info->set_pdom(); + } + dim3 gridDim = config.grid_dim(); + dim3 blockDim = config.block_dim(); + + gpgpu_t *gpu = context->get_device()->get_gpgpu(); + checkpoint *g_checkpoint; + g_checkpoint = new checkpoint(); + class memory_space *global_mem; + global_mem = gpu->get_global_memory(); -void setCuobjdumpidentifier(const char* identifier, std::list &cuobjdumpSectionList){ - printf("Adding identifier: %s\n", identifier); - cuobjdumpSectionList.front()->setIdentifier(identifier); -} + if(gpu->resume_option ==1 && (grid->get_uid()==gpu->resume_kernel)) + { + + char f1name[2048]; + snprintf(f1name,2048,"checkpoint_files/global_mem_%d.txt", grid->get_uid()); -void setCuobjdumpptxfilename(const char* filename, std::list &cuobjdumpSectionList){ - printf("Adding ptx filename: %s\n", filename); - cuobjdumpSection* x = cuobjdumpSectionList.front(); - if (dynamic_cast(x) == NULL){ - assert (0 && "You shouldn't be trying to add a ptxfilename to an elf section"); + g_checkpoint->load_global_mem(global_mem, f1name); + for (int i=0;iresume_CTA;i++) + grid->increment_cta_id(); } - (dynamic_cast(x))->setPTXfilename(filename); -} + if(gpu->resume_option==1 && (grid->get_uid()resume_kernel)) + { + char f1name[2048]; + snprintf(f1name,2048,"checkpoint_files/global_mem_%d.txt", grid->get_uid()); -void setCuobjdumpelffilename(const char* filename, std::list &cuobjdumpSectionList){ - if (dynamic_cast(cuobjdumpSectionList.front()) == NULL){ - assert (0 && "You shouldn't be trying to add a elffilename to an ptx section"); + g_checkpoint->load_global_mem(global_mem, f1name); + printf("Skipping kernel %d as resuming from kernel %d\n",grid->get_uid(),gpu->resume_kernel ); + ctx->api->g_cuda_launch_stack.pop_back(); + return g_last_cudaError = cudaSuccess; + } - (dynamic_cast(cuobjdumpSectionList.front()))->setELFfilename(filename); -} - -void setCuobjdumpsassfilename(const char* filename, std::list &cuobjdumpSectionList){ - if (dynamic_cast(cuobjdumpSectionList.front()) == NULL){ - assert (0 && "You shouldn't be trying to add a sassfilename to an ptx section"); + if(gpu->checkpoint_option==1 && (grid->get_uid()>gpu->checkpoint_kernel)) + { + printf("Skipping kernel %d as checkpoint from kernel %d\n",grid->get_uid(),gpu->checkpoint_kernel ); + ctx->api->g_cuda_launch_stack.pop_back(); + return g_last_cudaError = cudaSuccess; + } - (dynamic_cast(cuobjdumpSectionList.front()))->setSASSfilename(filename); + printf("GPGPU-Sim PTX: pushing kernel \'%s\' to stream %u, gridDim= (%u,%u,%u) blockDim = (%u,%u,%u) \n", + kname.c_str(), stream?stream->get_uid():0, gridDim.x,gridDim.y,gridDim.z,blockDim.x,blockDim.y,blockDim.z ); + stream_operation op(grid,g_ptx_sim_mode,stream); + g_stream_manager()->push(op); + ctx->api->g_cuda_launch_stack.pop_back(); + return g_last_cudaError = cudaSuccess; } + /******************************************************************************* * * * * @@ -1531,100 +1925,15 @@ __host__ const char* CUDARTAPI cudaGetErrorString(cudaError_t error) return strdup(buf); } -__host__ cudaError_t CUDARTAPI cudaConfigureCall(dim3 gridDim, dim3 blockDim, size_t sharedMem, cudaStream_t stream) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - struct CUstream_st *s = (struct CUstream_st *)stream; - CUctx_st *context = GPGPUSim_Context(); - context->g_cuda_launch_stack.push_back( kernel_config(gridDim,blockDim,sharedMem,s) ); - return g_last_cudaError = cudaSuccess; -} - __host__ cudaError_t CUDARTAPI cudaSetupArgument(const void *arg, size_t size, size_t offset) { - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - CUctx_st *context = GPGPUSim_Context(); - gpgpusim_ptx_assert( !context->g_cuda_launch_stack.empty(), "empty launch stack" ); - kernel_config &config = context->g_cuda_launch_stack.back(); - config.set_arg(arg,size,offset); - printf("GPGPU-Sim PTX: Setting up arguments for %zu bytes starting at 0x%llx..\n",size, (unsigned long long) arg); - - return g_last_cudaError = cudaSuccess; + return cudaSetupArgumentInternal(arg, size, offset); } __host__ cudaError_t CUDARTAPI cudaLaunch( const char *hostFun ) { - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - CUctx_st* context = GPGPUSim_Context(); - char *mode = getenv("PTX_SIM_MODE_FUNC"); - if( mode ) - sscanf(mode,"%u", &g_ptx_sim_mode); - gpgpusim_ptx_assert( !context->g_cuda_launch_stack.empty(), "empty launch stack" ); - kernel_config config = context->g_cuda_launch_stack.back(); - struct CUstream_st *stream = config.get_stream(); - printf("\nGPGPU-Sim PTX: cudaLaunch for 0x%p (mode=%s) on stream %u\n", hostFun, - g_ptx_sim_mode?"functional simulation":"performance simulation", stream?stream->get_uid():0 ); - kernel_info_t *grid = gpgpu_cuda_ptx_sim_init_grid(hostFun,config.get_args(),config.grid_dim(),config.block_dim(),context); - //do dynamic PDOM analysis for performance simulation scenario - std::string kname = grid->name(); - function_info *kernel_func_info = grid->entry(); - if (kernel_func_info->is_pdom_set()) { - printf("GPGPU-Sim PTX: PDOM analysis already done for %s \n", kname.c_str() ); - } else { - printf("GPGPU-Sim PTX: finding reconvergence points for \'%s\'...\n", kname.c_str() ); - kernel_func_info->do_pdom(); - kernel_func_info->set_pdom(); - } - dim3 gridDim = config.grid_dim(); - dim3 blockDim = config.block_dim(); - - gpgpu_t *gpu = context->get_device()->get_gpgpu(); - checkpoint *g_checkpoint; - g_checkpoint = new checkpoint(); - class memory_space *global_mem; - global_mem = gpu->get_global_memory(); - - if(gpu->resume_option ==1 && (grid->get_uid()==gpu->resume_kernel)) - { - - char f1name[2048]; - snprintf(f1name,2048,"checkpoint_files/global_mem_%d.txt", grid->get_uid()); - - g_checkpoint->load_global_mem(global_mem, f1name); - for (int i=0;iresume_CTA;i++) - grid->increment_cta_id(); - } - if(gpu->resume_option==1 && (grid->get_uid()resume_kernel)) - { - char f1name[2048]; - snprintf(f1name,2048,"checkpoint_files/global_mem_%d.txt", grid->get_uid()); - - g_checkpoint->load_global_mem(global_mem, f1name); - printf("Skipping kernel %d as resuming from kernel %d\n",grid->get_uid(),gpu->resume_kernel ); - context->g_cuda_launch_stack.pop_back(); - return g_last_cudaError = cudaSuccess; - - } - if(gpu->checkpoint_option==1 && (grid->get_uid()>gpu->checkpoint_kernel)) - { - printf("Skipping kernel %d as checkpoint from kernel %d\n",grid->get_uid(),gpu->checkpoint_kernel ); - context->g_cuda_launch_stack.pop_back(); - return g_last_cudaError = cudaSuccess; - - } - printf("GPGPU-Sim PTX: pushing kernel \'%s\' to stream %u, gridDim= (%u,%u,%u) blockDim = (%u,%u,%u) \n", - kname.c_str(), stream?stream->get_uid():0, gridDim.x,gridDim.y,gridDim.z,blockDim.x,blockDim.y,blockDim.z ); - stream_operation op(grid,g_ptx_sim_mode,stream); - g_stream_manager()->push(op); - context->g_cuda_launch_stack.pop_back(); - return g_last_cudaError = cudaSuccess; + return cudaLaunchInternal( hostFun ); } __host__ cudaError_t CUDARTAPI cudaLaunchKernel ( const char* hostFun, dim3 gridDim, dim3 blockDim, const void** args, size_t sharedMem, cudaStream_t stream ) @@ -1636,13 +1945,13 @@ __host__ cudaError_t CUDARTAPI cudaLaunchKernel ( const char* hostFun, dim3 grid CUctx_st *context = GPGPUSim_Context(); function_info *entry = context->get_kernel(hostFun); - cudaConfigureCall(gridDim, blockDim, sharedMem, stream); + cudaConfigureCallInternal(gridDim, blockDim, sharedMem, stream); for(unsigned i = 0; i < entry->num_args(); i++){ std::pair p = entry->get_param_config(i); - cudaSetupArgument(args[i], p.first, p.second); + cudaSetupArgumentInternal(args[i], p.first, p.second); } - cudaLaunch(hostFun); + cudaLaunchInternal(hostFun); return g_last_cudaError = cudaSuccess; } @@ -1925,83 +2234,32 @@ typedef int (*ExportedFunction)(); static ExportedFunction exportTable[3] = {&dummy0, &dummy0, &dummy0}; -__host__ cudaError_t CUDARTAPI cudaGetExportTable(const void **ppExportTable, const cudaUUID_t *pExportTableId) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("cudaGetExportTable: UUID = "); - for (int s = 0; s < 16; s++) { - printf("%#2x ", (unsigned char) (pExportTableId->bytes[s])); - } - *ppExportTable = &exportTable; - - printf("\n"); - return g_last_cudaError = cudaSuccess; -} - -#endif - - -/******************************************************************************* - * * - * * - * * - *******************************************************************************/ - -//#include "../../cuobjdump_to_ptxplus/cuobjdump_parser.h" - -//! Return the executable file of the process containing the PTX/SASS code -//! -//! This Function returns the executable file ran by the process. This -//! executable is supposed to contain the PTX/SASS code. It provides workaround -//! for processes running on valgrind by dereferencing /proc//exe within the -//! GPGPU-Sim process before calling cuobjdump to extract PTX/SASS. This is -//! needed because valgrind uses x86 emulation to detect memory leak. Other -//! processes (e.g. cuobjdump) reading /proc//exe will see the emulator -//! executable instead of the application binary. -//! -std::string get_app_binary(){ - char self_exe_path[1025]; -#ifdef __APPLE__ - uint32_t size = sizeof(self_exe_path); - if( _NSGetExecutablePath(self_exe_path,&size) != 0 ) { - printf("GPGPU-Sim ** ERROR: _NSGetExecutablePath input buffer too small\n"); - exit(1); - } -#else - std::stringstream exec_link; - exec_link << "/proc/self/exe"; - - ssize_t path_length = readlink(exec_link.str().c_str(), self_exe_path, 1024); - assert(path_length != -1); - self_exe_path[path_length] = '\0'; -#endif - - printf("self exe links to: %s\n", self_exe_path); - return self_exe_path; -} - -//above func gives abs path whereas this give just the name of application. -char* get_app_binary_name(std::string abs_path){ - char *self_exe_path; -#ifdef __APPLE__ - //TODO: get apple device and check the result. - printf("WARNING: not tested for Apple-mac devices \n"); - abort(); -#else - char* buf = strdup(abs_path.c_str()); - char *token = strtok(buf, "/"); - while(token !=NULL){ - self_exe_path = token; - token = strtok(NULL,"/"); - } -#endif - self_exe_path = strtok(self_exe_path, "."); - printf("self exe links to: %s\n", self_exe_path); - return self_exe_path; +__host__ cudaError_t CUDARTAPI cudaGetExportTable(const void **ppExportTable, const cudaUUID_t *pExportTableId) +{ + if(g_debug_execution >= 3){ + announce_call(__my_func__); + } + printf("cudaGetExportTable: UUID = "); + for (int s = 0; s < 16; s++) { + printf("%#2x ", (unsigned char) (pExportTableId->bytes[s])); + } + *ppExportTable = &exportTable; + + printf("\n"); + return g_last_cudaError = cudaSuccess; } +#endif + + +/******************************************************************************* + * * + * * + * * + *******************************************************************************/ + +//#include "../../cuobjdump_to_ptxplus/cuobjdump_parser.h" + //extracts all ptx files from binary and dumps into prog_name.unique_no.sm_<>.ptx files void gpgpu_context::extract_ptx_files_using_cuobjdump(CUctx_st *context){ extern bool g_cdp_enabled; @@ -2072,28 +2330,6 @@ void gpgpu_context::extract_ptx_files_using_cuobjdump(CUctx_st *context){ } -static int get_app_cuda_version() { - int app_cuda_version = 0; - char fname[1024]; - snprintf(fname,1024,"_app_cuda_version_XXXXXX"); - int fd=mkstemp(fname); - close(fd); - std::string app_cuda_version_command = "ldd " + get_app_binary() + " | grep libcudart.so | sed 's/.*libcudart.so.\\(.*\\) =>.*/\\1/' > " + fname; - system(app_cuda_version_command.c_str()); - FILE * cmd = fopen(fname, "r"); - char buf[256]; - while (fgets(buf, sizeof(buf), cmd) != 0) { - std::cout << buf; - app_cuda_version = atoi(buf); - } - fclose(cmd); - if ( app_cuda_version == 0 ) { - printf( "Error - Cannot detect the app's CUDA version.\n" ); - exit(1); - } - return app_cuda_version; -} - //! Call cuobjdump to extract everything (-elf -sass -ptx) /*! @@ -2475,11 +2711,6 @@ void gpgpu_context::cuobjdumpInit(){ } -//! Keep track of the association between filename and cubin handle -void cuobjdumpRegisterFatBinary(unsigned int handle, const char* filename, CUctx_st *context){ - context->fatbinmap[handle] = filename; -} - //! Either submit PTX for simulation or convert SASS to PTXPlus and submit it void gpgpu_context::cuobjdumpParseBinary(unsigned int handle){ @@ -2570,214 +2801,11 @@ void gpgpu_context::cuobjdumpParseBinary(unsigned int handle){ } } -void** cudaRegisterFatBinary( void *fatCubin, gpgpu_context* gpgpu_ctx = NULL) -{ - gpgpu_context *ctx; - if (gpgpu_ctx){ - ctx = gpgpu_ctx; - } else { - ctx = GPGPU_Context(); - } - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } -#if (CUDART_VERSION < 2010) - printf("GPGPU-Sim PTX: ERROR ** this version of GPGPU-Sim requires CUDA 2.1 or higher\n"); - exit(1); -#endif - CUctx_st *context = GPGPUSim_Context(); - static unsigned next_fat_bin_handle = 1; - if(context->get_device()->get_gpgpu()->get_config().use_cuobjdump()) { - // The following workaround has only been verified on 64-bit systems. - if (sizeof(void*) == 4) - printf("GPGPU-Sim PTX: FatBin file name extraction has not been tested on 32-bit system.\n"); - - // This code will get the CUDA version the app was compiled with. - // We need this to determine how to handle the parsing of the binary. - // Making this a runtime variable based on the app, enables GPGPU-Sim compiled - // with a newer version of CUDA to run apps compiled with older versions of - // CUDA. This is especially useful for PTXPLUS execution. - //Skip cuda version check for pytorch application - std::string app_binary_path = get_app_binary(); - int pos = app_binary_path.find("python"); - if (pos==std::string::npos){ - // Not pytorch app : checking cuda version - int app_cuda_version = get_app_cuda_version(); - assert( app_cuda_version == CUDART_VERSION / 1000 && "The app must be compiled with same major version as the simulator." ); - } - - //int app_cuda_version = get_app_cuda_version(); - //assert( app_cuda_version == CUDART_VERSION / 1000 && "The app must be compiled with same major version as the simulator." ); - const char* filename; -#if CUDART_VERSION < 6000 - // FatBin handle from the .fatbin.c file (one of the intermediate files generated by NVCC) - typedef struct {int m; int v; const unsigned long long* d; char* f;} __fatDeviceText __attribute__ ((aligned (8))); - __fatDeviceText * fatDeviceText = (__fatDeviceText *) fatCubin; - - // Extract the source code file name that generate the given FatBin. - // - Obtains the pointer to the actual fatbin structure from the FatBin handle (fatCubin). - // - An integer inside the fatbin structure contains the relative offset to the source code file name. - // - This offset differs among different CUDA and GCC versions. - char * pfatbin = (char*) fatDeviceText->d; - int offset = *((int*)(pfatbin+48)); - filename = (pfatbin+16+offset); -#else - filename = "default"; -#endif - - // The extracted file name is associated with a fat_cubin_handle passed - // into cudaLaunch(). Inside cudaLaunch(), the associated file name is - // used to find the PTX/SASS section from cuobjdump, which contains the - // PTX/SASS code for the launched kernel function. - // This allows us to work around the fact that cuobjdump only outputs the - // file name associated with each section. - unsigned long long fat_cubin_handle = next_fat_bin_handle; - next_fat_bin_handle++; - printf("GPGPU-Sim PTX: __cudaRegisterFatBinary, fat_cubin_handle = %llu, filename=%s\n", fat_cubin_handle, filename); - /*! - * This function extracts all data from all files in first call - * then for next calls, only returns the appropriate number - */ - assert(fat_cubin_handle >= 1); - if (fat_cubin_handle==1) ctx->cuobjdumpInit(); - cuobjdumpRegisterFatBinary(fat_cubin_handle, filename, context); - - return (void**)fat_cubin_handle; - } -#if (CUDART_VERSION < 8000) - else { - static unsigned source_num=1; - unsigned long long fat_cubin_handle = next_fat_bin_handle++; - __cudaFatCudaBinary *info = (__cudaFatCudaBinary *)fatCubin; - assert( info->version >= 3 ); - unsigned num_ptx_versions=0; - unsigned max_capability=0; - unsigned selected_capability=0; - bool found=false; - unsigned forced_max_capability = context->get_device()->get_gpgpu()->get_config().get_forced_max_capability(); - if (!info->ptx){ - printf("ERROR: Cannot find ptx code in cubin file\n" - "\tIf you are using CUDA 4.0 or higher, please enable -gpgpu_ptx_use_cuobjdump or downgrade to CUDA 3.1\n"); - exit(1); - } - while( info->ptx[num_ptx_versions].gpuProfileName != NULL ) { - unsigned capability=0; - sscanf(info->ptx[num_ptx_versions].gpuProfileName,"compute_%u",&capability); - printf("GPGPU-Sim PTX: __cudaRegisterFatBinary found PTX versions for '%s', ", info->ident); - printf("capability = %s\n", info->ptx[num_ptx_versions].gpuProfileName ); - if( forced_max_capability ) { - if( capability > max_capability && capability <= forced_max_capability ) { - found = true; - max_capability=capability; - selected_capability = num_ptx_versions; - } - } else { - if( capability > max_capability ) { - found = true; - max_capability=capability; - selected_capability = num_ptx_versions; - } - } - num_ptx_versions++; - } - if( found ) { - printf("GPGPU-Sim PTX: Loading PTX for %s, capability = %s\n", - info->ident, info->ptx[selected_capability].gpuProfileName ); - symbol_table *symtab; - const char *ptx = info->ptx[selected_capability].ptx; - if(context->get_device()->get_gpgpu()->get_config().convert_to_ptxplus() ) { - printf("GPGPU-Sim PTX: ERROR ** PTXPlus is only supported through cuobjdump\n" - "\tEither enable cuobjdump or disable PTXPlus in your configuration file\n"); - exit(1); - } else { - symtab=gpgpu_ptx_sim_load_ptx_from_string(ptx,source_num); - context->add_binary(symtab,fat_cubin_handle); - gpgpu_ptxinfo_load_from_string( ptx, source_num, max_capability, context->no_of_ptx ); - } - source_num++; - load_static_globals(symtab,STATIC_ALLOC_LIMIT,0xFFFFFFFF,context->get_device()->get_gpgpu()); - load_constants(symtab,STATIC_ALLOC_LIMIT,context->get_device()->get_gpgpu()); - } else { - printf("GPGPU-Sim PTX: warning -- did not find an appropriate PTX in cubin\n"); - } - return (void**)fat_cubin_handle; - } -#else - else { - printf("ERROR ** __cudaRegisterFatBinary() needs to be updated\n"); - abort(); - } -#endif -} - -void cudaRegisterFunction( - void **fatCubinHandle, - const char *hostFun, - char *deviceFun, - const char *deviceName, - int thread_limit, - uint3 *tid, - uint3 *bid, - dim3 *bDim, - dim3 *gDim, - gpgpu_context *gpgpu_ctx = NULL -) -{ - gpgpu_context *ctx; - if (gpgpu_ctx){ - ctx = gpgpu_ctx; - } else { - ctx = GPGPU_Context(); - } - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - CUctx_st *context = GPGPUSim_Context(); - unsigned fat_cubin_handle = (unsigned)(unsigned long long)fatCubinHandle; - printf("GPGPU-Sim PTX: __cudaRegisterFunction %s : hostFun 0x%p, fat_cubin_handle = %u\n", - deviceFun, hostFun, fat_cubin_handle); - if(context->get_device()->get_gpgpu()->get_config().use_cuobjdump()) - ctx->cuobjdumpParseBinary(fat_cubin_handle); - context->register_function( fat_cubin_handle, hostFun, deviceFun ); -} - -void cudaRegisterVar( - void **fatCubinHandle, - char *hostVar, //pointer to...something - char *deviceAddress, //name of variable - const char *deviceName, //name of variable (same as above) - int ext, - int size, - int constant, - int global, - gpgpu_context *gpgpu_ctx = NULL) -{ - gpgpu_context *ctx; - if (gpgpu_ctx){ - ctx = gpgpu_ctx; - } else { - ctx = GPGPU_Context(); - } - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("GPGPU-Sim PTX: __cudaRegisterVar: hostVar = %p; deviceAddress = %s; deviceName = %s\n", hostVar, deviceAddress, deviceName); - printf("GPGPU-Sim PTX: __cudaRegisterVar: Registering const memory space of %d bytes\n", size); - if(GPGPUSim_Context()->get_device()->get_gpgpu()->get_config().use_cuobjdump()) - ctx->cuobjdumpParseBinary((unsigned)(unsigned long long)fatCubinHandle); - fflush(stdout); - if ( constant && !global && !ext ) { - gpgpu_ptx_sim_register_const_variable(hostVar,deviceName,size); - } else if ( !constant && !global && !ext ) { - gpgpu_ptx_sim_register_global_variable(hostVar,deviceName,size); - } else cuda_not_implemented(__my_func__,__LINE__); -} - extern "C" { void** CUDARTAPI __cudaRegisterFatBinary( void *fatCubin ) { - return cudaRegisterFatBinary(fatCubin); + return cudaRegisterFatBinaryInternal(fatCubin); } void CUDARTAPI __cudaRegisterFunction( @@ -2791,7 +2819,7 @@ void CUDARTAPI __cudaRegisterFunction( dim3 *bDim, dim3 *gDim ) { - cudaRegisterFunction( + cudaRegisterFunctionInternal( fatCubinHandle, hostFun, deviceFun, @@ -2815,7 +2843,7 @@ extern void __cudaRegisterVar( int constant, int global ) { - cudaRegisterVar( + cudaRegisterVarInternal( fatCubinHandle, hostVar, deviceAddress, @@ -2825,6 +2853,12 @@ extern void __cudaRegisterVar( constant, global ); } + +__host__ cudaError_t CUDARTAPI cudaConfigureCall(dim3 gridDim, dim3 blockDim, size_t sharedMem, cudaStream_t stream) +{ + return cudaConfigureCallInternal(gridDim, blockDim, sharedMem, stream); +} + void __cudaUnregisterFatBinary(void **fatCubinHandle) { if(g_debug_execution >= 3){ @@ -4868,12 +4902,12 @@ CUresult CUDAAPI cuLaunchKernel(CUfunction f, const char *hostFun = (const char*) f; CUctx_st *context = GPGPUSim_Context(); function_info *entry = context->get_kernel(hostFun); - cudaConfigureCall(dim3(gridDimX, gridDimY, gridDimZ), dim3(blockDimX, blockDimY, blockDimZ), sharedMemBytes, (cudaStream_t) hStream); + cudaConfigureCallInternal(dim3(gridDimX, gridDimY, gridDimZ), dim3(blockDimX, blockDimY, blockDimZ), sharedMemBytes, (cudaStream_t) hStream); for(unsigned i = 0; i < entry->num_args(); i++){ std::pair p = entry->get_param_config(i); cudaSetupArgument(kernelParams[i], p.first, p.second); } - cudaLaunch(hostFun); + cudaLaunchInternal(hostFun); return CUDA_SUCCESS; } #endif /* CUDART_VERSION >= 4000 */ -- cgit v1.3 From 570ce0a3d6e049ec3ee42329fbf25019fd4eb2e5 Mon Sep 17 00:00:00 2001 From: Mengchi Zhang Date: Thu, 6 Jun 2019 01:57:40 -0400 Subject: Move 3 more var Signed-off-by: Mengchi Zhang --- libcuda/cuda_runtime_api.cc | 92 +++++++++++++++++++++++++-------------------- libcuda/gpgpu_context.h | 6 +++ 2 files changed, 58 insertions(+), 40 deletions(-) (limited to 'libcuda/cuda_runtime_api.cc') diff --git a/libcuda/cuda_runtime_api.cc b/libcuda/cuda_runtime_api.cc index ea96570..855f11d 100644 --- a/libcuda/cuda_runtime_api.cc +++ b/libcuda/cuda_runtime_api.cc @@ -313,10 +313,7 @@ struct CUctx_st { return i->second; } - std::mapfatbin_registered; - std::map fatbinmap; std::map g_mallocPtr_Size; - std::map name_symtab; std::map pinned_memory; //support for pinned memories added std::map pinned_memory_size; int no_of_ptx; @@ -647,8 +644,8 @@ static int get_app_cuda_version() { } //! Keep track of the association between filename and cubin handle -void cuobjdumpRegisterFatBinary(unsigned int handle, const char* filename, CUctx_st *context){ - context->fatbinmap[handle] = filename; +void gpgpu_context::cuobjdumpRegisterFatBinary(unsigned int handle, const char* filename, CUctx_st *context){ + fatbinmap[handle] = filename; } /******************************************************************************* @@ -725,7 +722,7 @@ void** cudaRegisterFatBinaryInternal( void *fatCubin, gpgpu_context* gpgpu_ctx = */ assert(fat_cubin_handle >= 1); if (fat_cubin_handle==1) ctx->cuobjdumpInit(); - cuobjdumpRegisterFatBinary(fat_cubin_handle, filename, context); + ctx->cuobjdumpRegisterFatBinary(fat_cubin_handle, filename, context); return (void**)fat_cubin_handle; } @@ -969,6 +966,46 @@ cudaError_t cudaLaunchInternal( const char *hostFun, gpgpu_context* gpgpu_ctx = return g_last_cudaError = cudaSuccess; } +#if CUDART_VERSION >= 6050 +CUresult +cuLinkAddFileInternal(CUlinkState state, CUjitInputType type, const char *path, + unsigned int numOptions, CUjit_option *options, void **optionValues, gpgpu_context* gpgpu_ctx = NULL) +{ + gpgpu_context *ctx; + if (gpgpu_ctx){ + ctx = gpgpu_ctx; + } else { + ctx = GPGPU_Context(); + } + if(g_debug_execution >= 3){ + announce_call(__my_func__); + } + static bool addedFile = false; + if (addedFile){ + printf("GPGPU-Sim PTX: ERROR: cuLinkAddFile does not support multiple files\n"); + abort(); + } + + //blocking + assert(type==CU_JIT_INPUT_PTX); + CUctx_st *context = GPGPUSim_Context(); + char *file = getenv("PTX_JIT_PATH"); + if(file==NULL){ + printf("GPGPU-Sim PTX: ERROR: PTX_JIT_PATH has not been set\n"); + abort(); + } + strcat(file,"/"); + strcat(file,path); + symbol_table *symtab = gpgpu_ptx_sim_load_ptx_from_filename( file ); + std::string fname(path); + ctx->name_symtab[fname] = symtab; + context->add_binary(symtab, 1); + load_static_globals(symtab,STATIC_ALLOC_LIMIT,0xFFFFFFFF,context->get_device()->get_gpgpu()); + load_constants(symtab,STATIC_ALLOC_LIMIT,context->get_device()->get_gpgpu()); + addedFile = true; + return CUDA_SUCCESS; +} +#endif /******************************************************************************* * * @@ -2715,12 +2752,12 @@ void gpgpu_context::cuobjdumpInit(){ void gpgpu_context::cuobjdumpParseBinary(unsigned int handle){ CUctx_st *context = GPGPUSim_Context(); - if(context->fatbin_registered[handle]) return; - context->fatbin_registered[handle] = true; - std::string fname = context->fatbinmap[handle]; + if(fatbin_registered[handle]) return; + fatbin_registered[handle] = true; + std::string fname = fatbinmap[handle]; - if (context->name_symtab.find(fname) != context->name_symtab.end()) { - symbol_table *symtab = context->name_symtab[fname]; + if (name_symtab.find(fname) != name_symtab.end()) { + symbol_table *symtab = name_symtab[fname]; context->add_binary(symtab, handle); return; } @@ -2737,7 +2774,7 @@ void gpgpu_context::cuobjdumpParseBinary(unsigned int handle){ symtab = gpgpu_ptx_sim_load_ptx_from_filename( ptx_filename.c_str() ); } } - context->name_symtab[fname] = symtab; + name_symtab[fname] = symtab; context->add_binary(symtab, handle); load_static_globals(symtab,STATIC_ALLOC_LIMIT,0xFFFFFFFF,context->get_device()->get_gpgpu()); load_constants(symtab,STATIC_ALLOC_LIMIT,context->get_device()->get_gpgpu()); @@ -2795,7 +2832,7 @@ void gpgpu_context::cuobjdumpParseBinary(unsigned int handle){ } load_static_globals(symtab,STATIC_ALLOC_LIMIT,0xFFFFFFFF,context->get_device()->get_gpgpu()); load_constants(symtab,STATIC_ALLOC_LIMIT,context->get_device()->get_gpgpu()); - context->name_symtab[fname] = symtab; + name_symtab[fname] = symtab; //TODO: Remove temporarily files as per configurations } @@ -3909,33 +3946,8 @@ CUresult CUDAAPI cuLinkAddFile(CUlinkState state, CUjitInputType type, const char *path, unsigned int numOptions, CUjit_option *options, void **optionValues) { - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - static bool addedFile = false; - if (addedFile){ - printf("GPGPU-Sim PTX: ERROR: cuLinkAddFile does not support multiple files\n"); - abort(); - } - - //blocking - assert(type==CU_JIT_INPUT_PTX); - CUctx_st *context = GPGPUSim_Context(); - char *file = getenv("PTX_JIT_PATH"); - if(file==NULL){ - printf("GPGPU-Sim PTX: ERROR: PTX_JIT_PATH has not been set\n"); - abort(); - } - strcat(file,"/"); - strcat(file,path); - symbol_table *symtab = gpgpu_ptx_sim_load_ptx_from_filename( file ); - std::string fname(path); - context->name_symtab[fname] = symtab; - context->add_binary(symtab, 1); - load_static_globals(symtab,STATIC_ALLOC_LIMIT,0xFFFFFFFF,context->get_device()->get_gpgpu()); - load_constants(symtab,STATIC_ALLOC_LIMIT,context->get_device()->get_gpgpu()); - addedFile = true; - return CUDA_SUCCESS; + return cuLinkAddFileInternal(state, type, path, + numOptions, options, optionValues) } #endif diff --git a/libcuda/gpgpu_context.h b/libcuda/gpgpu_context.h index 576ec67..0543ff8 100644 --- a/libcuda/gpgpu_context.h +++ b/libcuda/gpgpu_context.h @@ -3,11 +3,13 @@ #include #include #include +#include #include "cuda_api_object.h" class cuobjdumpSection; class cuobjdumpELFSection; class cuobjdumpPTXSection; +class symbol_table; class gpgpu_context { public: @@ -16,6 +18,9 @@ class gpgpu_context { } // global list std::list cuobjdumpSectionList; + std::mapfatbin_registered; + std::map fatbinmap; + std::map name_symtab; //maps sm version number to set of filenames std::map > version_filename; cuda_runtime_api* api; @@ -29,5 +34,6 @@ class gpgpu_context { cuobjdumpELFSection* findELFSection(const std::string identifier); cuobjdumpPTXSection* findPTXSection(const std::string identifier); void extract_ptx_files_using_cuobjdump(CUctx_st *context); + void cuobjdumpRegisterFatBinary(unsigned int handle, const char* filename, CUctx_st *context); }; #endif /* __gpgpu_context_h__ */ -- cgit v1.3 From ca08abd4340e990d0c135f93672184a8e2116ecd Mon Sep 17 00:00:00 2001 From: Mengchi Zhang Date: Thu, 6 Jun 2019 02:28:12 -0400 Subject: Miss semicolon Signed-off-by: Mengchi Zhang --- libcuda/cuda_runtime_api.cc | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) (limited to 'libcuda/cuda_runtime_api.cc') diff --git a/libcuda/cuda_runtime_api.cc b/libcuda/cuda_runtime_api.cc index 855f11d..66562aa 100644 --- a/libcuda/cuda_runtime_api.cc +++ b/libcuda/cuda_runtime_api.cc @@ -3947,7 +3947,7 @@ cuLinkAddFile(CUlinkState state, CUjitInputType type, const char *path, unsigned int numOptions, CUjit_option *options, void **optionValues) { return cuLinkAddFileInternal(state, type, path, - numOptions, options, optionValues) + numOptions, options, optionValues); } #endif -- cgit v1.3 From bc8d1e10507f043c916ee051dc9a687adc6d9b4b Mon Sep 17 00:00:00 2001 From: Mengchi Zhang Date: Thu, 6 Jun 2019 02:47:07 -0400 Subject: Move more vars Signed-off-by: Mengchi Zhang --- libcuda/cuda_runtime_api.cc | 117 ++++++++++++++++++++++++++------------------ libcuda/gpgpu_context.h | 11 +++++ 2 files changed, 81 insertions(+), 47 deletions(-) (limited to 'libcuda/cuda_runtime_api.cc') diff --git a/libcuda/cuda_runtime_api.cc b/libcuda/cuda_runtime_api.cc index 66562aa..02e2b2e 100644 --- a/libcuda/cuda_runtime_api.cc +++ b/libcuda/cuda_runtime_api.cc @@ -158,11 +158,11 @@ extern void exit_simulation(); static int load_static_globals( symbol_table *symtab, unsigned min_gaddr, unsigned max_gaddr, gpgpu_t *gpu ); static int load_constants( symbol_table *symtab, addr_t min_gaddr, gpgpu_t *gpu ); -static kernel_info_t *gpgpu_cuda_ptx_sim_init_grid( const char *kernel_key, - gpgpu_ptx_sim_arg_list_t args, - struct dim3 gridDim, - struct dim3 blockDim, - struct CUctx_st* context ); +//static kernel_info_t *gpgpu_cuda_ptx_sim_init_grid( const char *kernel_key, +// gpgpu_ptx_sim_arg_list_t args, +// struct dim3 gridDim, +// struct dim3 blockDim, +// struct CUctx_st* context ); /*DEVICE_BUILTIN*/ struct cudaArray @@ -313,7 +313,6 @@ struct CUctx_st { return i->second; } - std::map g_mallocPtr_Size; std::map pinned_memory; //support for pinned memories added std::map pinned_memory_size; int no_of_ptx; @@ -910,7 +909,7 @@ cudaError_t cudaLaunchInternal( const char *hostFun, gpgpu_context* gpgpu_ctx = struct CUstream_st *stream = config.get_stream(); printf("\nGPGPU-Sim PTX: cudaLaunch for 0x%p (mode=%s) on stream %u\n", hostFun, g_ptx_sim_mode?"functional simulation":"performance simulation", stream?stream->get_uid():0 ); - kernel_info_t *grid = gpgpu_cuda_ptx_sim_init_grid(hostFun,config.get_args(),config.grid_dim(),config.block_dim(),context); + kernel_info_t *grid = ctx->gpgpu_cuda_ptx_sim_init_grid(hostFun,config.get_args(),config.grid_dim(),config.block_dim(),context); //do dynamic PDOM analysis for performance simulation scenario std::string kname = grid->name(); function_info *kernel_func_info = grid->entry(); @@ -966,6 +965,66 @@ cudaError_t cudaLaunchInternal( const char *hostFun, gpgpu_context* gpgpu_ctx = return g_last_cudaError = cudaSuccess; } +cudaError_t cudaMallocInternal(void **devPtr, size_t size, gpgpu_context* gpgpu_ctx = NULL) +{ + gpgpu_context *ctx; + if (gpgpu_ctx){ + ctx = gpgpu_ctx; + } else { + ctx = GPGPU_Context(); + } + if(g_debug_execution >= 3){ + announce_call(__my_func__); + } + CUctx_st* context = GPGPUSim_Context(); + *devPtr = context->get_device()->get_gpgpu()->gpu_malloc(size); + if(g_debug_execution >= 3){ + printf("GPGPU-Sim PTX: cudaMallocing %zu bytes starting at 0x%llx..\n",size, (unsigned long long) *devPtr); + ctx->g_mallocPtr_Size[(unsigned long long)*devPtr] = size; + } + if ( *devPtr ) { + return g_last_cudaError = cudaSuccess; + } else { + return g_last_cudaError = cudaErrorMemoryAllocation; + } +} + +cudaError_t cudaHostGetDevicePointerInternal(void **pDevice, void *pHost, unsigned int flags, gpgpu_context* gpgpu_ctx = NULL) +{ + gpgpu_context *ctx; + if (gpgpu_ctx){ + ctx = gpgpu_ctx; + } else { + ctx = GPGPU_Context(); + } + if(g_debug_execution >= 3){ + announce_call(__my_func__); + } + if(g_debug_execution >= 3){ + announce_call(__my_func__); + } + //only cpu memory allocation happens in cudaHostAlloc. Linking with device pointer to pinned memory happens here. + //TODO: once kernel is executed, the contents in global pointer of GPU must be copied back to CPU host pointer! + flags=0; + CUctx_st* context = GPGPUSim_Context(); + gpgpu_t *gpu = context->get_device()->get_gpgpu(); + std::map::const_iterator i = context->pinned_memory_size.find(pHost); + assert(i != context->pinned_memory_size.end()); + size_t size = i->second; + *pDevice = gpu->gpu_malloc(size); + if(g_debug_execution >= 3){ + printf("GPGPU-Sim PTX: cudaMallocing %zu bytes starting at 0x%llx..\n",size, (unsigned long long) *pDevice); + ctx->g_mallocPtr_Size[(unsigned long long)*pDevice] = size; + } + if ( *pDevice ) { + context->pinned_memory[pHost]=pDevice; + //Copy contents in cpu to gpu + gpu->memcpy_to_gpu((size_t)*pDevice,pHost,size); + return g_last_cudaError = cudaSuccess; + } else { + return g_last_cudaError = cudaErrorMemoryAllocation; + } +} #if CUDART_VERSION >= 6050 CUresult cuLinkAddFileInternal(CUlinkState state, CUjitInputType type, const char *path, @@ -1027,20 +1086,7 @@ cudaError_t cudaPeekAtLastError(void) __host__ cudaError_t CUDARTAPI cudaMalloc(void **devPtr, size_t size) { - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - CUctx_st* context = GPGPUSim_Context(); - *devPtr = context->get_device()->get_gpgpu()->gpu_malloc(size); - if(g_debug_execution >= 3){ - printf("GPGPU-Sim PTX: cudaMallocing %zu bytes starting at 0x%llx..\n",size, (unsigned long long) *devPtr); - context->g_mallocPtr_Size[(unsigned long long)*devPtr] = size; - } - if ( *devPtr ) { - return g_last_cudaError = cudaSuccess; - } else { - return g_last_cudaError = cudaErrorMemoryAllocation; - } + return cudaMallocInternal(devPtr, size); } __host__ cudaError_t CUDARTAPI cudaMallocHost(void **ptr, size_t size) @@ -3111,30 +3157,7 @@ cudaError_t CUDARTAPI cudaHostAlloc(void **pHost, size_t bytes, unsigned int fl cudaError_t CUDARTAPI cudaHostGetDevicePointer(void **pDevice, void *pHost, unsigned int flags) { - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - //only cpu memory allocation happens in cudaHostAlloc. Linking with device pointer to pinned memory happens here. - //TODO: once kernel is executed, the contents in global pointer of GPU must be copied back to CPU host pointer! - flags=0; - CUctx_st* context = GPGPUSim_Context(); - gpgpu_t *gpu = context->get_device()->get_gpgpu(); - std::map::const_iterator i = context->pinned_memory_size.find(pHost); - assert(i != context->pinned_memory_size.end()); - size_t size = i->second; - *pDevice = gpu->gpu_malloc(size); - if(g_debug_execution >= 3){ - printf("GPGPU-Sim PTX: cudaMallocing %zu bytes starting at 0x%llx..\n",size, (unsigned long long) *pDevice); - context->g_mallocPtr_Size[(unsigned long long)*pDevice] = size; - } - if ( *pDevice ) { - context->pinned_memory[pHost]=pDevice; - //Copy contents in cpu to gpu - gpu->memcpy_to_gpu((size_t)*pDevice,pHost,size); - return g_last_cudaError = cudaSuccess; - } else { - return g_last_cudaError = cudaErrorMemoryAllocation; - } + return cudaHostGetDevicePointerInternal(pDevice, pHost, flags); } __host__ cudaError_t CUDARTAPI cudaPointerGetAttributes( @@ -3450,7 +3473,7 @@ static int load_constants( symbol_table *symtab, addr_t min_gaddr, gpgpu_t *gpu return nc_bytes; } -kernel_info_t *gpgpu_cuda_ptx_sim_init_grid( const char *hostFun, +kernel_info_t * gpgpu_context::gpgpu_cuda_ptx_sim_init_grid( const char *hostFun, gpgpu_ptx_sim_arg_list_t args, struct dim3 gridDim, struct dim3 blockDim, @@ -3482,7 +3505,7 @@ kernel_info_t *gpgpu_cuda_ptx_sim_init_grid( const char *hostFun, fflush(stdout); if(g_debug_execution >= 4){ - entry->ptx_jit_config(context->g_mallocPtr_Size, result->get_param_memory(), (gpgpu_t *) context->get_device()->get_gpgpu(), gridDim, blockDim); + entry->ptx_jit_config(g_mallocPtr_Size, result->get_param_memory(), (gpgpu_t *) context->get_device()->get_gpgpu(), gridDim, blockDim); } return result; diff --git a/libcuda/gpgpu_context.h b/libcuda/gpgpu_context.h index 0543ff8..f29e2e0 100644 --- a/libcuda/gpgpu_context.h +++ b/libcuda/gpgpu_context.h @@ -10,6 +10,10 @@ class cuobjdumpSection; class cuobjdumpELFSection; class cuobjdumpPTXSection; class symbol_table; +class gpgpu_ptx_sim_arg; +class kernel_info_t; + +typedef std::list gpgpu_ptx_sim_arg_list_t; class gpgpu_context { public: @@ -21,6 +25,7 @@ class gpgpu_context { std::mapfatbin_registered; std::map fatbinmap; std::map name_symtab; + std::map g_mallocPtr_Size; //maps sm version number to set of filenames std::map > version_filename; cuda_runtime_api* api; @@ -35,5 +40,11 @@ class gpgpu_context { cuobjdumpPTXSection* findPTXSection(const std::string identifier); void extract_ptx_files_using_cuobjdump(CUctx_st *context); void cuobjdumpRegisterFatBinary(unsigned int handle, const char* filename, CUctx_st *context); + kernel_info_t *gpgpu_cuda_ptx_sim_init_grid( const char *kernel_key, + gpgpu_ptx_sim_arg_list_t args, + struct dim3 gridDim, + struct dim3 blockDim, + struct CUctx_st* context ); + }; #endif /* __gpgpu_context_h__ */ -- cgit v1.3 From 7e286ec0ffc963d307551caced7f4a52241dced4 Mon Sep 17 00:00:00 2001 From: Mengchi Zhang Date: Thu, 6 Jun 2019 12:53:34 -0400 Subject: Move pinned_memory etc Signed-off-by: Mengchi Zhang --- libcuda/cuda_runtime_api.cc | 80 +++++++++++++++++++++++++++++---------------- libcuda/gpgpu_context.h | 3 ++ 2 files changed, 54 insertions(+), 29 deletions(-) (limited to 'libcuda/cuda_runtime_api.cc') diff --git a/libcuda/cuda_runtime_api.cc b/libcuda/cuda_runtime_api.cc index 02e2b2e..41f27bf 100644 --- a/libcuda/cuda_runtime_api.cc +++ b/libcuda/cuda_runtime_api.cc @@ -313,8 +313,6 @@ struct CUctx_st { return i->second; } - std::map pinned_memory; //support for pinned memories added - std::map pinned_memory_size; int no_of_ptx; typedef struct glbmap_entry glbmap_entry_t; @@ -989,6 +987,27 @@ cudaError_t cudaMallocInternal(void **devPtr, size_t size, gpgpu_context* gpgpu_ } } +cudaError_t cudaMallocHostInternal(void **ptr, size_t size, gpgpu_context* gpgpu_ctx = NULL) +{ + gpgpu_context *ctx; + if (gpgpu_ctx){ + ctx = gpgpu_ctx; + } else { + ctx = GPGPU_Context(); + } + if(g_debug_execution >= 3){ + announce_call(__my_func__); + } + *ptr = malloc(size); + if ( *ptr ) { + //track pinned memory size allocated in the host so that same amount of memory is also allocated in GPU. + ctx->pinned_memory_size[*ptr]=size; + return g_last_cudaError = cudaSuccess; + } else { + return g_last_cudaError = cudaErrorMemoryAllocation; + } +} + cudaError_t cudaHostGetDevicePointerInternal(void **pDevice, void *pHost, unsigned int flags, gpgpu_context* gpgpu_ctx = NULL) { gpgpu_context *ctx; @@ -1008,8 +1027,8 @@ cudaError_t cudaHostGetDevicePointerInternal(void **pDevice, void *pHost, unsign flags=0; CUctx_st* context = GPGPUSim_Context(); gpgpu_t *gpu = context->get_device()->get_gpgpu(); - std::map::const_iterator i = context->pinned_memory_size.find(pHost); - assert(i != context->pinned_memory_size.end()); + std::map::const_iterator i = ctx->pinned_memory_size.find(pHost); + assert(i != ctx->pinned_memory_size.end()); size_t size = i->second; *pDevice = gpu->gpu_malloc(size); if(g_debug_execution >= 3){ @@ -1017,7 +1036,7 @@ cudaError_t cudaHostGetDevicePointerInternal(void **pDevice, void *pHost, unsign ctx->g_mallocPtr_Size[(unsigned long long)*pDevice] = size; } if ( *pDevice ) { - context->pinned_memory[pHost]=pDevice; + ctx->pinned_memory[pHost]=pDevice; //Copy contents in cpu to gpu gpu->memcpy_to_gpu((size_t)*pDevice,pHost,size); return g_last_cudaError = cudaSuccess; @@ -1066,6 +1085,31 @@ cuLinkAddFileInternal(CUlinkState state, CUjitInputType type, const char *path, } #endif +#if (CUDART_VERSION >= 2010) + +cudaError_t cudaHostAllocInternal(void **pHost, size_t bytes, unsigned int flags, gpgpu_context* gpgpu_ctx = NULL) +{ + gpgpu_context *ctx; + if (gpgpu_ctx){ + ctx = gpgpu_ctx; + } else { + ctx = GPGPU_Context(); + } + if(g_debug_execution >= 3){ + announce_call(__my_func__); + } + *pHost = malloc(bytes); + //need to track the size allocated so that cudaHostGetDevicePointer() can function properly. + //TODO: vary this function behavior based on flags value (following nvidia documentation) + ctx->pinned_memory_size[*pHost]=bytes; + if( *pHost ) + return g_last_cudaError = cudaSuccess; + else + return g_last_cudaError = cudaErrorMemoryAllocation; +} + +#endif + /******************************************************************************* * * * * @@ -1091,18 +1135,7 @@ __host__ cudaError_t CUDARTAPI cudaMalloc(void **devPtr, size_t size) __host__ cudaError_t CUDARTAPI cudaMallocHost(void **ptr, size_t size) { - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - CUctx_st* context = GPGPUSim_Context(); - *ptr = malloc(size); - if ( *ptr ) { - //track pinned memory size allocated in the host so that same amount of memory is also allocated in GPU. - context->pinned_memory_size[*ptr]=size; - return g_last_cudaError = cudaSuccess; - } else { - return g_last_cudaError = cudaErrorMemoryAllocation; - } + return cudaMallocHostInternal(ptr, size); } __host__ cudaError_t CUDARTAPI cudaMallocPitch(void **devPtr, size_t *pitch, size_t width, size_t height) { @@ -3141,18 +3174,7 @@ cudaError_t cudaGLUnregisterBufferObject(GLuint bufferObj) cudaError_t CUDARTAPI cudaHostAlloc(void **pHost, size_t bytes, unsigned int flags) { - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - *pHost = malloc(bytes); - //need to track the size allocated so that cudaHostGetDevicePointer() can function properly. - //TODO: vary this function behavior based on flags value (following nvidia documentation) - CUctx_st* context = GPGPUSim_Context(); - context->pinned_memory_size[*pHost]=bytes; - if( *pHost ) - return g_last_cudaError = cudaSuccess; - else - return g_last_cudaError = cudaErrorMemoryAllocation; + return cudaHostAllocInternal(pHost, bytes, flags); } cudaError_t CUDARTAPI cudaHostGetDevicePointer(void **pDevice, void *pHost, unsigned int flags) diff --git a/libcuda/gpgpu_context.h b/libcuda/gpgpu_context.h index f29e2e0..6878d5c 100644 --- a/libcuda/gpgpu_context.h +++ b/libcuda/gpgpu_context.h @@ -28,6 +28,9 @@ class gpgpu_context { std::map g_mallocPtr_Size; //maps sm version number to set of filenames std::map > version_filename; + std::map pinned_memory; //support for pinned memories added + std::map pinned_memory_size; + // objects pointers for each file cuda_runtime_api* api; // member function list void cuobjdumpInit(); -- cgit v1.3 From ba8374c72558e4b89a0bddc973bcc87a10e2ab5f Mon Sep 17 00:00:00 2001 From: Mengchi Zhang Date: Thu, 6 Jun 2019 13:20:29 -0400 Subject: Move g_glbmap Signed-off-by: Mengchi Zhang --- libcuda/cuda_runtime_api.cc | 134 ++++++++++++++++++++++---------------------- libcuda/gpgpu_context.h | 15 +++++ 2 files changed, 82 insertions(+), 67 deletions(-) (limited to 'libcuda/cuda_runtime_api.cc') diff --git a/libcuda/cuda_runtime_api.cc b/libcuda/cuda_runtime_api.cc index 41f27bf..18a9abb 100644 --- a/libcuda/cuda_runtime_api.cc +++ b/libcuda/cuda_runtime_api.cc @@ -237,17 +237,6 @@ private: struct _cuda_device_id *m_next; }; -#ifndef OPENGL_SUPPORT -typedef unsigned long GLuint; -#endif - -struct glbmap_entry { - GLuint m_bufferObj; - void *m_devPtr; - size_t m_size; - struct glbmap_entry *m_next; -}; - struct CUctx_st { CUctx_st( _cuda_device_id *gpu ) { @@ -255,7 +244,6 @@ struct CUctx_st { m_binary_info.cmem = 0; m_binary_info.gmem = 0; no_of_ptx=0; - g_glbmap = NULL; } _cuda_device_id *get_device() { return m_gpu; } @@ -314,9 +302,6 @@ struct CUctx_st { } int no_of_ptx; - typedef struct glbmap_entry glbmap_entry_t; - - glbmap_entry_t* g_glbmap; private: _cuda_device_id *m_gpu; // selected gpu @@ -1044,6 +1029,72 @@ cudaError_t cudaHostGetDevicePointerInternal(void **pDevice, void *pHost, unsign return g_last_cudaError = cudaErrorMemoryAllocation; } } + +cudaError_t cudaGLMapBufferObjectInternal(void** devPtr, GLuint bufferObj, gpgpu_context* gpgpu_ctx = NULL) +{ + gpgpu_context *ctx; + if (gpgpu_ctx){ + ctx = gpgpu_ctx; + } else { + ctx = GPGPU_Context(); + } + if(g_debug_execution >= 3){ + announce_call(__my_func__); + } + if(g_debug_execution >= 3){ + announce_call(__my_func__); + } +#ifdef OPENGL_SUPPORT + GLint buffer_size=0; + CUctx_st* context = GPGPUSim_Context(); + + glbmap_entry_t *p = ctx->g_glbmap; + while ( p && p->m_bufferObj != bufferObj ) + p = p->m_next; + if ( p == NULL ) { + glBindBuffer(GL_ARRAY_BUFFER,bufferObj); + glGetBufferParameteriv(GL_ARRAY_BUFFER,GL_BUFFER_SIZE,&buffer_size); + assert( buffer_size != 0 ); + *devPtr = context->get_device()->get_gpgpu()->gpu_malloc(buffer_size); + + // create entry and insert to front of list + glbmap_entry_t *n = (glbmap_entry_t *) calloc(1,sizeof(glbmap_entry_t)); + n->m_next = ctx->g_glbmap; + ctx->g_glbmap = n; + + // initialize entry + n->m_bufferObj = bufferObj; + n->m_devPtr = *devPtr; + n->m_size = buffer_size; + + p = n; + } else { + buffer_size = p->m_size; + *devPtr = p->m_devPtr; + } + + if ( *devPtr ) { + char *data = (char *) calloc(p->m_size,1); + glGetBufferSubData(GL_ARRAY_BUFFER,0,buffer_size,data); + memcpy_to_gpu( (size_t) *devPtr, data, buffer_size ); + free(data); + printf("GPGPU-Sim PTX: cudaGLMapBufferObject %zu bytes starting at 0x%llx..\n", (size_t)buffer_size, + (unsigned long long) *devPtr); + return g_last_cudaError = cudaSuccess; + } else { + return g_last_cudaError = cudaErrorMemoryAllocation; + } + + return g_last_cudaError = cudaSuccess; +#else + fflush(stdout); + fflush(stderr); + printf("GPGPU-Sim PTX: GPGPU-Sim support for OpenGL integration disabled -- exiting\n"); + fflush(stdout); + exit(50); +#endif +} + #if CUDART_VERSION >= 6050 CUresult cuLinkAddFileInternal(CUlinkState state, CUjitInputType type, const char *path, @@ -3079,58 +3130,7 @@ cudaError_t cudaGLRegisterBufferObject(GLuint bufferObj) cudaError_t cudaGLMapBufferObject(void** devPtr, GLuint bufferObj) { - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } -#ifdef OPENGL_SUPPORT - GLint buffer_size=0; - CUctx_st* ctx = GPGPUSim_Context(); - - glbmap_entry_t *p = ctx->g_glbmap; - while ( p && p->m_bufferObj != bufferObj ) - p = p->m_next; - if ( p == NULL ) { - glBindBuffer(GL_ARRAY_BUFFER,bufferObj); - glGetBufferParameteriv(GL_ARRAY_BUFFER,GL_BUFFER_SIZE,&buffer_size); - assert( buffer_size != 0 ); - *devPtr = ctx->get_device()->get_gpgpu()->gpu_malloc(buffer_size); - - // create entry and insert to front of list - glbmap_entry_t *n = (glbmap_entry_t *) calloc(1,sizeof(glbmap_entry_t)); - n->m_next = ctx->g_glbmap; - ctx->g_glbmap = n; - - // initialize entry - n->m_bufferObj = bufferObj; - n->m_devPtr = *devPtr; - n->m_size = buffer_size; - - p = n; - } else { - buffer_size = p->m_size; - *devPtr = p->m_devPtr; - } - - if ( *devPtr ) { - char *data = (char *) calloc(p->m_size,1); - glGetBufferSubData(GL_ARRAY_BUFFER,0,buffer_size,data); - memcpy_to_gpu( (size_t) *devPtr, data, buffer_size ); - free(data); - printf("GPGPU-Sim PTX: cudaGLMapBufferObject %zu bytes starting at 0x%llx..\n", (size_t)buffer_size, - (unsigned long long) *devPtr); - return g_last_cudaError = cudaSuccess; - } else { - return g_last_cudaError = cudaErrorMemoryAllocation; - } - - return g_last_cudaError = cudaSuccess; -#else - fflush(stdout); - fflush(stderr); - printf("GPGPU-Sim PTX: GPGPU-Sim support for OpenGL integration disabled -- exiting\n"); - fflush(stdout); - exit(50); -#endif + return cudaGLMapBufferObjectInternal(devPtr, bufferObj); } cudaError_t cudaGLUnmapBufferObject(GLuint bufferObj) diff --git a/libcuda/gpgpu_context.h b/libcuda/gpgpu_context.h index 6878d5c..7569ea6 100644 --- a/libcuda/gpgpu_context.h +++ b/libcuda/gpgpu_context.h @@ -15,10 +15,24 @@ class kernel_info_t; typedef std::list gpgpu_ptx_sim_arg_list_t; +#ifndef OPENGL_SUPPORT +typedef unsigned long GLuint; +#endif + +struct glbmap_entry { + GLuint m_bufferObj; + void *m_devPtr; + size_t m_size; + struct glbmap_entry *m_next; +}; + +typedef struct glbmap_entry glbmap_entry_t; + class gpgpu_context { public: gpgpu_context() { api = new cuda_runtime_api(); + g_glbmap = NULL; } // global list std::list cuobjdumpSectionList; @@ -30,6 +44,7 @@ class gpgpu_context { std::map > version_filename; std::map pinned_memory; //support for pinned memories added std::map pinned_memory_size; + glbmap_entry_t* g_glbmap; // objects pointers for each file cuda_runtime_api* api; // member function list -- cgit v1.3 From 87409183125b863be0c3e0470b2417c3c92a748b Mon Sep 17 00:00:00 2001 From: Mengchi Zhang Date: Sun, 9 Jun 2019 02:47:39 -0400 Subject: Move some vars back to cuda_api_object Signed-off-by: Mengchi Zhang --- libcuda/cuda_api_object.h | 54 +++++++++++++++++++++++++++++++++++++++ libcuda/cuda_runtime_api.cc | 62 ++++++++++++++++++++++----------------------- libcuda/gpgpu_context.h | 53 -------------------------------------- 3 files changed, 85 insertions(+), 84 deletions(-) (limited to 'libcuda/cuda_runtime_api.cc') diff --git a/libcuda/cuda_api_object.h b/libcuda/cuda_api_object.h index 2001f91..41337c6 100644 --- a/libcuda/cuda_api_object.h +++ b/libcuda/cuda_api_object.h @@ -1,14 +1,68 @@ #ifndef __cuda_api_object_h__ #define __cuda_api_object_h__ + +#include +#include +#include +#include + class cuobjdumpSection; +class cuobjdumpELFSection; +class cuobjdumpPTXSection; +class symbol_table; +class gpgpu_ptx_sim_arg; class kernel_config; +class kernel_info_t; + +typedef std::list gpgpu_ptx_sim_arg_list_t; + +#ifndef OPENGL_SUPPORT +typedef unsigned long GLuint; +#endif + +struct glbmap_entry { + GLuint m_bufferObj; + void *m_devPtr; + size_t m_size; + struct glbmap_entry *m_next; +}; + +typedef struct glbmap_entry glbmap_entry_t; class cuda_runtime_api { public: + cuda_runtime_api() { + g_glbmap = NULL; + } // global list + std::list cuobjdumpSectionList; std::list libSectionList; std::list g_cuda_launch_stack; + std::mapfatbin_registered; + std::map fatbinmap; + std::map name_symtab; + std::map g_mallocPtr_Size; + //maps sm version number to set of filenames + std::map > version_filename; + std::map pinned_memory; //support for pinned memories added + std::map pinned_memory_size; + glbmap_entry_t* g_glbmap; // member function list + void cuobjdumpInit(); + void extract_code_using_cuobjdump(); + void extract_ptx_files_using_cuobjdump(CUctx_st *context); + void cuobjdumpParseBinary(unsigned int handle); + std::list pruneSectionList(CUctx_st *context); + std::list mergeMatchingSections(std::string identifier); + std::list mergeSections(); + cuobjdumpELFSection* findELFSection(const std::string identifier); + cuobjdumpPTXSection* findPTXSection(const std::string identifier); + void cuobjdumpRegisterFatBinary(unsigned int handle, const char* filename, CUctx_st *context); + kernel_info_t *gpgpu_cuda_ptx_sim_init_grid( const char *kernel_key, + gpgpu_ptx_sim_arg_list_t args, + struct dim3 gridDim, + struct dim3 blockDim, + struct CUctx_st* context ); }; #endif /* __cuda_api_object_h__ */ diff --git a/libcuda/cuda_runtime_api.cc b/libcuda/cuda_runtime_api.cc index 18a9abb..72ab002 100644 --- a/libcuda/cuda_runtime_api.cc +++ b/libcuda/cuda_runtime_api.cc @@ -626,7 +626,7 @@ static int get_app_cuda_version() { } //! Keep track of the association between filename and cubin handle -void gpgpu_context::cuobjdumpRegisterFatBinary(unsigned int handle, const char* filename, CUctx_st *context){ +void cuda_runtime_api::cuobjdumpRegisterFatBinary(unsigned int handle, const char* filename, CUctx_st *context){ fatbinmap[handle] = filename; } @@ -703,8 +703,8 @@ void** cudaRegisterFatBinaryInternal( void *fatCubin, gpgpu_context* gpgpu_ctx = * then for next calls, only returns the appropriate number */ assert(fat_cubin_handle >= 1); - if (fat_cubin_handle==1) ctx->cuobjdumpInit(); - ctx->cuobjdumpRegisterFatBinary(fat_cubin_handle, filename, context); + if (fat_cubin_handle==1) ctx->api->cuobjdumpInit(); + ctx->api->cuobjdumpRegisterFatBinary(fat_cubin_handle, filename, context); return (void**)fat_cubin_handle; } @@ -801,7 +801,7 @@ void cudaRegisterFunctionInternal( printf("GPGPU-Sim PTX: __cudaRegisterFunction %s : hostFun 0x%p, fat_cubin_handle = %u\n", deviceFun, hostFun, fat_cubin_handle); if(context->get_device()->get_gpgpu()->get_config().use_cuobjdump()) - ctx->cuobjdumpParseBinary(fat_cubin_handle); + ctx->api->cuobjdumpParseBinary(fat_cubin_handle); context->register_function( fat_cubin_handle, hostFun, deviceFun ); } @@ -828,7 +828,7 @@ void cudaRegisterVarInternal( printf("GPGPU-Sim PTX: __cudaRegisterVar: hostVar = %p; deviceAddress = %s; deviceName = %s\n", hostVar, deviceAddress, deviceName); printf("GPGPU-Sim PTX: __cudaRegisterVar: Registering const memory space of %d bytes\n", size); if(GPGPUSim_Context()->get_device()->get_gpgpu()->get_config().use_cuobjdump()) - ctx->cuobjdumpParseBinary((unsigned)(unsigned long long)fatCubinHandle); + ctx->api->cuobjdumpParseBinary((unsigned)(unsigned long long)fatCubinHandle); fflush(stdout); if ( constant && !global && !ext ) { gpgpu_ptx_sim_register_const_variable(hostVar,deviceName,size); @@ -892,7 +892,7 @@ cudaError_t cudaLaunchInternal( const char *hostFun, gpgpu_context* gpgpu_ctx = struct CUstream_st *stream = config.get_stream(); printf("\nGPGPU-Sim PTX: cudaLaunch for 0x%p (mode=%s) on stream %u\n", hostFun, g_ptx_sim_mode?"functional simulation":"performance simulation", stream?stream->get_uid():0 ); - kernel_info_t *grid = ctx->gpgpu_cuda_ptx_sim_init_grid(hostFun,config.get_args(),config.grid_dim(),config.block_dim(),context); + kernel_info_t *grid = ctx->api->gpgpu_cuda_ptx_sim_init_grid(hostFun,config.get_args(),config.grid_dim(),config.block_dim(),context); //do dynamic PDOM analysis for performance simulation scenario std::string kname = grid->name(); function_info *kernel_func_info = grid->entry(); @@ -963,7 +963,7 @@ cudaError_t cudaMallocInternal(void **devPtr, size_t size, gpgpu_context* gpgpu_ *devPtr = context->get_device()->get_gpgpu()->gpu_malloc(size); if(g_debug_execution >= 3){ printf("GPGPU-Sim PTX: cudaMallocing %zu bytes starting at 0x%llx..\n",size, (unsigned long long) *devPtr); - ctx->g_mallocPtr_Size[(unsigned long long)*devPtr] = size; + ctx->api->g_mallocPtr_Size[(unsigned long long)*devPtr] = size; } if ( *devPtr ) { return g_last_cudaError = cudaSuccess; @@ -986,7 +986,7 @@ cudaError_t cudaMallocHostInternal(void **ptr, size_t size, gpgpu_context* gpgpu *ptr = malloc(size); if ( *ptr ) { //track pinned memory size allocated in the host so that same amount of memory is also allocated in GPU. - ctx->pinned_memory_size[*ptr]=size; + ctx->api->pinned_memory_size[*ptr]=size; return g_last_cudaError = cudaSuccess; } else { return g_last_cudaError = cudaErrorMemoryAllocation; @@ -1012,16 +1012,16 @@ cudaError_t cudaHostGetDevicePointerInternal(void **pDevice, void *pHost, unsign flags=0; CUctx_st* context = GPGPUSim_Context(); gpgpu_t *gpu = context->get_device()->get_gpgpu(); - std::map::const_iterator i = ctx->pinned_memory_size.find(pHost); - assert(i != ctx->pinned_memory_size.end()); + std::map::const_iterator i = ctx->api->pinned_memory_size.find(pHost); + assert(i != ctx->api->pinned_memory_size.end()); size_t size = i->second; *pDevice = gpu->gpu_malloc(size); if(g_debug_execution >= 3){ printf("GPGPU-Sim PTX: cudaMallocing %zu bytes starting at 0x%llx..\n",size, (unsigned long long) *pDevice); - ctx->g_mallocPtr_Size[(unsigned long long)*pDevice] = size; + ctx->api->g_mallocPtr_Size[(unsigned long long)*pDevice] = size; } if ( *pDevice ) { - ctx->pinned_memory[pHost]=pDevice; + ctx->api->pinned_memory[pHost]=pDevice; //Copy contents in cpu to gpu gpu->memcpy_to_gpu((size_t)*pDevice,pHost,size); return g_last_cudaError = cudaSuccess; @@ -1048,7 +1048,7 @@ cudaError_t cudaGLMapBufferObjectInternal(void** devPtr, GLuint bufferObj, gpgpu GLint buffer_size=0; CUctx_st* context = GPGPUSim_Context(); - glbmap_entry_t *p = ctx->g_glbmap; + glbmap_entry_t *p = ctx->api->g_glbmap; while ( p && p->m_bufferObj != bufferObj ) p = p->m_next; if ( p == NULL ) { @@ -1059,8 +1059,8 @@ cudaError_t cudaGLMapBufferObjectInternal(void** devPtr, GLuint bufferObj, gpgpu // create entry and insert to front of list glbmap_entry_t *n = (glbmap_entry_t *) calloc(1,sizeof(glbmap_entry_t)); - n->m_next = ctx->g_glbmap; - ctx->g_glbmap = n; + n->m_next = ctx->api->g_glbmap; + ctx->api->g_glbmap = n; // initialize entry n->m_bufferObj = bufferObj; @@ -1127,7 +1127,7 @@ cuLinkAddFileInternal(CUlinkState state, CUjitInputType type, const char *path, strcat(file,path); symbol_table *symtab = gpgpu_ptx_sim_load_ptx_from_filename( file ); std::string fname(path); - ctx->name_symtab[fname] = symtab; + ctx->api->name_symtab[fname] = symtab; context->add_binary(symtab, 1); load_static_globals(symtab,STATIC_ALLOC_LIMIT,0xFFFFFFFF,context->get_device()->get_gpgpu()); load_constants(symtab,STATIC_ALLOC_LIMIT,context->get_device()->get_gpgpu()); @@ -1152,7 +1152,7 @@ cudaError_t cudaHostAllocInternal(void **pHost, size_t bytes, unsigned int flag *pHost = malloc(bytes); //need to track the size allocated so that cudaHostGetDevicePointer() can function properly. //TODO: vary this function behavior based on flags value (following nvidia documentation) - ctx->pinned_memory_size[*pHost]=bytes; + ctx->api->pinned_memory_size[*pHost]=bytes; if( *pHost ) return g_last_cudaError = cudaSuccess; else @@ -2428,7 +2428,7 @@ __host__ cudaError_t CUDARTAPI cudaGetExportTable(const void **ppExportTable, co //#include "../../cuobjdump_to_ptxplus/cuobjdump_parser.h" //extracts all ptx files from binary and dumps into prog_name.unique_no.sm_<>.ptx files -void gpgpu_context::extract_ptx_files_using_cuobjdump(CUctx_st *context){ +void cuda_runtime_api::extract_ptx_files_using_cuobjdump(CUctx_st *context){ extern bool g_cdp_enabled; char command[1000]; char *pytorch_bin = getenv("PYTORCH_BIN"); @@ -2506,7 +2506,7 @@ void gpgpu_context::extract_ptx_files_using_cuobjdump(CUctx_st *context){ * It is also responsible for extracting the libraries linked to the binary if the option is * enabled * */ -void gpgpu_context::extract_code_using_cuobjdump(){ +void cuda_runtime_api::extract_code_using_cuobjdump(){ CUctx_st *context = GPGPUSim_Context(); unsigned forced_max_capability = context->get_device()->get_gpgpu()->get_config().get_forced_max_capability(); @@ -2630,7 +2630,7 @@ void gpgpu_context::extract_code_using_cuobjdump(){ fclose(cuobjdump_in); std::getline(libsf, line); } - api->libSectionList = cuobjdumpSectionList; + libSectionList = cuobjdumpSectionList; //Restore the original section list cuobjdumpSectionList = tmpsl; @@ -2676,7 +2676,7 @@ void printSectionList(std::list sl) { } //! Remove unecessary sm versions from the section list -std::list gpgpu_context::pruneSectionList(CUctx_st *context) { +std::list cuda_runtime_api::pruneSectionList(CUctx_st *context) { unsigned forced_max_capability = context->get_device()->get_gpgpu()->get_config().get_forced_max_capability(); //For ptxplus, force the max capability to 19 if it's higher or unspecified(0) @@ -2729,7 +2729,7 @@ std::list gpgpu_context::pruneSectionList(CUctx_st *context) } //! Merge all PTX sections that have a specific identifier into one file -std::list gpgpu_context::mergeMatchingSections(std::string identifier){ +std::list cuda_runtime_api::mergeMatchingSections(std::string identifier){ const char *ptxcode = ""; std::list::iterator old_iter; cuobjdumpPTXSection* old_ptxsection = NULL; @@ -2772,7 +2772,7 @@ std::list gpgpu_context::mergeMatchingSections(std::string id } //! Merge any PTX sections with matching identifiers -std::list gpgpu_context::mergeSections(){ +std::list cuda_runtime_api::mergeSections(){ std::vector identifier; cuobjdumpPTXSection* ptxsection; @@ -2819,10 +2819,10 @@ cuobjdumpELFSection* findELFSectionInList(std::list sectionli } //! Find an ELF section in all the known lists -cuobjdumpELFSection* gpgpu_context::findELFSection(const std::string identifier){ +cuobjdumpELFSection* cuda_runtime_api::findELFSection(const std::string identifier){ cuobjdumpELFSection* sec = findELFSectionInList(cuobjdumpSectionList, identifier); if (sec!=NULL)return sec; - sec = findELFSectionInList(api->libSectionList, identifier); + sec = findELFSectionInList(libSectionList, identifier); if (sec!=NULL)return sec; std::cout << "Could not find " << identifier << std::endl; assert(0 && "Could not find the required ELF section"); @@ -2854,10 +2854,10 @@ cuobjdumpPTXSection* findPTXSectionInList(std::list §ionl } //! Find an PTX section in all the known lists -cuobjdumpPTXSection* gpgpu_context::findPTXSection(const std::string identifier){ +cuobjdumpPTXSection* cuda_runtime_api::findPTXSection(const std::string identifier){ cuobjdumpPTXSection* sec = findPTXSectionInList(cuobjdumpSectionList, identifier); if (sec!=NULL)return sec; - sec = findPTXSectionInList(api->libSectionList, identifier); + sec = findPTXSectionInList(libSectionList, identifier); if (sec!=NULL)return sec; std::cout << "Could not find " << identifier << std::endl; assert(0 && "Could not find the required PTX section"); @@ -2867,7 +2867,7 @@ cuobjdumpPTXSection* gpgpu_context::findPTXSection(const std::string identifier) //! Extract the code using cuobjdump and remove unnecessary sections -void gpgpu_context::cuobjdumpInit(){ +void cuda_runtime_api::cuobjdumpInit(){ CUctx_st *context = GPGPUSim_Context(); extract_code_using_cuobjdump(); //extract all the output of cuobjdump to _cuobjdump_*.* const char* pre_load = getenv("CUOBJDUMP_SIM_FILE"); @@ -2879,7 +2879,7 @@ void gpgpu_context::cuobjdumpInit(){ //! Either submit PTX for simulation or convert SASS to PTXPlus and submit it -void gpgpu_context::cuobjdumpParseBinary(unsigned int handle){ +void cuda_runtime_api::cuobjdumpParseBinary(unsigned int handle){ CUctx_st *context = GPGPUSim_Context(); if(fatbin_registered[handle]) return; @@ -3140,7 +3140,7 @@ cudaError_t cudaGLUnmapBufferObject(GLuint bufferObj) } #ifdef OPENGL_SUPPORT CUctx_st* ctx = GPGPUSim_Context(); - glbmap_entry_t *p = ctx->g_glbmap; + glbmap_entry_t *p = ctx->api->g_glbmap; while ( p && p->m_bufferObj != bufferObj ) p = p->m_next; if ( p == NULL ) @@ -3495,7 +3495,7 @@ static int load_constants( symbol_table *symtab, addr_t min_gaddr, gpgpu_t *gpu return nc_bytes; } -kernel_info_t * gpgpu_context::gpgpu_cuda_ptx_sim_init_grid( const char *hostFun, +kernel_info_t * cuda_runtime_api::gpgpu_cuda_ptx_sim_init_grid( const char *hostFun, gpgpu_ptx_sim_arg_list_t args, struct dim3 gridDim, struct dim3 blockDim, diff --git a/libcuda/gpgpu_context.h b/libcuda/gpgpu_context.h index 7569ea6..4622f00 100644 --- a/libcuda/gpgpu_context.h +++ b/libcuda/gpgpu_context.h @@ -1,68 +1,15 @@ #ifndef __gpgpu_context_h__ #define __gpgpu_context_h__ -#include -#include -#include -#include #include "cuda_api_object.h" -class cuobjdumpSection; -class cuobjdumpELFSection; -class cuobjdumpPTXSection; -class symbol_table; -class gpgpu_ptx_sim_arg; -class kernel_info_t; - -typedef std::list gpgpu_ptx_sim_arg_list_t; - -#ifndef OPENGL_SUPPORT -typedef unsigned long GLuint; -#endif - -struct glbmap_entry { - GLuint m_bufferObj; - void *m_devPtr; - size_t m_size; - struct glbmap_entry *m_next; -}; - -typedef struct glbmap_entry glbmap_entry_t; - class gpgpu_context { public: gpgpu_context() { api = new cuda_runtime_api(); - g_glbmap = NULL; } // global list - std::list cuobjdumpSectionList; - std::mapfatbin_registered; - std::map fatbinmap; - std::map name_symtab; - std::map g_mallocPtr_Size; - //maps sm version number to set of filenames - std::map > version_filename; - std::map pinned_memory; //support for pinned memories added - std::map pinned_memory_size; - glbmap_entry_t* g_glbmap; // objects pointers for each file cuda_runtime_api* api; // member function list - void cuobjdumpInit(); - void cuobjdumpParseBinary(unsigned int handle); - void extract_code_using_cuobjdump(); - std::list pruneSectionList(CUctx_st *context); - std::list mergeMatchingSections(std::string identifier); - std::list mergeSections(); - cuobjdumpELFSection* findELFSection(const std::string identifier); - cuobjdumpPTXSection* findPTXSection(const std::string identifier); - void extract_ptx_files_using_cuobjdump(CUctx_st *context); - void cuobjdumpRegisterFatBinary(unsigned int handle, const char* filename, CUctx_st *context); - kernel_info_t *gpgpu_cuda_ptx_sim_init_grid( const char *kernel_key, - gpgpu_ptx_sim_arg_list_t args, - struct dim3 gridDim, - struct dim3 blockDim, - struct CUctx_st* context ); - }; #endif /* __gpgpu_context_h__ */ -- cgit v1.3 From c7f515f6f5325c65f32dd64e1ad479660c751e99 Mon Sep 17 00:00:00 2001 From: tgrogers Date: Sun, 9 Jun 2019 22:20:15 -0400 Subject: A bunch of boilerbplate to get 10.1 to compile. Still does not yet run. The way CUDA calls kerenels (even on old code) has changed. --- Makefile | 2 ++ libcuda/cuda_api.h | 2 ++ libcuda/cuda_runtime_api.cc | 25 +++++++++++++++++++++++++ linux-so-version.txt | 4 ++++ setup_environment | 2 +- 5 files changed, 34 insertions(+), 1 deletion(-) (limited to 'libcuda/cuda_runtime_api.cc') diff --git a/Makefile b/Makefile index 3db8ce8..a69130c 100644 --- a/Makefile +++ b/Makefile @@ -164,6 +164,8 @@ $(SIM_LIB_DIR)/libcudart.so: makedirs $(LIBS) cudalib if [ ! -f $(SIM_LIB_DIR)/libcudart.so.9.0 ]; then ln -s libcudart.so $(SIM_LIB_DIR)/libcudart.so.9.0; fi if [ ! -f $(SIM_LIB_DIR)/libcudart.so.9.1 ]; then ln -s libcudart.so $(SIM_LIB_DIR)/libcudart.so.9.1; fi if [ ! -f $(SIM_LIB_DIR)/libcudart.so.9.2 ]; then ln -s libcudart.so $(SIM_LIB_DIR)/libcudart.so.9.2; fi + if [ ! -f $(SIM_LIB_DIR)/libcudart.so.10.0 ]; then ln -s libcudart.so $(SIM_LIB_DIR)/libcudart.so.10.0; fi + if [ ! -f $(SIM_LIB_DIR)/libcudart.so.10.1 ]; then ln -s libcudart.so $(SIM_LIB_DIR)/libcudart.so.10.1; fi $(SIM_LIB_DIR)/libcudart.dylib: makedirs $(LIBS) cudalib g++ -dynamiclib -Wl,-headerpad_max_install_names,-undefined,dynamic_lookup,-compatibility_version,1.1,-current_version,1.1\ diff --git a/libcuda/cuda_api.h b/libcuda/cuda_api.h index 3808e8a..7ee26dc 100644 --- a/libcuda/cuda_api.h +++ b/libcuda/cuda_api.h @@ -234,9 +234,11 @@ typedef struct CUgraphicsResource_st *CUgraphicsResource; /**< CUDA graphics int typedef unsigned long long CUtexObject; /**< An opaque value that represents a CUDA texture object */ typedef unsigned long long CUsurfObject; /**< An opaque value that represents a CUDA surface object */ +#if __CUDA_API_VERSION < 1010 typedef struct CUuuid_st { /**< CUDA definition of UUID */ char bytes[16]; } CUuuid; +#endif #if __CUDA_API_VERSION >= 4010 diff --git a/libcuda/cuda_runtime_api.cc b/libcuda/cuda_runtime_api.cc index 18a9abb..718db49 100644 --- a/libcuda/cuda_runtime_api.cc +++ b/libcuda/cuda_runtime_api.cc @@ -630,6 +630,8 @@ void gpgpu_context::cuobjdumpRegisterFatBinary(unsigned int handle, const char* fatbinmap[handle] = filename; } + + /******************************************************************************* * Add internal cuda runtime API call to accept gpgpu_context * *******************************************************************************/ @@ -2975,6 +2977,29 @@ void** CUDARTAPI __cudaRegisterFatBinary( void *fatCubin ) return cudaRegisterFatBinaryInternal(fatCubin); } +void CUDARTAPI __cudaRegisterFatBinaryEnd( void **fatCubinHandle ) +{ + +} + +unsigned CUDARTAPI __cudaPushCallConfiguration(dim3 gridDim, + dim3 blockDim, + size_t sharedMem = 0, + struct CUstream_st *stream = 0) +{ + +} + +cudaError_t CUDARTAPI __cudaPopCallConfiguration( + dim3 *gridDim, + dim3 *blockDim, + size_t *sharedMem, + void *stream +) +{ + return g_last_cudaError = cudaSuccess; +} + void CUDARTAPI __cudaRegisterFunction( void **fatCubinHandle, const char *hostFun, diff --git a/linux-so-version.txt b/linux-so-version.txt index a7c2d3c..45c40dd 100644 --- a/linux-so-version.txt +++ b/linux-so-version.txt @@ -4,5 +4,9 @@ libcudart.so.9.1{ }; libcudart.so.9.2{ }; +libcudart.so.10.0{ +}; +libcudart.so.10.1{ +}; libcuda.so.1{ }; diff --git a/setup_environment b/setup_environment index b420584..17891ce 100644 --- a/setup_environment +++ b/setup_environment @@ -51,7 +51,7 @@ CC_VERSION=`gcc --version | head -1 | awk '{for(i=1;i<=NF;i++){ if(match($i,/^[0 CUDA_VERSION_STRING=`$CUDA_INSTALL_PATH/bin/nvcc --version | awk '/release/ {print $5;}' | sed 's/,//'`; export CUDA_VERSION_NUMBER=`echo $CUDA_VERSION_STRING | sed 's/\./ /' | awk '{printf("%02u%02u", 10*int($1), 10*$2);}'` -if [ $CUDA_VERSION_NUMBER -gt 9100 -o $CUDA_VERSION_NUMBER -lt 2030 ]; then +if [ $CUDA_VERSION_NUMBER -gt 10100 -o $CUDA_VERSION_NUMBER -lt 2030 ]; then echo "ERROR ** GPGPU-Sim version $GPGPUSIM_VERSION_STRING not tested with CUDA version $CUDA_VERSION_STRING (please see README)"; return fi -- cgit v1.3 From 2ea18072618e7fe4e541f84de7d8575998299a1c Mon Sep 17 00:00:00 2001 From: tgrogers Date: Mon, 10 Jun 2019 09:19:30 -0400 Subject: Code to get CUDA 10 to work - looks like the gridDim/blockDim args given to cudaLaunchKernel are not complete garbage. We must rely on the PushConfig to get the proper sizing info --- libcuda/cuda_runtime_api.cc | 43 +++++++++++++++++++++++++++---------------- 1 file changed, 27 insertions(+), 16 deletions(-) (limited to 'libcuda/cuda_runtime_api.cc') diff --git a/libcuda/cuda_runtime_api.cc b/libcuda/cuda_runtime_api.cc index 718db49..500b5f2 100644 --- a/libcuda/cuda_runtime_api.cc +++ b/libcuda/cuda_runtime_api.cc @@ -2108,20 +2108,21 @@ __host__ cudaError_t CUDARTAPI cudaLaunch( const char *hostFun ) __host__ cudaError_t CUDARTAPI cudaLaunchKernel ( const char* hostFun, dim3 gridDim, dim3 blockDim, const void** args, size_t sharedMem, cudaStream_t stream ) { - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - CUctx_st *context = GPGPUSim_Context(); - function_info *entry = context->get_kernel(hostFun); - - cudaConfigureCallInternal(gridDim, blockDim, sharedMem, stream); - for(unsigned i = 0; i < entry->num_args(); i++){ - std::pair p = entry->get_param_config(i); - cudaSetupArgumentInternal(args[i], p.first, p.second); - } + if(g_debug_execution >= 3){ + announce_call(__my_func__); + } + CUctx_st *context = GPGPUSim_Context(); + function_info *entry = context->get_kernel(hostFun); +#if CUDART_VERSION < 10000 + cudaConfigureCallInternal(gridDim, blockDim, sharedMem, stream); +#endif + for(unsigned i = 0; i < entry->num_args(); i++){ + std::pair p = entry->get_param_config(i); + cudaSetupArgumentInternal(args[i], p.first, p.second); + } - cudaLaunchInternal(hostFun); - return g_last_cudaError = cudaSuccess; + cudaLaunchInternal(hostFun); + return g_last_cudaError = cudaSuccess; } @@ -2974,12 +2975,17 @@ extern "C" { void** CUDARTAPI __cudaRegisterFatBinary( void *fatCubin ) { + if(g_debug_execution >= 3){ + announce_call(__my_func__); + } return cudaRegisterFatBinaryInternal(fatCubin); } void CUDARTAPI __cudaRegisterFatBinaryEnd( void **fatCubinHandle ) { - + if(g_debug_execution >= 3){ + announce_call(__my_func__); + } } unsigned CUDARTAPI __cudaPushCallConfiguration(dim3 gridDim, @@ -2987,7 +2993,10 @@ unsigned CUDARTAPI __cudaPushCallConfiguration(dim3 gridDim, size_t sharedMem = 0, struct CUstream_st *stream = 0) { - + if(g_debug_execution >= 3){ + announce_call(__my_func__); + } + cudaConfigureCallInternal(gridDim, blockDim, sharedMem, stream); } cudaError_t CUDARTAPI __cudaPopCallConfiguration( @@ -2997,6 +3006,9 @@ cudaError_t CUDARTAPI __cudaPopCallConfiguration( void *stream ) { + if(g_debug_execution >= 3){ + announce_call(__my_func__); + } return g_last_cudaError = cudaSuccess; } @@ -3056,7 +3068,6 @@ void __cudaUnregisterFatBinary(void **fatCubinHandle) if(g_debug_execution >= 3){ announce_call(__my_func__); } - ; } cudaError_t cudaDeviceReset ( void ) { -- cgit v1.3 From c29246408c963ece65515fae92540e76ac71b72b Mon Sep 17 00:00:00 2001 From: Mengchi Zhang Date: Mon, 10 Jun 2019 11:17:23 -0400 Subject: Move some struct upward in cuda_runtime_api.cc Signed-off-by: Mengchi Zhang --- libcuda/cuda_api_object.h | 149 +++++++++++++++++++++++++++++++++++++++++--- libcuda/cuda_runtime_api.cc | 141 ----------------------------------------- libcuda/cuobjdump.h | 1 + 3 files changed, 143 insertions(+), 148 deletions(-) (limited to 'libcuda/cuda_runtime_api.cc') diff --git a/libcuda/cuda_api_object.h b/libcuda/cuda_api_object.h index 41337c6..73c077e 100644 --- a/libcuda/cuda_api_object.h +++ b/libcuda/cuda_api_object.h @@ -6,13 +6,10 @@ #include #include -class cuobjdumpSection; -class cuobjdumpELFSection; -class cuobjdumpPTXSection; -class symbol_table; -class gpgpu_ptx_sim_arg; -class kernel_config; -class kernel_info_t; +#include "../src/gpgpu-sim/gpu-sim.h" +#include "../src/cuda-sim/ptx_ir.h" +#include "../src/abstract_hardware_model.h" +#include "cuobjdump.h" typedef std::list gpgpu_ptx_sim_arg_list_t; @@ -29,6 +26,144 @@ struct glbmap_entry { typedef struct glbmap_entry glbmap_entry_t; +struct _cuda_device_id { + _cuda_device_id(gpgpu_sim* gpu) {m_id = 0; m_next = NULL; m_gpgpu=gpu;} + struct _cuda_device_id *next() { return m_next; } + unsigned num_shader() const { return m_gpgpu->get_config().num_shader(); } + int num_devices() const { + if( m_next == NULL ) return 1; + else return 1 + m_next->num_devices(); + } + struct _cuda_device_id *get_device( unsigned n ) + { + assert( n < (unsigned)num_devices() ); + struct _cuda_device_id *p=this; + for(unsigned i=0; im_next; + return p; + } + const struct cudaDeviceProp *get_prop() const + { + return m_gpgpu->get_prop(); + } + unsigned get_id() const { return m_id; } + + gpgpu_sim *get_gpgpu() { return m_gpgpu; } +private: + unsigned m_id; + class gpgpu_sim *m_gpgpu; + struct _cuda_device_id *m_next; +}; + +struct CUctx_st { + CUctx_st( _cuda_device_id *gpu ) + { + m_gpu = gpu; + m_binary_info.cmem = 0; + m_binary_info.gmem = 0; + no_of_ptx=0; + } + + _cuda_device_id *get_device() { return m_gpu; } + + void add_binary( symbol_table *symtab, unsigned fat_cubin_handle ) + { + m_code[fat_cubin_handle] = symtab; + m_last_fat_cubin_handle = fat_cubin_handle; + } + + void add_ptxinfo( const char *deviceFun, const struct gpgpu_ptx_sim_info &info ) + { + symbol *s = m_code[m_last_fat_cubin_handle]->lookup(deviceFun); + assert( s != NULL ); + function_info *f = s->get_pc(); + assert( f != NULL ); + f->set_kernel_info(info); + } + + void add_ptxinfo( const struct gpgpu_ptx_sim_info &info ) + { + m_binary_info = info; + } + + void register_function( unsigned fat_cubin_handle, const char *hostFun, const char *deviceFun ) + { + if( m_code.find(fat_cubin_handle) != m_code.end() ) { + symbol *s = m_code[fat_cubin_handle]->lookup(deviceFun); + if(s != NULL) { + function_info *f = s->get_pc(); + assert( f != NULL ); + m_kernel_lookup[hostFun] = f; + } + else { + printf("Warning: cannot find deviceFun %s\n", deviceFun); + m_kernel_lookup[hostFun] = NULL; + } + // assert( s != NULL ); + // function_info *f = s->get_pc(); + // assert( f != NULL ); + // m_kernel_lookup[hostFun] = f; + } else { + m_kernel_lookup[hostFun] = NULL; + } + } + + void register_hostFun_function( const char*hostFun, function_info* f){ + m_kernel_lookup[hostFun] = f; + } + + function_info *get_kernel(const char *hostFun) + { + std::map::iterator i=m_kernel_lookup.find(hostFun); + assert( i != m_kernel_lookup.end() ); + return i->second; + } + + int no_of_ptx; + +private: + _cuda_device_id *m_gpu; // selected gpu + std::map m_code; // fat binary handle => global symbol table + unsigned m_last_fat_cubin_handle; + std::map m_kernel_lookup; // unique id (CUDA app function address) => kernel entry point + struct gpgpu_ptx_sim_info m_binary_info; + +}; + +class kernel_config { +public: + kernel_config( dim3 GridDim, dim3 BlockDim, size_t sharedMem, struct CUstream_st *stream ) + { + m_GridDim=GridDim; + m_BlockDim=BlockDim; + m_sharedMem=sharedMem; + m_stream = stream; + } + kernel_config() + { + m_GridDim=dim3(-1,-1,-1); + m_BlockDim=dim3(-1,-1,-1); + m_sharedMem=0; + m_stream =NULL; + } + void set_arg( const void *arg, size_t size, size_t offset ) + { + m_args.push_front( gpgpu_ptx_sim_arg(arg,size,offset) ); + } + dim3 grid_dim() const { return m_GridDim; } + dim3 block_dim() const { return m_BlockDim; } + void set_grid_dim(dim3 *d) { m_GridDim = *d; } + void set_block_dim(dim3 *d) { m_BlockDim = *d; } + gpgpu_ptx_sim_arg_list_t get_args() { return m_args; } + struct CUstream_st *get_stream() { return m_stream; } + +private: + dim3 m_GridDim; + dim3 m_BlockDim; + size_t m_sharedMem; + struct CUstream_st *m_stream; + gpgpu_ptx_sim_arg_list_t m_args; +}; class cuda_runtime_api { public: diff --git a/libcuda/cuda_runtime_api.cc b/libcuda/cuda_runtime_api.cc index 72ab002..2e2b50b 100644 --- a/libcuda/cuda_runtime_api.cc +++ b/libcuda/cuda_runtime_api.cc @@ -141,8 +141,6 @@ #include "../src/gpgpusim_entrypoint.h" #include "../src/stream_manager.h" #include "../src/abstract_hardware_model.h" -typedef void * yyscan_t; -#include "cuobjdump.h" #include #include @@ -208,145 +206,6 @@ void register_ptx_function( const char *name, function_info *impl ) # endif #endif -struct _cuda_device_id { - _cuda_device_id(gpgpu_sim* gpu) {m_id = 0; m_next = NULL; m_gpgpu=gpu;} - struct _cuda_device_id *next() { return m_next; } - unsigned num_shader() const { return m_gpgpu->get_config().num_shader(); } - int num_devices() const { - if( m_next == NULL ) return 1; - else return 1 + m_next->num_devices(); - } - struct _cuda_device_id *get_device( unsigned n ) - { - assert( n < (unsigned)num_devices() ); - struct _cuda_device_id *p=this; - for(unsigned i=0; im_next; - return p; - } - const struct cudaDeviceProp *get_prop() const - { - return m_gpgpu->get_prop(); - } - unsigned get_id() const { return m_id; } - - gpgpu_sim *get_gpgpu() { return m_gpgpu; } -private: - unsigned m_id; - class gpgpu_sim *m_gpgpu; - struct _cuda_device_id *m_next; -}; - -struct CUctx_st { - CUctx_st( _cuda_device_id *gpu ) - { - m_gpu = gpu; - m_binary_info.cmem = 0; - m_binary_info.gmem = 0; - no_of_ptx=0; - } - - _cuda_device_id *get_device() { return m_gpu; } - - void add_binary( symbol_table *symtab, unsigned fat_cubin_handle ) - { - m_code[fat_cubin_handle] = symtab; - m_last_fat_cubin_handle = fat_cubin_handle; - } - - void add_ptxinfo( const char *deviceFun, const struct gpgpu_ptx_sim_info &info ) - { - symbol *s = m_code[m_last_fat_cubin_handle]->lookup(deviceFun); - assert( s != NULL ); - function_info *f = s->get_pc(); - assert( f != NULL ); - f->set_kernel_info(info); - } - - void add_ptxinfo( const struct gpgpu_ptx_sim_info &info ) - { - m_binary_info = info; - } - - void register_function( unsigned fat_cubin_handle, const char *hostFun, const char *deviceFun ) - { - if( m_code.find(fat_cubin_handle) != m_code.end() ) { - symbol *s = m_code[fat_cubin_handle]->lookup(deviceFun); - if(s != NULL) { - function_info *f = s->get_pc(); - assert( f != NULL ); - m_kernel_lookup[hostFun] = f; - } - else { - printf("Warning: cannot find deviceFun %s\n", deviceFun); - m_kernel_lookup[hostFun] = NULL; - } - // assert( s != NULL ); - // function_info *f = s->get_pc(); - // assert( f != NULL ); - // m_kernel_lookup[hostFun] = f; - } else { - m_kernel_lookup[hostFun] = NULL; - } - } - - void register_hostFun_function( const char*hostFun, function_info* f){ - m_kernel_lookup[hostFun] = f; - } - - function_info *get_kernel(const char *hostFun) - { - std::map::iterator i=m_kernel_lookup.find(hostFun); - assert( i != m_kernel_lookup.end() ); - return i->second; - } - - int no_of_ptx; - -private: - _cuda_device_id *m_gpu; // selected gpu - std::map m_code; // fat binary handle => global symbol table - unsigned m_last_fat_cubin_handle; - std::map m_kernel_lookup; // unique id (CUDA app function address) => kernel entry point - struct gpgpu_ptx_sim_info m_binary_info; - -}; - -class kernel_config { -public: - kernel_config( dim3 GridDim, dim3 BlockDim, size_t sharedMem, struct CUstream_st *stream ) - { - m_GridDim=GridDim; - m_BlockDim=BlockDim; - m_sharedMem=sharedMem; - m_stream = stream; - } - kernel_config() - { - m_GridDim=dim3(-1,-1,-1); - m_BlockDim=dim3(-1,-1,-1); - m_sharedMem=0; - m_stream =NULL; - } - void set_arg( const void *arg, size_t size, size_t offset ) - { - m_args.push_front( gpgpu_ptx_sim_arg(arg,size,offset) ); - } - dim3 grid_dim() const { return m_GridDim; } - dim3 block_dim() const { return m_BlockDim; } - void set_grid_dim(dim3 *d) { m_GridDim = *d; } - void set_block_dim(dim3 *d) { m_BlockDim = *d; } - gpgpu_ptx_sim_arg_list_t get_args() { return m_args; } - struct CUstream_st *get_stream() { return m_stream; } - -private: - dim3 m_GridDim; - dim3 m_BlockDim; - size_t m_sharedMem; - struct CUstream_st *m_stream; - gpgpu_ptx_sim_arg_list_t m_args; -}; - struct _cuda_device_id *GPGPUSim_Init() { //static _cuda_device_id *the_device = NULL; diff --git a/libcuda/cuobjdump.h b/libcuda/cuobjdump.h index 49af3e2..6ab6778 100644 --- a/libcuda/cuobjdump.h +++ b/libcuda/cuobjdump.h @@ -4,6 +4,7 @@ #include #include +typedef void * yyscan_t; struct cuobjdump_parser { yyscan_t scanner; int elfserial; -- cgit v1.3 From ce1fabd955ee208a38e73d5a5acfba2b26369a3d Mon Sep 17 00:00:00 2001 From: tgrogers Date: Sun, 9 Jun 2019 22:20:15 -0400 Subject: A bunch of boilerbplate to get 10.1 to compile. Still does not yet run. The way CUDA calls kerenels (even on old code) has changed. --- Makefile | 2 ++ libcuda/cuda_api.h | 2 ++ libcuda/cuda_runtime_api.cc | 23 +++++++++++++++++++++++ linux-so-version.txt | 4 ++++ setup_environment | 2 +- 5 files changed, 32 insertions(+), 1 deletion(-) (limited to 'libcuda/cuda_runtime_api.cc') diff --git a/Makefile b/Makefile index e1e9aaa..7a3bb47 100644 --- a/Makefile +++ b/Makefile @@ -165,6 +165,8 @@ $(SIM_LIB_DIR)/libcudart.so: makedirs $(LIBS) cudalib if [ ! -f $(SIM_LIB_DIR)/libcudart.so.9.0 ]; then ln -s libcudart.so $(SIM_LIB_DIR)/libcudart.so.9.0; fi if [ ! -f $(SIM_LIB_DIR)/libcudart.so.9.1 ]; then ln -s libcudart.so $(SIM_LIB_DIR)/libcudart.so.9.1; fi if [ ! -f $(SIM_LIB_DIR)/libcudart.so.9.2 ]; then ln -s libcudart.so $(SIM_LIB_DIR)/libcudart.so.9.2; fi + if [ ! -f $(SIM_LIB_DIR)/libcudart.so.10.0 ]; then ln -s libcudart.so $(SIM_LIB_DIR)/libcudart.so.10.0; fi + if [ ! -f $(SIM_LIB_DIR)/libcudart.so.10.1 ]; then ln -s libcudart.so $(SIM_LIB_DIR)/libcudart.so.10.1; fi $(SIM_LIB_DIR)/libcudart.dylib: makedirs $(LIBS) cudalib g++ -dynamiclib -Wl,-headerpad_max_install_names,-undefined,dynamic_lookup,-compatibility_version,1.1,-current_version,1.1\ diff --git a/libcuda/cuda_api.h b/libcuda/cuda_api.h index 3808e8a..7ee26dc 100644 --- a/libcuda/cuda_api.h +++ b/libcuda/cuda_api.h @@ -234,9 +234,11 @@ typedef struct CUgraphicsResource_st *CUgraphicsResource; /**< CUDA graphics int typedef unsigned long long CUtexObject; /**< An opaque value that represents a CUDA texture object */ typedef unsigned long long CUsurfObject; /**< An opaque value that represents a CUDA surface object */ +#if __CUDA_API_VERSION < 1010 typedef struct CUuuid_st { /**< CUDA definition of UUID */ char bytes[16]; } CUuuid; +#endif #if __CUDA_API_VERSION >= 4010 diff --git a/libcuda/cuda_runtime_api.cc b/libcuda/cuda_runtime_api.cc index 2cc84aa..fb3e07a 100644 --- a/libcuda/cuda_runtime_api.cc +++ b/libcuda/cuda_runtime_api.cc @@ -2763,6 +2763,29 @@ cudaError_t CUDARTAPI cudaDeviceSynchronize(void){ return g_last_cudaError = cudaSuccess; } +void CUDARTAPI __cudaRegisterFatBinaryEnd( void **fatCubinHandle ) +{ + +} + +unsigned CUDARTAPI __cudaPushCallConfiguration(dim3 gridDim, + dim3 blockDim, + size_t sharedMem = 0, + struct CUstream_st *stream = 0) +{ + +} + +cudaError_t CUDARTAPI __cudaPopCallConfiguration( + dim3 *gridDim, + dim3 *blockDim, + size_t *sharedMem, + void *stream +) +{ + return g_last_cudaError = cudaSuccess; +} + void CUDARTAPI __cudaRegisterFunction( void **fatCubinHandle, const char *hostFun, diff --git a/linux-so-version.txt b/linux-so-version.txt index a7c2d3c..45c40dd 100644 --- a/linux-so-version.txt +++ b/linux-so-version.txt @@ -4,5 +4,9 @@ libcudart.so.9.1{ }; libcudart.so.9.2{ }; +libcudart.so.10.0{ +}; +libcudart.so.10.1{ +}; libcuda.so.1{ }; diff --git a/setup_environment b/setup_environment index b420584..17891ce 100644 --- a/setup_environment +++ b/setup_environment @@ -51,7 +51,7 @@ CC_VERSION=`gcc --version | head -1 | awk '{for(i=1;i<=NF;i++){ if(match($i,/^[0 CUDA_VERSION_STRING=`$CUDA_INSTALL_PATH/bin/nvcc --version | awk '/release/ {print $5;}' | sed 's/,//'`; export CUDA_VERSION_NUMBER=`echo $CUDA_VERSION_STRING | sed 's/\./ /' | awk '{printf("%02u%02u", 10*int($1), 10*$2);}'` -if [ $CUDA_VERSION_NUMBER -gt 9100 -o $CUDA_VERSION_NUMBER -lt 2030 ]; then +if [ $CUDA_VERSION_NUMBER -gt 10100 -o $CUDA_VERSION_NUMBER -lt 2030 ]; then echo "ERROR ** GPGPU-Sim version $GPGPUSIM_VERSION_STRING not tested with CUDA version $CUDA_VERSION_STRING (please see README)"; return fi -- cgit v1.3 From fe58efe9c8ca38f7d0f3781e54b04bc526bdfd07 Mon Sep 17 00:00:00 2001 From: Mahmoud Date: Mon, 10 Jun 2019 18:06:42 -0400 Subject: fixing thrust error --- libcuda/cuda_runtime_api.cc | 3 +++ src/stream_manager.h | 2 ++ 2 files changed, 5 insertions(+) (limited to 'libcuda/cuda_runtime_api.cc') diff --git a/libcuda/cuda_runtime_api.cc b/libcuda/cuda_runtime_api.cc index fb3e07a..17a5c96 100644 --- a/libcuda/cuda_runtime_api.cc +++ b/libcuda/cuda_runtime_api.cc @@ -1498,6 +1498,9 @@ __host__ cudaError_t CUDARTAPI cudaLaunch( const char *hostFun ) gpgpusim_ptx_assert( !g_cuda_launch_stack.empty(), "empty launch stack" ); kernel_config config = g_cuda_launch_stack.back(); struct CUstream_st *stream = config.get_stream(); + if(g_stream_manager->is_blocking()) + stream = 0; + printf("\nGPGPU-Sim PTX: cudaLaunch for 0x%p (mode=%s) on stream %u\n", hostFun, g_ptx_sim_mode?"functional simulation":"performance simulation", stream?stream->get_uid():0 ); kernel_info_t *grid = gpgpu_cuda_ptx_sim_init_grid(hostFun,config.get_args(),config.grid_dim(),config.block_dim(),context); diff --git a/src/stream_manager.h b/src/stream_manager.h index 91d1b36..3fbdbaf 100644 --- a/src/stream_manager.h +++ b/src/stream_manager.h @@ -258,6 +258,8 @@ public: void pushCudaStreamWaitEventToAllStreams( CUevent_st *e, unsigned int flags ); bool operation(bool * sim); void stop_all_running_kernels(); + unsigned size() {return m_streams.size(); }; + bool is_blocking() {return m_cuda_launch_blocking; }; private: void print_impl( FILE *fp); -- cgit v1.3 From 269fbbdb72dacdf4dc3c482b213174336c3eeb1f Mon Sep 17 00:00:00 2001 From: Mengchi Zhang Date: Mon, 10 Jun 2019 22:42:27 -0400 Subject: Move getters for gpgpu_context to globals Signed-off-by: Mengchi Zhang --- libcuda/cuda_runtime_api.cc | 2 +- libcuda/gpgpu_context.h | 2 ++ 2 files changed, 3 insertions(+), 1 deletion(-) (limited to 'libcuda/cuda_runtime_api.cc') diff --git a/libcuda/cuda_runtime_api.cc b/libcuda/cuda_runtime_api.cc index 2e2b50b..c5ebc20 100644 --- a/libcuda/cuda_runtime_api.cc +++ b/libcuda/cuda_runtime_api.cc @@ -273,7 +273,7 @@ static CUctx_st* GPGPUSim_Context() return the_context; } -static gpgpu_context* GPGPU_Context() +gpgpu_context* GPGPU_Context() { static gpgpu_context *gpgpu_ctx = NULL; if( gpgpu_ctx == NULL ) { diff --git a/libcuda/gpgpu_context.h b/libcuda/gpgpu_context.h index 4622f00..6c8a293 100644 --- a/libcuda/gpgpu_context.h +++ b/libcuda/gpgpu_context.h @@ -12,4 +12,6 @@ class gpgpu_context { cuda_runtime_api* api; // member function list }; +gpgpu_context* GPGPU_Context(); + #endif /* __gpgpu_context_h__ */ -- cgit v1.3 From ba1ed2941753ae8406bc9ec4a0de1eddc6454a1c Mon Sep 17 00:00:00 2001 From: Mengchi Zhang Date: Wed, 12 Jun 2019 00:42:37 -0400 Subject: Move some vars Signed-off-by: Mengchi Zhang --- libcuda/cuda_api_object.h | 5 +++ libcuda/cuda_runtime_api.cc | 74 ++++++++++++++++++++++++++------------------- 2 files changed, 48 insertions(+), 31 deletions(-) (limited to 'libcuda/cuda_runtime_api.cc') diff --git a/libcuda/cuda_api_object.h b/libcuda/cuda_api_object.h index d931fd5..f9a4fde 100644 --- a/libcuda/cuda_api_object.h +++ b/libcuda/cuda_api_object.h @@ -171,6 +171,7 @@ class cuda_runtime_api { public: cuda_runtime_api() { g_glbmap = NULL; + g_active_device = 0; //active gpu that runs the code } // global list std::list cuobjdumpSectionList; @@ -185,6 +186,7 @@ class cuda_runtime_api { std::map pinned_memory; //support for pinned memories added std::map pinned_memory_size; glbmap_entry_t* g_glbmap; + int g_active_device; //active gpu that runs the code // member function list void cuobjdumpInit(); void extract_code_using_cuobjdump(); @@ -201,5 +203,8 @@ class cuda_runtime_api { struct dim3 gridDim, struct dim3 blockDim, struct CUctx_st* context ); + int load_static_globals( symbol_table *symtab, unsigned min_gaddr, unsigned max_gaddr, gpgpu_t *gpu ); + int load_constants( symbol_table *symtab, addr_t min_gaddr, gpgpu_t *gpu ); + }; #endif /* __cuda_api_object_h__ */ diff --git a/libcuda/cuda_runtime_api.cc b/libcuda/cuda_runtime_api.cc index c5ebc20..09a13a7 100644 --- a/libcuda/cuda_runtime_api.cc +++ b/libcuda/cuda_runtime_api.cc @@ -153,15 +153,6 @@ extern void synchronize(); extern void exit_simulation(); -static int load_static_globals( symbol_table *symtab, unsigned min_gaddr, unsigned max_gaddr, gpgpu_t *gpu ); -static int load_constants( symbol_table *symtab, addr_t min_gaddr, gpgpu_t *gpu ); - -//static kernel_info_t *gpgpu_cuda_ptx_sim_init_grid( const char *kernel_key, -// gpgpu_ptx_sim_arg_list_t args, -// struct dim3 gridDim, -// struct dim3 blockDim, -// struct CUctx_st* context ); - /*DEVICE_BUILTIN*/ struct cudaArray { @@ -353,7 +344,6 @@ typedef std::map event_tracker_t; int CUevent_st::m_next_event_uid; event_tracker_t g_timer_events; -int g_active_device = 0; //active gpu that runs the code extern int cuobjdump_lex_init(yyscan_t* scanner); extern void cuobjdump_set_in (FILE * _in_str ,yyscan_t yyscanner ); @@ -492,6 +482,41 @@ void cuda_runtime_api::cuobjdumpRegisterFatBinary(unsigned int handle, const cha /******************************************************************************* * Add internal cuda runtime API call to accept gpgpu_context * *******************************************************************************/ +cudaError_t cudaSetDeviceInternal(int device, gpgpu_context* gpgpu_ctx = NULL) +{ + gpgpu_context *ctx; + if (gpgpu_ctx){ + ctx = gpgpu_ctx; + } else { + ctx = GPGPU_Context(); + } + if(g_debug_execution >= 3){ + announce_call(__my_func__); + } + //set the active device to run cuda + if ( device <= GPGPUSim_Init()->num_devices() ) { + ctx->api->g_active_device = device; + return g_last_cudaError = cudaSuccess; + } else { + return g_last_cudaError = cudaErrorInvalidDevice; + } +} + +cudaError_t cudaGetDeviceInternal(int *device, gpgpu_context* gpgpu_ctx = NULL) +{ + gpgpu_context *ctx; + if (gpgpu_ctx){ + ctx = gpgpu_ctx; + } else { + ctx = GPGPU_Context(); + } + if(g_debug_execution >= 3){ + announce_call(__my_func__); + } + *device = ctx->api->g_active_device; + return g_last_cudaError = cudaSuccess; +} + void** cudaRegisterFatBinaryInternal( void *fatCubin, gpgpu_context* gpgpu_ctx = NULL) { @@ -618,8 +643,8 @@ void** cudaRegisterFatBinaryInternal( void *fatCubin, gpgpu_context* gpgpu_ctx = gpgpu_ptxinfo_load_from_string( ptx, source_num, max_capability, context->no_of_ptx ); } source_num++; - load_static_globals(symtab,STATIC_ALLOC_LIMIT,0xFFFFFFFF,context->get_device()->get_gpgpu()); - load_constants(symtab,STATIC_ALLOC_LIMIT,context->get_device()->get_gpgpu()); + ctx->api->load_static_globals(symtab,STATIC_ALLOC_LIMIT,0xFFFFFFFF,context->get_device()->get_gpgpu()); + ctx->api->load_constants(symtab,STATIC_ALLOC_LIMIT,context->get_device()->get_gpgpu()); } else { printf("GPGPU-Sim PTX: warning -- did not find an appropriate PTX in cubin\n"); } @@ -988,8 +1013,8 @@ cuLinkAddFileInternal(CUlinkState state, CUjitInputType type, const char *path, std::string fname(path); ctx->api->name_symtab[fname] = symtab; context->add_binary(symtab, 1); - load_static_globals(symtab,STATIC_ALLOC_LIMIT,0xFFFFFFFF,context->get_device()->get_gpgpu()); - load_constants(symtab,STATIC_ALLOC_LIMIT,context->get_device()->get_gpgpu()); + ctx->api->load_static_globals(symtab,STATIC_ALLOC_LIMIT,0xFFFFFFFF,context->get_device()->get_gpgpu()); + ctx->api->load_constants(symtab,STATIC_ALLOC_LIMIT,context->get_device()->get_gpgpu()); addedFile = true; return CUDA_SUCCESS; } @@ -1692,25 +1717,12 @@ __host__ cudaError_t CUDARTAPI cudaChooseDevice(int *device, const struct cudaDe __host__ cudaError_t CUDARTAPI cudaSetDevice(int device) { - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - //set the active device to run cuda - if ( device <= GPGPUSim_Init()->num_devices() ) { - g_active_device = device; - return g_last_cudaError = cudaSuccess; - } else { - return g_last_cudaError = cudaErrorInvalidDevice; - } + return cudaSetDeviceInternal(device); } __host__ cudaError_t CUDARTAPI cudaGetDevice(int *device) { - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - *device = g_active_device; - return g_last_cudaError = cudaSuccess; + return cudaGetDeviceInternal(device); } __host__ cudaError_t CUDARTAPI cudaDeviceGetLimit ( size_t* pValue, cudaLimit limit ) @@ -3269,7 +3281,7 @@ int CUDARTAPI __cudaSynchronizeThreads(void**, void*) /// static functions -static int load_static_globals( symbol_table *symtab, unsigned min_gaddr, unsigned max_gaddr, gpgpu_t *gpu ) +int cuda_runtime_api::load_static_globals( symbol_table *symtab, unsigned min_gaddr, unsigned max_gaddr, gpgpu_t *gpu ) { if(g_debug_execution >= 3){ announce_call(__my_func__); @@ -3308,7 +3320,7 @@ static int load_static_globals( symbol_table *symtab, unsigned min_gaddr, unsign return ng_bytes; } -static int load_constants( symbol_table *symtab, addr_t min_gaddr, gpgpu_t *gpu ) +int cuda_runtime_api::load_constants( symbol_table *symtab, addr_t min_gaddr, gpgpu_t *gpu ) { if(g_debug_execution >= 3){ announce_call(__my_func__); -- cgit v1.3 From 9e52c143a883f682c02d81149748cdf8aa5508f7 Mon Sep 17 00:00:00 2001 From: Mengchi Zhang Date: Wed, 12 Jun 2019 01:42:52 -0400 Subject: Move some function from ptx_loader to gpgpu_context Signed-off-by: Mengchi Zhang --- libcuda/cuda_api_object.h | 1 - libcuda/cuda_runtime_api.cc | 42 ++++++++++++++++++++--------------------- libcuda/gpgpu_context.h | 3 +++ libopencl/opencl_runtime_api.cc | 5 ++++- src/cuda-sim/ptx_loader.cc | 5 +++-- src/cuda-sim/ptx_loader.h | 2 -- 6 files changed, 31 insertions(+), 27 deletions(-) (limited to 'libcuda/cuda_runtime_api.cc') diff --git a/libcuda/cuda_api_object.h b/libcuda/cuda_api_object.h index f9a4fde..0054697 100644 --- a/libcuda/cuda_api_object.h +++ b/libcuda/cuda_api_object.h @@ -191,7 +191,6 @@ class cuda_runtime_api { void cuobjdumpInit(); void extract_code_using_cuobjdump(); void extract_ptx_files_using_cuobjdump(CUctx_st *context); - void cuobjdumpParseBinary(unsigned int handle); std::list pruneSectionList(CUctx_st *context); std::list mergeMatchingSections(std::string identifier); std::list mergeSections(); diff --git a/libcuda/cuda_runtime_api.cc b/libcuda/cuda_runtime_api.cc index 09a13a7..25642a7 100644 --- a/libcuda/cuda_runtime_api.cc +++ b/libcuda/cuda_runtime_api.cc @@ -685,7 +685,7 @@ void cudaRegisterFunctionInternal( printf("GPGPU-Sim PTX: __cudaRegisterFunction %s : hostFun 0x%p, fat_cubin_handle = %u\n", deviceFun, hostFun, fat_cubin_handle); if(context->get_device()->get_gpgpu()->get_config().use_cuobjdump()) - ctx->api->cuobjdumpParseBinary(fat_cubin_handle); + ctx->cuobjdumpParseBinary(fat_cubin_handle); context->register_function( fat_cubin_handle, hostFun, deviceFun ); } @@ -712,7 +712,7 @@ void cudaRegisterVarInternal( printf("GPGPU-Sim PTX: __cudaRegisterVar: hostVar = %p; deviceAddress = %s; deviceName = %s\n", hostVar, deviceAddress, deviceName); printf("GPGPU-Sim PTX: __cudaRegisterVar: Registering const memory space of %d bytes\n", size); if(GPGPUSim_Context()->get_device()->get_gpgpu()->get_config().use_cuobjdump()) - ctx->api->cuobjdumpParseBinary((unsigned)(unsigned long long)fatCubinHandle); + ctx->cuobjdumpParseBinary((unsigned)(unsigned long long)fatCubinHandle); fflush(stdout); if ( constant && !global && !ext ) { gpgpu_ptx_sim_register_const_variable(hostVar,deviceName,size); @@ -1009,7 +1009,7 @@ cuLinkAddFileInternal(CUlinkState state, CUjitInputType type, const char *path, } strcat(file,"/"); strcat(file,path); - symbol_table *symtab = gpgpu_ptx_sim_load_ptx_from_filename( file ); + symbol_table *symtab = ctx->gpgpu_ptx_sim_load_ptx_from_filename( file ); std::string fname(path); ctx->api->name_symtab[fname] = symtab; context->add_binary(symtab, 1); @@ -2750,15 +2750,15 @@ void cuda_runtime_api::cuobjdumpInit(){ //! Either submit PTX for simulation or convert SASS to PTXPlus and submit it -void cuda_runtime_api::cuobjdumpParseBinary(unsigned int handle){ +void gpgpu_context::cuobjdumpParseBinary(unsigned int handle){ CUctx_st *context = GPGPUSim_Context(); - if(fatbin_registered[handle]) return; - fatbin_registered[handle] = true; - std::string fname = fatbinmap[handle]; + if(api->fatbin_registered[handle]) return; + api->fatbin_registered[handle] = true; + std::string fname = api->fatbinmap[handle]; - if (name_symtab.find(fname) != name_symtab.end()) { - symbol_table *symtab = name_symtab[fname]; + if (api->name_symtab.find(fname) != api->name_symtab.end()) { + symbol_table *symtab = api->name_symtab[fname]; context->add_binary(symtab, handle); return; } @@ -2767,7 +2767,7 @@ void cuda_runtime_api::cuobjdumpParseBinary(unsigned int handle){ #if (CUDART_VERSION >= 6000) //loops through all ptx files from smallest sm version to largest std::map >::iterator itr_m; - for (itr_m = version_filename.begin(); itr_m!=version_filename.end(); itr_m++){ + for (itr_m = api->version_filename.begin(); itr_m!=api->version_filename.end(); itr_m++){ std::set::iterator itr_s; for (itr_s = itr_m->second.begin(); itr_s!=itr_m->second.end(); itr_s++){ std::string ptx_filename = *itr_s; @@ -2775,11 +2775,11 @@ void cuda_runtime_api::cuobjdumpParseBinary(unsigned int handle){ symtab = gpgpu_ptx_sim_load_ptx_from_filename( ptx_filename.c_str() ); } } - name_symtab[fname] = symtab; + api->name_symtab[fname] = symtab; context->add_binary(symtab, handle); - load_static_globals(symtab,STATIC_ALLOC_LIMIT,0xFFFFFFFF,context->get_device()->get_gpgpu()); - load_constants(symtab,STATIC_ALLOC_LIMIT,context->get_device()->get_gpgpu()); - for (itr_m = version_filename.begin(); itr_m!=version_filename.end(); itr_m++){ + api->load_static_globals(symtab,STATIC_ALLOC_LIMIT,0xFFFFFFFF,context->get_device()->get_gpgpu()); + api->load_constants(symtab,STATIC_ALLOC_LIMIT,context->get_device()->get_gpgpu()); + for (itr_m = api->version_filename.begin(); itr_m!=api->version_filename.end(); itr_m++){ std::set::iterator itr_s; for (itr_s = itr_m->second.begin(); itr_s!=itr_m->second.end(); itr_s++){ std::string ptx_filename = *itr_s; @@ -2791,8 +2791,8 @@ void cuda_runtime_api::cuobjdumpParseBinary(unsigned int handle){ #endif unsigned max_capability = 0; - for ( std::list::iterator iter = cuobjdumpSectionList.begin(); - iter != cuobjdumpSectionList.end(); + for ( std::list::iterator iter = api->cuobjdumpSectionList.begin(); + iter != api->cuobjdumpSectionList.end(); iter++){ unsigned capability = (*iter)->getArch(); if (capability > max_capability) max_capability = capability; @@ -2803,7 +2803,7 @@ void cuda_runtime_api::cuobjdumpParseBinary(unsigned int handle){ cuobjdumpPTXSection* ptx = NULL; const char* pre_load = getenv("CUOBJDUMP_SIM_FILE"); if(pre_load==NULL || strlen(pre_load)==0) - ptx = findPTXSection(fname); + ptx = api->findPTXSection(fname); char *ptxcode; const char *override_ptx_name = getenv("PTX_SIM_KERNELFILE"); if (override_ptx_name == NULL or getenv("PTX_SIM_USE_PTX_FILE") == NULL or strlen(getenv("PTX_SIM_USE_PTX_FILE"))==0) { @@ -2813,7 +2813,7 @@ void cuda_runtime_api::cuobjdumpParseBinary(unsigned int handle){ ptxcode = readfile(override_ptx_name); } if(context->get_device()->get_gpgpu()->get_config().convert_to_ptxplus() ) { - cuobjdumpELFSection* elfsection = findELFSection(ptx->getIdentifier()); + cuobjdumpELFSection* elfsection = api->findELFSection(ptx->getIdentifier()); assert (elfsection!= NULL); char *ptxplus_str = gpgpu_ptx_sim_convert_ptx_and_sass_to_ptxplus( ptx->getPTXfilename(), @@ -2831,9 +2831,9 @@ void cuda_runtime_api::cuobjdumpParseBinary(unsigned int handle){ context->add_binary(symtab, handle); gpgpu_ptxinfo_load_from_string( ptxcode, handle, max_capability, context->no_of_ptx ); } - load_static_globals(symtab,STATIC_ALLOC_LIMIT,0xFFFFFFFF,context->get_device()->get_gpgpu()); - load_constants(symtab,STATIC_ALLOC_LIMIT,context->get_device()->get_gpgpu()); - name_symtab[fname] = symtab; + api->load_static_globals(symtab,STATIC_ALLOC_LIMIT,0xFFFFFFFF,context->get_device()->get_gpgpu()); + api->load_constants(symtab,STATIC_ALLOC_LIMIT,context->get_device()->get_gpgpu()); + api->name_symtab[fname] = symtab; //TODO: Remove temporarily files as per configurations } diff --git a/libcuda/gpgpu_context.h b/libcuda/gpgpu_context.h index 6c8a293..16626eb 100644 --- a/libcuda/gpgpu_context.h +++ b/libcuda/gpgpu_context.h @@ -11,6 +11,9 @@ class gpgpu_context { // objects pointers for each file cuda_runtime_api* api; // member function list + void cuobjdumpParseBinary(unsigned int handle); + class symbol_table *gpgpu_ptx_sim_load_ptx_from_string( const char *p, unsigned source_num ); + class symbol_table *gpgpu_ptx_sim_load_ptx_from_filename( const char *filename ); }; gpgpu_context* GPGPU_Context(); diff --git a/libopencl/opencl_runtime_api.cc b/libopencl/opencl_runtime_api.cc index 97a54d8..0ec635e 100644 --- a/libopencl/opencl_runtime_api.cc +++ b/libopencl/opencl_runtime_api.cc @@ -84,6 +84,7 @@ #include "../src/gpgpusim_entrypoint.h" #include "../src/gpgpu-sim/gpu-sim.h" #include "../src/gpgpu-sim/shader.h" +#include "../libcuda/gpgpu_context.h" //# define __my_func__ __PRETTY_FUNCTION__ # if defined __cplusplus ? __GNUC_PREREQ (2, 6) : __GNUC_PREREQ (2, 4) @@ -424,6 +425,8 @@ void register_ptx_function( const char *name, function_info *impl ) void _cl_program::Build(const char *options) { + gpgpu_context *ctx; + ctx = GPGPU_Context(); printf("GPGPU-Sim OpenCL API: compiling OpenCL kernels...\n"); std::map::iterator i; for( i = m_pgm.begin(); i!= m_pgm.end(); i++ ) { @@ -576,7 +579,7 @@ void _cl_program::Build(const char *options) } } info.m_asm = tmp; - info.m_symtab = gpgpu_ptx_sim_load_ptx_from_string( tmp, source_num ); + info.m_symtab = ctx->gpgpu_ptx_sim_load_ptx_from_string( tmp, source_num ); gpgpu_ptxinfo_load_from_string( tmp, source_num ); free(tmp); } diff --git a/src/cuda-sim/ptx_loader.cc b/src/cuda-sim/ptx_loader.cc index f037c34..d7e9b71 100644 --- a/src/cuda-sim/ptx_loader.cc +++ b/src/cuda-sim/ptx_loader.cc @@ -33,6 +33,7 @@ #include #include #include +#include "../../libcuda/gpgpu_context.h" /// globals @@ -165,7 +166,7 @@ char* gpgpu_ptx_sim_convert_ptx_and_sass_to_ptxplus(const std::string ptxfilenam } -symbol_table *gpgpu_ptx_sim_load_ptx_from_string( const char *p, unsigned source_num ) +symbol_table *gpgpu_context::gpgpu_ptx_sim_load_ptx_from_string( const char *p, unsigned source_num ) { char buf[1024]; snprintf(buf,1024,"_%u.ptx", source_num ); @@ -200,7 +201,7 @@ symbol_table *gpgpu_ptx_sim_load_ptx_from_string( const char *p, unsigned source return symtab; } -symbol_table *gpgpu_ptx_sim_load_ptx_from_filename( const char *filename ) +symbol_table *gpgpu_context::gpgpu_ptx_sim_load_ptx_from_filename( const char *filename ) { symbol_table *symtab=init_parser(filename); printf("GPGPU-Sim PTX: finished parsing EMBEDDED .ptx file %s\n",filename); diff --git a/src/cuda-sim/ptx_loader.h b/src/cuda-sim/ptx_loader.h index c4d8292..2af611a 100644 --- a/src/cuda-sim/ptx_loader.h +++ b/src/cuda-sim/ptx_loader.h @@ -42,8 +42,6 @@ class ptxinfo_data{ extern bool g_override_embedded_ptx; extern int no_of_ptx; //counter to track number of ptx files to be extracted in an application. -class symbol_table *gpgpu_ptx_sim_load_ptx_from_string( const char *p, unsigned source_num ); -class symbol_table *gpgpu_ptx_sim_load_ptx_from_filename( const char *filename ); void gpgpu_ptxinfo_load_from_string( const char *p_for_info, unsigned source_num, unsigned sm_version=20, int no_of_ptx=0 ); void gpgpu_ptx_info_load_from_filename( const char *filename, unsigned sm_version ); char* gpgpu_ptx_sim_convert_ptx_and_sass_to_ptxplus(const std::string ptx_str, const std::string sass_str, const std::string elf_str); -- cgit v1.3 From c7d713104df5bab5583b3a0e96323cbe346f9759 Mon Sep 17 00:00:00 2001 From: Mengchi Zhang Date: Wed, 12 Jun 2019 09:42:25 -0400 Subject: Fix for 4.2 Signed-off-by: Mengchi Zhang --- libcuda/cuda_runtime_api.cc | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) (limited to 'libcuda/cuda_runtime_api.cc') diff --git a/libcuda/cuda_runtime_api.cc b/libcuda/cuda_runtime_api.cc index 25642a7..14e3329 100644 --- a/libcuda/cuda_runtime_api.cc +++ b/libcuda/cuda_runtime_api.cc @@ -638,7 +638,7 @@ void** cudaRegisterFatBinaryInternal( void *fatCubin, gpgpu_context* gpgpu_ctx = "\tEither enable cuobjdump or disable PTXPlus in your configuration file\n"); exit(1); } else { - symtab=gpgpu_ptx_sim_load_ptx_from_string(ptx,source_num); + symtab=ctx->gpgpu_ptx_sim_load_ptx_from_string(ptx,source_num); context->add_binary(symtab,fat_cubin_handle); gpgpu_ptxinfo_load_from_string( ptx, source_num, max_capability, context->no_of_ptx ); } -- cgit v1.3 From 7d02cbb061485db38ed8e5f6bf06c9b2fa40eed2 Mon Sep 17 00:00:00 2001 From: Mengchi Zhang Date: Wed, 12 Jun 2019 13:19:08 -0400 Subject: Integrate ptxinfo into gpgpu_context Signed-off-by: Mengchi Zhang --- libcuda/cuda_runtime_api.cc | 2 +- libcuda/gpgpu_context.h | 5 +++++ libopencl/opencl_runtime_api.cc | 2 +- src/cuda-sim/ptx_loader.cc | 43 +++++++++++++++++++---------------------- src/cuda-sim/ptx_loader.h | 2 -- 5 files changed, 27 insertions(+), 27 deletions(-) (limited to 'libcuda/cuda_runtime_api.cc') diff --git a/libcuda/cuda_runtime_api.cc b/libcuda/cuda_runtime_api.cc index 14e3329..0a6eabb 100644 --- a/libcuda/cuda_runtime_api.cc +++ b/libcuda/cuda_runtime_api.cc @@ -640,7 +640,7 @@ void** cudaRegisterFatBinaryInternal( void *fatCubin, gpgpu_context* gpgpu_ctx = } else { symtab=ctx->gpgpu_ptx_sim_load_ptx_from_string(ptx,source_num); context->add_binary(symtab,fat_cubin_handle); - gpgpu_ptxinfo_load_from_string( ptx, source_num, max_capability, context->no_of_ptx ); + ctx->gpgpu_ptxinfo_load_from_string( ptx, source_num, max_capability, context->no_of_ptx ); } source_num++; ctx->api->load_static_globals(symtab,STATIC_ALLOC_LIMIT,0xFFFFFFFF,context->get_device()->get_gpgpu()); diff --git a/libcuda/gpgpu_context.h b/libcuda/gpgpu_context.h index 16626eb..d3db1ad 100644 --- a/libcuda/gpgpu_context.h +++ b/libcuda/gpgpu_context.h @@ -1,19 +1,24 @@ #ifndef __gpgpu_context_h__ #define __gpgpu_context_h__ #include "cuda_api_object.h" +#include "../src/cuda-sim/ptx_loader.h" class gpgpu_context { public: gpgpu_context() { api = new cuda_runtime_api(); + ptxinfo = new ptxinfo_data(); } // global list // objects pointers for each file cuda_runtime_api* api; + ptxinfo_data* ptxinfo; // member function list void cuobjdumpParseBinary(unsigned int handle); class symbol_table *gpgpu_ptx_sim_load_ptx_from_string( const char *p, unsigned source_num ); class symbol_table *gpgpu_ptx_sim_load_ptx_from_filename( const char *filename ); + void gpgpu_ptx_info_load_from_filename( const char *filename, unsigned sm_version); + void gpgpu_ptxinfo_load_from_string( const char *p_for_info, unsigned source_num, unsigned sm_version=20, int no_of_ptx=0 ); }; gpgpu_context* GPGPU_Context(); diff --git a/libopencl/opencl_runtime_api.cc b/libopencl/opencl_runtime_api.cc index 0ec635e..50a02fa 100644 --- a/libopencl/opencl_runtime_api.cc +++ b/libopencl/opencl_runtime_api.cc @@ -580,7 +580,7 @@ void _cl_program::Build(const char *options) } info.m_asm = tmp; info.m_symtab = ctx->gpgpu_ptx_sim_load_ptx_from_string( tmp, source_num ); - gpgpu_ptxinfo_load_from_string( tmp, source_num ); + ctx->gpgpu_ptxinfo_load_from_string( tmp, source_num ); free(tmp); } printf("GPGPU-Sim OpenCL API: finished compiling OpenCL kernels.\n"); diff --git a/src/cuda-sim/ptx_loader.cc b/src/cuda-sim/ptx_loader.cc index 9857741..2a6d930 100644 --- a/src/cuda-sim/ptx_loader.cc +++ b/src/cuda-sim/ptx_loader.cc @@ -332,7 +332,7 @@ char* get_app_binary_name(){ return self_exe_path; } -void gpgpu_ptx_info_load_from_filename( const char *filename, unsigned sm_version) +void gpgpu_context::gpgpu_ptx_info_load_from_filename( const char *filename, unsigned sm_version) { std::string ptxas_filename(std::string(filename) + "as"); char buff[1024], extra_flags[1024]; @@ -352,17 +352,16 @@ void gpgpu_ptx_info_load_from_filename( const char *filename, unsigned sm_versio } FILE *ptxinfo_in; - ptxinfo_data ptxinfo; - ptxinfo.g_ptxinfo_filename = strdup(ptxas_filename.c_str()); - ptxinfo_in = fopen(ptxinfo.g_ptxinfo_filename,"r"); - ptxinfo_lex_init(&(ptxinfo.scanner)); - ptxinfo_set_in(ptxinfo_in, ptxinfo.scanner); - ptxinfo_parse(ptxinfo.scanner, &ptxinfo); - ptxinfo_lex_destroy(ptxinfo.scanner); + ptxinfo->g_ptxinfo_filename = strdup(ptxas_filename.c_str()); + ptxinfo_in = fopen(ptxinfo->g_ptxinfo_filename,"r"); + ptxinfo_lex_init(&(ptxinfo->scanner)); + ptxinfo_set_in(ptxinfo_in, ptxinfo->scanner); + ptxinfo_parse(ptxinfo->scanner, ptxinfo); + ptxinfo_lex_destroy(ptxinfo->scanner); fclose(ptxinfo_in); } -void gpgpu_ptxinfo_load_from_string( const char *p_for_info, unsigned source_num, unsigned sm_version, int no_of_ptx ) +void gpgpu_context::gpgpu_ptxinfo_load_from_string( const char *p_for_info, unsigned source_num, unsigned sm_version, int no_of_ptx ) { //do ptxas for individual files instead of one big embedded ptx. This prevents the duplicate defs and declarations. char ptx_file[1000]; @@ -420,14 +419,13 @@ void gpgpu_ptxinfo_load_from_string( const char *p_for_info, unsigned source_num if( result != 0 ) { // 65280 = duplicate errors if (result == 65280) { - ptxinfo_data ptxinfo; FILE *ptxinfo_in; ptxinfo_in = fopen(tempfile_ptxinfo,"r"); - ptxinfo.g_ptxinfo_filename = tempfile_ptxinfo; - ptxinfo_lex_init(&(ptxinfo.scanner)); - ptxinfo_set_in(ptxinfo_in, ptxinfo.scanner); - ptxinfo_parse(ptxinfo.scanner, &ptxinfo); - ptxinfo_lex_destroy(ptxinfo.scanner); + ptxinfo->g_ptxinfo_filename = tempfile_ptxinfo; + ptxinfo_lex_init(&(ptxinfo->scanner)); + ptxinfo_set_in(ptxinfo_in, ptxinfo->scanner); + ptxinfo_parse(ptxinfo->scanner, ptxinfo); + ptxinfo_lex_destroy(ptxinfo->scanner); fclose(ptxinfo_in); fix_duplicate_errors(fname2); @@ -507,18 +505,17 @@ void gpgpu_ptxinfo_load_from_string( const char *p_for_info, unsigned source_num } } - ptxinfo_data ptxinfo; if(no_of_ptx>0) - ptxinfo.g_ptxinfo_filename = final_tempfile_ptxinfo; + ptxinfo->g_ptxinfo_filename = final_tempfile_ptxinfo; else - ptxinfo.g_ptxinfo_filename = tempfile_ptxinfo; + ptxinfo->g_ptxinfo_filename = tempfile_ptxinfo; FILE *ptxinfo_in; - ptxinfo_in = fopen(ptxinfo.g_ptxinfo_filename,"r"); + ptxinfo_in = fopen(ptxinfo->g_ptxinfo_filename,"r"); - ptxinfo_lex_init(&(ptxinfo.scanner)); - ptxinfo_set_in(ptxinfo_in, ptxinfo.scanner); - ptxinfo_parse(ptxinfo.scanner, &ptxinfo); - ptxinfo_lex_destroy(ptxinfo.scanner); + ptxinfo_lex_init(&(ptxinfo->scanner)); + ptxinfo_set_in(ptxinfo_in, ptxinfo->scanner); + ptxinfo_parse(ptxinfo->scanner, ptxinfo); + ptxinfo_lex_destroy(ptxinfo->scanner); fclose(ptxinfo_in); snprintf(commandline,1024,"rm -f *info"); diff --git a/src/cuda-sim/ptx_loader.h b/src/cuda-sim/ptx_loader.h index c3ce888..3637bff 100644 --- a/src/cuda-sim/ptx_loader.h +++ b/src/cuda-sim/ptx_loader.h @@ -44,8 +44,6 @@ class ptxinfo_data{ extern bool g_override_embedded_ptx; extern int no_of_ptx; //counter to track number of ptx files to be extracted in an application. -void gpgpu_ptxinfo_load_from_string( const char *p_for_info, unsigned source_num, unsigned sm_version=20, int no_of_ptx=0 ); -void gpgpu_ptx_info_load_from_filename( const char *filename, unsigned sm_version ); char* gpgpu_ptx_sim_convert_ptx_and_sass_to_ptxplus(const std::string ptx_str, const std::string sass_str, const std::string elf_str); bool keep_intermediate_files(); -- cgit v1.3 From f63324eda157f742d06c889b1be73717771e60a5 Mon Sep 17 00:00:00 2001 From: tgrogers Date: Tue, 18 Jun 2019 22:34:07 -0400 Subject: Some stuff to get mnist to run with CUDA 9.1 --- libcuda/cuda_runtime_api.cc | 49 +++++++++++++++++++++++++++++++++++++++++++++ 1 file changed, 49 insertions(+) (limited to 'libcuda/cuda_runtime_api.cc') diff --git a/libcuda/cuda_runtime_api.cc b/libcuda/cuda_runtime_api.cc index f46d9f1..8ba2b0f 100644 --- a/libcuda/cuda_runtime_api.cc +++ b/libcuda/cuda_runtime_api.cc @@ -1694,6 +1694,12 @@ __host__ cudaError_t CUDARTAPI cudaDeviceGetAttribute(int *value, enum cudaDevic break; case 88: case 89: + case 90: + case 91: + case 92: + case 93: + case 94: + case 95: *value= 0; break; default: @@ -3246,6 +3252,49 @@ __host__ cudaError_t CUDARTAPI cudaDeviceSetLimit(enum cudaLimit limit, size_t v #endif +/** + * \brief Set attributes for a given function + * + * This function sets the attributes of a function specified via \p entry. + * The parameter \p entry must be a pointer to a function that executes + * on the device. The parameter specified by \p entry must be declared as a \p __global__ + * function. The enumeration defined by \p attr is set to the value defined by \p value + * If the specified function does not exist, then ::cudaErrorInvalidDeviceFunction is returned. + * If the specified attribute cannot be written, or if the value is incorrect, + * then ::cudaErrorInvalidValue is returned. + * + * Valid values for \p attr are: + * ::cuFuncAttrMaxDynamicSharedMem - Maximum size of dynamic shared memory per block + * ::cudaFuncAttributePreferredSharedMemoryCarveout - Preferred shared memory-L1 cache split ratio + * + * \param entry - Function to get attributes of + * \param attr - Attribute to set + * \param value - Value to set + * + * \return + * ::cudaSuccess, + * ::cudaErrorInitializationError, + * ::cudaErrorInvalidDeviceFunction, + * ::cudaErrorInvalidValue + * \notefnerr + * + * \ref ::cudaLaunchKernel(const T *func, dim3 gridDim, dim3 blockDim, void **args, size_t sharedMem, cudaStream_t stream) "cudaLaunchKernel (C++ API)", + * \ref ::cudaFuncSetCacheConfig(T*, enum cudaFuncCache) "cudaFuncSetCacheConfig (C++ API)", + * \ref ::cudaFuncGetAttributes(struct cudaFuncAttributes*, const void*) "cudaFuncGetAttributes (C API)", + * ::cudaSetDoubleForDevice, + * ::cudaSetDoubleForHost, + * \ref ::cudaSetupArgument(T, size_t) "cudaSetupArgument (C++ API)" + */ +cudaError_t CUDARTAPI cudaFuncSetAttribute(const void *func, enum cudaFuncAttribute attr, int value) +{ + if(g_debug_execution >= 3){ + announce_call(__my_func__); + } + printf("GPGPU-Sim PTX: Execution warning: ignoring call to \"%s ( func=%p, attr=%d, value=%d )\"\n", + __my_func__, func, attr, value ); + return g_last_cudaError = cudaSuccess; +} + cudaError_t CUDARTAPI cudaGLSetGLDevice(int device) { if(g_debug_execution >= 3){ -- cgit v1.3 From 48830687ede62b3acaebeba93633255b4d8ec9c8 Mon Sep 17 00:00:00 2001 From: tgrogers Date: Wed, 19 Jun 2019 10:22:41 -0400 Subject: Fixing 4.2 build --- libcuda/cuda_runtime_api.cc | 3 +++ 1 file changed, 3 insertions(+) (limited to 'libcuda/cuda_runtime_api.cc') diff --git a/libcuda/cuda_runtime_api.cc b/libcuda/cuda_runtime_api.cc index 8ba2b0f..015fbc0 100644 --- a/libcuda/cuda_runtime_api.cc +++ b/libcuda/cuda_runtime_api.cc @@ -3252,6 +3252,8 @@ __host__ cudaError_t CUDARTAPI cudaDeviceSetLimit(enum cudaLimit limit, size_t v #endif + +#if CUDART_VERSION >= 9000 /** * \brief Set attributes for a given function * @@ -3294,6 +3296,7 @@ cudaError_t CUDARTAPI cudaFuncSetAttribute(const void *func, enum cudaFuncAttrib __my_func__, func, attr, value ); return g_last_cudaError = cudaSuccess; } +#endif cudaError_t CUDARTAPI cudaGLSetGLDevice(int device) { -- cgit v1.3 From f23a5de2c7eec680fc8f5c6ba45c64fcd9544e65 Mon Sep 17 00:00:00 2001 From: Mahmoud Date: Sun, 23 Jun 2019 15:22:54 -0400 Subject: fixing the buffer limit for function names --- libcuda/cuda_runtime_api.cc | 4 ++++ src/cuda-sim/ptx_ir.cc | 4 ++-- 2 files changed, 6 insertions(+), 2 deletions(-) (limited to 'libcuda/cuda_runtime_api.cc') diff --git a/libcuda/cuda_runtime_api.cc b/libcuda/cuda_runtime_api.cc index 17a5c96..97396ac 100644 --- a/libcuda/cuda_runtime_api.cc +++ b/libcuda/cuda_runtime_api.cc @@ -987,6 +987,10 @@ cudaError_t CUDARTAPI cudaOccupancyMaxActiveBlocksPerMultiprocessorWithFlags(int * context->get_device()->get_gpgpu()->get_config().num_shader()); dim3 blockDim(blockSize); kernel_info_t result(gridDim, blockDim, entry); + //if(entry == NULL){ + // *numBlocks = 1; + // return g_last_cudaError = cudaErrorUnknown; + //} *numBlocks = context->get_device()->get_gpgpu()->get_max_cta(result); printf("Maximum size is %d with gridDim %d and blockDim %d\n", *numBlocks, gridDim.x, blockDim.x); return g_last_cudaError = cudaSuccess; diff --git a/src/cuda-sim/ptx_ir.cc b/src/cuda-sim/ptx_ir.cc index 4ad9ddf..c4d5a6c 100644 --- a/src/cuda-sim/ptx_ir.cc +++ b/src/cuda-sim/ptx_ir.cc @@ -177,8 +177,8 @@ void symbol_table::add_function( function_info *func, const char *filename, unsi //Jin: handle instruction group for cdp symbol_table* symbol_table::start_inst_group() { - char inst_group_name[1024]; - snprintf(inst_group_name, 1024, "%s_inst_group_%u", m_scope_name.c_str(), m_inst_group_id); + char inst_group_name[2048]; + snprintf(inst_group_name, 2048, "%s_inst_group_%u", m_scope_name.c_str(), m_inst_group_id); //previous added assert(m_inst_group_symtab.find(std::string(inst_group_name)) == m_inst_group_symtab.end()); -- cgit v1.3 From cb678c3670de4a435a3260ed80dc476da3860082 Mon Sep 17 00:00:00 2001 From: Mengchi Zhang Date: Mon, 1 Jul 2019 16:23:33 -0400 Subject: Move g_debug_ir_generation and GPGPUSim_Init Signed-off-by: Mengchi Zhang --- libcuda/cuda_runtime_api.cc | 674 ++++++++++++++++++++++------------------ libcuda/gpgpu_context.h | 2 + libopencl/opencl_runtime_api.cc | 4 +- src/cuda-sim/ptx_parser.cc | 3 +- src/cuda-sim/ptx_parser.h | 3 + src/gpgpusim_entrypoint.cc | 5 +- src/gpgpusim_entrypoint.h | 1 - 7 files changed, 375 insertions(+), 317 deletions(-) (limited to 'libcuda/cuda_runtime_api.cc') diff --git a/libcuda/cuda_runtime_api.cc b/libcuda/cuda_runtime_api.cc index 61f8415..a34727f 100644 --- a/libcuda/cuda_runtime_api.cc +++ b/libcuda/cuda_runtime_api.cc @@ -197,7 +197,7 @@ void register_ptx_function( const char *name, function_info *impl ) # endif #endif -struct _cuda_device_id *GPGPUSim_Init() +struct _cuda_device_id *gpgpu_context::GPGPUSim_Init() { //static _cuda_device_id *the_device = NULL; _cuda_device_id *the_device = GPGPUsim_ctx_ptr()->the_cude_device; @@ -255,9 +255,10 @@ struct _cuda_device_id *GPGPUSim_Init() static CUctx_st* GPGPUSim_Context() { //static CUctx_st *the_context = NULL; + gpgpu_context *cur_ctx = GPGPU_Context(); CUctx_st *the_context = GPGPUsim_ctx_ptr()->the_context; if( the_context == NULL ) { - _cuda_device_id *the_gpu = GPGPUSim_Init(); + _cuda_device_id *the_gpu = cur_ctx->GPGPUSim_Init(); GPGPUsim_ctx_ptr()->the_context = new CUctx_st(the_gpu); the_context = GPGPUsim_ctx_ptr()->the_context; } @@ -496,7 +497,7 @@ cudaError_t cudaSetDeviceInternal(int device, gpgpu_context* gpgpu_ctx = NULL) announce_call(__my_func__); } //set the active device to run cuda - if ( device <= GPGPUSim_Init()->num_devices() ) { + if ( device <= ctx->GPGPUSim_Init()->num_devices() ) { ctx->api->g_active_device = device; return g_last_cudaError = cudaSuccess; } else { @@ -519,6 +520,55 @@ cudaError_t cudaGetDeviceInternal(int *device, gpgpu_context* gpgpu_ctx = NULL) return g_last_cudaError = cudaSuccess; } +__host__ cudaError_t CUDARTAPI cudaDeviceGetLimitInternal( size_t* pValue, cudaLimit limit, gpgpu_context* gpgpu_ctx = NULL ) +{ + gpgpu_context *ctx; + if (gpgpu_ctx){ + ctx = gpgpu_ctx; + } else { + ctx = GPGPU_Context(); + } + if(g_debug_execution >= 3){ + announce_call(__my_func__); + } + _cuda_device_id *dev = ctx->GPGPUSim_Init(); + const struct cudaDeviceProp *prop = dev->get_prop(); + const gpgpu_sim_config& config=dev->get_gpgpu()->get_config(); + switch(limit) { + case 0: // cudaLimitStackSize + *pValue=config.stack_limit(); + break; + case 2: // cudaLimitMallocHeapSize + *pValue=config.heap_limit(); + break; +#if (CUDART_VERSION > 5050) + case 3: // cudaLimitDevRuntimeSyncDepth + if(prop->major > 2){ + *pValue=config.sync_depth_limit(); + break; + } + else{ + printf("ERROR:Limit %s is not supported on this architecture \n",limit); + abort(); + } + case 4: // cudaLimitDevRuntimePendingLaunchCount + if(prop->major > 2){ + *pValue=config.pending_launch_count_limit(); + break; + } + else{ + printf("ERROR:Limit %s is not supported on this architecture \n",limit); + abort(); + } +#endif + default: + printf("ERROR:Limit %s unimplemented \n",limit); + abort(); + } + return g_last_cudaError = cudaSuccess; + +} + void** cudaRegisterFatBinaryInternal( void *fatCubin, gpgpu_context* gpgpu_ctx = NULL) { @@ -739,6 +789,260 @@ cudaError_t cudaConfigureCallInternal(dim3 gridDim, dim3 blockDim, size_t shared return g_last_cudaError = cudaSuccess; } +__host__ cudaError_t CUDARTAPI cudaGetDeviceCountInternal(int *count, gpgpu_context* gpgpu_ctx = NULL) +{ + gpgpu_context *ctx; + if (gpgpu_ctx){ + ctx = gpgpu_ctx; + } else { + ctx = GPGPU_Context(); + } + if(g_debug_execution >= 3){ + announce_call(__my_func__); + } + _cuda_device_id *dev = ctx->GPGPUSim_Init(); + *count = dev->num_devices(); + return g_last_cudaError = cudaSuccess; +} + +__host__ cudaError_t CUDARTAPI cudaGetDevicePropertiesInternal(struct cudaDeviceProp *prop, int device, gpgpu_context* gpgpu_ctx = NULL) +{ + gpgpu_context *ctx; + if (gpgpu_ctx){ + ctx = gpgpu_ctx; + } else { + ctx = GPGPU_Context(); + } + if(g_debug_execution >= 3){ + announce_call(__my_func__); + } + _cuda_device_id *dev = ctx->GPGPUSim_Init(); + if (device <= dev->num_devices() ) { + *prop= *dev->get_prop(); + return g_last_cudaError = cudaSuccess; + } else { + return g_last_cudaError = cudaErrorInvalidDevice; + } +} + +#if (CUDART_VERSION > 5000) +__host__ cudaError_t CUDARTAPI cudaDeviceGetAttributeInternal(int *value, enum cudaDeviceAttr attr, int device, gpgpu_context* gpgpu_ctx = NULL) +{ + gpgpu_context *ctx; + if (gpgpu_ctx){ + ctx = gpgpu_ctx; + } else { + ctx = GPGPU_Context(); + } + if(g_debug_execution >= 3){ + announce_call(__my_func__); + } + const struct cudaDeviceProp *prop; + _cuda_device_id *dev = ctx->GPGPUSim_Init(); + if (device <= dev->num_devices() ) { + prop = dev->get_prop(); + switch (attr) { + case 1: + *value= prop->maxThreadsDim[0] * prop->maxThreadsDim[1] * prop->maxThreadsDim[2] * prop->maxGridSize[0] * prop->maxGridSize[1] * prop->maxGridSize[2]; + break; + case 2: + *value= prop->maxThreadsDim[0]; + break; + case 3: + *value= prop->maxThreadsDim[1]; + break; + case 4: + *value= prop->maxThreadsDim[2]; + break; + case 5: + *value= prop->maxGridSize[0]; + break; + case 6: + *value= prop->maxGridSize[1]; + break; + case 7: + *value= prop->maxGridSize[2]; + break; + case 8: + *value= prop->sharedMemPerBlock; + break; + case 9: + *value= prop->totalConstMem; + break; + case 10: + *value= prop->warpSize; + break; + case 11: + *value= 16;//dummy value + break; + case 12: + *value= prop->regsPerBlock; + break; + case 13: + *value= 1480000;//for 1080ti + break; + case 14: + *value= prop->textureAlignment ; + break; + case 15: + *value = 0; + break; + case 16: + *value= prop->multiProcessorCount ; + break; + case 17: + case 18: + case 19: + *value = 0; + break; + case 21: + case 22: + case 23: + case 24: + case 25: + case 26: + case 27: + case 28: + case 42: + case 45: + case 46: + case 47: + case 48: + case 49: + case 52: + case 53: + case 55: + case 56: + case 57: + case 58: + case 59: + case 60: + case 61: + case 62: + case 63: + case 64: + case 66: + case 67: + case 69: + case 70: + case 71: + case 73: + case 74: + case 77: + *value = 1000;//dummy value + break; + case 29: + case 43: + case 54: + case 65: + case 68: + case 72: + *value = 10;//dummy value + break; + case 30: + case 51: + *value = 128;//dummy value + break; + case 31: + *value = 1; + break; + case 32: + *value = 0; + break; + case 33: + case 50: + *value = 0;//dummy value + break; + case 34: + *value= 0; + break; + case 35: + *value = 0; + break; + case 36: + *value = 1250000;//CK value for 1080ti + break; + case 37: + *value = 352;//value for 1080ti + break; + case 38: + *value = 3000000;//value for 1080ti + break; + case 39: + *value= dev->get_gpgpu()->threads_per_core(); + break; + case 40: + *value= 0; + break; + case 41: + *value= 0; + break; + case 75://cudaDevAttrComputeCapabilityMajor + *value= prop->major ; + break; + case 76://cudaDevAttrComputeCapabilityMinor + *value= prop->minor ; + break; + case 78: + *value= 0 ; //TODO: as of now, we dont support stream priorities. + break; + case 79: + *value= 0; + break; + case 80: + *value= 0; + break; + #if (CUDART_VERSION > 5050) + case 81: + *value= prop->sharedMemPerMultiprocessor; + break; + case 82: + *value= prop->regsPerMultiprocessor; + break; + #endif + case 83: + case 84: + case 85: + case 86: + *value= 0; + break; + case 87: + *value= 4;//dummy value + break; + case 88: + case 89: + case 90: + case 91: + case 95: + *value= 0; + break; + default: + printf("ERROR: Attribute number %d unimplemented \n",attr); + abort(); + } + return g_last_cudaError = cudaSuccess; + } else { + return g_last_cudaError = cudaErrorInvalidDevice; + } +} +#endif + +__host__ cudaError_t CUDARTAPI cudaChooseDeviceInternal(int *device, const struct cudaDeviceProp *prop, gpgpu_context* gpgpu_ctx = NULL) +{ + gpgpu_context *ctx; + if (gpgpu_ctx){ + ctx = gpgpu_ctx; + } else { + ctx = GPGPU_Context(); + } + if(g_debug_execution >= 3){ + announce_call(__my_func__); + } + _cuda_device_id *dev = ctx->GPGPUSim_Init(); + *device = dev->get_id(); + return g_last_cudaError = cudaSuccess; +} + cudaError_t cudaSetupArgumentInternal(const void *arg, size_t size, size_t offset, gpgpu_context* gpgpu_ctx = NULL) { gpgpu_context *ctx; @@ -1058,6 +1362,52 @@ cudaError_t cudaHostAllocInternal(void **pHost, size_t bytes, unsigned int flag #endif +size_t getMaxThreadsPerBlock(struct cudaFuncAttributes *attr, gpgpu_context *ctx) { + _cuda_device_id *dev = ctx->GPGPUSim_Init(); + struct cudaDeviceProp prop; + + prop = *dev->get_prop(); + + size_t max = prop.maxThreadsPerBlock; + + if ((prop.regsPerBlock / attr->numRegs) < max) + max = prop.regsPerBlock / attr->numRegs; + + return max; +} + +cudaError_t CUDARTAPI cudaFuncGetAttributesInternal(struct cudaFuncAttributes *attr, const char *hostFun, gpgpu_context* gpgpu_ctx = NULL ) +{ + gpgpu_context *ctx; + if (gpgpu_ctx){ + ctx = gpgpu_ctx; + } else { + ctx = GPGPU_Context(); + } + if(g_debug_execution >= 3){ + announce_call(__my_func__); + } + CUctx_st *context = GPGPUSim_Context(); + function_info *entry = context->get_kernel(hostFun); + if( entry ) { + const struct gpgpu_ptx_sim_info *kinfo = entry->get_kernel_info(); + attr->sharedSizeBytes = kinfo->smem; + attr->constSizeBytes = kinfo->cmem; + attr->localSizeBytes = kinfo->lmem; + attr->numRegs = kinfo->regs; + if(kinfo->maxthreads > 0) + attr->maxThreadsPerBlock = kinfo->maxthreads; + else + attr->maxThreadsPerBlock = getMaxThreadsPerBlock(attr, ctx); +#if CUDART_VERSION >= 3000 + attr->ptxVersion = kinfo->ptx_version; + attr->binaryVersion = kinfo->sm_target; +#endif + } + return g_last_cudaError = cudaSuccess; +} + + /******************************************************************************* * * * * @@ -1503,235 +1853,24 @@ __host__ cudaError_t CUDARTAPI cudaGetSymbolSize(size_t *size, const char *symbo *******************************************************************************/ __host__ cudaError_t CUDARTAPI cudaGetDeviceCount(int *count) { - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - _cuda_device_id *dev = GPGPUSim_Init(); - *count = dev->num_devices(); - return g_last_cudaError = cudaSuccess; + return cudaGetDeviceCountInternal(count); } - -__host__ cudaError_t CUDARTAPI cudaGetDeviceProperties(struct cudaDeviceProp *prop, int device) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - _cuda_device_id *dev = GPGPUSim_Init(); - if (device <= dev->num_devices() ) { - *prop= *dev->get_prop(); - return g_last_cudaError = cudaSuccess; - } else { - return g_last_cudaError = cudaErrorInvalidDevice; - } + +__host__ cudaError_t CUDARTAPI cudaGetDeviceProperties(struct cudaDeviceProp *prop, int device) +{ + return cudaGetDevicePropertiesInternal(prop, device); } #if (CUDART_VERSION > 5000) __host__ cudaError_t CUDARTAPI cudaDeviceGetAttribute(int *value, enum cudaDeviceAttr attr, int device) { - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - const struct cudaDeviceProp *prop; - _cuda_device_id *dev = GPGPUSim_Init(); - if (device <= dev->num_devices() ) { - prop = dev->get_prop(); - switch (attr) { - case 1: - *value= prop->maxThreadsDim[0] * prop->maxThreadsDim[1] * prop->maxThreadsDim[2] * prop->maxGridSize[0] * prop->maxGridSize[1] * prop->maxGridSize[2]; - break; - case 2: - *value= prop->maxThreadsDim[0]; - break; - case 3: - *value= prop->maxThreadsDim[1]; - break; - case 4: - *value= prop->maxThreadsDim[2]; - break; - case 5: - *value= prop->maxGridSize[0]; - break; - case 6: - *value= prop->maxGridSize[1]; - break; - case 7: - *value= prop->maxGridSize[2]; - break; - case 8: - *value= prop->sharedMemPerBlock; - break; - case 9: - *value= prop->totalConstMem; - break; - case 10: - *value= prop->warpSize; - break; - case 11: - *value= 16;//dummy value - break; - case 12: - *value= prop->regsPerBlock; - break; - case 13: - *value= 1480000;//for 1080ti - break; - case 14: - *value= prop->textureAlignment ; - break; - case 15: - *value = 0; - break; - case 16: - *value= prop->multiProcessorCount ; - break; - case 17: - case 18: - case 19: - *value = 0; - break; - case 21: - case 22: - case 23: - case 24: - case 25: - case 26: - case 27: - case 28: - case 42: - case 45: - case 46: - case 47: - case 48: - case 49: - case 52: - case 53: - case 55: - case 56: - case 57: - case 58: - case 59: - case 60: - case 61: - case 62: - case 63: - case 64: - case 66: - case 67: - case 69: - case 70: - case 71: - case 73: - case 74: - case 77: - *value = 1000;//dummy value - break; - case 29: - case 43: - case 54: - case 65: - case 68: - case 72: - *value = 10;//dummy value - break; - case 30: - case 51: - *value = 128;//dummy value - break; - case 31: - *value = 1; - break; - case 32: - *value = 0; - break; - case 33: - case 50: - *value = 0;//dummy value - break; - case 34: - *value= 0; - break; - case 35: - *value = 0; - break; - case 36: - *value = 1250000;//CK value for 1080ti - break; - case 37: - *value = 352;//value for 1080ti - break; - case 38: - *value = 3000000;//value for 1080ti - break; - case 39: - *value= dev->get_gpgpu()->threads_per_core(); - break; - case 40: - *value= 0; - break; - case 41: - *value= 0; - break; - case 75://cudaDevAttrComputeCapabilityMajor - *value= prop->major ; - break; - case 76://cudaDevAttrComputeCapabilityMinor - *value= prop->minor ; - break; - case 78: - *value= 0 ; //TODO: as of now, we dont support stream priorities. - break; - case 79: - *value= 0; - break; - case 80: - *value= 0; - break; - #if (CUDART_VERSION > 5050) - case 81: - *value= prop->sharedMemPerMultiprocessor; - break; - case 82: - *value= prop->regsPerMultiprocessor; - break; - #endif - case 83: - case 84: - case 85: - case 86: - *value= 0; - break; - case 87: - *value= 4;//dummy value - break; - case 88: - case 89: - case 90: - case 91: - case 92: - case 93: - case 94: - case 95: - *value= 0; - break; - default: - printf("ERROR: Attribute number %d unimplemented \n",attr); - abort(); - } - return g_last_cudaError = cudaSuccess; - } else { - return g_last_cudaError = cudaErrorInvalidDevice; - } + return cudaDeviceGetAttributeInternal(value, attr, device); } #endif __host__ cudaError_t CUDARTAPI cudaChooseDevice(int *device, const struct cudaDeviceProp *prop) { - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - _cuda_device_id *dev = GPGPUSim_Init(); - *device = dev->get_id(); - return g_last_cudaError = cudaSuccess; + return cudaChooseDeviceInternal(device, prop); } __host__ cudaError_t CUDARTAPI cudaSetDevice(int device) @@ -1744,47 +1883,9 @@ __host__ cudaError_t CUDARTAPI cudaGetDevice(int *device) return cudaGetDeviceInternal(device); } -__host__ cudaError_t CUDARTAPI cudaDeviceGetLimit ( size_t* pValue, cudaLimit limit ) +__host__ cudaError_t CUDARTAPI cudaDeviceGetLimit( size_t* pValue, cudaLimit limit ) { - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - _cuda_device_id *dev = GPGPUSim_Init(); - const struct cudaDeviceProp *prop = dev->get_prop(); - const gpgpu_sim_config& config=dev->get_gpgpu()->get_config(); - switch(limit) { - case 0: // cudaLimitStackSize - *pValue=config.stack_limit(); - break; - case 2: // cudaLimitMallocHeapSize - *pValue=config.heap_limit(); - break; -#if (CUDART_VERSION > 5050) - case 3: // cudaLimitDevRuntimeSyncDepth - if(prop->major > 2){ - *pValue=config.sync_depth_limit(); - break; - } - else{ - printf("ERROR:Limit %s is not supported on this architecture \n",limit); - abort(); - } - case 4: // cudaLimitDevRuntimePendingLaunchCount - if(prop->major > 2){ - *pValue=config.pending_launch_count_limit(); - break; - } - else{ - printf("ERROR:Limit %s is not supported on this architecture \n",limit); - abort(); - } -#endif - default: - printf("ERROR:Limit %s unimplemented \n",limit); - abort(); - } - return g_last_cudaError = cudaSuccess; - + return cudaDeviceGetLimitInternal( pValue, limit ); } __host__ cudaError_t CUDARTAPI cudaStreamGetPriority ( cudaStream_t hStream, int* priority ) @@ -3167,58 +3268,9 @@ cudaError_t CUDARTAPI cudaSetDeviceFlags( int flags ) } } -size_t getMaxThreadsPerBlock(struct cudaFuncAttributes *attr) { - _cuda_device_id *dev = GPGPUSim_Init(); - struct cudaDeviceProp prop; - - prop = *dev->get_prop(); - - size_t max = prop.maxThreadsPerBlock; - - if ((prop.regsPerBlock / attr->numRegs) < max) - max = prop.regsPerBlock / attr->numRegs; - - return max; -} - cudaError_t CUDARTAPI cudaFuncGetAttributes(struct cudaFuncAttributes *attr, const char *hostFun ) { - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - CUctx_st *context = GPGPUSim_Context(); - function_info *entry = context->get_kernel(hostFun); - if( entry ) { - const struct gpgpu_ptx_sim_info *kinfo = entry->get_kernel_info(); - attr->sharedSizeBytes = kinfo->smem; - attr->constSizeBytes = kinfo->cmem; - attr->localSizeBytes = kinfo->lmem; - attr->numRegs = kinfo->regs; - if(kinfo->maxthreads > 0) - attr->maxThreadsPerBlock = kinfo->maxthreads; - else - attr->maxThreadsPerBlock = getMaxThreadsPerBlock(attr); -#if CUDART_VERSION >= 3000 - attr->ptxVersion = kinfo->ptx_version; - attr->binaryVersion = kinfo->sm_target; -#endif - } - return g_last_cudaError = cudaSuccess; -} - -cudaError_t CUDARTAPI cudaEventCreateWithFlags(cudaEvent_t *event, int flags) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - CUevent_st *e = new CUevent_st(flags==cudaEventBlockingSync); - g_timer_events[e->get_uid()] = e; -#if CUDART_VERSION >= 3000 - *event = e; -#else - *event = e->get_uid(); -#endif - return g_last_cudaError = cudaSuccess; + return cudaFuncGetAttributesInternal(attr, hostFun ); } cudaError_t CUDARTAPI cudaDriverGetVersion(int *driverVersion) diff --git a/libcuda/gpgpu_context.h b/libcuda/gpgpu_context.h index fa5d02b..dd8b51d 100644 --- a/libcuda/gpgpu_context.h +++ b/libcuda/gpgpu_context.h @@ -26,6 +26,8 @@ class gpgpu_context { void gpgpu_ptxinfo_load_from_string( const char *p_for_info, unsigned source_num, unsigned sm_version=20, int no_of_ptx=0 ); void print_ptx_file( const char *p, unsigned source_num, const char *filename ); class symbol_table* init_parser(const char*); + class gpgpu_sim *gpgpu_ptx_sim_init_perf(); + struct _cuda_device_id *GPGPUSim_Init(); }; gpgpu_context* GPGPU_Context(); diff --git a/libopencl/opencl_runtime_api.cc b/libopencl/opencl_runtime_api.cc index 50a02fa..e91e2e0 100644 --- a/libopencl/opencl_runtime_api.cc +++ b/libopencl/opencl_runtime_api.cc @@ -648,7 +648,9 @@ class _cl_device_id *GPGPUSim_Init() { static _cl_device_id *the_device = NULL; if( !the_device ) { - gpgpu_sim *the_gpu = gpgpu_ptx_sim_init_perf(); + gpgpu_context *ctx; + ctx = GPGPU_Context(); + gpgpu_sim *the_gpu = ctx->gpgpu_ptx_sim_init_perf(); the_device = new _cl_device_id(the_gpu); } start_sim_thread(2); diff --git a/src/cuda-sim/ptx_parser.cc b/src/cuda-sim/ptx_parser.cc index 2872b84..9094ec3 100644 --- a/src/cuda-sim/ptx_parser.cc +++ b/src/cuda-sim/ptx_parser.cc @@ -48,7 +48,6 @@ void set_ptx_warp_size(const struct core_config * warp_size) g_shader_core_config=warp_size; } -static bool g_debug_ir_generation=false; const char *g_filename; // the program intermediate representation... @@ -70,7 +69,7 @@ const char *decode_token( int type ) return g_ptx_token_decode[type].c_str(); } -void read_parser_environment_variables() +void ptx_recognizer::read_parser_environment_variables() { g_filename = getenv("PTX_SIM_KERNELFILE"); char *dbg_level = getenv("PTX_SIM_DEBUG"); diff --git a/src/cuda-sim/ptx_parser.h b/src/cuda-sim/ptx_parser.h index 25c01fe..bc9a872 100644 --- a/src/cuda-sim/ptx_parser.h +++ b/src/cuda-sim/ptx_parser.h @@ -62,6 +62,7 @@ class ptx_recognizer { g_error_detected = 0; g_entry_func_param_index=0; g_func_info = NULL; + g_debug_ir_generation=false; } // global list yyscan_t scanner; @@ -103,6 +104,7 @@ class ptx_recognizer { unsigned g_entry_func_param_index; function_info *g_func_info; operand_info g_return_var; + bool g_debug_ir_generation; // member function list void init_directive_state(); @@ -169,6 +171,7 @@ class ptx_recognizer { void start_inst_group(); void end_inst_group(); bool check_for_duplicates( const char *identifier ); + void read_parser_environment_variables(); }; diff --git a/src/gpgpusim_entrypoint.cc b/src/gpgpusim_entrypoint.cc index b09674a..476a9d4 100644 --- a/src/gpgpusim_entrypoint.cc +++ b/src/gpgpusim_entrypoint.cc @@ -35,6 +35,7 @@ #include "gpgpu-sim/gpu-sim.h" #include "gpgpu-sim/icnt_wrapper.h" #include "stream_manager.h" +#include "../libcuda/gpgpu_context.h" #define MAX(a,b) (((a)>(b))?(a):(b)) @@ -208,12 +209,12 @@ void exit_simulation() extern bool g_cuda_launch_blocking; -gpgpu_sim *gpgpu_ptx_sim_init_perf() +gpgpu_sim *gpgpu_context::gpgpu_ptx_sim_init_perf() { srand(1); print_splash(); read_sim_environment_variables(); - read_parser_environment_variables(); + ptx_parser->read_parser_environment_variables(); option_parser_t opp = option_parser_create(); ptx_reg_options(opp); diff --git a/src/gpgpusim_entrypoint.h b/src/gpgpusim_entrypoint.h index 9ce7fef..a443151 100644 --- a/src/gpgpusim_entrypoint.h +++ b/src/gpgpusim_entrypoint.h @@ -74,7 +74,6 @@ struct GPGPUsim_ctx { }; -class gpgpu_sim *gpgpu_ptx_sim_init_perf(); void start_sim_thread(int api); class gpgpu_sim* g_the_gpu(); -- cgit v1.3 From 3600ed5c59adafe40840524d19f622aa25e60dd6 Mon Sep 17 00:00:00 2001 From: Mengchi Zhang Date: Tue, 2 Jul 2019 00:22:15 -0400 Subject: Move ptxinfo_addinfo Signed-off-by: Mengchi Zhang --- libcuda/cuda_runtime_api.cc | 2 +- src/cuda-sim/ptx_loader.h | 2 +- src/cuda-sim/ptxinfo.y | 5 ++--- 3 files changed, 4 insertions(+), 5 deletions(-) (limited to 'libcuda/cuda_runtime_api.cc') diff --git a/libcuda/cuda_runtime_api.cc b/libcuda/cuda_runtime_api.cc index a34727f..eb645ce 100644 --- a/libcuda/cuda_runtime_api.cc +++ b/libcuda/cuda_runtime_api.cc @@ -274,7 +274,7 @@ gpgpu_context* GPGPU_Context() return gpgpu_ctx; } - void ptxinfo_addinfo() + void ptxinfo_data::ptxinfo_addinfo() { if(!get_ptxinfo_kname()){ /* This info is not per kernel (since CUDA 5.0 some info (e.g. gmem, and cmem) is added at the beginning for the whole binary ) */ diff --git a/src/cuda-sim/ptx_loader.h b/src/cuda-sim/ptx_loader.h index ee09e16..77f27c8 100644 --- a/src/cuda-sim/ptx_loader.h +++ b/src/cuda-sim/ptx_loader.h @@ -37,7 +37,7 @@ class ptxinfo_data{ char linebuf[PTXINFO_LINEBUF_SIZE]; unsigned col; const char *g_ptxinfo_filename; - + void ptxinfo_addinfo(); }; diff --git a/src/cuda-sim/ptxinfo.y b/src/cuda-sim/ptxinfo.y index 00c81e0..b303958 100644 --- a/src/cuda-sim/ptxinfo.y +++ b/src/cuda-sim/ptxinfo.y @@ -29,7 +29,7 @@ OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. %{ typedef void * yyscan_t; -class ptxinfo_data; +#include "ptx_loader.h" %} %define api.pure full @@ -79,7 +79,6 @@ class ptxinfo_data; static unsigned g_system; int ptxinfo_lex(YYSTYPE * yylval_param, yyscan_t yyscanner, ptxinfo_data* ptxinfo); void yyerror(yyscan_t yyscanner, ptxinfo_data* ptxinfo, const char* msg); - void ptxinfo_addinfo(); void ptxinfo_function(const char *fname ); void ptxinfo_regs( unsigned nregs ); void ptxinfo_lmem( unsigned declared, unsigned system ); @@ -104,7 +103,7 @@ line: HEADER INFO COLON line_info ; line_info: function_name - | function_info { ptxinfo_addinfo(); } + | function_info { ptxinfo->ptxinfo_addinfo(); } | gmem_info ; -- cgit v1.3 From 6e051c42b8c9c820af50a10025d182a96fdd12a6 Mon Sep 17 00:00:00 2001 From: Mengchi Zhang Date: Tue, 9 Jul 2019 10:06:31 -0400 Subject: Move m_ptx_save_converted_ptxplus Signed-off-by: Mengchi Zhang --- libcuda/cuda_runtime_api.cc | 2 +- src/cuda-sim/ptx_loader.cc | 5 ++--- src/cuda-sim/ptx_loader.h | 4 ++-- 3 files changed, 5 insertions(+), 6 deletions(-) (limited to 'libcuda/cuda_runtime_api.cc') diff --git a/libcuda/cuda_runtime_api.cc b/libcuda/cuda_runtime_api.cc index eb645ce..43c5e1f 100644 --- a/libcuda/cuda_runtime_api.cc +++ b/libcuda/cuda_runtime_api.cc @@ -2936,7 +2936,7 @@ void gpgpu_context::cuobjdumpParseBinary(unsigned int handle){ if(context->get_device()->get_gpgpu()->get_config().convert_to_ptxplus() ) { cuobjdumpELFSection* elfsection = api->findELFSection(ptx->getIdentifier()); assert (elfsection!= NULL); - char *ptxplus_str = gpgpu_ptx_sim_convert_ptx_and_sass_to_ptxplus( + char *ptxplus_str = ptxinfo->gpgpu_ptx_sim_convert_ptx_and_sass_to_ptxplus( ptx->getPTXfilename(), elfsection->getELFfilename(), elfsection->getSASSfilename()); diff --git a/src/cuda-sim/ptx_loader.cc b/src/cuda-sim/ptx_loader.cc index e0d9a11..2bf464c 100644 --- a/src/cuda-sim/ptx_loader.cc +++ b/src/cuda-sim/ptx_loader.cc @@ -58,7 +58,6 @@ extern int ptxinfo_lex_destroy(yyscan_t scanner); static bool g_save_embedded_ptx; static int g_occupancy_sm_number; -bool m_ptx_save_converted_ptxplus; bool ptxinfo_data::keep_intermediate_files() {return g_keep_intermediate_files;} @@ -71,7 +70,7 @@ void gpgpu_context::ptx_reg_options(option_parser_t opp) "keep intermediate files created by GPGPU-Sim when interfacing with external programs", "0"); option_parser_register(opp, "-gpgpu_ptx_save_converted_ptxplus", OPT_BOOL, - &m_ptx_save_converted_ptxplus, + &(ptxinfo->m_ptx_save_converted_ptxplus), "Saved converted ptxplus to a file", "0"); option_parser_register(opp, "-gpgpu_occupancy_sm_number", OPT_INT32, &g_occupancy_sm_number, @@ -106,7 +105,7 @@ void gpgpu_context::print_ptx_file( const char *p, unsigned source_num, const ch fflush(stdout); } -char* gpgpu_ptx_sim_convert_ptx_and_sass_to_ptxplus(const std::string ptxfilename, const std::string elffilename, const std::string sassfilename) +char* ptxinfo_data::gpgpu_ptx_sim_convert_ptx_and_sass_to_ptxplus(const std::string ptxfilename, const std::string elffilename, const std::string sassfilename) { printf("GPGPU-Sim PTX: converting EMBEDDED .ptx file to ptxplus \n"); diff --git a/src/cuda-sim/ptx_loader.h b/src/cuda-sim/ptx_loader.h index aa3e2f3..decfdc8 100644 --- a/src/cuda-sim/ptx_loader.h +++ b/src/cuda-sim/ptx_loader.h @@ -43,14 +43,14 @@ class ptxinfo_data{ const char *g_ptxinfo_filename; class gpgpu_context* gpgpu_ctx; bool g_keep_intermediate_files; + bool m_ptx_save_converted_ptxplus; void ptxinfo_addinfo(); bool keep_intermediate_files(); + char* gpgpu_ptx_sim_convert_ptx_and_sass_to_ptxplus(const std::string ptx_str, const std::string sass_str, const std::string elf_str); }; extern bool g_override_embedded_ptx; extern int no_of_ptx; //counter to track number of ptx files to be extracted in an application. -char* gpgpu_ptx_sim_convert_ptx_and_sass_to_ptxplus(const std::string ptx_str, const std::string sass_str, const std::string elf_str); - #endif -- cgit v1.3 From cda7a145b9e28eff0f3e9ac8197c2b6215755fc8 Mon Sep 17 00:00:00 2001 From: Mengchi Zhang Date: Tue, 9 Jul 2019 14:32:24 -0400 Subject: Move g_ptx_kernel_count Signed-off-by: Mengchi Zhang --- libcuda/cuda_api_object.h | 5 ++++- libcuda/cuda_runtime_api.cc | 2 +- libcuda/gpgpu_context.h | 2 +- libopencl/opencl_runtime_api.cc | 2 +- src/cuda-sim/cuda-sim.cc | 13 ++++++------- src/cuda-sim/cuda-sim.h | 14 ++++++++------ src/gpgpu-sim/gpu-sim.cc | 4 ++-- 7 files changed, 23 insertions(+), 19 deletions(-) (limited to 'libcuda/cuda_runtime_api.cc') diff --git a/libcuda/cuda_api_object.h b/libcuda/cuda_api_object.h index 0054697..db5e6a4 100644 --- a/libcuda/cuda_api_object.h +++ b/libcuda/cuda_api_object.h @@ -169,9 +169,10 @@ private: class cuda_runtime_api { public: - cuda_runtime_api() { + cuda_runtime_api( gpgpu_context* ctx ) { g_glbmap = NULL; g_active_device = 0; //active gpu that runs the code + gpgpu_ctx = ctx; } // global list std::list cuobjdumpSectionList; @@ -187,6 +188,8 @@ class cuda_runtime_api { std::map pinned_memory_size; glbmap_entry_t* g_glbmap; int g_active_device; //active gpu that runs the code + // backward pointer + class gpgpu_context* gpgpu_ctx; // member function list void cuobjdumpInit(); void extract_code_using_cuobjdump(); diff --git a/libcuda/cuda_runtime_api.cc b/libcuda/cuda_runtime_api.cc index 43c5e1f..43c8bae 100644 --- a/libcuda/cuda_runtime_api.cc +++ b/libcuda/cuda_runtime_api.cc @@ -3545,7 +3545,7 @@ kernel_info_t * cuda_runtime_api::gpgpu_cuda_ptx_sim_init_grid( const char *host } entry->finalize(result->get_param_memory()); - g_ptx_kernel_count++; + gpgpu_ctx->func_sim->g_ptx_kernel_count++; fflush(stdout); if(g_debug_execution >= 4){ diff --git a/libcuda/gpgpu_context.h b/libcuda/gpgpu_context.h index 2e21009..a2ae7b6 100644 --- a/libcuda/gpgpu_context.h +++ b/libcuda/gpgpu_context.h @@ -10,7 +10,7 @@ class gpgpu_context { public: gpgpu_context() { g_global_allfiles_symbol_table = NULL; - api = new cuda_runtime_api(); + api = new cuda_runtime_api(this); ptxinfo = new ptxinfo_data(this); ptx_parser = new ptx_recognizer(this); the_gpgpusim = new GPGPUsim_ctx(this); diff --git a/libopencl/opencl_runtime_api.cc b/libopencl/opencl_runtime_api.cc index d302ff8..0a6eb3e 100644 --- a/libopencl/opencl_runtime_api.cc +++ b/libopencl/opencl_runtime_api.cc @@ -956,7 +956,7 @@ clEnqueueNDRangeKernel(cl_command_queue command_queue, gpgpu_ptx_sim_memcpy_symbol( "%_global_launch_offset", zeros, 3 * sizeof(int), 0, 1, gpu ); gpgpu_ptx_sim_memcpy_symbol( "%_global_block_offset", zeros, 3 * sizeof(int), 0, 1, gpu ); } - kernel_info_t *grid = gpgpu_opencl_ptx_sim_init_grid(kernel->get_implementation(),params,GridDim,BlockDim,gpu); + kernel_info_t *grid = ctx->func_sim->gpgpu_opencl_ptx_sim_init_grid(kernel->get_implementation(),params,GridDim,BlockDim,gpu); if ( g_ptx_sim_mode ) ctx->func_sim->gpgpu_opencl_ptx_sim_main_func( grid ); else diff --git a/src/cuda-sim/cuda-sim.cc b/src/cuda-sim/cuda-sim.cc index e86395d..0ed125a 100644 --- a/src/cuda-sim/cuda-sim.cc +++ b/src/cuda-sim/cuda-sim.cc @@ -56,7 +56,6 @@ typedef void * yyscan_t; int gpgpu_ptx_instruction_classification; void ** g_inst_classification_stat = NULL; void ** g_inst_op_classification_stat= NULL; -int g_ptx_kernel_count = -1; // used for classification stat collection purposes int g_debug_execution = 0; int g_debug_thread_uid = 0; addr_t g_debug_pc = 0xBEEF1518; @@ -1480,7 +1479,7 @@ bool ptx_debug_exec_dump_cond(int thd_uid, addr_t pc) return false; } -void init_inst_classification_stat() +void cuda_sim::init_inst_classification_stat() { static std::set init; if( init.find(g_ptx_kernel_count) != init.end() ) @@ -1690,7 +1689,7 @@ void ptx_thread_info::ptx_exec_inst( warp_inst_t &inst, unsigned lane_id) ptx_file_line_stats_add_exec_count(pI); if ( gpgpu_ptx_instruction_classification ) { - init_inst_classification_stat(); + m_gpu->gpgpu_ctx->func_sim->init_inst_classification_stat(); unsigned space_type=0; switch ( pI->get_space().get_type() ) { case global_space: space_type = 10; break; @@ -1706,9 +1705,9 @@ void ptx_thread_info::ptx_exec_inst( warp_inst_t &inst, unsigned lane_id) space_type = 0 ; break; } - StatAddSample( g_inst_classification_stat[g_ptx_kernel_count], op_classification); - if (space_type) StatAddSample( g_inst_classification_stat[g_ptx_kernel_count], ( int )space_type); - StatAddSample( g_inst_op_classification_stat[g_ptx_kernel_count], (int) pI->get_opcode() ); + StatAddSample( g_inst_classification_stat[m_gpu->gpgpu_ctx->func_sim->g_ptx_kernel_count], op_classification); + if (space_type) StatAddSample( g_inst_classification_stat[m_gpu->gpgpu_ctx->func_sim->g_ptx_kernel_count], ( int )space_type); + StatAddSample( g_inst_op_classification_stat[m_gpu->gpgpu_ctx->func_sim->g_ptx_kernel_count], (int) pI->get_opcode() ); } if ( (m_gpu->gpgpu_ctx->func_sim->g_ptx_sim_num_insn % 100000) == 0 ) { dim3 ctaid = get_ctaid(); @@ -1917,7 +1916,7 @@ size_t get_kernel_code_size( class function_info *entry ) } -kernel_info_t *gpgpu_opencl_ptx_sim_init_grid(class function_info *entry, +kernel_info_t *cuda_sim::gpgpu_opencl_ptx_sim_init_grid(class function_info *entry, gpgpu_ptx_sim_arg_list_t args, struct dim3 gridDim, struct dim3 blockDim, diff --git a/src/cuda-sim/cuda-sim.h b/src/cuda-sim/cuda-sim.h index 16ee46e..e259f1f 100644 --- a/src/cuda-sim/cuda-sim.h +++ b/src/cuda-sim/cuda-sim.h @@ -46,13 +46,7 @@ extern int g_debug_execution; extern int g_debug_thread_uid; extern void ** g_inst_classification_stat; extern void ** g_inst_op_classification_stat; -extern int g_ptx_kernel_count; // used for classification stat collection purposes -extern class kernel_info_t *gpgpu_opencl_ptx_sim_init_grid(class function_info *entry, - gpgpu_ptx_sim_arg_list_t args, - struct dim3 gridDim, - struct dim3 blockDim, - class gpgpu_t *gpu ); extern void print_splash(); extern void gpgpu_ptx_sim_register_const_variable(void*, const char *deviceName, size_t size ); extern void gpgpu_ptx_sim_register_global_variable(void *hostVar, const char *deviceName, size_t size ); @@ -134,6 +128,7 @@ class cuda_sim { public: cuda_sim() { g_ptx_sim_num_insn = 0; + g_ptx_kernel_count = -1; // used for classification stat collection purposes } //global variables char *opcode_latency_int; @@ -151,10 +146,17 @@ class cuda_sim { int g_ptxinfo_error_detected; unsigned g_ptx_sim_num_insn; char *cdp_latency_str; + int g_ptx_kernel_count; // used for classification stat collection purposes //global functions void ptx_opcocde_latency_options (option_parser_t opp); void gpgpu_cuda_ptx_sim_main_func( kernel_info_t &kernel, bool openCL = false ); int gpgpu_opencl_ptx_sim_main_func( kernel_info_t *grid ); + void init_inst_classification_stat(); + kernel_info_t *gpgpu_opencl_ptx_sim_init_grid(class function_info *entry, + gpgpu_ptx_sim_arg_list_t args, + struct dim3 gridDim, + struct dim3 blockDim, + gpgpu_t *gpu ); }; #endif diff --git a/src/gpgpu-sim/gpu-sim.cc b/src/gpgpu-sim/gpu-sim.cc index 4f9ccbf..9f47067 100644 --- a/src/gpgpu-sim/gpu-sim.cc +++ b/src/gpgpu-sim/gpu-sim.cc @@ -1201,8 +1201,8 @@ void gpgpu_sim::gpu_print_stat() insn_warp_occ_print(stdout); } if ( gpgpu_ptx_instruction_classification ) { - StatDisp( g_inst_classification_stat[g_ptx_kernel_count]); - StatDisp( g_inst_op_classification_stat[g_ptx_kernel_count]); + StatDisp( g_inst_classification_stat[gpgpu_ctx->func_sim->g_ptx_kernel_count]); + StatDisp( g_inst_op_classification_stat[gpgpu_ctx->func_sim->g_ptx_kernel_count]); } #ifdef GPGPUSIM_POWER_MODEL -- cgit v1.3 From b3655bf28a7402db347f9d7f87049806b9315a05 Mon Sep 17 00:00:00 2001 From: Mengchi Zhang Date: Tue, 9 Jul 2019 15:03:18 -0400 Subject: Move g_global_name_lookup and g_constant_name_lookup Signed-off-by: Mengchi Zhang --- libcuda/cuda_runtime_api.cc | 4 ++-- src/cuda-sim/cuda-sim.cc | 14 ++++++-------- src/cuda-sim/cuda-sim.h | 6 ++++-- 3 files changed, 12 insertions(+), 12 deletions(-) (limited to 'libcuda/cuda_runtime_api.cc') diff --git a/libcuda/cuda_runtime_api.cc b/libcuda/cuda_runtime_api.cc index 43c8bae..bbbaf23 100644 --- a/libcuda/cuda_runtime_api.cc +++ b/libcuda/cuda_runtime_api.cc @@ -767,9 +767,9 @@ void cudaRegisterVarInternal( ctx->cuobjdumpParseBinary((unsigned)(unsigned long long)fatCubinHandle); fflush(stdout); if ( constant && !global && !ext ) { - gpgpu_ptx_sim_register_const_variable(hostVar,deviceName,size); + ctx->func_sim->gpgpu_ptx_sim_register_const_variable(hostVar,deviceName,size); } else if ( !constant && !global && !ext ) { - gpgpu_ptx_sim_register_global_variable(hostVar,deviceName,size); + ctx->func_sim->gpgpu_ptx_sim_register_global_variable(hostVar,deviceName,size); } else cuda_not_implemented(__my_func__,__LINE__); } diff --git a/src/cuda-sim/cuda-sim.cc b/src/cuda-sim/cuda-sim.cc index 0ed125a..b3e2965 100644 --- a/src/cuda-sim/cuda-sim.cc +++ b/src/cuda-sim/cuda-sim.cc @@ -1948,18 +1948,16 @@ void print_splash() } } -std::map g_const_name_lookup; // indexed by hostVar -std::map g_global_name_lookup; // indexed by hostVar std::set g_globals; std::set g_constants; -void gpgpu_ptx_sim_register_const_variable(void *hostVar, const char *deviceName, size_t size ) +void cuda_sim::gpgpu_ptx_sim_register_const_variable(void *hostVar, const char *deviceName, size_t size ) { printf("GPGPU-Sim PTX registering constant %s (%zu bytes) to name mapping\n", deviceName, size ); g_const_name_lookup[hostVar] = deviceName; } -void gpgpu_ptx_sim_register_global_variable(void *hostVar, const char *deviceName, size_t size ) +void cuda_sim::gpgpu_ptx_sim_register_global_variable(void *hostVar, const char *deviceName, size_t size ) { printf("GPGPU-Sim PTX registering global %s hostVar to name mapping\n", deviceName ); g_global_name_lookup[hostVar] = deviceName; @@ -1972,14 +1970,14 @@ void gpgpu_ptx_sim_memcpy_symbol(const char *hostVar, const void *src, size_t co memory_space_t mem_region = undefined_space; std::string sym_name; - std::map::iterator c=g_const_name_lookup.find(hostVar); - if ( c!=g_const_name_lookup.end() ) { + std::map::iterator c=gpu->gpgpu_ctx->func_sim->g_const_name_lookup.find(hostVar); + if ( c!=gpu->gpgpu_ctx->func_sim->g_const_name_lookup.end() ) { found_sym = true; sym_name = c->second; mem_region = const_space; } - std::map::iterator g=g_global_name_lookup.find(hostVar); - if ( g!=g_global_name_lookup.end() ) { + std::map::iterator g=gpu->gpgpu_ctx->func_sim->g_global_name_lookup.find(hostVar); + if ( g!=gpu->gpgpu_ctx->func_sim->g_global_name_lookup.end() ) { if ( found_sym ) { printf("Execution error: PTX symbol \"%s\" w/ hostVar=0x%Lx is declared both const and global?\n", sym_name.c_str(), (unsigned long long)hostVar ); diff --git a/src/cuda-sim/cuda-sim.h b/src/cuda-sim/cuda-sim.h index e259f1f..25ebf7b 100644 --- a/src/cuda-sim/cuda-sim.h +++ b/src/cuda-sim/cuda-sim.h @@ -48,8 +48,6 @@ extern void ** g_inst_classification_stat; extern void ** g_inst_op_classification_stat; extern void print_splash(); -extern void gpgpu_ptx_sim_register_const_variable(void*, const char *deviceName, size_t size ); -extern void gpgpu_ptx_sim_register_global_variable(void *hostVar, const char *deviceName, size_t size ); extern void gpgpu_ptx_sim_memcpy_symbol(const char *hostVar, const void *src, size_t count, size_t offset, int to, gpgpu_t *gpu ); extern void read_sim_environment_variables(); @@ -147,6 +145,8 @@ class cuda_sim { unsigned g_ptx_sim_num_insn; char *cdp_latency_str; int g_ptx_kernel_count; // used for classification stat collection purposes + std::map g_global_name_lookup; // indexed by hostVar + std::map g_const_name_lookup; // indexed by hostVar //global functions void ptx_opcocde_latency_options (option_parser_t opp); void gpgpu_cuda_ptx_sim_main_func( kernel_info_t &kernel, bool openCL = false ); @@ -157,6 +157,8 @@ class cuda_sim { struct dim3 gridDim, struct dim3 blockDim, gpgpu_t *gpu ); + void gpgpu_ptx_sim_register_global_variable(void *hostVar, const char *deviceName, size_t size ); + void gpgpu_ptx_sim_register_const_variable(void*, const char *deviceName, size_t size ); }; #endif -- cgit v1.3 From 4d4d5938d715d2b79a617c32583184426b4a642d Mon Sep 17 00:00:00 2001 From: Mengchi Zhang Date: Tue, 9 Jul 2019 23:16:17 -0400 Subject: Move g_ptx_sim_mode Signed-off-by: Mengchi Zhang --- libcuda/cuda_runtime_api.cc | 6 +++--- libcuda/gpgpu_context.h | 9 ++++++++- libopencl/opencl_runtime_api.cc | 6 +++--- src/cuda-sim/cuda-sim.cc | 6 ++---- src/cuda-sim/cuda-sim.h | 10 +++++++--- src/cuda-sim/cuda_device_runtime.cc | 7 ++++--- src/cuda-sim/cuda_device_runtime.h | 18 ++++++++++++++++-- src/gpgpu-sim/gpu-sim.cc | 4 ++-- src/gpgpusim_entrypoint.cc | 2 +- 9 files changed, 46 insertions(+), 22 deletions(-) (limited to 'libcuda/cuda_runtime_api.cc') diff --git a/libcuda/cuda_runtime_api.cc b/libcuda/cuda_runtime_api.cc index bbbaf23..59d2a60 100644 --- a/libcuda/cuda_runtime_api.cc +++ b/libcuda/cuda_runtime_api.cc @@ -1076,7 +1076,7 @@ cudaError_t cudaLaunchInternal( const char *hostFun, gpgpu_context* gpgpu_ctx = CUctx_st* context = GPGPUSim_Context(); char *mode = getenv("PTX_SIM_MODE_FUNC"); if( mode ) - sscanf(mode,"%u", &g_ptx_sim_mode); + sscanf(mode,"%u", &(ctx->func_sim->g_ptx_sim_mode)); gpgpusim_ptx_assert( !ctx->api->g_cuda_launch_stack.empty(), "empty launch stack" ); kernel_config config = ctx->api->g_cuda_launch_stack.back(); { @@ -1092,7 +1092,7 @@ cudaError_t cudaLaunchInternal( const char *hostFun, gpgpu_context* gpgpu_ctx = } struct CUstream_st *stream = config.get_stream(); printf("\nGPGPU-Sim PTX: cudaLaunch for 0x%p (mode=%s) on stream %u\n", hostFun, - g_ptx_sim_mode?"functional simulation":"performance simulation", stream?stream->get_uid():0 ); + (ctx->func_sim->g_ptx_sim_mode)?"functional simulation":"performance simulation", stream?stream->get_uid():0 ); kernel_info_t *grid = ctx->api->gpgpu_cuda_ptx_sim_init_grid(hostFun,config.get_args(),config.grid_dim(),config.block_dim(),context); //do dynamic PDOM analysis for performance simulation scenario std::string kname = grid->name(); @@ -1143,7 +1143,7 @@ cudaError_t cudaLaunchInternal( const char *hostFun, gpgpu_context* gpgpu_ctx = } printf("GPGPU-Sim PTX: pushing kernel \'%s\' to stream %u, gridDim= (%u,%u,%u) blockDim = (%u,%u,%u) \n", kname.c_str(), stream?stream->get_uid():0, gridDim.x,gridDim.y,gridDim.z,blockDim.x,blockDim.y,blockDim.z ); - stream_operation op(grid,g_ptx_sim_mode,stream); + stream_operation op(grid,ctx->func_sim->g_ptx_sim_mode,stream); g_stream_manager()->push(op); ctx->api->g_cuda_launch_stack.pop_back(); return g_last_cudaError = cudaSuccess; diff --git a/libcuda/gpgpu_context.h b/libcuda/gpgpu_context.h index a2ae7b6..3c9f87c 100644 --- a/libcuda/gpgpu_context.h +++ b/libcuda/gpgpu_context.h @@ -5,6 +5,7 @@ #include "../src/cuda-sim/ptx_parser.h" #include "../src/gpgpusim_entrypoint.h" #include "../src/cuda-sim/cuda-sim.h" +#include "../src/cuda-sim/cuda_device_runtime.h" class gpgpu_context { public: @@ -14,7 +15,10 @@ class gpgpu_context { ptxinfo = new ptxinfo_data(this); ptx_parser = new ptx_recognizer(this); the_gpgpusim = new GPGPUsim_ctx(this); - func_sim = new cuda_sim(); + func_sim = new cuda_sim(this); +#if (CUDART_VERSION >= 5000) + device_runtime = new cuda_device_runtime(this); +#endif } // global list symbol_table *g_global_allfiles_symbol_table; @@ -25,6 +29,9 @@ class gpgpu_context { ptx_recognizer* ptx_parser; GPGPUsim_ctx* the_gpgpusim; cuda_sim* func_sim; +#if (CUDART_VERSION >= 5000) + cuda_device_runtime* device_runtime; +#endif // member function list void cuobjdumpParseBinary(unsigned int handle); class symbol_table *gpgpu_ptx_sim_load_ptx_from_string( const char *p, unsigned source_num ); diff --git a/libopencl/opencl_runtime_api.cc b/libopencl/opencl_runtime_api.cc index 0a6eb3e..aaaec4f 100644 --- a/libopencl/opencl_runtime_api.cc +++ b/libopencl/opencl_runtime_api.cc @@ -884,9 +884,9 @@ clEnqueueNDRangeKernel(cl_command_queue command_queue, printf("\n\n\n"); char *mode = getenv("PTX_SIM_MODE_FUNC"); if ( mode ) - sscanf(mode,"%u", &g_ptx_sim_mode); + sscanf(mode,"%u", &(ctx->func_sim->g_ptx_sim_mode)); printf("GPGPU-Sim OpenCL API: clEnqueueNDRangeKernel '%s' (mode=%s)\n", kernel->name().c_str(), - g_ptx_sim_mode?"functional simulation":"performance simulation"); + (ctx->func_sim->g_ptx_sim_mode)?"functional simulation":"performance simulation"); if ( !work_dim || work_dim > 3 ) return CL_INVALID_WORK_DIMENSION; size_t _local_size[3]; if( local_work_size != NULL ) { @@ -957,7 +957,7 @@ clEnqueueNDRangeKernel(cl_command_queue command_queue, gpgpu_ptx_sim_memcpy_symbol( "%_global_block_offset", zeros, 3 * sizeof(int), 0, 1, gpu ); } kernel_info_t *grid = ctx->func_sim->gpgpu_opencl_ptx_sim_init_grid(kernel->get_implementation(),params,GridDim,BlockDim,gpu); - if ( g_ptx_sim_mode ) + if ( ctx->func_sim->g_ptx_sim_mode ) ctx->func_sim->gpgpu_opencl_ptx_sim_main_func( grid ); else gpgpu_opencl_ptx_sim_main_perf( grid ); diff --git a/src/cuda-sim/cuda-sim.cc b/src/cuda-sim/cuda-sim.cc index b3e2965..7a7d205 100644 --- a/src/cuda-sim/cuda-sim.cc +++ b/src/cuda-sim/cuda-sim.cc @@ -2033,13 +2033,11 @@ void gpgpu_ptx_sim_memcpy_symbol(const char *hostVar, const void *src, size_t co fflush(stdout); } -int g_ptx_sim_mode; // if non-zero run functional simulation only (i.e., no notion of a clock cycle) - extern int ptx_debug; bool g_cuda_launch_blocking = false; -void read_sim_environment_variables() +void cuda_sim::read_sim_environment_variables() { ptx_debug = 0; g_debug_execution = 0; @@ -2185,7 +2183,7 @@ void cuda_sim::gpgpu_cuda_ptx_sim_main_func( kernel_info_t &kernel, bool openCL cta.execute(cp_count,temp); #if (CUDART_VERSION >= 5000) - launch_all_device_kernels(); + gpgpu_ctx->device_runtime->launch_all_device_kernels(); #endif } else diff --git a/src/cuda-sim/cuda-sim.h b/src/cuda-sim/cuda-sim.h index 25ebf7b..3c4336d 100644 --- a/src/cuda-sim/cuda-sim.h +++ b/src/cuda-sim/cuda-sim.h @@ -36,12 +36,12 @@ #include #include"ptx_sim.h" +class gpgpu_context; class memory_space; class function_info; class symbol_table; extern const char *g_gpgpusim_version_string; -extern int g_ptx_sim_mode; extern int g_debug_execution; extern int g_debug_thread_uid; extern void ** g_inst_classification_stat; @@ -50,7 +50,6 @@ extern void ** g_inst_op_classification_stat; extern void print_splash(); extern void gpgpu_ptx_sim_memcpy_symbol(const char *hostVar, const void *src, size_t count, size_t offset, int to, gpgpu_t *gpu ); -extern void read_sim_environment_variables(); extern void ptxinfo_opencl_addinfo( std::map &kernels ); unsigned ptx_sim_init_thread( kernel_info_t &kernel, class ptx_thread_info** thread_info, @@ -124,9 +123,10 @@ struct gpgpu_ptx_sim_info get_ptxinfo(); class cuda_sim { public: - cuda_sim() { + cuda_sim( gpgpu_context* ctx ) { g_ptx_sim_num_insn = 0; g_ptx_kernel_count = -1; // used for classification stat collection purposes + gpgpu_ctx = ctx; } //global variables char *opcode_latency_int; @@ -147,6 +147,9 @@ class cuda_sim { int g_ptx_kernel_count; // used for classification stat collection purposes std::map g_global_name_lookup; // indexed by hostVar std::map g_const_name_lookup; // indexed by hostVar + int g_ptx_sim_mode; // if non-zero run functional simulation only (i.e., no notion of a clock cycle) + // backward pointer + class gpgpu_context* gpgpu_ctx; //global functions void ptx_opcocde_latency_options (option_parser_t opp); void gpgpu_cuda_ptx_sim_main_func( kernel_info_t &kernel, bool openCL = false ); @@ -159,6 +162,7 @@ class cuda_sim { gpgpu_t *gpu ); void gpgpu_ptx_sim_register_global_variable(void *hostVar, const char *deviceName, size_t size ); void gpgpu_ptx_sim_register_const_variable(void*, const char *deviceName, size_t size ); + void read_sim_environment_variables(); }; #endif diff --git a/src/cuda-sim/cuda_device_runtime.cc b/src/cuda-sim/cuda_device_runtime.cc index be8369f..354fa79 100644 --- a/src/cuda-sim/cuda_device_runtime.cc +++ b/src/cuda-sim/cuda_device_runtime.cc @@ -20,6 +20,7 @@ unsigned long long g_max_total_param_size = 0; #include "../stream_manager.h" #include "../gpgpusim_entrypoint.h" #include "cuda_device_runtime.h" +#include "../../libcuda/gpgpu_context.h" #define DEV_RUNTIME_REPORT(a) \ if( g_debug_execution ) { \ @@ -318,17 +319,17 @@ void gpgpusim_cuda_streamCreateWithFlags(const ptx_instruction * pI, ptx_thread_ } -void launch_one_device_kernel() { +void cuda_device_runtime::launch_one_device_kernel() { if(!g_cuda_device_launch_op.empty()) { device_launch_operation_t &op = g_cuda_device_launch_op.front(); - stream_operation stream_op = stream_operation(op.grid, g_ptx_sim_mode, op.stream); + stream_operation stream_op = stream_operation(op.grid, gpgpu_ctx->func_sim->g_ptx_sim_mode, op.stream); g_stream_manager()->push(stream_op); g_cuda_device_launch_op.pop_front(); } } -void launch_all_device_kernels() { +void cuda_device_runtime::launch_all_device_kernels() { while(!g_cuda_device_launch_op.empty()) { launch_one_device_kernel(); } diff --git a/src/cuda-sim/cuda_device_runtime.h b/src/cuda-sim/cuda_device_runtime.h index 6dbcd71..851fed2 100644 --- a/src/cuda-sim/cuda_device_runtime.h +++ b/src/cuda-sim/cuda_device_runtime.h @@ -6,6 +6,20 @@ void gpgpusim_cuda_getParameterBufferV2(const ptx_instruction * pI, ptx_thread_info * thread, const function_info * target_func); void gpgpusim_cuda_launchDeviceV2(const ptx_instruction * pI, ptx_thread_info * thread, const function_info * target_func); void gpgpusim_cuda_streamCreateWithFlags(const ptx_instruction * pI, ptx_thread_info * thread, const function_info * target_func); -void launch_all_device_kernels(); -void launch_one_device_kernel(); +#endif +#if (CUDART_VERSION >= 5000) + +class gpgpu_context; + +class cuda_device_runtime { + public: + cuda_device_runtime( gpgpu_context* ctx ) { + gpgpu_ctx = ctx; + } + // backward pointer + class gpgpu_context* gpgpu_ctx; + void launch_all_device_kernels(); + void launch_one_device_kernel(); +}; + #endif diff --git a/src/gpgpu-sim/gpu-sim.cc b/src/gpgpu-sim/gpu-sim.cc index 9f47067..bbcc078 100644 --- a/src/gpgpu-sim/gpu-sim.cc +++ b/src/gpgpu-sim/gpu-sim.cc @@ -503,7 +503,7 @@ void gpgpu_sim_config::reg_options(option_parser_t opp) &gpgpu_ptx_instruction_classification, "if enabled will classify ptx instruction types per kernel (Max 255 kernels now)", "0"); - option_parser_register(opp, "-gpgpu_ptx_sim_mode", OPT_INT32, &g_ptx_sim_mode, + option_parser_register(opp, "-gpgpu_ptx_sim_mode", OPT_INT32, &(gpgpu_ctx->func_sim->g_ptx_sim_mode), "Select between Performance (default) or Functional simulation (1)", "0"); option_parser_register(opp, "-gpgpu_clock_domains", OPT_CSTR, &gpgpu_clock_domains, @@ -1753,7 +1753,7 @@ void gpgpu_sim::cycle() #if (CUDART_VERSION >= 5000) //launch device kernel - launch_one_device_kernel(); + gpgpu_ctx->device_runtime->launch_one_device_kernel(); #endif } } diff --git a/src/gpgpusim_entrypoint.cc b/src/gpgpusim_entrypoint.cc index d9d1023..683a695 100644 --- a/src/gpgpusim_entrypoint.cc +++ b/src/gpgpusim_entrypoint.cc @@ -213,7 +213,7 @@ gpgpu_sim *gpgpu_context::gpgpu_ptx_sim_init_perf() { srand(1); print_splash(); - read_sim_environment_variables(); + func_sim->read_sim_environment_variables(); ptx_parser->read_parser_environment_variables(); option_parser_t opp = option_parser_create(); -- cgit v1.3 From 4671280cfe7252bf875d071e667f57064250a1b7 Mon Sep 17 00:00:00 2001 From: Mengchi Zhang Date: Fri, 12 Jul 2019 00:09:29 -0400 Subject: Move g_cdp_enabled Signed-off-by: Mengchi Zhang --- libcuda/cuda_api_object.h | 1 + libcuda/cuda_runtime_api.cc | 11 ++++------- src/abstract_hardware_model.cc | 3 --- src/cuda-sim/cuda_device_runtime.h | 2 ++ src/cuda-sim/ptx_loader.cc | 9 +++------ src/gpgpu-sim/gpu-sim.cc | 3 +-- 6 files changed, 11 insertions(+), 18 deletions(-) (limited to 'libcuda/cuda_runtime_api.cc') diff --git a/libcuda/cuda_api_object.h b/libcuda/cuda_api_object.h index db5e6a4..51416f2 100644 --- a/libcuda/cuda_api_object.h +++ b/libcuda/cuda_api_object.h @@ -199,6 +199,7 @@ class cuda_runtime_api { std::list mergeSections(); cuobjdumpELFSection* findELFSection(const std::string identifier); cuobjdumpPTXSection* findPTXSection(const std::string identifier); + cuobjdumpPTXSection* findPTXSectionInList(std::list §ionlist, const std::string identifier); void cuobjdumpRegisterFatBinary(unsigned int handle, const char* filename, CUctx_st *context); kernel_info_t *gpgpu_cuda_ptx_sim_init_grid( const char *kernel_key, gpgpu_ptx_sim_arg_list_t args, diff --git a/libcuda/cuda_runtime_api.cc b/libcuda/cuda_runtime_api.cc index 59d2a60..10a651a 100644 --- a/libcuda/cuda_runtime_api.cc +++ b/libcuda/cuda_runtime_api.cc @@ -2421,7 +2421,6 @@ __host__ cudaError_t CUDARTAPI cudaGetExportTable(const void **ppExportTable, co //extracts all ptx files from binary and dumps into prog_name.unique_no.sm_<>.ptx files void cuda_runtime_api::extract_ptx_files_using_cuobjdump(CUctx_st *context){ - extern bool g_cdp_enabled; char command[1000]; char *pytorch_bin = getenv("PYTORCH_BIN"); std::string app_binary = get_app_binary(); @@ -2442,7 +2441,7 @@ void cuda_runtime_api::extract_ptx_files_using_cuobjdump(CUctx_st *context){ printf("WARNING: Failed to execute cuobjdump to get list of ptx files \n"); exit(0); } - if(!g_cdp_enabled) { + if(!gpgpu_ctx->device_runtime->g_cdp_enabled) { //based on the list above, dump ptx files individually. Format of dumped ptx file is prog_name.unique_no.sm_<>.ptx std::ifstream infile(ptx_list_file_name); @@ -2515,7 +2514,6 @@ void cuda_runtime_api::extract_code_using_cuobjdump(){ } // Running cuobjdump using dynamic link to current process // Needs the option '-all' to extract PTX from CDP-enabled binary - extern bool g_cdp_enabled; //dump ptx for all individial ptx files into sepearte files which is later used by ptxas. int result=0; @@ -2530,7 +2528,7 @@ void cuda_runtime_api::extract_code_using_cuobjdump(){ snprintf(fname,1024,"_cuobjdump_complete_output_XXXXXX"); int fd=mkstemp(fname); close(fd); - if(!g_cdp_enabled) + if(!gpgpu_ctx->device_runtime->g_cdp_enabled) snprintf(command,1000,"$CUDA_INSTALL_PATH/bin/cuobjdump -ptx -elf -sass %s > %s", app_binary.c_str(), fname); else snprintf(command,1000,"$CUDA_INSTALL_PATH/bin/cuobjdump -ptx -elf -sass -all %s > %s", app_binary.c_str(), fname); @@ -2822,7 +2820,7 @@ cuobjdumpELFSection* cuda_runtime_api::findELFSection(const std::string identifi } //! Within the section list, find the PTX section corresponding to a given identifier -cuobjdumpPTXSection* findPTXSectionInList(std::list §ionlist, const std::string identifier){ +cuobjdumpPTXSection* cuda_runtime_api::findPTXSectionInList(std::list §ionlist, const std::string identifier){ std::list::iterator iter; for ( iter = sectionlist.begin(); iter != sectionlist.end(); @@ -2833,8 +2831,7 @@ cuobjdumpPTXSection* findPTXSectionInList(std::list §ionl if(ptxsection->getIdentifier() == identifier) return ptxsection; else { - extern bool g_cdp_enabled; - if(g_cdp_enabled) { + if(gpgpu_ctx->device_runtime->g_cdp_enabled) { printf("Warning: __cudaRegisterFatBinary needs %s, but find PTX section with %s\n", identifier.c_str(), ptxsection->getIdentifier().c_str()); return ptxsection; diff --git a/src/abstract_hardware_model.cc b/src/abstract_hardware_model.cc index 450087b..fde7874 100644 --- a/src/abstract_hardware_model.cc +++ b/src/abstract_hardware_model.cc @@ -691,9 +691,6 @@ void warp_inst_t::completed( unsigned long long cycle ) const ptx_file_line_stats_add_latency(pc, latency * active_count()); } -//Jin: CDP support -bool g_cdp_enabled; - unsigned kernel_info_t::m_next_uid = 1; /*A snapshot of the texture mappings needs to be stored in the kernel's info as diff --git a/src/cuda-sim/cuda_device_runtime.h b/src/cuda-sim/cuda_device_runtime.h index f37849b..7f7a0ca 100644 --- a/src/cuda-sim/cuda_device_runtime.h +++ b/src/cuda-sim/cuda_device_runtime.h @@ -51,6 +51,8 @@ class cuda_device_runtime { std::list g_cuda_device_launch_op; unsigned g_kernel_launch_latency; unsigned long long g_max_total_param_size; + bool g_cdp_enabled; + // backward pointer class gpgpu_context* gpgpu_ctx; #if (CUDART_VERSION >= 5000) diff --git a/src/cuda-sim/ptx_loader.cc b/src/cuda-sim/ptx_loader.cc index 6e36a62..dca3cec 100644 --- a/src/cuda-sim/ptx_loader.cc +++ b/src/cuda-sim/ptx_loader.cc @@ -330,8 +330,7 @@ void gpgpu_context::gpgpu_ptx_info_load_from_filename( const char *filename, uns std::string ptxas_filename(std::string(filename) + "as"); char buff[1024], extra_flags[1024]; extra_flags[0]=0; - extern bool g_cdp_enabled; - if(!g_cdp_enabled) + if(!device_runtime->g_cdp_enabled) snprintf(extra_flags,1024,"--gpu-name=sm_%u",sm_version); else snprintf(extra_flags,1024,"--compile-only --gpu-name=sm_%u",sm_version); @@ -398,8 +397,7 @@ void gpgpu_context::gpgpu_ptxinfo_load_from_string( const char *p_for_info, unsi "A register size/SM mismatch may result in occupancy differences." ); exit(1); } - extern bool g_cdp_enabled; - if(!g_cdp_enabled) + if(!device_runtime->g_cdp_enabled) snprintf(extra_flags,1024,"--gpu-name=sm_%u", g_occupancy_sm_number); else snprintf(extra_flags,1024,"--compile-only --gpu-name=sm_%u",g_occupancy_sm_number); @@ -467,8 +465,7 @@ void gpgpu_context::gpgpu_ptxinfo_load_from_string( const char *p_for_info, unsi #if CUDART_VERSION >= 3000 if (sm_version == 0) sm_version = 20; - extern bool g_cdp_enabled; - if(!g_cdp_enabled) + if(!device_runtime->g_cdp_enabled) snprintf(extra_flags,1024,"--gpu-name=sm_%u",sm_version); else snprintf(extra_flags,1024,"--compile-only --gpu-name=sm_%u",sm_version); diff --git a/src/gpgpu-sim/gpu-sim.cc b/src/gpgpu-sim/gpu-sim.cc index a3d6a8a..0481259 100644 --- a/src/gpgpu-sim/gpu-sim.cc +++ b/src/gpgpu-sim/gpu-sim.cc @@ -551,9 +551,8 @@ void gpgpu_sim_config::reg_options(option_parser_t opp) option_parser_register(opp, "-gpgpu_kernel_launch_latency", OPT_INT32, &(gpgpu_ctx->device_runtime->g_kernel_launch_latency), "Kernel launch latency in cycles. Default: 0", "0"); - extern bool g_cdp_enabled; option_parser_register(opp, "-gpgpu_cdp_enabled", OPT_BOOL, - &g_cdp_enabled, "Turn on CDP", + &(gpgpu_ctx->device_runtime->g_cdp_enabled), "Turn on CDP", "0"); } -- cgit v1.3 From 17815dfdd7b6e9b558997bc5a9ba157a0493da16 Mon Sep 17 00:00:00 2001 From: Mahmoud Date: Fri, 12 Jul 2019 21:31:54 -0400 Subject: fixing device and function parameters config --- configs/tested-cfgs/SM2_GTX480/gpgpusim.config | 1 + configs/tested-cfgs/SM6_TITANX/gpgpusim.config | 1 + configs/tested-cfgs/SM7_QV100/gpgpusim.config | 11 +++++++---- configs/tested-cfgs/SM7_TITANV/gpgpusim.config | 3 +++ libcuda/cuda_runtime_api.cc | 21 ++++++++++++--------- src/gpgpu-sim/gpu-sim.cc | 4 ++-- 6 files changed, 26 insertions(+), 15 deletions(-) (limited to 'libcuda/cuda_runtime_api.cc') diff --git a/configs/tested-cfgs/SM2_GTX480/gpgpusim.config b/configs/tested-cfgs/SM2_GTX480/gpgpusim.config index cf3627b..4a7a3c3 100644 --- a/configs/tested-cfgs/SM2_GTX480/gpgpusim.config +++ b/configs/tested-cfgs/SM2_GTX480/gpgpusim.config @@ -61,6 +61,7 @@ # Note: Hashing set index function (H) only applies to a set size of 32 or 64. -gpgpu_cache:dl1 N:32:128:4,L:L:m:N:H,S:64:8,8 -gpgpu_shmem_size 49152 +-gpgpu_shmem_sizeDefault 49152 -icnt_flit_size 40 -gmem_skip_L1D 0 -gpgpu_n_cluster_ejection_buffer_size 32 diff --git a/configs/tested-cfgs/SM6_TITANX/gpgpusim.config b/configs/tested-cfgs/SM6_TITANX/gpgpusim.config index 2fe898a..e6d8f1d 100644 --- a/configs/tested-cfgs/SM6_TITANX/gpgpusim.config +++ b/configs/tested-cfgs/SM6_TITANX/gpgpusim.config @@ -81,6 +81,7 @@ -gpgpu_cache:dl1PrefL1 S:4:128:48,L:L:s:N:L,A:256:8,16:0,32 -gpgpu_cache:dl1PrefShared S:4:128:48,L:L:s:N:L,A:256:8,16:0,32 -gpgpu_shmem_size 49152 +-gpgpu_shmem_sizeDefault 49152 -gpgpu_shmem_size_PrefL1 49152 -gpgpu_shmem_size_PrefShared 49152 # By default, L1 cache is disabled in Pascal P102. diff --git a/configs/tested-cfgs/SM7_QV100/gpgpusim.config b/configs/tested-cfgs/SM7_QV100/gpgpusim.config index 1a34d0f..5f64908 100644 --- a/configs/tested-cfgs/SM7_QV100/gpgpusim.config +++ b/configs/tested-cfgs/SM7_QV100/gpgpusim.config @@ -11,7 +11,7 @@ # functional simulator specification -gpgpu_ptx_instruction_classification 0 -gpgpu_ptx_sim_mode 0 --gpgpu_ptx_force_max_capability 60 +-gpgpu_ptx_force_max_capability 70 # Device Limits @@ -21,7 +21,7 @@ -gpgpu_runtime_pending_launch_count_limit 2048 # Compute Capability --gpgpu_compute_capability_major 6 +-gpgpu_compute_capability_major 7 -gpgpu_compute_capability_minor 0 # SASS execution (only supported with CUDA >= 4.0) @@ -44,12 +44,13 @@ # shader core pipeline config -gpgpu_shader_registers 65536 --gpgpu_occupancy_sm_number 60 +-gpgpu_registers_per_block 65536 +-gpgpu_occupancy_sm_number 70 # This implies a maximum of 64 warps/SM -gpgpu_shader_core_pipeline 2048:32 -gpgpu_shader_cta 32 --gpgpu_simd_model 1 +-gpgpu_simd_model 1 # Pipeline widths and number of FUs # ID_OC_SP,ID_OC_DP,ID_OC_INT,ID_OC_SFU,ID_OC_MEM,OC_EX_SP,OC_EX_DP,OC_EX_INT,OC_EX_SFU,OC_EX_MEM,EX_WB,ID_OC_TENSOR_CORE,OC_EX_TENSOR_CORE @@ -91,6 +92,8 @@ -mem_unit_ports 4 -gpgpu_cache:dl1 S:4:128:64,L:L:s:N:L,A:256:8,16:0,32 -gpgpu_shmem_size 98304 +-gpgpu_shmem_sizeDefault 98304 +-gpgpu_shmem_per_block 65536 -gmem_skip_L1D 0 -icnt_flit_size 40 -gpgpu_n_cluster_ejection_buffer_size 32 diff --git a/configs/tested-cfgs/SM7_TITANV/gpgpusim.config b/configs/tested-cfgs/SM7_TITANV/gpgpusim.config index b06f048..6c21dcb 100644 --- a/configs/tested-cfgs/SM7_TITANV/gpgpusim.config +++ b/configs/tested-cfgs/SM7_TITANV/gpgpusim.config @@ -44,6 +44,7 @@ # shader core pipeline config -gpgpu_shader_registers 65536 +-gpgpu_registers_per_block 65536 -gpgpu_occupancy_sm_number 70 # This implies a maximum of 64 warps/SM @@ -91,6 +92,8 @@ -mem_unit_ports 4 -gpgpu_cache:dl1 S:4:128:64,L:L:s:N:L,A:256:8,16:0,32 -gpgpu_shmem_size 98304 +-gpgpu_shmem_sizeDefault 98304 +-gpgpu_shmem_per_block 65536 -gmem_skip_L1D 0 -icnt_flit_size 40 -gpgpu_n_cluster_ejection_buffer_size 32 diff --git a/libcuda/cuda_runtime_api.cc b/libcuda/cuda_runtime_api.cc index c70a570..45511d4 100644 --- a/libcuda/cuda_runtime_api.cc +++ b/libcuda/cuda_runtime_api.cc @@ -383,7 +383,7 @@ struct _cuda_device_id *GPGPUSim_Init() prop->sharedMemPerMultiprocessor = the_gpu->shared_mem_size(); #endif prop->sharedMemPerBlock = the_gpu->shared_mem_per_block(); - prop->regsPerBlock = the_gpu->num_registers_per_core(); + prop->regsPerBlock = the_gpu->num_registers_per_block(); prop->warpSize = the_gpu->wrp_size(); prop->clockRate = the_gpu->shader_clock(); #if (CUDART_VERSION >= 2010) @@ -1014,7 +1014,7 @@ __host__ cudaError_t CUDARTAPI cudaDeviceGetAttribute(int *value, enum cudaDevic prop = dev->get_prop(); switch (attr) { case 1: - *value= prop->maxThreadsDim[0] * prop->maxThreadsDim[1] * prop->maxThreadsDim[2] * prop->maxGridSize[0] * prop->maxGridSize[1] * prop->maxGridSize[2]; + *value= prop->maxThreadsPerBlock; break; case 2: *value= prop->maxThreadsDim[0]; @@ -1504,13 +1504,13 @@ __host__ cudaError_t CUDARTAPI cudaLaunch( const char *hostFun ) { dim3 gridDim = config.grid_dim(); dim3 blockDim = config.block_dim(); - if (gridDim.x * gridDim.y * gridDim.z == 0 || blockDim.x * blockDim.y * blockDim.z == 0) - { + //if (gridDim.x * gridDim.y * gridDim.z == 0 || blockDim.x * blockDim.y * blockDim.z == 0) + //{ //can't launch - printf("can't launch a empty kernel\n"); - g_cuda_launch_stack.pop_back(); - return g_last_cudaError = cudaErrorInvalidConfiguration; - } + // printf("can't launch a empty kernel\n"); + // g_cuda_launch_stack.pop_back(); + // return g_last_cudaError = cudaErrorInvalidConfiguration; + //} } struct CUstream_st *stream = config.get_stream(); if(g_stream_manager->is_blocking()) @@ -3151,9 +3151,12 @@ size_t getMaxThreadsPerBlock(struct cudaFuncAttributes *attr) { size_t max = prop.maxThreadsPerBlock; - if ((prop.regsPerBlock / attr->numRegs) < max) + if (attr->numRegs && (prop.regsPerBlock / attr->numRegs) < max) max = prop.regsPerBlock / attr->numRegs; + if (attr->sharedSizeBytes && (prop.sharedMemPerBlock / attr->sharedSizeBytes) < max) + max = prop.sharedMemPerBlock / attr->sharedSizeBytes; + return max; } diff --git a/src/gpgpu-sim/gpu-sim.cc b/src/gpgpu-sim/gpu-sim.cc index 6de5845..72cb32b 100644 --- a/src/gpgpu-sim/gpu-sim.cc +++ b/src/gpgpu-sim/gpu-sim.cc @@ -320,7 +320,7 @@ void shader_core_config::reg_options(class OptionParser * opp) option_parser_register(opp, "-adaptive_volta_cache_config", OPT_BOOL, &adaptive_volta_cache_config, "adaptive_volta_cache_config", "0"); - option_parser_register(opp, "-gpgpu_shmem_size", OPT_UINT32, &gpgpu_shmem_sizeDefault, + option_parser_register(opp, "-gpgpu_shmem_sizeDefault", OPT_UINT32, &gpgpu_shmem_sizeDefault, "Size of shared memory per shader core (default 16kB)", "16384"); option_parser_register(opp, "-gpgpu_shmem_size_PrefL1", OPT_UINT32, &gpgpu_shmem_sizePrefL1, @@ -1065,7 +1065,7 @@ void gpgpu_sim::change_cache_config(FuncCache cache_config) if(cache_config != m_shader_config->m_L1D_config.get_cache_status()){ printf("FLUSH L1 Cache at configuration change between kernels\n"); for (unsigned i=0;in_simt_clusters;i++) { - m_cluster[i]->cache_flush(); + m_cluster[i]->cache_invalidate(); } } -- cgit v1.3 From 963947b33c99143afbb477a6d897245e56695b0b Mon Sep 17 00:00:00 2001 From: Mahmoud Date: Fri, 23 Aug 2019 13:22:20 -0400 Subject: fixing cuda 4 error --- libcuda/cuda_runtime_api.cc | 8 +++----- 1 file changed, 3 insertions(+), 5 deletions(-) (limited to 'libcuda/cuda_runtime_api.cc') diff --git a/libcuda/cuda_runtime_api.cc b/libcuda/cuda_runtime_api.cc index 9d685e6..e71db4c 100644 --- a/libcuda/cuda_runtime_api.cc +++ b/libcuda/cuda_runtime_api.cc @@ -1607,7 +1607,7 @@ cudaError_t CUDARTAPI cudaOccupancyMaxActiveBlocksPerMultiprocessorWithFlags(int #endif -#if (CUDART_VERSION > 5000) + /******************************************************************************* * * @@ -1625,6 +1625,7 @@ __host__ cudaError_t CUDARTAPI cudaMemset(void *mem, int c, size_t count) return g_last_cudaError = cudaSuccess; } +#if (CUDART_VERSION > 5000) __host__ cudaError_t CUDARTAPI cudaDeviceGetAttributeInternal(int *value, enum cudaDeviceAttr attr, int device, gpgpu_context* gpgpu_ctx = NULL) { gpgpu_context *ctx; @@ -1823,6 +1824,7 @@ __host__ cudaError_t CUDARTAPI cudaDeviceGetAttributeInternal(int *value, enum c return g_last_cudaError = cudaErrorInvalidDevice; } } +#endif //memset operation is done but i think its not async? __host__ cudaError_t CUDARTAPI cudaMemsetAsync(void *mem, int c, size_t count, cudaStream_t stream=0) @@ -3296,8 +3298,6 @@ cudaError_t CUDARTAPI cudaSetDeviceFlags( int flags ) } - - cudaError_t CUDARTAPI cudaFuncGetAttributes(struct cudaFuncAttributes *attr, const char *hostFun ) { return cudaFuncGetAttributesInternal(attr, hostFun ); @@ -3352,8 +3352,6 @@ __host__ cudaError_t CUDARTAPI cudaDeviceSetLimit(enum cudaLimit limit, size_t v #endif -#endif - #if CUDART_VERSION >= 9000 /** -- cgit v1.3 From 2a6788b59055b5ce694882a282af0cc6311854d4 Mon Sep 17 00:00:00 2001 From: Nick Date: Mon, 26 Aug 2019 13:42:10 -0400 Subject: Fix a bunch of outstanding warnings and undefined behavior --- libcuda/cuda_runtime_api.cc | 10 ++--- src/abstract_hardware_model.h | 18 ++++----- src/cuda-sim/cuda-sim.cc | 18 ++------- src/cuda-sim/instructions.cc | 74 ++++++++++++++++++------------------- src/cuda-sim/ptx_parser.cc | 2 +- src/gpgpu-sim/addrdec.cc | 2 +- src/gpgpu-sim/addrdec.h | 2 +- src/gpgpu-sim/dram.cc | 9 ++--- src/gpgpu-sim/gpu-cache.cc | 6 +-- src/gpgpu-sim/gpu-sim.cc | 10 ++--- src/gpgpu-sim/l2cache.cc | 6 +-- src/gpgpu-sim/local_interconnect.cc | 2 +- src/gpgpu-sim/scoreboard.cc | 4 +- src/gpgpu-sim/shader.cc | 14 +++---- src/gpgpu-sim/shader.h | 6 +-- 15 files changed, 82 insertions(+), 101 deletions(-) (limited to 'libcuda/cuda_runtime_api.cc') diff --git a/libcuda/cuda_runtime_api.cc b/libcuda/cuda_runtime_api.cc index 10a651a..43a5864 100644 --- a/libcuda/cuda_runtime_api.cc +++ b/libcuda/cuda_runtime_api.cc @@ -548,7 +548,7 @@ __host__ cudaError_t CUDARTAPI cudaDeviceGetLimitInternal( size_t* pValue, cudaL break; } else{ - printf("ERROR:Limit %s is not supported on this architecture \n",limit); + printf("ERROR:Limit %d is not supported on this architecture \n", limit); abort(); } case 4: // cudaLimitDevRuntimePendingLaunchCount @@ -557,12 +557,12 @@ __host__ cudaError_t CUDARTAPI cudaDeviceGetLimitInternal( size_t* pValue, cudaL break; } else{ - printf("ERROR:Limit %s is not supported on this architecture \n",limit); + printf("ERROR:Limit %d is not supported on this architecture \n",limit); abort(); } #endif default: - printf("ERROR:Limit %s unimplemented \n",limit); + printf("ERROR:Limit %d unimplemented \n",limit); abort(); } return g_last_cudaError = cudaSuccess; @@ -2471,7 +2471,6 @@ void cuda_runtime_api::extract_ptx_files_using_cuobjdump(CUctx_st *context){ while (std::getline(infile, line)) { //int pos = line.find(std::string(get_app_binary_name(app_binary))); - const char *ptx_file = line.c_str(); int pos1 = line.find("sm_"); int pos2 = line.find_last_of("."); if (pos1==std::string::npos&&pos2==std::string::npos){ @@ -2499,11 +2498,10 @@ void cuda_runtime_api::extract_ptx_files_using_cuobjdump(CUctx_st *context){ * */ void cuda_runtime_api::extract_code_using_cuobjdump(){ CUctx_st *context = GPGPUSim_Context(); - unsigned forced_max_capability = context->get_device()->get_gpgpu()->get_config().get_forced_max_capability(); //prevent the dumping by cuobjdump everytime we execute the code! const char *override_cuobjdump = getenv("CUOBJDUMP_SIM_FILE"); - char command[1000], ptx_file[1000]; + char command[1000]; std::string app_binary = get_app_binary(); //Running cuobjdump using dynamic link to current process snprintf(command,1000,"md5sum %s ", app_binary.c_str()); diff --git a/src/abstract_hardware_model.h b/src/abstract_hardware_model.h index d13b8c6..1982e04 100644 --- a/src/abstract_hardware_model.h +++ b/src/abstract_hardware_model.h @@ -519,10 +519,10 @@ private: int checkpoint_option; int checkpoint_kernel; int checkpoint_CTA; - int resume_option; - int resume_kernel; - int resume_CTA; - int checkpoint_CTA_t; + unsigned resume_option; + unsigned resume_kernel; + unsigned resume_CTA; + unsigned checkpoint_CTA_t; int checkpoint_insn_Y; int g_ptx_inst_debug_to_file; char* g_ptx_inst_debug_file; @@ -540,10 +540,10 @@ public: int checkpoint_option; int checkpoint_kernel; int checkpoint_CTA; - int resume_option; - int resume_kernel; - int resume_CTA; - int checkpoint_CTA_t; + unsigned resume_option; + unsigned resume_kernel; + unsigned resume_CTA; + unsigned checkpoint_CTA_t; int checkpoint_insn_Y; //Move some cycle core stats here instead of being global @@ -992,7 +992,7 @@ public: printf("Printing mem access generated\n"); std::list::iterator it; for (it = m_accessq.begin(); it != m_accessq.end(); ++it){ - printf("MEM_TXN_GEN:%s:%x, Size:%d \n",mem_access_type_str(it->get_type()), it->get_addr(),it->get_size()); + printf("MEM_TXN_GEN:%s:%llx, Size:%d \n",mem_access_type_str(it->get_type()), it->get_addr(),it->get_size()); } } } diff --git a/src/cuda-sim/cuda-sim.cc b/src/cuda-sim/cuda-sim.cc index b9e6552..f8d0b3e 100644 --- a/src/cuda-sim/cuda-sim.cc +++ b/src/cuda-sim/cuda-sim.cc @@ -1491,7 +1491,6 @@ static unsigned get_tex_datasize( const ptx_instruction *pI, ptx_thread_info *th const operand_info &src1 = pI->src1(); //the name of the texture std::string texname = src1.name(); - gpgpu_t *gpu = thread->get_gpu(); /* For programs with many streams, textures can be bound and unbound asynchronously. This means we need to use the kernel's "snapshot" of @@ -1577,7 +1576,7 @@ void ptx_thread_info::ptx_exec_inst( warp_inst_t &inst, unsigned lane_id) } //Tensorcore is warp synchronous operation. So these instructions needs to be executed only once. To make the simulation faster removing the redundant tensorcore operation - if(!tensorcore_op(inst_opcode)||(tensorcore_op(inst_opcode))&&(lane_id==0)){ + if(!tensorcore_op(inst_opcode)||((tensorcore_op(inst_opcode))&&(lane_id==0))){ switch ( inst_opcode ) { #define OP_DEF(OP,FUNC,STR,DST,CLASSIFICATION) case OP: FUNC(pI,this); op_classification = CLASSIFICATION; break; #define OP_W_DEF(OP,FUNC,STR,DST,CLASSIFICATION) case OP: FUNC(pI,get_core(),inst); op_classification = CLASSIFICATION; break; @@ -2141,13 +2140,7 @@ void cuda_sim::gpgpu_cuda_ptx_sim_main_func( kernel_info_t &kernel, bool openCL unsigned max_cta_tot = max_cta(kernel_info,kernel.threads_per_cta(), g_the_gpu()->getShaderCoreConfig()->warp_size, g_the_gpu()->getShaderCoreConfig()->n_thread_per_shader, g_the_gpu()->getShaderCoreConfig()->gpgpu_shmem_size, g_the_gpu()->getShaderCoreConfig()->gpgpu_shader_registers, g_the_gpu()->getShaderCoreConfig()->max_cta_per_core); printf("Max CTA : %d\n",max_cta_tot); - - - - - int inst_count=50; int cp_op= g_the_gpu()->checkpoint_option; - int cp_CTA = g_the_gpu()->checkpoint_CTA; int cp_kernel= g_the_gpu()->checkpoint_kernel; cp_count= g_the_gpu()->checkpoint_insn_Y; cp_cta_resume= g_the_gpu()->checkpoint_CTA_t; @@ -2184,7 +2177,7 @@ void cuda_sim::gpgpu_cuda_ptx_sim_main_func( kernel_info_t &kernel, bool openCL { char f1name[2048]; snprintf(f1name,2048,"checkpoint_files/global_mem_%d.txt", kernel.get_uid() ); - g_checkpoint->store_global_mem(g_the_gpu()->get_global_memory(), f1name , "%08x"); + g_checkpoint->store_global_mem(g_the_gpu()->get_global_memory(), f1name , (char *)"%08x"); } @@ -2312,18 +2305,15 @@ void functionalCoreSim::execute(int inst_count, unsigned ctaid_cp) checkpoint *g_checkpoint; g_checkpoint = new checkpoint(); - symbol * sym; ptx_reg_t regval; regval.u64= 123; - symbol_table * symtab= m_kernel->entry()->get_symtab(); - unsigned ctaid =m_kernel->get_next_cta_id_single(); if(m_gpu->checkpoint_option==1 && (m_kernel->get_uid()==m_gpu->checkpoint_kernel) && (ctaid_cp>=m_gpu->checkpoint_CTA) && (ctaid_cpcheckpoint_CTA_t)) { char fname[2048]; snprintf(fname,2048,"checkpoint_files/shared_mem_%d.txt",ctaid-1 ); - g_checkpoint->store_global_mem(m_thread[0]->m_shared_mem, fname , "%08x"); + g_checkpoint->store_global_mem(m_thread[0]->m_shared_mem, fname , (char *)"%08x"); for(int i=0; i<32*m_warp_count;i++) { char fname[2048]; @@ -2331,7 +2321,7 @@ void functionalCoreSim::execute(int inst_count, unsigned ctaid_cp) m_thread[i]->print_reg_thread(fname); char f1name[2048]; snprintf(f1name,2048,"checkpoint_files/local_mem_thread_%d_%d_reg.txt",i,ctaid-1 ); - g_checkpoint->store_global_mem(m_thread[i]->m_local_mem, f1name , "%08x"); + g_checkpoint->store_global_mem(m_thread[i]->m_local_mem, f1name , (char *)"%08x"); m_thread[i]->set_done(); m_thread[i]->exitCore(); m_thread[i]->registerExit(); diff --git a/src/cuda-sim/instructions.cc b/src/cuda-sim/instructions.cc index 58a077e..a44b03f 100644 --- a/src/cuda-sim/instructions.cc +++ b/src/cuda-sim/instructions.cc @@ -203,7 +203,7 @@ void ptx_thread_info::print_reg_thread(char * fname) const std::string &name = it->first->name(); const std::string &dec= it->first->decl_location(); unsigned size = it->first->get_size_in_bytes(); - fprintf(fp,"%s %llu %s %d\n",name.c_str(),it->second, dec.c_str(),size ); + fprintf(fp,"%s %llu %s %d\n", name.c_str(), it->second, dec.c_str(), size); } //m_regs.pop_back(); @@ -232,7 +232,6 @@ void ptx_thread_info::resume_reg_thread(char * fname, symbol_table * symtab) pch = strtok (NULL," "); data = atoi(pch); pch = strtok (NULL," "); - char * decl= pch; pch = strtok (NULL," "); size = atoi(pch); @@ -1819,9 +1818,7 @@ void mma_impl( const ptx_instruction *pI, core_t *core, warp_inst_t inst ) ptx_reg_t matrix_d[16][16]; ptx_reg_t src_data; ptx_thread_info *thread; - int stride; - unsigned wmma_type = pI->get_wmma_type(); unsigned a_layout = pI->get_wmma_layout(0); unsigned b_layout = pI->get_wmma_layout(1); unsigned type = pI->get_type(); @@ -1833,7 +1830,6 @@ void mma_impl( const ptx_instruction *pI, core_t *core, warp_inst_t inst ) tid= inst.warp_id_func()*core->get_warp_size(); else tid= inst.warp_id()*core->get_warp_size(); - unsigned thread_group_index; float temp; half temp2; @@ -1847,9 +1843,9 @@ void mma_impl( const ptx_instruction *pI, core_t *core, warp_inst_t inst ) ptx_reg_t v[8]; thread->get_vector_operand_values( src_a, v, nelem ); if(core->get_gpu()->gpgpu_ctx->debug_tensorcore){ - printf("Thread%d_Iteration=%d\n:",thrd,operand_num); - for(k=0;kset_vector_operand_values(dst,nw_data1,nw_data2,nw_data3,nw_data4); if(core->get_gpu()->gpgpu_ctx->debug_tensorcore) - printf("thread%d=%x,%x,%x,%x",thrd,nw_data1.s64,nw_data2.s64,nw_data3.s64,nw_data4.s64); + printf("thread%d=%llx,%llx,%llx,%llx", thrd, nw_data1.s64, nw_data2.s64, nw_data3.s64, nw_data4.s64); } else{ @@ -2298,9 +2294,8 @@ unsigned int saturatei(unsigned int a, unsigned int max) ptx_reg_t f2x( ptx_reg_t x, unsigned from_width, unsigned to_width, int to_sign, int rounding_mode, int saturation_mode ) { - half mytemp; - float myfloat; - half_float::half tmp_h; + half mytemp; + half_float::half tmp_h; //assert( from_width == 32); enum cudaRoundMode mode = cudaRoundZero; @@ -3085,7 +3080,7 @@ void mma_st_impl( const ptx_instruction *pI, core_t *core, warp_inst_t &inst ) size_t size; unsigned smid; int t; - int thrd,odd,inx,k; + int thrd, odd, inx, k; ptx_thread_info *thread; const operand_info &src = pI->operand_lookup(1); @@ -3105,15 +3100,15 @@ void mma_st_impl( const ptx_instruction *pI, core_t *core, warp_inst_t &inst ) _memory_op_t insn_memory_op = pI->has_memory_read() ? memory_load : memory_store; for (thrd=0; thrd < core->get_warp_size(); thrd++) { thread = core->get_thread_info()[tid+thrd]; - odd=thrd%2; - inx=thrd/2; - ptx_reg_t addr_reg = thread->get_operand_value(src1, src, type, thread, 1); + odd= thrd % 2; + inx= thrd / 2; + ptx_reg_t addr_reg = thread->get_operand_value(src1, src, type, thread, 1); ptx_reg_t src2_data = thread->get_operand_value(src2, src, type, thread, 1); const operand_info &src_a= pI->operand_lookup(1); unsigned nelem = src_a.get_vect_nelem(); ptx_reg_t* v= new ptx_reg_t[8]; thread->get_vector_operand_values( src_a, v, nelem ); - stride=src2_data.u32; + stride = src2_data.u32; memory_space_t space = pI->get_space(); @@ -3130,9 +3125,9 @@ void mma_st_impl( const ptx_instruction *pI, core_t *core, warp_inst_t &inst ) } decode_space(space,thread,src1,mem,addr); - type_info_key::type_decode(type,size,t); + type_info_key::type_decode(type, size, t); if(core->get_gpu()->gpgpu_ctx->debug_tensorcore) - printf("mma_st: thrd=%d,addr=%x, fp(size=%d), stride=%d\n",thrd,addr_reg.u32,size,src2_data.u32); + printf("mma_st: thrd=%d, addr=%x, fp(size=%zu), stride=%d\n", thrd, addr_reg.u32, size, src2_data.u32); addr_t new_addr = addr+thread_group_offset(thrd,wmma_type,wmma_layout,type,stride)*size/8; addr_t push_addr; @@ -3152,7 +3147,7 @@ void mma_st_impl( const ptx_instruction *pI, core_t *core, warp_inst_t &inst ) mem_txn_addr[num_mem_txn++]=push_addr; if(core->get_gpu()->gpgpu_ctx->debug_tensorcore){ - printf("wmma:store:thread%d=%x,%x,%x,%x,%x,%x,%x,%x\n",thrd,v[0].s64,v[1].s64,v[2].s64,v[3].s64,v[4].s64,v[5].s64,v[6].s64,v[7].s64); + printf("wmma:store:thread%d=%llx,%llx,%llx,%llx,%llx,%llx,%llx,%llx\n",thrd,v[0].s64,v[1].s64,v[2].s64,v[3].s64,v[4].s64,v[5].s64,v[6].s64,v[7].s64); float temp; int l; printf("thread=%d:",thrd); @@ -3179,7 +3174,7 @@ void mma_st_impl( const ptx_instruction *pI, core_t *core, warp_inst_t &inst ) } if(core->get_gpu()->gpgpu_ctx->debug_tensorcore) - printf("wmma:store:thread%d=%x,%x,%x,%x,%x,%x,%x,%x\n",thrd,nw_v[0].s64,nw_v[1].s64,nw_v[2].s64,nw_v[3].s64,nw_v[4].s64,nw_v[5].s64,nw_v[6].s64,nw_v[7].s64); + printf("wmma:store:thread%d=%llx,%llx,%llx,%llx,%llx,%llx,%llx,%llx\n",thrd,nw_v[0].s64,nw_v[1].s64,nw_v[2].s64,nw_v[3].s64,nw_v[4].s64,nw_v[5].s64,nw_v[6].s64,nw_v[7].s64); } } @@ -3238,11 +3233,11 @@ void mma_ld_impl( const ptx_instruction *pI, core_t *core, warp_inst_t &inst ) } decode_space(space,thread,src1,mem,addr); - type_info_key::type_decode(type,size,t); + type_info_key::type_decode(type, size, t); ptx_reg_t data[16]; if(core->get_gpu()->gpgpu_ctx->debug_tensorcore) - printf("mma_ld: thrd=%d,addr=%x, fpsize=%d, stride=%d\n",thrd,src1_data.u32,size,src2_data.u32); + printf("mma_ld: thrd=%d,addr=%x, fpsize=%zu, stride=%d\n", thrd, src1_data.u32, size, src2_data.u32); addr_t new_addr = addr+thread_group_offset(thrd,wmma_type,wmma_layout,type,stride)*size/8; addr_t fetch_addr; @@ -3341,7 +3336,7 @@ void mma_ld_impl( const ptx_instruction *pI, core_t *core, warp_inst_t &inst ) if(type==F16_TYPE){ printf("\nmma_ld:thread%d= ",thrd); for(i=0;i<16;i++){ - printf("%x ",data[i].u64); + printf("%llx ",data[i].u64); } printf("\n"); @@ -3361,7 +3356,7 @@ void mma_ld_impl( const ptx_instruction *pI, core_t *core, warp_inst_t &inst ) printf("\n"); printf("\nmma_ld:thread%d= ",thrd); for(i=0;i<8;i++){ - printf("%x ",data[i].u64); + printf("%llx ",data[i].u64); } printf("\n"); } @@ -3388,15 +3383,15 @@ void mma_ld_impl( const ptx_instruction *pI, core_t *core, warp_inst_t &inst ) else thread->set_wmma_vector_operand_values(dst,nw_data[0],nw_data[1],nw_data[2],nw_data[3],nw_data[4],nw_data[5],nw_data[6],nw_data[7]); if(core->get_gpu()->gpgpu_ctx->debug_tensorcore){ - printf("mma_ld:data[0].s64=%x,data[1].s64=%x,new_data[0].s64=%x\n",data[0].u64,data[1].u64,nw_data[0].u64); - printf("mma_ld:data[2].s64=%x,data[3].s64=%x,new_data[1].s64=%x\n",data[2].u64,data[3].u64,nw_data[1].u64); - printf("mma_ld:data[4].s64=%x,data[5].s64=%x,new_data[2].s64=%x\n",data[4].u64,data[5].u64,nw_data[2].u64); - printf("mma_ld:data[6].s64=%x,data[7].s64=%x,new_data[3].s64=%x\n",data[6].u64,data[7].u64,nw_data[3].u64); + printf("mma_ld:data[0].s64=%llx,data[1].s64=%llx,new_data[0].s64=%llx\n",data[0].u64,data[1].u64,nw_data[0].u64); + printf("mma_ld:data[2].s64=%llx,data[3].s64=%llx,new_data[1].s64=%llx\n",data[2].u64,data[3].u64,nw_data[1].u64); + printf("mma_ld:data[4].s64=%llx,data[5].s64=%llx,new_data[2].s64=%llx\n",data[4].u64,data[5].u64,nw_data[2].u64); + printf("mma_ld:data[6].s64=%llx,data[7].s64=%llx,new_data[3].s64=%llx\n",data[6].u64,data[7].u64,nw_data[3].u64); if(wmma_type!=LOAD_C){ - printf("mma_ld:data[8].s64=%x,data[9].s64=%x,new_data[4].s64=%x\n",data[8].u64,data[9].u64,nw_data[4].s64); - printf("mma_ld:data[10].s64=%x,data[11].s64=%x,new_data[5].s64=%x\n",data[10].u64,data[11].u64,nw_data[5].u64); - printf("mma_ld:data[12].s64=%x,data[13].s64=%x,new_data[6].s64=%x\n",data[12].u64,data[13].u64,nw_data[6].u64); - printf("mma_ld:data[14].s64=%x,data[15].s64=%x,new_data[7].s64=%x\n",data[14].u64,data[15].u64,nw_data[3].u64); + printf("mma_ld:data[8].s64=%llx,data[9].s64=%llx,new_data[4].s64=%llx\n",data[8].u64,data[9].u64,nw_data[4].s64); + printf("mma_ld:data[10].s64=%llx,data[11].s64=%llx,new_data[5].s64=%llx\n",data[10].u64,data[11].u64,nw_data[5].u64); + printf("mma_ld:data[12].s64=%llx,data[13].s64=%llx,new_data[6].s64=%llx\n",data[12].u64,data[13].u64,nw_data[6].u64); + printf("mma_ld:data[14].s64=%llx,data[15].s64=%llx,new_data[7].s64=%llx\n",data[14].u64,data[15].u64,nw_data[3].u64); } } } @@ -4132,9 +4127,9 @@ int prmt_mode_present(int mode) } return returnval; } -int read_byte(int mode,int control,int d_sel_index,signed long long value){ +int read_byte(int mode, int control, int d_sel_index, signed long long value){ - int returnval; + int returnval = 0; int prmt_f4e_mode[4][4]={{0,1,2,3},{1,2,3,4},{2,3,4,5},{3,4,5,6}}; int prmt_b4e_mode[4][4]={{0,7,6,5},{1,0,7,6},{2,1,0,7},{3,2,1,0}}; int prmt_rc8_mode[4][4]={{0,0,0,0},{1,1,1,1},{2,2,2,2},{3,3,3,3}}; @@ -4157,11 +4152,12 @@ int read_byte(int mode,int control,int d_sel_index,signed long long value){ case PRMT_RC8_MODE: returnval=prmt_rc8_mode[control][d_sel_index];break; case PRMT_ECL_MODE: returnval=prmt_ecl_mode[control][d_sel_index];break; case PRMT_ECR_MODE: returnval=prmt_ecr_mode[control][d_sel_index];break; - case PRMT_RC16_MODE: returnval=prmt_rc16_mode[control][d_sel_index];break; - default: printf("ERROR\n");break; + case PRMT_RC16_MODE: returnval=prmt_rc16_mode[control][d_sel_index];break; + // Change the default from printing "ERROR" to just asserting + default: assert(false); } } - return (returnval<<8*d_sel_index); + return (returnval << 8 * d_sel_index); } void prmt_impl( const ptx_instruction *pI, ptx_thread_info *thread ) { diff --git a/src/cuda-sim/ptx_parser.cc b/src/cuda-sim/ptx_parser.cc index 81b70af..a4f4a0c 100644 --- a/src/cuda-sim/ptx_parser.cc +++ b/src/cuda-sim/ptx_parser.cc @@ -421,7 +421,7 @@ void ptx_recognizer::add_identifier( const char *identifier, int array_dim, unsi assert( (num_bits%8) == 0 ); addr = g_current_symbol_table->get_sstarr_next(); addr_pad = pad_address(addr, num_bits/8, 128); - printf("from 0x%x to 0x%lx (sstarr memory space)\n", + printf("from 0x%llx to 0x%llx (sstarr memory space)\n", addr+addr_pad, addr+addr_pad + num_bits/8); fflush(stdout); diff --git a/src/gpgpu-sim/addrdec.cc b/src/gpgpu-sim/addrdec.cc index ca88ec9..09bbc3c 100644 --- a/src/gpgpu-sim/addrdec.cc +++ b/src/gpgpu-sim/addrdec.cc @@ -182,7 +182,7 @@ void linear_to_raw_address_translation::addrdec_tlx(new_addr_type addr, addrdec_ } assert(tlx->chip < m_n_channel); - assert(tlx->sub_partition < m_n_channel*m_n_sub_partition_in_channel); + assert(tlx->sub_partition < m_n_channel * m_n_sub_partition_in_channel); return; break; } diff --git a/src/gpgpu-sim/addrdec.h b/src/gpgpu-sim/addrdec.h index a5333fb..c9a1420 100644 --- a/src/gpgpu-sim/addrdec.h +++ b/src/gpgpu-sim/addrdec.h @@ -92,7 +92,7 @@ private: new_addr_type sub_partition_id_mask; unsigned int gap; - int m_n_channel; + unsigned m_n_channel; int m_n_sub_partition_in_channel; }; diff --git a/src/gpgpu-sim/dram.cc b/src/gpgpu-sim/dram.cc index d443d79..9c33822 100644 --- a/src/gpgpu-sim/dram.cc +++ b/src/gpgpu-sim/dram.cc @@ -482,7 +482,6 @@ void dram_t::cycle() bool memory_pending_rw_found=false; for (unsigned j=0;jnbk;j++) { - unsigned grp = get_bankgrp_number(j); if (bk[j]->mrq && (((bk[j]->curr_row == bk[j]->mrq->row) && (bk[j]->mrq->rw == READ) && (bk[j]->state == BANK_ACTIVE)) @@ -817,10 +816,10 @@ void dram_t::visualize() const void dram_t::print_stat( FILE* simFile ) { - fprintf(simFile,"DRAM (%llu): n_cmd=%llu n_nop=%llu n_act=%llu n_pre=%llu n_ref=%llu n_req=%llu n_rd=%llu n_write=%llu bw_util=%.4g ", + fprintf(simFile,"DRAM (%u): n_cmd=%llu n_nop=%llu n_act=%llu n_pre=%llu n_ref=%llu n_req=%llu n_rd=%llu n_write=%llu bw_util=%.4g ", id, n_cmd, n_nop, n_act, n_pre, n_ref, n_req, n_rd, n_wr, (float)bwutil/n_cmd); - fprintf(simFile, "mrqq: %d %.4g mrqsmax=%d ", max_mrqs, (float)ave_mrqs/n_cmd, max_mrqs_temp); + fprintf(simFile, "mrqq: %d %.4g mrqsmax=%llu ", max_mrqs, (float)ave_mrqs/n_cmd, max_mrqs_temp); fprintf(simFile, "\n"); fprintf(simFile, "dram_util_bins:"); for (unsigned i=0;i<10;i++) fprintf(simFile, " %d", dram_util_bins[i]); @@ -899,10 +898,10 @@ void dram_t::set_dram_power_stats( unsigned &cmd, unsigned dram_t::get_bankgrp_number(unsigned i) { if(m_config->dram_bnkgrp_indexing_policy == HIGHER_BITS) { //higher bits - return i>>m_config->bk_tag_length; + return i >> m_config->bk_tag_length; } else if (m_config->dram_bnkgrp_indexing_policy == LOWER_BITS) { //lower bits - return i&((m_config->nbkgrp-1)); + return i & ((m_config->nbkgrp - 1)); } else { assert(1); diff --git a/src/gpgpu-sim/gpu-cache.cc b/src/gpgpu-sim/gpu-cache.cc index 1705821..dec61db 100644 --- a/src/gpgpu-sim/gpu-cache.cc +++ b/src/gpgpu-sim/gpu-cache.cc @@ -777,7 +777,7 @@ void cache_stats::print_stats(FILE *fout, const char *cache_name) const{ } for (unsigned type = 0; type < NUM_MEM_ACCESS_TYPE; ++type) { if(total_access[type] > 0) - fprintf(fout, "\t%s[%s][%s] = %llu\n", + fprintf(fout, "\t%s[%s][%s] = %u\n", m_cache_name.c_str(), mem_access_type_str((enum mem_access_type)type), "TOTAL_ACCESS", @@ -790,7 +790,7 @@ void cache_stats::print_fail_stats(FILE *fout, const char *cache_name) const{ for (unsigned type = 0; type < NUM_MEM_ACCESS_TYPE; ++type) { for (unsigned fail = 0; fail < NUM_CACHE_RESERVATION_FAIL_STATUS; ++fail) { if(m_fail_stats[type][fail] > 0){ - fprintf(fout, "\t%s[%s][%s] = %u\n", + fprintf(fout, "\t%s[%s][%s] = %llu\n", m_cache_name.c_str(), mem_access_type_str((enum mem_access_type)type), cache_fail_status_str((enum cache_reservation_fail_reason)fail), @@ -1417,8 +1417,6 @@ data_cache::wr_miss_wa_lazy_fetch_on_read( new_addr_type addr, { new_addr_type block_addr = m_config.block_addr(addr); - new_addr_type mshr_addr = m_config.mshr_addr(mf->get_addr()); - //if the request writes to the whole cache line/sector, then, write and set cache line Modified. //and no need to send read request to memory or reserve mshr diff --git a/src/gpgpu-sim/gpu-sim.cc b/src/gpgpu-sim/gpu-sim.cc index e4ae04f..1f9a422 100644 --- a/src/gpgpu-sim/gpu-sim.cc +++ b/src/gpgpu-sim/gpu-sim.cc @@ -1101,8 +1101,8 @@ void gpgpu_sim::gpu_print_stat() printf("gpu_tot_sim_insn = %lld\n", gpu_tot_sim_insn+gpu_sim_insn); printf("gpu_tot_ipc = %12.4f\n", (float)(gpu_tot_sim_insn+gpu_sim_insn) / (gpu_tot_sim_cycle+gpu_sim_cycle)); printf("gpu_tot_issued_cta = %lld\n", gpu_tot_issued_cta + m_total_cta_launched); - printf("gpu_occupancy = %.4f\% \n", gpu_occupancy.get_occ_fraction() * 100); - printf("gpu_tot_occupancy = %.4f\% \n", (gpu_occupancy + gpu_tot_occupancy).get_occ_fraction() * 100); + printf("gpu_occupancy = %.4f%% \n", gpu_occupancy.get_occ_fraction() * 100); + printf("gpu_tot_occupancy = %.4f%% \n", (gpu_occupancy + gpu_tot_occupancy).get_occ_fraction() * 100); fprintf(statfout, "max_total_param_size = %llu\n", gpgpu_ctx->device_runtime->g_max_total_param_size); @@ -1343,7 +1343,7 @@ bool shader_core_ctx::occupy_shader_resource_1block(kernel_info_t & k, bool occu m_occupied_regs += (padded_cta_size * ((kernel_info->regs+3)&~3)); m_occupied_ctas++; - SHADER_DPRINTF(LIVENESS, "GPGPU-Sim uArch: Occupied %d threads, %d shared mem, %d registers, %d ctas\n", + SHADER_DPRINTF(LIVENESS, "GPGPU-Sim uArch: Occupied %u threads, %u shared mem, %u registers, %u ctas\n", m_occupied_n_threads, m_occupied_shmem, m_occupied_regs, m_occupied_ctas); } @@ -1460,7 +1460,7 @@ void shader_core_ctx::issue_block2core( kernel_info_t &kernel ) nthreads_in_block += ptx_sim_init_thread(kernel,&m_thread[i],m_sid,i,cta_size-(i-start_thread),m_config->n_thread_per_shader,this,free_cta_hw_id,warp_id,m_cluster->get_gpu()); m_threadState[i].m_active = true; // load thread local memory and register file - if(m_gpu->resume_option==1 && kernel.get_uid()==m_gpu->resume_kernel && ctaid>=m_gpu->resume_CTA && ctaidcheckpoint_CTA_t ) + if(m_gpu->resume_option == 1 && kernel.get_uid() == m_gpu->resume_kernel && ctaid >= m_gpu->resume_CTA && ctaid < m_gpu->checkpoint_CTA_t ) { char fname[2048]; snprintf(fname,2048,"checkpoint_files/thread_%d_%d_reg.txt",i%cta_size,ctaid ); @@ -1475,7 +1475,7 @@ void shader_core_ctx::issue_block2core( kernel_info_t &kernel ) assert( nthreads_in_block > 0 && nthreads_in_block <= m_config->n_thread_per_shader); // should be at least one, but less than max m_cta_status[free_cta_hw_id]=nthreads_in_block; - if(m_gpu->resume_option==1 && kernel.get_uid()==m_gpu->resume_kernel && ctaid>=m_gpu->resume_CTA && ctaidcheckpoint_CTA_t ) + if(m_gpu->resume_option == 1 && kernel.get_uid() == m_gpu->resume_kernel && ctaid >= m_gpu->resume_CTA && ctaid < m_gpu->checkpoint_CTA_t ) { char f1name[2048]; snprintf(f1name,2048,"checkpoint_files/shared_mem_%d.txt", ctaid); diff --git a/src/gpgpu-sim/l2cache.cc b/src/gpgpu-sim/l2cache.cc index 6540b52..862461f 100644 --- a/src/gpgpu-sim/l2cache.cc +++ b/src/gpgpu-sim/l2cache.cc @@ -82,7 +82,7 @@ void memory_partition_unit::handle_memcpy_to_gpu( size_t addr, unsigned global_s unsigned p = global_sub_partition_id_to_local_id(global_subpart_id); std::string mystring = mask.to_string(); - MEMPART_DPRINTF("Copy Engine Request Received For Address=%llx, local_subpart=%u, global_subpart=%u, sector_mask=%s \n", addr, p, global_subpart_id, mystring.c_str()); + MEMPART_DPRINTF("Copy Engine Request Received For Address=%zx, local_subpart=%u, global_subpart=%u, sector_mask=%s \n", addr, p, global_subpart_id, mystring.c_str()); m_sub_partition[p]->force_l2_tag_update(addr,m_gpu->gpu_sim_cycle+m_gpu->gpu_tot_sim_cycle, mask); } @@ -622,7 +622,7 @@ std::vector memory_sub_partition::breakdown_request_to_sector_reques } } else { - printf("Invalid sector received, address = 0x%06x, sector mask = %s, data size = %d", + printf("Invalid sector received, address = 0x%06llx, sector mask = %s, data size = %d", mf->get_addr(), mf->get_access_sector_mask(), mf->get_data_size()); assert(0 && "Undefined sector mask is received"); } @@ -657,7 +657,7 @@ std::vector memory_sub_partition::breakdown_request_to_sector_reques byte_sector_mask <<= SECTOR_SIZE; } } else { - printf("Invalid sector received, address = 0x%06x, sector mask = %d, byte mask = , data size = %d", + printf("Invalid sector received, address = 0x%06llx, sector mask = %d, byte mask = , data size = %u", mf->get_addr(), mf->get_access_sector_mask().count(), mf->get_data_size()); assert(0 && "Undefined data size is received"); } diff --git a/src/gpgpu-sim/local_interconnect.cc b/src/gpgpu-sim/local_interconnect.cc index 1416b2c..bb09d44 100644 --- a/src/gpgpu-sim/local_interconnect.cc +++ b/src/gpgpu-sim/local_interconnect.cc @@ -231,7 +231,7 @@ LocalInterconnect::LocalInterconnect(const struct inct_config& m_localinct_confi } LocalInterconnect::~LocalInterconnect(){ - for (int i=0; i inst_regs; - for(int iii=0;iiioutcount;iii++) + for(unsigned iii=0; iii < inst->outcount; iii++) inst_regs.insert(inst->out[iii]); - for(int jjj=0;jjjincount;jjj++) + for(unsigned jjj=0;jjjincount;jjj++) inst_regs.insert(inst->in[jjj]); if(inst->pred > 0) inst_regs.insert(inst->pred); diff --git a/src/gpgpu-sim/shader.cc b/src/gpgpu-sim/shader.cc index c697450..c365ebb 100644 --- a/src/gpgpu-sim/shader.cc +++ b/src/gpgpu-sim/shader.cc @@ -87,7 +87,7 @@ shader_core_ctx::shader_core_ctx( class gpgpu_sim *gpu, shader_core_stats *stats ) : core_t( gpu, NULL, config->warp_size, config->n_thread_per_shader ), m_barriers( this, config->max_warps_per_shader, config->max_cta_per_core, config->max_barriers_per_cta, config->warp_size ), - m_dynamic_warp_id(0), m_active_warps(0) + m_active_warps(0), m_dynamic_warp_id(0) { m_cluster = cluster; m_config = config; @@ -164,7 +164,7 @@ shader_core_ctx::shader_core_ctx( class gpgpu_sim *gpu, NUM_CONCRETE_SCHEDULERS; assert ( scheduler != NUM_CONCRETE_SCHEDULERS ); - for (int i = 0; i < m_config->gpgpu_num_sched_per_core; i++) { + for (unsigned i = 0; i < m_config->gpgpu_num_sched_per_core; i++) { switch( scheduler ) { case CONCRETE_SCHEDULER_LRR: @@ -263,7 +263,7 @@ shader_core_ctx::shader_core_ctx( class gpgpu_sim *gpu, //distribute i's evenly though schedulers; schedulers[i%m_config->gpgpu_num_sched_per_core]->add_supervised_warp_id(i); } - for ( int i = 0; i < m_config->gpgpu_num_sched_per_core; ++i ) { + for ( unsigned i = 0; i < m_config->gpgpu_num_sched_per_core; ++i ) { schedulers[i]->done_adding_supervised_warps(); } @@ -474,7 +474,7 @@ void shader_core_ctx::init_warps( unsigned cta_id, unsigned start_thread, unsign } m_simt_stack[i]->launch(start_pc,active_threads); - if(m_gpu->resume_option==1 && kernel_id==m_gpu->resume_kernel && ctaid>=m_gpu->resume_CTA && ctaidcheckpoint_CTA_t ) + if(m_gpu->resume_option == 1 && kernel_id == m_gpu->resume_kernel && ctaid >= m_gpu->resume_CTA && ctaid < m_gpu->checkpoint_CTA_t ) { char fname[2048]; snprintf(fname,2048,"checkpoint_files/warp_%d_%d_simt.txt",i%warp_per_cta,ctaid ); @@ -868,7 +868,7 @@ void shader_core_ctx::func_exec_inst( warp_inst_t &inst ) void shader_core_ctx::issue_warp( register_set& pipe_reg_set, const warp_inst_t* next_inst, const active_mask_t &active_mask, unsigned warp_id, unsigned sch_id ) { - warp_inst_t** pipe_reg = pipe_reg = pipe_reg_set.get_free(m_config->sub_core_model, sch_id); + warp_inst_t** pipe_reg = pipe_reg_set.get_free(m_config->sub_core_model, sch_id); assert(pipe_reg); m_warp[warp_id].ibuffer_free(); @@ -2134,7 +2134,7 @@ ldst_unit::ldst_unit( mem_fetch_interface *icnt, if(m_config->m_L1D_config.l1_latency > 0) { - for(int i=0; im_L1D_config.l1_latency; i++ ) + for(unsigned i = 0; i < m_config->m_L1D_config.l1_latency; i++ ) l1_latency_queue.push_back((mem_fetch*)NULL); } } @@ -2446,7 +2446,7 @@ void shader_core_ctx::register_cta_thread_exit( unsigned cta_num, kernel_info_t m_barriers.deallocate_barrier(cta_num); shader_CTA_count_unlog(m_sid, 1); - SHADER_DPRINTF(LIVENESS, "GPGPU-Sim uArch: Finished CTA #%d (%lld,%lld), %u CTAs running\n", + SHADER_DPRINTF(LIVENESS, "GPGPU-Sim uArch: Finished CTA #%u (%lld,%lld), %u CTAs running\n", cta_num, m_gpu->gpu_sim_cycle, m_gpu->gpu_tot_sim_cycle, m_n_active_cta); if( m_n_active_cta == 0 ) { diff --git a/src/gpgpu-sim/shader.h b/src/gpgpu-sim/shader.h index b0d7f7f..dbe2285 100644 --- a/src/gpgpu-sim/shader.h +++ b/src/gpgpu-sim/shader.h @@ -1392,9 +1392,9 @@ class shader_core_config : public core_config If we won't remove it, old regression will be broken. So to support the legacy config files it's best to handle in this way. */ - int num_config_to_read=N_PIPELINE_STAGES-2*(!gpgpu_tensor_core_avail); + int num_config_to_read= N_PIPELINE_STAGES - 2 * (!gpgpu_tensor_core_avail); - for (unsigned i = 0; i Date: Sun, 8 Sep 2019 00:56:20 -0400 Subject: Refactor GPGPUSim_Context and GPGPUSim_Init Signed-off-by: Mengchi Zhang --- libcuda/cuda_runtime_api.cc | 1531 ++++++++++++++++++++++++------------------- 1 file changed, 866 insertions(+), 665 deletions(-) (limited to 'libcuda/cuda_runtime_api.cc') diff --git a/libcuda/cuda_runtime_api.cc b/libcuda/cuda_runtime_api.cc index 716e297..175cbc5 100644 --- a/libcuda/cuda_runtime_api.cc +++ b/libcuda/cuda_runtime_api.cc @@ -199,8 +199,7 @@ void register_ptx_function( const char *name, function_info *impl ) struct _cuda_device_id *gpgpu_context::GPGPUSim_Init() { - //static _cuda_device_id *the_device = NULL; - _cuda_device_id *the_device = GPGPUsim_ctx_ptr()->the_cude_device; + _cuda_device_id *the_device = the_gpgpusim->the_cude_device; if( !the_device ) { gpgpu_sim *the_gpu = gpgpu_ptx_sim_init_perf(); @@ -252,15 +251,14 @@ struct _cuda_device_id *gpgpu_context::GPGPUSim_Init() return the_device; } -static CUctx_st* GPGPUSim_Context() +CUctx_st* GPGPUSim_Context(gpgpu_context * ctx) { //static CUctx_st *the_context = NULL; - gpgpu_context *cur_ctx = GPGPU_Context(); - CUctx_st *the_context = GPGPUsim_ctx_ptr()->the_context; + CUctx_st *the_context = ctx->the_gpgpusim->the_context; if( the_context == NULL ) { - _cuda_device_id *the_gpu = cur_ctx->GPGPUSim_Init(); - GPGPUsim_ctx_ptr()->the_context = new CUctx_st(the_gpu); - the_context = GPGPUsim_ctx_ptr()->the_context; + _cuda_device_id *the_gpu = ctx->GPGPUSim_Init(); + ctx->the_gpgpusim->the_context = new CUctx_st(the_gpu); + the_context = ctx->the_gpgpusim->the_context; } return the_context; } @@ -276,9 +274,9 @@ gpgpu_context* GPGPU_Context() void ptxinfo_data::ptxinfo_addinfo() { + CUctx_st *context = GPGPUSim_Context(gpgpu_ctx); if(!get_ptxinfo_kname()){ /* This info is not per kernel (since CUDA 5.0 some info (e.g. gmem, and cmem) is added at the beginning for the whole binary ) */ - CUctx_st *context = GPGPUSim_Context(); print_ptxinfo(); context->add_ptxinfo(get_ptxinfo()); clear_ptxinfo(); @@ -289,7 +287,6 @@ gpgpu_context* GPGPU_Context() clear_ptxinfo(); return; } - CUctx_st *context = GPGPUSim_Context(); print_ptxinfo(); context->add_ptxinfo( get_ptxinfo_kname(), get_ptxinfo() ); clear_ptxinfo(); @@ -585,7 +582,7 @@ void** cudaRegisterFatBinaryInternal( void *fatCubin, gpgpu_context* gpgpu_ctx = printf("GPGPU-Sim PTX: ERROR ** this version of GPGPU-Sim requires CUDA 2.1 or higher\n"); exit(1); #endif - CUctx_st *context = GPGPUSim_Context(); + CUctx_st *context = GPGPUSim_Context(ctx); static unsigned next_fat_bin_handle = 1; if(context->get_device()->get_gpgpu()->get_config().use_cuobjdump()) { // The following workaround has only been verified on 64-bit systems. @@ -732,7 +729,7 @@ void cudaRegisterFunctionInternal( if(g_debug_execution >= 3){ announce_call(__my_func__); } - CUctx_st *context = GPGPUSim_Context(); + CUctx_st *context = GPGPUSim_Context(ctx); unsigned fat_cubin_handle = (unsigned)(unsigned long long)fatCubinHandle; printf("GPGPU-Sim PTX: __cudaRegisterFunction %s : hostFun 0x%p, fat_cubin_handle = %u\n", deviceFun, hostFun, fat_cubin_handle); @@ -763,7 +760,7 @@ void cudaRegisterVarInternal( } printf("GPGPU-Sim PTX: __cudaRegisterVar: hostVar = %p; deviceAddress = %s; deviceName = %s\n", hostVar, deviceAddress, deviceName); printf("GPGPU-Sim PTX: __cudaRegisterVar: Registering const memory space of %d bytes\n", size); - if(GPGPUSim_Context()->get_device()->get_gpgpu()->get_config().use_cuobjdump()) + if(GPGPUSim_Context(ctx)->get_device()->get_gpgpu()->get_config().use_cuobjdump()) ctx->cuobjdumpParseBinary((unsigned)(unsigned long long)fatCubinHandle); fflush(stdout); if ( constant && !global && !ext ) { @@ -872,7 +869,7 @@ cudaError_t cudaLaunchInternal( const char *hostFun, gpgpu_context* gpgpu_ctx = if(g_debug_execution >= 3){ announce_call(__my_func__); } - CUctx_st* context = GPGPUSim_Context(); + CUctx_st* context = GPGPUSim_Context(ctx); char *mode = getenv("PTX_SIM_MODE_FUNC"); if( mode ) sscanf(mode,"%u", &(ctx->func_sim->g_ptx_sim_mode)); @@ -960,7 +957,7 @@ cudaError_t cudaMallocInternal(void **devPtr, size_t size, gpgpu_context* gpgpu_ if(g_debug_execution >= 3){ announce_call(__my_func__); } - CUctx_st* context = GPGPUSim_Context(); + CUctx_st* context = GPGPUSim_Context(ctx); *devPtr = context->get_device()->get_gpgpu()->gpu_malloc(size); if(g_debug_execution >= 3){ printf("GPGPU-Sim PTX: cudaMallocing %zu bytes starting at 0x%llx..\n",size, (unsigned long long) *devPtr); @@ -994,6 +991,29 @@ cudaError_t cudaMallocHostInternal(void **ptr, size_t size, gpgpu_context* gpgpu } } +__host__ cudaError_t CUDARTAPI cudaMallocPitchInternal(void **devPtr, size_t *pitch, size_t width, size_t height, gpgpu_context* gpgpu_ctx = NULL) +{ + gpgpu_context *ctx; + if (gpgpu_ctx){ + ctx = gpgpu_ctx; + } else { + ctx = GPGPU_Context(); + } + if(g_debug_execution >= 3){ + announce_call(__my_func__); + } + unsigned malloc_width_inbytes = width; + printf("GPGPU-Sim PTX: cudaMallocPitch (width = %d)\n", malloc_width_inbytes); + CUctx_st* context = GPGPUSim_Context(ctx); + *devPtr = context->get_device()->get_gpgpu()->gpu_malloc(malloc_width_inbytes*height); + pitch[0] = malloc_width_inbytes; + if ( *devPtr ) { + return g_last_cudaError = cudaSuccess; + } else { + return g_last_cudaError = cudaErrorMemoryAllocation; + } +} + cudaError_t cudaHostGetDevicePointerInternal(void **pDevice, void *pHost, unsigned int flags, gpgpu_context* gpgpu_ctx = NULL) { gpgpu_context *ctx; @@ -1011,7 +1031,7 @@ cudaError_t cudaHostGetDevicePointerInternal(void **pDevice, void *pHost, unsign //only cpu memory allocation happens in cudaHostAlloc. Linking with device pointer to pinned memory happens here. //TODO: once kernel is executed, the contents in global pointer of GPU must be copied back to CPU host pointer! flags=0; - CUctx_st* context = GPGPUSim_Context(); + CUctx_st* context = GPGPUSim_Context(ctx); gpgpu_t *gpu = context->get_device()->get_gpgpu(); std::map::const_iterator i = ctx->api->pinned_memory_size.find(pHost); assert(i != ctx->api->pinned_memory_size.end()); @@ -1031,7 +1051,182 @@ cudaError_t cudaHostGetDevicePointerInternal(void **pDevice, void *pHost, unsign } } -cudaError_t cudaGLMapBufferObjectInternal(void** devPtr, GLuint bufferObj, gpgpu_context* gpgpu_ctx = NULL) +__host__ cudaError_t CUDARTAPI cudaMallocArrayInternal(struct cudaArray **array, const struct cudaChannelFormatDesc *desc, size_t width, size_t height __dv(1), gpgpu_context* gpgpu_ctx = NULL) +{ + gpgpu_context *ctx; + if (gpgpu_ctx){ + ctx = gpgpu_ctx; + } else { + ctx = GPGPU_Context(); + } + if(g_debug_execution >= 3){ + announce_call(__my_func__); + } + unsigned size = width * height * ((desc->x + desc->y + desc->z + desc->w)/8); + CUctx_st* context = GPGPUSim_Context(ctx); + (*array) = (struct cudaArray*) malloc(sizeof(struct cudaArray)); + (*array)->desc = *desc; + (*array)->width = width; + (*array)->height = height; + (*array)->size = size; + (*array)->dimensions = 2; + ((*array)->devPtr32)= (int) (long long)context->get_device()->get_gpgpu()->gpu_mallocarray(size); + printf("GPGPU-Sim PTX: cudaMallocArray: devPtr32 = %d\n", ((*array)->devPtr32)); + ((*array)->devPtr) = (void*) (long long) ((*array)->devPtr32); + if ( ((*array)->devPtr) ) { + return g_last_cudaError = cudaSuccess; + } else { + return g_last_cudaError = cudaErrorMemoryAllocation; + } +} + +__host__ cudaError_t CUDARTAPI cudaMemcpyToArrayInternal(struct cudaArray *dst, size_t wOffset, size_t hOffset, const void *src, size_t count, enum cudaMemcpyKind kind, gpgpu_context* gpgpu_ctx = NULL) +{ + gpgpu_context *ctx; + if (gpgpu_ctx){ + ctx = gpgpu_ctx; + } else { + ctx = GPGPU_Context(); + } + if(g_debug_execution >= 3){ + announce_call(__my_func__); + } + CUctx_st *context = GPGPUSim_Context(ctx); + gpgpu_t *gpu = context->get_device()->get_gpgpu(); + size_t size = count; + printf("GPGPU-Sim PTX: cudaMemcpyToArray\n"); + if( kind == cudaMemcpyHostToDevice ) + gpu->memcpy_to_gpu( (size_t)(dst->devPtr), src, size); + else if( kind == cudaMemcpyDeviceToHost ) + gpu->memcpy_from_gpu( dst->devPtr, (size_t)src, size); + else if( kind == cudaMemcpyDeviceToDevice ) + gpu->memcpy_gpu_to_gpu( (size_t)(dst->devPtr), (size_t)src, size); + else { + printf("GPGPU-Sim PTX: cudaMemcpyToArray - ERROR : unsupported cudaMemcpyKind\n"); + abort(); + } + dst->devPtr32 = (unsigned) (size_t)(dst->devPtr); + return g_last_cudaError = cudaSuccess; +} + +__host__ cudaError_t CUDARTAPI cudaMemcpy2DInternal(void *dst, size_t dpitch, const void *src, size_t spitch, size_t width, size_t height, enum cudaMemcpyKind kind, gpgpu_context* gpgpu_ctx = NULL) +{ + gpgpu_context *ctx; + if (gpgpu_ctx){ + ctx = gpgpu_ctx; + } else { + ctx = GPGPU_Context(); + } + if(g_debug_execution >= 3){ + announce_call(__my_func__); + } + CUctx_st *context = GPGPUSim_Context(ctx); + gpgpu_t *gpu = context->get_device()->get_gpgpu(); + size_t size = spitch*height; + gpgpusim_ptx_assert( (dpitch==spitch), "different src and dst pitch not supported yet" ); + if( kind == cudaMemcpyHostToDevice ) + gpu->memcpy_to_gpu( (size_t)dst, src, size ); + else if( kind == cudaMemcpyDeviceToHost ) + gpu->memcpy_from_gpu( dst, (size_t)src, size ); + else if( kind == cudaMemcpyDeviceToDevice ) + gpu->memcpy_gpu_to_gpu( (size_t)dst, (size_t)src, size); + else { + printf("GPGPU-Sim PTX: cudaMemcpy2D - ERROR : unsupported cudaMemcpyKind\n"); + abort(); + } + return g_last_cudaError = cudaSuccess; +} + +__host__ cudaError_t CUDARTAPI cudaMemcpy2DToArrayInternal(struct cudaArray *dst, size_t wOffset, size_t hOffset, const void *src, size_t spitch, size_t width, size_t height, enum cudaMemcpyKind kind, gpgpu_context* gpgpu_ctx = NULL) +{ + gpgpu_context *ctx; + if (gpgpu_ctx){ + ctx = gpgpu_ctx; + } else { + ctx = GPGPU_Context(); + } + if(g_debug_execution >= 3){ + announce_call(__my_func__); + } + CUctx_st *context = GPGPUSim_Context(ctx); + gpgpu_t *gpu = context->get_device()->get_gpgpu(); + size_t size = spitch*height; + size_t channel_size = dst->desc.w+dst->desc.x+dst->desc.y+dst->desc.z; + gpgpusim_ptx_assert( ((channel_size%8) == 0), "none byte multiple destination channel size not supported (sz=%u)", channel_size ); + unsigned elem_size = channel_size/8; + gpgpusim_ptx_assert( (dst->dimensions==2), "copy to none 2D array not supported" ); + gpgpusim_ptx_assert( (wOffset==0), "non-zero wOffset not yet supported" ); + gpgpusim_ptx_assert( (hOffset==0), "non-zero hOffset not yet supported" ); + gpgpusim_ptx_assert( (dst->height == (int)height), "partial copy not supported" ); + gpgpusim_ptx_assert( (elem_size*dst->width == width), "partial copy not supported" ); + gpgpusim_ptx_assert( (spitch == width), "spitch != width not supported" ); + if( kind == cudaMemcpyHostToDevice ) + gpu->memcpy_to_gpu( (size_t)(dst->devPtr), src, size); + else if( kind == cudaMemcpyDeviceToHost ) + gpu->memcpy_from_gpu( dst->devPtr, (size_t)src, size); + else if( kind == cudaMemcpyDeviceToDevice ) + gpu->memcpy_gpu_to_gpu( (size_t)dst->devPtr, (size_t)src, size); + else { + printf("GPGPU-Sim PTX: cudaMemcpy2D - ERROR : unsupported cudaMemcpyKind\n"); + abort(); + } + dst->devPtr32 = (unsigned) (size_t)(dst->devPtr); + return g_last_cudaError = cudaSuccess; +} + +#if (CUDART_VERSION >= 8000) +cudaError_t CUDARTAPI cudaOccupancyMaxActiveBlocksPerMultiprocessorWithFlagsInternal(int* numBlocks, const char *hostFunc, int blockSize, size_t dynamicSMemSize, unsigned int flags, gpgpu_context* gpgpu_ctx = NULL) +{ + gpgpu_context *ctx; + if (gpgpu_ctx){ + ctx = gpgpu_ctx; + } else { + ctx = GPGPU_Context(); + } + printf("GPGPU-Sim PTX: cudaOccupancyMaxActiveBlocksPerMultiprocessorWithFlags %p\n", hostFunc); + CUctx_st *context = GPGPUSim_Context(ctx); + function_info *entry = context->get_kernel(hostFunc); + printf("Calculate Maxium Active Block with function ptr=%p, blockSize=%d, SMemSize=%d\n", hostFunc, blockSize, dynamicSMemSize); + if (flags == cudaOccupancyDefault) { + //create kernel_info based on entry + dim3 gridDim(context->get_device()->get_gpgpu()->max_cta_per_core() + * context->get_device()->get_gpgpu()->get_config().num_shader()); + dim3 blockDim(blockSize); + kernel_info_t result(gridDim, blockDim, entry); + //if(entry == NULL){ + // *numBlocks = 1; + // return g_last_cudaError = cudaErrorUnknown; + //} + *numBlocks = context->get_device()->get_gpgpu()->get_max_cta(result); + printf("Maximum size is %d with gridDim %d and blockDim %d\n", *numBlocks, gridDim.x, blockDim.x); + return g_last_cudaError = cudaSuccess; + } else { + cuda_not_implemented(__my_func__,__LINE__); + return g_last_cudaError = cudaErrorUnknown; + } +} + +#endif + +__host__ cudaError_t CUDARTAPI cudaMemsetInternal(void *mem, int c, size_t count, gpgpu_context* gpgpu_ctx = NULL) +{ + gpgpu_context *ctx; + if (gpgpu_ctx){ + ctx = gpgpu_ctx; + } else { + ctx = GPGPU_Context(); + } + if(g_debug_execution >= 3){ + announce_call(__my_func__); + } + CUctx_st *context = GPGPUSim_Context(ctx); + gpgpu_t *gpu = context->get_device()->get_gpgpu(); + gpu->gpu_memset((size_t)mem, c, count); + return g_last_cudaError = cudaSuccess; +} + +//memset operation is done but i think its not async? +__host__ cudaError_t CUDARTAPI cudaMemsetAsyncInternal(void *mem, int c, size_t count, cudaStream_t stream=0, gpgpu_context* gpgpu_ctx = NULL) { gpgpu_context *ctx; if (gpgpu_ctx){ @@ -1042,12 +1237,27 @@ cudaError_t cudaGLMapBufferObjectInternal(void** devPtr, GLuint bufferObj, gpgpu if(g_debug_execution >= 3){ announce_call(__my_func__); } + printf("GPGPU-Sim PTX: WARNING: Asynchronous memset not supported (%s)\n", __my_func__); + CUctx_st *context = GPGPUSim_Context(ctx); + gpgpu_t *gpu = context->get_device()->get_gpgpu(); + gpu->gpu_memset((size_t)mem, c, count); + return g_last_cudaError = cudaSuccess; +} + +cudaError_t cudaGLMapBufferObjectInternal(void** devPtr, GLuint bufferObj, gpgpu_context* gpgpu_ctx = NULL) +{ if(g_debug_execution >= 3){ announce_call(__my_func__); } #ifdef OPENGL_SUPPORT + gpgpu_context *ctx; + if (gpgpu_ctx){ + ctx = gpgpu_ctx; + } else { + ctx = GPGPU_Context(); + } GLint buffer_size=0; - CUctx_st* context = GPGPUSim_Context(); + CUctx_st* context = GPGPUSim_Context(ctx); glbmap_entry_t *p = ctx->api->g_glbmap; while ( p && p->m_bufferObj != bufferObj ) @@ -1118,7 +1328,7 @@ cuLinkAddFileInternal(CUlinkState state, CUjitInputType type, const char *path, //blocking assert(type==CU_JIT_INPUT_PTX); - CUctx_st *context = GPGPUSim_Context(); + CUctx_st *context = GPGPUSim_Context(ctx); char *file = getenv("PTX_JIT_PATH"); if(file==NULL){ printf("GPGPU-Sim PTX: ERROR: PTX_JIT_PATH has not been set\n"); @@ -1190,7 +1400,7 @@ cudaError_t CUDARTAPI cudaFuncGetAttributesInternal(struct cudaFuncAttributes *a if(g_debug_execution >= 3){ announce_call(__my_func__); } - CUctx_st *context = GPGPUSim_Context(); + CUctx_st *context = GPGPUSim_Context(ctx); function_info *entry = context->get_kernel(hostFun); if( entry ) { const struct gpgpu_ptx_sim_info *kinfo = entry->get_kernel_info(); @@ -1210,72 +1420,479 @@ cudaError_t CUDARTAPI cudaFuncGetAttributesInternal(struct cudaFuncAttributes *a return g_last_cudaError = cudaSuccess; } - -/******************************************************************************* - * * - * * - * * - *******************************************************************************/ - -extern "C" { - -/******************************************************************************* - * * - * * - * * - *******************************************************************************/ -cudaError_t cudaPeekAtLastError(void) -{ - return g_last_cudaError; -} - -__host__ cudaError_t CUDARTAPI cudaMalloc(void **devPtr, size_t size) +#if (CUDART_VERSION > 5000) +__host__ cudaError_t CUDARTAPI cudaDeviceGetAttributeInternal(int *value, enum cudaDeviceAttr attr, int device, gpgpu_context* gpgpu_ctx = NULL) { - return cudaMallocInternal(devPtr, size); -} + gpgpu_context *ctx; + if (gpgpu_ctx){ + ctx = gpgpu_ctx; + } else { + ctx = GPGPU_Context(); + } + if(g_debug_execution >= 3){ + announce_call(__my_func__); + } -__host__ cudaError_t CUDARTAPI cudaMallocHost(void **ptr, size_t size) -{ - return cudaMallocHostInternal(ptr, size); -} -__host__ cudaError_t CUDARTAPI cudaMallocPitch(void **devPtr, size_t *pitch, size_t width, size_t height) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - unsigned malloc_width_inbytes = width; - printf("GPGPU-Sim PTX: cudaMallocPitch (width = %d)\n", malloc_width_inbytes); - CUctx_st* ctx = GPGPUSim_Context(); - *devPtr = ctx->get_device()->get_gpgpu()->gpu_malloc(malloc_width_inbytes*height); - pitch[0] = malloc_width_inbytes; - if ( *devPtr ) { - return g_last_cudaError = cudaSuccess; - } else { - return g_last_cudaError = cudaErrorMemoryAllocation; - } -} + const struct cudaDeviceProp *prop; + _cuda_device_id *dev = ctx->GPGPUSim_Init(); -__host__ cudaError_t CUDARTAPI cudaMallocArray(struct cudaArray **array, const struct cudaChannelFormatDesc *desc, size_t width, size_t height __dv(1)) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - unsigned size = width * height * ((desc->x + desc->y + desc->z + desc->w)/8); - CUctx_st* context = GPGPUSim_Context(); - (*array) = (struct cudaArray*) malloc(sizeof(struct cudaArray)); - (*array)->desc = *desc; - (*array)->width = width; - (*array)->height = height; - (*array)->size = size; - (*array)->dimensions = 2; - ((*array)->devPtr32)= (int) (long long)context->get_device()->get_gpgpu()->gpu_mallocarray(size); - printf("GPGPU-Sim PTX: cudaMallocArray: devPtr32 = %d\n", ((*array)->devPtr32)); - ((*array)->devPtr) = (void*) (long long) ((*array)->devPtr32); - if ( ((*array)->devPtr) ) { - return g_last_cudaError = cudaSuccess; + if (device <= dev->num_devices() ) { + prop = dev->get_prop(); + switch (attr) { + case 1: + *value= prop->maxThreadsPerBlock; + break; + case 2: + *value= prop->maxThreadsDim[0]; + break; + case 3: + *value= prop->maxThreadsDim[1]; + break; + case 4: + *value= prop->maxThreadsDim[2]; + break; + case 5: + *value= prop->maxGridSize[0]; + break; + case 6: + *value= prop->maxGridSize[1]; + break; + case 7: + *value= prop->maxGridSize[2]; + break; + case 8: + *value= prop->sharedMemPerBlock; + break; + case 9: + *value= prop->totalConstMem; + break; + case 10: + *value= prop->warpSize; + break; + case 11: + *value= 16;//dummy value + break; + case 12: + *value= prop->regsPerBlock; + break; + case 13: + *value= 1480000;//for 1080ti + break; + case 14: + *value= prop->textureAlignment ; + break; + case 15: + *value = 0; + break; + case 16: + *value= prop->multiProcessorCount ; + break; + case 17: + case 18: + case 19: + *value = 0; + break; + case 21: + case 22: + case 23: + case 24: + case 25: + case 26: + case 27: + case 28: + case 42: + case 45: + case 46: + case 47: + case 48: + case 49: + case 52: + case 53: + case 55: + case 56: + case 57: + case 58: + case 59: + case 60: + case 61: + case 62: + case 63: + case 64: + case 66: + case 67: + case 69: + case 70: + case 71: + case 73: + case 74: + case 77: + *value = 1000;//dummy value + break; + case 29: + case 43: + case 54: + case 65: + case 68: + case 72: + *value = 10;//dummy value + break; + case 30: + case 51: + *value = 128;//dummy value + break; + case 31: + *value = 1; + break; + case 32: + *value = 0; + break; + case 33: + case 50: + *value = 0;//dummy value + break; + case 34: + *value= 0; + break; + case 35: + *value = 0; + break; + case 36: + *value = 1250000;//CK value for 1080ti + break; + case 37: + *value = 352;//value for 1080ti + break; + case 38: + *value = 3000000;//value for 1080ti + break; + case 39: + *value= dev->get_gpgpu()->threads_per_core(); + break; + case 40: + *value= 0; + break; + case 41: + *value= 0; + break; + case 75://cudaDevAttrComputeCapabilityMajor + *value= prop->major ; + break; + case 76://cudaDevAttrComputeCapabilityMinor + *value= prop->minor ; + break; + case 78: + *value= 0 ; //TODO: as of now, we dont support stream priorities. + break; + case 79: + *value= 0; + break; + case 80: + *value= 0; + break; + #if (CUDART_VERSION > 5050) + case 81: + *value= prop->sharedMemPerMultiprocessor; + break; + case 82: + *value= prop->regsPerMultiprocessor; + break; + #endif + case 83: + case 84: + case 85: + case 86: + *value= 0; + break; + case 87: + *value= 4;//dummy value + break; + case 88: + case 89: + *value= 0; + break; + default: + printf("ERROR: Attribute number %d unimplemented \n",attr); + abort(); + } + return g_last_cudaError = cudaSuccess; + } else { + return g_last_cudaError = cudaErrorInvalidDevice; + } +} +#endif + +__host__ cudaError_t CUDARTAPI cudaBindTextureInternal(size_t *offset, + const struct textureReference *texref, + const void *devPtr, + const struct cudaChannelFormatDesc *desc, + size_t size __dv(UINT_MAX), + gpgpu_context* gpgpu_ctx = NULL) +{ + gpgpu_context *ctx; + if (gpgpu_ctx){ + ctx = gpgpu_ctx; + } else { + ctx = GPGPU_Context(); + } + if(g_debug_execution >= 3){ + announce_call(__my_func__); + } + CUctx_st *context = GPGPUSim_Context(ctx); + gpgpu_t *gpu = context->get_device()->get_gpgpu(); + printf("GPGPU-Sim PTX: in cudaBindTexture: sizeof(struct textureReference) = %zu\n", sizeof(struct textureReference)); + struct cudaArray *array; + array = (struct cudaArray*) malloc(sizeof(struct cudaArray)); + array->desc = *desc; + array->size = size; + array->width = size; + array->height = 1; + array->dimensions = 1; + array->devPtr = (void*)devPtr; + array->devPtr32 = (int)(long long)devPtr; + offset = 0; + printf("GPGPU-Sim PTX: size = %zu\n", size); + printf("GPGPU-Sim PTX: texref = %p, array = %p\n", texref, array); + printf("GPGPU-Sim PTX: devPtr32 = %x\n", array->devPtr32); + printf("GPGPU-Sim PTX: Name corresponding to textureReference: %s\n", gpu->gpgpu_ptx_sim_findNamefromTexture(texref)); + printf("GPGPU-Sim PTX: ChannelFormatDesc: x=%d, y=%d, z=%d, w=%d\n", desc->x, desc->y, desc->z, desc->w); + printf("GPGPU-Sim PTX: Texture Normalized? = %d\n", texref->normalized); + gpu->gpgpu_ptx_sim_bindTextureToArray(texref, array); + devPtr = (void*)(long long)array->devPtr32; + printf("GPGPU-Sim PTX: devPtr = %p\n", devPtr); + return g_last_cudaError = cudaSuccess; +} + +__host__ cudaError_t CUDARTAPI cudaBindTextureToArrayInternal(const struct textureReference *texref, const struct cudaArray *array, const struct cudaChannelFormatDesc *desc, gpgpu_context* gpgpu_ctx = NULL) +{ + gpgpu_context *ctx; + if (gpgpu_ctx){ + ctx = gpgpu_ctx; + } else { + ctx = GPGPU_Context(); + } + if(g_debug_execution >= 3){ + announce_call(__my_func__); + } + CUctx_st *context = GPGPUSim_Context(ctx); + gpgpu_t *gpu = context->get_device()->get_gpgpu(); + printf("GPGPU-Sim PTX: in cudaBindTextureToArray: %p %p\n", texref, array); + printf("GPGPU-Sim PTX: devPtr32 = %x\n", array->devPtr32); + printf("GPGPU-Sim PTX: Name corresponding to textureReference: %s\n", gpu->gpgpu_ptx_sim_findNamefromTexture(texref)); + printf("GPGPU-Sim PTX: Texture Normalized? = %d\n", texref->normalized); + gpu->gpgpu_ptx_sim_bindTextureToArray(texref, array); + return g_last_cudaError = cudaSuccess; +} + +__host__ cudaError_t CUDARTAPI cudaUnbindTextureInternal(const struct textureReference *texref, gpgpu_context* gpgpu_ctx = NULL) +{ + gpgpu_context *ctx; + if (gpgpu_ctx){ + ctx = gpgpu_ctx; + } else { + ctx = GPGPU_Context(); + } + if(g_debug_execution >= 3){ + announce_call(__my_func__); + } + CUctx_st *context = GPGPUSim_Context(ctx); + gpgpu_t *gpu = context->get_device()->get_gpgpu(); + printf("GPGPU-Sim PTX: in cudaUnbindTexture: sizeof(struct textureReference) = %zu\n", sizeof(struct textureReference)); + printf("GPGPU-Sim PTX: Name corresponding to textureReference: %s\n", gpu->gpgpu_ptx_sim_findNamefromTexture(texref)); + + gpu->gpgpu_ptx_sim_unbindTexture(texref); + return g_last_cudaError = cudaSuccess; +} + +__host__ cudaError_t CUDARTAPI cudaLaunchKernelInternal( const char* hostFun, dim3 gridDim, dim3 blockDim, const void** args, size_t sharedMem, cudaStream_t stream, gpgpu_context* gpgpu_ctx = NULL ) +{ + gpgpu_context *ctx; + if (gpgpu_ctx){ + ctx = gpgpu_ctx; + } else { + ctx = GPGPU_Context(); + } + + if(g_debug_execution >= 3){ + announce_call(__my_func__); + } + CUctx_st *context = GPGPUSim_Context(ctx); + function_info *entry = context->get_kernel(hostFun); +#if CUDART_VERSION < 10000 + cudaConfigureCallInternal(gridDim, blockDim, sharedMem, stream, ctx); +#endif + for(unsigned i = 0; i < entry->num_args(); i++){ + std::pair p = entry->get_param_config(i); + cudaSetupArgumentInternal(args[i], p.first, p.second); + } + + cudaLaunchInternal(hostFun); + return g_last_cudaError = cudaSuccess; +} + +void __cudaRegisterTextureInternal( + void **fatCubinHandle, + const struct textureReference *hostVar, + const void **deviceAddress, + const char *deviceName, + int dim, + int norm, + int ext, + gpgpu_context* gpgpu_ctx = NULL +) //passes in a newly created textureReference +{ + gpgpu_context *ctx; + if (gpgpu_ctx){ + ctx = gpgpu_ctx; + } else { + ctx = GPGPU_Context(); + } + if(g_debug_execution >= 3){ + announce_call(__my_func__); + } + std::string devStr (deviceName); + #if (CUDART_VERSION > 4020) + if (devStr.size() > 2 && devStr.data()[0] == ':' && devStr.data()[1] == ':') + devStr = devStr.replace(0, 2, ""); + #endif + CUctx_st *context = GPGPUSim_Context(ctx); + gpgpu_t *gpu = context->get_device()->get_gpgpu(); + printf("GPGPU-Sim PTX: in __cudaRegisterTexture:\n"); + gpu->gpgpu_ptx_sim_bindNameToTexture(devStr.data(), hostVar, dim, norm, ext); + printf("GPGPU-Sim PTX: int dim = %d\n", dim); + printf("GPGPU-Sim PTX: int norm = %d\n", norm); + printf("GPGPU-Sim PTX: int ext = %d\n", ext); + printf("GPGPU-Sim PTX: Execution warning: Not finished implementing \"%s\"\n", __my_func__ ); +} + +cudaError_t cudaGLUnmapBufferObjectInternal(GLuint bufferObj, gpgpu_context* gpgpu_ctx = NULL) +{ + if(g_debug_execution >= 3){ + announce_call(__my_func__); + } +#ifdef OPENGL_SUPPORT + gpgpu_context *ctx; + if (gpgpu_ctx){ + ctx = gpgpu_ctx; } else { - return g_last_cudaError = cudaErrorMemoryAllocation; + ctx = GPGPU_Context(); } + CUctx_st* ctx = GPGPUSim_Context(ctx); + glbmap_entry_t *p = ctx->api->g_glbmap; + while ( p && p->m_bufferObj != bufferObj ) + p = p->m_next; + if ( p == NULL ) + return g_last_cudaError = cudaErrorUnknown; + + char *data = (char *) calloc(p->m_size,1); + memcpy_from_gpu( data,(size_t)p->m_devPtr,p->m_size ); + glBufferSubData(GL_ARRAY_BUFFER,0,p->m_size,data); + free(data); + + return g_last_cudaError = cudaSuccess; +#else + fflush(stdout); + fflush(stderr); + printf("GPGPU-Sim PTX: support for OpenGL integration disabled -- exiting\n"); + fflush(stdout); + exit(50); +#endif +} + +#if CUDART_VERSION >= 3000 + +__host__ cudaError_t CUDARTAPI cudaFuncSetCacheConfigInternal(const char *func, enum cudaFuncCache cacheConfig, gpgpu_context* gpgpu_ctx = NULL ) +{ + gpgpu_context *ctx; + if (gpgpu_ctx){ + ctx = gpgpu_ctx; + } else { + ctx = GPGPU_Context(); + } + if(g_debug_execution >= 3){ + announce_call(__my_func__); + } + CUctx_st *context = GPGPUSim_Context(ctx); + context->get_device()->get_gpgpu()->set_cache_config(context->get_kernel(func)->get_name(), (FuncCache)cacheConfig); + return g_last_cudaError = cudaSuccess; +} + +#endif + +#if CUDART_VERSION >= 4000 +CUresult CUDAAPI cuLaunchKernelInternal(CUfunction f, + unsigned int gridDimX, + unsigned int gridDimY, + unsigned int gridDimZ, + unsigned int blockDimX, + unsigned int blockDimY, + unsigned int blockDimZ, + unsigned int sharedMemBytes, + CUstream hStream, + void **kernelParams, + void **extra, + gpgpu_context* gpgpu_ctx = NULL) +{ + gpgpu_context *ctx; + if (gpgpu_ctx){ + ctx = gpgpu_ctx; + } else { + ctx = GPGPU_Context(); + } + if(g_debug_execution >= 3){ + announce_call(__my_func__); + } + if (extra!=NULL){ + printf("GPGPU-Sim CUDA DRIVER API: ERROR: Currently do not support void** extra.\n"); + abort(); + } + const char *hostFun = (const char*) f; + CUctx_st *context = GPGPUSim_Context(ctx); + function_info *entry = context->get_kernel(hostFun); + cudaConfigureCallInternal(dim3(gridDimX, gridDimY, gridDimZ), dim3(blockDimX, blockDimY, blockDimZ), sharedMemBytes, (cudaStream_t) hStream, ctx); + for(unsigned i = 0; i < entry->num_args(); i++){ + std::pair p = entry->get_param_config(i); + cudaSetupArgumentInternal(kernelParams[i], p.first, p.second, ctx); + } + cudaLaunchInternal(hostFun, ctx); + return CUDA_SUCCESS; +} +#endif /* CUDART_VERSION >= 4000 */ + +/******************************************************************************* + * * + * * + * * + *******************************************************************************/ + +extern "C" { + +/******************************************************************************* + * * + * * + * * + *******************************************************************************/ +cudaError_t cudaPeekAtLastError(void) +{ + return g_last_cudaError; +} + +__host__ cudaError_t CUDARTAPI cudaMalloc(void **devPtr, size_t size) +{ + return cudaMallocInternal(devPtr, size); +} + +__host__ cudaError_t CUDARTAPI cudaMallocHost(void **ptr, size_t size) +{ + return cudaMallocHostInternal(ptr, size); +} +__host__ cudaError_t CUDARTAPI cudaMallocPitch(void **devPtr, size_t *pitch, size_t width, size_t height) +{ + return cudaMallocPitchInternal(devPtr, pitch, width, height); +} + +__host__ cudaError_t CUDARTAPI cudaMallocArray(struct cudaArray **array, const struct cudaChannelFormatDesc *desc, size_t width, size_t height __dv(1)) +{ + return cudaMallocArrayInternal(array, desc, width, height __dv(1)); } __host__ cudaError_t CUDARTAPI cudaFree(void *devPtr) @@ -1344,213 +1961,18 @@ __host__ cudaError_t CUDARTAPI cudaMemcpy(void *dst, const void *src, size_t cou } else { printf("GPGPU-Sim PTX: cudaMemcpy - ERROR : unsupported cudaMemcpyKind\n"); - abort(); - } - return g_last_cudaError = cudaSuccess; -} - -__host__ cudaError_t CUDARTAPI cudaMemcpyToArray(struct cudaArray *dst, size_t wOffset, size_t hOffset, const void *src, size_t count, enum cudaMemcpyKind kind) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - CUctx_st *context = GPGPUSim_Context(); - gpgpu_t *gpu = context->get_device()->get_gpgpu(); - size_t size = count; - printf("GPGPU-Sim PTX: cudaMemcpyToArray\n"); - if( kind == cudaMemcpyHostToDevice ) - gpu->memcpy_to_gpu( (size_t)(dst->devPtr), src, size); - else if( kind == cudaMemcpyDeviceToHost ) - gpu->memcpy_from_gpu( dst->devPtr, (size_t)src, size); - else if( kind == cudaMemcpyDeviceToDevice ) - gpu->memcpy_gpu_to_gpu( (size_t)(dst->devPtr), (size_t)src, size); - else { - printf("GPGPU-Sim PTX: cudaMemcpyToArray - ERROR : unsupported cudaMemcpyKind\n"); - abort(); - } - dst->devPtr32 = (unsigned) (size_t)(dst->devPtr); - return g_last_cudaError = cudaSuccess; -} - - -__host__ cudaError_t CUDARTAPI cudaMemcpyFromArray(void *dst, const struct cudaArray *src, size_t wOffset, size_t hOffset, size_t count, enum cudaMemcpyKind kind) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - cuda_not_implemented(__my_func__,__LINE__); - return g_last_cudaError = cudaErrorUnknown; -} - - -__host__ cudaError_t CUDARTAPI cudaMemcpyArrayToArray(struct cudaArray *dst, size_t wOffsetDst, size_t hOffsetDst, const struct cudaArray *src, size_t wOffsetSrc, size_t hOffsetSrc, size_t count, enum cudaMemcpyKind kind __dv(cudaMemcpyDeviceToDevice)) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - cuda_not_implemented(__my_func__,__LINE__); - return g_last_cudaError = cudaErrorUnknown; -} - - -__host__ cudaError_t CUDARTAPI cudaMemcpy2D(void *dst, size_t dpitch, const void *src, size_t spitch, size_t width, size_t height, enum cudaMemcpyKind kind) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - CUctx_st *context = GPGPUSim_Context(); - gpgpu_t *gpu = context->get_device()->get_gpgpu(); - size_t size = spitch*height; - gpgpusim_ptx_assert( (dpitch==spitch), "different src and dst pitch not supported yet" ); - if( kind == cudaMemcpyHostToDevice ) - gpu->memcpy_to_gpu( (size_t)dst, src, size ); - else if( kind == cudaMemcpyDeviceToHost ) - gpu->memcpy_from_gpu( dst, (size_t)src, size ); - else if( kind == cudaMemcpyDeviceToDevice ) - gpu->memcpy_gpu_to_gpu( (size_t)dst, (size_t)src, size); - else { - printf("GPGPU-Sim PTX: cudaMemcpy2D - ERROR : unsupported cudaMemcpyKind\n"); - abort(); - } - return g_last_cudaError = cudaSuccess; -} - - -__host__ cudaError_t CUDARTAPI cudaMemcpy2DToArray(struct cudaArray *dst, size_t wOffset, size_t hOffset, const void *src, size_t spitch, size_t width, size_t height, enum cudaMemcpyKind kind) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - CUctx_st *context = GPGPUSim_Context(); - gpgpu_t *gpu = context->get_device()->get_gpgpu(); - size_t size = spitch*height; - size_t channel_size = dst->desc.w+dst->desc.x+dst->desc.y+dst->desc.z; - gpgpusim_ptx_assert( ((channel_size%8) == 0), "none byte multiple destination channel size not supported (sz=%u)", channel_size ); - unsigned elem_size = channel_size/8; - gpgpusim_ptx_assert( (dst->dimensions==2), "copy to none 2D array not supported" ); - gpgpusim_ptx_assert( (wOffset==0), "non-zero wOffset not yet supported" ); - gpgpusim_ptx_assert( (hOffset==0), "non-zero hOffset not yet supported" ); - gpgpusim_ptx_assert( (dst->height == (int)height), "partial copy not supported" ); - gpgpusim_ptx_assert( (elem_size*dst->width == width), "partial copy not supported" ); - gpgpusim_ptx_assert( (spitch == width), "spitch != width not supported" ); - if( kind == cudaMemcpyHostToDevice ) - gpu->memcpy_to_gpu( (size_t)(dst->devPtr), src, size); - else if( kind == cudaMemcpyDeviceToHost ) - gpu->memcpy_from_gpu( dst->devPtr, (size_t)src, size); - else if( kind == cudaMemcpyDeviceToDevice ) - gpu->memcpy_gpu_to_gpu( (size_t)dst->devPtr, (size_t)src, size); - else { - printf("GPGPU-Sim PTX: cudaMemcpy2D - ERROR : unsupported cudaMemcpyKind\n"); - abort(); - } - dst->devPtr32 = (unsigned) (size_t)(dst->devPtr); - return g_last_cudaError = cudaSuccess; -} - - -__host__ cudaError_t CUDARTAPI cudaMemcpy2DFromArray(void *dst, size_t dpitch, const struct cudaArray *src, size_t wOffset, size_t hOffset, size_t width, size_t height, enum cudaMemcpyKind kind) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - cuda_not_implemented(__my_func__,__LINE__); - return g_last_cudaError = cudaErrorUnknown; -} - - -__host__ cudaError_t CUDARTAPI cudaMemcpy2DArrayToArray(struct cudaArray *dst, size_t wOffsetDst, size_t hOffsetDst, const struct cudaArray *src, size_t wOffsetSrc, size_t hOffsetSrc, size_t width, size_t height, enum cudaMemcpyKind kind __dv(cudaMemcpyDeviceToDevice)) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - cuda_not_implemented(__my_func__,__LINE__); - return g_last_cudaError = cudaErrorUnknown; -} - - -__host__ cudaError_t CUDARTAPI cudaMemcpyToSymbol(const char *symbol, const void *src, size_t count, size_t offset __dv(0), enum cudaMemcpyKind kind __dv(cudaMemcpyHostToDevice)) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - //CUctx_st *context = GPGPUSim_Context(); - assert(kind == cudaMemcpyHostToDevice); - printf("GPGPU-Sim PTX: cudaMemcpyToSymbol: symbol = %p\n", symbol); - //stream_operation( const char *symbol, const void *src, size_t count, size_t offset ) - g_stream_manager()->push( stream_operation(src,symbol,count,offset,0) ); - //gpgpu_ptx_sim_memcpy_symbol(symbol,src,count,offset,1,context->get_device()->get_gpgpu()); - return g_last_cudaError = cudaSuccess; -} - - -__host__ cudaError_t CUDARTAPI cudaMemcpyFromSymbol(void *dst, const char *symbol, size_t count, size_t offset __dv(0), enum cudaMemcpyKind kind __dv(cudaMemcpyDeviceToHost)) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - //CUctx_st *context = GPGPUSim_Context(); - assert(kind == cudaMemcpyDeviceToHost); - printf("GPGPU-Sim PTX: cudaMemcpyFromSymbol: symbol = %p\n", symbol); - g_stream_manager()->push( stream_operation(symbol,dst,count,offset,0) ); - //gpgpu_ptx_sim_memcpy_symbol(symbol,dst,count,offset,0,context->get_device()->get_gpgpu()); - return g_last_cudaError = cudaSuccess; -} - -__host__ cudaError_t CUDARTAPI cudaMemGetInfo (size_t *free, size_t *total){ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - //placeholder; should interact with cudaMalloc and cudaFree? - *free = 10000000000; - *total = 10000000000; - - return g_last_cudaError = cudaSuccess; -} - -/******************************************************************************* - * * - * * - * * - *******************************************************************************/ - -__host__ cudaError_t CUDARTAPI cudaMemcpyAsync(void *dst, const void *src, size_t count, enum cudaMemcpyKind kind, cudaStream_t stream) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - struct CUstream_st *s = (struct CUstream_st *)stream; - switch( kind ) { - case cudaMemcpyHostToDevice: g_stream_manager()->push( stream_operation(src,(size_t)dst,count,s) ); break; - case cudaMemcpyDeviceToHost: g_stream_manager()->push( stream_operation((size_t)src,dst,count,s) ); break; - case cudaMemcpyDeviceToDevice: g_stream_manager()->push( stream_operation((size_t)src,(size_t)dst,count,s) ); break; - default: - abort(); - } - return g_last_cudaError = cudaSuccess; -} - - -__host__ cudaError_t CUDARTAPI cudaMemcpyToArrayAsync(struct cudaArray *dst, size_t wOffset, size_t hOffset, const void *src, size_t count, enum cudaMemcpyKind kind, cudaStream_t stream) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - cuda_not_implemented(__my_func__,__LINE__); - return g_last_cudaError = cudaErrorUnknown; -} - - -__host__ cudaError_t CUDARTAPI cudaMemcpyFromArrayAsync(void *dst, const struct cudaArray *src, size_t wOffset, size_t hOffset, size_t count, enum cudaMemcpyKind kind, cudaStream_t stream) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - cuda_not_implemented(__my_func__,__LINE__); - return g_last_cudaError = cudaErrorUnknown; + abort(); + } + return g_last_cudaError = cudaSuccess; +} + +__host__ cudaError_t CUDARTAPI cudaMemcpyToArray(struct cudaArray *dst, size_t wOffset, size_t hOffset, const void *src, size_t count, enum cudaMemcpyKind kind) +{ + return cudaMemcpyToArrayInternal(dst, wOffset, hOffset, src, count, kind); } -__host__ cudaError_t CUDARTAPI cudaMemcpy2DAsync(void *dst, size_t dpitch, const void *src, size_t spitch, size_t width, size_t height, enum cudaMemcpyKind kind, cudaStream_t stream) +__host__ cudaError_t CUDARTAPI cudaMemcpyFromArray(void *dst, const struct cudaArray *src, size_t wOffset, size_t hOffset, size_t count, enum cudaMemcpyKind kind) { if(g_debug_execution >= 3){ announce_call(__my_func__); @@ -1560,7 +1982,7 @@ __host__ cudaError_t CUDARTAPI cudaMemcpy2DAsync(void *dst, size_t dpitch, const } -__host__ cudaError_t CUDARTAPI cudaMemcpy2DToArrayAsync(struct cudaArray *dst, size_t wOffset, size_t hOffset, const void *src, size_t spitch, size_t width, size_t height, enum cudaMemcpyKind kind, cudaStream_t stream) +__host__ cudaError_t CUDARTAPI cudaMemcpyArrayToArray(struct cudaArray *dst, size_t wOffsetDst, size_t hOffsetDst, const struct cudaArray *src, size_t wOffsetSrc, size_t hOffsetSrc, size_t count, enum cudaMemcpyKind kind __dv(cudaMemcpyDeviceToDevice)) { if(g_debug_execution >= 3){ announce_call(__my_func__); @@ -1570,275 +1992,171 @@ __host__ cudaError_t CUDARTAPI cudaMemcpy2DToArrayAsync(struct cudaArray *dst, s } -__host__ cudaError_t CUDARTAPI cudaMemcpy2DFromArrayAsync(void *dst, size_t dpitch, const struct cudaArray *src, size_t wOffset, size_t hOffset, size_t width, size_t height, enum cudaMemcpyKind kind, cudaStream_t stream) +__host__ cudaError_t CUDARTAPI cudaMemcpy2D(void *dst, size_t dpitch, const void *src, size_t spitch, size_t width, size_t height, enum cudaMemcpyKind kind) { - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - cuda_not_implemented(__my_func__,__LINE__); - return g_last_cudaError = cudaErrorUnknown; + return cudaMemcpy2DInternal(dst, dpitch, src, spitch, width, height, kind); } -#if (CUDART_VERSION >= 8000) -cudaError_t CUDARTAPI cudaOccupancyMaxActiveBlocksPerMultiprocessorWithFlags(int* numBlocks, const char *hostFunc, int blockSize, size_t dynamicSMemSize, unsigned int flags) +__host__ cudaError_t CUDARTAPI cudaMemcpy2DToArray(struct cudaArray *dst, size_t wOffset, size_t hOffset, const void *src, size_t spitch, size_t width, size_t height, enum cudaMemcpyKind kind) { - printf("GPGPU-Sim PTX: cudaOccupancyMaxActiveBlocksPerMultiprocessorWithFlags %p\n", hostFunc); - CUctx_st *context = GPGPUSim_Context(); - function_info *entry = context->get_kernel(hostFunc); - printf("Calculate Maxium Active Block with function ptr=%p, blockSize=%d, SMemSize=%d\n", hostFunc, blockSize, dynamicSMemSize); - if (flags == cudaOccupancyDefault) { - //create kernel_info based on entry - dim3 gridDim(context->get_device()->get_gpgpu()->max_cta_per_core() - * context->get_device()->get_gpgpu()->get_config().num_shader()); - dim3 blockDim(blockSize); - kernel_info_t result(gridDim, blockDim, entry); - //if(entry == NULL){ - // *numBlocks = 1; - // return g_last_cudaError = cudaErrorUnknown; - //} - *numBlocks = context->get_device()->get_gpgpu()->get_max_cta(result); - printf("Maximum size is %d with gridDim %d and blockDim %d\n", *numBlocks, gridDim.x, blockDim.x); - return g_last_cudaError = cudaSuccess; - } else { - cuda_not_implemented(__my_func__,__LINE__); - return g_last_cudaError = cudaErrorUnknown; - } + return cudaMemcpy2DToArrayInternal(dst, wOffset, hOffset, src, spitch, width, height, kind); } -#endif - - - -/******************************************************************************* - * * - * * - * * - *******************************************************************************/ -__host__ cudaError_t CUDARTAPI cudaMemset(void *mem, int c, size_t count) +__host__ cudaError_t CUDARTAPI cudaMemcpy2DFromArray(void *dst, size_t dpitch, const struct cudaArray *src, size_t wOffset, size_t hOffset, size_t width, size_t height, enum cudaMemcpyKind kind) { if(g_debug_execution >= 3){ announce_call(__my_func__); } - CUctx_st *context = GPGPUSim_Context(); - gpgpu_t *gpu = context->get_device()->get_gpgpu(); - gpu->gpu_memset((size_t)mem, c, count); - return g_last_cudaError = cudaSuccess; + cuda_not_implemented(__my_func__,__LINE__); + return g_last_cudaError = cudaErrorUnknown; } -#if (CUDART_VERSION > 5000) -__host__ cudaError_t CUDARTAPI cudaDeviceGetAttributeInternal(int *value, enum cudaDeviceAttr attr, int device, gpgpu_context* gpgpu_ctx = NULL) -{ - gpgpu_context *ctx; - if (gpgpu_ctx){ - ctx = gpgpu_ctx; - } else { - ctx = GPGPU_Context(); - } - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - - const struct cudaDeviceProp *prop; - _cuda_device_id *dev = ctx->GPGPUSim_Init(); - - if (device <= dev->num_devices() ) { - prop = dev->get_prop(); - switch (attr) { - case 1: - *value= prop->maxThreadsPerBlock; - break; - case 2: - *value= prop->maxThreadsDim[0]; - break; - case 3: - *value= prop->maxThreadsDim[1]; - break; - case 4: - *value= prop->maxThreadsDim[2]; - break; - case 5: - *value= prop->maxGridSize[0]; - break; - case 6: - *value= prop->maxGridSize[1]; - break; - case 7: - *value= prop->maxGridSize[2]; - break; - case 8: - *value= prop->sharedMemPerBlock; - break; - case 9: - *value= prop->totalConstMem; - break; - case 10: - *value= prop->warpSize; - break; - case 11: - *value= 16;//dummy value - break; - case 12: - *value= prop->regsPerBlock; - break; - case 13: - *value= 1480000;//for 1080ti - break; - case 14: - *value= prop->textureAlignment ; - break; - case 15: - *value = 0; - break; - case 16: - *value= prop->multiProcessorCount ; - break; - case 17: - case 18: - case 19: - *value = 0; - break; - case 21: - case 22: - case 23: - case 24: - case 25: - case 26: - case 27: - case 28: - case 42: - case 45: - case 46: - case 47: - case 48: - case 49: - case 52: - case 53: - case 55: - case 56: - case 57: - case 58: - case 59: - case 60: - case 61: - case 62: - case 63: - case 64: - case 66: - case 67: - case 69: - case 70: - case 71: - case 73: - case 74: - case 77: - *value = 1000;//dummy value - break; - case 29: - case 43: - case 54: - case 65: - case 68: - case 72: - *value = 10;//dummy value - break; - case 30: - case 51: - *value = 128;//dummy value - break; - case 31: - *value = 1; - break; - case 32: - *value = 0; - break; - case 33: - case 50: - *value = 0;//dummy value - break; - case 34: - *value= 0; - break; - case 35: - *value = 0; - break; - case 36: - *value = 1250000;//CK value for 1080ti - break; - case 37: - *value = 352;//value for 1080ti - break; - case 38: - *value = 3000000;//value for 1080ti - break; - case 39: - *value= dev->get_gpgpu()->threads_per_core(); - break; - case 40: - *value= 0; - break; - case 41: - *value= 0; - break; - case 75://cudaDevAttrComputeCapabilityMajor - *value= prop->major ; - break; - case 76://cudaDevAttrComputeCapabilityMinor - *value= prop->minor ; - break; - case 78: - *value= 0 ; //TODO: as of now, we dont support stream priorities. - break; - case 79: - *value= 0; - break; - case 80: - *value= 0; - break; - #if (CUDART_VERSION > 5050) - case 81: - *value= prop->sharedMemPerMultiprocessor; - break; - case 82: - *value= prop->regsPerMultiprocessor; - break; - #endif - case 83: - case 84: - case 85: - case 86: - *value= 0; - break; - case 87: - *value= 4;//dummy value - break; - case 88: - case 89: - *value= 0; - break; - default: - printf("ERROR: Attribute number %d unimplemented \n",attr); - abort(); - } - return g_last_cudaError = cudaSuccess; - } else { - return g_last_cudaError = cudaErrorInvalidDevice; - } +__host__ cudaError_t CUDARTAPI cudaMemcpy2DArrayToArray(struct cudaArray *dst, size_t wOffsetDst, size_t hOffsetDst, const struct cudaArray *src, size_t wOffsetSrc, size_t hOffsetSrc, size_t width, size_t height, enum cudaMemcpyKind kind __dv(cudaMemcpyDeviceToDevice)) +{ + if(g_debug_execution >= 3){ + announce_call(__my_func__); + } + cuda_not_implemented(__my_func__,__LINE__); + return g_last_cudaError = cudaErrorUnknown; } -#endif -//memset operation is done but i think its not async? -__host__ cudaError_t CUDARTAPI cudaMemsetAsync(void *mem, int c, size_t count, cudaStream_t stream=0) +__host__ cudaError_t CUDARTAPI cudaMemcpyToSymbol(const char *symbol, const void *src, size_t count, size_t offset __dv(0), enum cudaMemcpyKind kind __dv(cudaMemcpyHostToDevice)) { if(g_debug_execution >= 3){ announce_call(__my_func__); } - printf("GPGPU-Sim PTX: WARNING: Asynchronous memset not supported (%s)\n", __my_func__); - CUctx_st *context = GPGPUSim_Context(); - gpgpu_t *gpu = context->get_device()->get_gpgpu(); - gpu->gpu_memset((size_t)mem, c, count); + //CUctx_st *context = GPGPUSim_Context(); + assert(kind == cudaMemcpyHostToDevice); + printf("GPGPU-Sim PTX: cudaMemcpyToSymbol: symbol = %p\n", symbol); + //stream_operation( const char *symbol, const void *src, size_t count, size_t offset ) + g_stream_manager()->push( stream_operation(src,symbol,count,offset,0) ); + //gpgpu_ptx_sim_memcpy_symbol(symbol,src,count,offset,1,context->get_device()->get_gpgpu()); + return g_last_cudaError = cudaSuccess; +} + + +__host__ cudaError_t CUDARTAPI cudaMemcpyFromSymbol(void *dst, const char *symbol, size_t count, size_t offset __dv(0), enum cudaMemcpyKind kind __dv(cudaMemcpyDeviceToHost)) +{ + if(g_debug_execution >= 3){ + announce_call(__my_func__); + } + //CUctx_st *context = GPGPUSim_Context(); + assert(kind == cudaMemcpyDeviceToHost); + printf("GPGPU-Sim PTX: cudaMemcpyFromSymbol: symbol = %p\n", symbol); + g_stream_manager()->push( stream_operation(symbol,dst,count,offset,0) ); + //gpgpu_ptx_sim_memcpy_symbol(symbol,dst,count,offset,0,context->get_device()->get_gpgpu()); + return g_last_cudaError = cudaSuccess; +} + +__host__ cudaError_t CUDARTAPI cudaMemGetInfo (size_t *free, size_t *total){ + if(g_debug_execution >= 3){ + announce_call(__my_func__); + } + //placeholder; should interact with cudaMalloc and cudaFree? + *free = 10000000000; + *total = 10000000000; + + return g_last_cudaError = cudaSuccess; +} + +/******************************************************************************* + * * + * * + * * + *******************************************************************************/ + +__host__ cudaError_t CUDARTAPI cudaMemcpyAsync(void *dst, const void *src, size_t count, enum cudaMemcpyKind kind, cudaStream_t stream) +{ + if(g_debug_execution >= 3){ + announce_call(__my_func__); + } + struct CUstream_st *s = (struct CUstream_st *)stream; + switch( kind ) { + case cudaMemcpyHostToDevice: g_stream_manager()->push( stream_operation(src,(size_t)dst,count,s) ); break; + case cudaMemcpyDeviceToHost: g_stream_manager()->push( stream_operation((size_t)src,dst,count,s) ); break; + case cudaMemcpyDeviceToDevice: g_stream_manager()->push( stream_operation((size_t)src,(size_t)dst,count,s) ); break; + default: + abort(); + } return g_last_cudaError = cudaSuccess; } + +__host__ cudaError_t CUDARTAPI cudaMemcpyToArrayAsync(struct cudaArray *dst, size_t wOffset, size_t hOffset, const void *src, size_t count, enum cudaMemcpyKind kind, cudaStream_t stream) +{ + if(g_debug_execution >= 3){ + announce_call(__my_func__); + } + cuda_not_implemented(__my_func__,__LINE__); + return g_last_cudaError = cudaErrorUnknown; +} + + +__host__ cudaError_t CUDARTAPI cudaMemcpyFromArrayAsync(void *dst, const struct cudaArray *src, size_t wOffset, size_t hOffset, size_t count, enum cudaMemcpyKind kind, cudaStream_t stream) +{ + if(g_debug_execution >= 3){ + announce_call(__my_func__); + } + cuda_not_implemented(__my_func__,__LINE__); + return g_last_cudaError = cudaErrorUnknown; +} + + +__host__ cudaError_t CUDARTAPI cudaMemcpy2DAsync(void *dst, size_t dpitch, const void *src, size_t spitch, size_t width, size_t height, enum cudaMemcpyKind kind, cudaStream_t stream) +{ + if(g_debug_execution >= 3){ + announce_call(__my_func__); + } + cuda_not_implemented(__my_func__,__LINE__); + return g_last_cudaError = cudaErrorUnknown; +} + + +__host__ cudaError_t CUDARTAPI cudaMemcpy2DToArrayAsync(struct cudaArray *dst, size_t wOffset, size_t hOffset, const void *src, size_t spitch, size_t width, size_t height, enum cudaMemcpyKind kind, cudaStream_t stream) +{ + if(g_debug_execution >= 3){ + announce_call(__my_func__); + } + cuda_not_implemented(__my_func__,__LINE__); + return g_last_cudaError = cudaErrorUnknown; +} + + +__host__ cudaError_t CUDARTAPI cudaMemcpy2DFromArrayAsync(void *dst, size_t dpitch, const struct cudaArray *src, size_t wOffset, size_t hOffset, size_t width, size_t height, enum cudaMemcpyKind kind, cudaStream_t stream) +{ + if(g_debug_execution >= 3){ + announce_call(__my_func__); + } + cuda_not_implemented(__my_func__,__LINE__); + return g_last_cudaError = cudaErrorUnknown; +} + +#if (CUDART_VERSION >= 8000) +cudaError_t CUDARTAPI cudaOccupancyMaxActiveBlocksPerMultiprocessorWithFlags(int* numBlocks, const char *hostFunc, int blockSize, size_t dynamicSMemSize, unsigned int flags) +{ + return cudaOccupancyMaxActiveBlocksPerMultiprocessorWithFlagsInternal(numBlocks, hostFunc, blockSize, dynamicSMemSize, flags); +} + +#endif + + + +/******************************************************************************* + * * + * * + * * + *******************************************************************************/ +__host__ cudaError_t CUDARTAPI cudaMemset(void *mem, int c, size_t count) +{ + return cudaMemsetInternal(mem, c, count); +} + +//memset operation is done but i think its not async? +__host__ cudaError_t CUDARTAPI cudaMemsetAsync(void *mem, int c, size_t count, cudaStream_t stream=0) +{ + return cudaMemsetAsyncInternal(mem, c, count, stream=0); +} + __host__ cudaError_t CUDARTAPI cudaMemset2D(void *mem, size_t pitch, int c, size_t width, size_t height) { if(g_debug_execution >= 3){ @@ -1986,61 +2304,18 @@ __host__ cudaError_t CUDARTAPI cudaBindTexture(size_t *offset, const struct cudaChannelFormatDesc *desc, size_t size __dv(UINT_MAX)) { - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - CUctx_st *context = GPGPUSim_Context(); - gpgpu_t *gpu = context->get_device()->get_gpgpu(); - printf("GPGPU-Sim PTX: in cudaBindTexture: sizeof(struct textureReference) = %zu\n", sizeof(struct textureReference)); - struct cudaArray *array; - array = (struct cudaArray*) malloc(sizeof(struct cudaArray)); - array->desc = *desc; - array->size = size; - array->width = size; - array->height = 1; - array->dimensions = 1; - array->devPtr = (void*)devPtr; - array->devPtr32 = (int)(long long)devPtr; - offset = 0; - printf("GPGPU-Sim PTX: size = %zu\n", size); - printf("GPGPU-Sim PTX: texref = %p, array = %p\n", texref, array); - printf("GPGPU-Sim PTX: devPtr32 = %x\n", array->devPtr32); - printf("GPGPU-Sim PTX: Name corresponding to textureReference: %s\n", gpu->gpgpu_ptx_sim_findNamefromTexture(texref)); - printf("GPGPU-Sim PTX: ChannelFormatDesc: x=%d, y=%d, z=%d, w=%d\n", desc->x, desc->y, desc->z, desc->w); - printf("GPGPU-Sim PTX: Texture Normalized? = %d\n", texref->normalized); - gpu->gpgpu_ptx_sim_bindTextureToArray(texref, array); - devPtr = (void*)(long long)array->devPtr32; - printf("GPGPU-Sim PTX: devPtr = %p\n", devPtr); - return g_last_cudaError = cudaSuccess; + return cudaBindTextureInternal(offset, texref, devPtr, desc, size __dv(UINT_MAX)); } __host__ cudaError_t CUDARTAPI cudaBindTextureToArray(const struct textureReference *texref, const struct cudaArray *array, const struct cudaChannelFormatDesc *desc) { - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - CUctx_st *context = GPGPUSim_Context(); - gpgpu_t *gpu = context->get_device()->get_gpgpu(); - printf("GPGPU-Sim PTX: in cudaBindTextureToArray: %p %p\n", texref, array); - printf("GPGPU-Sim PTX: devPtr32 = %x\n", array->devPtr32); - printf("GPGPU-Sim PTX: Name corresponding to textureReference: %s\n", gpu->gpgpu_ptx_sim_findNamefromTexture(texref)); - printf("GPGPU-Sim PTX: Texture Normalized? = %d\n", texref->normalized); - gpu->gpgpu_ptx_sim_bindTextureToArray(texref, array); - return g_last_cudaError = cudaSuccess; + return cudaBindTextureToArrayInternal(texref, array, desc); } -__host__ cudaError_t CUDARTAPI cudaUnbindTexture(const struct textureReference *texref){ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - CUctx_st *context = GPGPUSim_Context(); - gpgpu_t *gpu = context->get_device()->get_gpgpu(); - printf("GPGPU-Sim PTX: in cudaUnbindTexture: sizeof(struct textureReference) = %zu\n", sizeof(struct textureReference)); - printf("GPGPU-Sim PTX: Name corresponding to textureReference: %s\n", gpu->gpgpu_ptx_sim_findNamefromTexture(texref)); - - gpu->gpgpu_ptx_sim_unbindTexture(texref); - return g_last_cudaError = cudaSuccess; +__host__ cudaError_t CUDARTAPI cudaUnbindTexture(const struct textureReference *texref) +{ + return cudaUnbindTextureInternal(texref); } __host__ cudaError_t CUDARTAPI cudaGetTextureAlignmentOffset(size_t *offset, const struct textureReference *texref) @@ -2125,24 +2400,9 @@ __host__ cudaError_t CUDARTAPI cudaLaunch( const char *hostFun ) return cudaLaunchInternal( hostFun ); } -__host__ cudaError_t CUDARTAPI cudaLaunchKernel ( const char* hostFun, dim3 gridDim, dim3 blockDim, const void** args, size_t sharedMem, cudaStream_t stream ) +__host__ cudaError_t CUDARTAPI cudaLaunchKernel( const char* hostFun, dim3 gridDim, dim3 blockDim, const void** args, size_t sharedMem, cudaStream_t stream ) { - - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - CUctx_st *context = GPGPUSim_Context(); - function_info *entry = context->get_kernel(hostFun); -#if CUDART_VERSION < 10000 - cudaConfigureCallInternal(gridDim, blockDim, sharedMem, stream); -#endif - for(unsigned i = 0; i < entry->num_args(); i++){ - std::pair p = entry->get_param_config(i); - cudaSetupArgumentInternal(args[i], p.first, p.second); - } - - cudaLaunchInternal(hostFun); - return g_last_cudaError = cudaSuccess; + return cudaLaunchKernelInternal(hostFun, gridDim, blockDim, args, sharedMem, stream); } @@ -2528,7 +2788,7 @@ void cuda_runtime_api::extract_ptx_files_using_cuobjdump(CUctx_st *context){ * enabled * */ void cuda_runtime_api::extract_code_using_cuobjdump(){ - CUctx_st *context = GPGPUSim_Context(); + CUctx_st *context = GPGPUSim_Context(gpgpu_ctx); //prevent the dumping by cuobjdump everytime we execute the code! const char *override_cuobjdump = getenv("CUOBJDUMP_SIM_FILE"); @@ -2886,7 +3146,7 @@ cuobjdumpPTXSection* cuda_runtime_api::findPTXSection(const std::string identifi //! Extract the code using cuobjdump and remove unnecessary sections void cuda_runtime_api::cuobjdumpInit(){ - CUctx_st *context = GPGPUSim_Context(); + CUctx_st *context = GPGPUSim_Context(gpgpu_ctx); extract_code_using_cuobjdump(); //extract all the output of cuobjdump to _cuobjdump_*.* const char* pre_load = getenv("CUOBJDUMP_SIM_FILE"); if (pre_load ==NULL || strlen(pre_load)==0){ @@ -2899,7 +3159,7 @@ void cuda_runtime_api::cuobjdumpInit(){ //! Either submit PTX for simulation or convert SASS to PTXPlus and submit it void gpgpu_context::cuobjdumpParseBinary(unsigned int handle){ - CUctx_st *context = GPGPUSim_Context(); + CUctx_st *context = GPGPUSim_Context(this); if(api->fatbin_registered[handle]) return; api->fatbin_registered[handle] = true; std::string fname = api->fatbinmap[handle]; @@ -3140,22 +3400,7 @@ void __cudaRegisterTexture( int ext ) //passes in a newly created textureReference { - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - std::string devStr (deviceName); - #if (CUDART_VERSION > 4020) - if (devStr.size() > 2 && devStr.data()[0] == ':' && devStr.data()[1] == ':') - devStr = devStr.replace(0, 2, ""); - #endif - CUctx_st *context = GPGPUSim_Context(); - gpgpu_t *gpu = context->get_device()->get_gpgpu(); - printf("GPGPU-Sim PTX: in __cudaRegisterTexture:\n"); - gpu->gpgpu_ptx_sim_bindNameToTexture(devStr.data(), hostVar, dim, norm, ext); - printf("GPGPU-Sim PTX: int dim = %d\n", dim); - printf("GPGPU-Sim PTX: int norm = %d\n", norm); - printf("GPGPU-Sim PTX: int ext = %d\n", ext); - printf("GPGPU-Sim PTX: Execution warning: Not finished implementing \"%s\"\n", __my_func__ ); + __cudaRegisterTextureInternal(fatCubinHandle, hostVar, deviceAddress, deviceName, dim, norm, ext); } @@ -3187,30 +3432,7 @@ cudaError_t cudaGLMapBufferObject(void** devPtr, GLuint bufferObj) cudaError_t cudaGLUnmapBufferObject(GLuint bufferObj) { - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } -#ifdef OPENGL_SUPPORT - CUctx_st* ctx = GPGPUSim_Context(); - glbmap_entry_t *p = ctx->api->g_glbmap; - while ( p && p->m_bufferObj != bufferObj ) - p = p->m_next; - if ( p == NULL ) - return g_last_cudaError = cudaErrorUnknown; - - char *data = (char *) calloc(p->m_size,1); - memcpy_from_gpu( data,(size_t)p->m_devPtr,p->m_size ); - glBufferSubData(GL_ARRAY_BUFFER,0,p->m_size,data); - free(data); - - return g_last_cudaError = cudaSuccess; -#else - fflush(stdout); - fflush(stderr); - printf("GPGPU-Sim PTX: support for OpenGL integration disabled -- exiting\n"); - fflush(stdout); - exit(50); -#endif + return cudaGLUnmapBufferObjectInternal(bufferObj); } cudaError_t cudaGLUnregisterBufferObject(GLuint bufferObj) @@ -3322,12 +3544,7 @@ cudaError_t CUDARTAPI cudaRuntimeGetVersion(int *runtimeVersion) #if CUDART_VERSION >= 3000 __host__ cudaError_t CUDARTAPI cudaFuncSetCacheConfig(const char *func, enum cudaFuncCache cacheConfig ) { - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - CUctx_st *context = GPGPUSim_Context(); - context->get_device()->get_gpgpu()->set_cache_config(context->get_kernel(func)->get_name(), (FuncCache)cacheConfig); - return g_last_cudaError = cudaSuccess; + return cudaFuncSetCacheConfigInternal(func, cacheConfig); } //Jin: hack for cdp @@ -5033,23 +5250,7 @@ CUresult CUDAAPI cuLaunchKernel(CUfunction f, void **kernelParams, void **extra) { - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - if (extra!=NULL){ - printf("GPGPU-Sim CUDA DRIVER API: ERROR: Currently do not support void** extra.\n"); - abort(); - } - const char *hostFun = (const char*) f; - CUctx_st *context = GPGPUSim_Context(); - function_info *entry = context->get_kernel(hostFun); - cudaConfigureCallInternal(dim3(gridDimX, gridDimY, gridDimZ), dim3(blockDimX, blockDimY, blockDimZ), sharedMemBytes, (cudaStream_t) hStream); - for(unsigned i = 0; i < entry->num_args(); i++){ - std::pair p = entry->get_param_config(i); - cudaSetupArgument(kernelParams[i], p.first, p.second); - } - cudaLaunchInternal(hostFun); - return CUDA_SUCCESS; + return cuLaunchKernelInternal(f, gridDimX, gridDimY, gridDimZ, blockDimX, blockDimY, blockDimZ, sharedMemBytes, hStream, kernelParams, extra); } #endif /* CUDART_VERSION >= 4000 */ -- cgit v1.3 From 4671c7b3b3f1792f6f4b6c98e2bcffe7d072bac0 Mon Sep 17 00:00:00 2001 From: Mengchi Zhang Date: Wed, 11 Sep 2019 21:06:37 -0400 Subject: Fix init bug in cudaMallocArray Signed-off-by: Mengchi Zhang --- libcuda/cuda_runtime_api.cc | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) (limited to 'libcuda/cuda_runtime_api.cc') diff --git a/libcuda/cuda_runtime_api.cc b/libcuda/cuda_runtime_api.cc index 175cbc5..8930d61 100644 --- a/libcuda/cuda_runtime_api.cc +++ b/libcuda/cuda_runtime_api.cc @@ -1892,7 +1892,7 @@ __host__ cudaError_t CUDARTAPI cudaMallocPitch(void **devPtr, size_t *pitch, siz __host__ cudaError_t CUDARTAPI cudaMallocArray(struct cudaArray **array, const struct cudaChannelFormatDesc *desc, size_t width, size_t height __dv(1)) { - return cudaMallocArrayInternal(array, desc, width, height __dv(1)); + return cudaMallocArrayInternal(array, desc, width, height); } __host__ cudaError_t CUDARTAPI cudaFree(void *devPtr) -- cgit v1.3 From 5181fe19601cf10ce976c5a5965b993cf0213260 Mon Sep 17 00:00:00 2001 From: Mengchi Zhang Date: Wed, 11 Sep 2019 23:09:12 -0400 Subject: Remove g_stream_manager() Signed-off-by: Mengchi Zhang --- libcuda/cuda_runtime_api.cc | 348 ++++++++++++++++++++++-------------- src/abstract_hardware_model.cc | 8 +- src/cuda-sim/cuda-sim.cc | 2 +- src/cuda-sim/cuda_device_runtime.cc | 3 +- src/gpgpusim_entrypoint.cc | 4 - src/gpgpusim_entrypoint.h | 1 - 6 files changed, 224 insertions(+), 142 deletions(-) (limited to 'libcuda/cuda_runtime_api.cc') diff --git a/libcuda/cuda_runtime_api.cc b/libcuda/cuda_runtime_api.cc index 8930d61..9459b64 100644 --- a/libcuda/cuda_runtime_api.cc +++ b/libcuda/cuda_runtime_api.cc @@ -176,8 +176,6 @@ struct cudaArray cudaError_t g_last_cudaError = cudaSuccess; -//extern stream_manager *g_stream_manager(); - void register_ptx_function( const char *name, function_info *impl ) { // no longer need this @@ -941,7 +939,7 @@ cudaError_t cudaLaunchInternal( const char *hostFun, gpgpu_context* gpgpu_ctx = printf("GPGPU-Sim PTX: pushing kernel \'%s\' to stream %u, gridDim= (%u,%u,%u) blockDim = (%u,%u,%u) \n", kname.c_str(), stream?stream->get_uid():0, gridDim.x,gridDim.y,gridDim.z,blockDim.x,blockDim.y,blockDim.z ); stream_operation op(grid,ctx->func_sim->g_ptx_sim_mode,stream); - g_stream_manager()->push(op); + ctx->the_gpgpusim->g_stream_manager->push(op); ctx->api->g_cuda_launch_stack.pop_back(); return g_last_cudaError = cudaSuccess; } @@ -1080,6 +1078,50 @@ __host__ cudaError_t CUDARTAPI cudaMallocArrayInternal(struct cudaArray **array, } } +__host__ cudaError_t CUDARTAPI cudaMemcpyInternal(void *dst, const void *src, size_t count, enum cudaMemcpyKind kind, gpgpu_context* gpgpu_ctx = NULL) +{ + gpgpu_context *ctx; + if (gpgpu_ctx){ + ctx = gpgpu_ctx; + } else { + ctx = GPGPU_Context(); + } + if(g_debug_execution >= 3){ + announce_call(__my_func__); + } + //CUctx_st *context = GPGPUSim_Context(); + //gpgpu_t *gpu = context->get_device()->get_gpgpu(); + if(g_debug_execution >= 3) + printf("GPGPU-Sim PTX: cudaMemcpy(): devPtr = %p\n", dst); + if( kind == cudaMemcpyHostToDevice ) + ctx->the_gpgpusim->g_stream_manager->push( stream_operation(src,(size_t)dst,count,0) ); + else if( kind == cudaMemcpyDeviceToHost ) + ctx->the_gpgpusim->g_stream_manager->push( stream_operation((size_t)src,dst,count,0) ); + else if( kind == cudaMemcpyDeviceToDevice ) + ctx->the_gpgpusim->g_stream_manager->push( stream_operation((size_t)src,(size_t)dst,count,0) ); + else if ( kind == cudaMemcpyDefault ) { + if ((size_t)src >= GLOBAL_HEAP_START) { + if ((size_t)dst >= GLOBAL_HEAP_START) + ctx->the_gpgpusim->g_stream_manager->push( stream_operation((size_t)src,(size_t)dst,count,0) ); // device to device + else + ctx->the_gpgpusim->g_stream_manager->push( stream_operation((size_t)src,dst,count,0) ); // device to host + } + else { + if ((size_t)dst >= GLOBAL_HEAP_START) + ctx->the_gpgpusim->g_stream_manager->push( stream_operation(src,(size_t)dst,count,0) ); + else { + printf("GPGPU-Sim PTX: cudaMemcpy - ERROR : unsupported transfer: host to host\n"); + abort(); + } + } + } + else { + printf("GPGPU-Sim PTX: cudaMemcpy - ERROR : unsupported cudaMemcpyKind\n"); + abort(); + } + return g_last_cudaError = cudaSuccess; +} + __host__ cudaError_t CUDARTAPI cudaMemcpyToArrayInternal(struct cudaArray *dst, size_t wOffset, size_t hOffset, const void *src, size_t count, enum cudaMemcpyKind kind, gpgpu_context* gpgpu_ctx = NULL) { gpgpu_context *ctx; @@ -1174,6 +1216,69 @@ __host__ cudaError_t CUDARTAPI cudaMemcpy2DToArrayInternal(struct cudaArray *dst return g_last_cudaError = cudaSuccess; } +__host__ cudaError_t CUDARTAPI cudaMemcpyToSymbolInternal(const char *symbol, const void *src, size_t count, size_t offset __dv(0), enum cudaMemcpyKind kind __dv(cudaMemcpyHostToDevice), gpgpu_context* gpgpu_ctx = NULL) +{ + gpgpu_context *ctx; + if (gpgpu_ctx){ + ctx = gpgpu_ctx; + } else { + ctx = GPGPU_Context(); + } + if(g_debug_execution >= 3){ + announce_call(__my_func__); + } + //CUctx_st *context = GPGPUSim_Context(); + assert(kind == cudaMemcpyHostToDevice); + printf("GPGPU-Sim PTX: cudaMemcpyToSymbol: symbol = %p\n", symbol); + //stream_operation( const char *symbol, const void *src, size_t count, size_t offset ) + ctx->the_gpgpusim->g_stream_manager->push( stream_operation(src,symbol,count,offset,0) ); + //gpgpu_ptx_sim_memcpy_symbol(symbol,src,count,offset,1,context->get_device()->get_gpgpu()); + return g_last_cudaError = cudaSuccess; +} + + +__host__ cudaError_t CUDARTAPI cudaMemcpyFromSymbolInternal(void *dst, const char *symbol, size_t count, size_t offset __dv(0), enum cudaMemcpyKind kind __dv(cudaMemcpyDeviceToHost), gpgpu_context* gpgpu_ctx = NULL) +{ + gpgpu_context *ctx; + if (gpgpu_ctx){ + ctx = gpgpu_ctx; + } else { + ctx = GPGPU_Context(); + } + if(g_debug_execution >= 3){ + announce_call(__my_func__); + } + //CUctx_st *context = GPGPUSim_Context(); + assert(kind == cudaMemcpyDeviceToHost); + printf("GPGPU-Sim PTX: cudaMemcpyFromSymbol: symbol = %p\n", symbol); + ctx->the_gpgpusim->g_stream_manager->push( stream_operation(symbol,dst,count,offset,0) ); + //gpgpu_ptx_sim_memcpy_symbol(symbol,dst,count,offset,0,context->get_device()->get_gpgpu()); + return g_last_cudaError = cudaSuccess; +} + +__host__ cudaError_t CUDARTAPI cudaMemcpyAsyncInternal(void *dst, const void *src, size_t count, enum cudaMemcpyKind kind, cudaStream_t stream, gpgpu_context* gpgpu_ctx = NULL) +{ + gpgpu_context *ctx; + if (gpgpu_ctx){ + ctx = gpgpu_ctx; + } else { + ctx = GPGPU_Context(); + } + if(g_debug_execution >= 3){ + announce_call(__my_func__); + } + struct CUstream_st *s = (struct CUstream_st *)stream; + switch( kind ) { + case cudaMemcpyHostToDevice: ctx->the_gpgpusim->g_stream_manager->push( stream_operation(src,(size_t)dst,count,s) ); break; + case cudaMemcpyDeviceToHost: ctx->the_gpgpusim->g_stream_manager->push( stream_operation((size_t)src,dst,count,s) ); break; + case cudaMemcpyDeviceToDevice: ctx->the_gpgpusim->g_stream_manager->push( stream_operation((size_t)src,(size_t)dst,count,s) ); break; + default: + abort(); + } + return g_last_cudaError = cudaSuccess; +} + + #if (CUDART_VERSION >= 8000) cudaError_t CUDARTAPI cudaOccupancyMaxActiveBlocksPerMultiprocessorWithFlagsInternal(int* numBlocks, const char *hostFunc, int blockSize, size_t dynamicSMemSize, unsigned int flags, gpgpu_context* gpgpu_ctx = NULL) { @@ -1729,6 +1834,48 @@ __host__ cudaError_t CUDARTAPI cudaLaunchKernelInternal( const char* hostFun, di return g_last_cudaError = cudaSuccess; } +__host__ cudaError_t CUDARTAPI cudaStreamCreateInternal(cudaStream_t *stream, gpgpu_context* gpgpu_ctx = NULL) +{ + gpgpu_context *ctx; + if (gpgpu_ctx){ + ctx = gpgpu_ctx; + } else { + ctx = GPGPU_Context(); + } + if(g_debug_execution >= 3){ + announce_call(__my_func__); + } + printf("GPGPU-Sim PTX: cudaStreamCreate\n"); +#if (CUDART_VERSION >= 3000) + *stream = new struct CUstream_st(); + ctx->the_gpgpusim->g_stream_manager->add_stream(*stream); +#else + *stream = 0; + printf("GPGPU-Sim PTX: WARNING: Asynchronous kernel execution not supported (%s)\n", __my_func__); +#endif + return g_last_cudaError = cudaSuccess; +} + +__host__ cudaError_t CUDARTAPI cudaStreamDestroyInternal(cudaStream_t stream, gpgpu_context* gpgpu_ctx = NULL) +{ + gpgpu_context *ctx; + if (gpgpu_ctx){ + ctx = gpgpu_ctx; + } else { + ctx = GPGPU_Context(); + } + if(g_debug_execution >= 3){ + announce_call(__my_func__); + } +#if (CUDART_VERSION >= 3000) + //per-stream synchronization required for application using external libraries without explicit synchronization in the code to + //avoid the stream_manager from spinning forever to destroy non-empty streams without making any forward progress. + stream->synchronize(); + ctx->the_gpgpusim->g_stream_manager->destroy_stream(stream); +#endif + return g_last_cudaError = cudaSuccess; +} + void __cudaRegisterTextureInternal( void **fatCubinHandle, const struct textureReference *hostVar, @@ -1858,6 +2005,66 @@ CUresult CUDAAPI cuLaunchKernelInternal(CUfunction f, } #endif /* CUDART_VERSION >= 4000 */ +CUevent_st *get_event(cudaEvent_t event) +{ + unsigned event_uid; +#if CUDART_VERSION >= 3000 + event_uid = event->get_uid(); +#else + event_uid = event; +#endif + event_tracker_t::iterator e = g_timer_events.find(event_uid); + if( e == g_timer_events.end() ) + return NULL; + return e->second; +} + +__host__ cudaError_t CUDARTAPI cudaEventRecordInternal(cudaEvent_t event, cudaStream_t stream, gpgpu_context* gpgpu_ctx = NULL) +{ + gpgpu_context *ctx; + if (gpgpu_ctx){ + ctx = gpgpu_ctx; + } else { + ctx = GPGPU_Context(); + } + if(g_debug_execution >= 3){ + announce_call(__my_func__); + } + CUevent_st *e = get_event(event); + if( !e ) return g_last_cudaError = cudaErrorUnknown; + struct CUstream_st *s = (struct CUstream_st *)stream; + stream_operation op(e,s); + ctx->the_gpgpusim->g_stream_manager->push(op); + return g_last_cudaError = cudaSuccess; +} + +__host__ cudaError_t CUDARTAPI cudaStreamWaitEventInternal(cudaStream_t stream, cudaEvent_t event, unsigned int flags, gpgpu_context* gpgpu_ctx = NULL) +{ + gpgpu_context *ctx; + if (gpgpu_ctx){ + ctx = gpgpu_ctx; + } else { + ctx = GPGPU_Context(); + } + if(g_debug_execution >= 3){ + announce_call(__my_func__); + } + //reference: https://www.cs.cmu.edu/afs/cs/academic/class/15668-s11/www/cuda-doc/html/group__CUDART__STREAM_gfe68d207dc965685d92d3f03d77b0876.html + CUevent_st *e = get_event(event); + if( !e ){ + printf("GPGPU-Sim API: Warning: cudaEventRecord has not been called on event before calling cudaStreamWaitEvent.\nNothing to be done.\n"); + return g_last_cudaError = cudaSuccess; + } + if (!stream){ + ctx->the_gpgpusim->g_stream_manager->pushCudaStreamWaitEventToAllStreams(e, flags); + } else { + struct CUstream_st *s = (struct CUstream_st *)stream; + stream_operation op(s,e,flags); + ctx->the_gpgpusim->g_stream_manager->push(op); + } + return g_last_cudaError = cudaSuccess; +} + /******************************************************************************* * * * * @@ -1930,40 +2137,7 @@ __host__ cudaError_t CUDARTAPI cudaFreeArray(struct cudaArray *array) __host__ cudaError_t CUDARTAPI cudaMemcpy(void *dst, const void *src, size_t count, enum cudaMemcpyKind kind) { - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - //CUctx_st *context = GPGPUSim_Context(); - //gpgpu_t *gpu = context->get_device()->get_gpgpu(); - if(g_debug_execution >= 3) - printf("GPGPU-Sim PTX: cudaMemcpy(): devPtr = %p\n", dst); - if( kind == cudaMemcpyHostToDevice ) - g_stream_manager()->push( stream_operation(src,(size_t)dst,count,0) ); - else if( kind == cudaMemcpyDeviceToHost ) - g_stream_manager()->push( stream_operation((size_t)src,dst,count,0) ); - else if( kind == cudaMemcpyDeviceToDevice ) - g_stream_manager()->push( stream_operation((size_t)src,(size_t)dst,count,0) ); - else if ( kind == cudaMemcpyDefault ) { - if ((size_t)src >= GLOBAL_HEAP_START) { - if ((size_t)dst >= GLOBAL_HEAP_START) - g_stream_manager()->push( stream_operation((size_t)src,(size_t)dst,count,0) ); // device to device - else - g_stream_manager()->push( stream_operation((size_t)src,dst,count,0) ); // device to host - } - else { - if ((size_t)dst >= GLOBAL_HEAP_START) - g_stream_manager()->push( stream_operation(src,(size_t)dst,count,0) ); - else { - printf("GPGPU-Sim PTX: cudaMemcpy - ERROR : unsupported transfer: host to host\n"); - abort(); - } - } - } - else { - printf("GPGPU-Sim PTX: cudaMemcpy - ERROR : unsupported cudaMemcpyKind\n"); - abort(); - } - return g_last_cudaError = cudaSuccess; + return cudaMemcpyInternal(dst, src, count, kind); } __host__ cudaError_t CUDARTAPI cudaMemcpyToArray(struct cudaArray *dst, size_t wOffset, size_t hOffset, const void *src, size_t count, enum cudaMemcpyKind kind) @@ -2022,30 +2196,13 @@ __host__ cudaError_t CUDARTAPI cudaMemcpy2DArrayToArray(struct cudaArray *dst, s __host__ cudaError_t CUDARTAPI cudaMemcpyToSymbol(const char *symbol, const void *src, size_t count, size_t offset __dv(0), enum cudaMemcpyKind kind __dv(cudaMemcpyHostToDevice)) { - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - //CUctx_st *context = GPGPUSim_Context(); - assert(kind == cudaMemcpyHostToDevice); - printf("GPGPU-Sim PTX: cudaMemcpyToSymbol: symbol = %p\n", symbol); - //stream_operation( const char *symbol, const void *src, size_t count, size_t offset ) - g_stream_manager()->push( stream_operation(src,symbol,count,offset,0) ); - //gpgpu_ptx_sim_memcpy_symbol(symbol,src,count,offset,1,context->get_device()->get_gpgpu()); - return g_last_cudaError = cudaSuccess; + return cudaMemcpyToSymbolInternal(symbol, src, count, offset, kind); } __host__ cudaError_t CUDARTAPI cudaMemcpyFromSymbol(void *dst, const char *symbol, size_t count, size_t offset __dv(0), enum cudaMemcpyKind kind __dv(cudaMemcpyDeviceToHost)) { - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - //CUctx_st *context = GPGPUSim_Context(); - assert(kind == cudaMemcpyDeviceToHost); - printf("GPGPU-Sim PTX: cudaMemcpyFromSymbol: symbol = %p\n", symbol); - g_stream_manager()->push( stream_operation(symbol,dst,count,offset,0) ); - //gpgpu_ptx_sim_memcpy_symbol(symbol,dst,count,offset,0,context->get_device()->get_gpgpu()); - return g_last_cudaError = cudaSuccess; + return cudaMemcpyFromSymbolInternal(dst, symbol, count, offset, kind); } __host__ cudaError_t CUDARTAPI cudaMemGetInfo (size_t *free, size_t *total){ @@ -2067,18 +2224,7 @@ __host__ cudaError_t CUDARTAPI cudaMemGetInfo (size_t *free, size_t *total){ __host__ cudaError_t CUDARTAPI cudaMemcpyAsync(void *dst, const void *src, size_t count, enum cudaMemcpyKind kind, cudaStream_t stream) { - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - struct CUstream_st *s = (struct CUstream_st *)stream; - switch( kind ) { - case cudaMemcpyHostToDevice: g_stream_manager()->push( stream_operation(src,(size_t)dst,count,s) ); break; - case cudaMemcpyDeviceToHost: g_stream_manager()->push( stream_operation((size_t)src,dst,count,s) ); break; - case cudaMemcpyDeviceToDevice: g_stream_manager()->push( stream_operation((size_t)src,(size_t)dst,count,s) ); break; - default: - abort(); - } - return g_last_cudaError = cudaSuccess; + return cudaMemcpyAsyncInternal(dst, src, count, kind, stream); } @@ -2414,18 +2560,7 @@ __host__ cudaError_t CUDARTAPI cudaLaunchKernel( const char* hostFun, dim3 gridD __host__ cudaError_t CUDARTAPI cudaStreamCreate(cudaStream_t *stream) { - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("GPGPU-Sim PTX: cudaStreamCreate\n"); -#if (CUDART_VERSION >= 3000) - *stream = new struct CUstream_st(); - g_stream_manager()->add_stream(*stream); -#else - *stream = 0; - printf("GPGPU-Sim PTX: WARNING: Asynchronous kernel execution not supported (%s)\n", __my_func__); -#endif - return g_last_cudaError = cudaSuccess; + return cudaStreamCreateInternal(stream); } //TODO: introduce priorities @@ -2452,16 +2587,7 @@ __host__ __device__ cudaError_t CUDARTAPI cudaStreamCreateWithFlags(cudaStream_t __host__ cudaError_t CUDARTAPI cudaStreamDestroy(cudaStream_t stream) { - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } -#if (CUDART_VERSION >= 3000) - //per-stream synchronization required for application using external libraries without explicit synchronization in the code to - //avoid the stream_manager from spinning forever to destroy non-empty streams without making any forward progress. - stream->synchronize(); - g_stream_manager()->destroy_stream(stream); -#endif - return g_last_cudaError = cudaSuccess; + return cudaStreamDestroyInternal(stream); } __host__ cudaError_t CUDARTAPI cudaStreamSynchronize(cudaStream_t stream) @@ -2516,52 +2642,14 @@ __host__ cudaError_t CUDARTAPI cudaEventCreate(cudaEvent_t *event) return g_last_cudaError = cudaSuccess; } -CUevent_st *get_event(cudaEvent_t event) -{ - unsigned event_uid; -#if CUDART_VERSION >= 3000 - event_uid = event->get_uid(); -#else - event_uid = event; -#endif - event_tracker_t::iterator e = g_timer_events.find(event_uid); - if( e == g_timer_events.end() ) - return NULL; - return e->second; -} - __host__ cudaError_t CUDARTAPI cudaEventRecord(cudaEvent_t event, cudaStream_t stream) { - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - CUevent_st *e = get_event(event); - if( !e ) return g_last_cudaError = cudaErrorUnknown; - struct CUstream_st *s = (struct CUstream_st *)stream; - stream_operation op(e,s); - g_stream_manager()->push(op); - return g_last_cudaError = cudaSuccess; + return cudaEventRecordInternal(event, stream); } __host__ cudaError_t CUDARTAPI cudaStreamWaitEvent(cudaStream_t stream, cudaEvent_t event, unsigned int flags) { - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - //reference: https://www.cs.cmu.edu/afs/cs/academic/class/15668-s11/www/cuda-doc/html/group__CUDART__STREAM_gfe68d207dc965685d92d3f03d77b0876.html - CUevent_st *e = get_event(event); - if( !e ){ - printf("GPGPU-Sim API: Warning: cudaEventRecord has not been called on event before calling cudaStreamWaitEvent.\nNothing to be done.\n"); - return g_last_cudaError = cudaSuccess; - } - if (!stream){ - g_stream_manager()->pushCudaStreamWaitEventToAllStreams(e, flags); - } else { - struct CUstream_st *s = (struct CUstream_st *)stream; - stream_operation op(s,e,flags); - g_stream_manager()->push(op); - } - return g_last_cudaError = cudaSuccess; + return cudaStreamWaitEventInternal(stream, event, flags); } __host__ cudaError_t CUDARTAPI cudaEventQuery(cudaEvent_t event) diff --git a/src/abstract_hardware_model.cc b/src/abstract_hardware_model.cc index 9a91818..07232ee 100644 --- a/src/abstract_hardware_model.cc +++ b/src/abstract_hardware_model.cc @@ -808,14 +808,14 @@ void kernel_info_t::notify_parent_finished() { if(m_parent_kernel) { m_kernel_entry->gpgpu_ctx->device_runtime->g_total_param_size -= ((m_kernel_entry->get_args_aligned_size() + 255)/256*256); m_parent_kernel->remove_child(this); - g_stream_manager()->register_finished_kernel(m_parent_kernel->get_uid()); + m_kernel_entry->gpgpu_ctx->the_gpgpusim->g_stream_manager->register_finished_kernel(m_parent_kernel->get_uid()); } } CUstream_st * kernel_info_t::create_stream_cta(dim3 ctaid) { assert(get_default_stream_cta(ctaid)); CUstream_st * stream = new CUstream_st(); - g_stream_manager()->add_stream(stream); + m_kernel_entry->gpgpu_ctx->the_gpgpusim->g_stream_manager->add_stream(stream); assert(m_cta_streams.find(ctaid) != m_cta_streams.end()); assert(m_cta_streams[ctaid].size() >= 1); //must have default stream m_cta_streams[ctaid].push_back(stream); @@ -831,7 +831,7 @@ CUstream_st * kernel_info_t::get_default_stream_cta(dim3 ctaid) { else { m_cta_streams[ctaid] = std::list(); CUstream_st * stream = new CUstream_st(); - g_stream_manager()->add_stream(stream); + m_kernel_entry->gpgpu_ctx->the_gpgpusim->g_stream_manager->add_stream(stream); m_cta_streams[ctaid].push_back(stream); return stream; } @@ -863,7 +863,7 @@ void kernel_info_t::destroy_cta_streams() { for(auto s = m_cta_streams.begin(); s != m_cta_streams.end(); s++) { stream_size += s->second.size(); for(auto ss = s->second.begin(); ss != s->second.end(); ss++) - g_stream_manager()->destroy_stream(*ss); + m_kernel_entry->gpgpu_ctx->the_gpgpusim->g_stream_manager->destroy_stream(*ss); s->second.clear(); } printf("size %lu\n", stream_size); diff --git a/src/cuda-sim/cuda-sim.cc b/src/cuda-sim/cuda-sim.cc index 67d7f9b..fc1ea8e 100644 --- a/src/cuda-sim/cuda-sim.cc +++ b/src/cuda-sim/cuda-sim.cc @@ -2188,7 +2188,7 @@ void cuda_sim::gpgpu_cuda_ptx_sim_main_func( kernel_info_t &kernel, bool openCL //openCL kernel simulation calls don't register the kernel so we don't register its exit if(!openCL) { //extern stream_manager *g_stream_manager; - g_stream_manager()->register_finished_kernel(kernel.get_uid()); + gpgpu_ctx->the_gpgpusim->g_stream_manager->register_finished_kernel(kernel.get_uid()); } //******PRINTING******* diff --git a/src/cuda-sim/cuda_device_runtime.cc b/src/cuda-sim/cuda_device_runtime.cc index dc3adc3..4baced5 100644 --- a/src/cuda-sim/cuda_device_runtime.cc +++ b/src/cuda-sim/cuda_device_runtime.cc @@ -27,7 +27,6 @@ } -//extern stream_manager *g_stream_manager(); //Handling device runtime api: //void * cudaGetParameterBufferV2(void *func, dim3 gridDimension, dim3 blockDimension, unsigned int sharedMemSize) @@ -285,7 +284,7 @@ void cuda_device_runtime::launch_one_device_kernel() { device_launch_operation_t &op = g_cuda_device_launch_op.front(); stream_operation stream_op = stream_operation(op.grid, gpgpu_ctx->func_sim->g_ptx_sim_mode, op.stream); - g_stream_manager()->push(stream_op); + gpgpu_ctx->the_gpgpusim->g_stream_manager->push(stream_op); g_cuda_device_launch_op.pop_front(); } } diff --git a/src/gpgpusim_entrypoint.cc b/src/gpgpusim_entrypoint.cc index 19a525e..5ce40c6 100644 --- a/src/gpgpusim_entrypoint.cc +++ b/src/gpgpusim_entrypoint.cc @@ -52,10 +52,6 @@ GPGPUsim_ctx* GPGPUsim_ctx_ptr(){ return the_gpgpusim; } -class stream_manager* g_stream_manager() { - return GPGPUsim_ctx_ptr()->g_stream_manager; -} - static void print_simulation_time(); void *gpgpu_sim_thread_sequential(void*) diff --git a/src/gpgpusim_entrypoint.h b/src/gpgpusim_entrypoint.h index ba3aac9..b2e22e6 100644 --- a/src/gpgpusim_entrypoint.h +++ b/src/gpgpusim_entrypoint.h @@ -79,7 +79,6 @@ class GPGPUsim_ctx { void start_sim_thread(int api); struct GPGPUsim_ctx* GPGPUsim_ctx_ptr(); -class stream_manager* g_stream_manager(); int gpgpu_opencl_ptx_sim_main_perf( kernel_info_t *grid ); -- cgit v1.3 From 48887cfe0261abde6de23fc5d1d76694426b7e8e Mon Sep 17 00:00:00 2001 From: Mengchi Zhang Date: Thu, 12 Sep 2019 02:01:01 -0400 Subject: Remove GPGPUsim_ctx_ptr() Signed-off-by: Mengchi Zhang --- libcuda/cuda_runtime_api.cc | 116 ++++++++++++++++++++--------- libcuda/gpgpu_context.h | 5 ++ libopencl/opencl_runtime_api.cc | 8 +- src/cuda-sim/cuda-sim.cc | 2 +- src/gpgpu-sim/gpu-sim.cc | 4 +- src/gpgpusim_entrypoint.cc | 161 +++++++++++++++++++--------------------- src/gpgpusim_entrypoint.h | 6 -- 7 files changed, 167 insertions(+), 135 deletions(-) (limited to 'libcuda/cuda_runtime_api.cc') diff --git a/libcuda/cuda_runtime_api.cc b/libcuda/cuda_runtime_api.cc index 9459b64..5cefe60 100644 --- a/libcuda/cuda_runtime_api.cc +++ b/libcuda/cuda_runtime_api.cc @@ -150,9 +150,6 @@ #endif -extern void synchronize(); -extern void exit_simulation(); - /*DEVICE_BUILTIN*/ struct cudaArray { @@ -242,8 +239,8 @@ struct _cuda_device_id *gpgpu_context::GPGPUSim_Init() prop->maxThreadsPerMultiProcessor = the_gpu->threads_per_core(); #endif the_gpu->set_prop(prop); - GPGPUsim_ctx_ptr()->the_cude_device = new _cuda_device_id(the_gpu); - the_device = GPGPUsim_ctx_ptr()->the_cude_device; + the_gpgpusim->the_cude_device = new _cuda_device_id(the_gpu); + the_device = the_gpgpusim->the_cude_device; } start_sim_thread(1); return the_device; @@ -1876,6 +1873,28 @@ __host__ cudaError_t CUDARTAPI cudaStreamDestroyInternal(cudaStream_t stream, gp return g_last_cudaError = cudaSuccess; } +__host__ cudaError_t CUDARTAPI cudaStreamSynchronizeInternal(cudaStream_t stream, gpgpu_context* gpgpu_ctx = NULL) +{ + gpgpu_context *ctx; + if (gpgpu_ctx){ + ctx = gpgpu_ctx; + } else { + ctx = GPGPU_Context(); + } + if(g_debug_execution >= 3){ + announce_call(__my_func__); + } +#if (CUDART_VERSION >= 3000) + if( stream == NULL ) + ctx->synchronize(); + return g_last_cudaError = cudaSuccess; + stream->synchronize(); +#else + printf("GPGPU-Sim PTX: WARNING: Asynchronous kernel execution not supported (%s)\n", __my_func__); +#endif + return g_last_cudaError = cudaSuccess; +} + void __cudaRegisterTextureInternal( void **fatCubinHandle, const struct textureReference *hostVar, @@ -2065,6 +2084,53 @@ __host__ cudaError_t CUDARTAPI cudaStreamWaitEventInternal(cudaStream_t stream, return g_last_cudaError = cudaSuccess; } +__host__ cudaError_t CUDARTAPI cudaThreadExitInternal(gpgpu_context* gpgpu_ctx = NULL) +{ + gpgpu_context *ctx; + if (gpgpu_ctx){ + ctx = gpgpu_ctx; + } else { + ctx = GPGPU_Context(); + } + if(g_debug_execution >= 3){ + announce_call(__my_func__); + } + ctx->exit_simulation(); + return g_last_cudaError = cudaSuccess; +} + +__host__ cudaError_t CUDARTAPI cudaThreadSynchronizeInternal(gpgpu_context* gpgpu_ctx = NULL) +{ + gpgpu_context *ctx; + if (gpgpu_ctx){ + ctx = gpgpu_ctx; + } else { + ctx = GPGPU_Context(); + } + if(g_debug_execution >= 3){ + announce_call(__my_func__); + } + //Called on host side + ctx->synchronize(); + return g_last_cudaError = cudaSuccess; +} + +cudaError_t CUDARTAPI cudaDeviceSynchronizeInternal(gpgpu_context* gpgpu_ctx = NULL) +{ + gpgpu_context *ctx; + if (gpgpu_ctx){ + ctx = gpgpu_ctx; + } else { + ctx = GPGPU_Context(); + } + if(g_debug_execution >= 3){ + announce_call(__my_func__); + } + //Blocks until the device has completed all preceding requested tasks + ctx->synchronize(); + return g_last_cudaError = cudaSuccess; +} + /******************************************************************************* * * * * @@ -2592,18 +2658,7 @@ __host__ cudaError_t CUDARTAPI cudaStreamDestroy(cudaStream_t stream) __host__ cudaError_t CUDARTAPI cudaStreamSynchronize(cudaStream_t stream) { - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } -#if (CUDART_VERSION >= 3000) - if( stream == NULL ) - synchronize(); - return g_last_cudaError = cudaSuccess; - stream->synchronize(); -#else - printf("GPGPU-Sim PTX: WARNING: Asynchronous kernel execution not supported (%s)\n", __my_func__); -#endif - return g_last_cudaError = cudaSuccess; + return cudaStreamSynchronizeInternal(stream); } __host__ cudaError_t CUDARTAPI cudaStreamQuery(cudaStream_t stream) @@ -2722,22 +2777,13 @@ __host__ cudaError_t CUDARTAPI cudaEventElapsedTime(float *ms, cudaEvent_t start __host__ cudaError_t CUDARTAPI cudaThreadExit(void) { - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - exit_simulation(); - return g_last_cudaError = cudaSuccess; + return cudaThreadExitInternal(); } __host__ cudaError_t CUDARTAPI cudaThreadSynchronize(void) { - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - //Called on host side - synchronize(); - return g_last_cudaError = cudaSuccess; -}; + return cudaThreadSynchronizeInternal(); +} int CUDARTAPI __cudaSynchronizeThreads(void**, void*) { @@ -3441,15 +3487,11 @@ cudaError_t cudaDeviceReset ( void ) { } return g_last_cudaError = cudaSuccess; } -cudaError_t CUDARTAPI cudaDeviceSynchronize(void){ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - //Blocks until the device has completed all preceding requested tasks - synchronize(); - return g_last_cudaError = cudaSuccess; -} +cudaError_t CUDARTAPI cudaDeviceSynchronize(void) +{ + return cudaDeviceSynchronizeInternal(); +} void __cudaRegisterShared( void **fatCubinHandle, diff --git a/libcuda/gpgpu_context.h b/libcuda/gpgpu_context.h index 45c5cdd..61d7507 100644 --- a/libcuda/gpgpu_context.h +++ b/libcuda/gpgpu_context.h @@ -52,6 +52,10 @@ class gpgpu_context { cuda_device_runtime* device_runtime; ptx_stats* stats; // member function list + void synchronize(); + void exit_simulation(); + void print_simulation_time(); + int gpgpu_opencl_ptx_sim_main_perf( kernel_info_t *grid ); void cuobjdumpParseBinary(unsigned int handle); class symbol_table *gpgpu_ptx_sim_load_ptx_from_string( const char *p, unsigned source_num ); class symbol_table *gpgpu_ptx_sim_load_ptx_from_filename( const char *filename ); @@ -60,6 +64,7 @@ class gpgpu_context { void print_ptx_file( const char *p, unsigned source_num, const char *filename ); class symbol_table* init_parser(const char*); class gpgpu_sim *gpgpu_ptx_sim_init_perf(); + void start_sim_thread(int api); struct _cuda_device_id *GPGPUSim_Init(); void ptx_reg_options(option_parser_t opp); const ptx_instruction* pc_to_instruction(unsigned pc); diff --git a/libopencl/opencl_runtime_api.cc b/libopencl/opencl_runtime_api.cc index 7f029c7..b032c05 100644 --- a/libopencl/opencl_runtime_api.cc +++ b/libopencl/opencl_runtime_api.cc @@ -647,13 +647,13 @@ unsigned _cl_kernel::sm_context_uid = 0; class _cl_device_id *GPGPUSim_Init() { static _cl_device_id *the_device = NULL; + gpgpu_context *ctx; + ctx = GPGPU_Context(); if( !the_device ) { - gpgpu_context *ctx; - ctx = GPGPU_Context(); gpgpu_sim *the_gpu = ctx->gpgpu_ptx_sim_init_perf(); the_device = new _cl_device_id(the_gpu); } - start_sim_thread(2); + ctx->start_sim_thread(2); return the_device; } @@ -960,7 +960,7 @@ clEnqueueNDRangeKernel(cl_command_queue command_queue, if ( ctx->func_sim->g_ptx_sim_mode ) ctx->func_sim->gpgpu_opencl_ptx_sim_main_func( grid ); else - gpgpu_opencl_ptx_sim_main_perf( grid ); + ctx->gpgpu_opencl_ptx_sim_main_perf( grid ); return CL_SUCCESS; } diff --git a/src/cuda-sim/cuda-sim.cc b/src/cuda-sim/cuda-sim.cc index fc1ea8e..7a130ea 100644 --- a/src/cuda-sim/cuda-sim.cc +++ b/src/cuda-sim/cuda-sim.cc @@ -2203,7 +2203,7 @@ void cuda_sim::gpgpu_cuda_ptx_sim_main_func( kernel_info_t &kernel, bool openCL //g_simulation_starttime is initilized by gpgpu_ptx_sim_init_perf() in gpgpusim_entrypoint.cc upon starting gpgpu-sim time_t end_time, elapsed_time, days, hrs, minutes, sec; end_time = time((time_t *)NULL); - elapsed_time = MAX(end_time - GPGPUsim_ctx_ptr()->g_simulation_starttime, 1); + elapsed_time = MAX(end_time - gpgpu_ctx->the_gpgpusim->g_simulation_starttime, 1); //calculating and printing simulation time in terms of days, hours, minutes and seconds diff --git a/src/gpgpu-sim/gpu-sim.cc b/src/gpgpu-sim/gpu-sim.cc index d236c74..56ea8c4 100644 --- a/src/gpgpu-sim/gpu-sim.cc +++ b/src/gpgpu-sim/gpu-sim.cc @@ -1144,7 +1144,7 @@ void gpgpu_sim::gpu_print_stat() time_t curr_time; time(&curr_time); - unsigned long long elapsed_time = MAX( curr_time - GPGPUsim_ctx_ptr()->g_simulation_starttime, 1 ); + unsigned long long elapsed_time = MAX( curr_time - gpgpu_ctx->the_gpgpusim->g_simulation_starttime, 1 ); printf( "gpu_total_sim_rate=%u\n", (unsigned)( ( gpu_tot_sim_insn + gpu_sim_insn ) / elapsed_time ) ); //shader_print_l1_miss_stat( stdout ); @@ -1717,7 +1717,7 @@ void gpgpu_sim::cycle() time_t days, hrs, minutes, sec; time_t curr_time; time(&curr_time); - unsigned long long elapsed_time = MAX(curr_time - GPGPUsim_ctx_ptr()->g_simulation_starttime, 1); + unsigned long long elapsed_time = MAX(curr_time - gpgpu_ctx->the_gpgpusim->g_simulation_starttime, 1); if ( (elapsed_time - last_liveness_message_time) >= m_config.liveness_message_freq && DTRACE(LIVENESS) ) { days = elapsed_time/(3600*24); hrs = elapsed_time/3600 - 24*days; diff --git a/src/gpgpusim_entrypoint.cc b/src/gpgpusim_entrypoint.cc index 5ce40c6..846773d 100644 --- a/src/gpgpusim_entrypoint.cc +++ b/src/gpgpusim_entrypoint.cc @@ -43,38 +43,28 @@ static int sg_argc = 3; static const char *sg_argv[] = {"", "-config","gpgpusim.config"}; -GPGPUsim_ctx* the_gpgpusim = NULL; - -GPGPUsim_ctx* GPGPUsim_ctx_ptr(){ - if(the_gpgpusim == NULL) - the_gpgpusim = GPGPU_Context()->the_gpgpusim; - - return the_gpgpusim; -} - -static void print_simulation_time(); - -void *gpgpu_sim_thread_sequential(void*) +void * gpgpu_sim_thread_sequential(void * ctx_ptr) { + gpgpu_context * ctx = (gpgpu_context *)ctx_ptr; // at most one kernel running at a time bool done; do { - sem_wait(&(GPGPUsim_ctx_ptr()->g_sim_signal_start)); + sem_wait(&(ctx->the_gpgpusim->g_sim_signal_start)); done = true; - if( GPGPUsim_ctx_ptr()->g_the_gpu->get_more_cta_left() ) { + if( ctx->the_gpgpusim->g_the_gpu->get_more_cta_left() ) { done = false; - GPGPUsim_ctx_ptr()->g_the_gpu->init(); - while( GPGPUsim_ctx_ptr()->g_the_gpu->active() ) { - GPGPUsim_ctx_ptr()->g_the_gpu->cycle(); - GPGPUsim_ctx_ptr()->g_the_gpu->deadlock_check(); + ctx->the_gpgpusim->g_the_gpu->init(); + while( ctx->the_gpgpusim->g_the_gpu->active() ) { + ctx->the_gpgpusim->g_the_gpu->cycle(); + ctx->the_gpgpusim->g_the_gpu->deadlock_check(); } - GPGPUsim_ctx_ptr()->g_the_gpu->print_stats(); - GPGPUsim_ctx_ptr()->g_the_gpu->update_stats(); - print_simulation_time(); + ctx->the_gpgpusim->g_the_gpu->print_stats(); + ctx->the_gpgpusim->g_the_gpu->update_stats(); + ctx->print_simulation_time(); } - sem_post(&(GPGPUsim_ctx_ptr()->g_sim_signal_finish)); + sem_post(&(ctx->the_gpgpusim->g_sim_signal_finish)); } while(!done); - sem_post(&(GPGPUsim_ctx_ptr()->g_sim_signal_exit)); + sem_post(&(ctx->the_gpgpusim->g_sim_signal_exit)); return NULL; } @@ -86,8 +76,9 @@ static void termination_callback() fflush(stdout); } -void *gpgpu_sim_thread_concurrent(void*) +void *gpgpu_sim_thread_concurrent(void * ctx_ptr) { + gpgpu_context * ctx = (gpgpu_context *)ctx_ptr; atexit(termination_callback); // concurrent kernel execution simulation thread do { @@ -95,19 +86,19 @@ void *gpgpu_sim_thread_concurrent(void*) printf("GPGPU-Sim: *** simulation thread starting and spinning waiting for work ***\n"); fflush(stdout); } - while( GPGPUsim_ctx_ptr()->g_stream_manager->empty_protected() && !GPGPUsim_ctx_ptr()->g_sim_done ) + while( ctx->the_gpgpusim->g_stream_manager->empty_protected() && !ctx->the_gpgpusim->g_sim_done ) ; if(g_debug_execution >= 3) { printf("GPGPU-Sim: ** START simulation thread (detected work) **\n"); - GPGPUsim_ctx_ptr()->g_stream_manager->print(stdout); + ctx->the_gpgpusim->g_stream_manager->print(stdout); fflush(stdout); } - pthread_mutex_lock(&(GPGPUsim_ctx_ptr()->g_sim_lock)); - GPGPUsim_ctx_ptr()->g_sim_active = true; - pthread_mutex_unlock(&(GPGPUsim_ctx_ptr()->g_sim_lock)); + pthread_mutex_lock(&(ctx->the_gpgpusim->g_sim_lock)); + ctx->the_gpgpusim->g_sim_active = true; + pthread_mutex_unlock(&(ctx->the_gpgpusim->g_sim_lock)); bool active = false; bool sim_cycles = false; - GPGPUsim_ctx_ptr()->g_the_gpu->init(); + ctx->the_gpgpusim->g_the_gpu->init(); do { // check if a kernel has completed // launch operation on device if one is pending and can be run @@ -119,82 +110,82 @@ void *gpgpu_sim_thread_concurrent(void*) // another kernel, the gpu is not re-initialized and the inter-kernel // behaviour may be incorrect. Check that a kernel has finished and // no other kernel is currently running. - if(GPGPUsim_ctx_ptr()->g_stream_manager->operation(&sim_cycles) && !GPGPUsim_ctx_ptr()->g_the_gpu->active()) + if(ctx->the_gpgpusim->g_stream_manager->operation(&sim_cycles) && !ctx->the_gpgpusim->g_the_gpu->active()) break; //functional simulation - if( GPGPUsim_ctx_ptr()->g_the_gpu->is_functional_sim()) { - kernel_info_t * kernel = GPGPUsim_ctx_ptr()->g_the_gpu->get_functional_kernel(); + if( ctx->the_gpgpusim->g_the_gpu->is_functional_sim()) { + kernel_info_t * kernel = ctx->the_gpgpusim->g_the_gpu->get_functional_kernel(); assert(kernel); - GPGPUsim_ctx_ptr()->gpgpu_ctx->func_sim->gpgpu_cuda_ptx_sim_main_func(*kernel); - GPGPUsim_ctx_ptr()->g_the_gpu->finish_functional_sim(kernel); + ctx->the_gpgpusim->gpgpu_ctx->func_sim->gpgpu_cuda_ptx_sim_main_func(*kernel); + ctx->the_gpgpusim->g_the_gpu->finish_functional_sim(kernel); } //performance simulation - if( GPGPUsim_ctx_ptr()->g_the_gpu->active() ) { - GPGPUsim_ctx_ptr()->g_the_gpu->cycle(); + if( ctx->the_gpgpusim->g_the_gpu->active() ) { + ctx->the_gpgpusim->g_the_gpu->cycle(); sim_cycles = true; - GPGPUsim_ctx_ptr()->g_the_gpu->deadlock_check(); + ctx->the_gpgpusim->g_the_gpu->deadlock_check(); }else { - if(GPGPUsim_ctx_ptr()->g_the_gpu->cycle_insn_cta_max_hit()){ - GPGPUsim_ctx_ptr()->g_stream_manager->stop_all_running_kernels(); - GPGPUsim_ctx_ptr()->g_sim_done = true; - GPGPUsim_ctx_ptr()->break_limit = true; + if(ctx->the_gpgpusim->g_the_gpu->cycle_insn_cta_max_hit()){ + ctx->the_gpgpusim->g_stream_manager->stop_all_running_kernels(); + ctx->the_gpgpusim->g_sim_done = true; + ctx->the_gpgpusim->break_limit = true; } } - active=GPGPUsim_ctx_ptr()->g_the_gpu->active() || !(GPGPUsim_ctx_ptr()->g_stream_manager->empty_protected()); + active=ctx->the_gpgpusim->g_the_gpu->active() || !(ctx->the_gpgpusim->g_stream_manager->empty_protected()); - } while( active && !GPGPUsim_ctx_ptr()->g_sim_done); + } while( active && !ctx->the_gpgpusim->g_sim_done); if(g_debug_execution >= 3) { printf("GPGPU-Sim: ** STOP simulation thread (no work) **\n"); fflush(stdout); } if(sim_cycles) { - GPGPUsim_ctx_ptr()->g_the_gpu->print_stats(); - GPGPUsim_ctx_ptr()->g_the_gpu->update_stats(); - print_simulation_time(); + ctx->the_gpgpusim->g_the_gpu->print_stats(); + ctx->the_gpgpusim->g_the_gpu->update_stats(); + ctx->print_simulation_time(); } - pthread_mutex_lock(&(GPGPUsim_ctx_ptr()->g_sim_lock)); - GPGPUsim_ctx_ptr()->g_sim_active = false; - pthread_mutex_unlock(&(GPGPUsim_ctx_ptr()->g_sim_lock)); - } while( !GPGPUsim_ctx_ptr()->g_sim_done ); + pthread_mutex_lock(&(ctx->the_gpgpusim->g_sim_lock)); + ctx->the_gpgpusim->g_sim_active = false; + pthread_mutex_unlock(&(ctx->the_gpgpusim->g_sim_lock)); + } while( !ctx->the_gpgpusim->g_sim_done ); printf("GPGPU-Sim: *** simulation thread exiting ***\n"); fflush(stdout); - if(GPGPUsim_ctx_ptr()->break_limit) { + if(ctx->the_gpgpusim->break_limit) { printf("GPGPU-Sim: ** break due to reaching the maximum cycles (or instructions) **\n"); exit(1); } - sem_post(&(GPGPUsim_ctx_ptr()->g_sim_signal_exit)); + sem_post(&(ctx->the_gpgpusim->g_sim_signal_exit)); return NULL; } -void synchronize() +void gpgpu_context::synchronize() { printf("GPGPU-Sim: synchronize waiting for inactive GPU simulation\n"); - GPGPUsim_ctx_ptr()->g_stream_manager->print(stdout); + the_gpgpusim->g_stream_manager->print(stdout); fflush(stdout); // sem_wait(&g_sim_signal_finish); bool done = false; do { - pthread_mutex_lock(&(GPGPUsim_ctx_ptr()->g_sim_lock)); - done = ( GPGPUsim_ctx_ptr()->g_stream_manager->empty() && !GPGPUsim_ctx_ptr()->g_sim_active ) || GPGPUsim_ctx_ptr()->g_sim_done; - pthread_mutex_unlock(&(GPGPUsim_ctx_ptr()->g_sim_lock)); + pthread_mutex_lock(&(the_gpgpusim->g_sim_lock)); + done = ( the_gpgpusim->g_stream_manager->empty() && !the_gpgpusim->g_sim_active ) || the_gpgpusim->g_sim_done; + pthread_mutex_unlock(&(the_gpgpusim->g_sim_lock)); } while (!done); printf("GPGPU-Sim: detected inactive GPU simulation thread\n"); fflush(stdout); // sem_post(&g_sim_signal_start); } -void exit_simulation() +void gpgpu_context::exit_simulation() { - GPGPUsim_ctx_ptr()->g_sim_done=true; + the_gpgpusim->g_sim_done=true; printf("GPGPU-Sim: exit_simulation called\n"); fflush(stdout); - sem_wait(&(GPGPUsim_ctx_ptr()->g_sim_signal_exit)); + sem_wait(&(the_gpgpusim->g_sim_signal_exit)); printf("GPGPU-Sim: simulation thread signaled exit\n"); fflush(stdout); } @@ -211,8 +202,8 @@ gpgpu_sim *gpgpu_context::gpgpu_ptx_sim_init_perf() func_sim->ptx_opcocde_latency_options(opp); icnt_reg_options(opp); - GPGPUsim_ctx_ptr()->g_the_gpu_config = new gpgpu_sim_config(this); - GPGPUsim_ctx_ptr()->g_the_gpu_config->reg_options(opp); // register GPU microrachitecture options + the_gpgpusim->g_the_gpu_config = new gpgpu_sim_config(this); + the_gpgpusim->g_the_gpu_config->reg_options(opp); // register GPU microrachitecture options option_parser_cmdline(opp, sg_argc, sg_argv); // parse configuration options fprintf(stdout, "GPGPU-Sim: Configuration options:\n\n"); @@ -220,37 +211,37 @@ gpgpu_sim *gpgpu_context::gpgpu_ptx_sim_init_perf() // Set the Numeric locale to a standard locale where a decimal point is a "dot" not a "comma" // so it does the parsing correctly independent of the system environment variables assert(setlocale(LC_NUMERIC,"C")); - GPGPUsim_ctx_ptr()->g_the_gpu_config->init(); + the_gpgpusim->g_the_gpu_config->init(); - GPGPUsim_ctx_ptr()->g_the_gpu = new gpgpu_sim(*(GPGPUsim_ctx_ptr()->g_the_gpu_config), this); - GPGPUsim_ctx_ptr()->g_stream_manager = new stream_manager((GPGPUsim_ctx_ptr()->g_the_gpu), func_sim->g_cuda_launch_blocking); + the_gpgpusim->g_the_gpu = new gpgpu_sim(*(the_gpgpusim->g_the_gpu_config), this); + the_gpgpusim->g_stream_manager = new stream_manager((the_gpgpusim->g_the_gpu), func_sim->g_cuda_launch_blocking); - GPGPUsim_ctx_ptr()->g_simulation_starttime = time((time_t *)NULL); + the_gpgpusim->g_simulation_starttime = time((time_t *)NULL); - sem_init(&(GPGPUsim_ctx_ptr()->g_sim_signal_start),0,0); - sem_init(&(GPGPUsim_ctx_ptr()->g_sim_signal_finish),0,0); - sem_init(&(GPGPUsim_ctx_ptr()->g_sim_signal_exit),0,0); + sem_init(&(the_gpgpusim->g_sim_signal_start),0,0); + sem_init(&(the_gpgpusim->g_sim_signal_finish),0,0); + sem_init(&(the_gpgpusim->g_sim_signal_exit),0,0); - return GPGPUsim_ctx_ptr()->g_the_gpu; + return the_gpgpusim->g_the_gpu; } -void start_sim_thread(int api) +void gpgpu_context::start_sim_thread(int api) { - if( GPGPUsim_ctx_ptr()->g_sim_done ) { - GPGPUsim_ctx_ptr()->g_sim_done = false; + if( the_gpgpusim->g_sim_done ) { + the_gpgpusim->g_sim_done = false; if( api == 1 ) { - pthread_create(&(GPGPUsim_ctx_ptr()->g_simulation_thread),NULL,gpgpu_sim_thread_concurrent,NULL); + pthread_create(&(the_gpgpusim->g_simulation_thread),NULL,gpgpu_sim_thread_concurrent,(void *)this); } else { - pthread_create(&(GPGPUsim_ctx_ptr()->g_simulation_thread),NULL,gpgpu_sim_thread_sequential,NULL); + pthread_create(&(the_gpgpusim->g_simulation_thread),NULL,gpgpu_sim_thread_sequential,(void *)this); } } } -void print_simulation_time() +void gpgpu_context::print_simulation_time() { time_t current_time, difference, d, h, m, s; current_time = time((time_t *)NULL); - difference = MAX(current_time - GPGPUsim_ctx_ptr()->g_simulation_starttime, 1); + difference = MAX(current_time - the_gpgpusim->g_simulation_starttime, 1); d = difference/(3600*24); h = difference/3600 - 24*d; @@ -260,18 +251,18 @@ void print_simulation_time() fflush(stderr); printf("\n\ngpgpu_simulation_time = %u days, %u hrs, %u min, %u sec (%u sec)\n", (unsigned)d, (unsigned)h, (unsigned)m, (unsigned)s, (unsigned)difference ); - printf("gpgpu_simulation_rate = %u (inst/sec)\n", (unsigned)(GPGPUsim_ctx_ptr()->g_the_gpu->gpu_tot_sim_insn / difference) ); - const unsigned cycles_per_sec = (unsigned)(GPGPUsim_ctx_ptr()->g_the_gpu->gpu_tot_sim_cycle / difference); + printf("gpgpu_simulation_rate = %u (inst/sec)\n", (unsigned)(the_gpgpusim->g_the_gpu->gpu_tot_sim_insn / difference) ); + const unsigned cycles_per_sec = (unsigned)(the_gpgpusim->g_the_gpu->gpu_tot_sim_cycle / difference); printf("gpgpu_simulation_rate = %u (cycle/sec)\n", cycles_per_sec ); - printf("gpgpu_silicon_slowdown = %ux\n", GPGPUsim_ctx_ptr()->g_the_gpu->shader_clock() * 1000 / cycles_per_sec); + printf("gpgpu_silicon_slowdown = %ux\n", the_gpgpusim->g_the_gpu->shader_clock() * 1000 / cycles_per_sec); fflush(stdout); } -int gpgpu_opencl_ptx_sim_main_perf( kernel_info_t *grid ) +int gpgpu_context::gpgpu_opencl_ptx_sim_main_perf( kernel_info_t *grid ) { - GPGPUsim_ctx_ptr()->g_the_gpu->launch(grid); - sem_post(&(GPGPUsim_ctx_ptr()->g_sim_signal_start)); - sem_wait(&(GPGPUsim_ctx_ptr()->g_sim_signal_finish)); + the_gpgpusim->g_the_gpu->launch(grid); + sem_post(&(the_gpgpusim->g_sim_signal_start)); + sem_wait(&(the_gpgpusim->g_sim_signal_finish)); return 0; } diff --git a/src/gpgpusim_entrypoint.h b/src/gpgpusim_entrypoint.h index b2e22e6..9f408df 100644 --- a/src/gpgpusim_entrypoint.h +++ b/src/gpgpusim_entrypoint.h @@ -76,10 +76,4 @@ class GPGPUsim_ctx { }; -void start_sim_thread(int api); - -struct GPGPUsim_ctx* GPGPUsim_ctx_ptr(); - -int gpgpu_opencl_ptx_sim_main_perf( kernel_info_t *grid ); - #endif -- cgit v1.3 From cba77a151abfce2ccd556da69c42e5eb5d74b399 Mon Sep 17 00:00:00 2001 From: Nick Date: Fri, 13 Sep 2019 05:24:00 -0400 Subject: stop formatting where there are known problems ahead of time --- libcuda/cuda_runtime_api.cc | 3 ++- src/gpuwattch/memoryctrl.cc | 3 ++- 2 files changed, 4 insertions(+), 2 deletions(-) (limited to 'libcuda/cuda_runtime_api.cc') diff --git a/libcuda/cuda_runtime_api.cc b/libcuda/cuda_runtime_api.cc index 5cefe60..273194e 100644 --- a/libcuda/cuda_runtime_api.cc +++ b/libcuda/cuda_runtime_api.cc @@ -122,12 +122,13 @@ #endif #define __CUDA_RUNTIME_API_H__ - +// clang-format off #include "host_defines.h" #include "builtin_types.h" #include "driver_types.h" #include "cuda_api.h" #include "cudaProfiler.h" +// clang-format on #if (CUDART_VERSION < 8000) #include "__cudaFatFormat.h" #endif diff --git a/src/gpuwattch/memoryctrl.cc b/src/gpuwattch/memoryctrl.cc index d145ce1..f13c695 100644 --- a/src/gpuwattch/memoryctrl.cc +++ b/src/gpuwattch/memoryctrl.cc @@ -35,6 +35,7 @@ * Tayler Hetherington, University of British Columbia * * Ahmed ElTantawy, University of British Columbia * ********************************************************************/ +// clang-format off #include "io.h" #include "parameter.h" #include "const.h" @@ -48,7 +49,7 @@ #include #include "memoryctrl.h" #include "basic_components.h" - +// clang-format on /* overview of MC models: * McPAT memory controllers are modeled according to large number of industrial data points. * The Basic memory controller architecture is base on the Synopsis designs -- cgit v1.3 From e70fdaa33f694be7241833e8d5a161b432972dcb Mon Sep 17 00:00:00 2001 From: Nick Date: Fri, 13 Sep 2019 10:48:44 -0400 Subject: Add missing changes to the libcuda dir --- libcuda/cuda_api.h | 7589 +++++++++++++++------------ libcuda/cuda_api_object.h | 354 +- libcuda/cuda_runtime_api.cc | 11905 +++++++++++++++++++++--------------------- libcuda/cuobjdump.h | 131 +- libcuda/gpgpu_context.h | 137 +- 5 files changed, 10600 insertions(+), 9516 deletions(-) (limited to 'libcuda/cuda_runtime_api.cc') diff --git a/libcuda/cuda_api.h b/libcuda/cuda_api.h index 27983b4..5a970ba 100644 --- a/libcuda/cuda_api.h +++ b/libcuda/cuda_api.h @@ -63,7 +63,8 @@ typedef uint64_t cuuint64_t; /** * CUDA API versioning support */ -#if defined(__CUDA_API_VERSION_INTERNAL) || defined(__DOXYGEN_ONLY__) || defined(CUDA_ENABLE_DEPRECATED) +#if defined(__CUDA_API_VERSION_INTERNAL) || defined(__DOXYGEN_ONLY__) || \ + defined(CUDA_ENABLE_DEPRECATED) #define __CUDA_DEPRECATED #elif defined(_MSC_VER) #define __CUDA_DEPRECATED __declspec(deprecated) @@ -74,143 +75,148 @@ typedef uint64_t cuuint64_t; #endif #if defined(CUDA_FORCE_API_VERSION) - #if (CUDA_FORCE_API_VERSION == 3010) - #define __CUDA_API_VERSION 3010 - #else - #error "Unsupported value of CUDA_FORCE_API_VERSION" - #endif +#if (CUDA_FORCE_API_VERSION == 3010) +#define __CUDA_API_VERSION 3010 #else - #define __CUDA_API_VERSION 10010 +#error "Unsupported value of CUDA_FORCE_API_VERSION" +#endif +#else +#define __CUDA_API_VERSION 10010 #endif /* CUDA_FORCE_API_VERSION */ -#if defined(__CUDA_API_VERSION_INTERNAL) || defined(CUDA_API_PER_THREAD_DEFAULT_STREAM) - #define __CUDA_API_PER_THREAD_DEFAULT_STREAM - #define __CUDA_API_PTDS(api) api ## _ptds - #define __CUDA_API_PTSZ(api) api ## _ptsz +#if defined(__CUDA_API_VERSION_INTERNAL) || \ + defined(CUDA_API_PER_THREAD_DEFAULT_STREAM) +#define __CUDA_API_PER_THREAD_DEFAULT_STREAM +#define __CUDA_API_PTDS(api) api##_ptds +#define __CUDA_API_PTSZ(api) api##_ptsz #else - #define __CUDA_API_PTDS(api) api - #define __CUDA_API_PTSZ(api) api +#define __CUDA_API_PTDS(api) api +#define __CUDA_API_PTSZ(api) api #endif #if defined(__CUDA_API_VERSION_INTERNAL) || __CUDA_API_VERSION >= 3020 - #define cuDeviceTotalMem cuDeviceTotalMem_v2 - #define cuCtxCreate cuCtxCreate_v2 - #define cuModuleGetGlobal cuModuleGetGlobal_v2 - #define cuMemGetInfo cuMemGetInfo_v2 - #define cuMemAlloc cuMemAlloc_v2 - #define cuMemAllocPitch cuMemAllocPitch_v2 - #define cuMemFree cuMemFree_v2 - #define cuMemGetAddressRange cuMemGetAddressRange_v2 - #define cuMemAllocHost cuMemAllocHost_v2 - #define cuMemHostGetDevicePointer cuMemHostGetDevicePointer_v2 - #define cuMemcpyHtoD __CUDA_API_PTDS(cuMemcpyHtoD_v2) - #define cuMemcpyDtoH __CUDA_API_PTDS(cuMemcpyDtoH_v2) - #define cuMemcpyDtoD __CUDA_API_PTDS(cuMemcpyDtoD_v2) - #define cuMemcpyDtoA __CUDA_API_PTDS(cuMemcpyDtoA_v2) - #define cuMemcpyAtoD __CUDA_API_PTDS(cuMemcpyAtoD_v2) - #define cuMemcpyHtoA __CUDA_API_PTDS(cuMemcpyHtoA_v2) - #define cuMemcpyAtoH __CUDA_API_PTDS(cuMemcpyAtoH_v2) - #define cuMemcpyAtoA __CUDA_API_PTDS(cuMemcpyAtoA_v2) - #define cuMemcpyHtoAAsync __CUDA_API_PTSZ(cuMemcpyHtoAAsync_v2) - #define cuMemcpyAtoHAsync __CUDA_API_PTSZ(cuMemcpyAtoHAsync_v2) - #define cuMemcpy2D __CUDA_API_PTDS(cuMemcpy2D_v2) - #define cuMemcpy2DUnaligned __CUDA_API_PTDS(cuMemcpy2DUnaligned_v2) - #define cuMemcpy3D __CUDA_API_PTDS(cuMemcpy3D_v2) - #define cuMemcpyHtoDAsync __CUDA_API_PTSZ(cuMemcpyHtoDAsync_v2) - #define cuMemcpyDtoHAsync __CUDA_API_PTSZ(cuMemcpyDtoHAsync_v2) - #define cuMemcpyDtoDAsync __CUDA_API_PTSZ(cuMemcpyDtoDAsync_v2) - #define cuMemcpy2DAsync __CUDA_API_PTSZ(cuMemcpy2DAsync_v2) - #define cuMemcpy3DAsync __CUDA_API_PTSZ(cuMemcpy3DAsync_v2) - #define cuMemsetD8 __CUDA_API_PTDS(cuMemsetD8_v2) - #define cuMemsetD16 __CUDA_API_PTDS(cuMemsetD16_v2) - #define cuMemsetD32 __CUDA_API_PTDS(cuMemsetD32_v2) - #define cuMemsetD2D8 __CUDA_API_PTDS(cuMemsetD2D8_v2) - #define cuMemsetD2D16 __CUDA_API_PTDS(cuMemsetD2D16_v2) - #define cuMemsetD2D32 __CUDA_API_PTDS(cuMemsetD2D32_v2) - #define cuArrayCreate cuArrayCreate_v2 - #define cuArrayGetDescriptor cuArrayGetDescriptor_v2 - #define cuArray3DCreate cuArray3DCreate_v2 - #define cuArray3DGetDescriptor cuArray3DGetDescriptor_v2 - #define cuTexRefSetAddress cuTexRefSetAddress_v2 - #define cuTexRefGetAddress cuTexRefGetAddress_v2 - #define cuGraphicsResourceGetMappedPointer cuGraphicsResourceGetMappedPointer_v2 +#define cuDeviceTotalMem cuDeviceTotalMem_v2 +#define cuCtxCreate cuCtxCreate_v2 +#define cuModuleGetGlobal cuModuleGetGlobal_v2 +#define cuMemGetInfo cuMemGetInfo_v2 +#define cuMemAlloc cuMemAlloc_v2 +#define cuMemAllocPitch cuMemAllocPitch_v2 +#define cuMemFree cuMemFree_v2 +#define cuMemGetAddressRange cuMemGetAddressRange_v2 +#define cuMemAllocHost cuMemAllocHost_v2 +#define cuMemHostGetDevicePointer cuMemHostGetDevicePointer_v2 +#define cuMemcpyHtoD __CUDA_API_PTDS(cuMemcpyHtoD_v2) +#define cuMemcpyDtoH __CUDA_API_PTDS(cuMemcpyDtoH_v2) +#define cuMemcpyDtoD __CUDA_API_PTDS(cuMemcpyDtoD_v2) +#define cuMemcpyDtoA __CUDA_API_PTDS(cuMemcpyDtoA_v2) +#define cuMemcpyAtoD __CUDA_API_PTDS(cuMemcpyAtoD_v2) +#define cuMemcpyHtoA __CUDA_API_PTDS(cuMemcpyHtoA_v2) +#define cuMemcpyAtoH __CUDA_API_PTDS(cuMemcpyAtoH_v2) +#define cuMemcpyAtoA __CUDA_API_PTDS(cuMemcpyAtoA_v2) +#define cuMemcpyHtoAAsync __CUDA_API_PTSZ(cuMemcpyHtoAAsync_v2) +#define cuMemcpyAtoHAsync __CUDA_API_PTSZ(cuMemcpyAtoHAsync_v2) +#define cuMemcpy2D __CUDA_API_PTDS(cuMemcpy2D_v2) +#define cuMemcpy2DUnaligned __CUDA_API_PTDS(cuMemcpy2DUnaligned_v2) +#define cuMemcpy3D __CUDA_API_PTDS(cuMemcpy3D_v2) +#define cuMemcpyHtoDAsync __CUDA_API_PTSZ(cuMemcpyHtoDAsync_v2) +#define cuMemcpyDtoHAsync __CUDA_API_PTSZ(cuMemcpyDtoHAsync_v2) +#define cuMemcpyDtoDAsync __CUDA_API_PTSZ(cuMemcpyDtoDAsync_v2) +#define cuMemcpy2DAsync __CUDA_API_PTSZ(cuMemcpy2DAsync_v2) +#define cuMemcpy3DAsync __CUDA_API_PTSZ(cuMemcpy3DAsync_v2) +#define cuMemsetD8 __CUDA_API_PTDS(cuMemsetD8_v2) +#define cuMemsetD16 __CUDA_API_PTDS(cuMemsetD16_v2) +#define cuMemsetD32 __CUDA_API_PTDS(cuMemsetD32_v2) +#define cuMemsetD2D8 __CUDA_API_PTDS(cuMemsetD2D8_v2) +#define cuMemsetD2D16 __CUDA_API_PTDS(cuMemsetD2D16_v2) +#define cuMemsetD2D32 __CUDA_API_PTDS(cuMemsetD2D32_v2) +#define cuArrayCreate cuArrayCreate_v2 +#define cuArrayGetDescriptor cuArrayGetDescriptor_v2 +#define cuArray3DCreate cuArray3DCreate_v2 +#define cuArray3DGetDescriptor cuArray3DGetDescriptor_v2 +#define cuTexRefSetAddress cuTexRefSetAddress_v2 +#define cuTexRefGetAddress cuTexRefGetAddress_v2 +#define cuGraphicsResourceGetMappedPointer cuGraphicsResourceGetMappedPointer_v2 #endif /* __CUDA_API_VERSION_INTERNAL || __CUDA_API_VERSION >= 3020 */ #if defined(__CUDA_API_VERSION_INTERNAL) || __CUDA_API_VERSION >= 4000 - #define cuCtxDestroy cuCtxDestroy_v2 - #define cuCtxPopCurrent cuCtxPopCurrent_v2 - #define cuCtxPushCurrent cuCtxPushCurrent_v2 - #define cuStreamDestroy cuStreamDestroy_v2 - #define cuEventDestroy cuEventDestroy_v2 +#define cuCtxDestroy cuCtxDestroy_v2 +#define cuCtxPopCurrent cuCtxPopCurrent_v2 +#define cuCtxPushCurrent cuCtxPushCurrent_v2 +#define cuStreamDestroy cuStreamDestroy_v2 +#define cuEventDestroy cuEventDestroy_v2 #endif /* __CUDA_API_VERSION_INTERNAL || __CUDA_API_VERSION >= 4000 */ #if defined(__CUDA_API_VERSION_INTERNAL) || __CUDA_API_VERSION >= 4010 - #define cuTexRefSetAddress2D cuTexRefSetAddress2D_v3 +#define cuTexRefSetAddress2D cuTexRefSetAddress2D_v3 #endif /* __CUDA_API_VERSION_INTERNAL || __CUDA_API_VERSION >= 4010 */ #if defined(__CUDA_API_VERSION_INTERNAL) || __CUDA_API_VERSION >= 6050 - #define cuLinkCreate cuLinkCreate_v2 - #define cuLinkAddData cuLinkAddData_v2 - #define cuLinkAddFile cuLinkAddFile_v2 +#define cuLinkCreate cuLinkCreate_v2 +#define cuLinkAddData cuLinkAddData_v2 +#define cuLinkAddFile cuLinkAddFile_v2 #endif /* __CUDA_API_VERSION_INTERNAL || __CUDA_API_VERSION >= 6050 */ #if defined(__CUDA_API_VERSION_INTERNAL) || __CUDA_API_VERSION >= 6050 - #define cuMemHostRegister cuMemHostRegister_v2 - #define cuGraphicsResourceSetMapFlags cuGraphicsResourceSetMapFlags_v2 +#define cuMemHostRegister cuMemHostRegister_v2 +#define cuGraphicsResourceSetMapFlags cuGraphicsResourceSetMapFlags_v2 #endif /* __CUDA_API_VERSION_INTERNAL || __CUDA_API_VERSION >= 6050 */ #if defined(__CUDA_API_VERSION_INTERNAL) || __CUDA_API_VERSION >= 10010 - #define cuStreamBeginCapture __CUDA_API_PTSZ(cuStreamBeginCapture_v2) +#define cuStreamBeginCapture __CUDA_API_PTSZ(cuStreamBeginCapture_v2) #elif defined(__CUDA_API_PER_THREAD_DEFAULT_STREAM) - #define cuStreamBeginCapture __CUDA_API_PTSZ(cuStreamBeginCapture) +#define cuStreamBeginCapture __CUDA_API_PTSZ(cuStreamBeginCapture) #endif /* __CUDA_API_VERSION_INTERNAL || __CUDA_API_VERSION >= 10010 */ #if !defined(__CUDA_API_VERSION_INTERNAL) -#if defined(__CUDA_API_VERSION) && __CUDA_API_VERSION >= 3020 && __CUDA_API_VERSION < 4010 - #define cuTexRefSetAddress2D cuTexRefSetAddress2D_v2 -#endif /* __CUDA_API_VERSION && __CUDA_API_VERSION >= 3020 && __CUDA_API_VERSION < 4010 */ +#if defined(__CUDA_API_VERSION) && __CUDA_API_VERSION >= 3020 && \ + __CUDA_API_VERSION < 4010 +#define cuTexRefSetAddress2D cuTexRefSetAddress2D_v2 +#endif /* __CUDA_API_VERSION && __CUDA_API_VERSION >= 3020 && \ + __CUDA_API_VERSION < 4010 */ #endif /* __CUDA_API_VERSION_INTERNAL */ #if defined(__CUDA_API_PER_THREAD_DEFAULT_STREAM) - #define cuMemcpy __CUDA_API_PTDS(cuMemcpy) - #define cuMemcpyAsync __CUDA_API_PTSZ(cuMemcpyAsync) - #define cuMemcpyPeer __CUDA_API_PTDS(cuMemcpyPeer) - #define cuMemcpyPeerAsync __CUDA_API_PTSZ(cuMemcpyPeerAsync) - #define cuMemcpy3DPeer __CUDA_API_PTDS(cuMemcpy3DPeer) - #define cuMemcpy3DPeerAsync __CUDA_API_PTSZ(cuMemcpy3DPeerAsync) - #define cuMemPrefetchAsync __CUDA_API_PTSZ(cuMemPrefetchAsync) - - #define cuMemsetD8Async __CUDA_API_PTSZ(cuMemsetD8Async) - #define cuMemsetD16Async __CUDA_API_PTSZ(cuMemsetD16Async) - #define cuMemsetD32Async __CUDA_API_PTSZ(cuMemsetD32Async) - #define cuMemsetD2D8Async __CUDA_API_PTSZ(cuMemsetD2D8Async) - #define cuMemsetD2D16Async __CUDA_API_PTSZ(cuMemsetD2D16Async) - #define cuMemsetD2D32Async __CUDA_API_PTSZ(cuMemsetD2D32Async) - - #define cuStreamGetPriority __CUDA_API_PTSZ(cuStreamGetPriority) - #define cuStreamGetFlags __CUDA_API_PTSZ(cuStreamGetFlags) - #define cuStreamGetCtx __CUDA_API_PTSZ(cuStreamGetCtx) - #define cuStreamWaitEvent __CUDA_API_PTSZ(cuStreamWaitEvent) - #define cuStreamEndCapture __CUDA_API_PTSZ(cuStreamEndCapture) - #define cuStreamIsCapturing __CUDA_API_PTSZ(cuStreamIsCapturing) - #define cuStreamGetCaptureInfo __CUDA_API_PTSZ(cuStreamGetCaptureInfo) - #define cuStreamAddCallback __CUDA_API_PTSZ(cuStreamAddCallback) - #define cuStreamAttachMemAsync __CUDA_API_PTSZ(cuStreamAttachMemAsync) - #define cuStreamQuery __CUDA_API_PTSZ(cuStreamQuery) - #define cuStreamSynchronize __CUDA_API_PTSZ(cuStreamSynchronize) - #define cuEventRecord __CUDA_API_PTSZ(cuEventRecord) - #define cuLaunchKernel __CUDA_API_PTSZ(cuLaunchKernel) - #define cuLaunchHostFunc __CUDA_API_PTSZ(cuLaunchHostFunc) - #define cuGraphicsMapResources __CUDA_API_PTSZ(cuGraphicsMapResources) - #define cuGraphicsUnmapResources __CUDA_API_PTSZ(cuGraphicsUnmapResources) - - #define cuStreamWriteValue32 __CUDA_API_PTSZ(cuStreamWriteValue32) - #define cuStreamWaitValue32 __CUDA_API_PTSZ(cuStreamWaitValue32) - #define cuStreamWriteValue64 __CUDA_API_PTSZ(cuStreamWriteValue64) - #define cuStreamWaitValue64 __CUDA_API_PTSZ(cuStreamWaitValue64) - #define cuStreamBatchMemOp __CUDA_API_PTSZ(cuStreamBatchMemOp) - - #define cuLaunchCooperativeKernel __CUDA_API_PTSZ(cuLaunchCooperativeKernel) - - #define cuSignalExternalSemaphoresAsync __CUDA_API_PTSZ(cuSignalExternalSemaphoresAsync) - #define cuWaitExternalSemaphoresAsync __CUDA_API_PTSZ(cuWaitExternalSemaphoresAsync) - - #define cuGraphLaunch __CUDA_API_PTSZ(cuGraphLaunch) +#define cuMemcpy __CUDA_API_PTDS(cuMemcpy) +#define cuMemcpyAsync __CUDA_API_PTSZ(cuMemcpyAsync) +#define cuMemcpyPeer __CUDA_API_PTDS(cuMemcpyPeer) +#define cuMemcpyPeerAsync __CUDA_API_PTSZ(cuMemcpyPeerAsync) +#define cuMemcpy3DPeer __CUDA_API_PTDS(cuMemcpy3DPeer) +#define cuMemcpy3DPeerAsync __CUDA_API_PTSZ(cuMemcpy3DPeerAsync) +#define cuMemPrefetchAsync __CUDA_API_PTSZ(cuMemPrefetchAsync) + +#define cuMemsetD8Async __CUDA_API_PTSZ(cuMemsetD8Async) +#define cuMemsetD16Async __CUDA_API_PTSZ(cuMemsetD16Async) +#define cuMemsetD32Async __CUDA_API_PTSZ(cuMemsetD32Async) +#define cuMemsetD2D8Async __CUDA_API_PTSZ(cuMemsetD2D8Async) +#define cuMemsetD2D16Async __CUDA_API_PTSZ(cuMemsetD2D16Async) +#define cuMemsetD2D32Async __CUDA_API_PTSZ(cuMemsetD2D32Async) + +#define cuStreamGetPriority __CUDA_API_PTSZ(cuStreamGetPriority) +#define cuStreamGetFlags __CUDA_API_PTSZ(cuStreamGetFlags) +#define cuStreamGetCtx __CUDA_API_PTSZ(cuStreamGetCtx) +#define cuStreamWaitEvent __CUDA_API_PTSZ(cuStreamWaitEvent) +#define cuStreamEndCapture __CUDA_API_PTSZ(cuStreamEndCapture) +#define cuStreamIsCapturing __CUDA_API_PTSZ(cuStreamIsCapturing) +#define cuStreamGetCaptureInfo __CUDA_API_PTSZ(cuStreamGetCaptureInfo) +#define cuStreamAddCallback __CUDA_API_PTSZ(cuStreamAddCallback) +#define cuStreamAttachMemAsync __CUDA_API_PTSZ(cuStreamAttachMemAsync) +#define cuStreamQuery __CUDA_API_PTSZ(cuStreamQuery) +#define cuStreamSynchronize __CUDA_API_PTSZ(cuStreamSynchronize) +#define cuEventRecord __CUDA_API_PTSZ(cuEventRecord) +#define cuLaunchKernel __CUDA_API_PTSZ(cuLaunchKernel) +#define cuLaunchHostFunc __CUDA_API_PTSZ(cuLaunchHostFunc) +#define cuGraphicsMapResources __CUDA_API_PTSZ(cuGraphicsMapResources) +#define cuGraphicsUnmapResources __CUDA_API_PTSZ(cuGraphicsUnmapResources) + +#define cuStreamWriteValue32 __CUDA_API_PTSZ(cuStreamWriteValue32) +#define cuStreamWaitValue32 __CUDA_API_PTSZ(cuStreamWaitValue32) +#define cuStreamWriteValue64 __CUDA_API_PTSZ(cuStreamWriteValue64) +#define cuStreamWaitValue64 __CUDA_API_PTSZ(cuStreamWaitValue64) +#define cuStreamBatchMemOp __CUDA_API_PTSZ(cuStreamBatchMemOp) + +#define cuLaunchCooperativeKernel __CUDA_API_PTSZ(cuLaunchCooperativeKernel) + +#define cuSignalExternalSemaphoresAsync \ + __CUDA_API_PTSZ(cuSignalExternalSemaphoresAsync) +#define cuWaitExternalSemaphoresAsync \ + __CUDA_API_PTSZ(cuWaitExternalSemaphoresAsync) + +#define cuGraphLaunch __CUDA_API_PTSZ(cuGraphLaunch) #endif /** @@ -242,7 +248,8 @@ extern "C" { /** * CUDA device pointer - * CUdeviceptr is defined as an unsigned integer type whose size matches the size of a pointer on the target platform. + * CUdeviceptr is defined as an unsigned integer type whose size matches the + * size of a pointer on the target platform. */ #if __CUDA_API_VERSION >= 3020 @@ -254,29 +261,34 @@ typedef unsigned int CUdeviceptr; #endif /* __CUDA_API_VERSION >= 3020 */ -typedef int CUdevice; /**< CUDA device */ -typedef struct CUctx_st *CUcontext; /**< CUDA context */ -typedef struct CUmod_st *CUmodule; /**< CUDA module */ -typedef struct CUfunc_st *CUfunction; /**< CUDA function */ -typedef struct CUarray_st *CUarray; /**< CUDA array */ -typedef struct CUmipmappedArray_st *CUmipmappedArray; /**< CUDA mipmapped array */ -typedef struct CUtexref_st *CUtexref; /**< CUDA texture reference */ -typedef struct CUsurfref_st *CUsurfref; /**< CUDA surface reference */ -typedef struct CUevent_st *CUevent; /**< CUDA event */ -typedef struct CUstream_st *CUstream; /**< CUDA stream */ -typedef struct CUgraphicsResource_st *CUgraphicsResource; /**< CUDA graphics interop resource */ -typedef unsigned long long CUtexObject; /**< An opaque value that represents a CUDA texture object */ -typedef unsigned long long CUsurfObject; /**< An opaque value that represents a CUDA surface object */ -typedef struct CUextMemory_st *CUexternalMemory; /**< CUDA external memory */ -typedef struct CUextSemaphore_st *CUexternalSemaphore; /**< CUDA external semaphore */ -typedef struct CUgraph_st *CUgraph; /**< CUDA graph */ -typedef struct CUgraphNode_st *CUgraphNode; /**< CUDA graph node */ -typedef struct CUgraphExec_st *CUgraphExec; /**< CUDA executable graph */ +typedef int CUdevice; /**< CUDA device */ +typedef struct CUctx_st *CUcontext; /**< CUDA context */ +typedef struct CUmod_st *CUmodule; /**< CUDA module */ +typedef struct CUfunc_st *CUfunction; /**< CUDA function */ +typedef struct CUarray_st *CUarray; /**< CUDA array */ +typedef struct CUmipmappedArray_st + *CUmipmappedArray; /**< CUDA mipmapped array */ +typedef struct CUtexref_st *CUtexref; /**< CUDA texture reference */ +typedef struct CUsurfref_st *CUsurfref; /**< CUDA surface reference */ +typedef struct CUevent_st *CUevent; /**< CUDA event */ +typedef struct CUstream_st *CUstream; /**< CUDA stream */ +typedef struct CUgraphicsResource_st + *CUgraphicsResource; /**< CUDA graphics interop resource */ +typedef unsigned long long + CUtexObject; /**< An opaque value that represents a CUDA texture object */ +typedef unsigned long long + CUsurfObject; /**< An opaque value that represents a CUDA surface object */ +typedef struct CUextMemory_st *CUexternalMemory; /**< CUDA external memory */ +typedef struct CUextSemaphore_st + *CUexternalSemaphore; /**< CUDA external semaphore */ +typedef struct CUgraph_st *CUgraph; /**< CUDA graph */ +typedef struct CUgraphNode_st *CUgraphNode; /**< CUDA graph node */ +typedef struct CUgraphExec_st *CUgraphExec; /**< CUDA executable graph */ #ifndef CU_UUID_HAS_BEEN_DEFINED #define CU_UUID_HAS_BEEN_DEFINED -typedef struct CUuuid_st { /**< CUDA definition of UUID */ - char bytes[16]; +typedef struct CUuuid_st { /**< CUDA definition of UUID */ + char bytes[16]; } CUuuid; #endif @@ -291,21 +303,23 @@ typedef struct CUuuid_st { /**< CUDA definition o * CUDA IPC event handle */ typedef struct CUipcEventHandle_st { - char reserved[CU_IPC_HANDLE_SIZE]; + char reserved[CU_IPC_HANDLE_SIZE]; } CUipcEventHandle; /** * CUDA IPC mem handle */ typedef struct CUipcMemHandle_st { - char reserved[CU_IPC_HANDLE_SIZE]; + char reserved[CU_IPC_HANDLE_SIZE]; } CUipcMemHandle; /** * CUDA Ipc Mem Flags */ typedef enum CUipcMem_flags_enum { - CU_IPC_MEM_LAZY_ENABLE_PEER_ACCESS = 0x1 /**< Automatically enable peer access between remote devices as needed */ + CU_IPC_MEM_LAZY_ENABLE_PEER_ACCESS = + 0x1 /**< Automatically enable peer access between remote devices as needed + */ } CUipcMem_flags; #endif @@ -314,34 +328,41 @@ typedef enum CUipcMem_flags_enum { * CUDA Mem Attach Flags */ typedef enum CUmemAttach_flags_enum { - CU_MEM_ATTACH_GLOBAL = 0x1, /**< Memory can be accessed by any stream on any device */ - CU_MEM_ATTACH_HOST = 0x2, /**< Memory cannot be accessed by any stream on any device */ - CU_MEM_ATTACH_SINGLE = 0x4 /**< Memory can only be accessed by a single stream on the associated device */ + CU_MEM_ATTACH_GLOBAL = + 0x1, /**< Memory can be accessed by any stream on any device */ + CU_MEM_ATTACH_HOST = + 0x2, /**< Memory cannot be accessed by any stream on any device */ + CU_MEM_ATTACH_SINGLE = 0x4 /**< Memory can only be accessed by a single stream + on the associated device */ } CUmemAttach_flags; /** * Context creation flags */ typedef enum CUctx_flags_enum { - CU_CTX_SCHED_AUTO = 0x00, /**< Automatic scheduling */ - CU_CTX_SCHED_SPIN = 0x01, /**< Set spin as default scheduling */ - CU_CTX_SCHED_YIELD = 0x02, /**< Set yield as default scheduling */ - CU_CTX_SCHED_BLOCKING_SYNC = 0x04, /**< Set blocking synchronization as default scheduling */ - CU_CTX_BLOCKING_SYNC = 0x04, /**< Set blocking synchronization as default scheduling - * \deprecated This flag was deprecated as of CUDA 4.0 - * and was replaced with ::CU_CTX_SCHED_BLOCKING_SYNC. */ - CU_CTX_SCHED_MASK = 0x07, - CU_CTX_MAP_HOST = 0x08, /**< Support mapped pinned allocations */ - CU_CTX_LMEM_RESIZE_TO_MAX = 0x10, /**< Keep local memory allocation after launch */ - CU_CTX_FLAGS_MASK = 0x1f + CU_CTX_SCHED_AUTO = 0x00, /**< Automatic scheduling */ + CU_CTX_SCHED_SPIN = 0x01, /**< Set spin as default scheduling */ + CU_CTX_SCHED_YIELD = 0x02, /**< Set yield as default scheduling */ + CU_CTX_SCHED_BLOCKING_SYNC = + 0x04, /**< Set blocking synchronization as default scheduling */ + CU_CTX_BLOCKING_SYNC = + 0x04, /**< Set blocking synchronization as default scheduling + * \deprecated This flag was deprecated as of CUDA 4.0 + * and was replaced with ::CU_CTX_SCHED_BLOCKING_SYNC. */ + CU_CTX_SCHED_MASK = 0x07, + CU_CTX_MAP_HOST = 0x08, /**< Support mapped pinned allocations */ + CU_CTX_LMEM_RESIZE_TO_MAX = + 0x10, /**< Keep local memory allocation after launch */ + CU_CTX_FLAGS_MASK = 0x1f } CUctx_flags; /** * Stream creation flags */ typedef enum CUstream_flags_enum { - CU_STREAM_DEFAULT = 0x0, /**< Default stream flag */ - CU_STREAM_NON_BLOCKING = 0x1 /**< Stream does not synchronize with stream 0 (the NULL stream) */ + CU_STREAM_DEFAULT = 0x0, /**< Default stream flag */ + CU_STREAM_NON_BLOCKING = + 0x1 /**< Stream does not synchronize with stream 0 (the NULL stream) */ } CUstream_flags; /** @@ -352,7 +373,7 @@ typedef enum CUstream_flags_enum { * * See details of the \link_sync_behavior */ -#define CU_STREAM_LEGACY ((CUstream)0x1) +#define CU_STREAM_LEGACY ((CUstream)0x1) /** * Per-thread stream handle @@ -368,10 +389,11 @@ typedef enum CUstream_flags_enum { * Event creation flags */ typedef enum CUevent_flags_enum { - CU_EVENT_DEFAULT = 0x0, /**< Default event flag */ - CU_EVENT_BLOCKING_SYNC = 0x1, /**< Event uses blocking synchronization */ - CU_EVENT_DISABLE_TIMING = 0x2, /**< Event will not record timing data */ - CU_EVENT_INTERPROCESS = 0x4 /**< Event is suitable for interprocess use. CU_EVENT_DISABLE_TIMING must be set */ + CU_EVENT_DEFAULT = 0x0, /**< Default event flag */ + CU_EVENT_BLOCKING_SYNC = 0x1, /**< Event uses blocking synchronization */ + CU_EVENT_DISABLE_TIMING = 0x2, /**< Event will not record timing data */ + CU_EVENT_INTERPROCESS = 0x4 /**< Event is suitable for interprocess use. + CU_EVENT_DISABLE_TIMING must be set */ } CUevent_flags; #if __CUDA_API_VERSION >= 8000 @@ -379,80 +401,93 @@ typedef enum CUevent_flags_enum { * Flags for ::cuStreamWaitValue32 and ::cuStreamWaitValue64 */ typedef enum CUstreamWaitValue_flags_enum { - CU_STREAM_WAIT_VALUE_GEQ = 0x0, /**< Wait until (int32_t)(*addr - value) >= 0 (or int64_t for 64 bit - values). Note this is a cyclic comparison which ignores wraparound. - (Default behavior.) */ - CU_STREAM_WAIT_VALUE_EQ = 0x1, /**< Wait until *addr == value. */ - CU_STREAM_WAIT_VALUE_AND = 0x2, /**< Wait until (*addr & value) != 0. */ - CU_STREAM_WAIT_VALUE_NOR = 0x3, /**< Wait until ~(*addr | value) != 0. Support for this operation can be - queried with ::cuDeviceGetAttribute() and - ::CU_DEVICE_ATTRIBUTE_CAN_USE_STREAM_WAIT_VALUE_NOR.*/ - CU_STREAM_WAIT_VALUE_FLUSH = 1<<30 /**< Follow the wait operation with a flush of outstanding remote writes. This - means that, if a remote write operation is guaranteed to have reached the - device before the wait can be satisfied, that write is guaranteed to be - visible to downstream device work. The device is permitted to reorder - remote writes internally. For example, this flag would be required if - two remote writes arrive in a defined order, the wait is satisfied by the - second write, and downstream work needs to observe the first write. - Support for this operation is restricted to selected platforms and can be - queried with ::CU_DEVICE_ATTRIBUTE_CAN_USE_WAIT_VALUE_FLUSH.*/ + CU_STREAM_WAIT_VALUE_GEQ = + 0x0, /**< Wait until (int32_t)(*addr - value) >= 0 (or int64_t for 64 bit + values). Note this is a cyclic comparison which ignores + wraparound. (Default behavior.) */ + CU_STREAM_WAIT_VALUE_EQ = 0x1, /**< Wait until *addr == value. */ + CU_STREAM_WAIT_VALUE_AND = 0x2, /**< Wait until (*addr & value) != 0. */ + CU_STREAM_WAIT_VALUE_NOR = + 0x3, /**< Wait until ~(*addr | value) != 0. Support for this operation can + be queried with ::cuDeviceGetAttribute() and + ::CU_DEVICE_ATTRIBUTE_CAN_USE_STREAM_WAIT_VALUE_NOR.*/ + CU_STREAM_WAIT_VALUE_FLUSH = + 1 << 30 /**< Follow the wait operation with a flush of outstanding remote + writes. This means that, if a remote write operation is + guaranteed to have reached the device before the wait can be + satisfied, that write is guaranteed to be visible to downstream + device work. The device is permitted to reorder remote writes + internally. For example, this flag would be required if two + remote writes arrive in a defined order, the wait is satisfied + by the second write, and downstream work needs to observe the + first write. Support for this operation is restricted to + selected platforms and can be queried with + ::CU_DEVICE_ATTRIBUTE_CAN_USE_WAIT_VALUE_FLUSH.*/ } CUstreamWaitValue_flags; /** * Flags for ::cuStreamWriteValue32 */ typedef enum CUstreamWriteValue_flags_enum { - CU_STREAM_WRITE_VALUE_DEFAULT = 0x0, /**< Default behavior */ - CU_STREAM_WRITE_VALUE_NO_MEMORY_BARRIER = 0x1 /**< Permits the write to be reordered with writes which were issued - before it, as a performance optimization. Normally, - ::cuStreamWriteValue32 will provide a memory fence before the - write, which has similar semantics to - __threadfence_system() but is scoped to the stream - rather than a CUDA thread. */ + CU_STREAM_WRITE_VALUE_DEFAULT = 0x0, /**< Default behavior */ + CU_STREAM_WRITE_VALUE_NO_MEMORY_BARRIER = + 0x1 /**< Permits the write to be reordered with writes which were issued + before it, as a performance optimization. Normally, + ::cuStreamWriteValue32 will provide a memory fence before the + write, which has similar semantics to + __threadfence_system() but is scoped to the stream + rather than a CUDA thread. */ } CUstreamWriteValue_flags; /** * Operations for ::cuStreamBatchMemOp */ typedef enum CUstreamBatchMemOpType_enum { - CU_STREAM_MEM_OP_WAIT_VALUE_32 = 1, /**< Represents a ::cuStreamWaitValue32 operation */ - CU_STREAM_MEM_OP_WRITE_VALUE_32 = 2, /**< Represents a ::cuStreamWriteValue32 operation */ - CU_STREAM_MEM_OP_WAIT_VALUE_64 = 4, /**< Represents a ::cuStreamWaitValue64 operation */ - CU_STREAM_MEM_OP_WRITE_VALUE_64 = 5, /**< Represents a ::cuStreamWriteValue64 operation */ - CU_STREAM_MEM_OP_FLUSH_REMOTE_WRITES = 3 /**< This has the same effect as ::CU_STREAM_WAIT_VALUE_FLUSH, but as a - standalone operation. */ + CU_STREAM_MEM_OP_WAIT_VALUE_32 = + 1, /**< Represents a ::cuStreamWaitValue32 operation */ + CU_STREAM_MEM_OP_WRITE_VALUE_32 = + 2, /**< Represents a ::cuStreamWriteValue32 operation */ + CU_STREAM_MEM_OP_WAIT_VALUE_64 = + 4, /**< Represents a ::cuStreamWaitValue64 operation */ + CU_STREAM_MEM_OP_WRITE_VALUE_64 = + 5, /**< Represents a ::cuStreamWriteValue64 operation */ + CU_STREAM_MEM_OP_FLUSH_REMOTE_WRITES = + 3 /**< This has the same effect as ::CU_STREAM_WAIT_VALUE_FLUSH, but as a + standalone operation. */ } CUstreamBatchMemOpType; /** * Per-operation parameters for ::cuStreamBatchMemOp */ typedef union CUstreamBatchMemOpParams_union { + CUstreamBatchMemOpType operation; + struct CUstreamMemOpWaitValueParams_st { + CUstreamBatchMemOpType operation; + CUdeviceptr address; + union { + cuuint32_t value; + cuuint64_t value64; + }; + unsigned int flags; + CUdeviceptr + alias; /**< For driver internal use. Initial value is unimportant. */ + } waitValue; + struct CUstreamMemOpWriteValueParams_st { CUstreamBatchMemOpType operation; - struct CUstreamMemOpWaitValueParams_st { - CUstreamBatchMemOpType operation; - CUdeviceptr address; - union { - cuuint32_t value; - cuuint64_t value64; - }; - unsigned int flags; - CUdeviceptr alias; /**< For driver internal use. Initial value is unimportant. */ - } waitValue; - struct CUstreamMemOpWriteValueParams_st { - CUstreamBatchMemOpType operation; - CUdeviceptr address; - union { - cuuint32_t value; - cuuint64_t value64; - }; - unsigned int flags; - CUdeviceptr alias; /**< For driver internal use. Initial value is unimportant. */ - } writeValue; - struct CUstreamMemOpFlushRemoteWritesParams_st { - CUstreamBatchMemOpType operation; - unsigned int flags; - } flushRemoteWrites; - cuuint64_t pad[6]; + CUdeviceptr address; + union { + cuuint32_t value; + cuuint64_t value64; + }; + unsigned int flags; + CUdeviceptr + alias; /**< For driver internal use. Initial value is unimportant. */ + } writeValue; + struct CUstreamMemOpFlushRemoteWritesParams_st { + CUstreamBatchMemOpType operation; + unsigned int flags; + } flushRemoteWrites; + cuuint64_t pad[6]; } CUstreamBatchMemOpParams; #endif /* __CUDA_API_VERSION >= 8000 */ @@ -460,584 +495,739 @@ typedef union CUstreamBatchMemOpParams_union { * Occupancy calculator flag */ typedef enum CUoccupancy_flags_enum { - CU_OCCUPANCY_DEFAULT = 0x0, /**< Default behavior */ - CU_OCCUPANCY_DISABLE_CACHING_OVERRIDE = 0x1 /**< Assume global caching is enabled and cannot be automatically turned off */ + CU_OCCUPANCY_DEFAULT = 0x0, /**< Default behavior */ + CU_OCCUPANCY_DISABLE_CACHING_OVERRIDE = + 0x1 /**< Assume global caching is enabled and cannot be automatically + turned off */ } CUoccupancy_flags; /** * Array formats */ typedef enum CUarray_format_enum { - CU_AD_FORMAT_UNSIGNED_INT8 = 0x01, /**< Unsigned 8-bit integers */ - CU_AD_FORMAT_UNSIGNED_INT16 = 0x02, /**< Unsigned 16-bit integers */ - CU_AD_FORMAT_UNSIGNED_INT32 = 0x03, /**< Unsigned 32-bit integers */ - CU_AD_FORMAT_SIGNED_INT8 = 0x08, /**< Signed 8-bit integers */ - CU_AD_FORMAT_SIGNED_INT16 = 0x09, /**< Signed 16-bit integers */ - CU_AD_FORMAT_SIGNED_INT32 = 0x0a, /**< Signed 32-bit integers */ - CU_AD_FORMAT_HALF = 0x10, /**< 16-bit floating point */ - CU_AD_FORMAT_FLOAT = 0x20 /**< 32-bit floating point */ + CU_AD_FORMAT_UNSIGNED_INT8 = 0x01, /**< Unsigned 8-bit integers */ + CU_AD_FORMAT_UNSIGNED_INT16 = 0x02, /**< Unsigned 16-bit integers */ + CU_AD_FORMAT_UNSIGNED_INT32 = 0x03, /**< Unsigned 32-bit integers */ + CU_AD_FORMAT_SIGNED_INT8 = 0x08, /**< Signed 8-bit integers */ + CU_AD_FORMAT_SIGNED_INT16 = 0x09, /**< Signed 16-bit integers */ + CU_AD_FORMAT_SIGNED_INT32 = 0x0a, /**< Signed 32-bit integers */ + CU_AD_FORMAT_HALF = 0x10, /**< 16-bit floating point */ + CU_AD_FORMAT_FLOAT = 0x20 /**< 32-bit floating point */ } CUarray_format; /** * Texture reference addressing modes */ typedef enum CUaddress_mode_enum { - CU_TR_ADDRESS_MODE_WRAP = 0, /**< Wrapping address mode */ - CU_TR_ADDRESS_MODE_CLAMP = 1, /**< Clamp to edge address mode */ - CU_TR_ADDRESS_MODE_MIRROR = 2, /**< Mirror address mode */ - CU_TR_ADDRESS_MODE_BORDER = 3 /**< Border address mode */ + CU_TR_ADDRESS_MODE_WRAP = 0, /**< Wrapping address mode */ + CU_TR_ADDRESS_MODE_CLAMP = 1, /**< Clamp to edge address mode */ + CU_TR_ADDRESS_MODE_MIRROR = 2, /**< Mirror address mode */ + CU_TR_ADDRESS_MODE_BORDER = 3 /**< Border address mode */ } CUaddress_mode; /** * Texture reference filtering modes */ typedef enum CUfilter_mode_enum { - CU_TR_FILTER_MODE_POINT = 0, /**< Point filter mode */ - CU_TR_FILTER_MODE_LINEAR = 1 /**< Linear filter mode */ + CU_TR_FILTER_MODE_POINT = 0, /**< Point filter mode */ + CU_TR_FILTER_MODE_LINEAR = 1 /**< Linear filter mode */ } CUfilter_mode; /** * Device properties */ typedef enum CUdevice_attribute_enum { - CU_DEVICE_ATTRIBUTE_MAX_THREADS_PER_BLOCK = 1, /**< Maximum number of threads per block */ - CU_DEVICE_ATTRIBUTE_MAX_BLOCK_DIM_X = 2, /**< Maximum block dimension X */ - CU_DEVICE_ATTRIBUTE_MAX_BLOCK_DIM_Y = 3, /**< Maximum block dimension Y */ - CU_DEVICE_ATTRIBUTE_MAX_BLOCK_DIM_Z = 4, /**< Maximum block dimension Z */ - CU_DEVICE_ATTRIBUTE_MAX_GRID_DIM_X = 5, /**< Maximum grid dimension X */ - CU_DEVICE_ATTRIBUTE_MAX_GRID_DIM_Y = 6, /**< Maximum grid dimension Y */ - CU_DEVICE_ATTRIBUTE_MAX_GRID_DIM_Z = 7, /**< Maximum grid dimension Z */ - CU_DEVICE_ATTRIBUTE_MAX_SHARED_MEMORY_PER_BLOCK = 8, /**< Maximum shared memory available per block in bytes */ - CU_DEVICE_ATTRIBUTE_SHARED_MEMORY_PER_BLOCK = 8, /**< Deprecated, use CU_DEVICE_ATTRIBUTE_MAX_SHARED_MEMORY_PER_BLOCK */ - CU_DEVICE_ATTRIBUTE_TOTAL_CONSTANT_MEMORY = 9, /**< Memory available on device for __constant__ variables in a CUDA C kernel in bytes */ - CU_DEVICE_ATTRIBUTE_WARP_SIZE = 10, /**< Warp size in threads */ - CU_DEVICE_ATTRIBUTE_MAX_PITCH = 11, /**< Maximum pitch in bytes allowed by memory copies */ - CU_DEVICE_ATTRIBUTE_MAX_REGISTERS_PER_BLOCK = 12, /**< Maximum number of 32-bit registers available per block */ - CU_DEVICE_ATTRIBUTE_REGISTERS_PER_BLOCK = 12, /**< Deprecated, use CU_DEVICE_ATTRIBUTE_MAX_REGISTERS_PER_BLOCK */ - CU_DEVICE_ATTRIBUTE_CLOCK_RATE = 13, /**< Typical clock frequency in kilohertz */ - CU_DEVICE_ATTRIBUTE_TEXTURE_ALIGNMENT = 14, /**< Alignment requirement for textures */ - CU_DEVICE_ATTRIBUTE_GPU_OVERLAP = 15, /**< Device can possibly copy memory and execute a kernel concurrently. Deprecated. Use instead CU_DEVICE_ATTRIBUTE_ASYNC_ENGINE_COUNT. */ - CU_DEVICE_ATTRIBUTE_MULTIPROCESSOR_COUNT = 16, /**< Number of multiprocessors on device */ - CU_DEVICE_ATTRIBUTE_KERNEL_EXEC_TIMEOUT = 17, /**< Specifies whether there is a run time limit on kernels */ - CU_DEVICE_ATTRIBUTE_INTEGRATED = 18, /**< Device is integrated with host memory */ - CU_DEVICE_ATTRIBUTE_CAN_MAP_HOST_MEMORY = 19, /**< Device can map host memory into CUDA address space */ - CU_DEVICE_ATTRIBUTE_COMPUTE_MODE = 20, /**< Compute mode (See ::CUcomputemode for details) */ - CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE1D_WIDTH = 21, /**< Maximum 1D texture width */ - CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE2D_WIDTH = 22, /**< Maximum 2D texture width */ - CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE2D_HEIGHT = 23, /**< Maximum 2D texture height */ - CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE3D_WIDTH = 24, /**< Maximum 3D texture width */ - CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE3D_HEIGHT = 25, /**< Maximum 3D texture height */ - CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE3D_DEPTH = 26, /**< Maximum 3D texture depth */ - CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE2D_LAYERED_WIDTH = 27, /**< Maximum 2D layered texture width */ - CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE2D_LAYERED_HEIGHT = 28, /**< Maximum 2D layered texture height */ - CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE2D_LAYERED_LAYERS = 29, /**< Maximum layers in a 2D layered texture */ - CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE2D_ARRAY_WIDTH = 27, /**< Deprecated, use CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE2D_LAYERED_WIDTH */ - CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE2D_ARRAY_HEIGHT = 28, /**< Deprecated, use CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE2D_LAYERED_HEIGHT */ - CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE2D_ARRAY_NUMSLICES = 29, /**< Deprecated, use CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE2D_LAYERED_LAYERS */ - CU_DEVICE_ATTRIBUTE_SURFACE_ALIGNMENT = 30, /**< Alignment requirement for surfaces */ - CU_DEVICE_ATTRIBUTE_CONCURRENT_KERNELS = 31, /**< Device can possibly execute multiple kernels concurrently */ - CU_DEVICE_ATTRIBUTE_ECC_ENABLED = 32, /**< Device has ECC support enabled */ - CU_DEVICE_ATTRIBUTE_PCI_BUS_ID = 33, /**< PCI bus ID of the device */ - CU_DEVICE_ATTRIBUTE_PCI_DEVICE_ID = 34, /**< PCI device ID of the device */ - CU_DEVICE_ATTRIBUTE_TCC_DRIVER = 35, /**< Device is using TCC driver model */ - CU_DEVICE_ATTRIBUTE_MEMORY_CLOCK_RATE = 36, /**< Peak memory clock frequency in kilohertz */ - CU_DEVICE_ATTRIBUTE_GLOBAL_MEMORY_BUS_WIDTH = 37, /**< Global memory bus width in bits */ - CU_DEVICE_ATTRIBUTE_L2_CACHE_SIZE = 38, /**< Size of L2 cache in bytes */ - CU_DEVICE_ATTRIBUTE_MAX_THREADS_PER_MULTIPROCESSOR = 39, /**< Maximum resident threads per multiprocessor */ - CU_DEVICE_ATTRIBUTE_ASYNC_ENGINE_COUNT = 40, /**< Number of asynchronous engines */ - CU_DEVICE_ATTRIBUTE_UNIFIED_ADDRESSING = 41, /**< Device shares a unified address space with the host */ - CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE1D_LAYERED_WIDTH = 42, /**< Maximum 1D layered texture width */ - CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE1D_LAYERED_LAYERS = 43, /**< Maximum layers in a 1D layered texture */ - CU_DEVICE_ATTRIBUTE_CAN_TEX2D_GATHER = 44, /**< Deprecated, do not use. */ - CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE2D_GATHER_WIDTH = 45, /**< Maximum 2D texture width if CUDA_ARRAY3D_TEXTURE_GATHER is set */ - CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE2D_GATHER_HEIGHT = 46, /**< Maximum 2D texture height if CUDA_ARRAY3D_TEXTURE_GATHER is set */ - CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE3D_WIDTH_ALTERNATE = 47, /**< Alternate maximum 3D texture width */ - CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE3D_HEIGHT_ALTERNATE = 48,/**< Alternate maximum 3D texture height */ - CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE3D_DEPTH_ALTERNATE = 49, /**< Alternate maximum 3D texture depth */ - CU_DEVICE_ATTRIBUTE_PCI_DOMAIN_ID = 50, /**< PCI domain ID of the device */ - CU_DEVICE_ATTRIBUTE_TEXTURE_PITCH_ALIGNMENT = 51, /**< Pitch alignment requirement for textures */ - CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURECUBEMAP_WIDTH = 52, /**< Maximum cubemap texture width/height */ - CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURECUBEMAP_LAYERED_WIDTH = 53, /**< Maximum cubemap layered texture width/height */ - CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURECUBEMAP_LAYERED_LAYERS = 54, /**< Maximum layers in a cubemap layered texture */ - CU_DEVICE_ATTRIBUTE_MAXIMUM_SURFACE1D_WIDTH = 55, /**< Maximum 1D surface width */ - CU_DEVICE_ATTRIBUTE_MAXIMUM_SURFACE2D_WIDTH = 56, /**< Maximum 2D surface width */ - CU_DEVICE_ATTRIBUTE_MAXIMUM_SURFACE2D_HEIGHT = 57, /**< Maximum 2D surface height */ - CU_DEVICE_ATTRIBUTE_MAXIMUM_SURFACE3D_WIDTH = 58, /**< Maximum 3D surface width */ - CU_DEVICE_ATTRIBUTE_MAXIMUM_SURFACE3D_HEIGHT = 59, /**< Maximum 3D surface height */ - CU_DEVICE_ATTRIBUTE_MAXIMUM_SURFACE3D_DEPTH = 60, /**< Maximum 3D surface depth */ - CU_DEVICE_ATTRIBUTE_MAXIMUM_SURFACE1D_LAYERED_WIDTH = 61, /**< Maximum 1D layered surface width */ - CU_DEVICE_ATTRIBUTE_MAXIMUM_SURFACE1D_LAYERED_LAYERS = 62, /**< Maximum layers in a 1D layered surface */ - CU_DEVICE_ATTRIBUTE_MAXIMUM_SURFACE2D_LAYERED_WIDTH = 63, /**< Maximum 2D layered surface width */ - CU_DEVICE_ATTRIBUTE_MAXIMUM_SURFACE2D_LAYERED_HEIGHT = 64, /**< Maximum 2D layered surface height */ - CU_DEVICE_ATTRIBUTE_MAXIMUM_SURFACE2D_LAYERED_LAYERS = 65, /**< Maximum layers in a 2D layered surface */ - CU_DEVICE_ATTRIBUTE_MAXIMUM_SURFACECUBEMAP_WIDTH = 66, /**< Maximum cubemap surface width */ - CU_DEVICE_ATTRIBUTE_MAXIMUM_SURFACECUBEMAP_LAYERED_WIDTH = 67, /**< Maximum cubemap layered surface width */ - CU_DEVICE_ATTRIBUTE_MAXIMUM_SURFACECUBEMAP_LAYERED_LAYERS = 68, /**< Maximum layers in a cubemap layered surface */ - CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE1D_LINEAR_WIDTH = 69, /**< Maximum 1D linear texture width */ - CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE2D_LINEAR_WIDTH = 70, /**< Maximum 2D linear texture width */ - CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE2D_LINEAR_HEIGHT = 71, /**< Maximum 2D linear texture height */ - CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE2D_LINEAR_PITCH = 72, /**< Maximum 2D linear texture pitch in bytes */ - CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE2D_MIPMAPPED_WIDTH = 73, /**< Maximum mipmapped 2D texture width */ - CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE2D_MIPMAPPED_HEIGHT = 74,/**< Maximum mipmapped 2D texture height */ - CU_DEVICE_ATTRIBUTE_COMPUTE_CAPABILITY_MAJOR = 75, /**< Major compute capability version number */ - CU_DEVICE_ATTRIBUTE_COMPUTE_CAPABILITY_MINOR = 76, /**< Minor compute capability version number */ - CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE1D_MIPMAPPED_WIDTH = 77, /**< Maximum mipmapped 1D texture width */ - CU_DEVICE_ATTRIBUTE_STREAM_PRIORITIES_SUPPORTED = 78, /**< Device supports stream priorities */ - CU_DEVICE_ATTRIBUTE_GLOBAL_L1_CACHE_SUPPORTED = 79, /**< Device supports caching globals in L1 */ - CU_DEVICE_ATTRIBUTE_LOCAL_L1_CACHE_SUPPORTED = 80, /**< Device supports caching locals in L1 */ - CU_DEVICE_ATTRIBUTE_MAX_SHARED_MEMORY_PER_MULTIPROCESSOR = 81, /**< Maximum shared memory available per multiprocessor in bytes */ - CU_DEVICE_ATTRIBUTE_MAX_REGISTERS_PER_MULTIPROCESSOR = 82, /**< Maximum number of 32-bit registers available per multiprocessor */ - CU_DEVICE_ATTRIBUTE_MANAGED_MEMORY = 83, /**< Device can allocate managed memory on this system */ - CU_DEVICE_ATTRIBUTE_MULTI_GPU_BOARD = 84, /**< Device is on a multi-GPU board */ - CU_DEVICE_ATTRIBUTE_MULTI_GPU_BOARD_GROUP_ID = 85, /**< Unique id for a group of devices on the same multi-GPU board */ - CU_DEVICE_ATTRIBUTE_HOST_NATIVE_ATOMIC_SUPPORTED = 86, /**< Link between the device and the host supports native atomic operations (this is a placeholder attribute, and is not supported on any current hardware)*/ - CU_DEVICE_ATTRIBUTE_SINGLE_TO_DOUBLE_PRECISION_PERF_RATIO = 87, /**< Ratio of single precision performance (in floating-point operations per second) to double precision performance */ - CU_DEVICE_ATTRIBUTE_PAGEABLE_MEMORY_ACCESS = 88, /**< Device supports coherently accessing pageable memory without calling cudaHostRegister on it */ - CU_DEVICE_ATTRIBUTE_CONCURRENT_MANAGED_ACCESS = 89, /**< Device can coherently access managed memory concurrently with the CPU */ - CU_DEVICE_ATTRIBUTE_COMPUTE_PREEMPTION_SUPPORTED = 90, /**< Device supports compute preemption. */ - CU_DEVICE_ATTRIBUTE_CAN_USE_HOST_POINTER_FOR_REGISTERED_MEM = 91, /**< Device can access host registered memory at the same virtual address as the CPU */ - CU_DEVICE_ATTRIBUTE_CAN_USE_STREAM_MEM_OPS = 92, /**< ::cuStreamBatchMemOp and related APIs are supported. */ - CU_DEVICE_ATTRIBUTE_CAN_USE_64_BIT_STREAM_MEM_OPS = 93, /**< 64-bit operations are supported in ::cuStreamBatchMemOp and related APIs. */ - CU_DEVICE_ATTRIBUTE_CAN_USE_STREAM_WAIT_VALUE_NOR = 94, /**< ::CU_STREAM_WAIT_VALUE_NOR is supported. */ - CU_DEVICE_ATTRIBUTE_COOPERATIVE_LAUNCH = 95, /**< Device supports launching cooperative kernels via ::cuLaunchCooperativeKernel */ - CU_DEVICE_ATTRIBUTE_COOPERATIVE_MULTI_DEVICE_LAUNCH = 96, /**< Device can participate in cooperative kernels launched via ::cuLaunchCooperativeKernelMultiDevice */ - CU_DEVICE_ATTRIBUTE_MAX_SHARED_MEMORY_PER_BLOCK_OPTIN = 97, /**< Maximum optin shared memory per block */ - CU_DEVICE_ATTRIBUTE_CAN_FLUSH_REMOTE_WRITES = 98, /**< Both the ::CU_STREAM_WAIT_VALUE_FLUSH flag and the ::CU_STREAM_MEM_OP_FLUSH_REMOTE_WRITES MemOp are supported on the device. See \ref CUDA_MEMOP for additional details. */ - CU_DEVICE_ATTRIBUTE_HOST_REGISTER_SUPPORTED = 99, /**< Device supports host memory registration via ::cudaHostRegister. */ - CU_DEVICE_ATTRIBUTE_PAGEABLE_MEMORY_ACCESS_USES_HOST_PAGE_TABLES = 100, /**< Device accesses pageable memory via the host's page tables. */ - CU_DEVICE_ATTRIBUTE_DIRECT_MANAGED_MEM_ACCESS_FROM_HOST = 101, /**< The host can directly access managed memory on the device without migration. */ - CU_DEVICE_ATTRIBUTE_MAX + CU_DEVICE_ATTRIBUTE_MAX_THREADS_PER_BLOCK = + 1, /**< Maximum number of threads per block */ + CU_DEVICE_ATTRIBUTE_MAX_BLOCK_DIM_X = 2, /**< Maximum block dimension X */ + CU_DEVICE_ATTRIBUTE_MAX_BLOCK_DIM_Y = 3, /**< Maximum block dimension Y */ + CU_DEVICE_ATTRIBUTE_MAX_BLOCK_DIM_Z = 4, /**< Maximum block dimension Z */ + CU_DEVICE_ATTRIBUTE_MAX_GRID_DIM_X = 5, /**< Maximum grid dimension X */ + CU_DEVICE_ATTRIBUTE_MAX_GRID_DIM_Y = 6, /**< Maximum grid dimension Y */ + CU_DEVICE_ATTRIBUTE_MAX_GRID_DIM_Z = 7, /**< Maximum grid dimension Z */ + CU_DEVICE_ATTRIBUTE_MAX_SHARED_MEMORY_PER_BLOCK = + 8, /**< Maximum shared memory available per block in bytes */ + CU_DEVICE_ATTRIBUTE_SHARED_MEMORY_PER_BLOCK = + 8, /**< Deprecated, use CU_DEVICE_ATTRIBUTE_MAX_SHARED_MEMORY_PER_BLOCK */ + CU_DEVICE_ATTRIBUTE_TOTAL_CONSTANT_MEMORY = + 9, /**< Memory available on device for __constant__ variables in a CUDA C + kernel in bytes */ + CU_DEVICE_ATTRIBUTE_WARP_SIZE = 10, /**< Warp size in threads */ + CU_DEVICE_ATTRIBUTE_MAX_PITCH = + 11, /**< Maximum pitch in bytes allowed by memory copies */ + CU_DEVICE_ATTRIBUTE_MAX_REGISTERS_PER_BLOCK = + 12, /**< Maximum number of 32-bit registers available per block */ + CU_DEVICE_ATTRIBUTE_REGISTERS_PER_BLOCK = + 12, /**< Deprecated, use CU_DEVICE_ATTRIBUTE_MAX_REGISTERS_PER_BLOCK */ + CU_DEVICE_ATTRIBUTE_CLOCK_RATE = + 13, /**< Typical clock frequency in kilohertz */ + CU_DEVICE_ATTRIBUTE_TEXTURE_ALIGNMENT = + 14, /**< Alignment requirement for textures */ + CU_DEVICE_ATTRIBUTE_GPU_OVERLAP = + 15, /**< Device can possibly copy memory and execute a kernel + concurrently. Deprecated. Use instead + CU_DEVICE_ATTRIBUTE_ASYNC_ENGINE_COUNT. */ + CU_DEVICE_ATTRIBUTE_MULTIPROCESSOR_COUNT = + 16, /**< Number of multiprocessors on device */ + CU_DEVICE_ATTRIBUTE_KERNEL_EXEC_TIMEOUT = + 17, /**< Specifies whether there is a run time limit on kernels */ + CU_DEVICE_ATTRIBUTE_INTEGRATED = + 18, /**< Device is integrated with host memory */ + CU_DEVICE_ATTRIBUTE_CAN_MAP_HOST_MEMORY = + 19, /**< Device can map host memory into CUDA address space */ + CU_DEVICE_ATTRIBUTE_COMPUTE_MODE = + 20, /**< Compute mode (See ::CUcomputemode for details) */ + CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE1D_WIDTH = + 21, /**< Maximum 1D texture width */ + CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE2D_WIDTH = + 22, /**< Maximum 2D texture width */ + CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE2D_HEIGHT = + 23, /**< Maximum 2D texture height */ + CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE3D_WIDTH = + 24, /**< Maximum 3D texture width */ + CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE3D_HEIGHT = + 25, /**< Maximum 3D texture height */ + CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE3D_DEPTH = + 26, /**< Maximum 3D texture depth */ + CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE2D_LAYERED_WIDTH = + 27, /**< Maximum 2D layered texture width */ + CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE2D_LAYERED_HEIGHT = + 28, /**< Maximum 2D layered texture height */ + CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE2D_LAYERED_LAYERS = + 29, /**< Maximum layers in a 2D layered texture */ + CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE2D_ARRAY_WIDTH = + 27, /**< Deprecated, use + CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE2D_LAYERED_WIDTH */ + CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE2D_ARRAY_HEIGHT = + 28, /**< Deprecated, use + CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE2D_LAYERED_HEIGHT */ + CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE2D_ARRAY_NUMSLICES = + 29, /**< Deprecated, use + CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE2D_LAYERED_LAYERS */ + CU_DEVICE_ATTRIBUTE_SURFACE_ALIGNMENT = + 30, /**< Alignment requirement for surfaces */ + CU_DEVICE_ATTRIBUTE_CONCURRENT_KERNELS = + 31, /**< Device can possibly execute multiple kernels concurrently */ + CU_DEVICE_ATTRIBUTE_ECC_ENABLED = 32, /**< Device has ECC support enabled */ + CU_DEVICE_ATTRIBUTE_PCI_BUS_ID = 33, /**< PCI bus ID of the device */ + CU_DEVICE_ATTRIBUTE_PCI_DEVICE_ID = 34, /**< PCI device ID of the device */ + CU_DEVICE_ATTRIBUTE_TCC_DRIVER = 35, /**< Device is using TCC driver model */ + CU_DEVICE_ATTRIBUTE_MEMORY_CLOCK_RATE = + 36, /**< Peak memory clock frequency in kilohertz */ + CU_DEVICE_ATTRIBUTE_GLOBAL_MEMORY_BUS_WIDTH = + 37, /**< Global memory bus width in bits */ + CU_DEVICE_ATTRIBUTE_L2_CACHE_SIZE = 38, /**< Size of L2 cache in bytes */ + CU_DEVICE_ATTRIBUTE_MAX_THREADS_PER_MULTIPROCESSOR = + 39, /**< Maximum resident threads per multiprocessor */ + CU_DEVICE_ATTRIBUTE_ASYNC_ENGINE_COUNT = + 40, /**< Number of asynchronous engines */ + CU_DEVICE_ATTRIBUTE_UNIFIED_ADDRESSING = + 41, /**< Device shares a unified address space with the host */ + CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE1D_LAYERED_WIDTH = + 42, /**< Maximum 1D layered texture width */ + CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE1D_LAYERED_LAYERS = + 43, /**< Maximum layers in a 1D layered texture */ + CU_DEVICE_ATTRIBUTE_CAN_TEX2D_GATHER = 44, /**< Deprecated, do not use. */ + CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE2D_GATHER_WIDTH = + 45, /**< Maximum 2D texture width if CUDA_ARRAY3D_TEXTURE_GATHER is set */ + CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE2D_GATHER_HEIGHT = + 46, /**< Maximum 2D texture height if CUDA_ARRAY3D_TEXTURE_GATHER is set + */ + CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE3D_WIDTH_ALTERNATE = + 47, /**< Alternate maximum 3D texture width */ + CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE3D_HEIGHT_ALTERNATE = + 48, /**< Alternate maximum 3D texture height */ + CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE3D_DEPTH_ALTERNATE = + 49, /**< Alternate maximum 3D texture depth */ + CU_DEVICE_ATTRIBUTE_PCI_DOMAIN_ID = 50, /**< PCI domain ID of the device */ + CU_DEVICE_ATTRIBUTE_TEXTURE_PITCH_ALIGNMENT = + 51, /**< Pitch alignment requirement for textures */ + CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURECUBEMAP_WIDTH = + 52, /**< Maximum cubemap texture width/height */ + CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURECUBEMAP_LAYERED_WIDTH = + 53, /**< Maximum cubemap layered texture width/height */ + CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURECUBEMAP_LAYERED_LAYERS = + 54, /**< Maximum layers in a cubemap layered texture */ + CU_DEVICE_ATTRIBUTE_MAXIMUM_SURFACE1D_WIDTH = + 55, /**< Maximum 1D surface width */ + CU_DEVICE_ATTRIBUTE_MAXIMUM_SURFACE2D_WIDTH = + 56, /**< Maximum 2D surface width */ + CU_DEVICE_ATTRIBUTE_MAXIMUM_SURFACE2D_HEIGHT = + 57, /**< Maximum 2D surface height */ + CU_DEVICE_ATTRIBUTE_MAXIMUM_SURFACE3D_WIDTH = + 58, /**< Maximum 3D surface width */ + CU_DEVICE_ATTRIBUTE_MAXIMUM_SURFACE3D_HEIGHT = + 59, /**< Maximum 3D surface height */ + CU_DEVICE_ATTRIBUTE_MAXIMUM_SURFACE3D_DEPTH = + 60, /**< Maximum 3D surface depth */ + CU_DEVICE_ATTRIBUTE_MAXIMUM_SURFACE1D_LAYERED_WIDTH = + 61, /**< Maximum 1D layered surface width */ + CU_DEVICE_ATTRIBUTE_MAXIMUM_SURFACE1D_LAYERED_LAYERS = + 62, /**< Maximum layers in a 1D layered surface */ + CU_DEVICE_ATTRIBUTE_MAXIMUM_SURFACE2D_LAYERED_WIDTH = + 63, /**< Maximum 2D layered surface width */ + CU_DEVICE_ATTRIBUTE_MAXIMUM_SURFACE2D_LAYERED_HEIGHT = + 64, /**< Maximum 2D layered surface height */ + CU_DEVICE_ATTRIBUTE_MAXIMUM_SURFACE2D_LAYERED_LAYERS = + 65, /**< Maximum layers in a 2D layered surface */ + CU_DEVICE_ATTRIBUTE_MAXIMUM_SURFACECUBEMAP_WIDTH = + 66, /**< Maximum cubemap surface width */ + CU_DEVICE_ATTRIBUTE_MAXIMUM_SURFACECUBEMAP_LAYERED_WIDTH = + 67, /**< Maximum cubemap layered surface width */ + CU_DEVICE_ATTRIBUTE_MAXIMUM_SURFACECUBEMAP_LAYERED_LAYERS = + 68, /**< Maximum layers in a cubemap layered surface */ + CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE1D_LINEAR_WIDTH = + 69, /**< Maximum 1D linear texture width */ + CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE2D_LINEAR_WIDTH = + 70, /**< Maximum 2D linear texture width */ + CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE2D_LINEAR_HEIGHT = + 71, /**< Maximum 2D linear texture height */ + CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE2D_LINEAR_PITCH = + 72, /**< Maximum 2D linear texture pitch in bytes */ + CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE2D_MIPMAPPED_WIDTH = + 73, /**< Maximum mipmapped 2D texture width */ + CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE2D_MIPMAPPED_HEIGHT = + 74, /**< Maximum mipmapped 2D texture height */ + CU_DEVICE_ATTRIBUTE_COMPUTE_CAPABILITY_MAJOR = + 75, /**< Major compute capability version number */ + CU_DEVICE_ATTRIBUTE_COMPUTE_CAPABILITY_MINOR = + 76, /**< Minor compute capability version number */ + CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE1D_MIPMAPPED_WIDTH = + 77, /**< Maximum mipmapped 1D texture width */ + CU_DEVICE_ATTRIBUTE_STREAM_PRIORITIES_SUPPORTED = + 78, /**< Device supports stream priorities */ + CU_DEVICE_ATTRIBUTE_GLOBAL_L1_CACHE_SUPPORTED = + 79, /**< Device supports caching globals in L1 */ + CU_DEVICE_ATTRIBUTE_LOCAL_L1_CACHE_SUPPORTED = + 80, /**< Device supports caching locals in L1 */ + CU_DEVICE_ATTRIBUTE_MAX_SHARED_MEMORY_PER_MULTIPROCESSOR = + 81, /**< Maximum shared memory available per multiprocessor in bytes */ + CU_DEVICE_ATTRIBUTE_MAX_REGISTERS_PER_MULTIPROCESSOR = + 82, /**< Maximum number of 32-bit registers available per multiprocessor + */ + CU_DEVICE_ATTRIBUTE_MANAGED_MEMORY = + 83, /**< Device can allocate managed memory on this system */ + CU_DEVICE_ATTRIBUTE_MULTI_GPU_BOARD = + 84, /**< Device is on a multi-GPU board */ + CU_DEVICE_ATTRIBUTE_MULTI_GPU_BOARD_GROUP_ID = + 85, /**< Unique id for a group of devices on the same multi-GPU board */ + CU_DEVICE_ATTRIBUTE_HOST_NATIVE_ATOMIC_SUPPORTED = + 86, /**< Link between the device and the host supports native atomic + operations (this is a placeholder attribute, and is not supported + on any current hardware)*/ + CU_DEVICE_ATTRIBUTE_SINGLE_TO_DOUBLE_PRECISION_PERF_RATIO = + 87, /**< Ratio of single precision performance (in floating-point + operations per second) to double precision performance */ + CU_DEVICE_ATTRIBUTE_PAGEABLE_MEMORY_ACCESS = + 88, /**< Device supports coherently accessing pageable memory without + calling cudaHostRegister on it */ + CU_DEVICE_ATTRIBUTE_CONCURRENT_MANAGED_ACCESS = + 89, /**< Device can coherently access managed memory concurrently with the + CPU */ + CU_DEVICE_ATTRIBUTE_COMPUTE_PREEMPTION_SUPPORTED = + 90, /**< Device supports compute preemption. */ + CU_DEVICE_ATTRIBUTE_CAN_USE_HOST_POINTER_FOR_REGISTERED_MEM = + 91, /**< Device can access host registered memory at the same virtual + address as the CPU */ + CU_DEVICE_ATTRIBUTE_CAN_USE_STREAM_MEM_OPS = + 92, /**< ::cuStreamBatchMemOp and related APIs are supported. */ + CU_DEVICE_ATTRIBUTE_CAN_USE_64_BIT_STREAM_MEM_OPS = + 93, /**< 64-bit operations are supported in ::cuStreamBatchMemOp and + related APIs. */ + CU_DEVICE_ATTRIBUTE_CAN_USE_STREAM_WAIT_VALUE_NOR = + 94, /**< ::CU_STREAM_WAIT_VALUE_NOR is supported. */ + CU_DEVICE_ATTRIBUTE_COOPERATIVE_LAUNCH = + 95, /**< Device supports launching cooperative kernels via + ::cuLaunchCooperativeKernel */ + CU_DEVICE_ATTRIBUTE_COOPERATIVE_MULTI_DEVICE_LAUNCH = + 96, /**< Device can participate in cooperative kernels launched via + ::cuLaunchCooperativeKernelMultiDevice */ + CU_DEVICE_ATTRIBUTE_MAX_SHARED_MEMORY_PER_BLOCK_OPTIN = + 97, /**< Maximum optin shared memory per block */ + CU_DEVICE_ATTRIBUTE_CAN_FLUSH_REMOTE_WRITES = + 98, /**< Both the ::CU_STREAM_WAIT_VALUE_FLUSH flag and the + ::CU_STREAM_MEM_OP_FLUSH_REMOTE_WRITES MemOp are supported on the + device. See \ref CUDA_MEMOP for additional details. */ + CU_DEVICE_ATTRIBUTE_HOST_REGISTER_SUPPORTED = + 99, /**< Device supports host memory registration via ::cudaHostRegister. + */ + CU_DEVICE_ATTRIBUTE_PAGEABLE_MEMORY_ACCESS_USES_HOST_PAGE_TABLES = + 100, /**< Device accesses pageable memory via the host's page tables. */ + CU_DEVICE_ATTRIBUTE_DIRECT_MANAGED_MEM_ACCESS_FROM_HOST = + 101, /**< The host can directly access managed memory on the device + without migration. */ + CU_DEVICE_ATTRIBUTE_MAX } CUdevice_attribute; /** * Legacy device properties */ typedef struct CUdevprop_st { - int maxThreadsPerBlock; /**< Maximum number of threads per block */ - int maxThreadsDim[3]; /**< Maximum size of each dimension of a block */ - int maxGridSize[3]; /**< Maximum size of each dimension of a grid */ - int sharedMemPerBlock; /**< Shared memory available per block in bytes */ - int totalConstantMemory; /**< Constant memory available on device in bytes */ - int SIMDWidth; /**< Warp size in threads */ - int memPitch; /**< Maximum pitch in bytes allowed by memory copies */ - int regsPerBlock; /**< 32-bit registers available per block */ - int clockRate; /**< Clock frequency in kilohertz */ - int textureAlign; /**< Alignment requirement for textures */ + int maxThreadsPerBlock; /**< Maximum number of threads per block */ + int maxThreadsDim[3]; /**< Maximum size of each dimension of a block */ + int maxGridSize[3]; /**< Maximum size of each dimension of a grid */ + int sharedMemPerBlock; /**< Shared memory available per block in bytes */ + int totalConstantMemory; /**< Constant memory available on device in bytes */ + int SIMDWidth; /**< Warp size in threads */ + int memPitch; /**< Maximum pitch in bytes allowed by memory copies */ + int regsPerBlock; /**< 32-bit registers available per block */ + int clockRate; /**< Clock frequency in kilohertz */ + int textureAlign; /**< Alignment requirement for textures */ } CUdevprop; /** * Pointer information */ typedef enum CUpointer_attribute_enum { - CU_POINTER_ATTRIBUTE_CONTEXT = 1, /**< The ::CUcontext on which a pointer was allocated or registered */ - CU_POINTER_ATTRIBUTE_MEMORY_TYPE = 2, /**< The ::CUmemorytype describing the physical location of a pointer */ - CU_POINTER_ATTRIBUTE_DEVICE_POINTER = 3, /**< The address at which a pointer's memory may be accessed on the device */ - CU_POINTER_ATTRIBUTE_HOST_POINTER = 4, /**< The address at which a pointer's memory may be accessed on the host */ - CU_POINTER_ATTRIBUTE_P2P_TOKENS = 5, /**< A pair of tokens for use with the nv-p2p.h Linux kernel interface */ - CU_POINTER_ATTRIBUTE_SYNC_MEMOPS = 6, /**< Synchronize every synchronous memory operation initiated on this region */ - CU_POINTER_ATTRIBUTE_BUFFER_ID = 7, /**< A process-wide unique ID for an allocated memory region*/ - CU_POINTER_ATTRIBUTE_IS_MANAGED = 8, /**< Indicates if the pointer points to managed memory */ - CU_POINTER_ATTRIBUTE_DEVICE_ORDINAL = 9 /**< A device ordinal of a device on which a pointer was allocated or registered */ + CU_POINTER_ATTRIBUTE_CONTEXT = + 1, /**< The ::CUcontext on which a pointer was allocated or registered */ + CU_POINTER_ATTRIBUTE_MEMORY_TYPE = 2, /**< The ::CUmemorytype describing the + physical location of a pointer */ + CU_POINTER_ATTRIBUTE_DEVICE_POINTER = + 3, /**< The address at which a pointer's memory may be accessed on the + device */ + CU_POINTER_ATTRIBUTE_HOST_POINTER = + 4, /**< The address at which a pointer's memory may be accessed on the + host */ + CU_POINTER_ATTRIBUTE_P2P_TOKENS = 5, /**< A pair of tokens for use with the + nv-p2p.h Linux kernel interface */ + CU_POINTER_ATTRIBUTE_SYNC_MEMOPS = + 6, /**< Synchronize every synchronous memory operation initiated on this + region */ + CU_POINTER_ATTRIBUTE_BUFFER_ID = + 7, /**< A process-wide unique ID for an allocated memory region*/ + CU_POINTER_ATTRIBUTE_IS_MANAGED = + 8, /**< Indicates if the pointer points to managed memory */ + CU_POINTER_ATTRIBUTE_DEVICE_ORDINAL = + 9 /**< A device ordinal of a device on which a pointer was allocated or + registered */ } CUpointer_attribute; /** * Function properties */ typedef enum CUfunction_attribute_enum { - /** - * The maximum number of threads per block, beyond which a launch of the - * function would fail. This number depends on both the function and the - * device on which the function is currently loaded. - */ - CU_FUNC_ATTRIBUTE_MAX_THREADS_PER_BLOCK = 0, - - /** - * The size in bytes of statically-allocated shared memory required by - * this function. This does not include dynamically-allocated shared - * memory requested by the user at runtime. - */ - CU_FUNC_ATTRIBUTE_SHARED_SIZE_BYTES = 1, - - /** - * The size in bytes of user-allocated constant memory required by this - * function. - */ - CU_FUNC_ATTRIBUTE_CONST_SIZE_BYTES = 2, - - /** - * The size in bytes of local memory used by each thread of this function. - */ - CU_FUNC_ATTRIBUTE_LOCAL_SIZE_BYTES = 3, - - /** - * The number of registers used by each thread of this function. - */ - CU_FUNC_ATTRIBUTE_NUM_REGS = 4, - - /** - * The PTX virtual architecture version for which the function was - * compiled. This value is the major PTX version * 10 + the minor PTX - * version, so a PTX version 1.3 function would return the value 13. - * Note that this may return the undefined value of 0 for cubins - * compiled prior to CUDA 3.0. - */ - CU_FUNC_ATTRIBUTE_PTX_VERSION = 5, - - /** - * The binary architecture version for which the function was compiled. - * This value is the major binary version * 10 + the minor binary version, - * so a binary version 1.3 function would return the value 13. Note that - * this will return a value of 10 for legacy cubins that do not have a - * properly-encoded binary architecture version. - */ - CU_FUNC_ATTRIBUTE_BINARY_VERSION = 6, - - /** - * The attribute to indicate whether the function has been compiled with - * user specified option "-Xptxas --dlcm=ca" set . - */ - CU_FUNC_ATTRIBUTE_CACHE_MODE_CA = 7, - - /** - * The maximum size in bytes of dynamically-allocated shared memory that can be used by - * this function. If the user-specified dynamic shared memory size is larger than this - * value, the launch will fail. - * See ::cuFuncSetAttribute - */ - CU_FUNC_ATTRIBUTE_MAX_DYNAMIC_SHARED_SIZE_BYTES = 8, - - /** - * On devices where the L1 cache and shared memory use the same hardware resources, - * this sets the shared memory carveout preference, in percent of the total shared memory. - * Refer to ::CU_DEVICE_ATTRIBUTE_MAX_SHARED_MEMORY_PER_MULTIPROCESSOR. - * This is only a hint, and the driver can choose a different ratio if required to execute the function. - * See ::cuFuncSetAttribute - */ - CU_FUNC_ATTRIBUTE_PREFERRED_SHARED_MEMORY_CARVEOUT = 9, - - CU_FUNC_ATTRIBUTE_MAX + /** + * The maximum number of threads per block, beyond which a launch of the + * function would fail. This number depends on both the function and the + * device on which the function is currently loaded. + */ + CU_FUNC_ATTRIBUTE_MAX_THREADS_PER_BLOCK = 0, + + /** + * The size in bytes of statically-allocated shared memory required by + * this function. This does not include dynamically-allocated shared + * memory requested by the user at runtime. + */ + CU_FUNC_ATTRIBUTE_SHARED_SIZE_BYTES = 1, + + /** + * The size in bytes of user-allocated constant memory required by this + * function. + */ + CU_FUNC_ATTRIBUTE_CONST_SIZE_BYTES = 2, + + /** + * The size in bytes of local memory used by each thread of this function. + */ + CU_FUNC_ATTRIBUTE_LOCAL_SIZE_BYTES = 3, + + /** + * The number of registers used by each thread of this function. + */ + CU_FUNC_ATTRIBUTE_NUM_REGS = 4, + + /** + * The PTX virtual architecture version for which the function was + * compiled. This value is the major PTX version * 10 + the minor PTX + * version, so a PTX version 1.3 function would return the value 13. + * Note that this may return the undefined value of 0 for cubins + * compiled prior to CUDA 3.0. + */ + CU_FUNC_ATTRIBUTE_PTX_VERSION = 5, + + /** + * The binary architecture version for which the function was compiled. + * This value is the major binary version * 10 + the minor binary version, + * so a binary version 1.3 function would return the value 13. Note that + * this will return a value of 10 for legacy cubins that do not have a + * properly-encoded binary architecture version. + */ + CU_FUNC_ATTRIBUTE_BINARY_VERSION = 6, + + /** + * The attribute to indicate whether the function has been compiled with + * user specified option "-Xptxas --dlcm=ca" set . + */ + CU_FUNC_ATTRIBUTE_CACHE_MODE_CA = 7, + + /** + * The maximum size in bytes of dynamically-allocated shared memory that can + * be used by this function. If the user-specified dynamic shared memory size + * is larger than this value, the launch will fail. See ::cuFuncSetAttribute + */ + CU_FUNC_ATTRIBUTE_MAX_DYNAMIC_SHARED_SIZE_BYTES = 8, + + /** + * On devices where the L1 cache and shared memory use the same hardware + * resources, this sets the shared memory carveout preference, in percent of + * the total shared memory. Refer to + * ::CU_DEVICE_ATTRIBUTE_MAX_SHARED_MEMORY_PER_MULTIPROCESSOR. This is only a + * hint, and the driver can choose a different ratio if required to execute + * the function. See ::cuFuncSetAttribute + */ + CU_FUNC_ATTRIBUTE_PREFERRED_SHARED_MEMORY_CARVEOUT = 9, + + CU_FUNC_ATTRIBUTE_MAX } CUfunction_attribute; /** * Function cache configurations */ typedef enum CUfunc_cache_enum { - CU_FUNC_CACHE_PREFER_NONE = 0x00, /**< no preference for shared memory or L1 (default) */ - CU_FUNC_CACHE_PREFER_SHARED = 0x01, /**< prefer larger shared memory and smaller L1 cache */ - CU_FUNC_CACHE_PREFER_L1 = 0x02, /**< prefer larger L1 cache and smaller shared memory */ - CU_FUNC_CACHE_PREFER_EQUAL = 0x03 /**< prefer equal sized L1 cache and shared memory */ + CU_FUNC_CACHE_PREFER_NONE = + 0x00, /**< no preference for shared memory or L1 (default) */ + CU_FUNC_CACHE_PREFER_SHARED = + 0x01, /**< prefer larger shared memory and smaller L1 cache */ + CU_FUNC_CACHE_PREFER_L1 = + 0x02, /**< prefer larger L1 cache and smaller shared memory */ + CU_FUNC_CACHE_PREFER_EQUAL = + 0x03 /**< prefer equal sized L1 cache and shared memory */ } CUfunc_cache; /** * Shared memory configurations */ typedef enum CUsharedconfig_enum { - CU_SHARED_MEM_CONFIG_DEFAULT_BANK_SIZE = 0x00, /**< set default shared memory bank size */ - CU_SHARED_MEM_CONFIG_FOUR_BYTE_BANK_SIZE = 0x01, /**< set shared memory bank width to four bytes */ - CU_SHARED_MEM_CONFIG_EIGHT_BYTE_BANK_SIZE = 0x02 /**< set shared memory bank width to eight bytes */ + CU_SHARED_MEM_CONFIG_DEFAULT_BANK_SIZE = + 0x00, /**< set default shared memory bank size */ + CU_SHARED_MEM_CONFIG_FOUR_BYTE_BANK_SIZE = + 0x01, /**< set shared memory bank width to four bytes */ + CU_SHARED_MEM_CONFIG_EIGHT_BYTE_BANK_SIZE = + 0x02 /**< set shared memory bank width to eight bytes */ } CUsharedconfig; /** - * Shared memory carveout configurations. These may be passed to ::cuFuncSetAttribute + * Shared memory carveout configurations. These may be passed to + * ::cuFuncSetAttribute */ typedef enum CUshared_carveout_enum { - CU_SHAREDMEM_CARVEOUT_DEFAULT = -1, /**< No preference for shared memory or L1 (default) */ - CU_SHAREDMEM_CARVEOUT_MAX_SHARED = 100, /**< Prefer maximum available shared memory, minimum L1 cache */ - CU_SHAREDMEM_CARVEOUT_MAX_L1 = 0 /**< Prefer maximum available L1 cache, minimum shared memory */ + CU_SHAREDMEM_CARVEOUT_DEFAULT = + -1, /**< No preference for shared memory or L1 (default) */ + CU_SHAREDMEM_CARVEOUT_MAX_SHARED = + 100, /**< Prefer maximum available shared memory, minimum L1 cache */ + CU_SHAREDMEM_CARVEOUT_MAX_L1 = + 0 /**< Prefer maximum available L1 cache, minimum shared memory */ } CUshared_carveout; /** * Memory types */ typedef enum CUmemorytype_enum { - CU_MEMORYTYPE_HOST = 0x01, /**< Host memory */ - CU_MEMORYTYPE_DEVICE = 0x02, /**< Device memory */ - CU_MEMORYTYPE_ARRAY = 0x03, /**< Array memory */ - CU_MEMORYTYPE_UNIFIED = 0x04 /**< Unified device or host memory */ + CU_MEMORYTYPE_HOST = 0x01, /**< Host memory */ + CU_MEMORYTYPE_DEVICE = 0x02, /**< Device memory */ + CU_MEMORYTYPE_ARRAY = 0x03, /**< Array memory */ + CU_MEMORYTYPE_UNIFIED = 0x04 /**< Unified device or host memory */ } CUmemorytype; /** * Compute Modes */ typedef enum CUcomputemode_enum { - CU_COMPUTEMODE_DEFAULT = 0, /**< Default compute mode (Multiple contexts allowed per device) */ - CU_COMPUTEMODE_PROHIBITED = 2, /**< Compute-prohibited mode (No contexts can be created on this device at this time) */ - CU_COMPUTEMODE_EXCLUSIVE_PROCESS = 3 /**< Compute-exclusive-process mode (Only one context used by a single process can be present on this device at a time) */ + CU_COMPUTEMODE_DEFAULT = + 0, /**< Default compute mode (Multiple contexts allowed per device) */ + CU_COMPUTEMODE_PROHIBITED = 2, /**< Compute-prohibited mode (No contexts can + be created on this device at this time) */ + CU_COMPUTEMODE_EXCLUSIVE_PROCESS = + 3 /**< Compute-exclusive-process mode (Only one context used by a single + process can be present on this device at a time) */ } CUcomputemode; /** * Memory advise values */ typedef enum CUmem_advise_enum { - CU_MEM_ADVISE_SET_READ_MOSTLY = 1, /**< Data will mostly be read and only occassionally be written to */ - CU_MEM_ADVISE_UNSET_READ_MOSTLY = 2, /**< Undo the effect of ::CU_MEM_ADVISE_SET_READ_MOSTLY */ - CU_MEM_ADVISE_SET_PREFERRED_LOCATION = 3, /**< Set the preferred location for the data as the specified device */ - CU_MEM_ADVISE_UNSET_PREFERRED_LOCATION = 4, /**< Clear the preferred location for the data */ - CU_MEM_ADVISE_SET_ACCESSED_BY = 5, /**< Data will be accessed by the specified device, so prevent page faults as much as possible */ - CU_MEM_ADVISE_UNSET_ACCESSED_BY = 6 /**< Let the Unified Memory subsystem decide on the page faulting policy for the specified device */ + CU_MEM_ADVISE_SET_READ_MOSTLY = + 1, /**< Data will mostly be read and only occassionally be written to */ + CU_MEM_ADVISE_UNSET_READ_MOSTLY = + 2, /**< Undo the effect of ::CU_MEM_ADVISE_SET_READ_MOSTLY */ + CU_MEM_ADVISE_SET_PREFERRED_LOCATION = + 3, /**< Set the preferred location for the data as the specified device */ + CU_MEM_ADVISE_UNSET_PREFERRED_LOCATION = + 4, /**< Clear the preferred location for the data */ + CU_MEM_ADVISE_SET_ACCESSED_BY = + 5, /**< Data will be accessed by the specified device, so prevent page + faults as much as possible */ + CU_MEM_ADVISE_UNSET_ACCESSED_BY = + 6 /**< Let the Unified Memory subsystem decide on the page faulting policy + for the specified device */ } CUmem_advise; typedef enum CUmem_range_attribute_enum { - CU_MEM_RANGE_ATTRIBUTE_READ_MOSTLY = 1, /**< Whether the range will mostly be read and only occassionally be written to */ - CU_MEM_RANGE_ATTRIBUTE_PREFERRED_LOCATION = 2, /**< The preferred location of the range */ - CU_MEM_RANGE_ATTRIBUTE_ACCESSED_BY = 3, /**< Memory range has ::CU_MEM_ADVISE_SET_ACCESSED_BY set for specified device */ - CU_MEM_RANGE_ATTRIBUTE_LAST_PREFETCH_LOCATION = 4 /**< The last location to which the range was prefetched */ + CU_MEM_RANGE_ATTRIBUTE_READ_MOSTLY = + 1, /**< Whether the range will mostly be read and only occassionally be + written to */ + CU_MEM_RANGE_ATTRIBUTE_PREFERRED_LOCATION = + 2, /**< The preferred location of the range */ + CU_MEM_RANGE_ATTRIBUTE_ACCESSED_BY = + 3, /**< Memory range has ::CU_MEM_ADVISE_SET_ACCESSED_BY set for specified + device */ + CU_MEM_RANGE_ATTRIBUTE_LAST_PREFETCH_LOCATION = + 4 /**< The last location to which the range was prefetched */ } CUmem_range_attribute; /** * Online compiler and linker options */ -typedef enum CUjit_option_enum -{ - /** - * Max number of registers that a thread may use.\n - * Option type: unsigned int\n - * Applies to: compiler only - */ - CU_JIT_MAX_REGISTERS = 0, - - /** - * IN: Specifies minimum number of threads per block to target compilation - * for\n - * OUT: Returns the number of threads the compiler actually targeted. - * This restricts the resource utilization fo the compiler (e.g. max - * registers) such that a block with the given number of threads should be - * able to launch based on register limitations. Note, this option does not - * currently take into account any other resource limitations, such as - * shared memory utilization.\n - * Cannot be combined with ::CU_JIT_TARGET.\n - * Option type: unsigned int\n - * Applies to: compiler only - */ - CU_JIT_THREADS_PER_BLOCK, - - /** - * Overwrites the option value with the total wall clock time, in - * milliseconds, spent in the compiler and linker\n - * Option type: float\n - * Applies to: compiler and linker - */ - CU_JIT_WALL_TIME, - - /** - * Pointer to a buffer in which to print any log messages - * that are informational in nature (the buffer size is specified via - * option ::CU_JIT_INFO_LOG_BUFFER_SIZE_BYTES)\n - * Option type: char *\n - * Applies to: compiler and linker - */ - CU_JIT_INFO_LOG_BUFFER, - - /** - * IN: Log buffer size in bytes. Log messages will be capped at this size - * (including null terminator)\n - * OUT: Amount of log buffer filled with messages\n - * Option type: unsigned int\n - * Applies to: compiler and linker - */ - CU_JIT_INFO_LOG_BUFFER_SIZE_BYTES, - - /** - * Pointer to a buffer in which to print any log messages that - * reflect errors (the buffer size is specified via option - * ::CU_JIT_ERROR_LOG_BUFFER_SIZE_BYTES)\n - * Option type: char *\n - * Applies to: compiler and linker - */ - CU_JIT_ERROR_LOG_BUFFER, - - /** - * IN: Log buffer size in bytes. Log messages will be capped at this size - * (including null terminator)\n - * OUT: Amount of log buffer filled with messages\n - * Option type: unsigned int\n - * Applies to: compiler and linker - */ - CU_JIT_ERROR_LOG_BUFFER_SIZE_BYTES, - - /** - * Level of optimizations to apply to generated code (0 - 4), with 4 - * being the default and highest level of optimizations.\n - * Option type: unsigned int\n - * Applies to: compiler only - */ - CU_JIT_OPTIMIZATION_LEVEL, - - /** - * No option value required. Determines the target based on the current - * attached context (default)\n - * Option type: No option value needed\n - * Applies to: compiler and linker - */ - CU_JIT_TARGET_FROM_CUCONTEXT, - - /** - * Target is chosen based on supplied ::CUjit_target. Cannot be - * combined with ::CU_JIT_THREADS_PER_BLOCK.\n - * Option type: unsigned int for enumerated type ::CUjit_target\n - * Applies to: compiler and linker - */ - CU_JIT_TARGET, - - /** - * Specifies choice of fallback strategy if matching cubin is not found. - * Choice is based on supplied ::CUjit_fallback. This option cannot be - * used with cuLink* APIs as the linker requires exact matches.\n - * Option type: unsigned int for enumerated type ::CUjit_fallback\n - * Applies to: compiler only - */ - CU_JIT_FALLBACK_STRATEGY, - - /** - * Specifies whether to create debug information in output (-g) - * (0: false, default)\n - * Option type: int\n - * Applies to: compiler and linker - */ - CU_JIT_GENERATE_DEBUG_INFO, - - /** - * Generate verbose log messages (0: false, default)\n - * Option type: int\n - * Applies to: compiler and linker - */ - CU_JIT_LOG_VERBOSE, - - /** - * Generate line number information (-lineinfo) (0: false, default)\n - * Option type: int\n - * Applies to: compiler only - */ - CU_JIT_GENERATE_LINE_INFO, - - /** - * Specifies whether to enable caching explicitly (-dlcm) \n - * Choice is based on supplied ::CUjit_cacheMode_enum.\n - * Option type: unsigned int for enumerated type ::CUjit_cacheMode_enum\n - * Applies to: compiler only - */ - CU_JIT_CACHE_MODE, - - /** - * The below jit options are used for internal purposes only, in this version of CUDA - */ - CU_JIT_NEW_SM3X_OPT, - CU_JIT_FAST_COMPILE, - - /** - * Array of device symbol names that will be relocated to the corresponing - * host addresses stored in ::CU_JIT_GLOBAL_SYMBOL_ADDRESSES.\n - * Must contain ::CU_JIT_GLOBAL_SYMBOL_COUNT entries.\n - * When loding a device module, driver will relocate all encountered - * unresolved symbols to the host addresses.\n - * It is only allowed to register symbols that correspond to unresolved - * global variables.\n - * It is illegal to register the same device symbol at multiple addresses.\n - * Option type: const char **\n - * Applies to: dynamic linker only - */ - CU_JIT_GLOBAL_SYMBOL_NAMES, - - /** - * Array of host addresses that will be used to relocate corresponding - * device symbols stored in ::CU_JIT_GLOBAL_SYMBOL_NAMES.\n - * Must contain ::CU_JIT_GLOBAL_SYMBOL_COUNT entries.\n - * Option type: void **\n - * Applies to: dynamic linker only - */ - CU_JIT_GLOBAL_SYMBOL_ADDRESSES, - - /** - * Number of entries in ::CU_JIT_GLOBAL_SYMBOL_NAMES and - * ::CU_JIT_GLOBAL_SYMBOL_ADDRESSES arrays.\n - * Option type: unsigned int\n - * Applies to: dynamic linker only - */ - CU_JIT_GLOBAL_SYMBOL_COUNT, - - CU_JIT_NUM_OPTIONS +typedef enum CUjit_option_enum { + /** + * Max number of registers that a thread may use.\n + * Option type: unsigned int\n + * Applies to: compiler only + */ + CU_JIT_MAX_REGISTERS = 0, + + /** + * IN: Specifies minimum number of threads per block to target compilation + * for\n + * OUT: Returns the number of threads the compiler actually targeted. + * This restricts the resource utilization fo the compiler (e.g. max + * registers) such that a block with the given number of threads should be + * able to launch based on register limitations. Note, this option does not + * currently take into account any other resource limitations, such as + * shared memory utilization.\n + * Cannot be combined with ::CU_JIT_TARGET.\n + * Option type: unsigned int\n + * Applies to: compiler only + */ + CU_JIT_THREADS_PER_BLOCK, + + /** + * Overwrites the option value with the total wall clock time, in + * milliseconds, spent in the compiler and linker\n + * Option type: float\n + * Applies to: compiler and linker + */ + CU_JIT_WALL_TIME, + + /** + * Pointer to a buffer in which to print any log messages + * that are informational in nature (the buffer size is specified via + * option ::CU_JIT_INFO_LOG_BUFFER_SIZE_BYTES)\n + * Option type: char *\n + * Applies to: compiler and linker + */ + CU_JIT_INFO_LOG_BUFFER, + + /** + * IN: Log buffer size in bytes. Log messages will be capped at this size + * (including null terminator)\n + * OUT: Amount of log buffer filled with messages\n + * Option type: unsigned int\n + * Applies to: compiler and linker + */ + CU_JIT_INFO_LOG_BUFFER_SIZE_BYTES, + + /** + * Pointer to a buffer in which to print any log messages that + * reflect errors (the buffer size is specified via option + * ::CU_JIT_ERROR_LOG_BUFFER_SIZE_BYTES)\n + * Option type: char *\n + * Applies to: compiler and linker + */ + CU_JIT_ERROR_LOG_BUFFER, + + /** + * IN: Log buffer size in bytes. Log messages will be capped at this size + * (including null terminator)\n + * OUT: Amount of log buffer filled with messages\n + * Option type: unsigned int\n + * Applies to: compiler and linker + */ + CU_JIT_ERROR_LOG_BUFFER_SIZE_BYTES, + + /** + * Level of optimizations to apply to generated code (0 - 4), with 4 + * being the default and highest level of optimizations.\n + * Option type: unsigned int\n + * Applies to: compiler only + */ + CU_JIT_OPTIMIZATION_LEVEL, + + /** + * No option value required. Determines the target based on the current + * attached context (default)\n + * Option type: No option value needed\n + * Applies to: compiler and linker + */ + CU_JIT_TARGET_FROM_CUCONTEXT, + + /** + * Target is chosen based on supplied ::CUjit_target. Cannot be + * combined with ::CU_JIT_THREADS_PER_BLOCK.\n + * Option type: unsigned int for enumerated type ::CUjit_target\n + * Applies to: compiler and linker + */ + CU_JIT_TARGET, + + /** + * Specifies choice of fallback strategy if matching cubin is not found. + * Choice is based on supplied ::CUjit_fallback. This option cannot be + * used with cuLink* APIs as the linker requires exact matches.\n + * Option type: unsigned int for enumerated type ::CUjit_fallback\n + * Applies to: compiler only + */ + CU_JIT_FALLBACK_STRATEGY, + + /** + * Specifies whether to create debug information in output (-g) + * (0: false, default)\n + * Option type: int\n + * Applies to: compiler and linker + */ + CU_JIT_GENERATE_DEBUG_INFO, + + /** + * Generate verbose log messages (0: false, default)\n + * Option type: int\n + * Applies to: compiler and linker + */ + CU_JIT_LOG_VERBOSE, + + /** + * Generate line number information (-lineinfo) (0: false, default)\n + * Option type: int\n + * Applies to: compiler only + */ + CU_JIT_GENERATE_LINE_INFO, + + /** + * Specifies whether to enable caching explicitly (-dlcm) \n + * Choice is based on supplied ::CUjit_cacheMode_enum.\n + * Option type: unsigned int for enumerated type ::CUjit_cacheMode_enum\n + * Applies to: compiler only + */ + CU_JIT_CACHE_MODE, + + /** + * The below jit options are used for internal purposes only, in this version + * of CUDA + */ + CU_JIT_NEW_SM3X_OPT, + CU_JIT_FAST_COMPILE, + + /** + * Array of device symbol names that will be relocated to the corresponing + * host addresses stored in ::CU_JIT_GLOBAL_SYMBOL_ADDRESSES.\n + * Must contain ::CU_JIT_GLOBAL_SYMBOL_COUNT entries.\n + * When loding a device module, driver will relocate all encountered + * unresolved symbols to the host addresses.\n + * It is only allowed to register symbols that correspond to unresolved + * global variables.\n + * It is illegal to register the same device symbol at multiple addresses.\n + * Option type: const char **\n + * Applies to: dynamic linker only + */ + CU_JIT_GLOBAL_SYMBOL_NAMES, + + /** + * Array of host addresses that will be used to relocate corresponding + * device symbols stored in ::CU_JIT_GLOBAL_SYMBOL_NAMES.\n + * Must contain ::CU_JIT_GLOBAL_SYMBOL_COUNT entries.\n + * Option type: void **\n + * Applies to: dynamic linker only + */ + CU_JIT_GLOBAL_SYMBOL_ADDRESSES, + + /** + * Number of entries in ::CU_JIT_GLOBAL_SYMBOL_NAMES and + * ::CU_JIT_GLOBAL_SYMBOL_ADDRESSES arrays.\n + * Option type: unsigned int\n + * Applies to: dynamic linker only + */ + CU_JIT_GLOBAL_SYMBOL_COUNT, + + CU_JIT_NUM_OPTIONS } CUjit_option; /** * Online compilation targets */ -typedef enum CUjit_target_enum -{ - CU_TARGET_COMPUTE_20 = 20, /**< Compute device class 2.0 */ - CU_TARGET_COMPUTE_21 = 21, /**< Compute device class 2.1 */ - CU_TARGET_COMPUTE_30 = 30, /**< Compute device class 3.0 */ - CU_TARGET_COMPUTE_32 = 32, /**< Compute device class 3.2 */ - CU_TARGET_COMPUTE_35 = 35, /**< Compute device class 3.5 */ - CU_TARGET_COMPUTE_37 = 37, /**< Compute device class 3.7 */ - CU_TARGET_COMPUTE_50 = 50, /**< Compute device class 5.0 */ - CU_TARGET_COMPUTE_52 = 52, /**< Compute device class 5.2 */ - CU_TARGET_COMPUTE_53 = 53, /**< Compute device class 5.3 */ - CU_TARGET_COMPUTE_60 = 60, /**< Compute device class 6.0.*/ - CU_TARGET_COMPUTE_61 = 61, /**< Compute device class 6.1.*/ - CU_TARGET_COMPUTE_62 = 62, /**< Compute device class 6.2.*/ - CU_TARGET_COMPUTE_70 = 70, /**< Compute device class 7.0.*/ - CU_TARGET_COMPUTE_72 = 72, /**< Compute device class 7.2.*/ - CU_TARGET_COMPUTE_75 = 75 /**< Compute device class 7.5.*/ +typedef enum CUjit_target_enum { + CU_TARGET_COMPUTE_20 = 20, /**< Compute device class 2.0 */ + CU_TARGET_COMPUTE_21 = 21, /**< Compute device class 2.1 */ + CU_TARGET_COMPUTE_30 = 30, /**< Compute device class 3.0 */ + CU_TARGET_COMPUTE_32 = 32, /**< Compute device class 3.2 */ + CU_TARGET_COMPUTE_35 = 35, /**< Compute device class 3.5 */ + CU_TARGET_COMPUTE_37 = 37, /**< Compute device class 3.7 */ + CU_TARGET_COMPUTE_50 = 50, /**< Compute device class 5.0 */ + CU_TARGET_COMPUTE_52 = 52, /**< Compute device class 5.2 */ + CU_TARGET_COMPUTE_53 = 53, /**< Compute device class 5.3 */ + CU_TARGET_COMPUTE_60 = 60, /**< Compute device class 6.0.*/ + CU_TARGET_COMPUTE_61 = 61, /**< Compute device class 6.1.*/ + CU_TARGET_COMPUTE_62 = 62, /**< Compute device class 6.2.*/ + CU_TARGET_COMPUTE_70 = 70, /**< Compute device class 7.0.*/ + CU_TARGET_COMPUTE_72 = 72, /**< Compute device class 7.2.*/ + CU_TARGET_COMPUTE_75 = 75 /**< Compute device class 7.5.*/ } CUjit_target; /** * Cubin matching fallback strategies */ -typedef enum CUjit_fallback_enum -{ - CU_PREFER_PTX = 0, /**< Prefer to compile ptx if exact binary match not found */ +typedef enum CUjit_fallback_enum { + CU_PREFER_PTX = + 0, /**< Prefer to compile ptx if exact binary match not found */ - CU_PREFER_BINARY /**< Prefer to fall back to compatible binary code if exact match not found */ + CU_PREFER_BINARY /**< Prefer to fall back to compatible binary code if exact + match not found */ } CUjit_fallback; /** * Caching modes for dlcm */ -typedef enum CUjit_cacheMode_enum -{ - CU_JIT_CACHE_OPTION_NONE = 0, /**< Compile with no -dlcm flag specified */ - CU_JIT_CACHE_OPTION_CG, /**< Compile with L1 cache disabled */ - CU_JIT_CACHE_OPTION_CA /**< Compile with L1 cache enabled */ +typedef enum CUjit_cacheMode_enum { + CU_JIT_CACHE_OPTION_NONE = 0, /**< Compile with no -dlcm flag specified */ + CU_JIT_CACHE_OPTION_CG, /**< Compile with L1 cache disabled */ + CU_JIT_CACHE_OPTION_CA /**< Compile with L1 cache enabled */ } CUjit_cacheMode; /** * Device code formats */ -typedef enum CUjitInputType_enum -{ - /** - * Compiled device-class-specific device code\n - * Applicable options: none - */ - CU_JIT_INPUT_CUBIN = 0, - - /** - * PTX source code\n - * Applicable options: PTX compiler options - */ - CU_JIT_INPUT_PTX, - - /** - * Bundle of multiple cubins and/or PTX of some device code\n - * Applicable options: PTX compiler options, ::CU_JIT_FALLBACK_STRATEGY - */ - CU_JIT_INPUT_FATBINARY, - - /** - * Host object with embedded device code\n - * Applicable options: PTX compiler options, ::CU_JIT_FALLBACK_STRATEGY - */ - CU_JIT_INPUT_OBJECT, - - /** - * Archive of host objects with embedded device code\n - * Applicable options: PTX compiler options, ::CU_JIT_FALLBACK_STRATEGY - */ - CU_JIT_INPUT_LIBRARY, - - CU_JIT_NUM_INPUT_TYPES +typedef enum CUjitInputType_enum { + /** + * Compiled device-class-specific device code\n + * Applicable options: none + */ + CU_JIT_INPUT_CUBIN = 0, + + /** + * PTX source code\n + * Applicable options: PTX compiler options + */ + CU_JIT_INPUT_PTX, + + /** + * Bundle of multiple cubins and/or PTX of some device code\n + * Applicable options: PTX compiler options, ::CU_JIT_FALLBACK_STRATEGY + */ + CU_JIT_INPUT_FATBINARY, + + /** + * Host object with embedded device code\n + * Applicable options: PTX compiler options, ::CU_JIT_FALLBACK_STRATEGY + */ + CU_JIT_INPUT_OBJECT, + + /** + * Archive of host objects with embedded device code\n + * Applicable options: PTX compiler options, ::CU_JIT_FALLBACK_STRATEGY + */ + CU_JIT_INPUT_LIBRARY, + + CU_JIT_NUM_INPUT_TYPES } CUjitInputType; #if __CUDA_API_VERSION >= 5050 @@ -1048,55 +1238,59 @@ typedef struct CUlinkState_st *CUlinkState; * Flags to register a graphics resource */ typedef enum CUgraphicsRegisterFlags_enum { - CU_GRAPHICS_REGISTER_FLAGS_NONE = 0x00, - CU_GRAPHICS_REGISTER_FLAGS_READ_ONLY = 0x01, - CU_GRAPHICS_REGISTER_FLAGS_WRITE_DISCARD = 0x02, - CU_GRAPHICS_REGISTER_FLAGS_SURFACE_LDST = 0x04, - CU_GRAPHICS_REGISTER_FLAGS_TEXTURE_GATHER = 0x08 + CU_GRAPHICS_REGISTER_FLAGS_NONE = 0x00, + CU_GRAPHICS_REGISTER_FLAGS_READ_ONLY = 0x01, + CU_GRAPHICS_REGISTER_FLAGS_WRITE_DISCARD = 0x02, + CU_GRAPHICS_REGISTER_FLAGS_SURFACE_LDST = 0x04, + CU_GRAPHICS_REGISTER_FLAGS_TEXTURE_GATHER = 0x08 } CUgraphicsRegisterFlags; /** * Flags for mapping and unmapping interop resources */ typedef enum CUgraphicsMapResourceFlags_enum { - CU_GRAPHICS_MAP_RESOURCE_FLAGS_NONE = 0x00, - CU_GRAPHICS_MAP_RESOURCE_FLAGS_READ_ONLY = 0x01, - CU_GRAPHICS_MAP_RESOURCE_FLAGS_WRITE_DISCARD = 0x02 + CU_GRAPHICS_MAP_RESOURCE_FLAGS_NONE = 0x00, + CU_GRAPHICS_MAP_RESOURCE_FLAGS_READ_ONLY = 0x01, + CU_GRAPHICS_MAP_RESOURCE_FLAGS_WRITE_DISCARD = 0x02 } CUgraphicsMapResourceFlags; /** * Array indices for cube faces */ typedef enum CUarray_cubemap_face_enum { - CU_CUBEMAP_FACE_POSITIVE_X = 0x00, /**< Positive X face of cubemap */ - CU_CUBEMAP_FACE_NEGATIVE_X = 0x01, /**< Negative X face of cubemap */ - CU_CUBEMAP_FACE_POSITIVE_Y = 0x02, /**< Positive Y face of cubemap */ - CU_CUBEMAP_FACE_NEGATIVE_Y = 0x03, /**< Negative Y face of cubemap */ - CU_CUBEMAP_FACE_POSITIVE_Z = 0x04, /**< Positive Z face of cubemap */ - CU_CUBEMAP_FACE_NEGATIVE_Z = 0x05 /**< Negative Z face of cubemap */ + CU_CUBEMAP_FACE_POSITIVE_X = 0x00, /**< Positive X face of cubemap */ + CU_CUBEMAP_FACE_NEGATIVE_X = 0x01, /**< Negative X face of cubemap */ + CU_CUBEMAP_FACE_POSITIVE_Y = 0x02, /**< Positive Y face of cubemap */ + CU_CUBEMAP_FACE_NEGATIVE_Y = 0x03, /**< Negative Y face of cubemap */ + CU_CUBEMAP_FACE_POSITIVE_Z = 0x04, /**< Positive Z face of cubemap */ + CU_CUBEMAP_FACE_NEGATIVE_Z = 0x05 /**< Negative Z face of cubemap */ } CUarray_cubemap_face; /** * Limits */ typedef enum CUlimit_enum { - CU_LIMIT_STACK_SIZE = 0x00, /**< GPU thread stack size */ - CU_LIMIT_PRINTF_FIFO_SIZE = 0x01, /**< GPU printf FIFO size */ - CU_LIMIT_MALLOC_HEAP_SIZE = 0x02, /**< GPU malloc heap size */ - CU_LIMIT_DEV_RUNTIME_SYNC_DEPTH = 0x03, /**< GPU device runtime launch synchronize depth */ - CU_LIMIT_DEV_RUNTIME_PENDING_LAUNCH_COUNT = 0x04, /**< GPU device runtime pending launch count */ - CU_LIMIT_MAX_L2_FETCH_GRANULARITY = 0x05, /**< A value between 0 and 128 that indicates the maximum fetch granularity of L2 (in Bytes). This is a hint */ - CU_LIMIT_MAX + CU_LIMIT_STACK_SIZE = 0x00, /**< GPU thread stack size */ + CU_LIMIT_PRINTF_FIFO_SIZE = 0x01, /**< GPU printf FIFO size */ + CU_LIMIT_MALLOC_HEAP_SIZE = 0x02, /**< GPU malloc heap size */ + CU_LIMIT_DEV_RUNTIME_SYNC_DEPTH = + 0x03, /**< GPU device runtime launch synchronize depth */ + CU_LIMIT_DEV_RUNTIME_PENDING_LAUNCH_COUNT = + 0x04, /**< GPU device runtime pending launch count */ + CU_LIMIT_MAX_L2_FETCH_GRANULARITY = + 0x05, /**< A value between 0 and 128 that indicates the maximum fetch + granularity of L2 (in Bytes). This is a hint */ + CU_LIMIT_MAX } CUlimit; /** * Resource types */ typedef enum CUresourcetype_enum { - CU_RESOURCE_TYPE_ARRAY = 0x00, /**< Array resoure */ - CU_RESOURCE_TYPE_MIPMAPPED_ARRAY = 0x01, /**< Mipmapped array resource */ - CU_RESOURCE_TYPE_LINEAR = 0x02, /**< Linear resource */ - CU_RESOURCE_TYPE_PITCH2D = 0x03 /**< Pitch 2D resource */ + CU_RESOURCE_TYPE_ARRAY = 0x00, /**< Array resoure */ + CU_RESOURCE_TYPE_MIPMAPPED_ARRAY = 0x01, /**< Mipmapped array resource */ + CU_RESOURCE_TYPE_LINEAR = 0x02, /**< Linear resource */ + CU_RESOURCE_TYPE_PITCH2D = 0x03 /**< Pitch 2D resource */ } CUresourcetype; #ifdef _WIN32 @@ -1111,65 +1305,69 @@ typedef enum CUresourcetype_enum { * CUDA host function * \param userData Argument value passed to the function */ -typedef void (CUDA_CB *CUhostFn)(void *userData); +typedef void(CUDA_CB *CUhostFn)(void *userData); /** * GPU kernel node parameters */ typedef struct CUDA_KERNEL_NODE_PARAMS_st { - CUfunction func; /**< Kernel to launch */ - unsigned int gridDimX; /**< Width of grid in blocks */ - unsigned int gridDimY; /**< Height of grid in blocks */ - unsigned int gridDimZ; /**< Depth of grid in blocks */ - unsigned int blockDimX; /**< X dimension of each thread block */ - unsigned int blockDimY; /**< Y dimension of each thread block */ - unsigned int blockDimZ; /**< Z dimension of each thread block */ - unsigned int sharedMemBytes; /**< Dynamic shared-memory size per thread block in bytes */ - void **kernelParams; /**< Array of pointers to kernel parameters */ - void **extra; /**< Extra options */ + CUfunction func; /**< Kernel to launch */ + unsigned int gridDimX; /**< Width of grid in blocks */ + unsigned int gridDimY; /**< Height of grid in blocks */ + unsigned int gridDimZ; /**< Depth of grid in blocks */ + unsigned int blockDimX; /**< X dimension of each thread block */ + unsigned int blockDimY; /**< Y dimension of each thread block */ + unsigned int blockDimZ; /**< Z dimension of each thread block */ + unsigned int sharedMemBytes; /**< Dynamic shared-memory size per thread block + in bytes */ + void **kernelParams; /**< Array of pointers to kernel parameters */ + void **extra; /**< Extra options */ } CUDA_KERNEL_NODE_PARAMS; /** * Memset node parameters */ typedef struct CUDA_MEMSET_NODE_PARAMS_st { - CUdeviceptr dst; /**< Destination device pointer */ - size_t pitch; /**< Pitch of destination device pointer. Unused if height is 1 */ - unsigned int value; /**< Value to be set */ - unsigned int elementSize; /**< Size of each element in bytes. Must be 1, 2, or 4. */ - size_t width; /**< Width in bytes, of the row */ - size_t height; /**< Number of rows */ + CUdeviceptr dst; /**< Destination device pointer */ + size_t + pitch; /**< Pitch of destination device pointer. Unused if height is 1 */ + unsigned int value; /**< Value to be set */ + unsigned int + elementSize; /**< Size of each element in bytes. Must be 1, 2, or 4. */ + size_t width; /**< Width in bytes, of the row */ + size_t height; /**< Number of rows */ } CUDA_MEMSET_NODE_PARAMS; /** * Host node parameters */ typedef struct CUDA_HOST_NODE_PARAMS_st { - CUhostFn fn; /**< The function to call when the node executes */ - void* userData; /**< Argument to pass to the function */ + CUhostFn fn; /**< The function to call when the node executes */ + void *userData; /**< Argument to pass to the function */ } CUDA_HOST_NODE_PARAMS; /** * Graph node types */ typedef enum CUgraphNodeType_enum { - CU_GRAPH_NODE_TYPE_KERNEL = 0, /**< GPU kernel node */ - CU_GRAPH_NODE_TYPE_MEMCPY = 1, /**< Memcpy node */ - CU_GRAPH_NODE_TYPE_MEMSET = 2, /**< Memset node */ - CU_GRAPH_NODE_TYPE_HOST = 3, /**< Host (executable) node */ - CU_GRAPH_NODE_TYPE_GRAPH = 4, /**< Node which executes an embedded graph */ - CU_GRAPH_NODE_TYPE_EMPTY = 5, /**< Empty (no-op) node */ - CU_GRAPH_NODE_TYPE_COUNT + CU_GRAPH_NODE_TYPE_KERNEL = 0, /**< GPU kernel node */ + CU_GRAPH_NODE_TYPE_MEMCPY = 1, /**< Memcpy node */ + CU_GRAPH_NODE_TYPE_MEMSET = 2, /**< Memset node */ + CU_GRAPH_NODE_TYPE_HOST = 3, /**< Host (executable) node */ + CU_GRAPH_NODE_TYPE_GRAPH = 4, /**< Node which executes an embedded graph */ + CU_GRAPH_NODE_TYPE_EMPTY = 5, /**< Empty (no-op) node */ + CU_GRAPH_NODE_TYPE_COUNT } CUgraphNodeType; /** * Possible stream capture statuses returned by ::cuStreamIsCapturing */ typedef enum CUstreamCaptureStatus_enum { - CU_STREAM_CAPTURE_STATUS_NONE = 0, /**< Stream is not capturing */ - CU_STREAM_CAPTURE_STATUS_ACTIVE = 1, /**< Stream is actively capturing */ - CU_STREAM_CAPTURE_STATUS_INVALIDATED = 2 /**< Stream is part of a capture sequence that - has been invalidated, but not terminated */ + CU_STREAM_CAPTURE_STATUS_NONE = 0, /**< Stream is not capturing */ + CU_STREAM_CAPTURE_STATUS_ACTIVE = 1, /**< Stream is actively capturing */ + CU_STREAM_CAPTURE_STATUS_INVALIDATED = + 2 /**< Stream is part of a capture sequence that + has been invalidated, but not terminated */ } CUstreamCaptureStatus; #endif /* __CUDA_API_VERSION >= 10000 */ @@ -1181,532 +1379,541 @@ typedef enum CUstreamCaptureStatus_enum { * ::cuStreamBeginCapture and ::cuThreadExchangeStreamCaptureMode */ typedef enum CUstreamCaptureMode_enum { - CU_STREAM_CAPTURE_MODE_GLOBAL = 0, - CU_STREAM_CAPTURE_MODE_THREAD_LOCAL = 1, - CU_STREAM_CAPTURE_MODE_RELAXED = 2 + CU_STREAM_CAPTURE_MODE_GLOBAL = 0, + CU_STREAM_CAPTURE_MODE_THREAD_LOCAL = 1, + CU_STREAM_CAPTURE_MODE_RELAXED = 2 } CUstreamCaptureMode; #endif /* __CUDA_API_VERSION >= 10010 */ -/** - * Error codes - */ -typedef enum cudaError_enum { - /** - * The API call returned with no errors. In the case of query calls, this - * also means that the operation being queried is complete (see - * ::cuEventQuery() and ::cuStreamQuery()). - */ - CUDA_SUCCESS = 0, - - /** - * This indicates that one or more of the parameters passed to the API call - * is not within an acceptable range of values. - */ - CUDA_ERROR_INVALID_VALUE = 1, - - /** - * The API call failed because it was unable to allocate enough memory to - * perform the requested operation. - */ - CUDA_ERROR_OUT_OF_MEMORY = 2, - - /** - * This indicates that the CUDA driver has not been initialized with - * ::cuInit() or that initialization has failed. - */ - CUDA_ERROR_NOT_INITIALIZED = 3, - - /** - * This indicates that the CUDA driver is in the process of shutting down. - */ - CUDA_ERROR_DEINITIALIZED = 4, - - /** - * This indicates profiler is not initialized for this run. This can - * happen when the application is running with external profiling tools - * like visual profiler. - */ - CUDA_ERROR_PROFILER_DISABLED = 5, - - /** - * \deprecated - * This error return is deprecated as of CUDA 5.0. It is no longer an error - * to attempt to enable/disable the profiling via ::cuProfilerStart or - * ::cuProfilerStop without initialization. - */ - CUDA_ERROR_PROFILER_NOT_INITIALIZED = 6, - - /** - * \deprecated - * This error return is deprecated as of CUDA 5.0. It is no longer an error - * to call cuProfilerStart() when profiling is already enabled. - */ - CUDA_ERROR_PROFILER_ALREADY_STARTED = 7, - - /** - * \deprecated - * This error return is deprecated as of CUDA 5.0. It is no longer an error - * to call cuProfilerStop() when profiling is already disabled. - */ - CUDA_ERROR_PROFILER_ALREADY_STOPPED = 8, - - /** - * This indicates that no CUDA-capable devices were detected by the installed - * CUDA driver. - */ - CUDA_ERROR_NO_DEVICE = 100, - - /** - * This indicates that the device ordinal supplied by the user does not - * correspond to a valid CUDA device. - */ - CUDA_ERROR_INVALID_DEVICE = 101, - - - /** - * This indicates that the device kernel image is invalid. This can also - * indicate an invalid CUDA module. - */ - CUDA_ERROR_INVALID_IMAGE = 200, - - /** - * This most frequently indicates that there is no context bound to the - * current thread. This can also be returned if the context passed to an - * API call is not a valid handle (such as a context that has had - * ::cuCtxDestroy() invoked on it). This can also be returned if a user - * mixes different API versions (i.e. 3010 context with 3020 API calls). - * See ::cuCtxGetApiVersion() for more details. - */ - CUDA_ERROR_INVALID_CONTEXT = 201, - - /** - * This indicated that the context being supplied as a parameter to the - * API call was already the active context. - * \deprecated - * This error return is deprecated as of CUDA 3.2. It is no longer an - * error to attempt to push the active context via ::cuCtxPushCurrent(). - */ - CUDA_ERROR_CONTEXT_ALREADY_CURRENT = 202, - - /** - * This indicates that a map or register operation has failed. - */ - CUDA_ERROR_MAP_FAILED = 205, - - /** - * This indicates that an unmap or unregister operation has failed. - */ - CUDA_ERROR_UNMAP_FAILED = 206, - - /** - * This indicates that the specified array is currently mapped and thus - * cannot be destroyed. - */ - CUDA_ERROR_ARRAY_IS_MAPPED = 207, - - /** - * This indicates that the resource is already mapped. - */ - CUDA_ERROR_ALREADY_MAPPED = 208, - - /** - * This indicates that there is no kernel image available that is suitable - * for the device. This can occur when a user specifies code generation - * options for a particular CUDA source file that do not include the - * corresponding device configuration. - */ - CUDA_ERROR_NO_BINARY_FOR_GPU = 209, - - /** - * This indicates that a resource has already been acquired. - */ - CUDA_ERROR_ALREADY_ACQUIRED = 210, - - /** - * This indicates that a resource is not mapped. - */ - CUDA_ERROR_NOT_MAPPED = 211, - - /** - * This indicates that a mapped resource is not available for access as an - * array. - */ - CUDA_ERROR_NOT_MAPPED_AS_ARRAY = 212, - - /** - * This indicates that a mapped resource is not available for access as a - * pointer. - */ - CUDA_ERROR_NOT_MAPPED_AS_POINTER = 213, - - /** - * This indicates that an uncorrectable ECC error was detected during - * execution. - */ - CUDA_ERROR_ECC_UNCORRECTABLE = 214, - - /** - * This indicates that the ::CUlimit passed to the API call is not - * supported by the active device. - */ - CUDA_ERROR_UNSUPPORTED_LIMIT = 215, - - /** - * This indicates that the ::CUcontext passed to the API call can - * only be bound to a single CPU thread at a time but is already - * bound to a CPU thread. - */ - CUDA_ERROR_CONTEXT_ALREADY_IN_USE = 216, - - /** - * This indicates that peer access is not supported across the given - * devices. - */ - CUDA_ERROR_PEER_ACCESS_UNSUPPORTED = 217, - - /** - * This indicates that a PTX JIT compilation failed. - */ - CUDA_ERROR_INVALID_PTX = 218, - - /** - * This indicates an error with OpenGL or DirectX context. - */ - CUDA_ERROR_INVALID_GRAPHICS_CONTEXT = 219, - - /** - * This indicates that an uncorrectable NVLink error was detected during the - * execution. - */ - CUDA_ERROR_NVLINK_UNCORRECTABLE = 220, - - /** - * This indicates that the PTX JIT compiler library was not found. - */ - CUDA_ERROR_JIT_COMPILER_NOT_FOUND = 221, - - /** - * This indicates that the device kernel source is invalid. - */ - CUDA_ERROR_INVALID_SOURCE = 300, - - /** - * This indicates that the file specified was not found. - */ - CUDA_ERROR_FILE_NOT_FOUND = 301, - - /** - * This indicates that a link to a shared object failed to resolve. - */ - CUDA_ERROR_SHARED_OBJECT_SYMBOL_NOT_FOUND = 302, - - /** - * This indicates that initialization of a shared object failed. - */ - CUDA_ERROR_SHARED_OBJECT_INIT_FAILED = 303, - - /** - * This indicates that an OS call failed. - */ - CUDA_ERROR_OPERATING_SYSTEM = 304, - - /** - * This indicates that a resource handle passed to the API call was not - * valid. Resource handles are opaque types like ::CUstream and ::CUevent. - */ - CUDA_ERROR_INVALID_HANDLE = 400, - - /** - * This indicates that a resource required by the API call is not in a - * valid state to perform the requested operation. - */ - CUDA_ERROR_ILLEGAL_STATE = 401, - - /** - * This indicates that a named symbol was not found. Examples of symbols - * are global/constant variable names, texture names, and surface names. - */ - CUDA_ERROR_NOT_FOUND = 500, - - /** - * This indicates that asynchronous operations issued previously have not - * completed yet. This result is not actually an error, but must be indicated - * differently than ::CUDA_SUCCESS (which indicates completion). Calls that - * may return this value include ::cuEventQuery() and ::cuStreamQuery(). - */ - CUDA_ERROR_NOT_READY = 600, - - /** - * While executing a kernel, the device encountered a - * load or store instruction on an invalid memory address. - * This leaves the process in an inconsistent state and any further CUDA work - * will return the same error. To continue using CUDA, the process must be terminated - * and relaunched. - */ - CUDA_ERROR_ILLEGAL_ADDRESS = 700, - - /** - * This indicates that a launch did not occur because it did not have - * appropriate resources. This error usually indicates that the user has - * attempted to pass too many arguments to the device kernel, or the - * kernel launch specifies too many threads for the kernel's register - * count. Passing arguments of the wrong size (i.e. a 64-bit pointer - * when a 32-bit int is expected) is equivalent to passing too many - * arguments and can also result in this error. - */ - CUDA_ERROR_LAUNCH_OUT_OF_RESOURCES = 701, - - /** - * This indicates that the device kernel took too long to execute. This can - * only occur if timeouts are enabled - see the device attribute - * ::CU_DEVICE_ATTRIBUTE_KERNEL_EXEC_TIMEOUT for more information. - * This leaves the process in an inconsistent state and any further CUDA work - * will return the same error. To continue using CUDA, the process must be terminated - * and relaunched. - */ - CUDA_ERROR_LAUNCH_TIMEOUT = 702, - - /** - * This error indicates a kernel launch that uses an incompatible texturing - * mode. - */ - CUDA_ERROR_LAUNCH_INCOMPATIBLE_TEXTURING = 703, - - /** - * This error indicates that a call to ::cuCtxEnablePeerAccess() is - * trying to re-enable peer access to a context which has already - * had peer access to it enabled. - */ - CUDA_ERROR_PEER_ACCESS_ALREADY_ENABLED = 704, - - /** - * This error indicates that ::cuCtxDisablePeerAccess() is - * trying to disable peer access which has not been enabled yet - * via ::cuCtxEnablePeerAccess(). - */ - CUDA_ERROR_PEER_ACCESS_NOT_ENABLED = 705, - - /** - * This error indicates that the primary context for the specified device - * has already been initialized. - */ - CUDA_ERROR_PRIMARY_CONTEXT_ACTIVE = 708, - - /** - * This error indicates that the context current to the calling thread - * has been destroyed using ::cuCtxDestroy, or is a primary context which - * has not yet been initialized. - */ - CUDA_ERROR_CONTEXT_IS_DESTROYED = 709, - - /** - * A device-side assert triggered during kernel execution. The context - * cannot be used anymore, and must be destroyed. All existing device - * memory allocations from this context are invalid and must be - * reconstructed if the program is to continue using CUDA. - */ - CUDA_ERROR_ASSERT = 710, - - /** - * This error indicates that the hardware resources required to enable - * peer access have been exhausted for one or more of the devices - * passed to ::cuCtxEnablePeerAccess(). - */ - CUDA_ERROR_TOO_MANY_PEERS = 711, - - /** - * This error indicates that the memory range passed to ::cuMemHostRegister() - * has already been registered. - */ - CUDA_ERROR_HOST_MEMORY_ALREADY_REGISTERED = 712, - - /** - * This error indicates that the pointer passed to ::cuMemHostUnregister() - * does not correspond to any currently registered memory region. - */ - CUDA_ERROR_HOST_MEMORY_NOT_REGISTERED = 713, - - /** - * While executing a kernel, the device encountered a stack error. - * This can be due to stack corruption or exceeding the stack size limit. - * This leaves the process in an inconsistent state and any further CUDA work - * will return the same error. To continue using CUDA, the process must be terminated - * and relaunched. - */ - CUDA_ERROR_HARDWARE_STACK_ERROR = 714, - - /** - * While executing a kernel, the device encountered an illegal instruction. - * This leaves the process in an inconsistent state and any further CUDA work - * will return the same error. To continue using CUDA, the process must be terminated - * and relaunched. - */ - CUDA_ERROR_ILLEGAL_INSTRUCTION = 715, - - /** - * While executing a kernel, the device encountered a load or store instruction - * on a memory address which is not aligned. - * This leaves the process in an inconsistent state and any further CUDA work - * will return the same error. To continue using CUDA, the process must be terminated - * and relaunched. - */ - CUDA_ERROR_MISALIGNED_ADDRESS = 716, - - /** - * While executing a kernel, the device encountered an instruction - * which can only operate on memory locations in certain address spaces - * (global, shared, or local), but was supplied a memory address not - * belonging to an allowed address space. - * This leaves the process in an inconsistent state and any further CUDA work - * will return the same error. To continue using CUDA, the process must be terminated - * and relaunched. - */ - CUDA_ERROR_INVALID_ADDRESS_SPACE = 717, - - /** - * While executing a kernel, the device program counter wrapped its address space. - * This leaves the process in an inconsistent state and any further CUDA work - * will return the same error. To continue using CUDA, the process must be terminated - * and relaunched. - */ - CUDA_ERROR_INVALID_PC = 718, - - /** - * An exception occurred on the device while executing a kernel. Common - * causes include dereferencing an invalid device pointer and accessing - * out of bounds shared memory. Less common cases can be system specific - more - * information about these cases can be found in the system specific user guide. - * This leaves the process in an inconsistent state and any further CUDA work - * will return the same error. To continue using CUDA, the process must be terminated - * and relaunched. - */ - CUDA_ERROR_LAUNCH_FAILED = 719, - - /** - * This error indicates that the number of blocks launched per grid for a kernel that was - * launched via either ::cuLaunchCooperativeKernel or ::cuLaunchCooperativeKernelMultiDevice - * exceeds the maximum number of blocks as allowed by ::cuOccupancyMaxActiveBlocksPerMultiprocessor - * or ::cuOccupancyMaxActiveBlocksPerMultiprocessorWithFlags times the number of multiprocessors - * as specified by the device attribute ::CU_DEVICE_ATTRIBUTE_MULTIPROCESSOR_COUNT. - */ - CUDA_ERROR_COOPERATIVE_LAUNCH_TOO_LARGE = 720, - - /** - * This error indicates that the attempted operation is not permitted. - */ - CUDA_ERROR_NOT_PERMITTED = 800, - - /** - * This error indicates that the attempted operation is not supported - * on the current system or device. - */ - CUDA_ERROR_NOT_SUPPORTED = 801, - - /** - * This error indicates that the system is not yet ready to start any CUDA - * work. To continue using CUDA, verify the system configuration is in a - * valid state and all required driver daemons are actively running. - * More information about this error can be found in the system specific - * user guide. - */ - CUDA_ERROR_SYSTEM_NOT_READY = 802, - - /** - * This error indicates that there is a mismatch between the versions of - * the display driver and the CUDA driver. Refer to the compatibility documentation - * for supported versions. - */ - CUDA_ERROR_SYSTEM_DRIVER_MISMATCH = 803, - - /** - * This error indicates that the system was upgraded to run with forward compatibility - * but the visible hardware detected by CUDA does not support this configuration. - * Refer to the compatibility documentation for the supported hardware matrix or ensure - * that only supported hardware is visible during initialization via the CUDA_VISIBLE_DEVICES - * environment variable. - */ - CUDA_ERROR_COMPAT_NOT_SUPPORTED_ON_DEVICE = 804, - - /** - * This error indicates that the operation is not permitted when - * the stream is capturing. - */ - CUDA_ERROR_STREAM_CAPTURE_UNSUPPORTED = 900, - - /** - * This error indicates that the current capture sequence on the stream - * has been invalidated due to a previous error. - */ - CUDA_ERROR_STREAM_CAPTURE_INVALIDATED = 901, - - /** - * This error indicates that the operation would have resulted in a merge - * of two independent capture sequences. - */ - CUDA_ERROR_STREAM_CAPTURE_MERGE = 902, - - /** - * This error indicates that the capture was not initiated in this stream. - */ - CUDA_ERROR_STREAM_CAPTURE_UNMATCHED = 903, - - /** - * This error indicates that the capture sequence contains a fork that was - * not joined to the primary stream. - */ - CUDA_ERROR_STREAM_CAPTURE_UNJOINED = 904, - - /** - * This error indicates that a dependency would have been created which - * crosses the capture sequence boundary. Only implicit in-stream ordering - * dependencies are allowed to cross the boundary. - */ - CUDA_ERROR_STREAM_CAPTURE_ISOLATION = 905, - - /** - * This error indicates a disallowed implicit dependency on a current capture - * sequence from cudaStreamLegacy. - */ - CUDA_ERROR_STREAM_CAPTURE_IMPLICIT = 906, - - /** - * This error indicates that the operation is not permitted on an event which - * was last recorded in a capturing stream. - */ - CUDA_ERROR_CAPTURED_EVENT = 907, - - /** - * A stream capture sequence not initiated with the ::CU_STREAM_CAPTURE_MODE_RELAXED - * argument to ::cuStreamBeginCapture was passed to ::cuStreamEndCapture in a - * different thread. - */ - CUDA_ERROR_STREAM_CAPTURE_WRONG_THREAD = 908, - - /** - * This indicates that an unknown internal error has occurred. - */ - CUDA_ERROR_UNKNOWN = 999 +/** + * Error codes + */ +typedef enum cudaError_enum { + /** + * The API call returned with no errors. In the case of query calls, this + * also means that the operation being queried is complete (see + * ::cuEventQuery() and ::cuStreamQuery()). + */ + CUDA_SUCCESS = 0, + + /** + * This indicates that one or more of the parameters passed to the API call + * is not within an acceptable range of values. + */ + CUDA_ERROR_INVALID_VALUE = 1, + + /** + * The API call failed because it was unable to allocate enough memory to + * perform the requested operation. + */ + CUDA_ERROR_OUT_OF_MEMORY = 2, + + /** + * This indicates that the CUDA driver has not been initialized with + * ::cuInit() or that initialization has failed. + */ + CUDA_ERROR_NOT_INITIALIZED = 3, + + /** + * This indicates that the CUDA driver is in the process of shutting down. + */ + CUDA_ERROR_DEINITIALIZED = 4, + + /** + * This indicates profiler is not initialized for this run. This can + * happen when the application is running with external profiling tools + * like visual profiler. + */ + CUDA_ERROR_PROFILER_DISABLED = 5, + + /** + * \deprecated + * This error return is deprecated as of CUDA 5.0. It is no longer an error + * to attempt to enable/disable the profiling via ::cuProfilerStart or + * ::cuProfilerStop without initialization. + */ + CUDA_ERROR_PROFILER_NOT_INITIALIZED = 6, + + /** + * \deprecated + * This error return is deprecated as of CUDA 5.0. It is no longer an error + * to call cuProfilerStart() when profiling is already enabled. + */ + CUDA_ERROR_PROFILER_ALREADY_STARTED = 7, + + /** + * \deprecated + * This error return is deprecated as of CUDA 5.0. It is no longer an error + * to call cuProfilerStop() when profiling is already disabled. + */ + CUDA_ERROR_PROFILER_ALREADY_STOPPED = 8, + + /** + * This indicates that no CUDA-capable devices were detected by the installed + * CUDA driver. + */ + CUDA_ERROR_NO_DEVICE = 100, + + /** + * This indicates that the device ordinal supplied by the user does not + * correspond to a valid CUDA device. + */ + CUDA_ERROR_INVALID_DEVICE = 101, + + /** + * This indicates that the device kernel image is invalid. This can also + * indicate an invalid CUDA module. + */ + CUDA_ERROR_INVALID_IMAGE = 200, + + /** + * This most frequently indicates that there is no context bound to the + * current thread. This can also be returned if the context passed to an + * API call is not a valid handle (such as a context that has had + * ::cuCtxDestroy() invoked on it). This can also be returned if a user + * mixes different API versions (i.e. 3010 context with 3020 API calls). + * See ::cuCtxGetApiVersion() for more details. + */ + CUDA_ERROR_INVALID_CONTEXT = 201, + + /** + * This indicated that the context being supplied as a parameter to the + * API call was already the active context. + * \deprecated + * This error return is deprecated as of CUDA 3.2. It is no longer an + * error to attempt to push the active context via ::cuCtxPushCurrent(). + */ + CUDA_ERROR_CONTEXT_ALREADY_CURRENT = 202, + + /** + * This indicates that a map or register operation has failed. + */ + CUDA_ERROR_MAP_FAILED = 205, + + /** + * This indicates that an unmap or unregister operation has failed. + */ + CUDA_ERROR_UNMAP_FAILED = 206, + + /** + * This indicates that the specified array is currently mapped and thus + * cannot be destroyed. + */ + CUDA_ERROR_ARRAY_IS_MAPPED = 207, + + /** + * This indicates that the resource is already mapped. + */ + CUDA_ERROR_ALREADY_MAPPED = 208, + + /** + * This indicates that there is no kernel image available that is suitable + * for the device. This can occur when a user specifies code generation + * options for a particular CUDA source file that do not include the + * corresponding device configuration. + */ + CUDA_ERROR_NO_BINARY_FOR_GPU = 209, + + /** + * This indicates that a resource has already been acquired. + */ + CUDA_ERROR_ALREADY_ACQUIRED = 210, + + /** + * This indicates that a resource is not mapped. + */ + CUDA_ERROR_NOT_MAPPED = 211, + + /** + * This indicates that a mapped resource is not available for access as an + * array. + */ + CUDA_ERROR_NOT_MAPPED_AS_ARRAY = 212, + + /** + * This indicates that a mapped resource is not available for access as a + * pointer. + */ + CUDA_ERROR_NOT_MAPPED_AS_POINTER = 213, + + /** + * This indicates that an uncorrectable ECC error was detected during + * execution. + */ + CUDA_ERROR_ECC_UNCORRECTABLE = 214, + + /** + * This indicates that the ::CUlimit passed to the API call is not + * supported by the active device. + */ + CUDA_ERROR_UNSUPPORTED_LIMIT = 215, + + /** + * This indicates that the ::CUcontext passed to the API call can + * only be bound to a single CPU thread at a time but is already + * bound to a CPU thread. + */ + CUDA_ERROR_CONTEXT_ALREADY_IN_USE = 216, + + /** + * This indicates that peer access is not supported across the given + * devices. + */ + CUDA_ERROR_PEER_ACCESS_UNSUPPORTED = 217, + + /** + * This indicates that a PTX JIT compilation failed. + */ + CUDA_ERROR_INVALID_PTX = 218, + + /** + * This indicates an error with OpenGL or DirectX context. + */ + CUDA_ERROR_INVALID_GRAPHICS_CONTEXT = 219, + + /** + * This indicates that an uncorrectable NVLink error was detected during the + * execution. + */ + CUDA_ERROR_NVLINK_UNCORRECTABLE = 220, + + /** + * This indicates that the PTX JIT compiler library was not found. + */ + CUDA_ERROR_JIT_COMPILER_NOT_FOUND = 221, + + /** + * This indicates that the device kernel source is invalid. + */ + CUDA_ERROR_INVALID_SOURCE = 300, + + /** + * This indicates that the file specified was not found. + */ + CUDA_ERROR_FILE_NOT_FOUND = 301, + + /** + * This indicates that a link to a shared object failed to resolve. + */ + CUDA_ERROR_SHARED_OBJECT_SYMBOL_NOT_FOUND = 302, + + /** + * This indicates that initialization of a shared object failed. + */ + CUDA_ERROR_SHARED_OBJECT_INIT_FAILED = 303, + + /** + * This indicates that an OS call failed. + */ + CUDA_ERROR_OPERATING_SYSTEM = 304, + + /** + * This indicates that a resource handle passed to the API call was not + * valid. Resource handles are opaque types like ::CUstream and ::CUevent. + */ + CUDA_ERROR_INVALID_HANDLE = 400, + + /** + * This indicates that a resource required by the API call is not in a + * valid state to perform the requested operation. + */ + CUDA_ERROR_ILLEGAL_STATE = 401, + + /** + * This indicates that a named symbol was not found. Examples of symbols + * are global/constant variable names, texture names, and surface names. + */ + CUDA_ERROR_NOT_FOUND = 500, + + /** + * This indicates that asynchronous operations issued previously have not + * completed yet. This result is not actually an error, but must be indicated + * differently than ::CUDA_SUCCESS (which indicates completion). Calls that + * may return this value include ::cuEventQuery() and ::cuStreamQuery(). + */ + CUDA_ERROR_NOT_READY = 600, + + /** + * While executing a kernel, the device encountered a + * load or store instruction on an invalid memory address. + * This leaves the process in an inconsistent state and any further CUDA work + * will return the same error. To continue using CUDA, the process must be + * terminated and relaunched. + */ + CUDA_ERROR_ILLEGAL_ADDRESS = 700, + + /** + * This indicates that a launch did not occur because it did not have + * appropriate resources. This error usually indicates that the user has + * attempted to pass too many arguments to the device kernel, or the + * kernel launch specifies too many threads for the kernel's register + * count. Passing arguments of the wrong size (i.e. a 64-bit pointer + * when a 32-bit int is expected) is equivalent to passing too many + * arguments and can also result in this error. + */ + CUDA_ERROR_LAUNCH_OUT_OF_RESOURCES = 701, + + /** + * This indicates that the device kernel took too long to execute. This can + * only occur if timeouts are enabled - see the device attribute + * ::CU_DEVICE_ATTRIBUTE_KERNEL_EXEC_TIMEOUT for more information. + * This leaves the process in an inconsistent state and any further CUDA work + * will return the same error. To continue using CUDA, the process must be + * terminated and relaunched. + */ + CUDA_ERROR_LAUNCH_TIMEOUT = 702, + + /** + * This error indicates a kernel launch that uses an incompatible texturing + * mode. + */ + CUDA_ERROR_LAUNCH_INCOMPATIBLE_TEXTURING = 703, + + /** + * This error indicates that a call to ::cuCtxEnablePeerAccess() is + * trying to re-enable peer access to a context which has already + * had peer access to it enabled. + */ + CUDA_ERROR_PEER_ACCESS_ALREADY_ENABLED = 704, + + /** + * This error indicates that ::cuCtxDisablePeerAccess() is + * trying to disable peer access which has not been enabled yet + * via ::cuCtxEnablePeerAccess(). + */ + CUDA_ERROR_PEER_ACCESS_NOT_ENABLED = 705, + + /** + * This error indicates that the primary context for the specified device + * has already been initialized. + */ + CUDA_ERROR_PRIMARY_CONTEXT_ACTIVE = 708, + + /** + * This error indicates that the context current to the calling thread + * has been destroyed using ::cuCtxDestroy, or is a primary context which + * has not yet been initialized. + */ + CUDA_ERROR_CONTEXT_IS_DESTROYED = 709, + + /** + * A device-side assert triggered during kernel execution. The context + * cannot be used anymore, and must be destroyed. All existing device + * memory allocations from this context are invalid and must be + * reconstructed if the program is to continue using CUDA. + */ + CUDA_ERROR_ASSERT = 710, + + /** + * This error indicates that the hardware resources required to enable + * peer access have been exhausted for one or more of the devices + * passed to ::cuCtxEnablePeerAccess(). + */ + CUDA_ERROR_TOO_MANY_PEERS = 711, + + /** + * This error indicates that the memory range passed to ::cuMemHostRegister() + * has already been registered. + */ + CUDA_ERROR_HOST_MEMORY_ALREADY_REGISTERED = 712, + + /** + * This error indicates that the pointer passed to ::cuMemHostUnregister() + * does not correspond to any currently registered memory region. + */ + CUDA_ERROR_HOST_MEMORY_NOT_REGISTERED = 713, + + /** + * While executing a kernel, the device encountered a stack error. + * This can be due to stack corruption or exceeding the stack size limit. + * This leaves the process in an inconsistent state and any further CUDA work + * will return the same error. To continue using CUDA, the process must be + * terminated and relaunched. + */ + CUDA_ERROR_HARDWARE_STACK_ERROR = 714, + + /** + * While executing a kernel, the device encountered an illegal instruction. + * This leaves the process in an inconsistent state and any further CUDA work + * will return the same error. To continue using CUDA, the process must be + * terminated and relaunched. + */ + CUDA_ERROR_ILLEGAL_INSTRUCTION = 715, + + /** + * While executing a kernel, the device encountered a load or store + * instruction on a memory address which is not aligned. This leaves the + * process in an inconsistent state and any further CUDA work will return the + * same error. To continue using CUDA, the process must be terminated and + * relaunched. + */ + CUDA_ERROR_MISALIGNED_ADDRESS = 716, + + /** + * While executing a kernel, the device encountered an instruction + * which can only operate on memory locations in certain address spaces + * (global, shared, or local), but was supplied a memory address not + * belonging to an allowed address space. + * This leaves the process in an inconsistent state and any further CUDA work + * will return the same error. To continue using CUDA, the process must be + * terminated and relaunched. + */ + CUDA_ERROR_INVALID_ADDRESS_SPACE = 717, + + /** + * While executing a kernel, the device program counter wrapped its address + * space. This leaves the process in an inconsistent state and any further + * CUDA work will return the same error. To continue using CUDA, the process + * must be terminated and relaunched. + */ + CUDA_ERROR_INVALID_PC = 718, + + /** + * An exception occurred on the device while executing a kernel. Common + * causes include dereferencing an invalid device pointer and accessing + * out of bounds shared memory. Less common cases can be system specific - + * more information about these cases can be found in the system specific user + * guide. This leaves the process in an inconsistent state and any further + * CUDA work will return the same error. To continue using CUDA, the process + * must be terminated and relaunched. + */ + CUDA_ERROR_LAUNCH_FAILED = 719, + + /** + * This error indicates that the number of blocks launched per grid for a + * kernel that was launched via either ::cuLaunchCooperativeKernel or + * ::cuLaunchCooperativeKernelMultiDevice exceeds the maximum number of blocks + * as allowed by ::cuOccupancyMaxActiveBlocksPerMultiprocessor or + * ::cuOccupancyMaxActiveBlocksPerMultiprocessorWithFlags times the number of + * multiprocessors as specified by the device attribute + * ::CU_DEVICE_ATTRIBUTE_MULTIPROCESSOR_COUNT. + */ + CUDA_ERROR_COOPERATIVE_LAUNCH_TOO_LARGE = 720, + + /** + * This error indicates that the attempted operation is not permitted. + */ + CUDA_ERROR_NOT_PERMITTED = 800, + + /** + * This error indicates that the attempted operation is not supported + * on the current system or device. + */ + CUDA_ERROR_NOT_SUPPORTED = 801, + + /** + * This error indicates that the system is not yet ready to start any CUDA + * work. To continue using CUDA, verify the system configuration is in a + * valid state and all required driver daemons are actively running. + * More information about this error can be found in the system specific + * user guide. + */ + CUDA_ERROR_SYSTEM_NOT_READY = 802, + + /** + * This error indicates that there is a mismatch between the versions of + * the display driver and the CUDA driver. Refer to the compatibility + * documentation for supported versions. + */ + CUDA_ERROR_SYSTEM_DRIVER_MISMATCH = 803, + + /** + * This error indicates that the system was upgraded to run with forward + * compatibility but the visible hardware detected by CUDA does not support + * this configuration. Refer to the compatibility documentation for the + * supported hardware matrix or ensure that only supported hardware is visible + * during initialization via the CUDA_VISIBLE_DEVICES environment variable. + */ + CUDA_ERROR_COMPAT_NOT_SUPPORTED_ON_DEVICE = 804, + + /** + * This error indicates that the operation is not permitted when + * the stream is capturing. + */ + CUDA_ERROR_STREAM_CAPTURE_UNSUPPORTED = 900, + + /** + * This error indicates that the current capture sequence on the stream + * has been invalidated due to a previous error. + */ + CUDA_ERROR_STREAM_CAPTURE_INVALIDATED = 901, + + /** + * This error indicates that the operation would have resulted in a merge + * of two independent capture sequences. + */ + CUDA_ERROR_STREAM_CAPTURE_MERGE = 902, + + /** + * This error indicates that the capture was not initiated in this stream. + */ + CUDA_ERROR_STREAM_CAPTURE_UNMATCHED = 903, + + /** + * This error indicates that the capture sequence contains a fork that was + * not joined to the primary stream. + */ + CUDA_ERROR_STREAM_CAPTURE_UNJOINED = 904, + + /** + * This error indicates that a dependency would have been created which + * crosses the capture sequence boundary. Only implicit in-stream ordering + * dependencies are allowed to cross the boundary. + */ + CUDA_ERROR_STREAM_CAPTURE_ISOLATION = 905, + + /** + * This error indicates a disallowed implicit dependency on a current capture + * sequence from cudaStreamLegacy. + */ + CUDA_ERROR_STREAM_CAPTURE_IMPLICIT = 906, + + /** + * This error indicates that the operation is not permitted on an event which + * was last recorded in a capturing stream. + */ + CUDA_ERROR_CAPTURED_EVENT = 907, + + /** + * A stream capture sequence not initiated with the + * ::CU_STREAM_CAPTURE_MODE_RELAXED argument to ::cuStreamBeginCapture was + * passed to ::cuStreamEndCapture in a different thread. + */ + CUDA_ERROR_STREAM_CAPTURE_WRONG_THREAD = 908, + + /** + * This indicates that an unknown internal error has occurred. + */ + CUDA_ERROR_UNKNOWN = 999 } CUresult; /** * P2P Attributes */ typedef enum CUdevice_P2PAttribute_enum { - CU_DEVICE_P2P_ATTRIBUTE_PERFORMANCE_RANK = 0x01, /**< A relative value indicating the performance of the link between two devices */ - CU_DEVICE_P2P_ATTRIBUTE_ACCESS_SUPPORTED = 0x02, /**< P2P Access is enable */ - CU_DEVICE_P2P_ATTRIBUTE_NATIVE_ATOMIC_SUPPORTED = 0x03, /**< Atomic operation over the link supported */ - CU_DEVICE_P2P_ATTRIBUTE_ACCESS_ACCESS_SUPPORTED = 0x04, /**< \deprecated use CU_DEVICE_P2P_ATTRIBUTE_CUDA_ARRAY_ACCESS_SUPPORTED instead */ - CU_DEVICE_P2P_ATTRIBUTE_CUDA_ARRAY_ACCESS_SUPPORTED = 0x04 /**< Accessing CUDA arrays over the link supported */ + CU_DEVICE_P2P_ATTRIBUTE_PERFORMANCE_RANK = + 0x01, /**< A relative value indicating the performance of the link between + two devices */ + CU_DEVICE_P2P_ATTRIBUTE_ACCESS_SUPPORTED = 0x02, /**< P2P Access is enable */ + CU_DEVICE_P2P_ATTRIBUTE_NATIVE_ATOMIC_SUPPORTED = + 0x03, /**< Atomic operation over the link supported */ + CU_DEVICE_P2P_ATTRIBUTE_ACCESS_ACCESS_SUPPORTED = + 0x04, /**< \deprecated use + CU_DEVICE_P2P_ATTRIBUTE_CUDA_ARRAY_ACCESS_SUPPORTED instead */ + CU_DEVICE_P2P_ATTRIBUTE_CUDA_ARRAY_ACCESS_SUPPORTED = + 0x04 /**< Accessing CUDA arrays over the link supported */ } CUdevice_P2PAttribute; /** * CUDA stream callback - * \param hStream The stream the callback was added to, as passed to ::cuStreamAddCallback. May be NULL. - * \param status ::CUDA_SUCCESS or any persistent error on the stream. - * \param userData User parameter provided at registration. + * \param hStream The stream the callback was added to, as passed to + * ::cuStreamAddCallback. May be NULL. \param status ::CUDA_SUCCESS or any + * persistent error on the stream. \param userData User parameter provided at + * registration. */ -typedef void (CUDA_CB *CUstreamCallback)(CUstream hStream, CUresult status, void *userData); +typedef void(CUDA_CB *CUstreamCallback)(CUstream hStream, CUresult status, + void *userData); /** * Block size to per-block dynamic shared memory mapping for a certain @@ -1714,20 +1921,20 @@ typedef void (CUDA_CB *CUstreamCallback)(CUstream hStream, CUresult status, void * * \return The dynamic shared memory needed by a block. */ -typedef size_t (CUDA_CB *CUoccupancyB2DSize)(int blockSize); +typedef size_t(CUDA_CB *CUoccupancyB2DSize)(int blockSize); /** * If set, host memory is portable between CUDA contexts. * Flag for ::cuMemHostAlloc() */ -#define CU_MEMHOSTALLOC_PORTABLE 0x01 +#define CU_MEMHOSTALLOC_PORTABLE 0x01 /** * If set, host memory is mapped into CUDA address space and * ::cuMemHostGetDevicePointer() may be called on the host pointer. * Flag for ::cuMemHostAlloc() */ -#define CU_MEMHOSTALLOC_DEVICEMAP 0x02 +#define CU_MEMHOSTALLOC_DEVICEMAP 0x02 /** * If set, host memory is allocated as write-combined - fast to write, @@ -1735,20 +1942,20 @@ typedef size_t (CUDA_CB *CUoccupancyB2DSize)(int blockSize); * (MOVNTDQA). * Flag for ::cuMemHostAlloc() */ -#define CU_MEMHOSTALLOC_WRITECOMBINED 0x04 +#define CU_MEMHOSTALLOC_WRITECOMBINED 0x04 /** * If set, host memory is portable between CUDA contexts. * Flag for ::cuMemHostRegister() */ -#define CU_MEMHOSTREGISTER_PORTABLE 0x01 +#define CU_MEMHOSTREGISTER_PORTABLE 0x01 /** * If set, host memory is mapped into CUDA address space and * ::cuMemHostGetDevicePointer() may be called on the host pointer. * Flag for ::cuMemHostRegister() */ -#define CU_MEMHOSTREGISTER_DEVICEMAP 0x02 +#define CU_MEMHOSTREGISTER_DEVICEMAP 0x02 /** * If set, the passed memory pointer is treated as pointing to some @@ -1762,7 +1969,7 @@ typedef size_t (CUDA_CB *CUoccupancyB2DSize)(int blockSize); * is returned. * Flag for ::cuMemHostRegister() */ -#define CU_MEMHOSTREGISTER_IOMEMORY 0x04 +#define CU_MEMHOSTREGISTER_IOMEMORY 0x04 #if __CUDA_API_VERSION >= 3020 @@ -1770,118 +1977,125 @@ typedef size_t (CUDA_CB *CUoccupancyB2DSize)(int blockSize); * 2D memory copy parameters */ typedef struct CUDA_MEMCPY2D_st { - size_t srcXInBytes; /**< Source X in bytes */ - size_t srcY; /**< Source Y */ - - CUmemorytype srcMemoryType; /**< Source memory type (host, device, array) */ - const void *srcHost; /**< Source host pointer */ - CUdeviceptr srcDevice; /**< Source device pointer */ - CUarray srcArray; /**< Source array reference */ - size_t srcPitch; /**< Source pitch (ignored when src is array) */ - - size_t dstXInBytes; /**< Destination X in bytes */ - size_t dstY; /**< Destination Y */ - - CUmemorytype dstMemoryType; /**< Destination memory type (host, device, array) */ - void *dstHost; /**< Destination host pointer */ - CUdeviceptr dstDevice; /**< Destination device pointer */ - CUarray dstArray; /**< Destination array reference */ - size_t dstPitch; /**< Destination pitch (ignored when dst is array) */ - - size_t WidthInBytes; /**< Width of 2D memory copy in bytes */ - size_t Height; /**< Height of 2D memory copy */ + size_t srcXInBytes; /**< Source X in bytes */ + size_t srcY; /**< Source Y */ + + CUmemorytype srcMemoryType; /**< Source memory type (host, device, array) */ + const void *srcHost; /**< Source host pointer */ + CUdeviceptr srcDevice; /**< Source device pointer */ + CUarray srcArray; /**< Source array reference */ + size_t srcPitch; /**< Source pitch (ignored when src is array) */ + + size_t dstXInBytes; /**< Destination X in bytes */ + size_t dstY; /**< Destination Y */ + + CUmemorytype + dstMemoryType; /**< Destination memory type (host, device, array) */ + void *dstHost; /**< Destination host pointer */ + CUdeviceptr dstDevice; /**< Destination device pointer */ + CUarray dstArray; /**< Destination array reference */ + size_t dstPitch; /**< Destination pitch (ignored when dst is array) */ + + size_t WidthInBytes; /**< Width of 2D memory copy in bytes */ + size_t Height; /**< Height of 2D memory copy */ } CUDA_MEMCPY2D; /** * 3D memory copy parameters */ typedef struct CUDA_MEMCPY3D_st { - size_t srcXInBytes; /**< Source X in bytes */ - size_t srcY; /**< Source Y */ - size_t srcZ; /**< Source Z */ - size_t srcLOD; /**< Source LOD */ - CUmemorytype srcMemoryType; /**< Source memory type (host, device, array) */ - const void *srcHost; /**< Source host pointer */ - CUdeviceptr srcDevice; /**< Source device pointer */ - CUarray srcArray; /**< Source array reference */ - void *reserved0; /**< Must be NULL */ - size_t srcPitch; /**< Source pitch (ignored when src is array) */ - size_t srcHeight; /**< Source height (ignored when src is array; may be 0 if Depth==1) */ - - size_t dstXInBytes; /**< Destination X in bytes */ - size_t dstY; /**< Destination Y */ - size_t dstZ; /**< Destination Z */ - size_t dstLOD; /**< Destination LOD */ - CUmemorytype dstMemoryType; /**< Destination memory type (host, device, array) */ - void *dstHost; /**< Destination host pointer */ - CUdeviceptr dstDevice; /**< Destination device pointer */ - CUarray dstArray; /**< Destination array reference */ - void *reserved1; /**< Must be NULL */ - size_t dstPitch; /**< Destination pitch (ignored when dst is array) */ - size_t dstHeight; /**< Destination height (ignored when dst is array; may be 0 if Depth==1) */ - - size_t WidthInBytes; /**< Width of 3D memory copy in bytes */ - size_t Height; /**< Height of 3D memory copy */ - size_t Depth; /**< Depth of 3D memory copy */ + size_t srcXInBytes; /**< Source X in bytes */ + size_t srcY; /**< Source Y */ + size_t srcZ; /**< Source Z */ + size_t srcLOD; /**< Source LOD */ + CUmemorytype srcMemoryType; /**< Source memory type (host, device, array) */ + const void *srcHost; /**< Source host pointer */ + CUdeviceptr srcDevice; /**< Source device pointer */ + CUarray srcArray; /**< Source array reference */ + void *reserved0; /**< Must be NULL */ + size_t srcPitch; /**< Source pitch (ignored when src is array) */ + size_t srcHeight; /**< Source height (ignored when src is array; may be 0 if + Depth==1) */ + + size_t dstXInBytes; /**< Destination X in bytes */ + size_t dstY; /**< Destination Y */ + size_t dstZ; /**< Destination Z */ + size_t dstLOD; /**< Destination LOD */ + CUmemorytype + dstMemoryType; /**< Destination memory type (host, device, array) */ + void *dstHost; /**< Destination host pointer */ + CUdeviceptr dstDevice; /**< Destination device pointer */ + CUarray dstArray; /**< Destination array reference */ + void *reserved1; /**< Must be NULL */ + size_t dstPitch; /**< Destination pitch (ignored when dst is array) */ + size_t dstHeight; /**< Destination height (ignored when dst is array; may be 0 + if Depth==1) */ + + size_t WidthInBytes; /**< Width of 3D memory copy in bytes */ + size_t Height; /**< Height of 3D memory copy */ + size_t Depth; /**< Depth of 3D memory copy */ } CUDA_MEMCPY3D; /** * 3D memory cross-context copy parameters */ typedef struct CUDA_MEMCPY3D_PEER_st { - size_t srcXInBytes; /**< Source X in bytes */ - size_t srcY; /**< Source Y */ - size_t srcZ; /**< Source Z */ - size_t srcLOD; /**< Source LOD */ - CUmemorytype srcMemoryType; /**< Source memory type (host, device, array) */ - const void *srcHost; /**< Source host pointer */ - CUdeviceptr srcDevice; /**< Source device pointer */ - CUarray srcArray; /**< Source array reference */ - CUcontext srcContext; /**< Source context (ignored with srcMemoryType is ::CU_MEMORYTYPE_ARRAY) */ - size_t srcPitch; /**< Source pitch (ignored when src is array) */ - size_t srcHeight; /**< Source height (ignored when src is array; may be 0 if Depth==1) */ - - size_t dstXInBytes; /**< Destination X in bytes */ - size_t dstY; /**< Destination Y */ - size_t dstZ; /**< Destination Z */ - size_t dstLOD; /**< Destination LOD */ - CUmemorytype dstMemoryType; /**< Destination memory type (host, device, array) */ - void *dstHost; /**< Destination host pointer */ - CUdeviceptr dstDevice; /**< Destination device pointer */ - CUarray dstArray; /**< Destination array reference */ - CUcontext dstContext; /**< Destination context (ignored with dstMemoryType is ::CU_MEMORYTYPE_ARRAY) */ - size_t dstPitch; /**< Destination pitch (ignored when dst is array) */ - size_t dstHeight; /**< Destination height (ignored when dst is array; may be 0 if Depth==1) */ - - size_t WidthInBytes; /**< Width of 3D memory copy in bytes */ - size_t Height; /**< Height of 3D memory copy */ - size_t Depth; /**< Depth of 3D memory copy */ + size_t srcXInBytes; /**< Source X in bytes */ + size_t srcY; /**< Source Y */ + size_t srcZ; /**< Source Z */ + size_t srcLOD; /**< Source LOD */ + CUmemorytype srcMemoryType; /**< Source memory type (host, device, array) */ + const void *srcHost; /**< Source host pointer */ + CUdeviceptr srcDevice; /**< Source device pointer */ + CUarray srcArray; /**< Source array reference */ + CUcontext srcContext; /**< Source context (ignored with srcMemoryType is + ::CU_MEMORYTYPE_ARRAY) */ + size_t srcPitch; /**< Source pitch (ignored when src is array) */ + size_t srcHeight; /**< Source height (ignored when src is array; may be 0 if + Depth==1) */ + + size_t dstXInBytes; /**< Destination X in bytes */ + size_t dstY; /**< Destination Y */ + size_t dstZ; /**< Destination Z */ + size_t dstLOD; /**< Destination LOD */ + CUmemorytype + dstMemoryType; /**< Destination memory type (host, device, array) */ + void *dstHost; /**< Destination host pointer */ + CUdeviceptr dstDevice; /**< Destination device pointer */ + CUarray dstArray; /**< Destination array reference */ + CUcontext dstContext; /**< Destination context (ignored with dstMemoryType is + ::CU_MEMORYTYPE_ARRAY) */ + size_t dstPitch; /**< Destination pitch (ignored when dst is array) */ + size_t dstHeight; /**< Destination height (ignored when dst is array; may be 0 + if Depth==1) */ + + size_t WidthInBytes; /**< Width of 3D memory copy in bytes */ + size_t Height; /**< Height of 3D memory copy */ + size_t Depth; /**< Depth of 3D memory copy */ } CUDA_MEMCPY3D_PEER; /** * Array descriptor */ -typedef struct CUDA_ARRAY_DESCRIPTOR_st -{ - size_t Width; /**< Width of array */ - size_t Height; /**< Height of array */ +typedef struct CUDA_ARRAY_DESCRIPTOR_st { + size_t Width; /**< Width of array */ + size_t Height; /**< Height of array */ - CUarray_format Format; /**< Array format */ - unsigned int NumChannels; /**< Channels per array element */ + CUarray_format Format; /**< Array format */ + unsigned int NumChannels; /**< Channels per array element */ } CUDA_ARRAY_DESCRIPTOR; /** * 3D array descriptor */ -typedef struct CUDA_ARRAY3D_DESCRIPTOR_st -{ - size_t Width; /**< Width of 3D array */ - size_t Height; /**< Height of 3D array */ - size_t Depth; /**< Depth of 3D array */ +typedef struct CUDA_ARRAY3D_DESCRIPTOR_st { + size_t Width; /**< Width of 3D array */ + size_t Height; /**< Height of 3D array */ + size_t Depth; /**< Depth of 3D array */ - CUarray_format Format; /**< Array format */ - unsigned int NumChannels; /**< Channels per array element */ - unsigned int Flags; /**< Flags */ + CUarray_format Format; /**< Array format */ + unsigned int NumChannels; /**< Channels per array element */ + unsigned int Flags; /**< Flags */ } CUDA_ARRAY3D_DESCRIPTOR; #endif /* __CUDA_API_VERSION >= 3020 */ @@ -1891,119 +2105,125 @@ typedef struct CUDA_ARRAY3D_DESCRIPTOR_st /** * CUDA Resource descriptor */ -typedef struct CUDA_RESOURCE_DESC_st -{ - CUresourcetype resType; /**< Resource type */ +typedef struct CUDA_RESOURCE_DESC_st { + CUresourcetype resType; /**< Resource type */ - union { - struct { - CUarray hArray; /**< CUDA array */ - } array; - struct { - CUmipmappedArray hMipmappedArray; /**< CUDA mipmapped array */ - } mipmap; - struct { - CUdeviceptr devPtr; /**< Device pointer */ - CUarray_format format; /**< Array format */ - unsigned int numChannels; /**< Channels per array element */ - size_t sizeInBytes; /**< Size in bytes */ - } linear; - struct { - CUdeviceptr devPtr; /**< Device pointer */ - CUarray_format format; /**< Array format */ - unsigned int numChannels; /**< Channels per array element */ - size_t width; /**< Width of the array in elements */ - size_t height; /**< Height of the array in elements */ - size_t pitchInBytes; /**< Pitch between two rows in bytes */ - } pitch2D; - struct { - int reserved[32]; - } reserved; - } res; - - unsigned int flags; /**< Flags (must be zero) */ + union { + struct { + CUarray hArray; /**< CUDA array */ + } array; + struct { + CUmipmappedArray hMipmappedArray; /**< CUDA mipmapped array */ + } mipmap; + struct { + CUdeviceptr devPtr; /**< Device pointer */ + CUarray_format format; /**< Array format */ + unsigned int numChannels; /**< Channels per array element */ + size_t sizeInBytes; /**< Size in bytes */ + } linear; + struct { + CUdeviceptr devPtr; /**< Device pointer */ + CUarray_format format; /**< Array format */ + unsigned int numChannels; /**< Channels per array element */ + size_t width; /**< Width of the array in elements */ + size_t height; /**< Height of the array in elements */ + size_t pitchInBytes; /**< Pitch between two rows in bytes */ + } pitch2D; + struct { + int reserved[32]; + } reserved; + } res; + + unsigned int flags; /**< Flags (must be zero) */ } CUDA_RESOURCE_DESC; /** * Texture descriptor */ typedef struct CUDA_TEXTURE_DESC_st { - CUaddress_mode addressMode[3]; /**< Address modes */ - CUfilter_mode filterMode; /**< Filter mode */ - unsigned int flags; /**< Flags */ - unsigned int maxAnisotropy; /**< Maximum anisotropy ratio */ - CUfilter_mode mipmapFilterMode; /**< Mipmap filter mode */ - float mipmapLevelBias; /**< Mipmap level bias */ - float minMipmapLevelClamp; /**< Mipmap minimum level clamp */ - float maxMipmapLevelClamp; /**< Mipmap maximum level clamp */ - float borderColor[4]; /**< Border Color */ - int reserved[12]; + CUaddress_mode addressMode[3]; /**< Address modes */ + CUfilter_mode filterMode; /**< Filter mode */ + unsigned int flags; /**< Flags */ + unsigned int maxAnisotropy; /**< Maximum anisotropy ratio */ + CUfilter_mode mipmapFilterMode; /**< Mipmap filter mode */ + float mipmapLevelBias; /**< Mipmap level bias */ + float minMipmapLevelClamp; /**< Mipmap minimum level clamp */ + float maxMipmapLevelClamp; /**< Mipmap maximum level clamp */ + float borderColor[4]; /**< Border Color */ + int reserved[12]; } CUDA_TEXTURE_DESC; /** * Resource view format */ -typedef enum CUresourceViewFormat_enum -{ - CU_RES_VIEW_FORMAT_NONE = 0x00, /**< No resource view format (use underlying resource format) */ - CU_RES_VIEW_FORMAT_UINT_1X8 = 0x01, /**< 1 channel unsigned 8-bit integers */ - CU_RES_VIEW_FORMAT_UINT_2X8 = 0x02, /**< 2 channel unsigned 8-bit integers */ - CU_RES_VIEW_FORMAT_UINT_4X8 = 0x03, /**< 4 channel unsigned 8-bit integers */ - CU_RES_VIEW_FORMAT_SINT_1X8 = 0x04, /**< 1 channel signed 8-bit integers */ - CU_RES_VIEW_FORMAT_SINT_2X8 = 0x05, /**< 2 channel signed 8-bit integers */ - CU_RES_VIEW_FORMAT_SINT_4X8 = 0x06, /**< 4 channel signed 8-bit integers */ - CU_RES_VIEW_FORMAT_UINT_1X16 = 0x07, /**< 1 channel unsigned 16-bit integers */ - CU_RES_VIEW_FORMAT_UINT_2X16 = 0x08, /**< 2 channel unsigned 16-bit integers */ - CU_RES_VIEW_FORMAT_UINT_4X16 = 0x09, /**< 4 channel unsigned 16-bit integers */ - CU_RES_VIEW_FORMAT_SINT_1X16 = 0x0a, /**< 1 channel signed 16-bit integers */ - CU_RES_VIEW_FORMAT_SINT_2X16 = 0x0b, /**< 2 channel signed 16-bit integers */ - CU_RES_VIEW_FORMAT_SINT_4X16 = 0x0c, /**< 4 channel signed 16-bit integers */ - CU_RES_VIEW_FORMAT_UINT_1X32 = 0x0d, /**< 1 channel unsigned 32-bit integers */ - CU_RES_VIEW_FORMAT_UINT_2X32 = 0x0e, /**< 2 channel unsigned 32-bit integers */ - CU_RES_VIEW_FORMAT_UINT_4X32 = 0x0f, /**< 4 channel unsigned 32-bit integers */ - CU_RES_VIEW_FORMAT_SINT_1X32 = 0x10, /**< 1 channel signed 32-bit integers */ - CU_RES_VIEW_FORMAT_SINT_2X32 = 0x11, /**< 2 channel signed 32-bit integers */ - CU_RES_VIEW_FORMAT_SINT_4X32 = 0x12, /**< 4 channel signed 32-bit integers */ - CU_RES_VIEW_FORMAT_FLOAT_1X16 = 0x13, /**< 1 channel 16-bit floating point */ - CU_RES_VIEW_FORMAT_FLOAT_2X16 = 0x14, /**< 2 channel 16-bit floating point */ - CU_RES_VIEW_FORMAT_FLOAT_4X16 = 0x15, /**< 4 channel 16-bit floating point */ - CU_RES_VIEW_FORMAT_FLOAT_1X32 = 0x16, /**< 1 channel 32-bit floating point */ - CU_RES_VIEW_FORMAT_FLOAT_2X32 = 0x17, /**< 2 channel 32-bit floating point */ - CU_RES_VIEW_FORMAT_FLOAT_4X32 = 0x18, /**< 4 channel 32-bit floating point */ - CU_RES_VIEW_FORMAT_UNSIGNED_BC1 = 0x19, /**< Block compressed 1 */ - CU_RES_VIEW_FORMAT_UNSIGNED_BC2 = 0x1a, /**< Block compressed 2 */ - CU_RES_VIEW_FORMAT_UNSIGNED_BC3 = 0x1b, /**< Block compressed 3 */ - CU_RES_VIEW_FORMAT_UNSIGNED_BC4 = 0x1c, /**< Block compressed 4 unsigned */ - CU_RES_VIEW_FORMAT_SIGNED_BC4 = 0x1d, /**< Block compressed 4 signed */ - CU_RES_VIEW_FORMAT_UNSIGNED_BC5 = 0x1e, /**< Block compressed 5 unsigned */ - CU_RES_VIEW_FORMAT_SIGNED_BC5 = 0x1f, /**< Block compressed 5 signed */ - CU_RES_VIEW_FORMAT_UNSIGNED_BC6H = 0x20, /**< Block compressed 6 unsigned half-float */ - CU_RES_VIEW_FORMAT_SIGNED_BC6H = 0x21, /**< Block compressed 6 signed half-float */ - CU_RES_VIEW_FORMAT_UNSIGNED_BC7 = 0x22 /**< Block compressed 7 */ +typedef enum CUresourceViewFormat_enum { + CU_RES_VIEW_FORMAT_NONE = + 0x00, /**< No resource view format (use underlying resource format) */ + CU_RES_VIEW_FORMAT_UINT_1X8 = 0x01, /**< 1 channel unsigned 8-bit integers */ + CU_RES_VIEW_FORMAT_UINT_2X8 = 0x02, /**< 2 channel unsigned 8-bit integers */ + CU_RES_VIEW_FORMAT_UINT_4X8 = 0x03, /**< 4 channel unsigned 8-bit integers */ + CU_RES_VIEW_FORMAT_SINT_1X8 = 0x04, /**< 1 channel signed 8-bit integers */ + CU_RES_VIEW_FORMAT_SINT_2X8 = 0x05, /**< 2 channel signed 8-bit integers */ + CU_RES_VIEW_FORMAT_SINT_4X8 = 0x06, /**< 4 channel signed 8-bit integers */ + CU_RES_VIEW_FORMAT_UINT_1X16 = + 0x07, /**< 1 channel unsigned 16-bit integers */ + CU_RES_VIEW_FORMAT_UINT_2X16 = + 0x08, /**< 2 channel unsigned 16-bit integers */ + CU_RES_VIEW_FORMAT_UINT_4X16 = + 0x09, /**< 4 channel unsigned 16-bit integers */ + CU_RES_VIEW_FORMAT_SINT_1X16 = 0x0a, /**< 1 channel signed 16-bit integers */ + CU_RES_VIEW_FORMAT_SINT_2X16 = 0x0b, /**< 2 channel signed 16-bit integers */ + CU_RES_VIEW_FORMAT_SINT_4X16 = 0x0c, /**< 4 channel signed 16-bit integers */ + CU_RES_VIEW_FORMAT_UINT_1X32 = + 0x0d, /**< 1 channel unsigned 32-bit integers */ + CU_RES_VIEW_FORMAT_UINT_2X32 = + 0x0e, /**< 2 channel unsigned 32-bit integers */ + CU_RES_VIEW_FORMAT_UINT_4X32 = + 0x0f, /**< 4 channel unsigned 32-bit integers */ + CU_RES_VIEW_FORMAT_SINT_1X32 = 0x10, /**< 1 channel signed 32-bit integers */ + CU_RES_VIEW_FORMAT_SINT_2X32 = 0x11, /**< 2 channel signed 32-bit integers */ + CU_RES_VIEW_FORMAT_SINT_4X32 = 0x12, /**< 4 channel signed 32-bit integers */ + CU_RES_VIEW_FORMAT_FLOAT_1X16 = 0x13, /**< 1 channel 16-bit floating point */ + CU_RES_VIEW_FORMAT_FLOAT_2X16 = 0x14, /**< 2 channel 16-bit floating point */ + CU_RES_VIEW_FORMAT_FLOAT_4X16 = 0x15, /**< 4 channel 16-bit floating point */ + CU_RES_VIEW_FORMAT_FLOAT_1X32 = 0x16, /**< 1 channel 32-bit floating point */ + CU_RES_VIEW_FORMAT_FLOAT_2X32 = 0x17, /**< 2 channel 32-bit floating point */ + CU_RES_VIEW_FORMAT_FLOAT_4X32 = 0x18, /**< 4 channel 32-bit floating point */ + CU_RES_VIEW_FORMAT_UNSIGNED_BC1 = 0x19, /**< Block compressed 1 */ + CU_RES_VIEW_FORMAT_UNSIGNED_BC2 = 0x1a, /**< Block compressed 2 */ + CU_RES_VIEW_FORMAT_UNSIGNED_BC3 = 0x1b, /**< Block compressed 3 */ + CU_RES_VIEW_FORMAT_UNSIGNED_BC4 = 0x1c, /**< Block compressed 4 unsigned */ + CU_RES_VIEW_FORMAT_SIGNED_BC4 = 0x1d, /**< Block compressed 4 signed */ + CU_RES_VIEW_FORMAT_UNSIGNED_BC5 = 0x1e, /**< Block compressed 5 unsigned */ + CU_RES_VIEW_FORMAT_SIGNED_BC5 = 0x1f, /**< Block compressed 5 signed */ + CU_RES_VIEW_FORMAT_UNSIGNED_BC6H = + 0x20, /**< Block compressed 6 unsigned half-float */ + CU_RES_VIEW_FORMAT_SIGNED_BC6H = + 0x21, /**< Block compressed 6 signed half-float */ + CU_RES_VIEW_FORMAT_UNSIGNED_BC7 = 0x22 /**< Block compressed 7 */ } CUresourceViewFormat; /** * Resource view descriptor */ -typedef struct CUDA_RESOURCE_VIEW_DESC_st -{ - CUresourceViewFormat format; /**< Resource view format */ - size_t width; /**< Width of the resource view */ - size_t height; /**< Height of the resource view */ - size_t depth; /**< Depth of the resource view */ - unsigned int firstMipmapLevel; /**< First defined mipmap level */ - unsigned int lastMipmapLevel; /**< Last defined mipmap level */ - unsigned int firstLayer; /**< First layer index */ - unsigned int lastLayer; /**< Last layer index */ - unsigned int reserved[16]; +typedef struct CUDA_RESOURCE_VIEW_DESC_st { + CUresourceViewFormat format; /**< Resource view format */ + size_t width; /**< Width of the resource view */ + size_t height; /**< Height of the resource view */ + size_t depth; /**< Depth of the resource view */ + unsigned int firstMipmapLevel; /**< First defined mipmap level */ + unsigned int lastMipmapLevel; /**< Last defined mipmap level */ + unsigned int firstLayer; /**< First layer index */ + unsigned int lastLayer; /**< Last layer index */ + unsigned int reserved[16]; } CUDA_RESOURCE_VIEW_DESC; /** * GPU Direct v3 tokens */ typedef struct CUDA_POINTER_ATTRIBUTE_P2P_TOKENS_st { - unsigned long long p2pToken; - unsigned int vaSpaceToken; + unsigned long long p2pToken; + unsigned int vaSpaceToken; } CUDA_POINTER_ATTRIBUTE_P2P_TOKENS; #endif /* __CUDA_API_VERSION >= 5000 */ @@ -2014,16 +2234,17 @@ typedef struct CUDA_POINTER_ATTRIBUTE_P2P_TOKENS_st { * Kernel launch parameters */ typedef struct CUDA_LAUNCH_PARAMS_st { - CUfunction function; /**< Kernel to launch */ - unsigned int gridDimX; /**< Width of grid in blocks */ - unsigned int gridDimY; /**< Height of grid in blocks */ - unsigned int gridDimZ; /**< Depth of grid in blocks */ - unsigned int blockDimX; /**< X dimension of each thread block */ - unsigned int blockDimY; /**< Y dimension of each thread block */ - unsigned int blockDimZ; /**< Z dimension of each thread block */ - unsigned int sharedMemBytes; /**< Dynamic shared-memory size per thread block in bytes */ - CUstream hStream; /**< Stream identifier */ - void **kernelParams; /**< Array of pointers to kernel parameters */ + CUfunction function; /**< Kernel to launch */ + unsigned int gridDimX; /**< Width of grid in blocks */ + unsigned int gridDimY; /**< Height of grid in blocks */ + unsigned int gridDimZ; /**< Depth of grid in blocks */ + unsigned int blockDimX; /**< X dimension of each thread block */ + unsigned int blockDimY; /**< Y dimension of each thread block */ + unsigned int blockDimZ; /**< Z dimension of each thread block */ + unsigned int sharedMemBytes; /**< Dynamic shared-memory size per thread block + in bytes */ + CUstream hStream; /**< Stream identifier */ + void **kernelParams; /**< Array of pointers to kernel parameters */ } CUDA_LAUNCH_PARAMS; #endif /* __CUDA_API_VERSION >= 9000 */ @@ -2034,277 +2255,278 @@ typedef struct CUDA_LAUNCH_PARAMS_st { * External memory handle types */ typedef enum CUexternalMemoryHandleType_enum { - /** - * Handle is an opaque file descriptor - */ - CU_EXTERNAL_MEMORY_HANDLE_TYPE_OPAQUE_FD = 1, - /** - * Handle is an opaque shared NT handle - */ - CU_EXTERNAL_MEMORY_HANDLE_TYPE_OPAQUE_WIN32 = 2, - /** - * Handle is an opaque, globally shared handle - */ - CU_EXTERNAL_MEMORY_HANDLE_TYPE_OPAQUE_WIN32_KMT = 3, - /** - * Handle is a D3D12 heap object - */ - CU_EXTERNAL_MEMORY_HANDLE_TYPE_D3D12_HEAP = 4, - /** - * Handle is a D3D12 committed resource - */ - CU_EXTERNAL_MEMORY_HANDLE_TYPE_D3D12_RESOURCE = 5 + /** + * Handle is an opaque file descriptor + */ + CU_EXTERNAL_MEMORY_HANDLE_TYPE_OPAQUE_FD = 1, + /** + * Handle is an opaque shared NT handle + */ + CU_EXTERNAL_MEMORY_HANDLE_TYPE_OPAQUE_WIN32 = 2, + /** + * Handle is an opaque, globally shared handle + */ + CU_EXTERNAL_MEMORY_HANDLE_TYPE_OPAQUE_WIN32_KMT = 3, + /** + * Handle is a D3D12 heap object + */ + CU_EXTERNAL_MEMORY_HANDLE_TYPE_D3D12_HEAP = 4, + /** + * Handle is a D3D12 committed resource + */ + CU_EXTERNAL_MEMORY_HANDLE_TYPE_D3D12_RESOURCE = 5 } CUexternalMemoryHandleType; /** * Indicates that the external memory object is a dedicated resource */ -#define CUDA_EXTERNAL_MEMORY_DEDICATED 0x1 +#define CUDA_EXTERNAL_MEMORY_DEDICATED 0x1 /** * External memory handle descriptor */ typedef struct CUDA_EXTERNAL_MEMORY_HANDLE_DESC_st { - /** - * Type of the handle - */ - CUexternalMemoryHandleType type; - union { - /** - * File descriptor referencing the memory object. Valid - * when type is - * ::CU_EXTERNAL_MEMORY_HANDLE_TYPE_OPAQUE_FD - */ - int fd; - /** - * Win32 handle referencing the semaphore object. Valid when - * type is one of the following: - * - ::CU_EXTERNAL_MEMORY_HANDLE_TYPE_OPAQUE_WIN32 - * - ::CU_EXTERNAL_MEMORY_HANDLE_TYPE_OPAQUE_WIN32_KMT - * - ::CU_EXTERNAL_MEMORY_HANDLE_TYPE_D3D12_HEAP - * - ::CU_EXTERNAL_MEMORY_HANDLE_TYPE_D3D12_RESOURCE - * Exactly one of 'handle' and 'name' must be non-NULL. If - * type is - * ::CU_EXTERNAL_SEMAPHORE_HANDLE_TYPE_OPAQUE_WIN32_KMT - * then 'name' must be NULL. - */ - struct { - /** - * Valid NT handle. Must be NULL if 'name' is non-NULL - */ - void *handle; - /** - * Name of a valid memory object. - * Must be NULL if 'handle' is non-NULL. - */ - const void *name; - } win32; - } handle; - /** - * Size of the memory allocation + /** + * Type of the handle + */ + CUexternalMemoryHandleType type; + union { + /** + * File descriptor referencing the memory object. Valid + * when type is + * ::CU_EXTERNAL_MEMORY_HANDLE_TYPE_OPAQUE_FD + */ + int fd; + /** + * Win32 handle referencing the semaphore object. Valid when + * type is one of the following: + * - ::CU_EXTERNAL_MEMORY_HANDLE_TYPE_OPAQUE_WIN32 + * - ::CU_EXTERNAL_MEMORY_HANDLE_TYPE_OPAQUE_WIN32_KMT + * - ::CU_EXTERNAL_MEMORY_HANDLE_TYPE_D3D12_HEAP + * - ::CU_EXTERNAL_MEMORY_HANDLE_TYPE_D3D12_RESOURCE + * Exactly one of 'handle' and 'name' must be non-NULL. If + * type is + * ::CU_EXTERNAL_SEMAPHORE_HANDLE_TYPE_OPAQUE_WIN32_KMT + * then 'name' must be NULL. */ - unsigned long long size; - /** - * Flags must either be zero or ::CUDA_EXTERNAL_MEMORY_DEDICATED - */ - unsigned int flags; - unsigned int reserved[16]; + struct { + /** + * Valid NT handle. Must be NULL if 'name' is non-NULL + */ + void *handle; + /** + * Name of a valid memory object. + * Must be NULL if 'handle' is non-NULL. + */ + const void *name; + } win32; + } handle; + /** + * Size of the memory allocation + */ + unsigned long long size; + /** + * Flags must either be zero or ::CUDA_EXTERNAL_MEMORY_DEDICATED + */ + unsigned int flags; + unsigned int reserved[16]; } CUDA_EXTERNAL_MEMORY_HANDLE_DESC; /** * External memory buffer descriptor */ typedef struct CUDA_EXTERNAL_MEMORY_BUFFER_DESC_st { - /** - * Offset into the memory object where the buffer's base is - */ - unsigned long long offset; - /** - * Size of the buffer - */ - unsigned long long size; - /** - * Flags reserved for future use. Must be zero. - */ - unsigned int flags; - unsigned int reserved[16]; + /** + * Offset into the memory object where the buffer's base is + */ + unsigned long long offset; + /** + * Size of the buffer + */ + unsigned long long size; + /** + * Flags reserved for future use. Must be zero. + */ + unsigned int flags; + unsigned int reserved[16]; } CUDA_EXTERNAL_MEMORY_BUFFER_DESC; /** * External memory mipmap descriptor */ typedef struct CUDA_EXTERNAL_MEMORY_MIPMAPPED_ARRAY_DESC_st { - /** - * Offset into the memory object where the base level of the - * mipmap chain is. - */ - unsigned long long offset; - /** - * Format, dimension and type of base level of the mipmap chain - */ - CUDA_ARRAY3D_DESCRIPTOR arrayDesc; - /** - * Total number of levels in the mipmap chain - */ - unsigned int numLevels; - unsigned int reserved[16]; + /** + * Offset into the memory object where the base level of the + * mipmap chain is. + */ + unsigned long long offset; + /** + * Format, dimension and type of base level of the mipmap chain + */ + CUDA_ARRAY3D_DESCRIPTOR arrayDesc; + /** + * Total number of levels in the mipmap chain + */ + unsigned int numLevels; + unsigned int reserved[16]; } CUDA_EXTERNAL_MEMORY_MIPMAPPED_ARRAY_DESC; /** * External semaphore handle types */ typedef enum CUexternalSemaphoreHandleType_enum { - /** - * Handle is an opaque file descriptor - */ - CU_EXTERNAL_SEMAPHORE_HANDLE_TYPE_OPAQUE_FD = 1, - /** - * Handle is an opaque shared NT handle - */ - CU_EXTERNAL_SEMAPHORE_HANDLE_TYPE_OPAQUE_WIN32 = 2, - /** - * Handle is an opaque, globally shared handle - */ - CU_EXTERNAL_SEMAPHORE_HANDLE_TYPE_OPAQUE_WIN32_KMT = 3, - /** - * Handle is a shared NT handle referencing a D3D12 fence object - */ - CU_EXTERNAL_SEMAPHORE_HANDLE_TYPE_D3D12_FENCE = 4 + /** + * Handle is an opaque file descriptor + */ + CU_EXTERNAL_SEMAPHORE_HANDLE_TYPE_OPAQUE_FD = 1, + /** + * Handle is an opaque shared NT handle + */ + CU_EXTERNAL_SEMAPHORE_HANDLE_TYPE_OPAQUE_WIN32 = 2, + /** + * Handle is an opaque, globally shared handle + */ + CU_EXTERNAL_SEMAPHORE_HANDLE_TYPE_OPAQUE_WIN32_KMT = 3, + /** + * Handle is a shared NT handle referencing a D3D12 fence object + */ + CU_EXTERNAL_SEMAPHORE_HANDLE_TYPE_D3D12_FENCE = 4 } CUexternalSemaphoreHandleType; /** * External semaphore handle descriptor */ typedef struct CUDA_EXTERNAL_SEMAPHORE_HANDLE_DESC_st { - /** - * Type of the handle - */ - CUexternalSemaphoreHandleType type; - union { - /** - * File descriptor referencing the semaphore object. Valid - * when type is - * ::CU_EXTERNAL_SEMAPHORE_HANDLE_TYPE_OPAQUE_FD - */ - int fd; - /** - * Win32 handle referencing the semaphore object. Valid when - * type is one of the following: - * - ::CU_EXTERNAL_SEMAPHORE_HANDLE_TYPE_OPAQUE_WIN32 - * - ::CU_EXTERNAL_SEMAPHORE_HANDLE_TYPE_OPAQUE_WIN32_KMT - * - ::CU_EXTERNAL_SEMAPHORE_HANDLE_TYPE_D3D12_FENCE - * Exactly one of 'handle' and 'name' must be non-NULL. If - * type is - * ::CU_EXTERNAL_SEMAPHORE_HANDLE_TYPE_OPAQUE_WIN32_KMT - * then 'name' must be NULL. - */ - struct { - /** - * Valid NT handle. Must be NULL if 'name' is non-NULL - */ - void *handle; - /** - * Name of a valid synchronization primitive. - * Must be NULL if 'handle' is non-NULL. - */ - const void *name; - } win32; - } handle; - /** - * Flags reserved for the future. Must be zero. + /** + * Type of the handle + */ + CUexternalSemaphoreHandleType type; + union { + /** + * File descriptor referencing the semaphore object. Valid + * when type is + * ::CU_EXTERNAL_SEMAPHORE_HANDLE_TYPE_OPAQUE_FD + */ + int fd; + /** + * Win32 handle referencing the semaphore object. Valid when + * type is one of the following: + * - ::CU_EXTERNAL_SEMAPHORE_HANDLE_TYPE_OPAQUE_WIN32 + * - ::CU_EXTERNAL_SEMAPHORE_HANDLE_TYPE_OPAQUE_WIN32_KMT + * - ::CU_EXTERNAL_SEMAPHORE_HANDLE_TYPE_D3D12_FENCE + * Exactly one of 'handle' and 'name' must be non-NULL. If + * type is + * ::CU_EXTERNAL_SEMAPHORE_HANDLE_TYPE_OPAQUE_WIN32_KMT + * then 'name' must be NULL. */ - unsigned int flags; - unsigned int reserved[16]; + struct { + /** + * Valid NT handle. Must be NULL if 'name' is non-NULL + */ + void *handle; + /** + * Name of a valid synchronization primitive. + * Must be NULL if 'handle' is non-NULL. + */ + const void *name; + } win32; + } handle; + /** + * Flags reserved for the future. Must be zero. + */ + unsigned int flags; + unsigned int reserved[16]; } CUDA_EXTERNAL_SEMAPHORE_HANDLE_DESC; /** * External semaphore signal parameters */ typedef struct CUDA_EXTERNAL_SEMAPHORE_SIGNAL_PARAMS_st { - struct { - /** - * Parameters for fence objects - */ - struct { - /** - * Value of fence to be signaled - */ - unsigned long long value; - } fence; - unsigned int reserved[16]; - } params; + struct { /** - * Flags reserved for the future. Must be zero. + * Parameters for fence objects */ - unsigned int flags; + struct { + /** + * Value of fence to be signaled + */ + unsigned long long value; + } fence; unsigned int reserved[16]; + } params; + /** + * Flags reserved for the future. Must be zero. + */ + unsigned int flags; + unsigned int reserved[16]; } CUDA_EXTERNAL_SEMAPHORE_SIGNAL_PARAMS; /** * External semaphore wait parameters */ typedef struct CUDA_EXTERNAL_SEMAPHORE_WAIT_PARAMS_st { - struct { - /** - * Parameters for fence objects - */ - struct { - /** - * Value of fence to be waited on - */ - unsigned long long value; - } fence; - unsigned int reserved[16]; - } params; + struct { /** - * Flags reserved for the future. Must be zero. + * Parameters for fence objects */ - unsigned int flags; + struct { + /** + * Value of fence to be waited on + */ + unsigned long long value; + } fence; unsigned int reserved[16]; + } params; + /** + * Flags reserved for the future. Must be zero. + */ + unsigned int flags; + unsigned int reserved[16]; } CUDA_EXTERNAL_SEMAPHORE_WAIT_PARAMS; - #endif /* __CUDA_API_VERSION >= 10000 */ /** - * If set, each kernel launched as part of ::cuLaunchCooperativeKernelMultiDevice only - * waits for prior work in the stream corresponding to that GPU to complete before the - * kernel begins execution. + * If set, each kernel launched as part of + * ::cuLaunchCooperativeKernelMultiDevice only waits for prior work in the + * stream corresponding to that GPU to complete before the kernel begins + * execution. */ -#define CUDA_COOPERATIVE_LAUNCH_MULTI_DEVICE_NO_PRE_LAUNCH_SYNC 0x01 +#define CUDA_COOPERATIVE_LAUNCH_MULTI_DEVICE_NO_PRE_LAUNCH_SYNC 0x01 /** * If set, any subsequent work pushed in a stream that participated in a call to - * ::cuLaunchCooperativeKernelMultiDevice will only wait for the kernel launched on - * the GPU corresponding to that stream to complete before it begins execution. + * ::cuLaunchCooperativeKernelMultiDevice will only wait for the kernel launched + * on the GPU corresponding to that stream to complete before it begins + * execution. */ -#define CUDA_COOPERATIVE_LAUNCH_MULTI_DEVICE_NO_POST_LAUNCH_SYNC 0x02 +#define CUDA_COOPERATIVE_LAUNCH_MULTI_DEVICE_NO_POST_LAUNCH_SYNC 0x02 /** - * If set, the CUDA array is a collection of layers, where each layer is either a 1D - * or a 2D array and the Depth member of CUDA_ARRAY3D_DESCRIPTOR specifies the number - * of layers, not the depth of a 3D array. + * If set, the CUDA array is a collection of layers, where each layer is either + * a 1D or a 2D array and the Depth member of CUDA_ARRAY3D_DESCRIPTOR specifies + * the number of layers, not the depth of a 3D array. */ -#define CUDA_ARRAY3D_LAYERED 0x01 +#define CUDA_ARRAY3D_LAYERED 0x01 /** * Deprecated, use CUDA_ARRAY3D_LAYERED */ -#define CUDA_ARRAY3D_2DARRAY 0x01 +#define CUDA_ARRAY3D_2DARRAY 0x01 /** * This flag must be set in order to bind a surface reference * to the CUDA array */ -#define CUDA_ARRAY3D_SURFACE_LDST 0x02 +#define CUDA_ARRAY3D_SURFACE_LDST 0x02 /** - * If set, the CUDA array is a collection of six 2D arrays, representing faces of a cube. The - * width of such a CUDA array must be equal to its height, and Depth must be six. - * If ::CUDA_ARRAY3D_LAYERED flag is also set, then the CUDA array is a collection of cubemaps - * and Depth must be a multiple of six. + * If set, the CUDA array is a collection of six 2D arrays, representing faces + * of a cube. The width of such a CUDA array must be equal to its height, and + * Depth must be six. If ::CUDA_ARRAY3D_LAYERED flag is also set, then the CUDA + * array is a collection of cubemaps and Depth must be a multiple of six. */ -#define CUDA_ARRAY3D_CUBEMAP 0x04 +#define CUDA_ARRAY3D_CUBEMAP 0x04 /** * This flag must be set in order to perform texture gather operations @@ -2335,25 +2557,25 @@ typedef struct CUDA_EXTERNAL_SEMAPHORE_WAIT_PARAMS_st { * in the range [0,1]. * Flag for ::cuTexRefSetFlags() */ -#define CU_TRSF_READ_AS_INTEGER 0x01 +#define CU_TRSF_READ_AS_INTEGER 0x01 /** * Use normalized texture coordinates in the range [0,1) instead of [0,dim). * Flag for ::cuTexRefSetFlags() */ -#define CU_TRSF_NORMALIZED_COORDINATES 0x02 +#define CU_TRSF_NORMALIZED_COORDINATES 0x02 /** * Perform sRGB->linear conversion during texture read. * Flag for ::cuTexRefSetFlags() */ -#define CU_TRSF_SRGB 0x10 +#define CU_TRSF_SRGB 0x10 /** * End of array terminator for the \p extra parameter to * ::cuLaunchKernel */ -#define CU_LAUNCH_PARAM_END ((void*)0x00) +#define CU_LAUNCH_PARAM_END ((void *)0x00) /** * Indicator that the next value in the \p extra parameter to @@ -2364,7 +2586,7 @@ typedef struct CUDA_EXTERNAL_SEMAPHORE_WAIT_PARAMS_st { * \p extra array, then ::CU_LAUNCH_PARAM_BUFFER_POINTER will have no * effect. */ -#define CU_LAUNCH_PARAM_BUFFER_POINTER ((void*)0x01) +#define CU_LAUNCH_PARAM_BUFFER_POINTER ((void *)0x01) /** * Indicator that the next value in the \p extra parameter to @@ -2374,7 +2596,7 @@ typedef struct CUDA_EXTERNAL_SEMAPHORE_WAIT_PARAMS_st { * in the \p extra array if the value associated with * ::CU_LAUNCH_PARAM_BUFFER_SIZE is not zero. */ -#define CU_LAUNCH_PARAM_BUFFER_SIZE ((void*)0x02) +#define CU_LAUNCH_PARAM_BUFFER_SIZE ((void *)0x02) /** * For texture references loaded into the module, use default texunit from @@ -2385,12 +2607,12 @@ typedef struct CUDA_EXTERNAL_SEMAPHORE_WAIT_PARAMS_st { /** * Device that represents the CPU */ -#define CU_DEVICE_CPU ((CUdevice)-1) +#define CU_DEVICE_CPU ((CUdevice)-1) /** * Device that represents an invalid device */ -#define CU_DEVICE_INVALID ((CUdevice)-2) +#define CU_DEVICE_INVALID ((CUdevice)-2) /** @} */ /* END CUDA_TYPES */ @@ -2684,7 +2906,8 @@ CUresult CUDAAPI cuDeviceGetUuid(CUuuid *uuid, CUdevice dev); * ::cuDeviceTotalMem, * ::cudaGetDeviceProperties */ -CUresult CUDAAPI cuDeviceGetLuid(char *luid, unsigned int *deviceNodeMask, CUdevice dev); +CUresult CUDAAPI cuDeviceGetLuid(char *luid, unsigned int *deviceNodeMask, + CUdevice dev); #endif #if __CUDA_API_VERSION >= 3020 @@ -2841,8 +3064,8 @@ CUresult CUDAAPI cuDeviceTotalMem(size_t *bytes, CUdevice dev); * can have multiple CUDA contexts present at a single time. * - ::CU_COMPUTEMODE_PROHIBITED: Compute-prohibited mode - Device is * prohibited from creating new CUDA contexts. - * - ::CU_COMPUTEMODE_EXCLUSIVE_PROCESS: Compute-exclusive-process mode - Device - * can have only one context used by a single process at a time. + * - ::CU_COMPUTEMODE_EXCLUSIVE_PROCESS: Compute-exclusive-process mode - + * Device can have only one context used by a single process at a time. * - ::CU_DEVICE_ATTRIBUTE_CONCURRENT_KERNELS: 1 if the device supports * executing multiple kernels within the same context simultaneously, or 0 if * not. It is not guaranteed that multiple kernels will be resident @@ -2851,51 +3074,66 @@ CUresult CUDAAPI cuDeviceTotalMem(size_t *bytes, CUdevice dev); * - ::CU_DEVICE_ATTRIBUTE_ECC_ENABLED: 1 if error correction is enabled on the * device, 0 if error correction is disabled or not supported by the device; * - ::CU_DEVICE_ATTRIBUTE_PCI_BUS_ID: PCI bus identifier of the device; - * - ::CU_DEVICE_ATTRIBUTE_PCI_DEVICE_ID: PCI device (also known as slot) identifier - * of the device; + * - ::CU_DEVICE_ATTRIBUTE_PCI_DEVICE_ID: PCI device (also known as slot) + * identifier of the device; * - ::CU_DEVICE_ATTRIBUTE_PCI_DOMAIN_ID: PCI domain identifier of the device - * - ::CU_DEVICE_ATTRIBUTE_TCC_DRIVER: 1 if the device is using a TCC driver. TCC - * is only available on Tesla hardware running Windows Vista or later; - * - ::CU_DEVICE_ATTRIBUTE_MEMORY_CLOCK_RATE: Peak memory clock frequency in kilohertz; - * - ::CU_DEVICE_ATTRIBUTE_GLOBAL_MEMORY_BUS_WIDTH: Global memory bus width in bits; - * - ::CU_DEVICE_ATTRIBUTE_L2_CACHE_SIZE: Size of L2 cache in bytes. 0 if the device doesn't have L2 cache; - * - ::CU_DEVICE_ATTRIBUTE_MAX_THREADS_PER_MULTIPROCESSOR: Maximum resident threads per multiprocessor; - * - ::CU_DEVICE_ATTRIBUTE_UNIFIED_ADDRESSING: 1 if the device shares a unified address space with - * the host, or 0 if not; - * - ::CU_DEVICE_ATTRIBUTE_COMPUTE_CAPABILITY_MAJOR: Major compute capability version number; - * - ::CU_DEVICE_ATTRIBUTE_COMPUTE_CAPABILITY_MINOR: Minor compute capability version number; - * - ::CU_DEVICE_ATTRIBUTE_GLOBAL_L1_CACHE_SUPPORTED: 1 if device supports caching globals - * in L1 cache, 0 if caching globals in L1 cache is not supported by the device; - * - ::CU_DEVICE_ATTRIBUTE_LOCAL_L1_CACHE_SUPPORTED: 1 if device supports caching locals - * in L1 cache, 0 if caching locals in L1 cache is not supported by the device; - * - ::CU_DEVICE_ATTRIBUTE_MAX_SHARED_MEMORY_PER_MULTIPROCESSOR: Maximum amount of - * shared memory available to a multiprocessor in bytes; this amount is shared - * by all thread blocks simultaneously resident on a multiprocessor; - * - ::CU_DEVICE_ATTRIBUTE_MAX_REGISTERS_PER_MULTIPROCESSOR: Maximum number of 32-bit - * registers available to a multiprocessor; this number is shared by all thread - * blocks simultaneously resident on a multiprocessor; - * - ::CU_DEVICE_ATTRIBUTE_MANAGED_MEMORY: 1 if device supports allocating managed memory - * on this system, 0 if allocating managed memory is not supported by the device on this system. - * - ::CU_DEVICE_ATTRIBUTE_MULTI_GPU_BOARD: 1 if device is on a multi-GPU board, 0 if not. - * - ::CU_DEVICE_ATTRIBUTE_MULTI_GPU_BOARD_GROUP_ID: Unique identifier for a group of devices - * associated with the same board. Devices on the same multi-GPU board will share the same identifier. - * - ::CU_DEVICE_ATTRIBUTE_HOST_NATIVE_ATOMIC_SUPPORTED: 1 if Link between the device and the host - * supports native atomic operations. - * - ::CU_DEVICE_ATTRIBUTE_SINGLE_TO_DOUBLE_PRECISION_PERF_RATIO: Ratio of single precision performance - * (in floating-point operations per second) to double precision performance. - * - ::CU_DEVICE_ATTRIBUTE_PAGEABLE_MEMORY_ACCESS: Device suppports coherently accessing - * pageable memory without calling cudaHostRegister on it. - * - ::CU_DEVICE_ATTRIBUTE_CONCURRENT_MANAGED_ACCESS: Device can coherently access managed memory - * concurrently with the CPU. - * - ::CU_DEVICE_ATTRIBUTE_COMPUTE_PREEMPTION_SUPPORTED: Device supports Compute Preemption. - * - ::CU_DEVICE_ATTRIBUTE_CAN_USE_HOST_POINTER_FOR_REGISTERED_MEM: Device can access host registered - * memory at the same virtual address as the CPU. - * - ::CU_DEVICE_ATTRIBUTE_MAX_SHARED_MEMORY_PER_BLOCK_OPTIN: The maximum per block shared memory size - * suported on this device. This is the maximum value that can be opted into when using the cuFuncSetAttribute() call. - * For more details see ::CU_FUNC_ATTRIBUTE_MAX_DYNAMIC_SHARED_SIZE_BYTES - * - ::CU_DEVICE_ATTRIBUTE_PAGEABLE_MEMORY_ACCESS_USES_HOST_PAGE_TABLES: Device accesses pageable memory via the host's - * page tables. - * - ::CU_DEVICE_ATTRIBUTE_DIRECT_MANAGED_MEM_ACCESS_FROM_HOST: The host can directly access managed memory on the device without migration. + * - ::CU_DEVICE_ATTRIBUTE_TCC_DRIVER: 1 if the device is using a TCC driver. + * TCC is only available on Tesla hardware running Windows Vista or later; + * - ::CU_DEVICE_ATTRIBUTE_MEMORY_CLOCK_RATE: Peak memory clock frequency in + * kilohertz; + * - ::CU_DEVICE_ATTRIBUTE_GLOBAL_MEMORY_BUS_WIDTH: Global memory bus width in + * bits; + * - ::CU_DEVICE_ATTRIBUTE_L2_CACHE_SIZE: Size of L2 cache in bytes. 0 if the + * device doesn't have L2 cache; + * - ::CU_DEVICE_ATTRIBUTE_MAX_THREADS_PER_MULTIPROCESSOR: Maximum resident + * threads per multiprocessor; + * - ::CU_DEVICE_ATTRIBUTE_UNIFIED_ADDRESSING: 1 if the device shares a unified + * address space with the host, or 0 if not; + * - ::CU_DEVICE_ATTRIBUTE_COMPUTE_CAPABILITY_MAJOR: Major compute capability + * version number; + * - ::CU_DEVICE_ATTRIBUTE_COMPUTE_CAPABILITY_MINOR: Minor compute capability + * version number; + * - ::CU_DEVICE_ATTRIBUTE_GLOBAL_L1_CACHE_SUPPORTED: 1 if device supports + * caching globals in L1 cache, 0 if caching globals in L1 cache is not + * supported by the device; + * - ::CU_DEVICE_ATTRIBUTE_LOCAL_L1_CACHE_SUPPORTED: 1 if device supports + * caching locals in L1 cache, 0 if caching locals in L1 cache is not supported + * by the device; + * - ::CU_DEVICE_ATTRIBUTE_MAX_SHARED_MEMORY_PER_MULTIPROCESSOR: Maximum amount + * of shared memory available to a multiprocessor in bytes; this amount is + * shared by all thread blocks simultaneously resident on a multiprocessor; + * - ::CU_DEVICE_ATTRIBUTE_MAX_REGISTERS_PER_MULTIPROCESSOR: Maximum number of + * 32-bit registers available to a multiprocessor; this number is shared by all + * thread blocks simultaneously resident on a multiprocessor; + * - ::CU_DEVICE_ATTRIBUTE_MANAGED_MEMORY: 1 if device supports allocating + * managed memory on this system, 0 if allocating managed memory is not + * supported by the device on this system. + * - ::CU_DEVICE_ATTRIBUTE_MULTI_GPU_BOARD: 1 if device is on a multi-GPU board, + * 0 if not. + * - ::CU_DEVICE_ATTRIBUTE_MULTI_GPU_BOARD_GROUP_ID: Unique identifier for a + * group of devices associated with the same board. Devices on the same + * multi-GPU board will share the same identifier. + * - ::CU_DEVICE_ATTRIBUTE_HOST_NATIVE_ATOMIC_SUPPORTED: 1 if Link between the + * device and the host supports native atomic operations. + * - ::CU_DEVICE_ATTRIBUTE_SINGLE_TO_DOUBLE_PRECISION_PERF_RATIO: Ratio of + * single precision performance (in floating-point operations per second) to + * double precision performance. + * - ::CU_DEVICE_ATTRIBUTE_PAGEABLE_MEMORY_ACCESS: Device suppports coherently + * accessing pageable memory without calling cudaHostRegister on it. + * - ::CU_DEVICE_ATTRIBUTE_CONCURRENT_MANAGED_ACCESS: Device can coherently + * access managed memory concurrently with the CPU. + * - ::CU_DEVICE_ATTRIBUTE_COMPUTE_PREEMPTION_SUPPORTED: Device supports Compute + * Preemption. + * - ::CU_DEVICE_ATTRIBUTE_CAN_USE_HOST_POINTER_FOR_REGISTERED_MEM: Device can + * access host registered memory at the same virtual address as the CPU. + * - ::CU_DEVICE_ATTRIBUTE_MAX_SHARED_MEMORY_PER_BLOCK_OPTIN: The maximum per + * block shared memory size suported on this device. This is the maximum value + * that can be opted into when using the cuFuncSetAttribute() call. For more + * details see ::CU_FUNC_ATTRIBUTE_MAX_DYNAMIC_SHARED_SIZE_BYTES + * - ::CU_DEVICE_ATTRIBUTE_PAGEABLE_MEMORY_ACCESS_USES_HOST_PAGE_TABLES: Device + * accesses pageable memory via the host's page tables. + * - ::CU_DEVICE_ATTRIBUTE_DIRECT_MANAGED_MEM_ACCESS_FROM_HOST: The host can + * directly access managed memory on the device without migration. * * \param pi - Returned device attribute value * \param attrib - Device attribute to query @@ -2919,7 +3157,8 @@ CUresult CUDAAPI cuDeviceTotalMem(size_t *bytes, CUdevice dev); * ::cudaDeviceGetAttribute, * ::cudaGetDeviceProperties */ -CUresult CUDAAPI cuDeviceGetAttribute(int *pi, CUdevice_attribute attrib, CUdevice dev); +CUresult CUDAAPI cuDeviceGetAttribute(int *pi, CUdevice_attribute attrib, + CUdevice dev); /** @} */ /* END CUDA_DEVICE */ @@ -2940,7 +3179,8 @@ CUresult CUDAAPI cuDeviceGetAttribute(int *pi, CUdevice_attribute attrib, CUdevi * * \deprecated * - * This function was deprecated as of CUDA 5.0 and replaced by ::cuDeviceGetAttribute(). + * This function was deprecated as of CUDA 5.0 and replaced by + ::cuDeviceGetAttribute(). * * Returns in \p *prop the properties of device \p dev. The ::CUdevprop * structure is defined as: @@ -2997,7 +3237,8 @@ CUresult CUDAAPI cuDeviceGetAttribute(int *pi, CUdevice_attribute attrib, CUdevi * ::cuDeviceGet, * ::cuDeviceTotalMem */ -__CUDA_DEPRECATED CUresult CUDAAPI cuDeviceGetProperties(CUdevprop *prop, CUdevice dev); +__CUDA_DEPRECATED CUresult CUDAAPI cuDeviceGetProperties(CUdevprop *prop, + CUdevice dev); /** * \brief Returns the compute capability of the device @@ -3031,21 +3272,23 @@ __CUDA_DEPRECATED CUresult CUDAAPI cuDeviceGetProperties(CUdevprop *prop, CUdevi * ::cuDeviceGet, * ::cuDeviceTotalMem */ -__CUDA_DEPRECATED CUresult CUDAAPI cuDeviceComputeCapability(int *major, int *minor, CUdevice dev); +__CUDA_DEPRECATED CUresult CUDAAPI cuDeviceComputeCapability(int *major, + int *minor, + CUdevice dev); /** @} */ /* END CUDA_DEVICE_DEPRECATED */ /** * \defgroup CUDA_PRIMARY_CTX Primary Context Management * - * ___MANBRIEF___ primary context management functions of the low-level CUDA driver - * API (___CURRENT_FILE___) ___ENDMANBRIEF___ + * ___MANBRIEF___ primary context management functions of the low-level CUDA + * driver API (___CURRENT_FILE___) ___ENDMANBRIEF___ * - * This section describes the primary context management functions of the low-level - * CUDA driver application programming interface. + * This section describes the primary context management functions of the + * low-level CUDA driver application programming interface. * - * The primary context is unique per device and shared with the CUDA runtime API. - * These functions allow integration with other libraries using CUDA. + * The primary context is unique per device and shared with the CUDA runtime + * API. These functions allow integration with other libraries using CUDA. * * @{ */ @@ -3058,18 +3301,18 @@ __CUDA_DEPRECATED CUresult CUDAAPI cuDeviceComputeCapability(int *major, int *mi * Retains the primary context on the device, creating it if necessary, * increasing its usage count. The caller must call * ::cuDevicePrimaryCtxRelease() when done using the context. - * Unlike ::cuCtxCreate() the newly created context is not pushed onto the stack. + * Unlike ::cuCtxCreate() the newly created context is not pushed onto the + * stack. * * Context creation will fail with ::CUDA_ERROR_UNKNOWN if the compute mode of - * the device is ::CU_COMPUTEMODE_PROHIBITED. The function ::cuDeviceGetAttribute() - * can be used with ::CU_DEVICE_ATTRIBUTE_COMPUTE_MODE to determine the compute mode - * of the device. - * The nvidia-smi tool can be used to set the compute mode for - * devices. Documentation for nvidia-smi can be obtained by passing a - * -h option to it. + * the device is ::CU_COMPUTEMODE_PROHIBITED. The function + * ::cuDeviceGetAttribute() can be used with ::CU_DEVICE_ATTRIBUTE_COMPUTE_MODE + * to determine the compute mode of the device. The nvidia-smi tool can + * be used to set the compute mode for devices. Documentation for + * nvidia-smi can be obtained by passing a -h option to it. * - * Please note that the primary context always supports pinned allocations. Other - * flags can be specified by ::cuDevicePrimaryCtxSetFlags(). + * Please note that the primary context always supports pinned allocations. + * Other flags can be specified by ::cuDevicePrimaryCtxSetFlags(). * * \param pctx - Returned context handle of the new context * \param dev - Device for which primary context is requested @@ -3224,7 +3467,8 @@ CUresult CUDAAPI cuDevicePrimaryCtxSetFlags(CUdevice dev, unsigned int flags); * ::cuCtxGetFlags, * ::cudaGetDeviceFlags */ -CUresult CUDAAPI cuDevicePrimaryCtxGetState(CUdevice dev, unsigned int *flags, int *active); +CUresult CUDAAPI cuDevicePrimaryCtxGetState(CUdevice dev, unsigned int *flags, + int *active); /** * \brief Destroy all allocations and reset all state on the primary context @@ -3268,7 +3512,6 @@ CUresult CUDAAPI cuDevicePrimaryCtxReset(CUdevice dev); /** @} */ /* END CUDA_PRIMARY_CTX */ - /** * \defgroup CUDA_CTX Context Management * @@ -3294,8 +3537,8 @@ CUresult CUDAAPI cuDevicePrimaryCtxReset(CUdevice dev); * \p flags parameter is described below. The context is created with a usage * count of 1 and the caller of ::cuCtxCreate() must call ::cuCtxDestroy() * when done using the context. If a context is already current to the thread, - * it is supplanted by the newly created context and may be restored by a subsequent - * call to ::cuCtxPopCurrent(). + * it is supplanted by the newly created context and may be restored by a + * subsequent call to ::cuCtxPopCurrent(). * * The three LSBs of the \p flags parameter can be used to control how the OS * thread, which owns the CUDA context at the time of an API call, interacts @@ -3340,12 +3583,11 @@ CUresult CUDAAPI cuDevicePrimaryCtxReset(CUdevice dev); * memory usage at the cost of potentially increased memory usage. * * Context creation will fail with ::CUDA_ERROR_UNKNOWN if the compute mode of - * the device is ::CU_COMPUTEMODE_PROHIBITED. The function ::cuDeviceGetAttribute() - * can be used with ::CU_DEVICE_ATTRIBUTE_COMPUTE_MODE to determine the - * compute mode of the device. The nvidia-smi tool can be used to set - * the compute mode for * devices. - * Documentation for nvidia-smi can be obtained by passing a - * -h option to it. + * the device is ::CU_COMPUTEMODE_PROHIBITED. The function + * ::cuDeviceGetAttribute() can be used with ::CU_DEVICE_ATTRIBUTE_COMPUTE_MODE + * to determine the compute mode of the device. The nvidia-smi tool can + * be used to set the compute mode for * devices. Documentation for + * nvidia-smi can be obtained by passing a -h option to it. * * \param pctx - Returned context handle of the new context * \param flags - Context creation flags @@ -3692,9 +3934,9 @@ CUresult CUDAAPI cuCtxSynchronize(void); * than 3.5 will result in the error ::CUDA_ERROR_UNSUPPORTED_LIMIT being * returned. * - * - ::CU_LIMIT_MAX_L2_FETCH_GRANULARITY controls the L2 cache fetch granularity. - * Values can range from 0B to 128B. This is purely a performance hint and - * it can be ignored or clamped depending on the platform. + * - ::CU_LIMIT_MAX_L2_FETCH_GRANULARITY controls the L2 cache fetch + * granularity. Values can range from 0B to 128B. This is purely a performance + * hint and it can be ignored or clamped depending on the platform. * * \param limit - Limit to set * \param value - Size of limit @@ -3767,17 +4009,19 @@ CUresult CUDAAPI cuCtxGetLimit(size_t *pvalue, CUlimit limit); * \brief Returns the preferred cache configuration for the current context. * * On devices where the L1 cache and shared memory use the same hardware - * resources, this function returns through \p pconfig the preferred cache configuration - * for the current context. This is only a preference. The driver will use - * the requested configuration if possible, but it is free to choose a different - * configuration if required to execute functions. + * resources, this function returns through \p pconfig the preferred cache + * configuration for the current context. This is only a preference. The driver + * will use the requested configuration if possible, but it is free to choose a + * different configuration if required to execute functions. * * This will return a \p pconfig of ::CU_FUNC_CACHE_PREFER_NONE on devices * where the size of the L1 cache and shared memory are fixed. * * The supported cache configurations are: - * - ::CU_FUNC_CACHE_PREFER_NONE: no preference for shared memory or L1 (default) - * - ::CU_FUNC_CACHE_PREFER_SHARED: prefer larger shared memory and smaller L1 cache + * - ::CU_FUNC_CACHE_PREFER_NONE: no preference for shared memory or L1 + * (default) + * - ::CU_FUNC_CACHE_PREFER_SHARED: prefer larger shared memory and smaller L1 + * cache * - ::CU_FUNC_CACHE_PREFER_L1: prefer larger L1 cache and smaller shared memory * - ::CU_FUNC_CACHE_PREFER_EQUAL: prefer equal sized L1 cache and shared memory * @@ -3827,8 +4071,10 @@ CUresult CUDAAPI cuCtxGetCacheConfig(CUfunc_cache *pconfig); * preference setting may insert a device-side synchronization point. * * The supported cache configurations are: - * - ::CU_FUNC_CACHE_PREFER_NONE: no preference for shared memory or L1 (default) - * - ::CU_FUNC_CACHE_PREFER_SHARED: prefer larger shared memory and smaller L1 cache + * - ::CU_FUNC_CACHE_PREFER_NONE: no preference for shared memory or L1 + * (default) + * - ::CU_FUNC_CACHE_PREFER_SHARED: prefer larger shared memory and smaller L1 + * cache * - ::CU_FUNC_CACHE_PREFER_L1: prefer larger L1 cache and smaller shared memory * - ::CU_FUNC_CACHE_PREFER_EQUAL: prefer equal sized L1 cache and shared memory * @@ -3860,10 +4106,12 @@ CUresult CUDAAPI cuCtxSetCacheConfig(CUfunc_cache config); #if __CUDA_API_VERSION >= 4020 /** - * \brief Returns the current shared memory configuration for the current context. + * \brief Returns the current shared memory configuration for the current + * context. * - * This function will return in \p pConfig the current size of shared memory banks - * in the current context. On devices with configurable shared memory banks, + * This function will return in \p pConfig the current size of shared memory + * banks in the current context. On devices with configurable shared memory + * banks, * ::cuCtxSetSharedMemConfig can be used to change this setting, so that all * subsequent kernel launches will by default use the new bank size. When * ::cuCtxGetSharedMemConfig is called on devices without configurable shared @@ -3913,19 +4161,19 @@ CUresult CUDAAPI cuCtxGetSharedMemConfig(CUsharedconfig *pConfig); * * Changing the shared memory bank size will not increase shared memory usage * or affect occupancy of kernels, but may have major effects on performance. - * Larger bank sizes will allow for greater potential bandwidth to shared memory, - * but will change what kinds of accesses to shared memory will result in bank - * conflicts. + * Larger bank sizes will allow for greater potential bandwidth to shared + * memory, but will change what kinds of accesses to shared memory will result + * in bank conflicts. * * This function will do nothing on devices with fixed shared memory bank size. * * The supported bank configurations are: - * - ::CU_SHARED_MEM_CONFIG_DEFAULT_BANK_SIZE: set bank width to the default initial - * setting (currently, four bytes). + * - ::CU_SHARED_MEM_CONFIG_DEFAULT_BANK_SIZE: set bank width to the default + * initial setting (currently, four bytes). * - ::CU_SHARED_MEM_CONFIG_FOUR_BYTE_BANK_SIZE: set shared memory bank width to * be natively four bytes. - * - ::CU_SHARED_MEM_CONFIG_EIGHT_BYTE_BANK_SIZE: set shared memory bank width to - * be natively eight bytes. + * - ::CU_SHARED_MEM_CONFIG_EIGHT_BYTE_BANK_SIZE: set shared memory bank width + * to be natively eight bytes. * * \param config - requested shared memory configuration * @@ -3964,9 +4212,9 @@ CUresult CUDAAPI cuCtxSetSharedMemConfig(CUsharedconfig config); * used to create the currently bound context. * * Note that new API versions are only introduced when context capabilities are - * changed that break binary compatibility, so the API version and driver version - * may be different. For example, it is valid for the API version to be 3020 while - * the driver version is 4020. + * changed that break binary compatibility, so the API version and driver + * version may be different. For example, it is valid for the API version to be + * 3020 while the driver version is 4020. * * \param ctx - Context to check * \param version - Pointer to version @@ -3997,26 +4245,25 @@ CUresult CUDAAPI cuCtxGetApiVersion(CUcontext ctx, unsigned int *version); * \brief Returns numerical values that correspond to the least and * greatest stream priorities. * - * Returns in \p *leastPriority and \p *greatestPriority the numerical values that correspond - * to the least and greatest stream priorities respectively. Stream priorities - * follow a convention where lower numbers imply greater priorities. The range of - * meaningful stream priorities is given by [\p *greatestPriority, \p *leastPriority]. - * If the user attempts to create a stream with a priority value that is - * outside the meaningful range as specified by this API, the priority is - * automatically clamped down or up to either \p *leastPriority or \p *greatestPriority - * respectively. See ::cuStreamCreateWithPriority for details on creating a - * priority stream. - * A NULL may be passed in for \p *leastPriority or \p *greatestPriority if the value - * is not desired. + * Returns in \p *leastPriority and \p *greatestPriority the numerical values + * that correspond to the least and greatest stream priorities respectively. + * Stream priorities follow a convention where lower numbers imply greater + * priorities. The range of meaningful stream priorities is given by [\p + * *greatestPriority, \p *leastPriority]. If the user attempts to create a + * stream with a priority value that is outside the meaningful range as + * specified by this API, the priority is automatically clamped down or up to + * either \p *leastPriority or \p *greatestPriority respectively. See + * ::cuStreamCreateWithPriority for details on creating a priority stream. A + * NULL may be passed in for \p *leastPriority or \p *greatestPriority if the + * value is not desired. * - * This function will return '0' in both \p *leastPriority and \p *greatestPriority if - * the current context's device does not support stream priorities - * (see ::cuDeviceGetAttribute). + * This function will return '0' in both \p *leastPriority and \p + * *greatestPriority if the current context's device does not support stream + * priorities (see ::cuDeviceGetAttribute). * - * \param leastPriority - Pointer to an int in which the numerical value for least - * stream priority is returned - * \param greatestPriority - Pointer to an int in which the numerical value for greatest - * stream priority is returned + * \param leastPriority - Pointer to an int in which the numerical value for + * least stream priority is returned \param greatestPriority - Pointer to an int + * in which the numerical value for greatest stream priority is returned * * \return * ::CUDA_SUCCESS, @@ -4031,7 +4278,8 @@ CUresult CUDAAPI cuCtxGetApiVersion(CUcontext ctx, unsigned int *version); * ::cuCtxSynchronize, * ::cudaDeviceGetStreamPriorityRange */ -CUresult CUDAAPI cuCtxGetStreamPriorityRange(int *leastPriority, int *greatestPriority); +CUresult CUDAAPI cuCtxGetStreamPriorityRange(int *leastPriority, + int *greatestPriority); /** @} */ /* END CUDA_CTX */ @@ -4041,8 +4289,8 @@ CUresult CUDAAPI cuCtxGetStreamPriorityRange(int *leastPriority, int *greatestPr * ___MANBRIEF___ deprecated context management functions of the low-level CUDA * driver API (___CURRENT_FILE___) ___ENDMANBRIEF___ * - * This section describes the deprecated context management functions of the low-level - * CUDA driver application programming interface. + * This section describes the deprecated context management functions of the + * low-level CUDA driver application programming interface. * * @{ */ @@ -4086,7 +4334,8 @@ CUresult CUDAAPI cuCtxGetStreamPriorityRange(int *leastPriority, int *greatestPr * ::cuCtxSetLimit, * ::cuCtxSynchronize */ -__CUDA_DEPRECATED CUresult CUDAAPI cuCtxAttach(CUcontext *pctx, unsigned int flags); +__CUDA_DEPRECATED CUresult CUDAAPI cuCtxAttach(CUcontext *pctx, + unsigned int flags); /** * \brief Decrement a context's usage-count @@ -4126,7 +4375,6 @@ __CUDA_DEPRECATED CUresult CUDAAPI cuCtxDetach(CUcontext ctx); /** @} */ /* END CUDA_CTX_DEPRECATED */ - /** * \defgroup CUDA_MODULE Module Management * @@ -4257,7 +4505,9 @@ CUresult CUDAAPI cuModuleLoadData(CUmodule *module, const void *image); * ::cuModuleLoadFatBinary, * ::cuModuleUnload */ -CUresult CUDAAPI cuModuleLoadDataEx(CUmodule *module, const void *image, unsigned int numOptions, CUjit_option *options, void **optionValues); +CUresult CUDAAPI cuModuleLoadDataEx(CUmodule *module, const void *image, + unsigned int numOptions, + CUjit_option *options, void **optionValues); /** * \brief Load a module's data @@ -4354,7 +4604,8 @@ CUresult CUDAAPI cuModuleUnload(CUmodule hmod); * ::cuModuleLoadFatBinary, * ::cuModuleUnload */ -CUresult CUDAAPI cuModuleGetFunction(CUfunction *hfunc, CUmodule hmod, const char *name); +CUresult CUDAAPI cuModuleGetFunction(CUfunction *hfunc, CUmodule hmod, + const char *name); #if __CUDA_API_VERSION >= 3020 /** @@ -4390,7 +4641,8 @@ CUresult CUDAAPI cuModuleGetFunction(CUfunction *hfunc, CUmodule hmod, const cha * ::cudaGetSymbolAddress, * ::cudaGetSymbolSize */ -CUresult CUDAAPI cuModuleGetGlobal(CUdeviceptr *dptr, size_t *bytes, CUmodule hmod, const char *name); +CUresult CUDAAPI cuModuleGetGlobal(CUdeviceptr *dptr, size_t *bytes, + CUmodule hmod, const char *name); #endif /* __CUDA_API_VERSION >= 3020 */ /** @@ -4425,7 +4677,8 @@ CUresult CUDAAPI cuModuleGetGlobal(CUdeviceptr *dptr, size_t *bytes, CUmodule hm * ::cuModuleUnload, * ::cudaGetTextureReference */ -CUresult CUDAAPI cuModuleGetTexRef(CUtexref *pTexRef, CUmodule hmod, const char *name); +CUresult CUDAAPI cuModuleGetTexRef(CUtexref *pTexRef, CUmodule hmod, + const char *name); /** * \brief Returns a handle to a surface reference @@ -4457,7 +4710,8 @@ CUresult CUDAAPI cuModuleGetTexRef(CUtexref *pTexRef, CUmodule hmod, const char * ::cuModuleUnload, * ::cudaGetSurfaceReference */ -CUresult CUDAAPI cuModuleGetSurfRef(CUsurfref *pSurfRef, CUmodule hmod, const char *name); +CUresult CUDAAPI cuModuleGetSurfRef(CUsurfref *pSurfRef, CUmodule hmod, + const char *name); #if __CUDA_API_VERSION >= 5050 @@ -4499,14 +4753,14 @@ CUresult CUDAAPI cuModuleGetSurfRef(CUsurfref *pSurfRef, CUmodule hmod, const ch * ::cuLinkComplete, * ::cuLinkDestroy */ -CUresult CUDAAPI -cuLinkCreate(unsigned int numOptions, CUjit_option *options, void **optionValues, CUlinkState *stateOut); +CUresult CUDAAPI cuLinkCreate(unsigned int numOptions, CUjit_option *options, + void **optionValues, CUlinkState *stateOut); /** * \brief Add an input to a pending linker invocation * - * Ownership of \p data is retained by the caller. No reference is retained to any - * inputs after this call returns. + * Ownership of \p data is retained by the caller. No reference is retained to + * any inputs after this call returns. * * This method accepts only compiler options, which are used if the data must * be compiled from PTX, and does not accept any of @@ -4519,8 +4773,9 @@ cuLinkCreate(unsigned int numOptions, CUjit_option *options, void **optionValues * \param size The length of the input data. * \param name An optional name for this input in log messages. * \param numOptions Size of options. - * \param options Options to be applied only for this input (overrides options from ::cuLinkCreate). - * \param optionValues Array of option values, each cast to void *. + * \param options Options to be applied only for this input (overrides + * options from ::cuLinkCreate). \param optionValues Array of option values, + * each cast to void *. * * \return * ::CUDA_SUCCESS, @@ -4536,9 +4791,10 @@ cuLinkCreate(unsigned int numOptions, CUjit_option *options, void **optionValues * ::cuLinkComplete, * ::cuLinkDestroy */ -CUresult CUDAAPI -cuLinkAddData(CUlinkState state, CUjitInputType type, void *data, size_t size, const char *name, - unsigned int numOptions, CUjit_option *options, void **optionValues); +CUresult CUDAAPI cuLinkAddData(CUlinkState state, CUjitInputType type, + void *data, size_t size, const char *name, + unsigned int numOptions, CUjit_option *options, + void **optionValues); /** * \brief Add a file input to a pending linker invocation @@ -4557,8 +4813,9 @@ cuLinkAddData(CUlinkState state, CUjitInputType type, void *data, size_t size, c * \param type The type of the input data * \param path Path to the input file * \param numOptions Size of options - * \param options Options to be applied only for this input (overrides options from ::cuLinkCreate) - * \param optionValues Array of option values, each cast to void * + * \param options Options to be applied only for this input (overrides + * options from ::cuLinkCreate) \param optionValues Array of option values, each + * cast to void * * * \return * ::CUDA_SUCCESS, @@ -4575,17 +4832,17 @@ cuLinkAddData(CUlinkState state, CUjitInputType type, void *data, size_t size, c * ::cuLinkComplete, * ::cuLinkDestroy */ -CUresult CUDAAPI -cuLinkAddFile(CUlinkState state, CUjitInputType type, const char *path, - unsigned int numOptions, CUjit_option *options, void **optionValues); +CUresult CUDAAPI cuLinkAddFile(CUlinkState state, CUjitInputType type, + const char *path, unsigned int numOptions, + CUjit_option *options, void **optionValues); /** * \brief Complete a pending linker invocation * - * Completes the pending linker action and returns the cubin image for the linked - * device code, which can be used with ::cuModuleLoadData. The cubin is owned by - * \p state, so it should be loaded before \p state is destroyed via ::cuLinkDestroy. - * This call does not destroy \p state. + * Completes the pending linker action and returns the cubin image for the + * linked device code, which can be used with ::cuModuleLoadData. The cubin is + * owned by \p state, so it should be loaded before \p state is destroyed via + * ::cuLinkDestroy. This call does not destroy \p state. * * \param state A pending linker invocation * \param cubinOut On success, this will point to the output image @@ -4602,8 +4859,8 @@ cuLinkAddFile(CUlinkState state, CUjitInputType type, const char *path, * ::cuLinkDestroy, * ::cuModuleLoadData */ -CUresult CUDAAPI -cuLinkComplete(CUlinkState state, void **cubinOut, size_t *sizeOut); +CUresult CUDAAPI cuLinkComplete(CUlinkState state, void **cubinOut, + size_t *sizeOut); /** * \brief Destroys state for a JIT linker invocation. @@ -4616,14 +4873,12 @@ cuLinkComplete(CUlinkState state, void **cubinOut, size_t *sizeOut); * * \sa ::cuLinkCreate */ -CUresult CUDAAPI -cuLinkDestroy(CUlinkState state); +CUresult CUDAAPI cuLinkDestroy(CUlinkState state); #endif /* __CUDA_API_VERSION >= 5050 */ /** @} */ /* END CUDA_MODULE */ - /** * \defgroup CUDA_MEM Memory Management * @@ -4658,7 +4913,8 @@ cuLinkDestroy(CUlinkState state); * ::cuArrayDestroy, ::cuArrayGetDescriptor, ::cuMemAlloc, ::cuMemAllocHost, * ::cuMemAllocPitch, ::cuMemcpy2D, ::cuMemcpy2DAsync, ::cuMemcpy2DUnaligned, * ::cuMemcpy3D, ::cuMemcpy3DAsync, ::cuMemcpyAtoA, ::cuMemcpyAtoD, - * ::cuMemcpyAtoH, ::cuMemcpyAtoHAsync, ::cuMemcpyDtoA, ::cuMemcpyDtoD, ::cuMemcpyDtoDAsync, + * ::cuMemcpyAtoH, ::cuMemcpyAtoHAsync, ::cuMemcpyDtoA, ::cuMemcpyDtoD, + * ::cuMemcpyDtoDAsync, * ::cuMemcpyDtoH, ::cuMemcpyDtoHAsync, ::cuMemcpyHtoA, ::cuMemcpyHtoAAsync, * ::cuMemcpyHtoD, ::cuMemcpyHtoDAsync, ::cuMemFree, ::cuMemFreeHost, * ::cuMemGetAddressRange, ::cuMemHostAlloc, @@ -4692,7 +4948,8 @@ CUresult CUDAAPI cuMemGetInfo(size_t *free, size_t *total); * ::cuArrayDestroy, ::cuArrayGetDescriptor, ::cuMemAllocHost, * ::cuMemAllocPitch, ::cuMemcpy2D, ::cuMemcpy2DAsync, ::cuMemcpy2DUnaligned, * ::cuMemcpy3D, ::cuMemcpy3DAsync, ::cuMemcpyAtoA, ::cuMemcpyAtoD, - * ::cuMemcpyAtoH, ::cuMemcpyAtoHAsync, ::cuMemcpyDtoA, ::cuMemcpyDtoD, ::cuMemcpyDtoDAsync, + * ::cuMemcpyAtoH, ::cuMemcpyAtoHAsync, ::cuMemcpyDtoA, ::cuMemcpyDtoD, + * ::cuMemcpyDtoDAsync, * ::cuMemcpyDtoH, ::cuMemcpyDtoHAsync, ::cuMemcpyHtoA, ::cuMemcpyHtoAAsync, * ::cuMemcpyHtoD, ::cuMemcpyHtoDAsync, ::cuMemFree, ::cuMemFreeHost, * ::cuMemGetAddressRange, ::cuMemGetInfo, ::cuMemHostAlloc, @@ -4754,7 +5011,8 @@ CUresult CUDAAPI cuMemAlloc(CUdeviceptr *dptr, size_t bytesize); * ::cuArrayDestroy, ::cuArrayGetDescriptor, ::cuMemAlloc, ::cuMemAllocHost, * ::cuMemcpy2D, ::cuMemcpy2DAsync, ::cuMemcpy2DUnaligned, * ::cuMemcpy3D, ::cuMemcpy3DAsync, ::cuMemcpyAtoA, ::cuMemcpyAtoD, - * ::cuMemcpyAtoH, ::cuMemcpyAtoHAsync, ::cuMemcpyDtoA, ::cuMemcpyDtoD, ::cuMemcpyDtoDAsync, + * ::cuMemcpyAtoH, ::cuMemcpyAtoHAsync, ::cuMemcpyDtoA, ::cuMemcpyDtoD, + ::cuMemcpyDtoDAsync, * ::cuMemcpyDtoH, ::cuMemcpyDtoHAsync, ::cuMemcpyHtoA, ::cuMemcpyHtoAAsync, * ::cuMemcpyHtoD, ::cuMemcpyHtoDAsync, ::cuMemFree, ::cuMemFreeHost, * ::cuMemGetAddressRange, ::cuMemGetInfo, ::cuMemHostAlloc, @@ -4762,7 +5020,9 @@ CUresult CUDAAPI cuMemAlloc(CUdeviceptr *dptr, size_t bytesize); * ::cuMemsetD2D32, ::cuMemsetD8, ::cuMemsetD16, ::cuMemsetD32, * ::cudaMallocPitch */ -CUresult CUDAAPI cuMemAllocPitch(CUdeviceptr *dptr, size_t *pPitch, size_t WidthInBytes, size_t Height, unsigned int ElementSizeBytes); +CUresult CUDAAPI cuMemAllocPitch(CUdeviceptr *dptr, size_t *pPitch, + size_t WidthInBytes, size_t Height, + unsigned int ElementSizeBytes); /** * \brief Frees device memory @@ -4784,7 +5044,8 @@ CUresult CUDAAPI cuMemAllocPitch(CUdeviceptr *dptr, size_t *pPitch, size_t Width * ::cuArrayDestroy, ::cuArrayGetDescriptor, ::cuMemAlloc, ::cuMemAllocHost, * ::cuMemAllocPitch, ::cuMemcpy2D, ::cuMemcpy2DAsync, ::cuMemcpy2DUnaligned, * ::cuMemcpy3D, ::cuMemcpy3DAsync, ::cuMemcpyAtoA, ::cuMemcpyAtoD, - * ::cuMemcpyAtoH, ::cuMemcpyAtoHAsync, ::cuMemcpyDtoA, ::cuMemcpyDtoD, ::cuMemcpyDtoDAsync, + * ::cuMemcpyAtoH, ::cuMemcpyAtoHAsync, ::cuMemcpyDtoA, ::cuMemcpyDtoD, + * ::cuMemcpyDtoDAsync, * ::cuMemcpyDtoH, ::cuMemcpyDtoHAsync, ::cuMemcpyHtoA, ::cuMemcpyHtoAAsync, * ::cuMemcpyHtoD, ::cuMemcpyHtoDAsync, ::cuMemFreeHost, * ::cuMemGetAddressRange, ::cuMemGetInfo, ::cuMemHostAlloc, @@ -4819,14 +5080,16 @@ CUresult CUDAAPI cuMemFree(CUdeviceptr dptr); * ::cuArrayDestroy, ::cuArrayGetDescriptor, ::cuMemAlloc, ::cuMemAllocHost, * ::cuMemAllocPitch, ::cuMemcpy2D, ::cuMemcpy2DAsync, ::cuMemcpy2DUnaligned, * ::cuMemcpy3D, ::cuMemcpy3DAsync, ::cuMemcpyAtoA, ::cuMemcpyAtoD, - * ::cuMemcpyAtoH, ::cuMemcpyAtoHAsync, ::cuMemcpyDtoA, ::cuMemcpyDtoD, ::cuMemcpyDtoDAsync, + * ::cuMemcpyAtoH, ::cuMemcpyAtoHAsync, ::cuMemcpyDtoA, ::cuMemcpyDtoD, + * ::cuMemcpyDtoDAsync, * ::cuMemcpyDtoH, ::cuMemcpyDtoHAsync, ::cuMemcpyHtoA, ::cuMemcpyHtoAAsync, * ::cuMemcpyHtoD, ::cuMemcpyHtoDAsync, ::cuMemFree, ::cuMemFreeHost, * ::cuMemGetInfo, ::cuMemHostAlloc, * ::cuMemHostGetDevicePointer, ::cuMemsetD2D8, ::cuMemsetD2D16, * ::cuMemsetD2D32, ::cuMemsetD8, ::cuMemsetD16, ::cuMemsetD32 */ -CUresult CUDAAPI cuMemGetAddressRange(CUdeviceptr *pbase, size_t *psize, CUdeviceptr dptr); +CUresult CUDAAPI cuMemGetAddressRange(CUdeviceptr *pbase, size_t *psize, + CUdeviceptr dptr); /** * \brief Allocates page-locked host memory @@ -4843,11 +5106,11 @@ CUresult CUDAAPI cuMemGetAddressRange(CUdeviceptr *pbase, size_t *psize, CUdevic * staging areas for data exchange between host and device. * * Note all host memory allocated using ::cuMemHostAlloc() will automatically - * be immediately accessible to all contexts on all devices which support unified - * addressing (as may be queried using ::CU_DEVICE_ATTRIBUTE_UNIFIED_ADDRESSING). - * The device pointer that may be used to access this host memory from those - * contexts is always equal to the returned host pointer \p *pp. - * See \ref CUDA_UNIFIED for additional details. + * be immediately accessible to all contexts on all devices which support + * unified addressing (as may be queried using + * ::CU_DEVICE_ATTRIBUTE_UNIFIED_ADDRESSING). The device pointer that may be + * used to access this host memory from those contexts is always equal to the + * returned host pointer \p *pp. See \ref CUDA_UNIFIED for additional details. * * \param pp - Returned host pointer to page-locked memory * \param bytesize - Requested allocation size in bytes @@ -4865,7 +5128,8 @@ CUresult CUDAAPI cuMemGetAddressRange(CUdeviceptr *pbase, size_t *psize, CUdevic * ::cuArrayDestroy, ::cuArrayGetDescriptor, ::cuMemAlloc, * ::cuMemAllocPitch, ::cuMemcpy2D, ::cuMemcpy2DAsync, ::cuMemcpy2DUnaligned, * ::cuMemcpy3D, ::cuMemcpy3DAsync, ::cuMemcpyAtoA, ::cuMemcpyAtoD, - * ::cuMemcpyAtoH, ::cuMemcpyAtoHAsync, ::cuMemcpyDtoA, ::cuMemcpyDtoD, ::cuMemcpyDtoDAsync, + * ::cuMemcpyAtoH, ::cuMemcpyAtoHAsync, ::cuMemcpyDtoA, ::cuMemcpyDtoD, + * ::cuMemcpyDtoDAsync, * ::cuMemcpyDtoH, ::cuMemcpyDtoHAsync, ::cuMemcpyHtoA, ::cuMemcpyHtoAAsync, * ::cuMemcpyHtoD, ::cuMemcpyHtoDAsync, ::cuMemFree, ::cuMemFreeHost, * ::cuMemGetAddressRange, ::cuMemGetInfo, ::cuMemHostAlloc, @@ -4896,7 +5160,8 @@ CUresult CUDAAPI cuMemAllocHost(void **pp, size_t bytesize); * ::cuArrayDestroy, ::cuArrayGetDescriptor, ::cuMemAlloc, ::cuMemAllocHost, * ::cuMemAllocPitch, ::cuMemcpy2D, ::cuMemcpy2DAsync, ::cuMemcpy2DUnaligned, * ::cuMemcpy3D, ::cuMemcpy3DAsync, ::cuMemcpyAtoA, ::cuMemcpyAtoD, - * ::cuMemcpyAtoH, ::cuMemcpyAtoHAsync, ::cuMemcpyDtoA, ::cuMemcpyDtoD, ::cuMemcpyDtoDAsync, + * ::cuMemcpyAtoH, ::cuMemcpyAtoHAsync, ::cuMemcpyDtoA, ::cuMemcpyDtoD, + * ::cuMemcpyDtoDAsync, * ::cuMemcpyDtoH, ::cuMemcpyDtoHAsync, ::cuMemcpyHtoA, ::cuMemcpyHtoAAsync, * ::cuMemcpyHtoD, ::cuMemcpyHtoDAsync, ::cuMemFree, * ::cuMemGetAddressRange, ::cuMemGetInfo, ::cuMemHostAlloc, @@ -4952,13 +5217,14 @@ CUresult CUDAAPI cuMemFreeHost(void *p); * The memory allocated by this function must be freed with ::cuMemFreeHost(). * * Note all host memory allocated using ::cuMemHostAlloc() will automatically - * be immediately accessible to all contexts on all devices which support unified - * addressing (as may be queried using ::CU_DEVICE_ATTRIBUTE_UNIFIED_ADDRESSING). - * Unless the flag ::CU_MEMHOSTALLOC_WRITECOMBINED is specified, the device pointer - * that may be used to access this host memory from those contexts is always equal - * to the returned host pointer \p *pp. If the flag ::CU_MEMHOSTALLOC_WRITECOMBINED - * is specified, then the function ::cuMemHostGetDevicePointer() must be used - * to query the device pointer, even if the context supports unified addressing. + * be immediately accessible to all contexts on all devices which support + * unified addressing (as may be queried using + * ::CU_DEVICE_ATTRIBUTE_UNIFIED_ADDRESSING). Unless the flag + * ::CU_MEMHOSTALLOC_WRITECOMBINED is specified, the device pointer that may be + * used to access this host memory from those contexts is always equal to the + * returned host pointer \p *pp. If the flag ::CU_MEMHOSTALLOC_WRITECOMBINED is + * specified, then the function ::cuMemHostGetDevicePointer() must be used to + * query the device pointer, even if the context supports unified addressing. * See \ref CUDA_UNIFIED for additional details. * * \param pp - Returned host pointer to page-locked memory @@ -4978,7 +5244,8 @@ CUresult CUDAAPI cuMemFreeHost(void *p); * ::cuArrayDestroy, ::cuArrayGetDescriptor, ::cuMemAlloc, ::cuMemAllocHost, * ::cuMemAllocPitch, ::cuMemcpy2D, ::cuMemcpy2DAsync, ::cuMemcpy2DUnaligned, * ::cuMemcpy3D, ::cuMemcpy3DAsync, ::cuMemcpyAtoA, ::cuMemcpyAtoD, - * ::cuMemcpyAtoH, ::cuMemcpyAtoHAsync, ::cuMemcpyDtoA, ::cuMemcpyDtoD, ::cuMemcpyDtoDAsync, + * ::cuMemcpyAtoH, ::cuMemcpyAtoHAsync, ::cuMemcpyDtoA, ::cuMemcpyDtoD, + * ::cuMemcpyDtoDAsync, * ::cuMemcpyDtoH, ::cuMemcpyDtoHAsync, ::cuMemcpyHtoA, ::cuMemcpyHtoAAsync, * ::cuMemcpyHtoD, ::cuMemcpyHtoDAsync, ::cuMemFree, ::cuMemFreeHost, * ::cuMemGetAddressRange, ::cuMemGetInfo, @@ -5003,16 +5270,18 @@ CUresult CUDAAPI cuMemHostAlloc(void **pp, size_t bytesize, unsigned int Flags); * ::CU_DEVICE_ATTRIBUTE_CAN_USE_HOST_POINTER_FOR_REGISTERED_MEM, the memory * can also be accessed from the device using the host pointer \p p. * The device pointer returned by ::cuMemHostGetDevicePointer() may or may not - * match the original host pointer \p p and depends on the devices visible to the - * application. If all devices visible to the application have a non-zero value for the - * device attribute, the device pointer returned by ::cuMemHostGetDevicePointer() - * will match the original pointer \p p. If any device visible to the application - * has a zero value for the device attribute, the device pointer returned by + * match the original host pointer \p p and depends on the devices visible to + * the application. If all devices visible to the application have a non-zero + * value for the device attribute, the device pointer returned by + * ::cuMemHostGetDevicePointer() will match the original pointer \p p. If any + * device visible to the application has a zero value for the device attribute, + * the device pointer returned by * ::cuMemHostGetDevicePointer() will not match the original host pointer \p p, - * but it will be suitable for use on all devices provided Unified Virtual Addressing - * is enabled. In such systems, it is valid to access the memory using either pointer - * on devices that have a non-zero value for the device attribute. Note however that - * such devices should access the memory using only of the two pointers and not both. + * but it will be suitable for use on all devices provided Unified Virtual + * Addressing is enabled. In such systems, it is valid to access the memory + * using either pointer on devices that have a non-zero value for the device + * attribute. Note however that such devices should access the memory using only + * of the two pointers and not both. * * \p Flags provides for future releases. For now, it must be set to 0. * @@ -5032,7 +5301,8 @@ CUresult CUDAAPI cuMemHostAlloc(void **pp, size_t bytesize, unsigned int Flags); * ::cuArrayDestroy, ::cuArrayGetDescriptor, ::cuMemAlloc, ::cuMemAllocHost, * ::cuMemAllocPitch, ::cuMemcpy2D, ::cuMemcpy2DAsync, ::cuMemcpy2DUnaligned, * ::cuMemcpy3D, ::cuMemcpy3DAsync, ::cuMemcpyAtoA, ::cuMemcpyAtoD, - * ::cuMemcpyAtoH, ::cuMemcpyAtoHAsync, ::cuMemcpyDtoA, ::cuMemcpyDtoD, ::cuMemcpyDtoDAsync, + * ::cuMemcpyAtoH, ::cuMemcpyAtoHAsync, ::cuMemcpyDtoA, ::cuMemcpyDtoD, + * ::cuMemcpyDtoDAsync, * ::cuMemcpyDtoH, ::cuMemcpyDtoHAsync, ::cuMemcpyHtoA, ::cuMemcpyHtoAAsync, * ::cuMemcpyHtoD, ::cuMemcpyHtoDAsync, ::cuMemFree, ::cuMemFreeHost, * ::cuMemGetAddressRange, ::cuMemGetInfo, ::cuMemHostAlloc, @@ -5040,7 +5310,8 @@ CUresult CUDAAPI cuMemHostAlloc(void **pp, size_t bytesize, unsigned int Flags); * ::cuMemsetD2D32, ::cuMemsetD8, ::cuMemsetD16, ::cuMemsetD32, * ::cudaHostGetDevicePointer */ -CUresult CUDAAPI cuMemHostGetDevicePointer(CUdeviceptr *pdptr, void *p, unsigned int Flags); +CUresult CUDAAPI cuMemHostGetDevicePointer(CUdeviceptr *pdptr, void *p, + unsigned int Flags); #endif /* __CUDA_API_VERSION >= 3020 */ /** @@ -5073,7 +5344,8 @@ CUresult CUDAAPI cuMemHostGetFlags(unsigned int *pFlags, void *p); #if __CUDA_API_VERSION >= 6000 /** - * \brief Allocates memory that will be automatically managed by the Unified Memory system + * \brief Allocates memory that will be automatically managed by the Unified + * Memory system * * Allocates \p bytesize bytes of managed memory on the device and returns in * \p *dptr a pointer to the allocated memory. If the device doesn't support @@ -5082,80 +5354,97 @@ CUresult CUDAAPI cuMemHostGetFlags(unsigned int *pFlags, void *p); * ::CU_DEVICE_ATTRIBUTE_MANAGED_MEMORY. The allocated memory is suitably * aligned for any kind of variable. The memory is not cleared. If \p bytesize * is 0, ::cuMemAllocManaged returns ::CUDA_ERROR_INVALID_VALUE. The pointer - * is valid on the CPU and on all GPUs in the system that support managed memory. - * All accesses to this pointer must obey the Unified Memory programming model. + * is valid on the CPU and on all GPUs in the system that support managed + * memory. All accesses to this pointer must obey the Unified Memory programming + * model. * * \p flags specifies the default stream association for this allocation. * \p flags must be one of ::CU_MEM_ATTACH_GLOBAL or ::CU_MEM_ATTACH_HOST. If * ::CU_MEM_ATTACH_GLOBAL is specified, then this memory is accessible from * any stream on any device. If ::CU_MEM_ATTACH_HOST is specified, then the * allocation should not be accessed from devices that have a zero value for the - * device attribute ::CU_DEVICE_ATTRIBUTE_CONCURRENT_MANAGED_ACCESS; an explicit call to + * device attribute ::CU_DEVICE_ATTRIBUTE_CONCURRENT_MANAGED_ACCESS; an explicit + * call to * ::cuStreamAttachMemAsync will be required to enable access on such devices. * * If the association is later changed via ::cuStreamAttachMemAsync to - * a single stream, the default association as specifed during ::cuMemAllocManaged - * is restored when that stream is destroyed. For __managed__ variables, the - * default association is always ::CU_MEM_ATTACH_GLOBAL. Note that destroying a - * stream is an asynchronous operation, and as a result, the change to default - * association won't happen until all work in the stream has completed. - * - * Memory allocated with ::cuMemAllocManaged should be released with ::cuMemFree. - * - * Device memory oversubscription is possible for GPUs that have a non-zero value for the - * device attribute ::CU_DEVICE_ATTRIBUTE_CONCURRENT_MANAGED_ACCESS. Managed memory on - * such GPUs may be evicted from device memory to host memory at any time by the Unified + * a single stream, the default association as specifed during + * ::cuMemAllocManaged is restored when that stream is destroyed. For + * __managed__ variables, the default association is always + * ::CU_MEM_ATTACH_GLOBAL. Note that destroying a stream is an asynchronous + * operation, and as a result, the change to default association won't happen + * until all work in the stream has completed. + * + * Memory allocated with ::cuMemAllocManaged should be released with + * ::cuMemFree. + * + * Device memory oversubscription is possible for GPUs that have a non-zero + * value for the device attribute + * ::CU_DEVICE_ATTRIBUTE_CONCURRENT_MANAGED_ACCESS. Managed memory on such GPUs + * may be evicted from device memory to host memory at any time by the Unified * Memory driver in order to make room for other allocations. * - * In a multi-GPU system where all GPUs have a non-zero value for the device attribute - * ::CU_DEVICE_ATTRIBUTE_CONCURRENT_MANAGED_ACCESS, managed memory may not be populated when this - * API returns and instead may be populated on access. In such systems, managed memory can - * migrate to any processor's memory at any time. The Unified Memory driver will employ heuristics to - * maintain data locality and prevent excessive page faults to the extent possible. The application - * can also guide the driver about memory usage patterns via ::cuMemAdvise. The application - * can also explicitly migrate memory to a desired processor's memory via + * In a multi-GPU system where all GPUs have a non-zero value for the device + * attribute + * ::CU_DEVICE_ATTRIBUTE_CONCURRENT_MANAGED_ACCESS, managed memory may not be + * populated when this API returns and instead may be populated on access. In + * such systems, managed memory can migrate to any processor's memory at any + * time. The Unified Memory driver will employ heuristics to maintain data + * locality and prevent excessive page faults to the extent possible. The + * application can also guide the driver about memory usage patterns via + * ::cuMemAdvise. The application can also explicitly migrate memory to a + * desired processor's memory via * ::cuMemPrefetchAsync. * - * In a multi-GPU system where all of the GPUs have a zero value for the device attribute - * ::CU_DEVICE_ATTRIBUTE_CONCURRENT_MANAGED_ACCESS and all the GPUs have peer-to-peer support - * with each other, the physical storage for managed memory is created on the GPU which is active - * at the time ::cuMemAllocManaged is called. All other GPUs will reference the data at reduced - * bandwidth via peer mappings over the PCIe bus. The Unified Memory driver does not migrate - * memory among such GPUs. - * - * In a multi-GPU system where not all GPUs have peer-to-peer support with each other and - * where the value of the device attribute ::CU_DEVICE_ATTRIBUTE_CONCURRENT_MANAGED_ACCESS - * is zero for at least one of those GPUs, the location chosen for physical storage of managed - * memory is system-dependent. - * - On Linux, the location chosen will be device memory as long as the current set of active - * contexts are on devices that either have peer-to-peer support with each other or have a - * non-zero value for the device attribute ::CU_DEVICE_ATTRIBUTE_CONCURRENT_MANAGED_ACCESS. - * If there is an active context on a GPU that does not have a non-zero value for that device - * attribute and it does not have peer-to-peer support with the other devices that have active - * contexts on them, then the location for physical storage will be 'zero-copy' or host memory. - * Note that this means that managed memory that is located in device memory is migrated to - * host memory if a new context is created on a GPU that doesn't have a non-zero value for - * the device attribute and does not support peer-to-peer with at least one of the other devices - * that has an active context. This in turn implies that context creation may fail if there is - * insufficient host memory to migrate all managed allocations. - * - On Windows, the physical storage is always created in 'zero-copy' or host memory. - * All GPUs will reference the data at reduced bandwidth over the PCIe bus. In these - * circumstances, use of the environment variable CUDA_VISIBLE_DEVICES is recommended to - * restrict CUDA to only use those GPUs that have peer-to-peer support. - * Alternatively, users can also set CUDA_MANAGED_FORCE_DEVICE_ALLOC to a - * non-zero value to force the driver to always use device memory for physical storage. - * When this environment variable is set to a non-zero value, all contexts created in - * that process on devices that support managed memory have to be peer-to-peer compatible - * with each other. Context creation will fail if a context is created on a device that - * supports managed memory and is not peer-to-peer compatible with any of the other - * managed memory supporting devices on which contexts were previously created, even if - * those contexts have been destroyed. These environment variables are described - * in the CUDA programming guide under the "CUDA environment variables" section. + * In a multi-GPU system where all of the GPUs have a zero value for the device + * attribute + * ::CU_DEVICE_ATTRIBUTE_CONCURRENT_MANAGED_ACCESS and all the GPUs have + * peer-to-peer support with each other, the physical storage for managed memory + * is created on the GPU which is active at the time ::cuMemAllocManaged is + * called. All other GPUs will reference the data at reduced bandwidth via peer + * mappings over the PCIe bus. The Unified Memory driver does not migrate memory + * among such GPUs. + * + * In a multi-GPU system where not all GPUs have peer-to-peer support with each + * other and where the value of the device attribute + * ::CU_DEVICE_ATTRIBUTE_CONCURRENT_MANAGED_ACCESS is zero for at least one of + * those GPUs, the location chosen for physical storage of managed memory is + * system-dependent. + * - On Linux, the location chosen will be device memory as long as the current + * set of active contexts are on devices that either have peer-to-peer support + * with each other or have a non-zero value for the device attribute + * ::CU_DEVICE_ATTRIBUTE_CONCURRENT_MANAGED_ACCESS. If there is an active + * context on a GPU that does not have a non-zero value for that device + * attribute and it does not have peer-to-peer support with the other devices + * that have active contexts on them, then the location for physical storage + * will be 'zero-copy' or host memory. Note that this means that managed memory + * that is located in device memory is migrated to host memory if a new context + * is created on a GPU that doesn't have a non-zero value for the device + * attribute and does not support peer-to-peer with at least one of the other + * devices that has an active context. This in turn implies that context + * creation may fail if there is insufficient host memory to migrate all managed + * allocations. + * - On Windows, the physical storage is always created in 'zero-copy' or host + * memory. All GPUs will reference the data at reduced bandwidth over the PCIe + * bus. In these circumstances, use of the environment variable + * CUDA_VISIBLE_DEVICES is recommended to restrict CUDA to only use those GPUs + * that have peer-to-peer support. Alternatively, users can also set + * CUDA_MANAGED_FORCE_DEVICE_ALLOC to a non-zero value to force the driver to + * always use device memory for physical storage. When this environment variable + * is set to a non-zero value, all contexts created in that process on devices + * that support managed memory have to be peer-to-peer compatible with each + * other. Context creation will fail if a context is created on a device that + * supports managed memory and is not peer-to-peer compatible with any of the + * other managed memory supporting devices on which contexts were previously + * created, even if those contexts have been destroyed. These environment + * variables are described in the CUDA programming guide under the "CUDA + * environment variables" section. * - On ARM, managed memory is not available on discrete gpu with Drive PX-2. * * \param dptr - Returned device pointer * \param bytesize - Requested allocation size in bytes - * \param flags - Must be one of ::CU_MEM_ATTACH_GLOBAL or ::CU_MEM_ATTACH_HOST + * \param flags - Must be one of ::CU_MEM_ATTACH_GLOBAL or + * ::CU_MEM_ATTACH_HOST * * \return * ::CUDA_SUCCESS, @@ -5171,7 +5460,8 @@ CUresult CUDAAPI cuMemHostGetFlags(unsigned int *pFlags, void *p); * ::cuArrayDestroy, ::cuArrayGetDescriptor, ::cuMemAllocHost, * ::cuMemAllocPitch, ::cuMemcpy2D, ::cuMemcpy2DAsync, ::cuMemcpy2DUnaligned, * ::cuMemcpy3D, ::cuMemcpy3DAsync, ::cuMemcpyAtoA, ::cuMemcpyAtoD, - * ::cuMemcpyAtoH, ::cuMemcpyAtoHAsync, ::cuMemcpyDtoA, ::cuMemcpyDtoD, ::cuMemcpyDtoDAsync, + * ::cuMemcpyAtoH, ::cuMemcpyAtoHAsync, ::cuMemcpyDtoA, ::cuMemcpyDtoD, + * ::cuMemcpyDtoDAsync, * ::cuMemcpyDtoH, ::cuMemcpyDtoHAsync, ::cuMemcpyHtoA, ::cuMemcpyHtoAAsync, * ::cuMemcpyHtoD, ::cuMemcpyHtoDAsync, ::cuMemFree, ::cuMemFreeHost, * ::cuMemGetAddressRange, ::cuMemGetInfo, ::cuMemHostAlloc, @@ -5180,7 +5470,8 @@ CUresult CUDAAPI cuMemHostGetFlags(unsigned int *pFlags, void *p); * ::cuDeviceGetAttribute, ::cuStreamAttachMemAsync, * ::cudaMallocManaged */ -CUresult CUDAAPI cuMemAllocManaged(CUdeviceptr *dptr, size_t bytesize, unsigned int flags); +CUresult CUDAAPI cuMemAllocManaged(CUdeviceptr *dptr, size_t bytesize, + unsigned int flags); #endif /* __CUDA_API_VERSION >= 6000 */ @@ -5197,7 +5488,8 @@ CUresult CUDAAPI cuMemAllocManaged(CUdeviceptr *dptr, size_t bytesize, unsigned * [domain]:[bus]:[device].[function] * [domain]:[bus]:[device] * [bus]:[device].[function] - * where \p domain, \p bus, \p device, and \p function are all hexadecimal values + * where \p domain, \p bus, \p device, and \p function are all hexadecimal + * values * * \return * ::CUDA_SUCCESS, @@ -5222,10 +5514,10 @@ CUresult CUDAAPI cuDeviceGetByPCIBusId(CUdevice *dev, const char *pciBusId); * string pointed to by \p pciBusId. \p len specifies the maximum length of the * string that may be returned. * - * \param pciBusId - Returned identifier string for the device in the following format - * [domain]:[bus]:[device].[function] - * where \p domain, \p bus, \p device, and \p function are all hexadecimal values. - * pciBusId should be large enough to store 13 characters including the NULL-terminator. + * \param pciBusId - Returned identifier string for the device in the following + * format [domain]:[bus]:[device].[function] where \p domain, \p bus, \p device, + * and \p function are all hexadecimal values. pciBusId should be large enough + * to store 13 characters including the NULL-terminator. * * \param len - Maximum length of string to store in \p name * @@ -5330,7 +5622,8 @@ CUresult CUDAAPI cuIpcGetEventHandle(CUipcEventHandle *pHandle, CUevent event); * ::cuIpcCloseMemHandle, * ::cudaIpcOpenEventHandle */ -CUresult CUDAAPI cuIpcOpenEventHandle(CUevent *phEvent, CUipcEventHandle handle); +CUresult CUDAAPI cuIpcOpenEventHandle(CUevent *phEvent, + CUipcEventHandle handle); /** * \brief Gets an interprocess memory handle for an existing device memory @@ -5403,7 +5696,8 @@ CUresult CUDAAPI cuIpcGetMemHandle(CUipcMemHandle *pHandle, CUdeviceptr dptr); * * \param pdptr - Returned device pointer * \param handle - ::CUipcMemHandle to open - * \param Flags - Flags for this operation. Must be specified as ::CU_IPC_MEM_LAZY_ENABLE_PEER_ACCESS + * \param Flags - Flags for this operation. Must be specified as + * ::CU_IPC_MEM_LAZY_ENABLE_PEER_ACCESS * * \returns * ::CUDA_SUCCESS, @@ -5414,7 +5708,8 @@ CUresult CUDAAPI cuIpcGetMemHandle(CUipcMemHandle *pHandle, CUdeviceptr dptr); * ::CUDA_ERROR_INVALID_VALUE * * \note No guarantees are made about the address returned in \p *pdptr. - * In particular, multiple processes may not receive the same address for the same \p handle. + * In particular, multiple processes may not receive the same address for the + * same \p handle. * * \sa * ::cuMemAlloc, @@ -5427,7 +5722,8 @@ CUresult CUDAAPI cuIpcGetMemHandle(CUipcMemHandle *pHandle, CUdeviceptr dptr); * ::cuDeviceCanAccessPeer, * ::cudaIpcOpenMemHandle */ -CUresult CUDAAPI cuIpcOpenMemHandle(CUdeviceptr *pdptr, CUipcMemHandle handle, unsigned int Flags); +CUresult CUDAAPI cuIpcOpenMemHandle(CUdeviceptr *pdptr, CUipcMemHandle handle, + unsigned int Flags); /** * \brief Close memory mapped with ::cuIpcOpenMemHandle @@ -5470,14 +5766,14 @@ CUresult CUDAAPI cuIpcCloseMemHandle(CUdeviceptr dptr); * * Page-locks the memory range specified by \p p and \p bytesize and maps it * for the device(s) as specified by \p Flags. This memory range also is added - * to the same tracking mechanism as ::cuMemHostAlloc to automatically accelerate - * calls to functions such as ::cuMemcpyHtoD(). Since the memory can be accessed - * directly by the device, it can be read or written with much higher bandwidth - * than pageable memory that has not been registered. Page-locking excessive - * amounts of memory may degrade system performance, since it reduces the amount - * of memory available to the system for paging. As a result, this function is - * best used sparingly to register staging areas for data exchange between - * host and device. + * to the same tracking mechanism as ::cuMemHostAlloc to automatically + * accelerate calls to functions such as ::cuMemcpyHtoD(). Since the memory can + * be accessed directly by the device, it can be read or written with much + * higher bandwidth than pageable memory that has not been registered. + * Page-locking excessive amounts of memory may degrade system performance, + * since it reduces the amount of memory available to the system for paging. As + * a result, this function is best used sparingly to register staging areas for + * data exchange between host and device. * * This function has limited support on Mac OS X. OS 10.7 or higher is required. * @@ -5510,16 +5806,18 @@ CUresult CUDAAPI cuIpcCloseMemHandle(CUdeviceptr dptr); * ::CU_DEVICE_ATTRIBUTE_CAN_USE_HOST_POINTER_FOR_REGISTERED_MEM, the memory * can also be accessed from the device using the host pointer \p p. * The device pointer returned by ::cuMemHostGetDevicePointer() may or may not - * match the original host pointer \p ptr and depends on the devices visible to the - * application. If all devices visible to the application have a non-zero value for the - * device attribute, the device pointer returned by ::cuMemHostGetDevicePointer() - * will match the original pointer \p ptr. If any device visible to the application - * has a zero value for the device attribute, the device pointer returned by - * ::cuMemHostGetDevicePointer() will not match the original host pointer \p ptr, - * but it will be suitable for use on all devices provided Unified Virtual Addressing - * is enabled. In such systems, it is valid to access the memory using either pointer - * on devices that have a non-zero value for the device attribute. Note however that - * such devices should access the memory using only of the two pointers and not both. + * match the original host pointer \p ptr and depends on the devices visible to + * the application. If all devices visible to the application have a non-zero + * value for the device attribute, the device pointer returned by + * ::cuMemHostGetDevicePointer() will match the original pointer \p ptr. If any + * device visible to the application has a zero value for the device attribute, + * the device pointer returned by + * ::cuMemHostGetDevicePointer() will not match the original host pointer \p + * ptr, but it will be suitable for use on all devices provided Unified Virtual + * Addressing is enabled. In such systems, it is valid to access the memory + * using either pointer on devices that have a non-zero value for the device + * attribute. Note however that such devices should access the memory using only + * of the two pointers and not both. * * The memory page-locked by this function must be unregistered with * ::cuMemHostUnregister(). @@ -5546,7 +5844,8 @@ CUresult CUDAAPI cuIpcCloseMemHandle(CUdeviceptr dptr); * ::cuMemHostGetDevicePointer, * ::cudaHostRegister */ -CUresult CUDAAPI cuMemHostRegister(void *p, size_t bytesize, unsigned int Flags); +CUresult CUDAAPI cuMemHostRegister(void *p, size_t bytesize, + unsigned int Flags); /** * \brief Unregisters a memory range that was registered with cuMemHostRegister. @@ -5578,11 +5877,11 @@ CUresult CUDAAPI cuMemHostUnregister(void *p); * \brief Copies memory * * Copies data between two pointers. - * \p dst and \p src are base pointers of the destination and source, respectively. - * \p ByteCount specifies the number of bytes to copy. - * Note that this function infers the type of the transfer (host to host, host to - * device, device to device, or device to host) from the pointer values. This - * function is only allowed in contexts which support unified addressing. + * \p dst and \p src are base pointers of the destination and source, + * respectively. \p ByteCount specifies the number of bytes to copy. Note that + * this function infers the type of the transfer (host to host, host to device, + * device to device, or device to host) from the pointer values. This function + * is only allowed in contexts which support unified addressing. * * \param dst - Destination unified virtual address space pointer * \param src - Source unified virtual address space pointer @@ -5637,11 +5936,14 @@ CUresult CUDAAPI cuMemcpy(CUdeviceptr dst, CUdeviceptr src, size_t ByteCount); * \notefnerr * \note_sync * - * \sa ::cuMemcpyDtoD, ::cuMemcpy3DPeer, ::cuMemcpyDtoDAsync, ::cuMemcpyPeerAsync, + * \sa ::cuMemcpyDtoD, ::cuMemcpy3DPeer, ::cuMemcpyDtoDAsync, + * ::cuMemcpyPeerAsync, * ::cuMemcpy3DPeerAsync, * ::cudaMemcpyPeer */ -CUresult CUDAAPI cuMemcpyPeer(CUdeviceptr dstDevice, CUcontext dstContext, CUdeviceptr srcDevice, CUcontext srcContext, size_t ByteCount); +CUresult CUDAAPI cuMemcpyPeer(CUdeviceptr dstDevice, CUcontext dstContext, + CUdeviceptr srcDevice, CUcontext srcContext, + size_t ByteCount); #endif /* __CUDA_API_VERSION >= 4000 */ @@ -5670,7 +5972,8 @@ CUresult CUDAAPI cuMemcpyPeer(CUdeviceptr dstDevice, CUcontext dstContext, CUdev * ::cuArrayDestroy, ::cuArrayGetDescriptor, ::cuMemAlloc, ::cuMemAllocHost, * ::cuMemAllocPitch, ::cuMemcpy2D, ::cuMemcpy2DAsync, ::cuMemcpy2DUnaligned, * ::cuMemcpy3D, ::cuMemcpy3DAsync, ::cuMemcpyAtoA, ::cuMemcpyAtoD, - * ::cuMemcpyAtoH, ::cuMemcpyAtoHAsync, ::cuMemcpyDtoA, ::cuMemcpyDtoD, ::cuMemcpyDtoDAsync, + * ::cuMemcpyAtoH, ::cuMemcpyAtoHAsync, ::cuMemcpyDtoA, ::cuMemcpyDtoD, + * ::cuMemcpyDtoDAsync, * ::cuMemcpyDtoH, ::cuMemcpyDtoHAsync, ::cuMemcpyHtoA, ::cuMemcpyHtoAAsync, * ::cuMemcpyHtoDAsync, ::cuMemFree, ::cuMemFreeHost, * ::cuMemGetAddressRange, ::cuMemGetInfo, ::cuMemHostAlloc, @@ -5679,7 +5982,8 @@ CUresult CUDAAPI cuMemcpyPeer(CUdeviceptr dstDevice, CUcontext dstContext, CUdev * ::cudaMemcpy, * ::cudaMemcpyToSymbol */ -CUresult CUDAAPI cuMemcpyHtoD(CUdeviceptr dstDevice, const void *srcHost, size_t ByteCount); +CUresult CUDAAPI cuMemcpyHtoD(CUdeviceptr dstDevice, const void *srcHost, + size_t ByteCount); /** * \brief Copies memory from Device to Host @@ -5705,7 +6009,8 @@ CUresult CUDAAPI cuMemcpyHtoD(CUdeviceptr dstDevice, const void *srcHost, size_t * ::cuArrayDestroy, ::cuArrayGetDescriptor, ::cuMemAlloc, ::cuMemAllocHost, * ::cuMemAllocPitch, ::cuMemcpy2D, ::cuMemcpy2DAsync, ::cuMemcpy2DUnaligned, * ::cuMemcpy3D, ::cuMemcpy3DAsync, ::cuMemcpyAtoA, ::cuMemcpyAtoD, - * ::cuMemcpyAtoH, ::cuMemcpyAtoHAsync, ::cuMemcpyDtoA, ::cuMemcpyDtoD, ::cuMemcpyDtoDAsync, + * ::cuMemcpyAtoH, ::cuMemcpyAtoHAsync, ::cuMemcpyDtoA, ::cuMemcpyDtoD, + * ::cuMemcpyDtoDAsync, * ::cuMemcpyDtoHAsync, ::cuMemcpyHtoA, ::cuMemcpyHtoAAsync, * ::cuMemcpyHtoD, ::cuMemcpyHtoDAsync, ::cuMemFree, ::cuMemFreeHost, * ::cuMemGetAddressRange, ::cuMemGetInfo, ::cuMemHostAlloc, @@ -5714,7 +6019,8 @@ CUresult CUDAAPI cuMemcpyHtoD(CUdeviceptr dstDevice, const void *srcHost, size_t * ::cudaMemcpy, * ::cudaMemcpyFromSymbol */ -CUresult CUDAAPI cuMemcpyDtoH(void *dstHost, CUdeviceptr srcDevice, size_t ByteCount); +CUresult CUDAAPI cuMemcpyDtoH(void *dstHost, CUdeviceptr srcDevice, + size_t ByteCount); /** * \brief Copies memory from Device to Device @@ -5750,7 +6056,8 @@ CUresult CUDAAPI cuMemcpyDtoH(void *dstHost, CUdeviceptr srcDevice, size_t ByteC * ::cudaMemcpyToSymbol, * ::cudaMemcpyFromSymbol */ -CUresult CUDAAPI cuMemcpyDtoD(CUdeviceptr dstDevice, CUdeviceptr srcDevice, size_t ByteCount); +CUresult CUDAAPI cuMemcpyDtoD(CUdeviceptr dstDevice, CUdeviceptr srcDevice, + size_t ByteCount); /** * \brief Copies memory from Device to Array @@ -5786,7 +6093,8 @@ CUresult CUDAAPI cuMemcpyDtoD(CUdeviceptr dstDevice, CUdeviceptr srcDevice, size * ::cuMemsetD2D32, ::cuMemsetD8, ::cuMemsetD16, ::cuMemsetD32, * ::cudaMemcpyToArray */ -CUresult CUDAAPI cuMemcpyDtoA(CUarray dstArray, size_t dstOffset, CUdeviceptr srcDevice, size_t ByteCount); +CUresult CUDAAPI cuMemcpyDtoA(CUarray dstArray, size_t dstOffset, + CUdeviceptr srcDevice, size_t ByteCount); /** * \brief Copies memory from Array to Device @@ -5816,7 +6124,8 @@ CUresult CUDAAPI cuMemcpyDtoA(CUarray dstArray, size_t dstOffset, CUdeviceptr sr * ::cuArrayDestroy, ::cuArrayGetDescriptor, ::cuMemAlloc, ::cuMemAllocHost, * ::cuMemAllocPitch, ::cuMemcpy2D, ::cuMemcpy2DAsync, ::cuMemcpy2DUnaligned, * ::cuMemcpy3D, ::cuMemcpy3DAsync, ::cuMemcpyAtoA, - * ::cuMemcpyAtoH, ::cuMemcpyAtoHAsync, ::cuMemcpyDtoA, ::cuMemcpyDtoD, ::cuMemcpyDtoDAsync, + * ::cuMemcpyAtoH, ::cuMemcpyAtoHAsync, ::cuMemcpyDtoA, ::cuMemcpyDtoD, + * ::cuMemcpyDtoDAsync, * ::cuMemcpyDtoH, ::cuMemcpyDtoHAsync, ::cuMemcpyHtoA, ::cuMemcpyHtoAAsync, * ::cuMemcpyHtoD, ::cuMemcpyHtoDAsync, ::cuMemFree, ::cuMemFreeHost, * ::cuMemGetAddressRange, ::cuMemGetInfo, ::cuMemHostAlloc, @@ -5824,15 +6133,16 @@ CUresult CUDAAPI cuMemcpyDtoA(CUarray dstArray, size_t dstOffset, CUdeviceptr sr * ::cuMemsetD2D32, ::cuMemsetD8, ::cuMemsetD16, ::cuMemsetD32, * ::cudaMemcpyFromArray */ -CUresult CUDAAPI cuMemcpyAtoD(CUdeviceptr dstDevice, CUarray srcArray, size_t srcOffset, size_t ByteCount); +CUresult CUDAAPI cuMemcpyAtoD(CUdeviceptr dstDevice, CUarray srcArray, + size_t srcOffset, size_t ByteCount); /** * \brief Copies memory from Host to Array * * Copies from host memory to a 1D CUDA array. \p dstArray and \p dstOffset * specify the CUDA array handle and starting offset in bytes of the destination - * data. \p pSrc specifies the base address of the source. \p ByteCount specifies - * the number of bytes to copy. + * data. \p pSrc specifies the base address of the source. \p ByteCount + * specifies the number of bytes to copy. * * \param dstArray - Destination array * \param dstOffset - Offset in bytes of destination array @@ -5852,7 +6162,8 @@ CUresult CUDAAPI cuMemcpyAtoD(CUdeviceptr dstDevice, CUarray srcArray, size_t sr * ::cuArrayDestroy, ::cuArrayGetDescriptor, ::cuMemAlloc, ::cuMemAllocHost, * ::cuMemAllocPitch, ::cuMemcpy2D, ::cuMemcpy2DAsync, ::cuMemcpy2DUnaligned, * ::cuMemcpy3D, ::cuMemcpy3DAsync, ::cuMemcpyAtoA, ::cuMemcpyAtoD, - * ::cuMemcpyAtoH, ::cuMemcpyAtoHAsync, ::cuMemcpyDtoA, ::cuMemcpyDtoD, ::cuMemcpyDtoDAsync, + * ::cuMemcpyAtoH, ::cuMemcpyAtoHAsync, ::cuMemcpyDtoA, ::cuMemcpyDtoD, + * ::cuMemcpyDtoDAsync, * ::cuMemcpyDtoH, ::cuMemcpyDtoHAsync, ::cuMemcpyHtoAAsync, * ::cuMemcpyHtoD, ::cuMemcpyHtoDAsync, ::cuMemFree, ::cuMemFreeHost, * ::cuMemGetAddressRange, ::cuMemGetInfo, ::cuMemHostAlloc, @@ -5860,7 +6171,8 @@ CUresult CUDAAPI cuMemcpyAtoD(CUdeviceptr dstDevice, CUarray srcArray, size_t sr * ::cuMemsetD2D32, ::cuMemsetD8, ::cuMemsetD16, ::cuMemsetD32, * ::cudaMemcpyToArray */ -CUresult CUDAAPI cuMemcpyHtoA(CUarray dstArray, size_t dstOffset, const void *srcHost, size_t ByteCount); +CUresult CUDAAPI cuMemcpyHtoA(CUarray dstArray, size_t dstOffset, + const void *srcHost, size_t ByteCount); /** * \brief Copies memory from Array to Host @@ -5896,7 +6208,8 @@ CUresult CUDAAPI cuMemcpyHtoA(CUarray dstArray, size_t dstOffset, const void *sr * ::cuMemsetD2D32, ::cuMemsetD8, ::cuMemsetD16, ::cuMemsetD32, * ::cudaMemcpyFromArray */ -CUresult CUDAAPI cuMemcpyAtoH(void *dstHost, CUarray srcArray, size_t srcOffset, size_t ByteCount); +CUresult CUDAAPI cuMemcpyAtoH(void *dstHost, CUarray srcArray, size_t srcOffset, + size_t ByteCount); /** * \brief Copies memory from Array to Array @@ -5928,7 +6241,8 @@ CUresult CUDAAPI cuMemcpyAtoH(void *dstHost, CUarray srcArray, size_t srcOffset, * ::cuArrayDestroy, ::cuArrayGetDescriptor, ::cuMemAlloc, ::cuMemAllocHost, * ::cuMemAllocPitch, ::cuMemcpy2D, ::cuMemcpy2DAsync, ::cuMemcpy2DUnaligned, * ::cuMemcpy3D, ::cuMemcpy3DAsync, ::cuMemcpyAtoD, - * ::cuMemcpyAtoH, ::cuMemcpyAtoHAsync, ::cuMemcpyDtoA, ::cuMemcpyDtoD, ::cuMemcpyDtoDAsync, + * ::cuMemcpyAtoH, ::cuMemcpyAtoHAsync, ::cuMemcpyDtoA, ::cuMemcpyDtoD, + * ::cuMemcpyDtoDAsync, * ::cuMemcpyDtoH, ::cuMemcpyDtoHAsync, ::cuMemcpyHtoA, ::cuMemcpyHtoAAsync, * ::cuMemcpyHtoD, ::cuMemcpyHtoDAsync, ::cuMemFree, ::cuMemFreeHost, * ::cuMemGetAddressRange, ::cuMemGetInfo, ::cuMemHostAlloc, @@ -5936,7 +6250,9 @@ CUresult CUDAAPI cuMemcpyAtoH(void *dstHost, CUarray srcArray, size_t srcOffset, * ::cuMemsetD2D32, ::cuMemsetD8, ::cuMemsetD16, ::cuMemsetD32, * ::cudaMemcpyArrayToArray */ -CUresult CUDAAPI cuMemcpyAtoA(CUarray dstArray, size_t dstOffset, CUarray srcArray, size_t srcOffset, size_t ByteCount); +CUresult CUDAAPI cuMemcpyAtoA(CUarray dstArray, size_t dstOffset, + CUarray srcArray, size_t srcOffset, + size_t ByteCount); /** * \brief Copies memory for 2D arrays @@ -6090,7 +6406,8 @@ CUresult CUDAAPI cuMemcpyAtoA(CUarray dstArray, size_t dstOffset, CUarray srcArr * ::cuArrayDestroy, ::cuArrayGetDescriptor, ::cuMemAlloc, ::cuMemAllocHost, * ::cuMemAllocPitch, ::cuMemcpy2DAsync, ::cuMemcpy2DUnaligned, * ::cuMemcpy3D, ::cuMemcpy3DAsync, ::cuMemcpyAtoA, ::cuMemcpyAtoD, - * ::cuMemcpyAtoH, ::cuMemcpyAtoHAsync, ::cuMemcpyDtoA, ::cuMemcpyDtoD, ::cuMemcpyDtoDAsync, + * ::cuMemcpyAtoH, ::cuMemcpyAtoHAsync, ::cuMemcpyDtoA, ::cuMemcpyDtoD, + ::cuMemcpyDtoDAsync, * ::cuMemcpyDtoH, ::cuMemcpyDtoHAsync, ::cuMemcpyHtoA, ::cuMemcpyHtoAAsync, * ::cuMemcpyHtoD, ::cuMemcpyHtoDAsync, ::cuMemFree, ::cuMemFreeHost, * ::cuMemGetAddressRange, ::cuMemGetInfo, ::cuMemHostAlloc, @@ -6252,7 +6569,8 @@ CUresult CUDAAPI cuMemcpy2D(const CUDA_MEMCPY2D *pCopy); * ::cuArrayDestroy, ::cuArrayGetDescriptor, ::cuMemAlloc, ::cuMemAllocHost, * ::cuMemAllocPitch, ::cuMemcpy2D, ::cuMemcpy2DAsync, * ::cuMemcpy3D, ::cuMemcpy3DAsync, ::cuMemcpyAtoA, ::cuMemcpyAtoD, - * ::cuMemcpyAtoH, ::cuMemcpyAtoHAsync, ::cuMemcpyDtoA, ::cuMemcpyDtoD, ::cuMemcpyDtoDAsync, + * ::cuMemcpyAtoH, ::cuMemcpyAtoHAsync, ::cuMemcpyDtoA, ::cuMemcpyDtoD, + ::cuMemcpyDtoDAsync, * ::cuMemcpyDtoH, ::cuMemcpyDtoHAsync, ::cuMemcpyHtoA, ::cuMemcpyHtoAAsync, * ::cuMemcpyHtoD, ::cuMemcpyHtoDAsync, ::cuMemFree, ::cuMemFreeHost, * ::cuMemGetAddressRange, ::cuMemGetInfo, ::cuMemHostAlloc, @@ -6280,7 +6598,8 @@ CUresult CUDAAPI cuMemcpy2DUnaligned(const CUDA_MEMCPY2D *pCopy); CUdeviceptr srcDevice; CUarray srcArray; unsigned int srcPitch; // ignored when src is array - unsigned int srcHeight; // ignored when src is array; may be 0 if Depth==1 + unsigned int srcHeight; // ignored when src is array; may be 0 + if Depth==1 unsigned int dstXInBytes, dstY, dstZ; unsigned int dstLOD; @@ -6289,7 +6608,8 @@ CUresult CUDAAPI cuMemcpy2DUnaligned(const CUDA_MEMCPY2D *pCopy); CUdeviceptr dstDevice; CUarray dstArray; unsigned int dstPitch; // ignored when dst is array - unsigned int dstHeight; // ignored when dst is array; may be 0 if Depth==1 + unsigned int dstHeight; // ignored when dst is array; may be 0 + if Depth==1 unsigned int WidthInBytes; unsigned int Height; @@ -6361,7 +6681,8 @@ CUresult CUDAAPI cuMemcpy2DUnaligned(const CUDA_MEMCPY2D *pCopy); * \par * For host pointers, the starting address is * \code - void* Start = (void*)((char*)srcHost+(srcZ*srcHeight+srcY)*srcPitch + srcXInBytes); + void* Start = (void*)((char*)srcHost+(srcZ*srcHeight+srcY)*srcPitch + + srcXInBytes); * \endcode * * \par @@ -6380,7 +6701,8 @@ CUresult CUDAAPI cuMemcpy2DUnaligned(const CUDA_MEMCPY2D *pCopy); * \par * For host pointers, the base address is * \code - void* dstStart = (void*)((char*)dstHost+(dstZ*dstHeight+dstY)*dstPitch + dstXInBytes); + void* dstStart = (void*)((char*)dstHost+(dstZ*dstHeight+dstY)*dstPitch + + dstXInBytes); * \endcode * * \par @@ -6423,7 +6745,8 @@ CUresult CUDAAPI cuMemcpy2DUnaligned(const CUDA_MEMCPY2D *pCopy); * ::cuArrayDestroy, ::cuArrayGetDescriptor, ::cuMemAlloc, ::cuMemAllocHost, * ::cuMemAllocPitch, ::cuMemcpy2D, ::cuMemcpy2DAsync, ::cuMemcpy2DUnaligned, * ::cuMemcpy3DAsync, ::cuMemcpyAtoA, ::cuMemcpyAtoD, - * ::cuMemcpyAtoH, ::cuMemcpyAtoHAsync, ::cuMemcpyDtoA, ::cuMemcpyDtoD, ::cuMemcpyDtoDAsync, + * ::cuMemcpyAtoH, ::cuMemcpyAtoHAsync, ::cuMemcpyDtoA, ::cuMemcpyDtoD, + ::cuMemcpyDtoDAsync, * ::cuMemcpyDtoH, ::cuMemcpyDtoHAsync, ::cuMemcpyHtoA, ::cuMemcpyHtoAAsync, * ::cuMemcpyHtoD, ::cuMemcpyHtoDAsync, ::cuMemFree, ::cuMemFreeHost, * ::cuMemGetAddressRange, ::cuMemGetInfo, ::cuMemHostAlloc, @@ -6463,11 +6786,11 @@ CUresult CUDAAPI cuMemcpy3DPeer(const CUDA_MEMCPY3D_PEER *pCopy); * \brief Copies memory asynchronously * * Copies data between two pointers. - * \p dst and \p src are base pointers of the destination and source, respectively. - * \p ByteCount specifies the number of bytes to copy. - * Note that this function infers the type of the transfer (host to host, host to - * device, device to device, or device to host) from the pointer values. This - * function is only allowed in contexts which support unified addressing. + * \p dst and \p src are base pointers of the destination and source, + * respectively. \p ByteCount specifies the number of bytes to copy. Note that + * this function infers the type of the transfer (host to host, host to device, + * device to device, or device to host) from the pointer values. This function + * is only allowed in contexts which support unified addressing. * * \param dst - Destination unified virtual address space pointer * \param src - Source unified virtual address space pointer @@ -6501,7 +6824,8 @@ CUresult CUDAAPI cuMemcpy3DPeer(const CUDA_MEMCPY3D_PEER *pCopy); * ::cudaMemcpyToSymbolAsync, * ::cudaMemcpyFromSymbolAsync */ -CUresult CUDAAPI cuMemcpyAsync(CUdeviceptr dst, CUdeviceptr src, size_t ByteCount, CUstream hStream); +CUresult CUDAAPI cuMemcpyAsync(CUdeviceptr dst, CUdeviceptr src, + size_t ByteCount, CUstream hStream); /** * \brief Copies device memory between two contexts asynchronously. @@ -6534,7 +6858,9 @@ CUresult CUDAAPI cuMemcpyAsync(CUdeviceptr dst, CUdeviceptr src, size_t ByteCoun * ::cuMemcpy3DPeerAsync, * ::cudaMemcpyPeerAsync */ -CUresult CUDAAPI cuMemcpyPeerAsync(CUdeviceptr dstDevice, CUcontext dstContext, CUdeviceptr srcDevice, CUcontext srcContext, size_t ByteCount, CUstream hStream); +CUresult CUDAAPI cuMemcpyPeerAsync(CUdeviceptr dstDevice, CUcontext dstContext, + CUdeviceptr srcDevice, CUcontext srcContext, + size_t ByteCount, CUstream hStream); #endif /* __CUDA_API_VERSION >= 4000 */ #if __CUDA_API_VERSION >= 3020 @@ -6565,7 +6891,8 @@ CUresult CUDAAPI cuMemcpyPeerAsync(CUdeviceptr dstDevice, CUcontext dstContext, * ::cuArrayDestroy, ::cuArrayGetDescriptor, ::cuMemAlloc, ::cuMemAllocHost, * ::cuMemAllocPitch, ::cuMemcpy2D, ::cuMemcpy2DAsync, ::cuMemcpy2DUnaligned, * ::cuMemcpy3D, ::cuMemcpy3DAsync, ::cuMemcpyAtoA, ::cuMemcpyAtoD, - * ::cuMemcpyAtoH, ::cuMemcpyAtoHAsync, ::cuMemcpyDtoA, ::cuMemcpyDtoD, ::cuMemcpyDtoDAsync, + * ::cuMemcpyAtoH, ::cuMemcpyAtoHAsync, ::cuMemcpyDtoA, ::cuMemcpyDtoD, + * ::cuMemcpyDtoDAsync, * ::cuMemcpyDtoH, ::cuMemcpyDtoHAsync, ::cuMemcpyHtoA, ::cuMemcpyHtoAAsync, * ::cuMemcpyHtoD, ::cuMemFree, ::cuMemFreeHost, * ::cuMemGetAddressRange, ::cuMemGetInfo, ::cuMemHostAlloc, @@ -6576,7 +6903,8 @@ CUresult CUDAAPI cuMemcpyPeerAsync(CUdeviceptr dstDevice, CUcontext dstContext, * ::cudaMemcpyAsync, * ::cudaMemcpyToSymbolAsync */ -CUresult CUDAAPI cuMemcpyHtoDAsync(CUdeviceptr dstDevice, const void *srcHost, size_t ByteCount, CUstream hStream); +CUresult CUDAAPI cuMemcpyHtoDAsync(CUdeviceptr dstDevice, const void *srcHost, + size_t ByteCount, CUstream hStream); /** * \brief Copies memory from Device to Host @@ -6605,7 +6933,8 @@ CUresult CUDAAPI cuMemcpyHtoDAsync(CUdeviceptr dstDevice, const void *srcHost, s * ::cuArrayDestroy, ::cuArrayGetDescriptor, ::cuMemAlloc, ::cuMemAllocHost, * ::cuMemAllocPitch, ::cuMemcpy2D, ::cuMemcpy2DAsync, ::cuMemcpy2DUnaligned, * ::cuMemcpy3D, ::cuMemcpy3DAsync, ::cuMemcpyAtoA, ::cuMemcpyAtoD, - * ::cuMemcpyAtoH, ::cuMemcpyAtoHAsync, ::cuMemcpyDtoA, ::cuMemcpyDtoD, ::cuMemcpyDtoDAsync, + * ::cuMemcpyAtoH, ::cuMemcpyAtoHAsync, ::cuMemcpyDtoA, ::cuMemcpyDtoD, + * ::cuMemcpyDtoDAsync, * ::cuMemcpyDtoH, ::cuMemcpyHtoA, ::cuMemcpyHtoAAsync, * ::cuMemcpyHtoD, ::cuMemcpyHtoDAsync, ::cuMemFree, ::cuMemFreeHost, * ::cuMemGetAddressRange, ::cuMemGetInfo, ::cuMemHostAlloc, @@ -6616,7 +6945,8 @@ CUresult CUDAAPI cuMemcpyHtoDAsync(CUdeviceptr dstDevice, const void *srcHost, s * ::cudaMemcpyAsync, * ::cudaMemcpyFromSymbolAsync */ -CUresult CUDAAPI cuMemcpyDtoHAsync(void *dstHost, CUdeviceptr srcDevice, size_t ByteCount, CUstream hStream); +CUresult CUDAAPI cuMemcpyDtoHAsync(void *dstHost, CUdeviceptr srcDevice, + size_t ByteCount, CUstream hStream); /** * \brief Copies memory from Device to Device @@ -6657,7 +6987,8 @@ CUresult CUDAAPI cuMemcpyDtoHAsync(void *dstHost, CUdeviceptr srcDevice, size_t * ::cudaMemcpyToSymbolAsync, * ::cudaMemcpyFromSymbolAsync */ -CUresult CUDAAPI cuMemcpyDtoDAsync(CUdeviceptr dstDevice, CUdeviceptr srcDevice, size_t ByteCount, CUstream hStream); +CUresult CUDAAPI cuMemcpyDtoDAsync(CUdeviceptr dstDevice, CUdeviceptr srcDevice, + size_t ByteCount, CUstream hStream); /** * \brief Copies memory from Host to Array @@ -6688,7 +7019,8 @@ CUresult CUDAAPI cuMemcpyDtoDAsync(CUdeviceptr dstDevice, CUdeviceptr srcDevice, * ::cuArrayDestroy, ::cuArrayGetDescriptor, ::cuMemAlloc, ::cuMemAllocHost, * ::cuMemAllocPitch, ::cuMemcpy2D, ::cuMemcpy2DAsync, ::cuMemcpy2DUnaligned, * ::cuMemcpy3D, ::cuMemcpy3DAsync, ::cuMemcpyAtoA, ::cuMemcpyAtoD, - * ::cuMemcpyAtoH, ::cuMemcpyAtoHAsync, ::cuMemcpyDtoA, ::cuMemcpyDtoD, ::cuMemcpyDtoDAsync, + * ::cuMemcpyAtoH, ::cuMemcpyAtoHAsync, ::cuMemcpyDtoA, ::cuMemcpyDtoD, + * ::cuMemcpyDtoDAsync, * ::cuMemcpyDtoH, ::cuMemcpyDtoHAsync, ::cuMemcpyHtoA, * ::cuMemcpyHtoD, ::cuMemcpyHtoDAsync, ::cuMemFree, ::cuMemFreeHost, * ::cuMemGetAddressRange, ::cuMemGetInfo, ::cuMemHostAlloc, @@ -6698,7 +7030,9 @@ CUresult CUDAAPI cuMemcpyDtoDAsync(CUdeviceptr dstDevice, CUdeviceptr srcDevice, * ::cuMemsetD32, ::cuMemsetD32Async, * ::cudaMemcpyToArrayAsync */ -CUresult CUDAAPI cuMemcpyHtoAAsync(CUarray dstArray, size_t dstOffset, const void *srcHost, size_t ByteCount, CUstream hStream); +CUresult CUDAAPI cuMemcpyHtoAAsync(CUarray dstArray, size_t dstOffset, + const void *srcHost, size_t ByteCount, + CUstream hStream); /** * \brief Copies memory from Array to Host @@ -6739,7 +7073,9 @@ CUresult CUDAAPI cuMemcpyHtoAAsync(CUarray dstArray, size_t dstOffset, const voi * ::cuMemsetD32, ::cuMemsetD32Async, * ::cudaMemcpyFromArrayAsync */ -CUresult CUDAAPI cuMemcpyAtoHAsync(void *dstHost, CUarray srcArray, size_t srcOffset, size_t ByteCount, CUstream hStream); +CUresult CUDAAPI cuMemcpyAtoHAsync(void *dstHost, CUarray srcArray, + size_t srcOffset, size_t ByteCount, + CUstream hStream); /** * \brief Copies memory for 2D arrays @@ -6896,7 +7232,8 @@ CUresult CUDAAPI cuMemcpyAtoHAsync(void *dstHost, CUarray srcArray, size_t srcOf * ::cuArrayDestroy, ::cuArrayGetDescriptor, ::cuMemAlloc, ::cuMemAllocHost, * ::cuMemAllocPitch, ::cuMemcpy2D, ::cuMemcpy2DUnaligned, * ::cuMemcpy3D, ::cuMemcpy3DAsync, ::cuMemcpyAtoA, ::cuMemcpyAtoD, - * ::cuMemcpyAtoH, ::cuMemcpyAtoHAsync, ::cuMemcpyDtoA, ::cuMemcpyDtoD, ::cuMemcpyDtoDAsync, + * ::cuMemcpyAtoH, ::cuMemcpyAtoHAsync, ::cuMemcpyDtoA, ::cuMemcpyDtoD, + ::cuMemcpyDtoDAsync, * ::cuMemcpyDtoH, ::cuMemcpyDtoHAsync, ::cuMemcpyHtoA, ::cuMemcpyHtoAAsync, * ::cuMemcpyHtoD, ::cuMemcpyHtoDAsync, ::cuMemFree, ::cuMemFreeHost, * ::cuMemGetAddressRange, ::cuMemGetInfo, ::cuMemHostAlloc, @@ -6926,7 +7263,8 @@ CUresult CUDAAPI cuMemcpy2DAsync(const CUDA_MEMCPY2D *pCopy, CUstream hStream); CUdeviceptr srcDevice; CUarray srcArray; unsigned int srcPitch; // ignored when src is array - unsigned int srcHeight; // ignored when src is array; may be 0 if Depth==1 + unsigned int srcHeight; // ignored when src is array; may be 0 + if Depth==1 unsigned int dstXInBytes, dstY, dstZ; unsigned int dstLOD; @@ -6935,7 +7273,8 @@ CUresult CUDAAPI cuMemcpy2DAsync(const CUDA_MEMCPY2D *pCopy, CUstream hStream); CUdeviceptr dstDevice; CUarray dstArray; unsigned int dstPitch; // ignored when dst is array - unsigned int dstHeight; // ignored when dst is array; may be 0 if Depth==1 + unsigned int dstHeight; // ignored when dst is array; may be 0 + if Depth==1 unsigned int WidthInBytes; unsigned int Height; @@ -7007,7 +7346,8 @@ CUresult CUDAAPI cuMemcpy2DAsync(const CUDA_MEMCPY2D *pCopy, CUstream hStream); * \par * For host pointers, the starting address is * \code - void* Start = (void*)((char*)srcHost+(srcZ*srcHeight+srcY)*srcPitch + srcXInBytes); + void* Start = (void*)((char*)srcHost+(srcZ*srcHeight+srcY)*srcPitch + + srcXInBytes); * \endcode * * \par @@ -7026,7 +7366,8 @@ CUresult CUDAAPI cuMemcpy2DAsync(const CUDA_MEMCPY2D *pCopy, CUstream hStream); * \par * For host pointers, the base address is * \code - void* dstStart = (void*)((char*)dstHost+(dstZ*dstHeight+dstY)*dstPitch + dstXInBytes); + void* dstStart = (void*)((char*)dstHost+(dstZ*dstHeight+dstY)*dstPitch + + dstXInBytes); * \endcode * * \par @@ -7072,7 +7413,8 @@ CUresult CUDAAPI cuMemcpy2DAsync(const CUDA_MEMCPY2D *pCopy, CUstream hStream); * ::cuArrayDestroy, ::cuArrayGetDescriptor, ::cuMemAlloc, ::cuMemAllocHost, * ::cuMemAllocPitch, ::cuMemcpy2D, ::cuMemcpy2DAsync, ::cuMemcpy2DUnaligned, * ::cuMemcpy3D, ::cuMemcpyAtoA, ::cuMemcpyAtoD, - * ::cuMemcpyAtoH, ::cuMemcpyAtoHAsync, ::cuMemcpyDtoA, ::cuMemcpyDtoD, ::cuMemcpyDtoDAsync, + * ::cuMemcpyAtoH, ::cuMemcpyAtoHAsync, ::cuMemcpyDtoA, ::cuMemcpyDtoD, + ::cuMemcpyDtoDAsync, * ::cuMemcpyDtoH, ::cuMemcpyDtoHAsync, ::cuMemcpyHtoA, ::cuMemcpyHtoAAsync, * ::cuMemcpyHtoD, ::cuMemcpyHtoDAsync, ::cuMemFree, ::cuMemFreeHost, * ::cuMemGetAddressRange, ::cuMemGetInfo, ::cuMemHostAlloc, @@ -7110,7 +7452,8 @@ CUresult CUDAAPI cuMemcpy3DAsync(const CUDA_MEMCPY3D *pCopy, CUstream hStream); * ::cuMemcpy3DPeerAsync, * ::cudaMemcpy3DPeerAsync */ -CUresult CUDAAPI cuMemcpy3DPeerAsync(const CUDA_MEMCPY3D_PEER *pCopy, CUstream hStream); +CUresult CUDAAPI cuMemcpy3DPeerAsync(const CUDA_MEMCPY3D_PEER *pCopy, + CUstream hStream); #endif /* __CUDA_API_VERSION >= 4000 */ #if __CUDA_API_VERSION >= 3020 @@ -7137,7 +7480,8 @@ CUresult CUDAAPI cuMemcpy3DPeerAsync(const CUDA_MEMCPY3D_PEER *pCopy, CUstream h * ::cuArrayDestroy, ::cuArrayGetDescriptor, ::cuMemAlloc, ::cuMemAllocHost, * ::cuMemAllocPitch, ::cuMemcpy2D, ::cuMemcpy2DAsync, ::cuMemcpy2DUnaligned, * ::cuMemcpy3D, ::cuMemcpy3DAsync, ::cuMemcpyAtoA, ::cuMemcpyAtoD, - * ::cuMemcpyAtoH, ::cuMemcpyAtoHAsync, ::cuMemcpyDtoA, ::cuMemcpyDtoD, ::cuMemcpyDtoDAsync, + * ::cuMemcpyAtoH, ::cuMemcpyAtoHAsync, ::cuMemcpyDtoA, ::cuMemcpyDtoD, + * ::cuMemcpyDtoDAsync, * ::cuMemcpyDtoH, ::cuMemcpyDtoHAsync, ::cuMemcpyHtoA, ::cuMemcpyHtoAAsync, * ::cuMemcpyHtoD, ::cuMemcpyHtoDAsync, ::cuMemFree, ::cuMemFreeHost, * ::cuMemGetAddressRange, ::cuMemGetInfo, ::cuMemHostAlloc, @@ -7172,7 +7516,8 @@ CUresult CUDAAPI cuMemsetD8(CUdeviceptr dstDevice, unsigned char uc, size_t N); * ::cuArrayDestroy, ::cuArrayGetDescriptor, ::cuMemAlloc, ::cuMemAllocHost, * ::cuMemAllocPitch, ::cuMemcpy2D, ::cuMemcpy2DAsync, ::cuMemcpy2DUnaligned, * ::cuMemcpy3D, ::cuMemcpy3DAsync, ::cuMemcpyAtoA, ::cuMemcpyAtoD, - * ::cuMemcpyAtoH, ::cuMemcpyAtoHAsync, ::cuMemcpyDtoA, ::cuMemcpyDtoD, ::cuMemcpyDtoDAsync, + * ::cuMemcpyAtoH, ::cuMemcpyAtoHAsync, ::cuMemcpyDtoA, ::cuMemcpyDtoD, + * ::cuMemcpyDtoDAsync, * ::cuMemcpyDtoH, ::cuMemcpyDtoHAsync, ::cuMemcpyHtoA, ::cuMemcpyHtoAAsync, * ::cuMemcpyHtoD, ::cuMemcpyHtoDAsync, ::cuMemFree, ::cuMemFreeHost, * ::cuMemGetAddressRange, ::cuMemGetInfo, ::cuMemHostAlloc, @@ -7182,7 +7527,8 @@ CUresult CUDAAPI cuMemsetD8(CUdeviceptr dstDevice, unsigned char uc, size_t N); * ::cuMemsetD32, ::cuMemsetD32Async, * ::cudaMemset */ -CUresult CUDAAPI cuMemsetD16(CUdeviceptr dstDevice, unsigned short us, size_t N); +CUresult CUDAAPI cuMemsetD16(CUdeviceptr dstDevice, unsigned short us, + size_t N); /** * \brief Initializes device memory @@ -7207,7 +7553,8 @@ CUresult CUDAAPI cuMemsetD16(CUdeviceptr dstDevice, unsigned short us, size_t N) * ::cuArrayDestroy, ::cuArrayGetDescriptor, ::cuMemAlloc, ::cuMemAllocHost, * ::cuMemAllocPitch, ::cuMemcpy2D, ::cuMemcpy2DAsync, ::cuMemcpy2DUnaligned, * ::cuMemcpy3D, ::cuMemcpy3DAsync, ::cuMemcpyAtoA, ::cuMemcpyAtoD, - * ::cuMemcpyAtoH, ::cuMemcpyAtoHAsync, ::cuMemcpyDtoA, ::cuMemcpyDtoD, ::cuMemcpyDtoDAsync, + * ::cuMemcpyAtoH, ::cuMemcpyAtoHAsync, ::cuMemcpyDtoA, ::cuMemcpyDtoD, + * ::cuMemcpyDtoDAsync, * ::cuMemcpyDtoH, ::cuMemcpyDtoHAsync, ::cuMemcpyHtoA, ::cuMemcpyHtoAAsync, * ::cuMemcpyHtoD, ::cuMemcpyHtoDAsync, ::cuMemFree, ::cuMemFreeHost, * ::cuMemGetAddressRange, ::cuMemGetInfo, ::cuMemHostAlloc, @@ -7247,7 +7594,8 @@ CUresult CUDAAPI cuMemsetD32(CUdeviceptr dstDevice, unsigned int ui, size_t N); * ::cuArrayDestroy, ::cuArrayGetDescriptor, ::cuMemAlloc, ::cuMemAllocHost, * ::cuMemAllocPitch, ::cuMemcpy2D, ::cuMemcpy2DAsync, ::cuMemcpy2DUnaligned, * ::cuMemcpy3D, ::cuMemcpy3DAsync, ::cuMemcpyAtoA, ::cuMemcpyAtoD, - * ::cuMemcpyAtoH, ::cuMemcpyAtoHAsync, ::cuMemcpyDtoA, ::cuMemcpyDtoD, ::cuMemcpyDtoDAsync, + * ::cuMemcpyAtoH, ::cuMemcpyAtoHAsync, ::cuMemcpyDtoA, ::cuMemcpyDtoD, + * ::cuMemcpyDtoDAsync, * ::cuMemcpyDtoH, ::cuMemcpyDtoHAsync, ::cuMemcpyHtoA, ::cuMemcpyHtoAAsync, * ::cuMemcpyHtoD, ::cuMemcpyHtoDAsync, ::cuMemFree, ::cuMemFreeHost, * ::cuMemGetAddressRange, ::cuMemGetInfo, ::cuMemHostAlloc, @@ -7257,7 +7605,8 @@ CUresult CUDAAPI cuMemsetD32(CUdeviceptr dstDevice, unsigned int ui, size_t N); * ::cuMemsetD32, ::cuMemsetD32Async, * ::cudaMemset2D */ -CUresult CUDAAPI cuMemsetD2D8(CUdeviceptr dstDevice, size_t dstPitch, unsigned char uc, size_t Width, size_t Height); +CUresult CUDAAPI cuMemsetD2D8(CUdeviceptr dstDevice, size_t dstPitch, + unsigned char uc, size_t Width, size_t Height); /** * \brief Initializes device memory @@ -7288,7 +7637,8 @@ CUresult CUDAAPI cuMemsetD2D8(CUdeviceptr dstDevice, size_t dstPitch, unsigned c * ::cuArrayDestroy, ::cuArrayGetDescriptor, ::cuMemAlloc, ::cuMemAllocHost, * ::cuMemAllocPitch, ::cuMemcpy2D, ::cuMemcpy2DAsync, ::cuMemcpy2DUnaligned, * ::cuMemcpy3D, ::cuMemcpy3DAsync, ::cuMemcpyAtoA, ::cuMemcpyAtoD, - * ::cuMemcpyAtoH, ::cuMemcpyAtoHAsync, ::cuMemcpyDtoA, ::cuMemcpyDtoD, ::cuMemcpyDtoDAsync, + * ::cuMemcpyAtoH, ::cuMemcpyAtoHAsync, ::cuMemcpyDtoA, ::cuMemcpyDtoD, + * ::cuMemcpyDtoDAsync, * ::cuMemcpyDtoH, ::cuMemcpyDtoHAsync, ::cuMemcpyHtoA, ::cuMemcpyHtoAAsync, * ::cuMemcpyHtoD, ::cuMemcpyHtoDAsync, ::cuMemFree, ::cuMemFreeHost, * ::cuMemGetAddressRange, ::cuMemGetInfo, ::cuMemHostAlloc, @@ -7298,7 +7648,8 @@ CUresult CUDAAPI cuMemsetD2D8(CUdeviceptr dstDevice, size_t dstPitch, unsigned c * ::cuMemsetD32, ::cuMemsetD32Async, * ::cudaMemset2D */ -CUresult CUDAAPI cuMemsetD2D16(CUdeviceptr dstDevice, size_t dstPitch, unsigned short us, size_t Width, size_t Height); +CUresult CUDAAPI cuMemsetD2D16(CUdeviceptr dstDevice, size_t dstPitch, + unsigned short us, size_t Width, size_t Height); /** * \brief Initializes device memory @@ -7329,7 +7680,8 @@ CUresult CUDAAPI cuMemsetD2D16(CUdeviceptr dstDevice, size_t dstPitch, unsigned * ::cuArrayDestroy, ::cuArrayGetDescriptor, ::cuMemAlloc, ::cuMemAllocHost, * ::cuMemAllocPitch, ::cuMemcpy2D, ::cuMemcpy2DAsync, ::cuMemcpy2DUnaligned, * ::cuMemcpy3D, ::cuMemcpy3DAsync, ::cuMemcpyAtoA, ::cuMemcpyAtoD, - * ::cuMemcpyAtoH, ::cuMemcpyAtoHAsync, ::cuMemcpyDtoA, ::cuMemcpyDtoD, ::cuMemcpyDtoDAsync, + * ::cuMemcpyAtoH, ::cuMemcpyAtoHAsync, ::cuMemcpyDtoA, ::cuMemcpyDtoD, + * ::cuMemcpyDtoDAsync, * ::cuMemcpyDtoH, ::cuMemcpyDtoHAsync, ::cuMemcpyHtoA, ::cuMemcpyHtoAAsync, * ::cuMemcpyHtoD, ::cuMemcpyHtoDAsync, ::cuMemFree, ::cuMemFreeHost, * ::cuMemGetAddressRange, ::cuMemGetInfo, ::cuMemHostAlloc, @@ -7339,7 +7691,8 @@ CUresult CUDAAPI cuMemsetD2D16(CUdeviceptr dstDevice, size_t dstPitch, unsigned * ::cuMemsetD32, ::cuMemsetD32Async, * ::cudaMemset2D */ -CUresult CUDAAPI cuMemsetD2D32(CUdeviceptr dstDevice, size_t dstPitch, unsigned int ui, size_t Width, size_t Height); +CUresult CUDAAPI cuMemsetD2D32(CUdeviceptr dstDevice, size_t dstPitch, + unsigned int ui, size_t Width, size_t Height); /** * \brief Sets device memory @@ -7366,7 +7719,8 @@ CUresult CUDAAPI cuMemsetD2D32(CUdeviceptr dstDevice, size_t dstPitch, unsigned * ::cuArrayDestroy, ::cuArrayGetDescriptor, ::cuMemAlloc, ::cuMemAllocHost, * ::cuMemAllocPitch, ::cuMemcpy2D, ::cuMemcpy2DAsync, ::cuMemcpy2DUnaligned, * ::cuMemcpy3D, ::cuMemcpy3DAsync, ::cuMemcpyAtoA, ::cuMemcpyAtoD, - * ::cuMemcpyAtoH, ::cuMemcpyAtoHAsync, ::cuMemcpyDtoA, ::cuMemcpyDtoD, ::cuMemcpyDtoDAsync, + * ::cuMemcpyAtoH, ::cuMemcpyAtoHAsync, ::cuMemcpyDtoA, ::cuMemcpyDtoD, + * ::cuMemcpyDtoDAsync, * ::cuMemcpyDtoH, ::cuMemcpyDtoHAsync, ::cuMemcpyHtoA, ::cuMemcpyHtoAAsync, * ::cuMemcpyHtoD, ::cuMemcpyHtoDAsync, ::cuMemFree, ::cuMemFreeHost, * ::cuMemGetAddressRange, ::cuMemGetInfo, ::cuMemHostAlloc, @@ -7376,7 +7730,8 @@ CUresult CUDAAPI cuMemsetD2D32(CUdeviceptr dstDevice, size_t dstPitch, unsigned * ::cuMemsetD32, ::cuMemsetD32Async, * ::cudaMemsetAsync */ -CUresult CUDAAPI cuMemsetD8Async(CUdeviceptr dstDevice, unsigned char uc, size_t N, CUstream hStream); +CUresult CUDAAPI cuMemsetD8Async(CUdeviceptr dstDevice, unsigned char uc, + size_t N, CUstream hStream); /** * \brief Sets device memory @@ -7403,7 +7758,8 @@ CUresult CUDAAPI cuMemsetD8Async(CUdeviceptr dstDevice, unsigned char uc, size_t * ::cuArrayDestroy, ::cuArrayGetDescriptor, ::cuMemAlloc, ::cuMemAllocHost, * ::cuMemAllocPitch, ::cuMemcpy2D, ::cuMemcpy2DAsync, ::cuMemcpy2DUnaligned, * ::cuMemcpy3D, ::cuMemcpy3DAsync, ::cuMemcpyAtoA, ::cuMemcpyAtoD, - * ::cuMemcpyAtoH, ::cuMemcpyAtoHAsync, ::cuMemcpyDtoA, ::cuMemcpyDtoD, ::cuMemcpyDtoDAsync, + * ::cuMemcpyAtoH, ::cuMemcpyAtoHAsync, ::cuMemcpyDtoA, ::cuMemcpyDtoD, + * ::cuMemcpyDtoDAsync, * ::cuMemcpyDtoH, ::cuMemcpyDtoHAsync, ::cuMemcpyHtoA, ::cuMemcpyHtoAAsync, * ::cuMemcpyHtoD, ::cuMemcpyHtoDAsync, ::cuMemFree, ::cuMemFreeHost, * ::cuMemGetAddressRange, ::cuMemGetInfo, ::cuMemHostAlloc, @@ -7413,7 +7769,8 @@ CUresult CUDAAPI cuMemsetD8Async(CUdeviceptr dstDevice, unsigned char uc, size_t * ::cuMemsetD32, ::cuMemsetD32Async, * ::cudaMemsetAsync */ -CUresult CUDAAPI cuMemsetD16Async(CUdeviceptr dstDevice, unsigned short us, size_t N, CUstream hStream); +CUresult CUDAAPI cuMemsetD16Async(CUdeviceptr dstDevice, unsigned short us, + size_t N, CUstream hStream); /** * \brief Sets device memory @@ -7440,16 +7797,19 @@ CUresult CUDAAPI cuMemsetD16Async(CUdeviceptr dstDevice, unsigned short us, size * ::cuArrayDestroy, ::cuArrayGetDescriptor, ::cuMemAlloc, ::cuMemAllocHost, * ::cuMemAllocPitch, ::cuMemcpy2D, ::cuMemcpy2DAsync, ::cuMemcpy2DUnaligned, * ::cuMemcpy3D, ::cuMemcpy3DAsync, ::cuMemcpyAtoA, ::cuMemcpyAtoD, - * ::cuMemcpyAtoH, ::cuMemcpyAtoHAsync, ::cuMemcpyDtoA, ::cuMemcpyDtoD, ::cuMemcpyDtoDAsync, + * ::cuMemcpyAtoH, ::cuMemcpyAtoHAsync, ::cuMemcpyDtoA, ::cuMemcpyDtoD, + * ::cuMemcpyDtoDAsync, * ::cuMemcpyDtoH, ::cuMemcpyDtoHAsync, ::cuMemcpyHtoA, ::cuMemcpyHtoAAsync, * ::cuMemcpyHtoD, ::cuMemcpyHtoDAsync, ::cuMemFree, ::cuMemFreeHost, * ::cuMemGetAddressRange, ::cuMemGetInfo, ::cuMemHostAlloc, * ::cuMemHostGetDevicePointer, ::cuMemsetD2D8, ::cuMemsetD2D8Async, * ::cuMemsetD2D16, ::cuMemsetD2D16Async, ::cuMemsetD2D32, ::cuMemsetD2D32Async, - * ::cuMemsetD8, ::cuMemsetD8Async, ::cuMemsetD16, ::cuMemsetD16Async, ::cuMemsetD32, + * ::cuMemsetD8, ::cuMemsetD8Async, ::cuMemsetD16, ::cuMemsetD16Async, + * ::cuMemsetD32, * ::cudaMemsetAsync */ -CUresult CUDAAPI cuMemsetD32Async(CUdeviceptr dstDevice, unsigned int ui, size_t N, CUstream hStream); +CUresult CUDAAPI cuMemsetD32Async(CUdeviceptr dstDevice, unsigned int ui, + size_t N, CUstream hStream); /** * \brief Sets device memory @@ -7481,7 +7841,8 @@ CUresult CUDAAPI cuMemsetD32Async(CUdeviceptr dstDevice, unsigned int ui, size_t * ::cuArrayDestroy, ::cuArrayGetDescriptor, ::cuMemAlloc, ::cuMemAllocHost, * ::cuMemAllocPitch, ::cuMemcpy2D, ::cuMemcpy2DAsync, ::cuMemcpy2DUnaligned, * ::cuMemcpy3D, ::cuMemcpy3DAsync, ::cuMemcpyAtoA, ::cuMemcpyAtoD, - * ::cuMemcpyAtoH, ::cuMemcpyAtoHAsync, ::cuMemcpyDtoA, ::cuMemcpyDtoD, ::cuMemcpyDtoDAsync, + * ::cuMemcpyAtoH, ::cuMemcpyAtoHAsync, ::cuMemcpyDtoA, ::cuMemcpyDtoD, + * ::cuMemcpyDtoDAsync, * ::cuMemcpyDtoH, ::cuMemcpyDtoHAsync, ::cuMemcpyHtoA, ::cuMemcpyHtoAAsync, * ::cuMemcpyHtoD, ::cuMemcpyHtoDAsync, ::cuMemFree, ::cuMemFreeHost, * ::cuMemGetAddressRange, ::cuMemGetInfo, ::cuMemHostAlloc, @@ -7491,7 +7852,9 @@ CUresult CUDAAPI cuMemsetD32Async(CUdeviceptr dstDevice, unsigned int ui, size_t * ::cuMemsetD32, ::cuMemsetD32Async, * ::cudaMemset2DAsync */ -CUresult CUDAAPI cuMemsetD2D8Async(CUdeviceptr dstDevice, size_t dstPitch, unsigned char uc, size_t Width, size_t Height, CUstream hStream); +CUresult CUDAAPI cuMemsetD2D8Async(CUdeviceptr dstDevice, size_t dstPitch, + unsigned char uc, size_t Width, + size_t Height, CUstream hStream); /** * \brief Sets device memory @@ -7524,7 +7887,8 @@ CUresult CUDAAPI cuMemsetD2D8Async(CUdeviceptr dstDevice, size_t dstPitch, unsig * ::cuArrayDestroy, ::cuArrayGetDescriptor, ::cuMemAlloc, ::cuMemAllocHost, * ::cuMemAllocPitch, ::cuMemcpy2D, ::cuMemcpy2DAsync, ::cuMemcpy2DUnaligned, * ::cuMemcpy3D, ::cuMemcpy3DAsync, ::cuMemcpyAtoA, ::cuMemcpyAtoD, - * ::cuMemcpyAtoH, ::cuMemcpyAtoHAsync, ::cuMemcpyDtoA, ::cuMemcpyDtoD, ::cuMemcpyDtoDAsync, + * ::cuMemcpyAtoH, ::cuMemcpyAtoHAsync, ::cuMemcpyDtoA, ::cuMemcpyDtoD, + * ::cuMemcpyDtoDAsync, * ::cuMemcpyDtoH, ::cuMemcpyDtoHAsync, ::cuMemcpyHtoA, ::cuMemcpyHtoAAsync, * ::cuMemcpyHtoD, ::cuMemcpyHtoDAsync, ::cuMemFree, ::cuMemFreeHost, * ::cuMemGetAddressRange, ::cuMemGetInfo, ::cuMemHostAlloc, @@ -7534,7 +7898,9 @@ CUresult CUDAAPI cuMemsetD2D8Async(CUdeviceptr dstDevice, size_t dstPitch, unsig * ::cuMemsetD32, ::cuMemsetD32Async, * ::cudaMemset2DAsync */ -CUresult CUDAAPI cuMemsetD2D16Async(CUdeviceptr dstDevice, size_t dstPitch, unsigned short us, size_t Width, size_t Height, CUstream hStream); +CUresult CUDAAPI cuMemsetD2D16Async(CUdeviceptr dstDevice, size_t dstPitch, + unsigned short us, size_t Width, + size_t Height, CUstream hStream); /** * \brief Sets device memory @@ -7567,7 +7933,8 @@ CUresult CUDAAPI cuMemsetD2D16Async(CUdeviceptr dstDevice, size_t dstPitch, unsi * ::cuArrayDestroy, ::cuArrayGetDescriptor, ::cuMemAlloc, ::cuMemAllocHost, * ::cuMemAllocPitch, ::cuMemcpy2D, ::cuMemcpy2DAsync, ::cuMemcpy2DUnaligned, * ::cuMemcpy3D, ::cuMemcpy3DAsync, ::cuMemcpyAtoA, ::cuMemcpyAtoD, - * ::cuMemcpyAtoH, ::cuMemcpyAtoHAsync, ::cuMemcpyDtoA, ::cuMemcpyDtoD, ::cuMemcpyDtoDAsync, + * ::cuMemcpyAtoH, ::cuMemcpyAtoHAsync, ::cuMemcpyDtoA, ::cuMemcpyDtoD, + * ::cuMemcpyDtoDAsync, * ::cuMemcpyDtoH, ::cuMemcpyDtoHAsync, ::cuMemcpyHtoA, ::cuMemcpyHtoAAsync, * ::cuMemcpyHtoD, ::cuMemcpyHtoDAsync, ::cuMemFree, ::cuMemFreeHost, * ::cuMemGetAddressRange, ::cuMemGetInfo, ::cuMemHostAlloc, @@ -7577,7 +7944,9 @@ CUresult CUDAAPI cuMemsetD2D16Async(CUdeviceptr dstDevice, size_t dstPitch, unsi * ::cuMemsetD32, ::cuMemsetD32Async, * ::cudaMemset2DAsync */ -CUresult CUDAAPI cuMemsetD2D32Async(CUdeviceptr dstDevice, size_t dstPitch, unsigned int ui, size_t Width, size_t Height, CUstream hStream); +CUresult CUDAAPI cuMemsetD2D32Async(CUdeviceptr dstDevice, size_t dstPitch, + unsigned int ui, size_t Width, + size_t Height, CUstream hStream); /** * \brief Creates a 1D or 2D CUDA array @@ -7673,7 +8042,8 @@ CUresult CUDAAPI cuMemsetD2D32Async(CUdeviceptr dstDevice, size_t dstPitch, unsi * ::cuArrayDestroy, ::cuArrayGetDescriptor, ::cuMemAlloc, ::cuMemAllocHost, * ::cuMemAllocPitch, ::cuMemcpy2D, ::cuMemcpy2DAsync, ::cuMemcpy2DUnaligned, * ::cuMemcpy3D, ::cuMemcpy3DAsync, ::cuMemcpyAtoA, ::cuMemcpyAtoD, - * ::cuMemcpyAtoH, ::cuMemcpyAtoHAsync, ::cuMemcpyDtoA, ::cuMemcpyDtoD, ::cuMemcpyDtoDAsync, + * ::cuMemcpyAtoH, ::cuMemcpyAtoHAsync, ::cuMemcpyDtoA, ::cuMemcpyDtoD, + ::cuMemcpyDtoDAsync, * ::cuMemcpyDtoH, ::cuMemcpyDtoHAsync, ::cuMemcpyHtoA, ::cuMemcpyHtoAAsync, * ::cuMemcpyHtoD, ::cuMemcpyHtoDAsync, ::cuMemFree, ::cuMemFreeHost, * ::cuMemGetAddressRange, ::cuMemGetInfo, ::cuMemHostAlloc, @@ -7681,7 +8051,8 @@ CUresult CUDAAPI cuMemsetD2D32Async(CUdeviceptr dstDevice, size_t dstPitch, unsi * ::cuMemsetD2D32, ::cuMemsetD8, ::cuMemsetD16, ::cuMemsetD32, * ::cudaMallocArray */ -CUresult CUDAAPI cuArrayCreate(CUarray *pHandle, const CUDA_ARRAY_DESCRIPTOR *pAllocateArray); +CUresult CUDAAPI cuArrayCreate(CUarray *pHandle, + const CUDA_ARRAY_DESCRIPTOR *pAllocateArray); /** * \brief Get a 1D or 2D CUDA array descriptor @@ -7707,7 +8078,8 @@ CUresult CUDAAPI cuArrayCreate(CUarray *pHandle, const CUDA_ARRAY_DESCRIPTOR *pA * ::cuArrayDestroy, ::cuMemAlloc, ::cuMemAllocHost, * ::cuMemAllocPitch, ::cuMemcpy2D, ::cuMemcpy2DAsync, ::cuMemcpy2DUnaligned, * ::cuMemcpy3D, ::cuMemcpy3DAsync, ::cuMemcpyAtoA, ::cuMemcpyAtoD, - * ::cuMemcpyAtoH, ::cuMemcpyAtoHAsync, ::cuMemcpyDtoA, ::cuMemcpyDtoD, ::cuMemcpyDtoDAsync, + * ::cuMemcpyAtoH, ::cuMemcpyAtoHAsync, ::cuMemcpyDtoA, ::cuMemcpyDtoD, + * ::cuMemcpyDtoDAsync, * ::cuMemcpyDtoH, ::cuMemcpyDtoHAsync, ::cuMemcpyHtoA, ::cuMemcpyHtoAAsync, * ::cuMemcpyHtoD, ::cuMemcpyHtoDAsync, ::cuMemFree, ::cuMemFreeHost, * ::cuMemGetAddressRange, ::cuMemGetInfo, ::cuMemHostAlloc, @@ -7715,10 +8087,10 @@ CUresult CUDAAPI cuArrayCreate(CUarray *pHandle, const CUDA_ARRAY_DESCRIPTOR *pA * ::cuMemsetD2D32, ::cuMemsetD8, ::cuMemsetD16, ::cuMemsetD32, * ::cudaArrayGetInfo */ -CUresult CUDAAPI cuArrayGetDescriptor(CUDA_ARRAY_DESCRIPTOR *pArrayDescriptor, CUarray hArray); +CUresult CUDAAPI cuArrayGetDescriptor(CUDA_ARRAY_DESCRIPTOR *pArrayDescriptor, + CUarray hArray); #endif /* __CUDA_API_VERSION >= 3020 */ - /** * \brief Destroys a CUDA array * @@ -7740,7 +8112,8 @@ CUresult CUDAAPI cuArrayGetDescriptor(CUDA_ARRAY_DESCRIPTOR *pArrayDescriptor, C * ::cuArrayGetDescriptor, ::cuMemAlloc, ::cuMemAllocHost, * ::cuMemAllocPitch, ::cuMemcpy2D, ::cuMemcpy2DAsync, ::cuMemcpy2DUnaligned, * ::cuMemcpy3D, ::cuMemcpy3DAsync, ::cuMemcpyAtoA, ::cuMemcpyAtoD, - * ::cuMemcpyAtoH, ::cuMemcpyAtoHAsync, ::cuMemcpyDtoA, ::cuMemcpyDtoD, ::cuMemcpyDtoDAsync, + * ::cuMemcpyAtoH, ::cuMemcpyAtoHAsync, ::cuMemcpyDtoA, ::cuMemcpyDtoD, + * ::cuMemcpyDtoDAsync, * ::cuMemcpyDtoH, ::cuMemcpyDtoHAsync, ::cuMemcpyHtoA, ::cuMemcpyHtoAAsync, * ::cuMemcpyHtoD, ::cuMemcpyHtoDAsync, ::cuMemFree, ::cuMemFreeHost, * ::cuMemGetAddressRange, ::cuMemGetInfo, ::cuMemHostAlloc, @@ -7771,26 +8144,40 @@ CUresult CUDAAPI cuArrayDestroy(CUarray hArray); * where: * * - \p Width, \p Height, and \p Depth are the width, height, and depth of the - * CUDA array (in elements); the following types of CUDA arrays can be allocated: - * - A 1D array is allocated if \p Height and \p Depth extents are both zero. + * CUDA array (in elements); the following types of CUDA arrays can be + allocated: + * - A 1D array is allocated if \p Height and \p Depth extents are both + zero. * - A 2D array is allocated if only \p Depth extent is zero. * - A 3D array is allocated if all three extents are non-zero. * - A 1D layered CUDA array is allocated if only \p Height is zero and the - * ::CUDA_ARRAY3D_LAYERED flag is set. Each layer is a 1D array. The number + * ::CUDA_ARRAY3D_LAYERED flag is set. Each layer is a 1D array. The + number * of layers is determined by the depth extent. - * - A 2D layered CUDA array is allocated if all three extents are non-zero and - * the ::CUDA_ARRAY3D_LAYERED flag is set. Each layer is a 2D array. The number + * - A 2D layered CUDA array is allocated if all three extents are non-zero + and + * the ::CUDA_ARRAY3D_LAYERED flag is set. Each layer is a 2D array. The + number * of layers is determined by the depth extent. - * - A cubemap CUDA array is allocated if all three extents are non-zero and the - * ::CUDA_ARRAY3D_CUBEMAP flag is set. \p Width must be equal to \p Height, and - * \p Depth must be six. A cubemap is a special type of 2D layered CUDA array, - * where the six layers represent the six faces of a cube. The order of the six + * - A cubemap CUDA array is allocated if all three extents are non-zero and + the + * ::CUDA_ARRAY3D_CUBEMAP flag is set. \p Width must be equal to \p + Height, and + * \p Depth must be six. A cubemap is a special type of 2D layered CUDA + array, + * where the six layers represent the six faces of a cube. The order of + the six * layers in memory is the same as that listed in ::CUarray_cubemap_face. - * - A cubemap layered CUDA array is allocated if all three extents are non-zero, - * and both, ::CUDA_ARRAY3D_CUBEMAP and ::CUDA_ARRAY3D_LAYERED flags are set. - * \p Width must be equal to \p Height, and \p Depth must be a multiple of six. - * A cubemap layered CUDA array is a special type of 2D layered CUDA array that - * consists of a collection of cubemaps. The first six layers represent the first + * - A cubemap layered CUDA array is allocated if all three extents are + non-zero, + * and both, ::CUDA_ARRAY3D_CUBEMAP and ::CUDA_ARRAY3D_LAYERED flags are + set. + * \p Width must be equal to \p Height, and \p Depth must be a multiple of + six. + * A cubemap layered CUDA array is a special type of 2D layered CUDA array + that + * consists of a collection of cubemaps. The first six layers represent + the first * cubemap, the next six layers form the second cubemap, and so on. * * - ::Format specifies the format of the elements; ::CUarray_format is @@ -7812,29 +8199,41 @@ CUresult CUDAAPI cuArrayDestroy(CUarray hArray); * element; it may be 1, 2, or 4; * * - ::Flags may be set to - * - ::CUDA_ARRAY3D_LAYERED to enable creation of layered CUDA arrays. If this flag is set, + * - ::CUDA_ARRAY3D_LAYERED to enable creation of layered CUDA arrays. If this + flag is set, * \p Depth specifies the number of layers, not the depth of a 3D array. - * - ::CUDA_ARRAY3D_SURFACE_LDST to enable surface references to be bound to the CUDA array. - * If this flag is not set, ::cuSurfRefSetArray will fail when attempting to bind the CUDA array + * - ::CUDA_ARRAY3D_SURFACE_LDST to enable surface references to be bound to + the CUDA array. + * If this flag is not set, ::cuSurfRefSetArray will fail when attempting to + bind the CUDA array * to a surface reference. - * - ::CUDA_ARRAY3D_CUBEMAP to enable creation of cubemaps. If this flag is set, \p Width must be - * equal to \p Height, and \p Depth must be six. If the ::CUDA_ARRAY3D_LAYERED flag is also set, + * - ::CUDA_ARRAY3D_CUBEMAP to enable creation of cubemaps. If this flag is + set, \p Width must be + * equal to \p Height, and \p Depth must be six. If the + ::CUDA_ARRAY3D_LAYERED flag is also set, * then \p Depth must be a multiple of six. - * - ::CUDA_ARRAY3D_TEXTURE_GATHER to indicate that the CUDA array will be used for texture gather. + * - ::CUDA_ARRAY3D_TEXTURE_GATHER to indicate that the CUDA array will be + used for texture gather. * Texture gather can only be performed on 2D CUDA arrays. * - * \p Width, \p Height and \p Depth must meet certain size requirements as listed in the following table. - * All values are specified in elements. Note that for brevity's sake, the full name of the device attribute + * \p Width, \p Height and \p Depth must meet certain size requirements as + listed in the following table. + * All values are specified in elements. Note that for brevity's sake, the full + name of the device attribute * is not specified. For ex., TEXTURE1D_WIDTH refers to the device attribute * ::CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE1D_WIDTH. * - * Note that 2D CUDA arrays have different size requirements if the ::CUDA_ARRAY3D_TEXTURE_GATHER flag - * is set. \p Width and \p Height must not be greater than ::CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE2D_GATHER_WIDTH - * and ::CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE2D_GATHER_HEIGHT respectively, in that case. + * Note that 2D CUDA arrays have different size requirements if the + ::CUDA_ARRAY3D_TEXTURE_GATHER flag + * is set. \p Width and \p Height must not be greater than + ::CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE2D_GATHER_WIDTH + * and ::CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE2D_GATHER_HEIGHT respectively, in + that case. * * * - * * @@ -7861,13 +8260,16 @@ CUresult CUDAAPI cuArrayDestroy(CUarray hArray); * * - * + * * * - * - * *
CUDA array typeValid extents that must always be met
{(width range in elements), (height range), + * 		Valid extents that must always be met
{(width range in elements), + (height range), * (depth range)}		Valid extents with CUDA_ARRAY3D_SURFACE_LDST set
* {(width range in elements), (height range), (depth range)}
{ (1,SURFACE2D_LAYERED_WIDTH), (1,SURFACE2D_LAYERED_HEIGHT), * (1,SURFACE2D_LAYERED_LAYERS) }
Cubemap{ (1,TEXTURECUBEMAP_WIDTH), (1,TEXTURECUBEMAP_WIDTH), 6 }{ (1,TEXTURECUBEMAP_WIDTH), (1,TEXTURECUBEMAP_WIDTH), 6 + }{ (1,SURFACECUBEMAP_WIDTH), * (1,SURFACECUBEMAP_WIDTH), 6 }
Cubemap Layered{ (1,TEXTURECUBEMAP_LAYERED_WIDTH), (1,TEXTURECUBEMAP_LAYERED_WIDTH), + * { (1,TEXTURECUBEMAP_LAYERED_WIDTH), + (1,TEXTURECUBEMAP_LAYERED_WIDTH), * (1,TEXTURECUBEMAP_LAYERED_LAYERS) }{ (1,SURFACECUBEMAP_LAYERED_WIDTH), (1,SURFACECUBEMAP_LAYERED_WIDTH), + * { (1,SURFACECUBEMAP_LAYERED_WIDTH), + (1,SURFACECUBEMAP_LAYERED_WIDTH), * (1,SURFACECUBEMAP_LAYERED_LAYERS) }
* @@ -7921,7 +8323,8 @@ CUresult CUDAAPI cuArrayDestroy(CUarray hArray); * ::cuArrayDestroy, ::cuArrayGetDescriptor, ::cuMemAlloc, ::cuMemAllocHost, * ::cuMemAllocPitch, ::cuMemcpy2D, ::cuMemcpy2DAsync, ::cuMemcpy2DUnaligned, * ::cuMemcpy3D, ::cuMemcpy3DAsync, ::cuMemcpyAtoA, ::cuMemcpyAtoD, - * ::cuMemcpyAtoH, ::cuMemcpyAtoHAsync, ::cuMemcpyDtoA, ::cuMemcpyDtoD, ::cuMemcpyDtoDAsync, + * ::cuMemcpyAtoH, ::cuMemcpyAtoHAsync, ::cuMemcpyDtoA, ::cuMemcpyDtoD, + ::cuMemcpyDtoDAsync, * ::cuMemcpyDtoH, ::cuMemcpyDtoHAsync, ::cuMemcpyHtoA, ::cuMemcpyHtoAAsync, * ::cuMemcpyHtoD, ::cuMemcpyHtoDAsync, ::cuMemFree, ::cuMemFreeHost, * ::cuMemGetAddressRange, ::cuMemGetInfo, ::cuMemHostAlloc, @@ -7929,7 +8332,8 @@ CUresult CUDAAPI cuArrayDestroy(CUarray hArray); * ::cuMemsetD2D32, ::cuMemsetD8, ::cuMemsetD16, ::cuMemsetD32, * ::cudaMalloc3DArray */ -CUresult CUDAAPI cuArray3DCreate(CUarray *pHandle, const CUDA_ARRAY3D_DESCRIPTOR *pAllocateArray); +CUresult CUDAAPI cuArray3DCreate(CUarray *pHandle, + const CUDA_ARRAY3D_DESCRIPTOR *pAllocateArray); /** * \brief Get a 3D CUDA array descriptor @@ -7959,7 +8363,8 @@ CUresult CUDAAPI cuArray3DCreate(CUarray *pHandle, const CUDA_ARRAY3D_DESCRIPTOR * ::cuArrayDestroy, ::cuArrayGetDescriptor, ::cuMemAlloc, ::cuMemAllocHost, * ::cuMemAllocPitch, ::cuMemcpy2D, ::cuMemcpy2DAsync, ::cuMemcpy2DUnaligned, * ::cuMemcpy3D, ::cuMemcpy3DAsync, ::cuMemcpyAtoA, ::cuMemcpyAtoD, - * ::cuMemcpyAtoH, ::cuMemcpyAtoHAsync, ::cuMemcpyDtoA, ::cuMemcpyDtoD, ::cuMemcpyDtoDAsync, + * ::cuMemcpyAtoH, ::cuMemcpyAtoHAsync, ::cuMemcpyDtoA, ::cuMemcpyDtoD, + * ::cuMemcpyDtoDAsync, * ::cuMemcpyDtoH, ::cuMemcpyDtoHAsync, ::cuMemcpyHtoA, ::cuMemcpyHtoAAsync, * ::cuMemcpyHtoD, ::cuMemcpyHtoDAsync, ::cuMemFree, ::cuMemFreeHost, * ::cuMemGetAddressRange, ::cuMemGetInfo, ::cuMemHostAlloc, @@ -7967,7 +8372,8 @@ CUresult CUDAAPI cuArray3DCreate(CUarray *pHandle, const CUDA_ARRAY3D_DESCRIPTOR * ::cuMemsetD2D32, ::cuMemsetD8, ::cuMemsetD16, ::cuMemsetD32, * ::cudaArrayGetInfo */ -CUresult CUDAAPI cuArray3DGetDescriptor(CUDA_ARRAY3D_DESCRIPTOR *pArrayDescriptor, CUarray hArray); +CUresult CUDAAPI cuArray3DGetDescriptor( + CUDA_ARRAY3D_DESCRIPTOR *pArrayDescriptor, CUarray hArray); #endif /* __CUDA_API_VERSION >= 3020 */ #if __CUDA_API_VERSION >= 5000 @@ -7975,9 +8381,12 @@ CUresult CUDAAPI cuArray3DGetDescriptor(CUDA_ARRAY3D_DESCRIPTOR *pArrayDescripto /** * \brief Creates a CUDA mipmapped array * - * Creates a CUDA mipmapped array according to the ::CUDA_ARRAY3D_DESCRIPTOR structure - * \p pMipmappedArrayDesc and returns a handle to the new CUDA mipmapped array in \p *pHandle. - * \p numMipmapLevels specifies the number of mipmap levels to be allocated. This value is + * Creates a CUDA mipmapped array according to the ::CUDA_ARRAY3D_DESCRIPTOR + structure + * \p pMipmappedArrayDesc and returns a handle to the new CUDA mipmapped array + in \p *pHandle. + * \p numMipmapLevels specifies the number of mipmap levels to be allocated. + This value is * clamped to the range [1, 1 + floor(log2(max(width, height, depth)))]. * * The ::CUDA_ARRAY3D_DESCRIPTOR is defined as: @@ -7995,26 +8404,41 @@ CUresult CUDAAPI cuArray3DGetDescriptor(CUDA_ARRAY3D_DESCRIPTOR *pArrayDescripto * where: * * - \p Width, \p Height, and \p Depth are the width, height, and depth of the - * CUDA array (in elements); the following types of CUDA arrays can be allocated: - * - A 1D mipmapped array is allocated if \p Height and \p Depth extents are both zero. + * CUDA array (in elements); the following types of CUDA arrays can be + allocated: + * - A 1D mipmapped array is allocated if \p Height and \p Depth extents are + both zero. * - A 2D mipmapped array is allocated if only \p Depth extent is zero. * - A 3D mipmapped array is allocated if all three extents are non-zero. - * - A 1D layered CUDA mipmapped array is allocated if only \p Height is zero and the - * ::CUDA_ARRAY3D_LAYERED flag is set. Each layer is a 1D array. The number + * - A 1D layered CUDA mipmapped array is allocated if only \p Height is + zero and the + * ::CUDA_ARRAY3D_LAYERED flag is set. Each layer is a 1D array. The + number * of layers is determined by the depth extent. - * - A 2D layered CUDA mipmapped array is allocated if all three extents are non-zero and - * the ::CUDA_ARRAY3D_LAYERED flag is set. Each layer is a 2D array. The number + * - A 2D layered CUDA mipmapped array is allocated if all three extents are + non-zero and + * the ::CUDA_ARRAY3D_LAYERED flag is set. Each layer is a 2D array. The + number * of layers is determined by the depth extent. - * - A cubemap CUDA mipmapped array is allocated if all three extents are non-zero and the - * ::CUDA_ARRAY3D_CUBEMAP flag is set. \p Width must be equal to \p Height, and - * \p Depth must be six. A cubemap is a special type of 2D layered CUDA array, - * where the six layers represent the six faces of a cube. The order of the six + * - A cubemap CUDA mipmapped array is allocated if all three extents are + non-zero and the + * ::CUDA_ARRAY3D_CUBEMAP flag is set. \p Width must be equal to \p + Height, and + * \p Depth must be six. A cubemap is a special type of 2D layered CUDA + array, + * where the six layers represent the six faces of a cube. The order of + the six * layers in memory is the same as that listed in ::CUarray_cubemap_face. - * - A cubemap layered CUDA mipmapped array is allocated if all three extents are non-zero, - * and both, ::CUDA_ARRAY3D_CUBEMAP and ::CUDA_ARRAY3D_LAYERED flags are set. - * \p Width must be equal to \p Height, and \p Depth must be a multiple of six. - * A cubemap layered CUDA array is a special type of 2D layered CUDA array that - * consists of a collection of cubemaps. The first six layers represent the first + * - A cubemap layered CUDA mipmapped array is allocated if all three + extents are non-zero, + * and both, ::CUDA_ARRAY3D_CUBEMAP and ::CUDA_ARRAY3D_LAYERED flags are + set. + * \p Width must be equal to \p Height, and \p Depth must be a multiple of + six. + * A cubemap layered CUDA array is a special type of 2D layered CUDA array + that + * consists of a collection of cubemaps. The first six layers represent + the first * cubemap, the next six layers form the second cubemap, and so on. * * - ::Format specifies the format of the elements; ::CUarray_format is @@ -8036,25 +8460,35 @@ CUresult CUDAAPI cuArray3DGetDescriptor(CUDA_ARRAY3D_DESCRIPTOR *pArrayDescripto * element; it may be 1, 2, or 4; * * - ::Flags may be set to - * - ::CUDA_ARRAY3D_LAYERED to enable creation of layered CUDA mipmapped arrays. If this flag is set, + * - ::CUDA_ARRAY3D_LAYERED to enable creation of layered CUDA mipmapped + arrays. If this flag is set, * \p Depth specifies the number of layers, not the depth of a 3D array. - * - ::CUDA_ARRAY3D_SURFACE_LDST to enable surface references to be bound to individual mipmap levels of - * the CUDA mipmapped array. If this flag is not set, ::cuSurfRefSetArray will fail when attempting to + * - ::CUDA_ARRAY3D_SURFACE_LDST to enable surface references to be bound to + individual mipmap levels of + * the CUDA mipmapped array. If this flag is not set, ::cuSurfRefSetArray + will fail when attempting to * bind a mipmap level of the CUDA mipmapped array to a surface reference. - * - ::CUDA_ARRAY3D_CUBEMAP to enable creation of mipmapped cubemaps. If this flag is set, \p Width must be - * equal to \p Height, and \p Depth must be six. If the ::CUDA_ARRAY3D_LAYERED flag is also set, + * - ::CUDA_ARRAY3D_CUBEMAP to enable creation of mipmapped cubemaps. If this + flag is set, \p Width must be + * equal to \p Height, and \p Depth must be six. If the + ::CUDA_ARRAY3D_LAYERED flag is also set, * then \p Depth must be a multiple of six. - * - ::CUDA_ARRAY3D_TEXTURE_GATHER to indicate that the CUDA mipmapped array will be used for texture gather. + * - ::CUDA_ARRAY3D_TEXTURE_GATHER to indicate that the CUDA mipmapped array + will be used for texture gather. * Texture gather can only be performed on 2D CUDA mipmapped arrays. * - * \p Width, \p Height and \p Depth must meet certain size requirements as listed in the following table. - * All values are specified in elements. Note that for brevity's sake, the full name of the device attribute - * is not specified. For ex., TEXTURE1D_MIPMAPPED_WIDTH refers to the device attribute + * \p Width, \p Height and \p Depth must meet certain size requirements as + listed in the following table. + * All values are specified in elements. Note that for brevity's sake, the full + name of the device attribute + * is not specified. For ex., TEXTURE1D_MIPMAPPED_WIDTH refers to the device + attribute * ::CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE1D_MIPMAPPED_WIDTH. * * * - * * @@ -8062,7 +8496,8 @@ CUresult CUDAAPI cuArray3DGetDescriptor(CUDA_ARRAY3D_DESCRIPTOR *pArrayDescripto * * * - * + * * * * * - * + * * * - * - * *
CUDA array typeValid extents that must always be met
{(width range in elements), (height range), + * 		Valid extents that must always be met
{(width range in elements), + (height range), * (depth range)}		Valid extents with CUDA_ARRAY3D_SURFACE_LDST set
* {(width range in elements), (height range), (depth range)}
{ (1,TEXTURE1D_MIPMAPPED_WIDTH), 0, 0 }{ (1,SURFACE1D_WIDTH), 0, 0 }
2D{ (1,TEXTURE2D_MIPMAPPED_WIDTH), (1,TEXTURE2D_MIPMAPPED_HEIGHT), 0 }{ (1,TEXTURE2D_MIPMAPPED_WIDTH), (1,TEXTURE2D_MIPMAPPED_HEIGHT), 0 + }{ (1,SURFACE2D_WIDTH), (1,SURFACE2D_HEIGHT), 0 }
3D{ (1,TEXTURE3D_WIDTH), (1,TEXTURE3D_HEIGHT), (1,TEXTURE3D_DEPTH) } @@ -8081,13 +8516,16 @@ CUresult CUDAAPI cuArray3DGetDescriptor(CUDA_ARRAY3D_DESCRIPTOR *pArrayDescripto * { (1,SURFACE2D_LAYERED_WIDTH), (1,SURFACE2D_LAYERED_HEIGHT), * (1,SURFACE2D_LAYERED_LAYERS) }
Cubemap{ (1,TEXTURECUBEMAP_WIDTH), (1,TEXTURECUBEMAP_WIDTH), 6 }{ (1,TEXTURECUBEMAP_WIDTH), (1,TEXTURECUBEMAP_WIDTH), 6 + }{ (1,SURFACECUBEMAP_WIDTH), * (1,SURFACECUBEMAP_WIDTH), 6 }
Cubemap Layered{ (1,TEXTURECUBEMAP_LAYERED_WIDTH), (1,TEXTURECUBEMAP_LAYERED_WIDTH), + * { (1,TEXTURECUBEMAP_LAYERED_WIDTH), + (1,TEXTURECUBEMAP_LAYERED_WIDTH), * (1,TEXTURECUBEMAP_LAYERED_LAYERS) }{ (1,SURFACECUBEMAP_LAYERED_WIDTH), (1,SURFACECUBEMAP_LAYERED_WIDTH), + * { (1,SURFACECUBEMAP_LAYERED_WIDTH), + (1,SURFACECUBEMAP_LAYERED_WIDTH), * (1,SURFACECUBEMAP_LAYERED_LAYERS) }
* @@ -8112,7 +8550,10 @@ CUresult CUDAAPI cuArray3DGetDescriptor(CUDA_ARRAY3D_DESCRIPTOR *pArrayDescripto * ::cuArrayCreate, * ::cudaMallocMipmappedArray */ -CUresult CUDAAPI cuMipmappedArrayCreate(CUmipmappedArray *pHandle, const CUDA_ARRAY3D_DESCRIPTOR *pMipmappedArrayDesc, unsigned int numMipmapLevels); +CUresult CUDAAPI +cuMipmappedArrayCreate(CUmipmappedArray *pHandle, + const CUDA_ARRAY3D_DESCRIPTOR *pMipmappedArrayDesc, + unsigned int numMipmapLevels); /** * \brief Gets a mipmap level of a CUDA mipmapped array @@ -8120,7 +8561,8 @@ CUresult CUDAAPI cuMipmappedArrayCreate(CUmipmappedArray *pHandle, const CUDA_AR * Returns in \p *pLevelArray a CUDA array that represents a single mipmap level * of the CUDA mipmapped array \p hMipmappedArray. * - * If \p level is greater than the maximum number of levels in this mipmapped array, + * If \p level is greater than the maximum number of levels in this mipmapped + * array, * ::CUDA_ERROR_INVALID_VALUE is returned. * * \param pLevelArray - Returned mipmap level CUDA array @@ -8142,7 +8584,9 @@ CUresult CUDAAPI cuMipmappedArrayCreate(CUmipmappedArray *pHandle, const CUDA_AR * ::cuArrayCreate, * ::cudaGetMipmappedArrayLevel */ -CUresult CUDAAPI cuMipmappedArrayGetLevel(CUarray *pLevelArray, CUmipmappedArray hMipmappedArray, unsigned int level); +CUresult CUDAAPI cuMipmappedArrayGetLevel(CUarray *pLevelArray, + CUmipmappedArray hMipmappedArray, + unsigned int level); /** * \brief Destroys a CUDA mipmapped array @@ -8219,7 +8663,8 @@ CUresult CUDAAPI cuMipmappedArrayDestroy(CUmipmappedArray hMipmappedArray); * used to specify that the CUDA driver should infer the location of the * pointer from its value. * - * \section CUDA_UNIFIED_automaphost Automatic Mapping of Host Allocated Host Memory + * \section CUDA_UNIFIED_automaphost Automatic Mapping of Host Allocated Host + * Memory * * All host memory allocated in all contexts using ::cuMemAllocHost() and * ::cuMemHostAlloc() is always directly accessible from all contexts on @@ -8230,8 +8675,8 @@ CUresult CUDAAPI cuMipmappedArrayDestroy(CUmipmappedArray hMipmappedArray); * The pointer value through which allocated host memory may be accessed * in kernels on all devices that support unified addressing is the same * as the pointer value through which that memory is accessed on the host, - * so it is not necessary to call ::cuMemHostGetDevicePointer() to get the device - * pointer for these allocations. + * so it is not necessary to call ::cuMemHostGetDevicePointer() to get the + * device pointer for these allocations. * * Note that this is not the case for memory allocated using the flag * ::CU_MEMHOSTALLOC_WRITECOMBINED, as discussed below. @@ -8350,32 +8795,33 @@ CUresult CUDAAPI cuMipmappedArrayDestroy(CUmipmappedArray hMipmappedArray); * * - ::CU_POINTER_ATTRIBUTE_SYNC_MEMOPS: * - * A boolean attribute which when set, ensures that synchronous memory operations - * initiated on the region of memory that \p ptr points to will always synchronize. - * See further documentation in the section titled "API synchronization behavior" - * to learn more about cases when synchronous memory operations can - * exhibit asynchronous behavior. + * A boolean attribute which when set, ensures that synchronous memory + * operations initiated on the region of memory that \p ptr points to will + * always synchronize. See further documentation in the section titled "API + * synchronization behavior" to learn more about cases when synchronous memory + * operations can exhibit asynchronous behavior. * * - ::CU_POINTER_ATTRIBUTE_BUFFER_ID: * - * Returns in \p *data a buffer ID which is guaranteed to be unique within the process. - * \p data must point to an unsigned long long. + * Returns in \p *data a buffer ID which is guaranteed to be unique within + * the process. \p data must point to an unsigned long long. * - * \p ptr must be a pointer to memory obtained from a CUDA memory allocation API. - * Every memory allocation from any of the CUDA memory allocation APIs will - * have a unique ID over a process lifetime. Subsequent allocations do not reuse IDs - * from previous freed allocations. IDs are only unique within a single process. + * \p ptr must be a pointer to memory obtained from a CUDA memory + * allocation API. Every memory allocation from any of the CUDA memory + * allocation APIs will have a unique ID over a process lifetime. Subsequent + * allocations do not reuse IDs from previous freed allocations. IDs are only + * unique within a single process. * * * - ::CU_POINTER_ATTRIBUTE_IS_MANAGED: * - * Returns in \p *data a boolean that indicates whether the pointer points to - * managed memory or not. + * Returns in \p *data a boolean that indicates whether the pointer points + * to managed memory or not. * * - ::CU_POINTER_ATTRIBUTE_DEVICE_ORDINAL: * - * Returns in \p *data an integer representing a device ordinal of a device against - * which the memory was allocated or registered. + * Returns in \p *data an integer representing a device ordinal of a device + * against which the memory was allocated or registered. * * \par * @@ -8419,7 +8865,9 @@ CUresult CUDAAPI cuMipmappedArrayDestroy(CUmipmappedArray hMipmappedArray); * ::cuMemHostUnregister, * ::cudaPointerGetAttributes */ -CUresult CUDAAPI cuPointerGetAttribute(void *data, CUpointer_attribute attribute, CUdeviceptr ptr); +CUresult CUDAAPI cuPointerGetAttribute(void *data, + CUpointer_attribute attribute, + CUdeviceptr ptr); #endif /* __CUDA_API_VERSION >= 4000 */ #if __CUDA_API_VERSION >= 8000 @@ -8428,46 +8876,51 @@ CUresult CUDAAPI cuPointerGetAttribute(void *data, CUpointer_attribute attribute * * Prefetches memory to the specified destination device. \p devPtr is the * base device pointer of the memory to be prefetched and \p dstDevice is the - * destination device. \p count specifies the number of bytes to copy. \p hStream - * is the stream in which the operation is enqueued. The memory range must refer - * to managed memory allocated via ::cuMemAllocManaged or declared via __managed__ variables. - * - * Passing in CU_DEVICE_CPU for \p dstDevice will prefetch the data to host memory. If - * \p dstDevice is a GPU, then the device attribute ::CU_DEVICE_ATTRIBUTE_CONCURRENT_MANAGED_ACCESS - * must be non-zero. Additionally, \p hStream must be associated with a device that has a - * non-zero value for the device attribute ::CU_DEVICE_ATTRIBUTE_CONCURRENT_MANAGED_ACCESS. - * - * The start address and end address of the memory range will be rounded down and rounded up - * respectively to be aligned to CPU page size before the prefetch operation is enqueued - * in the stream. - * - * If no physical memory has been allocated for this region, then this memory region - * will be populated and mapped on the destination device. If there's insufficient - * memory to prefetch the desired region, the Unified Memory driver may evict pages from other - * ::cuMemAllocManaged allocations to host memory in order to make room. Device memory - * allocated using ::cuMemAlloc or ::cuArrayCreate will not be evicted. - * - * By default, any mappings to the previous location of the migrated pages are removed and - * mappings for the new location are only setup on \p dstDevice. The exact behavior however - * also depends on the settings applied to this memory range via ::cuMemAdvise as described - * below: - * - * If ::CU_MEM_ADVISE_SET_READ_MOSTLY was set on any subset of this memory range, - * then that subset will create a read-only copy of the pages on \p dstDevice. - * - * If ::CU_MEM_ADVISE_SET_PREFERRED_LOCATION was called on any subset of this memory - * range, then the pages will be migrated to \p dstDevice even if \p dstDevice is not the - * preferred location of any pages in the memory range. - * - * If ::CU_MEM_ADVISE_SET_ACCESSED_BY was called on any subset of this memory range, - * then mappings to those pages from all the appropriate processors are updated to - * refer to the new location if establishing such a mapping is possible. Otherwise, - * those mappings are cleared. - * - * Note that this API is not required for functionality and only serves to improve performance - * by allowing the application to migrate data to a suitable location before it is accessed. - * Memory accesses to this range are always coherent and are allowed even when the data is - * actively being migrated. + * destination device. \p count specifies the number of bytes to copy. \p + * hStream is the stream in which the operation is enqueued. The memory range + * must refer to managed memory allocated via ::cuMemAllocManaged or declared + * via __managed__ variables. + * + * Passing in CU_DEVICE_CPU for \p dstDevice will prefetch the data to host + * memory. If \p dstDevice is a GPU, then the device attribute + * ::CU_DEVICE_ATTRIBUTE_CONCURRENT_MANAGED_ACCESS must be non-zero. + * Additionally, \p hStream must be associated with a device that has a non-zero + * value for the device attribute + * ::CU_DEVICE_ATTRIBUTE_CONCURRENT_MANAGED_ACCESS. + * + * The start address and end address of the memory range will be rounded down + * and rounded up respectively to be aligned to CPU page size before the + * prefetch operation is enqueued in the stream. + * + * If no physical memory has been allocated for this region, then this memory + * region will be populated and mapped on the destination device. If there's + * insufficient memory to prefetch the desired region, the Unified Memory driver + * may evict pages from other + * ::cuMemAllocManaged allocations to host memory in order to make room. Device + * memory allocated using ::cuMemAlloc or ::cuArrayCreate will not be evicted. + * + * By default, any mappings to the previous location of the migrated pages are + * removed and mappings for the new location are only setup on \p dstDevice. The + * exact behavior however also depends on the settings applied to this memory + * range via ::cuMemAdvise as described below: + * + * If ::CU_MEM_ADVISE_SET_READ_MOSTLY was set on any subset of this memory + * range, then that subset will create a read-only copy of the pages on \p + * dstDevice. + * + * If ::CU_MEM_ADVISE_SET_PREFERRED_LOCATION was called on any subset of this + * memory range, then the pages will be migrated to \p dstDevice even if \p + * dstDevice is not the preferred location of any pages in the memory range. + * + * If ::CU_MEM_ADVISE_SET_ACCESSED_BY was called on any subset of this memory + * range, then mappings to those pages from all the appropriate processors are + * updated to refer to the new location if establishing such a mapping is + * possible. Otherwise, those mappings are cleared. + * + * Note that this API is not required for functionality and only serves to + * improve performance by allowing the application to migrate data to a suitable + * location before it is accessed. Memory accesses to this range are always + * coherent and are allowed even when the data is actively being migrated. * * Note that this function is asynchronous with respect to the host and all work * on other devices. @@ -8489,102 +8942,131 @@ CUresult CUDAAPI cuPointerGetAttribute(void *data, CUpointer_attribute attribute * ::cuMemcpy3DPeerAsync, ::cuMemAdvise, * ::cudaMemPrefetchAsync */ -CUresult CUDAAPI cuMemPrefetchAsync(CUdeviceptr devPtr, size_t count, CUdevice dstDevice, CUstream hStream); +CUresult CUDAAPI cuMemPrefetchAsync(CUdeviceptr devPtr, size_t count, + CUdevice dstDevice, CUstream hStream); /** * \brief Advise about the usage of a given memory range * - * Advise the Unified Memory subsystem about the usage pattern for the memory range - * starting at \p devPtr with a size of \p count bytes. The start address and end address of the memory - * range will be rounded down and rounded up respectively to be aligned to CPU page size before the - * advice is applied. The memory range must refer to managed memory allocated via ::cuMemAllocManaged - * or declared via __managed__ variables. The memory range could also refer to system-allocated pageable - * memory provided it represents a valid, host-accessible region of memory and all additional constraints - * imposed by \p advice as outlined below are also satisfied. Specifying an invalid system-allocated pageable - * memory range results in an error being returned. + * Advise the Unified Memory subsystem about the usage pattern for the memory + * range starting at \p devPtr with a size of \p count bytes. The start address + * and end address of the memory range will be rounded down and rounded up + * respectively to be aligned to CPU page size before the advice is applied. The + * memory range must refer to managed memory allocated via ::cuMemAllocManaged + * or declared via __managed__ variables. The memory range could also refer to + * system-allocated pageable memory provided it represents a valid, + * host-accessible region of memory and all additional constraints imposed by \p + * advice as outlined below are also satisfied. Specifying an invalid + * system-allocated pageable memory range results in an error being returned. * * The \p advice parameter can take the following values: - * - ::CU_MEM_ADVISE_SET_READ_MOSTLY: This implies that the data is mostly going to be read - * from and only occasionally written to. Any read accesses from any processor to this region will create a - * read-only copy of at least the accessed pages in that processor's memory. Additionally, if ::cuMemPrefetchAsync - * is called on this region, it will create a read-only copy of the data on the destination processor. - * If any processor writes to this region, all copies of the corresponding page will be invalidated - * except for the one where the write occurred. The \p device argument is ignored for this advice. - * Note that for a page to be read-duplicated, the accessing processor must either be the CPU or a GPU - * that has a non-zero value for the device attribute ::CU_DEVICE_ATTRIBUTE_CONCURRENT_MANAGED_ACCESS. - * Also, if a context is created on a device that does not have the device attribute - * ::CU_DEVICE_ATTRIBUTE_CONCURRENT_MANAGED_ACCESS set, then read-duplication will not occur until - * all such contexts are destroyed. - * If the memory region refers to valid system-allocated pageable memory, then the accessing device must - * have a non-zero value for the device attribute ::CU_DEVICE_ATTRIBUTE_PAGEABLE_MEMORY_ACCESS for a read-only - * copy to be created on that device. Note however that if the accessing device also has a non-zero value for the - * device attribute ::CU_DEVICE_ATTRIBUTE_PAGEABLE_MEMORY_ACCESS_USES_HOST_PAGE_TABLES, then setting this advice - * will not create a read-only copy when that device accesses this memory region. - * - * - ::CU_MEM_ADVISE_UNSET_READ_MOSTLY: Undoes the effect of ::CU_MEM_ADVISE_SET_READ_MOSTLY and also prevents the - * Unified Memory driver from attempting heuristic read-duplication on the memory range. Any read-duplicated - * copies of the data will be collapsed into a single copy. The location for the collapsed - * copy will be the preferred location if the page has a preferred location and one of the read-duplicated - * copies was resident at that location. Otherwise, the location chosen is arbitrary. - * - * - ::CU_MEM_ADVISE_SET_PREFERRED_LOCATION: This advice sets the preferred location for the - * data to be the memory belonging to \p device. Passing in CU_DEVICE_CPU for \p device sets the - * preferred location as host memory. If \p device is a GPU, then it must have a non-zero value for the - * device attribute ::CU_DEVICE_ATTRIBUTE_CONCURRENT_MANAGED_ACCESS. Setting the preferred location - * does not cause data to migrate to that location immediately. Instead, it guides the migration policy - * when a fault occurs on that memory region. If the data is already in its preferred location and the - * faulting processor can establish a mapping without requiring the data to be migrated, then - * data migration will be avoided. On the other hand, if the data is not in its preferred location - * or if a direct mapping cannot be established, then it will be migrated to the processor accessing - * it. It is important to note that setting the preferred location does not prevent data prefetching - * done using ::cuMemPrefetchAsync. - * Having a preferred location can override the page thrash detection and resolution logic in the Unified - * Memory driver. Normally, if a page is detected to be constantly thrashing between for example host and device - * memory, the page may eventually be pinned to host memory by the Unified Memory driver. But - * if the preferred location is set as device memory, then the page will continue to thrash indefinitely. - * If ::CU_MEM_ADVISE_SET_READ_MOSTLY is also set on this memory region or any subset of it, then the - * policies associated with that advice will override the policies of this advice, unless read accesses from - * \p device will not result in a read-only copy being created on that device as outlined in description for - * the advice ::CU_MEM_ADVISE_SET_READ_MOSTLY. - * If the memory region refers to valid system-allocated pageable memory, then \p device must have a non-zero - * value for the device attribute ::CU_DEVICE_ATTRIBUTE_PAGEABLE_MEMORY_ACCESS. Additionally, if \p device has - * a non-zero value for the device attribute ::CU_DEVICE_ATTRIBUTE_PAGEABLE_MEMORY_ACCESS_USES_HOST_PAGE_TABLES, - * then this call has no effect. Note however that this behavior may change in the future. - * - * - ::CU_MEM_ADVISE_UNSET_PREFERRED_LOCATION: Undoes the effect of ::CU_MEM_ADVISE_SET_PREFERRED_LOCATION - * and changes the preferred location to none. - * - * - ::CU_MEM_ADVISE_SET_ACCESSED_BY: This advice implies that the data will be accessed by \p device. - * Passing in ::CU_DEVICE_CPU for \p device will set the advice for the CPU. If \p device is a GPU, then - * the device attribute ::CU_DEVICE_ATTRIBUTE_CONCURRENT_MANAGED_ACCESS must be non-zero. - * This advice does not cause data migration and has no impact on the location of the data per se. Instead, - * it causes the data to always be mapped in the specified processor's page tables, as long as the - * location of the data permits a mapping to be established. If the data gets migrated for any reason, - * the mappings are updated accordingly. - * This advice is recommended in scenarios where data locality is not important, but avoiding faults is. - * Consider for example a system containing multiple GPUs with peer-to-peer access enabled, where the - * data located on one GPU is occasionally accessed by peer GPUs. In such scenarios, migrating data - * over to the other GPUs is not as important because the accesses are infrequent and the overhead of - * migration may be too high. But preventing faults can still help improve performance, and so having - * a mapping set up in advance is useful. Note that on CPU access of this data, the data may be migrated - * to host memory because the CPU typically cannot access device memory directly. Any GPU that had the - * ::CU_MEM_ADVISE_SET_ACCESSED_BY flag set for this data will now have its mapping updated to point to the - * page in host memory. - * If ::CU_MEM_ADVISE_SET_READ_MOSTLY is also set on this memory region or any subset of it, then the - * policies associated with that advice will override the policies of this advice. Additionally, if the - * preferred location of this memory region or any subset of it is also \p device, then the policies - * associated with ::CU_MEM_ADVISE_SET_PREFERRED_LOCATION will override the policies of this advice. - * If the memory region refers to valid system-allocated pageable memory, then \p device must have a non-zero - * value for the device attribute ::CU_DEVICE_ATTRIBUTE_PAGEABLE_MEMORY_ACCESS. Additionally, if \p device has - * a non-zero value for the device attribute ::CU_DEVICE_ATTRIBUTE_PAGEABLE_MEMORY_ACCESS_USES_HOST_PAGE_TABLES, - * then this call has no effect. - * - * - ::CU_MEM_ADVISE_UNSET_ACCESSED_BY: Undoes the effect of ::CU_MEM_ADVISE_SET_ACCESSED_BY. Any mappings to - * the data from \p device may be removed at any time causing accesses to result in non-fatal page faults. - * If the memory region refers to valid system-allocated pageable memory, then \p device must have a non-zero - * value for the device attribute ::CU_DEVICE_ATTRIBUTE_PAGEABLE_MEMORY_ACCESS. Additionally, if \p device has - * a non-zero value for the device attribute ::CU_DEVICE_ATTRIBUTE_PAGEABLE_MEMORY_ACCESS_USES_HOST_PAGE_TABLES, - * then this call has no effect. + * - ::CU_MEM_ADVISE_SET_READ_MOSTLY: This implies that the data is mostly going + * to be read from and only occasionally written to. Any read accesses from any + * processor to this region will create a read-only copy of at least the + * accessed pages in that processor's memory. Additionally, if + * ::cuMemPrefetchAsync is called on this region, it will create a read-only + * copy of the data on the destination processor. If any processor writes to + * this region, all copies of the corresponding page will be invalidated except + * for the one where the write occurred. The \p device argument is ignored for + * this advice. Note that for a page to be read-duplicated, the accessing + * processor must either be the CPU or a GPU that has a non-zero value for the + * device attribute ::CU_DEVICE_ATTRIBUTE_CONCURRENT_MANAGED_ACCESS. Also, if a + * context is created on a device that does not have the device attribute + * ::CU_DEVICE_ATTRIBUTE_CONCURRENT_MANAGED_ACCESS set, then read-duplication + * will not occur until all such contexts are destroyed. If the memory region + * refers to valid system-allocated pageable memory, then the accessing device + * must have a non-zero value for the device attribute + * ::CU_DEVICE_ATTRIBUTE_PAGEABLE_MEMORY_ACCESS for a read-only copy to be + * created on that device. Note however that if the accessing device also has a + * non-zero value for the device attribute + * ::CU_DEVICE_ATTRIBUTE_PAGEABLE_MEMORY_ACCESS_USES_HOST_PAGE_TABLES, then + * setting this advice will not create a read-only copy when that device + * accesses this memory region. + * + * - ::CU_MEM_ADVISE_UNSET_READ_MOSTLY: Undoes the effect of + * ::CU_MEM_ADVISE_SET_READ_MOSTLY and also prevents the Unified Memory driver + * from attempting heuristic read-duplication on the memory range. Any + * read-duplicated copies of the data will be collapsed into a single copy. The + * location for the collapsed copy will be the preferred location if the page + * has a preferred location and one of the read-duplicated copies was resident + * at that location. Otherwise, the location chosen is arbitrary. + * + * - ::CU_MEM_ADVISE_SET_PREFERRED_LOCATION: This advice sets the preferred + * location for the data to be the memory belonging to \p device. Passing in + * CU_DEVICE_CPU for \p device sets the preferred location as host memory. If \p + * device is a GPU, then it must have a non-zero value for the device attribute + * ::CU_DEVICE_ATTRIBUTE_CONCURRENT_MANAGED_ACCESS. Setting the preferred + * location does not cause data to migrate to that location immediately. + * Instead, it guides the migration policy when a fault occurs on that memory + * region. If the data is already in its preferred location and the faulting + * processor can establish a mapping without requiring the data to be migrated, + * then data migration will be avoided. On the other hand, if the data is not in + * its preferred location or if a direct mapping cannot be established, then it + * will be migrated to the processor accessing it. It is important to note that + * setting the preferred location does not prevent data prefetching done using + * ::cuMemPrefetchAsync. Having a preferred location can override the page + * thrash detection and resolution logic in the Unified Memory driver. Normally, + * if a page is detected to be constantly thrashing between for example host and + * device memory, the page may eventually be pinned to host memory by the + * Unified Memory driver. But if the preferred location is set as device memory, + * then the page will continue to thrash indefinitely. If + * ::CU_MEM_ADVISE_SET_READ_MOSTLY is also set on this memory region or any + * subset of it, then the policies associated with that advice will override the + * policies of this advice, unless read accesses from \p device will not result + * in a read-only copy being created on that device as outlined in description + * for the advice ::CU_MEM_ADVISE_SET_READ_MOSTLY. If the memory region refers + * to valid system-allocated pageable memory, then \p device must have a + * non-zero value for the device attribute + * ::CU_DEVICE_ATTRIBUTE_PAGEABLE_MEMORY_ACCESS. Additionally, if \p device has + * a non-zero value for the device attribute + * ::CU_DEVICE_ATTRIBUTE_PAGEABLE_MEMORY_ACCESS_USES_HOST_PAGE_TABLES, then this + * call has no effect. Note however that this behavior may change in the future. + * + * - ::CU_MEM_ADVISE_UNSET_PREFERRED_LOCATION: Undoes the effect of + * ::CU_MEM_ADVISE_SET_PREFERRED_LOCATION and changes the preferred location to + * none. + * + * - ::CU_MEM_ADVISE_SET_ACCESSED_BY: This advice implies that the data will be + * accessed by \p device. Passing in ::CU_DEVICE_CPU for \p device will set the + * advice for the CPU. If \p device is a GPU, then the device attribute + * ::CU_DEVICE_ATTRIBUTE_CONCURRENT_MANAGED_ACCESS must be non-zero. This advice + * does not cause data migration and has no impact on the location of the data + * per se. Instead, it causes the data to always be mapped in the specified + * processor's page tables, as long as the location of the data permits a + * mapping to be established. If the data gets migrated for any reason, the + * mappings are updated accordingly. This advice is recommended in scenarios + * where data locality is not important, but avoiding faults is. Consider for + * example a system containing multiple GPUs with peer-to-peer access enabled, + * where the data located on one GPU is occasionally accessed by peer GPUs. In + * such scenarios, migrating data over to the other GPUs is not as important + * because the accesses are infrequent and the overhead of migration may be too + * high. But preventing faults can still help improve performance, and so having + * a mapping set up in advance is useful. Note that on CPU access of this data, + * the data may be migrated to host memory because the CPU typically cannot + * access device memory directly. Any GPU that had the + * ::CU_MEM_ADVISE_SET_ACCESSED_BY flag set for this data will now have its + * mapping updated to point to the page in host memory. If + * ::CU_MEM_ADVISE_SET_READ_MOSTLY is also set on this memory region or any + * subset of it, then the policies associated with that advice will override the + * policies of this advice. Additionally, if the preferred location of this + * memory region or any subset of it is also \p device, then the policies + * associated with ::CU_MEM_ADVISE_SET_PREFERRED_LOCATION will override the + * policies of this advice. If the memory region refers to valid + * system-allocated pageable memory, then \p device must have a non-zero value + * for the device attribute ::CU_DEVICE_ATTRIBUTE_PAGEABLE_MEMORY_ACCESS. + * Additionally, if \p device has a non-zero value for the device attribute + * ::CU_DEVICE_ATTRIBUTE_PAGEABLE_MEMORY_ACCESS_USES_HOST_PAGE_TABLES, then this + * call has no effect. + * + * - ::CU_MEM_ADVISE_UNSET_ACCESSED_BY: Undoes the effect of + * ::CU_MEM_ADVISE_SET_ACCESSED_BY. Any mappings to the data from \p device may + * be removed at any time causing accesses to result in non-fatal page faults. + * If the memory region refers to valid system-allocated pageable memory, then + * \p device must have a non-zero value for the device attribute + * ::CU_DEVICE_ATTRIBUTE_PAGEABLE_MEMORY_ACCESS. Additionally, if \p device has + * a non-zero value for the device attribute + * ::CU_DEVICE_ATTRIBUTE_PAGEABLE_MEMORY_ACCESS_USES_HOST_PAGE_TABLES, then this + * call has no effect. * * \param devPtr - Pointer to memory to set the advice for * \param count - Size in bytes of the memory range @@ -8603,44 +9085,56 @@ CUresult CUDAAPI cuMemPrefetchAsync(CUdeviceptr devPtr, size_t count, CUdevice d * ::cuMemcpy3DPeerAsync, ::cuMemPrefetchAsync, * ::cudaMemAdvise */ -CUresult CUDAAPI cuMemAdvise(CUdeviceptr devPtr, size_t count, CUmem_advise advice, CUdevice device); +CUresult CUDAAPI cuMemAdvise(CUdeviceptr devPtr, size_t count, + CUmem_advise advice, CUdevice device); /** * \brief Query an attribute of a given memory range * - * Query an attribute about the memory range starting at \p devPtr with a size of \p count bytes. The - * memory range must refer to managed memory allocated via ::cuMemAllocManaged or declared via + * Query an attribute about the memory range starting at \p devPtr with a size + * of \p count bytes. The memory range must refer to managed memory allocated + * via ::cuMemAllocManaged or declared via * __managed__ variables. * * The \p attribute parameter can take the following values: - * - ::CU_MEM_RANGE_ATTRIBUTE_READ_MOSTLY: If this attribute is specified, \p data will be interpreted - * as a 32-bit integer, and \p dataSize must be 4. The result returned will be 1 if all pages in the given - * memory range have read-duplication enabled, or 0 otherwise. - * - ::CU_MEM_RANGE_ATTRIBUTE_PREFERRED_LOCATION: If this attribute is specified, \p data will be - * interpreted as a 32-bit integer, and \p dataSize must be 4. The result returned will be a GPU device - * id if all pages in the memory range have that GPU as their preferred location, or it will be CU_DEVICE_CPU - * if all pages in the memory range have the CPU as their preferred location, or it will be CU_DEVICE_INVALID - * if either all the pages don't have the same preferred location or some of the pages don't have a - * preferred location at all. Note that the actual location of the pages in the memory range at the time of - * the query may be different from the preferred location. - * - ::CU_MEM_RANGE_ATTRIBUTE_ACCESSED_BY: If this attribute is specified, \p data will be interpreted - * as an array of 32-bit integers, and \p dataSize must be a non-zero multiple of 4. The result returned - * will be a list of device ids that had ::CU_MEM_ADVISE_SET_ACCESSED_BY set for that entire memory range. - * If any device does not have that advice set for the entire memory range, that device will not be included. - * If \p data is larger than the number of devices that have that advice set for that memory range, - * CU_DEVICE_INVALID will be returned in all the extra space provided. For ex., if \p dataSize is 12 - * (i.e. \p data has 3 elements) and only device 0 has the advice set, then the result returned will be - * { 0, CU_DEVICE_INVALID, CU_DEVICE_INVALID }. If \p data is smaller than the number of devices that have - * that advice set, then only as many devices will be returned as can fit in the array. There is no - * guarantee on which specific devices will be returned, however. - * - ::CU_MEM_RANGE_ATTRIBUTE_LAST_PREFETCH_LOCATION: If this attribute is specified, \p data will be - * interpreted as a 32-bit integer, and \p dataSize must be 4. The result returned will be the last location - * to which all pages in the memory range were prefetched explicitly via ::cuMemPrefetchAsync. This will either be - * a GPU id or CU_DEVICE_CPU depending on whether the last location for prefetch was a GPU or the CPU - * respectively. If any page in the memory range was never explicitly prefetched or if all pages were not - * prefetched to the same location, CU_DEVICE_INVALID will be returned. Note that this simply returns the - * last location that the applicaton requested to prefetch the memory range to. It gives no indication as to - * whether the prefetch operation to that location has completed or even begun. + * - ::CU_MEM_RANGE_ATTRIBUTE_READ_MOSTLY: If this attribute is specified, \p + * data will be interpreted as a 32-bit integer, and \p dataSize must be 4. The + * result returned will be 1 if all pages in the given memory range have + * read-duplication enabled, or 0 otherwise. + * - ::CU_MEM_RANGE_ATTRIBUTE_PREFERRED_LOCATION: If this attribute is + * specified, \p data will be interpreted as a 32-bit integer, and \p dataSize + * must be 4. The result returned will be a GPU device id if all pages in the + * memory range have that GPU as their preferred location, or it will be + * CU_DEVICE_CPU if all pages in the memory range have the CPU as their + * preferred location, or it will be CU_DEVICE_INVALID if either all the pages + * don't have the same preferred location or some of the pages don't have a + * preferred location at all. Note that the actual location of the pages in the + * memory range at the time of the query may be different from the preferred + * location. + * - ::CU_MEM_RANGE_ATTRIBUTE_ACCESSED_BY: If this attribute is specified, \p + * data will be interpreted as an array of 32-bit integers, and \p dataSize must + * be a non-zero multiple of 4. The result returned will be a list of device ids + * that had ::CU_MEM_ADVISE_SET_ACCESSED_BY set for that entire memory range. If + * any device does not have that advice set for the entire memory range, that + * device will not be included. If \p data is larger than the number of devices + * that have that advice set for that memory range, CU_DEVICE_INVALID will be + * returned in all the extra space provided. For ex., if \p dataSize is 12 (i.e. + * \p data has 3 elements) and only device 0 has the advice set, then the result + * returned will be { 0, CU_DEVICE_INVALID, CU_DEVICE_INVALID }. If \p data is + * smaller than the number of devices that have that advice set, then only as + * many devices will be returned as can fit in the array. There is no guarantee + * on which specific devices will be returned, however. + * - ::CU_MEM_RANGE_ATTRIBUTE_LAST_PREFETCH_LOCATION: If this attribute is + * specified, \p data will be interpreted as a 32-bit integer, and \p dataSize + * must be 4. The result returned will be the last location to which all pages + * in the memory range were prefetched explicitly via ::cuMemPrefetchAsync. This + * will either be a GPU id or CU_DEVICE_CPU depending on whether the last + * location for prefetch was a GPU or the CPU respectively. If any page in the + * memory range was never explicitly prefetched or if all pages were not + * prefetched to the same location, CU_DEVICE_INVALID will be returned. Note + * that this simply returns the last location that the applicaton requested to + * prefetch the memory range to. It gives no indication as to whether the + * prefetch operation to that location has completed or even begun. * * \param data - A pointers to a memory location where the result * of each attribute query will be written to. @@ -8661,19 +9155,23 @@ CUresult CUDAAPI cuMemAdvise(CUdeviceptr devPtr, size_t count, CUmem_advise advi * ::cuMemAdvise, * ::cudaMemRangeGetAttribute */ -CUresult CUDAAPI cuMemRangeGetAttribute(void *data, size_t dataSize, CUmem_range_attribute attribute, CUdeviceptr devPtr, size_t count); +CUresult CUDAAPI cuMemRangeGetAttribute(void *data, size_t dataSize, + CUmem_range_attribute attribute, + CUdeviceptr devPtr, size_t count); /** * \brief Query attributes of a given memory range. * - * Query attributes of the memory range starting at \p devPtr with a size of \p count bytes. The - * memory range must refer to managed memory allocated via ::cuMemAllocManaged or declared via - * __managed__ variables. The \p attributes array will be interpreted to have \p numAttributes - * entries. The \p dataSizes array will also be interpreted to have \p numAttributes entries. - * The results of the query will be stored in \p data. + * Query attributes of the memory range starting at \p devPtr with a size of \p + * count bytes. The memory range must refer to managed memory allocated via + * ::cuMemAllocManaged or declared via + * __managed__ variables. The \p attributes array will be interpreted to have \p + * numAttributes entries. The \p dataSizes array will also be interpreted to + * have \p numAttributes entries. The results of the query will be stored in \p + * data. * - * The list of supported attributes are given below. Please refer to ::cuMemRangeGetAttribute for - * attribute descriptions and restrictions. + * The list of supported attributes are given below. Please refer to + * ::cuMemRangeGetAttribute for attribute descriptions and restrictions. * * - ::CU_MEM_RANGE_ATTRIBUTE_READ_MOSTLY * - ::CU_MEM_RANGE_ATTRIBUTE_PREFERRED_LOCATION @@ -8681,13 +9179,12 @@ CUresult CUDAAPI cuMemRangeGetAttribute(void *data, size_t dataSize, CUmem_range * - ::CU_MEM_RANGE_ATTRIBUTE_LAST_PREFETCH_LOCATION * * \param data - A two-dimensional array containing pointers to memory - * locations where the result of each attribute query will be written to. - * \param dataSizes - Array containing the sizes of each result - * \param attributes - An array of attributes to query - * (numAttributes and the number of attributes in this array should match) - * \param numAttributes - Number of attributes to query - * \param devPtr - Start of the range to query - * \param count - Size of the range to query + * locations where the result of each attribute query + * will be written to. \param dataSizes - Array containing the sizes of each + * result \param attributes - An array of attributes to query (numAttributes + * and the number of attributes in this array should match) \param numAttributes + * - Number of attributes to query \param devPtr - Start of the range to + * query \param count - Size of the range to query * * \return * ::CUDA_SUCCESS, @@ -8701,7 +9198,10 @@ CUresult CUDAAPI cuMemRangeGetAttribute(void *data, size_t dataSize, CUmem_range * ::cuMemPrefetchAsync, * ::cudaMemRangeGetAttributes */ -CUresult CUDAAPI cuMemRangeGetAttributes(void **data, size_t *dataSizes, CUmem_range_attribute *attributes, size_t numAttributes, CUdeviceptr devPtr, size_t count); +CUresult CUDAAPI cuMemRangeGetAttributes(void **data, size_t *dataSizes, + CUmem_range_attribute *attributes, + size_t numAttributes, + CUdeviceptr devPtr, size_t count); #endif /* __CUDA_API_VERSION >= 8000 */ #if __CUDA_API_VERSION >= 6000 @@ -8713,18 +9213,19 @@ CUresult CUDAAPI cuMemRangeGetAttributes(void **data, size_t *dataSizes, CUmem_r * - ::CU_POINTER_ATTRIBUTE_SYNC_MEMOPS: * * A boolean attribute that can either be set (1) or unset (0). When set, - * the region of memory that \p ptr points to is guaranteed to always synchronize - * memory operations that are synchronous. If there are some previously initiated - * synchronous memory operations that are pending when this attribute is set, the - * function does not return until those memory operations are complete. - * See further documentation in the section titled "API synchronization behavior" - * to learn more about cases when synchronous memory operations can - * exhibit asynchronous behavior. - * \p value will be considered as a pointer to an unsigned integer to which this attribute is to be set. + * the region of memory that \p ptr points to is guaranteed to always + * synchronize memory operations that are synchronous. If there are some + * previously initiated synchronous memory operations that are pending when this + * attribute is set, the function does not return until those memory operations + * are complete. See further documentation in the section titled "API + * synchronization behavior" to learn more about cases when synchronous memory + * operations can exhibit asynchronous behavior. \p value will be considered as + * a pointer to an unsigned integer to which this attribute is to be set. * * \param value - Pointer to memory containing the value to be set * \param attribute - Pointer attribute to set - * \param ptr - Pointer to a memory region allocated using CUDA memory allocation APIs + * \param ptr - Pointer to a memory region allocated using CUDA memory + * allocation APIs * * \return * ::CUDA_SUCCESS, @@ -8745,14 +9246,17 @@ CUresult CUDAAPI cuMemRangeGetAttributes(void **data, size_t *dataSizes, CUmem_r * ::cuMemHostRegister, * ::cuMemHostUnregister */ -CUresult CUDAAPI cuPointerSetAttribute(const void *value, CUpointer_attribute attribute, CUdeviceptr ptr); +CUresult CUDAAPI cuPointerSetAttribute(const void *value, + CUpointer_attribute attribute, + CUdeviceptr ptr); #endif /* __CUDA_API_VERSION >= 6000 */ #if __CUDA_API_VERSION >= 7000 /** * \brief Returns information about a pointer. * - * The supported attributes are (refer to ::cuPointerGetAttribute for attribute descriptions and restrictions): + * The supported attributes are (refer to ::cuPointerGetAttribute for attribute + * descriptions and restrictions): * * - ::CU_POINTER_ATTRIBUTE_CONTEXT * - ::CU_POINTER_ATTRIBUTE_MEMORY_TYPE @@ -8765,17 +9269,18 @@ CUresult CUDAAPI cuPointerSetAttribute(const void *value, CUpointer_attribute at * * \param numAttributes - Number of attributes to query * \param attributes - An array of attributes to query - * (numAttributes and the number of attributes in this array should match) - * \param data - A two-dimensional array containing pointers to memory - * locations where the result of each attribute query will be written to. - * \param ptr - Pointer to query + * (numAttributes and the number of attributes in this + * array should match) \param data - A two-dimensional array containing + * pointers to memory locations where the result of each attribute query will be + * written to. \param ptr - Pointer to query * - * Unlike ::cuPointerGetAttribute, this function will not return an error when the \p ptr - * encountered is not a valid CUDA pointer. Instead, the attributes are assigned default NULL values - * and CUDA_SUCCESS is returned. + * Unlike ::cuPointerGetAttribute, this function will not return an error when + * the \p ptr encountered is not a valid CUDA pointer. Instead, the attributes + * are assigned default NULL values and CUDA_SUCCESS is returned. * - * If \p ptr was not allocated by, mapped by, or registered with a ::CUcontext which uses UVA - * (Unified Virtual Addressing), ::CUDA_ERROR_INVALID_CONTEXT is returned. + * If \p ptr was not allocated by, mapped by, or registered with a ::CUcontext + * which uses UVA (Unified Virtual Addressing), ::CUDA_ERROR_INVALID_CONTEXT is + * returned. * * \return * ::CUDA_SUCCESS, @@ -8790,7 +9295,9 @@ CUresult CUDAAPI cuPointerSetAttribute(const void *value, CUpointer_attribute at * ::cuPointerSetAttribute, * ::cudaPointerGetAttributes */ -CUresult CUDAAPI cuPointerGetAttributes(unsigned int numAttributes, CUpointer_attribute *attributes, void **data, CUdeviceptr ptr); +CUresult CUDAAPI cuPointerGetAttributes(unsigned int numAttributes, + CUpointer_attribute *attributes, + void **data, CUdeviceptr ptr); #endif /* __CUDA_API_VERSION >= 7000 */ /** @} */ /* END CUDA_UNIFIED */ @@ -8814,8 +9321,9 @@ CUresult CUDAAPI cuPointerGetAttributes(unsigned int numAttributes, CUpointer_at * determines behaviors of the stream. Valid values for \p Flags are: * - ::CU_STREAM_DEFAULT: Default stream creation flag. * - ::CU_STREAM_NON_BLOCKING: Specifies that work running in the created - * stream may run concurrently with work in stream 0 (the NULL stream), and that - * the created stream should perform no implicit synchronization with stream 0. + * stream may run concurrently with work in stream 0 (the NULL stream), and + * that the created stream should perform no implicit synchronization with + * stream 0. * * \param phStream - Returned newly created stream * \param Flags - Parameters for stream creation @@ -8845,22 +9353,23 @@ CUresult CUDAAPI cuStreamCreate(CUstream *phStream, unsigned int Flags); /** * \brief Create a stream with the given priority * - * Creates a stream with the specified priority and returns a handle in \p phStream. - * This API alters the scheduler priority of work in the stream. Work in a higher - * priority stream may preempt work already executing in a low priority stream. + * Creates a stream with the specified priority and returns a handle in \p + * phStream. This API alters the scheduler priority of work in the stream. Work + * in a higher priority stream may preempt work already executing in a low + * priority stream. * - * \p priority follows a convention where lower numbers represent higher priorities. - * '0' represents default priority. The range of meaningful numerical priorities can - * be queried using ::cuCtxGetStreamPriorityRange. If the specified priority is - * outside the numerical range returned by ::cuCtxGetStreamPriorityRange, - * it will automatically be clamped to the lowest or the highest number in the range. + * \p priority follows a convention where lower numbers represent higher + * priorities. '0' represents default priority. The range of meaningful + * numerical priorities can be queried using ::cuCtxGetStreamPriorityRange. If + * the specified priority is outside the numerical range returned by + * ::cuCtxGetStreamPriorityRange, it will automatically be clamped to the lowest + * or the highest number in the range. * * \param phStream - Returned newly created stream - * \param flags - Flags for stream creation. See ::cuStreamCreate for a list of - * valid flags - * \param priority - Stream priority. Lower numbers represent higher priorities. - * See ::cuCtxGetStreamPriorityRange for more information about - * meaningful stream priorities that can be passed. + * \param flags - Flags for stream creation. See ::cuStreamCreate for a + * list of valid flags \param priority - Stream priority. Lower numbers + * represent higher priorities. See ::cuCtxGetStreamPriorityRange for more + * information about meaningful stream priorities that can be passed. * * \return * ::CUDA_SUCCESS, @@ -8875,8 +9384,8 @@ CUresult CUDAAPI cuStreamCreate(CUstream *phStream, unsigned int Flags); * with compute capability 3.5 or higher. * * \note In the current implementation, only compute kernels launched in - * priority streams are affected by the stream's priority. Stream priorities have - * no effect on host-to-device and device-to-host memory operations. + * priority streams are affected by the stream's priority. Stream priorities + * have no effect on host-to-device and device-to-host memory operations. * * \sa ::cuStreamDestroy, * ::cuStreamCreate, @@ -8889,21 +9398,22 @@ CUresult CUDAAPI cuStreamCreate(CUstream *phStream, unsigned int Flags); * ::cuStreamAddCallback, * ::cudaStreamCreateWithPriority */ -CUresult CUDAAPI cuStreamCreateWithPriority(CUstream *phStream, unsigned int flags, int priority); - +CUresult CUDAAPI cuStreamCreateWithPriority(CUstream *phStream, + unsigned int flags, int priority); /** * \brief Query the priority of a given stream * - * Query the priority of a stream created using ::cuStreamCreate or ::cuStreamCreateWithPriority - * and return the priority in \p priority. Note that if the stream was created with a - * priority outside the numerical range returned by ::cuCtxGetStreamPriorityRange, - * this function returns the clamped priority. - * See ::cuStreamCreateWithPriority for details about priority clamping. + * Query the priority of a stream created using ::cuStreamCreate or + * ::cuStreamCreateWithPriority and return the priority in \p priority. Note + * that if the stream was created with a priority outside the numerical range + * returned by ::cuCtxGetStreamPriorityRange, this function returns the clamped + * priority. See ::cuStreamCreateWithPriority for details about priority + * clamping. * * \param hStream - Handle to the stream to be queried - * \param priority - Pointer to a signed integer in which the stream's priority is returned - * \return + * \param priority - Pointer to a signed integer in which the stream's + * priority is returned \return * ::CUDA_SUCCESS, * ::CUDA_ERROR_DEINITIALIZED, * ::CUDA_ERROR_NOT_INITIALIZED, @@ -8925,15 +9435,14 @@ CUresult CUDAAPI cuStreamGetPriority(CUstream hStream, int *priority); /** * \brief Query the flags of a given stream * - * Query the flags of a stream created using ::cuStreamCreate or ::cuStreamCreateWithPriority - * and return the flags in \p flags. + * Query the flags of a stream created using ::cuStreamCreate or + * ::cuStreamCreateWithPriority and return the flags in \p flags. * * \param hStream - Handle to the stream to be queried - * \param flags - Pointer to an unsigned integer in which the stream's flags are returned - * The value returned in \p flags is a logical 'OR' of all flags that - * were used while creating this stream. See ::cuStreamCreate for the list - * of valid flags - * \return + * \param flags - Pointer to an unsigned integer in which the stream's + * flags are returned The value returned in \p flags is a logical 'OR' of all + * flags that were used while creating this stream. See ::cuStreamCreate for the + * list of valid flags \return * ::CUDA_SUCCESS, * ::CUDA_ERROR_DEINITIALIZED, * ::CUDA_ERROR_NOT_INITIALIZED, @@ -8959,16 +9468,19 @@ CUresult CUDAAPI cuStreamGetFlags(CUstream hStream, unsigned int *flags); * * The stream handle \p hStream can refer to any of the following: *
    - *
  • a stream created via any of the CUDA driver APIs such as ::cuStreamCreate - * and ::cuStreamCreateWithPriority, or their runtime API equivalents such as - * ::cudaStreamCreate, ::cudaStreamCreateWithFlags and ::cudaStreamCreateWithPriority. - * The returned context is the context that was active in the calling thread when the - * stream was created. Passing an invalid handle will result in undefined behavior.
    • - *
  • any of the special streams such as the NULL stream, ::CU_STREAM_LEGACY and - * ::CU_STREAM_PER_THREAD. The runtime API equivalents of these are also accepted, - * which are NULL, ::cudaStreamLegacy and ::cudaStreamPerThread respectively. - * Specifying any of the special handles will return the context current to the - * calling thread. If no context is current to the calling thread, + *
  • a stream created via any of the CUDA driver APIs such as + * ::cuStreamCreate and ::cuStreamCreateWithPriority, or their runtime API + * equivalents such as + * ::cudaStreamCreate, ::cudaStreamCreateWithFlags and + * ::cudaStreamCreateWithPriority. The returned context is the context that was + * active in the calling thread when the stream was created. Passing an invalid + * handle will result in undefined behavior.
  • any of the special streams + * such as the NULL stream, ::CU_STREAM_LEGACY and + * ::CU_STREAM_PER_THREAD. The runtime API equivalents of these are also + * accepted, which are NULL, ::cudaStreamLegacy and ::cudaStreamPerThread + * respectively. Specifying any of the special handles will return the context + * current to the calling thread. If no context is current to the calling + * thread, * ::CUDA_ERROR_INVALID_CONTEXT is returned.
    • *
* @@ -9002,9 +9514,10 @@ CUresult CUDAAPI cuStreamGetCtx(CUstream hStream, CUcontext *pctx); * \brief Make a compute stream wait on an event * * Makes all future work submitted to \p hStream wait for all work captured in - * \p hEvent. See ::cuEventRecord() for details on what is captured by an event. - * The synchronization will be performed efficiently on the device when applicable. - * \p hEvent may be from a different context or device than \p hStream. + * \p hEvent. See ::cuEventRecord() for details on what is captured by an + * event. The synchronization will be performed efficiently on the device when + * applicable. \p hEvent may be from a different context or device than \p + * hStream. * * \param hStream - Stream to wait * \param hEvent - Event to wait on (may not be NULL) @@ -9027,7 +9540,8 @@ CUresult CUDAAPI cuStreamGetCtx(CUstream hStream, CUcontext *pctx); * ::cuStreamDestroy, * ::cudaStreamWaitEvent */ -CUresult CUDAAPI cuStreamWaitEvent(CUstream hStream, CUevent hEvent, unsigned int Flags); +CUresult CUDAAPI cuStreamWaitEvent(CUstream hStream, CUevent hEvent, + unsigned int Flags); /** * \brief Add a callback to a compute stream @@ -9078,9 +9592,9 @@ CUresult CUDAAPI cuStreamWaitEvent(CUstream hStream, CUevent hEvent, unsigned in * * * \param hStream - Stream to add callback to - * \param callback - The function to call once preceding stream operations are complete - * \param userData - User specified data to be passed to the callback function - * \param flags - Reserved for future use, must be 0 + * \param callback - The function to call once preceding stream operations are + * complete \param userData - User specified data to be passed to the callback + * function \param flags - Reserved for future use, must be 0 * * \return * ::CUDA_SUCCESS, @@ -9102,32 +9616,36 @@ CUresult CUDAAPI cuStreamWaitEvent(CUstream hStream, CUevent hEvent, unsigned in * ::cuStreamLaunchHostFunc, * ::cudaStreamAddCallback */ -CUresult CUDAAPI cuStreamAddCallback(CUstream hStream, CUstreamCallback callback, void *userData, unsigned int flags); +CUresult CUDAAPI cuStreamAddCallback(CUstream hStream, + CUstreamCallback callback, void *userData, + unsigned int flags); #if __CUDA_API_VERSION >= 10000 /** * \brief Begins graph capture on a stream * - * Begin graph capture on \p hStream. When a stream is in capture mode, all operations - * pushed into the stream will not be executed, but will instead be captured into - * a graph, which will be returned via ::cuStreamEndCapture. Capture may not be initiated - * if \p stream is CU_STREAM_LEGACY. Capture must be ended on the same stream in which - * it was initiated, and it may only be initiated if the stream is not already in capture - * mode. The capture mode may be queried via ::cuStreamIsCapturing. A unique id - * representing the capture sequence may be queried via ::cuStreamGetCaptureInfo. + * Begin graph capture on \p hStream. When a stream is in capture mode, all + * operations pushed into the stream will not be executed, but will instead be + * captured into a graph, which will be returned via ::cuStreamEndCapture. + * Capture may not be initiated if \p stream is CU_STREAM_LEGACY. Capture must + * be ended on the same stream in which it was initiated, and it may only be + * initiated if the stream is not already in capture mode. The capture mode may + * be queried via ::cuStreamIsCapturing. A unique id representing the capture + * sequence may be queried via ::cuStreamGetCaptureInfo. * - * If \p mode is not ::CU_STREAM_CAPTURE_MODE_RELAXED, ::cuStreamEndCapture must be - * called on this stream from the same thread. + * If \p mode is not ::CU_STREAM_CAPTURE_MODE_RELAXED, ::cuStreamEndCapture must + * be called on this stream from the same thread. * * \param hStream - Stream in which to initiate capture - * \param mode - Controls the interaction of this capture sequence with other API - * calls that are potentially unsafe. For more details see + * \param mode - Controls the interaction of this capture sequence with other + * API calls that are potentially unsafe. For more details see * ::cuThreadExchangeStreamCaptureMode. * - * \note Kernels captured using this API must not use texture and surface references. - * Reading or writing through any texture or surface reference is undefined - * behavior. This restriction does not apply to texture and surface objects. + * \note Kernels captured using this API must not use texture and surface + * references. Reading or writing through any texture or surface reference is + * undefined behavior. This restriction does not apply to texture and surface + * objects. * * \return * ::CUDA_SUCCESS, @@ -9142,7 +9660,8 @@ CUresult CUDAAPI cuStreamAddCallback(CUstream hStream, CUstreamCallback callback * ::cuStreamEndCapture, * ::cuThreadExchangeStreamCaptureMode */ -CUresult CUDAAPI cuStreamBeginCapture(CUstream hStream, CUstreamCaptureMode mode); +CUresult CUDAAPI cuStreamBeginCapture(CUstream hStream, + CUstreamCaptureMode mode); #endif /* __CUDA_API_VERSION >= 10000 */ #if __CUDA_API_VERSION >= 10010 @@ -9150,9 +9669,12 @@ CUresult CUDAAPI cuStreamBeginCapture(CUstream hStream, CUstreamCaptureMode mode /** * \brief Swaps the stream capture interaction mode for a thread * - * Sets the calling thread's stream capture interaction mode to the value contained - * in \p *mode, and overwrites \p *mode with the previous mode for the thread. To - * facilitate deterministic behavior across function or module boundaries, callers + * Sets the calling thread's stream capture interaction mode to the value + contained + * in \p *mode, and overwrites \p *mode with the previous mode for the thread. + To + * facilitate deterministic behavior across function or module boundaries, + callers * are encouraged to use this API in a push-pop fashion: \code CUstreamCaptureMode mode = desiredMode; cuThreadExchangeStreamCaptureMode(&mode); @@ -9160,30 +9682,43 @@ CUresult CUDAAPI cuStreamBeginCapture(CUstream hStream, CUstreamCaptureMode mode cuThreadExchangeStreamCaptureMode(&mode); // restore previous mode * \endcode * - * During stream capture (see ::cuStreamBeginCapture), some actions, such as a call + * During stream capture (see ::cuStreamBeginCapture), some actions, such as a + call * to ::cudaMalloc, may be unsafe. In the case of ::cudaMalloc, the operation is - * not enqueued asynchronously to a stream, and is not observed by stream capture. + * not enqueued asynchronously to a stream, and is not observed by stream + capture. * Therefore, if the sequence of operations captured via ::cuStreamBeginCapture * depended on the allocation being replayed whenever the graph is launched, the * captured graph would be invalid. * - * Therefore, stream capture places restrictions on API calls that can be made within - * or concurrently to a ::cuStreamBeginCapture-::cuStreamEndCapture sequence. This + * Therefore, stream capture places restrictions on API calls that can be made + within + * or concurrently to a ::cuStreamBeginCapture-::cuStreamEndCapture sequence. + This * behavior can be controlled via this API and flags to ::cuStreamBeginCapture. * * A thread's mode is one of the following: - * - \p CU_STREAM_CAPTURE_MODE_GLOBAL: This is the default mode. If the local thread has + * - \p CU_STREAM_CAPTURE_MODE_GLOBAL: This is the default mode. If the local + thread has * an ongoing capture sequence that was not initiated with - * \p CU_STREAM_CAPTURE_MODE_RELAXED at \p cuStreamBeginCapture, or if any other thread - * has a concurrent capture sequence initiated with \p CU_STREAM_CAPTURE_MODE_GLOBAL, + * \p CU_STREAM_CAPTURE_MODE_RELAXED at \p cuStreamBeginCapture, or if any + other thread + * has a concurrent capture sequence initiated with \p + CU_STREAM_CAPTURE_MODE_GLOBAL, * this thread is prohibited from potentially unsafe API calls. - * - \p CU_STREAM_CAPTURE_MODE_THREAD_LOCAL: If the local thread has an ongoing capture - * sequence not initiated with \p CU_STREAM_CAPTURE_MODE_RELAXED, it is prohibited - * from potentially unsafe API calls. Concurrent capture sequences in other threads + * - \p CU_STREAM_CAPTURE_MODE_THREAD_LOCAL: If the local thread has an ongoing + capture + * sequence not initiated with \p CU_STREAM_CAPTURE_MODE_RELAXED, it is + prohibited + * from potentially unsafe API calls. Concurrent capture sequences in other + threads * are ignored. - * - \p CU_STREAM_CAPTURE_MODE_RELAXED: The local thread is not prohibited from potentially - * unsafe API calls. Note that the thread is still prohibited from API calls which - * necessarily conflict with stream capture, for example, attempting ::cuEventQuery + * - \p CU_STREAM_CAPTURE_MODE_RELAXED: The local thread is not prohibited from + potentially + * unsafe API calls. Note that the thread is still prohibited from API calls + which + * necessarily conflict with stream capture, for example, attempting + ::cuEventQuery * on an event that was last recorded inside a capture sequence. * * \param mode - Pointer to mode value to swap with the current mode @@ -9207,9 +9742,9 @@ CUresult CUDAAPI cuThreadExchangeStreamCaptureMode(CUstreamCaptureMode *mode); * \brief Ends capture on a stream, returning the captured graph * * End capture on \p hStream, returning the captured graph via \p phGraph. - * Capture must have been initiated on \p hStream via a call to ::cuStreamBeginCapture. - * If capture was invalidated, due to a violation of the rules of stream capture, then - * a NULL graph will be returned. + * Capture must have been initiated on \p hStream via a call to + * ::cuStreamBeginCapture. If capture was invalidated, due to a violation of the + * rules of stream capture, then a NULL graph will be returned. * * If the \p mode argument to ::cuStreamBeginCapture was not * ::CU_STREAM_CAPTURE_MODE_RELAXED, this call must be from the same thread as @@ -9236,14 +9771,14 @@ CUresult CUDAAPI cuStreamEndCapture(CUstream hStream, CUgraph *phGraph); /** * \brief Returns a stream's capture status * - * Return the capture status of \p hStream via \p captureStatus. After a successful - * call, \p *captureStatus will contain one of the following: + * Return the capture status of \p hStream via \p captureStatus. After a + * successful call, \p *captureStatus will contain one of the following: * - ::CU_STREAM_CAPTURE_STATUS_NONE: The stream is not capturing. * - ::CU_STREAM_CAPTURE_STATUS_ACTIVE: The stream is capturing. - * - ::CU_STREAM_CAPTURE_STATUS_INVALIDATED: The stream was capturing but an error - * has invalidated the capture sequence. The capture sequence must be terminated - * with ::cuStreamEndCapture on the stream where it was initiated in order to - * continue using \p hStream. + * - ::CU_STREAM_CAPTURE_STATUS_INVALIDATED: The stream was capturing but an + * error has invalidated the capture sequence. The capture sequence must be + * terminated with ::cuStreamEndCapture on the stream where it was initiated in + * order to continue using \p hStream. * * Note that, if this is called on ::CU_STREAM_LEGACY (the "null stream") while * a blocking stream in the same context is capturing, it will return @@ -9271,7 +9806,8 @@ CUresult CUDAAPI cuStreamEndCapture(CUstream hStream, CUgraph *phGraph); * ::cuStreamBeginCapture, * ::cuStreamEndCapture */ -CUresult CUDAAPI cuStreamIsCapturing(CUstream hStream, CUstreamCaptureStatus *captureStatus); +CUresult CUDAAPI cuStreamIsCapturing(CUstream hStream, + CUstreamCaptureStatus *captureStatus); #endif /* __CUDA_API_VERSION >= 10000 */ @@ -9283,8 +9819,9 @@ CUresult CUDAAPI cuStreamIsCapturing(CUstream hStream, CUstreamCaptureStatus *ca * Query the capture status of a stream and and get an id for * the capture sequence, which is unique over the lifetime of the process. * - * If called on ::CU_STREAM_LEGACY (the "null stream") while a stream not created - * with ::CU_STREAM_NON_BLOCKING is capturing, returns ::CUDA_ERROR_STREAM_CAPTURE_IMPLICIT. + * If called on ::CU_STREAM_LEGACY (the "null stream") while a stream not + * created with ::CU_STREAM_NON_BLOCKING is capturing, returns + * ::CUDA_ERROR_STREAM_CAPTURE_IMPLICIT. * * A valid id is returned only if both of the following are true: * - the call returns CUDA_SUCCESS @@ -9299,7 +9836,9 @@ CUresult CUDAAPI cuStreamIsCapturing(CUstream hStream, CUstreamCaptureStatus *ca * ::cuStreamBeginCapture, * ::cuStreamIsCapturing */ - CUresult CUDAAPI cuStreamGetCaptureInfo(CUstream hStream, CUstreamCaptureStatus *captureStatus, cuuint64_t *id); +CUresult CUDAAPI cuStreamGetCaptureInfo(CUstream hStream, + CUstreamCaptureStatus *captureStatus, + cuuint64_t *id); #endif /* __CUDA_API_VERSION >= 10010 */ @@ -9334,19 +9873,21 @@ CUresult CUDAAPI cuStreamIsCapturing(CUstream hStream, CUstreamCaptureStatus *ca * If the ::CU_MEM_ATTACH_GLOBAL flag is specified, the memory can be accessed * by any stream on any device. * If the ::CU_MEM_ATTACH_HOST flag is specified, the program makes a guarantee - * that it won't access the memory on the device from any stream on a device that - * has a zero value for the device attribute ::CU_DEVICE_ATTRIBUTE_CONCURRENT_MANAGED_ACCESS. - * If the ::CU_MEM_ATTACH_SINGLE flag is specified and \p hStream is associated with - * a device that has a zero value for the device attribute ::CU_DEVICE_ATTRIBUTE_CONCURRENT_MANAGED_ACCESS, - * the program makes a guarantee that it will only access the memory on the device - * from \p hStream. It is illegal to attach singly to the NULL stream, because the - * NULL stream is a virtual global stream and not a specific stream. An error will - * be returned in this case. - * - * When memory is associated with a single stream, the Unified Memory system will - * allow CPU access to this memory region so long as all operations in \p hStream - * have completed, regardless of whether other streams are active. In effect, - * this constrains exclusive ownership of the managed memory region by + * that it won't access the memory on the device from any stream on a device + * that has a zero value for the device attribute + * ::CU_DEVICE_ATTRIBUTE_CONCURRENT_MANAGED_ACCESS. If the + * ::CU_MEM_ATTACH_SINGLE flag is specified and \p hStream is associated with a + * device that has a zero value for the device attribute + * ::CU_DEVICE_ATTRIBUTE_CONCURRENT_MANAGED_ACCESS, the program makes a + * guarantee that it will only access the memory on the device from \p hStream. + * It is illegal to attach singly to the NULL stream, because the NULL stream is + * a virtual global stream and not a specific stream. An error will be returned + * in this case. + * + * When memory is associated with a single stream, the Unified Memory system + * will allow CPU access to this memory region so long as all operations in \p + * hStream have completed, regardless of whether other streams are active. In + * effect, this constrains exclusive ownership of the managed memory region by * an active GPU to per-stream activity instead of whole-GPU activity. * * Accessing memory on the device from streams that are not associated with @@ -9359,12 +9900,13 @@ CUresult CUDAAPI cuStreamIsCapturing(CUstream hStream, CUstreamCaptureStatus *ca * at all times. Data visibility and coherency will be changed appropriately * for all kernels which follow a stream-association change. * - * If \p hStream is destroyed while data is associated with it, the association is - * removed and the association reverts to the default visibility of the allocation - * as specified at ::cuMemAllocManaged. For __managed__ variables, the default - * association is always ::CU_MEM_ATTACH_GLOBAL. Note that destroying a stream is an - * asynchronous operation, and as a result, the change to default association won't - * happen until all work in the stream has completed. + * If \p hStream is destroyed while data is associated with it, the association + * is removed and the association reverts to the default visibility of the + * allocation as specified at ::cuMemAllocManaged. For __managed__ variables, + * the default association is always ::CU_MEM_ATTACH_GLOBAL. Note that + * destroying a stream is an asynchronous operation, and as a result, the change + * to default association won't happen until all work in the stream has + * completed. * * \param hStream - Stream in which to enqueue the attach operation * \param dptr - Pointer to memory (must be a pointer to managed memory or @@ -9391,7 +9933,8 @@ CUresult CUDAAPI cuStreamIsCapturing(CUstream hStream, CUstreamCaptureStatus *ca * ::cuMemAllocManaged, * ::cudaStreamAttachMemAsync */ -CUresult CUDAAPI cuStreamAttachMemAsync(CUstream hStream, CUdeviceptr dptr, size_t length, unsigned int flags); +CUresult CUDAAPI cuStreamAttachMemAsync(CUstream hStream, CUdeviceptr dptr, + size_t length, unsigned int flags); #endif /* __CUDA_API_VERSION >= 6000 */ @@ -9488,7 +10031,6 @@ CUresult CUDAAPI cuStreamDestroy(CUstream hStream); /** @} */ /* END CUDA_STREAM */ - /** * \defgroup CUDA_EVENT Event Management * @@ -9504,13 +10046,13 @@ CUresult CUDAAPI cuStreamDestroy(CUstream hStream); /** * \brief Creates an event * - * Creates an event *phEvent for the current context with the flags specified via - * \p Flags. Valid flags include: + * Creates an event *phEvent for the current context with the flags specified + * via \p Flags. Valid flags include: * - ::CU_EVENT_DEFAULT: Default event creation flag. - * - ::CU_EVENT_BLOCKING_SYNC: Specifies that the created event should use blocking - * synchronization. A CPU thread that uses ::cuEventSynchronize() to wait on - * an event created with this flag will block until the event has actually - * been recorded. + * - ::CU_EVENT_BLOCKING_SYNC: Specifies that the created event should use + * blocking synchronization. A CPU thread that uses ::cuEventSynchronize() to + * wait on an event created with this flag will block until the event has + * actually been recorded. * - ::CU_EVENT_DISABLE_TIMING: Specifies that the created event does not need * to record timing data. Events created with this flag specified and * the ::CU_EVENT_BLOCKING_SYNC flag not specified will provide the best @@ -9719,147 +10261,152 @@ CUresult CUDAAPI cuEventDestroy(CUevent hEvent); * ::cuEventDestroy, * ::cudaEventElapsedTime */ -CUresult CUDAAPI cuEventElapsedTime(float *pMilliseconds, CUevent hStart, CUevent hEnd); +CUresult CUDAAPI cuEventElapsedTime(float *pMilliseconds, CUevent hStart, + CUevent hEnd); /** @} */ /* END CUDA_EVENT */ /** * \defgroup CUDA_EXTRES_INTEROP External Resource Interoperability * - * ___MANBRIEF___ External resource interoperability functions of the low-level CUDA driver API + * ___MANBRIEF___ External resource interoperability functions of the low-level + * CUDA driver API * (___CURRENT_FILE___) ___ENDMANBRIEF___ * - * This section describes the external resource interoperability functions of the low-level CUDA - * driver application programming interface. + * This section describes the external resource interoperability functions of + * the low-level CUDA driver application programming interface. * * @{ */ #if __CUDA_API_VERSION >= 10000 - /** - * \brief Imports an external memory object - * - * Imports an externally allocated memory object and returns - * a handle to that in \p extMem_out. - * - * The properties of the handle being imported must be described in - * \p memHandleDesc. The ::CUDA_EXTERNAL_MEMORY_HANDLE_DESC structure - * is defined as follows: - * - * \code - typedef struct CUDA_EXTERNAL_MEMORY_HANDLE_DESC_st { - CUexternalMemoryHandleType type; - union { - int fd; - struct { - void *handle; - const void *name; - } win32; - } handle; - unsigned long long size; - unsigned int flags; - } CUDA_EXTERNAL_MEMORY_HANDLE_DESC; - * \endcode - * - * where ::CUDA_EXTERNAL_MEMORY_HANDLE_DESC::type specifies the type - * of handle being imported. ::CUexternalMemoryHandleType is - * defined as: - * - * \code - typedef enum CUexternalMemoryHandleType_enum { - CU_EXTERNAL_MEMORY_HANDLE_TYPE_OPAQUE_FD = 1, - CU_EXTERNAL_MEMORY_HANDLE_TYPE_OPAQUE_WIN32 = 2, - CU_EXTERNAL_MEMORY_HANDLE_TYPE_OPAQUE_WIN32_KMT = 3, - CU_EXTERNAL_MEMORY_HANDLE_TYPE_D3D12_HEAP = 4, - CU_EXTERNAL_MEMORY_HANDLE_TYPE_D3D12_RESOURCE = 5 - } CUexternalMemoryHandleType; - * \endcode - * - * If ::CUDA_EXTERNAL_MEMORY_HANDLE_DESC::type is - * ::CU_EXTERNAL_MEMORY_HANDLE_TYPE_OPAQUE_FD, then - * ::CUDA_EXTERNAL_MEMORY_HANDLE_DESC::handle::fd must be a valid - * file descriptor referencing a memory object. Ownership of - * the file descriptor is transferred to the CUDA driver when the - * handle is imported successfully. Performing any operations on the - * file descriptor after it is imported results in undefined behavior. - * - * If ::CUDA_EXTERNAL_MEMORY_HANDLE_DESC::type is - * ::CU_EXTERNAL_MEMORY_HANDLE_TYPE_OPAQUE_WIN32, then exactly one - * of ::CUDA_EXTERNAL_MEMORY_HANDLE_DESC::handle::win32::handle and - * ::CUDA_EXTERNAL_MEMORY_HANDLE_DESC::handle::win32::name must not be - * NULL. If ::CUDA_EXTERNAL_MEMORY_HANDLE_DESC::handle::win32::handle - * is not NULL, then it must represent a valid shared NT handle that - * references a memory object. Ownership of this handle is - * not transferred to CUDA after the import operation, so the - * application must release the handle using the appropriate system - * call. If ::CUDA_EXTERNAL_MEMORY_HANDLE_DESC::handle::win32::name - * is not NULL, then it must point to a NULL-terminated array of - * UTF-16 characters that refers to a memory object. - * - * If ::CUDA_EXTERNAL_MEMORY_HANDLE_DESC::type is - * ::CU_EXTERNAL_MEMORY_HANDLE_TYPE_OPAQUE_WIN32_KMT, then - * ::CUDA_EXTERNAL_MEMORY_HANDLE_DESC::handle::win32::handle must - * be non-NULL and - * ::CUDA_EXTERNAL_MEMORY_HANDLE_DESC::handle::win32::name - * must be NULL. The handle specified must be a globally shared KMT - * handle. This handle does not hold a reference to the underlying - * object, and thus will be invalid when all references to the - * memory object are destroyed. - * - * If ::CUDA_EXTERNAL_MEMORY_HANDLE_DESC::type is - * ::CU_EXTERNAL_MEMORY_HANDLE_TYPE_D3D12_HEAP, then exactly one - * of ::CUDA_EXTERNAL_MEMORY_HANDLE_DESC::handle::win32::handle and - * ::CUDA_EXTERNAL_MEMORY_HANDLE_DESC::handle::win32::name must not be - * NULL. If ::CUDA_EXTERNAL_MEMORY_HANDLE_DESC::handle::win32::handle - * is not NULL, then it must represent a valid shared NT handle that - * is returned by ID3DDevice::CreateSharedHandle when referring to a - * ID3D12Heap object. This handle holds a reference to the underlying - * object. If ::CUDA_EXTERNAL_MEMORY_HANDLE_DESC::handle::win32::name - * is not NULL, then it must point to a NULL-terminated array of - * UTF-16 characters that refers to a ID3D12Heap object. - * - * If ::CUDA_EXTERNAL_MEMORY_HANDLE_DESC::type is - * ::CU_EXTERNAL_MEMORY_HANDLE_TYPE_D3D12_RESOURCE, then exactly one - * of ::CUDA_EXTERNAL_MEMORY_HANDLE_DESC::handle::win32::handle and - * ::CUDA_EXTERNAL_MEMORY_HANDLE_DESC::handle::win32::name must not be - * NULL. If ::CUDA_EXTERNAL_MEMORY_HANDLE_DESC::handle::win32::handle - * is not NULL, then it must represent a valid shared NT handle that - * is returned by ID3DDevice::CreateSharedHandle when referring to a - * ID3D12Resource object. This handle holds a reference to the - * underlying object. If - * ::CUDA_EXTERNAL_MEMORY_HANDLE_DESC::handle::win32::name - * is not NULL, then it must point to a NULL-terminated array of - * UTF-16 characters that refers to a ID3D12Resource object. - * - * The size of the memory object must be specified in - * ::CUDA_EXTERNAL_MEMORY_HANDLE_DESC::size. - * - * Specifying the flag ::CUDA_EXTERNAL_MEMORY_DEDICATED in - * ::CUDA_EXTERNAL_MEMORY_HANDLE_DESC::flags indicates that the - * resource is a dedicated resource. The definition of what a - * dedicated resource is outside the scope of this extension. - * - * \param extMem_out - Returned handle to an external memory object - * \param memHandleDesc - Memory import handle descriptor - * - * \return - * ::CUDA_SUCCESS, - * ::CUDA_ERROR_NOT_INITIALIZED, - * ::CUDA_ERROR_INVALID_HANDLE - * \notefnerr - * - * \note If the Vulkan memory imported into CUDA is mapped on the CPU then the - * application must use vkInvalidateMappedMemoryRanges/vkFlushMappedMemoryRanges - * as well as appropriate Vulkan pipeline barriers to maintain coherence between - * CPU and GPU. For more information on these APIs, please refer to "Synchronization - * and Cache Control" chapter from Vulkan specification. - * - * \sa ::cuDestroyExternalMemory, - * ::cuExternalMemoryGetMappedBuffer, - * ::cuExternalMemoryGetMappedMipmappedArray - */ -CUresult CUDAAPI cuImportExternalMemory(CUexternalMemory *extMem_out, const CUDA_EXTERNAL_MEMORY_HANDLE_DESC *memHandleDesc); +/** +* \brief Imports an external memory object +* +* Imports an externally allocated memory object and returns +* a handle to that in \p extMem_out. +* +* The properties of the handle being imported must be described in +* \p memHandleDesc. The ::CUDA_EXTERNAL_MEMORY_HANDLE_DESC structure +* is defined as follows: +* +* \code + typedef struct CUDA_EXTERNAL_MEMORY_HANDLE_DESC_st { + CUexternalMemoryHandleType type; + union { + int fd; + struct { + void *handle; + const void *name; + } win32; + } handle; + unsigned long long size; + unsigned int flags; + } CUDA_EXTERNAL_MEMORY_HANDLE_DESC; +* \endcode +* +* where ::CUDA_EXTERNAL_MEMORY_HANDLE_DESC::type specifies the type +* of handle being imported. ::CUexternalMemoryHandleType is +* defined as: +* +* \code + typedef enum CUexternalMemoryHandleType_enum { + CU_EXTERNAL_MEMORY_HANDLE_TYPE_OPAQUE_FD = 1, + CU_EXTERNAL_MEMORY_HANDLE_TYPE_OPAQUE_WIN32 = 2, + CU_EXTERNAL_MEMORY_HANDLE_TYPE_OPAQUE_WIN32_KMT = 3, + CU_EXTERNAL_MEMORY_HANDLE_TYPE_D3D12_HEAP = 4, + CU_EXTERNAL_MEMORY_HANDLE_TYPE_D3D12_RESOURCE = 5 + } CUexternalMemoryHandleType; +* \endcode +* +* If ::CUDA_EXTERNAL_MEMORY_HANDLE_DESC::type is +* ::CU_EXTERNAL_MEMORY_HANDLE_TYPE_OPAQUE_FD, then +* ::CUDA_EXTERNAL_MEMORY_HANDLE_DESC::handle::fd must be a valid +* file descriptor referencing a memory object. Ownership of +* the file descriptor is transferred to the CUDA driver when the +* handle is imported successfully. Performing any operations on the +* file descriptor after it is imported results in undefined behavior. +* +* If ::CUDA_EXTERNAL_MEMORY_HANDLE_DESC::type is +* ::CU_EXTERNAL_MEMORY_HANDLE_TYPE_OPAQUE_WIN32, then exactly one +* of ::CUDA_EXTERNAL_MEMORY_HANDLE_DESC::handle::win32::handle and +* ::CUDA_EXTERNAL_MEMORY_HANDLE_DESC::handle::win32::name must not be +* NULL. If ::CUDA_EXTERNAL_MEMORY_HANDLE_DESC::handle::win32::handle +* is not NULL, then it must represent a valid shared NT handle that +* references a memory object. Ownership of this handle is +* not transferred to CUDA after the import operation, so the +* application must release the handle using the appropriate system +* call. If ::CUDA_EXTERNAL_MEMORY_HANDLE_DESC::handle::win32::name +* is not NULL, then it must point to a NULL-terminated array of +* UTF-16 characters that refers to a memory object. +* +* If ::CUDA_EXTERNAL_MEMORY_HANDLE_DESC::type is +* ::CU_EXTERNAL_MEMORY_HANDLE_TYPE_OPAQUE_WIN32_KMT, then +* ::CUDA_EXTERNAL_MEMORY_HANDLE_DESC::handle::win32::handle must +* be non-NULL and +* ::CUDA_EXTERNAL_MEMORY_HANDLE_DESC::handle::win32::name +* must be NULL. The handle specified must be a globally shared KMT +* handle. This handle does not hold a reference to the underlying +* object, and thus will be invalid when all references to the +* memory object are destroyed. +* +* If ::CUDA_EXTERNAL_MEMORY_HANDLE_DESC::type is +* ::CU_EXTERNAL_MEMORY_HANDLE_TYPE_D3D12_HEAP, then exactly one +* of ::CUDA_EXTERNAL_MEMORY_HANDLE_DESC::handle::win32::handle and +* ::CUDA_EXTERNAL_MEMORY_HANDLE_DESC::handle::win32::name must not be +* NULL. If ::CUDA_EXTERNAL_MEMORY_HANDLE_DESC::handle::win32::handle +* is not NULL, then it must represent a valid shared NT handle that +* is returned by ID3DDevice::CreateSharedHandle when referring to a +* ID3D12Heap object. This handle holds a reference to the underlying +* object. If ::CUDA_EXTERNAL_MEMORY_HANDLE_DESC::handle::win32::name +* is not NULL, then it must point to a NULL-terminated array of +* UTF-16 characters that refers to a ID3D12Heap object. +* +* If ::CUDA_EXTERNAL_MEMORY_HANDLE_DESC::type is +* ::CU_EXTERNAL_MEMORY_HANDLE_TYPE_D3D12_RESOURCE, then exactly one +* of ::CUDA_EXTERNAL_MEMORY_HANDLE_DESC::handle::win32::handle and +* ::CUDA_EXTERNAL_MEMORY_HANDLE_DESC::handle::win32::name must not be +* NULL. If ::CUDA_EXTERNAL_MEMORY_HANDLE_DESC::handle::win32::handle +* is not NULL, then it must represent a valid shared NT handle that +* is returned by ID3DDevice::CreateSharedHandle when referring to a +* ID3D12Resource object. This handle holds a reference to the +* underlying object. If +* ::CUDA_EXTERNAL_MEMORY_HANDLE_DESC::handle::win32::name +* is not NULL, then it must point to a NULL-terminated array of +* UTF-16 characters that refers to a ID3D12Resource object. +* +* The size of the memory object must be specified in +* ::CUDA_EXTERNAL_MEMORY_HANDLE_DESC::size. +* +* Specifying the flag ::CUDA_EXTERNAL_MEMORY_DEDICATED in +* ::CUDA_EXTERNAL_MEMORY_HANDLE_DESC::flags indicates that the +* resource is a dedicated resource. The definition of what a +* dedicated resource is outside the scope of this extension. +* +* \param extMem_out - Returned handle to an external memory object +* \param memHandleDesc - Memory import handle descriptor +* +* \return +* ::CUDA_SUCCESS, +* ::CUDA_ERROR_NOT_INITIALIZED, +* ::CUDA_ERROR_INVALID_HANDLE +* \notefnerr +* +* \note If the Vulkan memory imported into CUDA is mapped on the CPU then the +* application must use vkInvalidateMappedMemoryRanges/vkFlushMappedMemoryRanges +* as well as appropriate Vulkan pipeline barriers to maintain coherence between +* CPU and GPU. For more information on these APIs, please refer to +"Synchronization +* and Cache Control" chapter from Vulkan specification. +* +* \sa ::cuDestroyExternalMemory, +* ::cuExternalMemoryGetMappedBuffer, +* ::cuExternalMemoryGetMappedMipmappedArray +*/ +CUresult CUDAAPI +cuImportExternalMemory(CUexternalMemory *extMem_out, + const CUDA_EXTERNAL_MEMORY_HANDLE_DESC *memHandleDesc); /** * \brief Maps a buffer onto an imported memory object @@ -9912,7 +10459,9 @@ CUresult CUDAAPI cuImportExternalMemory(CUexternalMemory *extMem_out, const CUDA * ::cuDestroyExternalMemory, * ::cuExternalMemoryGetMappedMipmappedArray */ -CUresult CUDAAPI cuExternalMemoryGetMappedBuffer(CUdeviceptr *devPtr, CUexternalMemory extMem, const CUDA_EXTERNAL_MEMORY_BUFFER_DESC *bufferDesc); +CUresult CUDAAPI cuExternalMemoryGetMappedBuffer( + CUdeviceptr *devPtr, CUexternalMemory extMem, + const CUDA_EXTERNAL_MEMORY_BUFFER_DESC *bufferDesc); /** * \brief Maps a CUDA mipmapped array onto an external memory object @@ -9945,7 +10494,8 @@ CUresult CUDAAPI cuExternalMemoryGetMappedBuffer(CUdeviceptr *devPtr, CUexternal * ::CUDA_EXTERNAL_MEMORY_MIPMAPPED_ARRAY_DESC::numLevels specifies * the total number of levels in the mipmap chain. * - * The returned CUDA mipmapped array must be freed using ::cuMipmappedArrayDestroy. + * The returned CUDA mipmapped array must be freed using + ::cuMipmappedArrayDestroy. * * \param mipmap - Returned CUDA mipmapped array * \param extMem - Handle to external memory object @@ -9961,7 +10511,9 @@ CUresult CUDAAPI cuExternalMemoryGetMappedBuffer(CUdeviceptr *devPtr, CUexternal * ::cuDestroyExternalMemory, * ::cuExternalMemoryGetMappedBuffer */ -CUresult CUDAAPI cuExternalMemoryGetMappedMipmappedArray(CUmipmappedArray *mipmap, CUexternalMemory extMem, const CUDA_EXTERNAL_MEMORY_MIPMAPPED_ARRAY_DESC *mipmapDesc); +CUresult CUDAAPI cuExternalMemoryGetMappedMipmappedArray( + CUmipmappedArray *mipmap, CUexternalMemory extMem, + const CUDA_EXTERNAL_MEMORY_MIPMAPPED_ARRAY_DESC *mipmapDesc); /** * \brief Destroys an external memory object. @@ -10080,7 +10632,9 @@ CUresult CUDAAPI cuDestroyExternalMemory(CUexternalMemory extMem); * ::cuSignalExternalSemaphoresAsync, * ::cuWaitExternalSemaphoresAsync */ -CUresult CUDAAPI cuImportExternalSemaphore(CUexternalSemaphore *extSem_out, const CUDA_EXTERNAL_SEMAPHORE_HANDLE_DESC *semHandleDesc); +CUresult CUDAAPI cuImportExternalSemaphore( + CUexternalSemaphore *extSem_out, + const CUDA_EXTERNAL_SEMAPHORE_HANDLE_DESC *semHandleDesc); /** * \brief Signals a set of external semaphore objects @@ -10118,7 +10672,10 @@ CUresult CUDAAPI cuImportExternalSemaphore(CUexternalSemaphore *extSem_out, cons * ::cuDestroyExternalSemaphore, * ::cuWaitExternalSemaphoresAsync */ -CUresult CUDAAPI cuSignalExternalSemaphoresAsync(const CUexternalSemaphore *extSemArray, const CUDA_EXTERNAL_SEMAPHORE_SIGNAL_PARAMS *paramsArray, unsigned int numExtSems, CUstream stream); +CUresult CUDAAPI cuSignalExternalSemaphoresAsync( + const CUexternalSemaphore *extSemArray, + const CUDA_EXTERNAL_SEMAPHORE_SIGNAL_PARAMS *paramsArray, + unsigned int numExtSems, CUstream stream); /** * \brief Waits on a set of external semaphore objects @@ -10160,7 +10717,10 @@ CUresult CUDAAPI cuSignalExternalSemaphoresAsync(const CUexternalSemaphore *extS * ::cuDestroyExternalSemaphore, * ::cuSignalExternalSemaphoresAsync */ -CUresult CUDAAPI cuWaitExternalSemaphoresAsync(const CUexternalSemaphore *extSemArray, const CUDA_EXTERNAL_SEMAPHORE_WAIT_PARAMS *paramsArray, unsigned int numExtSems, CUstream stream); +CUresult CUDAAPI cuWaitExternalSemaphoresAsync( + const CUexternalSemaphore *extSemArray, + const CUDA_EXTERNAL_SEMAPHORE_WAIT_PARAMS *paramsArray, + unsigned int numExtSems, CUstream stream); /** * \brief Destroys an external semaphore @@ -10247,7 +10807,8 @@ CUresult CUDAAPI cuDestroyExternalSemaphore(CUexternalSemaphore extSem); * Support for this can be queried with ::cuDeviceGetAttribute() and * ::CU_DEVICE_ATTRIBUTE_CAN_USE_STREAM_MEM_OPS. * - * Support for CU_STREAM_WAIT_VALUE_NOR can be queried with ::cuDeviceGetAttribute() and + * Support for CU_STREAM_WAIT_VALUE_NOR can be queried with + * ::cuDeviceGetAttribute() and * ::CU_DEVICE_ATTRIBUTE_CAN_USE_STREAM_WAIT_VALUE_NOR. * * \param stream The stream to synchronize on the memory location. @@ -10268,7 +10829,8 @@ CUresult CUDAAPI cuDestroyExternalSemaphore(CUexternalSemaphore extSem); * ::cuMemHostRegister, * ::cuStreamWaitEvent */ -CUresult CUDAAPI cuStreamWaitValue32(CUstream stream, CUdeviceptr addr, cuuint32_t value, unsigned int flags); +CUresult CUDAAPI cuStreamWaitValue32(CUstream stream, CUdeviceptr addr, + cuuint32_t value, unsigned int flags); /** * \brief Wait on a memory location @@ -10303,7 +10865,8 @@ CUresult CUDAAPI cuStreamWaitValue32(CUstream stream, CUdeviceptr addr, cuuint32 * ::cuMemHostRegister, * ::cuStreamWaitEvent */ -CUresult CUDAAPI cuStreamWaitValue64(CUstream stream, CUdeviceptr addr, cuuint64_t value, unsigned int flags); +CUresult CUDAAPI cuStreamWaitValue64(CUstream stream, CUdeviceptr addr, + cuuint64_t value, unsigned int flags); /** * \brief Write a value to memory @@ -10338,7 +10901,8 @@ CUresult CUDAAPI cuStreamWaitValue64(CUstream stream, CUdeviceptr addr, cuuint64 * ::cuMemHostRegister, * ::cuEventRecord */ -CUresult CUDAAPI cuStreamWriteValue32(CUstream stream, CUdeviceptr addr, cuuint32_t value, unsigned int flags); +CUresult CUDAAPI cuStreamWriteValue32(CUstream stream, CUdeviceptr addr, + cuuint32_t value, unsigned int flags); /** * \brief Write a value to memory @@ -10372,15 +10936,17 @@ CUresult CUDAAPI cuStreamWriteValue32(CUstream stream, CUdeviceptr addr, cuuint3 * ::cuMemHostRegister, * ::cuEventRecord */ -CUresult CUDAAPI cuStreamWriteValue64(CUstream stream, CUdeviceptr addr, cuuint64_t value, unsigned int flags); +CUresult CUDAAPI cuStreamWriteValue64(CUstream stream, CUdeviceptr addr, + cuuint64_t value, unsigned int flags); /** * \brief Batch operations to synchronize the stream via memory operations * - * This is a batch version of ::cuStreamWaitValue32() and ::cuStreamWriteValue32(). - * Batching operations may avoid some performance overhead in both the API call - * and the device execution versus adding them to the stream in separate API - * calls. The operations are enqueued in the order they appear in the array. + * This is a batch version of ::cuStreamWaitValue32() and + * ::cuStreamWriteValue32(). Batching operations may avoid some performance + * overhead in both the API call and the device execution versus adding them to + * the stream in separate API calls. The operations are enqueued in the order + * they appear in the array. * * See ::CUstreamBatchMemOpType for the full set of supported operations, and * ::cuStreamWaitValue32(), ::cuStreamWaitValue64(), ::cuStreamWriteValue32(), @@ -10407,7 +10973,9 @@ CUresult CUDAAPI cuStreamWriteValue64(CUstream stream, CUdeviceptr addr, cuuint6 * ::cuStreamWriteValue64, * ::cuMemHostRegister */ -CUresult CUDAAPI cuStreamBatchMemOp(CUstream stream, unsigned int count, CUstreamBatchMemOpParams *paramArray, unsigned int flags); +CUresult CUDAAPI cuStreamBatchMemOp(CUstream stream, unsigned int count, + CUstreamBatchMemOpParams *paramArray, + unsigned int flags); #endif /* __CUDA_API_VERSION >= 8000 */ /** @} */ /* END CUDA_MEMOP */ @@ -10456,10 +11024,10 @@ CUresult CUDAAPI cuStreamBatchMemOp(CUstream stream, unsigned int count, CUstrea * version. * - ::CU_FUNC_CACHE_MODE_CA: The attribute to indicate whether the function has * been compiled with user specified option "-Xptxas --dlcm=ca" set . - * - ::CU_FUNC_ATTRIBUTE_MAX_DYNAMIC_SHARED_SIZE_BYTES: The maximum size in bytes of - * dynamically-allocated shared memory. - * - ::CU_FUNC_ATTRIBUTE_PREFERRED_SHARED_MEMORY_CARVEOUT: Preferred shared memory-L1 - * cache split ratio in percent of total shared memory. + * - ::CU_FUNC_ATTRIBUTE_MAX_DYNAMIC_SHARED_SIZE_BYTES: The maximum size in + * bytes of dynamically-allocated shared memory. + * - ::CU_FUNC_ATTRIBUTE_PREFERRED_SHARED_MEMORY_CARVEOUT: Preferred shared + * memory-L1 cache split ratio in percent of total shared memory. * * \param pi - Returned attribute value * \param attrib - Attribute requested @@ -10481,33 +11049,35 @@ CUresult CUDAAPI cuStreamBatchMemOp(CUstream stream, unsigned int count, CUstrea * ::cudaFuncGetAttributes * ::cudaFuncSetAttribute */ -CUresult CUDAAPI cuFuncGetAttribute(int *pi, CUfunction_attribute attrib, CUfunction hfunc); +CUresult CUDAAPI cuFuncGetAttribute(int *pi, CUfunction_attribute attrib, + CUfunction hfunc); #if __CUDA_API_VERSION >= 9000 /** * \brief Sets information about a function * - * This call sets the value of a specified attribute \p attrib on the kernel given - * by \p hfunc to an integer value specified by \p val - * This function returns CUDA_SUCCESS if the new value of the attribute could be - * successfully set. If the set fails, this call will return an error. - * Not all attributes can have values set. Attempting to set a value on a read-only - * attribute will result in an error (CUDA_ERROR_INVALID_VALUE) + * This call sets the value of a specified attribute \p attrib on the kernel + * given by \p hfunc to an integer value specified by \p val This function + * returns CUDA_SUCCESS if the new value of the attribute could be successfully + * set. If the set fails, this call will return an error. Not all attributes can + * have values set. Attempting to set a value on a read-only attribute will + * result in an error (CUDA_ERROR_INVALID_VALUE) * * Supported attributes for the cuFuncSetAttribute call are: - * - ::CU_FUNC_ATTRIBUTE_MAX_DYNAMIC_SHARED_SIZE_BYTES: This maximum size in bytes of - * dynamically-allocated shared memory. The value should contain the requested - * maximum size of dynamically-allocated shared memory. The sum of this value and - * the function attribute ::CU_FUNC_ATTRIBUTE_SHARED_SIZE_BYTES cannot exceed the - * device attribute ::CU_DEVICE_ATTRIBUTE_MAX_SHARED_MEMORY_PER_BLOCK_OPTIN. - * The maximal size of requestable dynamic shared memory may differ by GPU - * architecture. - * - ::CU_FUNC_ATTRIBUTE_PREFERRED_SHARED_MEMORY_CARVEOUT: On devices where the L1 - * cache and shared memory use the same hardware resources, this sets the shared memory - * carveout preference, in percent of the total shared memory. - * See ::CU_DEVICE_ATTRIBUTE_MAX_SHARED_MEMORY_PER_MULTIPROCESSOR - * This is only a hint, and the driver can choose a different ratio if required to execute the function. + * - ::CU_FUNC_ATTRIBUTE_MAX_DYNAMIC_SHARED_SIZE_BYTES: This maximum size in + * bytes of dynamically-allocated shared memory. The value should contain the + * requested maximum size of dynamically-allocated shared memory. The sum of + * this value and the function attribute ::CU_FUNC_ATTRIBUTE_SHARED_SIZE_BYTES + * cannot exceed the device attribute + * ::CU_DEVICE_ATTRIBUTE_MAX_SHARED_MEMORY_PER_BLOCK_OPTIN. The maximal size of + * requestable dynamic shared memory may differ by GPU architecture. + * - ::CU_FUNC_ATTRIBUTE_PREFERRED_SHARED_MEMORY_CARVEOUT: On devices where the + * L1 cache and shared memory use the same hardware resources, this sets the + * shared memory carveout preference, in percent of the total shared memory. See + * ::CU_DEVICE_ATTRIBUTE_MAX_SHARED_MEMORY_PER_MULTIPROCESSOR This is only a + * hint, and the driver can choose a different ratio if required to execute the + * function. * * \param hfunc - Function to query attribute of * \param attrib - Attribute requested @@ -10529,8 +11099,9 @@ CUresult CUDAAPI cuFuncGetAttribute(int *pi, CUfunction_attribute attrib, CUfunc * ::cudaFuncGetAttributes * ::cudaFuncSetAttribute */ -CUresult CUDAAPI cuFuncSetAttribute(CUfunction hfunc, CUfunction_attribute attrib, int value); -#endif // __CUDA_API_VERSION >= 9000 +CUresult CUDAAPI cuFuncSetAttribute(CUfunction hfunc, + CUfunction_attribute attrib, int value); +#endif // __CUDA_API_VERSION >= 9000 /** * \brief Sets the preferred cache configuration for a device function @@ -10552,8 +11123,10 @@ CUresult CUDAAPI cuFuncSetAttribute(CUfunction hfunc, CUfunction_attribute attri * * * The supported cache configurations are: - * - ::CU_FUNC_CACHE_PREFER_NONE: no preference for shared memory or L1 (default) - * - ::CU_FUNC_CACHE_PREFER_SHARED: prefer larger shared memory and smaller L1 cache + * - ::CU_FUNC_CACHE_PREFER_NONE: no preference for shared memory or L1 + * (default) + * - ::CU_FUNC_CACHE_PREFER_SHARED: prefer larger shared memory and smaller L1 + * cache * - ::CU_FUNC_CACHE_PREFER_L1: prefer larger L1 cache and smaller shared memory * - ::CU_FUNC_CACHE_PREFER_EQUAL: prefer equal sized L1 cache and shared memory * @@ -10594,9 +11167,9 @@ CUresult CUDAAPI cuFuncSetCacheConfig(CUfunction hfunc, CUfunc_cache config); * * Changing the shared memory bank size will not increase shared memory usage * or affect occupancy of kernels, but may have major effects on performance. - * Larger bank sizes will allow for greater potential bandwidth to shared memory, - * but will change what kinds of accesses to shared memory will result in bank - * conflicts. + * Larger bank sizes will allow for greater potential bandwidth to shared + * memory, but will change what kinds of accesses to shared memory will result + * in bank conflicts. * * This function will do nothing on devices with fixed shared memory bank size. * @@ -10605,8 +11178,8 @@ CUresult CUDAAPI cuFuncSetCacheConfig(CUfunction hfunc, CUfunc_cache config); * configuration when launching this function. * - ::CU_SHARED_MEM_CONFIG_FOUR_BYTE_BANK_SIZE: set shared memory bank width to * be natively four bytes when launching this function. - * - ::CU_SHARED_MEM_CONFIG_EIGHT_BYTE_BANK_SIZE: set shared memory bank width to - * be natively eight bytes when launching this function. + * - ::CU_SHARED_MEM_CONFIG_EIGHT_BYTE_BANK_SIZE: set shared memory bank width + * to be natively eight bytes when launching this function. * * \param hfunc - kernel to be given a shared memory config * \param config - requested shared memory configuration @@ -10627,7 +11200,8 @@ CUresult CUDAAPI cuFuncSetCacheConfig(CUfunction hfunc, CUfunc_cache config); * ::cuLaunchKernel, * ::cudaFuncSetSharedMemConfig */ -CUresult CUDAAPI cuFuncSetSharedMemConfig(CUfunction hfunc, CUsharedconfig config); +CUresult CUDAAPI cuFuncSetSharedMemConfig(CUfunction hfunc, + CUsharedconfig config); #endif #if __CUDA_API_VERSION >= 4000 @@ -10742,21 +11316,17 @@ CUresult CUDAAPI cuFuncSetSharedMemConfig(CUfunction hfunc, CUsharedconfig confi * ::cuFuncGetAttribute, * ::cudaLaunchKernel */ -CUresult CUDAAPI cuLaunchKernel(CUfunction f, - unsigned int gridDimX, - unsigned int gridDimY, - unsigned int gridDimZ, - unsigned int blockDimX, - unsigned int blockDimY, +CUresult CUDAAPI cuLaunchKernel(CUfunction f, unsigned int gridDimX, + unsigned int gridDimY, unsigned int gridDimZ, + unsigned int blockDimX, unsigned int blockDimY, unsigned int blockDimZ, - unsigned int sharedMemBytes, - CUstream hStream, - void **kernelParams, - void **extra); + unsigned int sharedMemBytes, CUstream hStream, + void **kernelParams, void **extra); #endif /* __CUDA_API_VERSION >= 4000 */ #if __CUDA_API_VERSION >= 9000 /** - * \brief Launches a CUDA function where thread blocks can cooperate and synchronize as they execute + * \brief Launches a CUDA function where thread blocks can cooperate and + * synchronize as they execute * * Invokes the kernel \p f on a \p gridDimX x \p gridDimY x \p gridDimZ * grid of blocks. Each block contains \p blockDimX x \p blockDimY x @@ -10768,10 +11338,12 @@ CUresult CUDAAPI cuLaunchKernel(CUfunction f, * The device on which this kernel is invoked must have a non-zero value for * the device attribute ::CU_DEVICE_ATTRIBUTE_COOPERATIVE_LAUNCH. * - * The total number of blocks launched cannot exceed the maximum number of blocks per - * multiprocessor as returned by ::cuOccupancyMaxActiveBlocksPerMultiprocessor (or - * ::cuOccupancyMaxActiveBlocksPerMultiprocessorWithFlags) times the number of multiprocessors - * as specified by the device attribute ::CU_DEVICE_ATTRIBUTE_MULTIPROCESSOR_COUNT. + * The total number of blocks launched cannot exceed the maximum number of + * blocks per multiprocessor as returned by + * ::cuOccupancyMaxActiveBlocksPerMultiprocessor (or + * ::cuOccupancyMaxActiveBlocksPerMultiprocessorWithFlags) times the number of + * multiprocessors as specified by the device attribute + * ::CU_DEVICE_ATTRIBUTE_MULTIPROCESSOR_COUNT. * * The kernel cannot make use of CUDA dynamic parallelism. * @@ -10786,15 +11358,15 @@ CUresult CUDAAPI cuLaunchKernel(CUfunction f, * Calling ::cuLaunchCooperativeKernel() sets persistent function state that is * the same as function state set through ::cuLaunchKernel API * - * When the kernel \p f is launched via ::cuLaunchCooperativeKernel(), the previous - * block shape, shared size and parameter info associated with \p f - * is overwritten. + * When the kernel \p f is launched via ::cuLaunchCooperativeKernel(), the + * previous block shape, shared size and parameter info associated with \p f is + * overwritten. * - * Note that to use ::cuLaunchCooperativeKernel(), the kernel \p f must either have - * been compiled with toolchain version 3.2 or later so that it will + * Note that to use ::cuLaunchCooperativeKernel(), the kernel \p f must either + * have been compiled with toolchain version 3.2 or later so that it will * contain kernel parameter information, or have no kernel parameters. - * If either of these conditions is not met, then ::cuLaunchCooperativeKernel() will - * return ::CUDA_ERROR_INVALID_IMAGE. + * If either of these conditions is not met, then ::cuLaunchCooperativeKernel() + * will return ::CUDA_ERROR_INVALID_IMAGE. * * \param f - Kernel to launch * \param gridDimX - Width of grid in blocks @@ -10831,49 +11403,64 @@ CUresult CUDAAPI cuLaunchKernel(CUfunction f, * ::cuLaunchCooperativeKernelMultiDevice, * ::cudaLaunchCooperativeKernel */ -CUresult CUDAAPI cuLaunchCooperativeKernel(CUfunction f, - unsigned int gridDimX, - unsigned int gridDimY, - unsigned int gridDimZ, - unsigned int blockDimX, - unsigned int blockDimY, - unsigned int blockDimZ, - unsigned int sharedMemBytes, - CUstream hStream, - void **kernelParams); +CUresult CUDAAPI cuLaunchCooperativeKernel( + CUfunction f, unsigned int gridDimX, unsigned int gridDimY, + unsigned int gridDimZ, unsigned int blockDimX, unsigned int blockDimY, + unsigned int blockDimZ, unsigned int sharedMemBytes, CUstream hStream, + void **kernelParams); /** - * \brief Launches CUDA functions on multiple devices where thread blocks can cooperate and synchronize as they execute + * \brief Launches CUDA functions on multiple devices where thread blocks can + cooperate and synchronize as they execute * - * Invokes kernels as specified in the \p launchParamsList array where each element - * of the array specifies all the parameters required to perform a single kernel launch. - * These kernels can cooperate and synchronize as they execute. The size of the array is + * Invokes kernels as specified in the \p launchParamsList array where each + element + * of the array specifies all the parameters required to perform a single kernel + launch. + * These kernels can cooperate and synchronize as they execute. The size of the + array is * specified by \p numDevices. * - * No two kernels can be launched on the same device. All the devices targeted by this - * multi-device launch must be identical. All devices must have a non-zero value for the + * No two kernels can be launched on the same device. All the devices targeted + by this + * multi-device launch must be identical. All devices must have a non-zero value + for the * device attribute ::CU_DEVICE_ATTRIBUTE_COOPERATIVE_MULTI_DEVICE_LAUNCH. * - * All kernels launched must be identical with respect to the compiled code. Note that - * any __device__, __constant__ or __managed__ variables present in the module that owns - * the kernel launched on each device, are independently instantiated on every device. - * It is the application's responsiblity to ensure these variables are initialized and + * All kernels launched must be identical with respect to the compiled code. + Note that + * any __device__, __constant__ or __managed__ variables present in the module + that owns + * the kernel launched on each device, are independently instantiated on every + device. + * It is the application's responsiblity to ensure these variables are + initialized and * used appropriately. * - * The size of the grids as specified in blocks, the size of the blocks themselves - * and the amount of shared memory used by each thread block must also match across + * The size of the grids as specified in blocks, the size of the blocks + themselves + * and the amount of shared memory used by each thread block must also match + across * all launched kernels. * - * The streams used to launch these kernels must have been created via either ::cuStreamCreate - * or ::cuStreamCreateWithPriority. The NULL stream or ::CU_STREAM_LEGACY or ::CU_STREAM_PER_THREAD + * The streams used to launch these kernels must have been created via either + ::cuStreamCreate + * or ::cuStreamCreateWithPriority. The NULL stream or ::CU_STREAM_LEGACY or + ::CU_STREAM_PER_THREAD * cannot be used. * - * The total number of blocks launched per kernel cannot exceed the maximum number of blocks - * per multiprocessor as returned by ::cuOccupancyMaxActiveBlocksPerMultiprocessor (or - * ::cuOccupancyMaxActiveBlocksPerMultiprocessorWithFlags) times the number of multiprocessors - * as specified by the device attribute ::CU_DEVICE_ATTRIBUTE_MULTIPROCESSOR_COUNT. Since the - * total number of blocks launched per device has to match across all devices, the maximum - * number of blocks that can be launched per device will be limited by the device with the + * The total number of blocks launched per kernel cannot exceed the maximum + number of blocks + * per multiprocessor as returned by + ::cuOccupancyMaxActiveBlocksPerMultiprocessor (or + * ::cuOccupancyMaxActiveBlocksPerMultiprocessorWithFlags) times the number of + multiprocessors + * as specified by the device attribute + ::CU_DEVICE_ATTRIBUTE_MULTIPROCESSOR_COUNT. Since the + * total number of blocks launched per device has to match across all devices, + the maximum + * number of blocks that can be launched per device will be limited by the + device with the * least number of multiprocessors. * * The kernels cannot make use of CUDA dynamic parallelism. @@ -10895,56 +11482,87 @@ CUresult CUDAAPI cuLaunchCooperativeKernel(CUfunction f, } CUDA_LAUNCH_PARAMS; * \endcode * where: - * - ::CUDA_LAUNCH_PARAMS::function specifies the kernel to be launched. All functions must + * - ::CUDA_LAUNCH_PARAMS::function specifies the kernel to be launched. All + functions must * be identical with respect to the compiled code. - * - ::CUDA_LAUNCH_PARAMS::gridDimX is the width of the grid in blocks. This must match across + * - ::CUDA_LAUNCH_PARAMS::gridDimX is the width of the grid in blocks. This + must match across * all kernels launched. - * - ::CUDA_LAUNCH_PARAMS::gridDimY is the height of the grid in blocks. This must match across + * - ::CUDA_LAUNCH_PARAMS::gridDimY is the height of the grid in blocks. This + must match across * all kernels launched. - * - ::CUDA_LAUNCH_PARAMS::gridDimZ is the depth of the grid in blocks. This must match across + * - ::CUDA_LAUNCH_PARAMS::gridDimZ is the depth of the grid in blocks. This + must match across * all kernels launched. - * - ::CUDA_LAUNCH_PARAMS::blockDimX is the X dimension of each thread block. This must match across + * - ::CUDA_LAUNCH_PARAMS::blockDimX is the X dimension of each thread block. + This must match across * all kernels launched. - * - ::CUDA_LAUNCH_PARAMS::blockDimX is the Y dimension of each thread block. This must match across + * - ::CUDA_LAUNCH_PARAMS::blockDimX is the Y dimension of each thread block. + This must match across * all kernels launched. - * - ::CUDA_LAUNCH_PARAMS::blockDimZ is the Z dimension of each thread block. This must match across + * - ::CUDA_LAUNCH_PARAMS::blockDimZ is the Z dimension of each thread block. + This must match across * all kernels launched. - * - ::CUDA_LAUNCH_PARAMS::sharedMemBytes is the dynamic shared-memory size per thread block in bytes. + * - ::CUDA_LAUNCH_PARAMS::sharedMemBytes is the dynamic shared-memory size per + thread block in bytes. * This must match across all kernels launched. - * - ::CUDA_LAUNCH_PARAMS::hStream is the handle to the stream to perform the launch in. This cannot - * be the NULL stream or ::CU_STREAM_LEGACY or ::CU_STREAM_PER_THREAD. The CUDA context associated - * with this stream must match that associated with ::CUDA_LAUNCH_PARAMS::function. - * - ::CUDA_LAUNCH_PARAMS::kernelParams is an array of pointers to kernel parameters. If - * ::CUDA_LAUNCH_PARAMS::function has N parameters, then ::CUDA_LAUNCH_PARAMS::kernelParams - * needs to be an array of N pointers. Each of ::CUDA_LAUNCH_PARAMS::kernelParams[0] through - * ::CUDA_LAUNCH_PARAMS::kernelParams[N-1] must point to a region of memory from which the actual - * kernel parameter will be copied. The number of kernel parameters and their offsets and sizes - * do not need to be specified as that information is retrieved directly from the kernel's image. - * - * By default, the kernel won't begin execution on any GPU until all prior work in all the specified + * - ::CUDA_LAUNCH_PARAMS::hStream is the handle to the stream to perform the + launch in. This cannot + * be the NULL stream or ::CU_STREAM_LEGACY or ::CU_STREAM_PER_THREAD. The + CUDA context associated + * with this stream must match that associated with + ::CUDA_LAUNCH_PARAMS::function. + * - ::CUDA_LAUNCH_PARAMS::kernelParams is an array of pointers to kernel + parameters. If + * ::CUDA_LAUNCH_PARAMS::function has N parameters, then + ::CUDA_LAUNCH_PARAMS::kernelParams + * needs to be an array of N pointers. Each of + ::CUDA_LAUNCH_PARAMS::kernelParams[0] through + * ::CUDA_LAUNCH_PARAMS::kernelParams[N-1] must point to a region of memory + from which the actual + * kernel parameter will be copied. The number of kernel parameters and their + offsets and sizes + * do not need to be specified as that information is retrieved directly from + the kernel's image. + * + * By default, the kernel won't begin execution on any GPU until all prior work + in all the specified * streams has completed. This behavior can be overridden by specifying the flag - * ::CUDA_COOPERATIVE_LAUNCH_MULTI_DEVICE_NO_PRE_LAUNCH_SYNC. When this flag is specified, each kernel - * will only wait for prior work in the stream corresponding to that GPU to complete before it begins + * ::CUDA_COOPERATIVE_LAUNCH_MULTI_DEVICE_NO_PRE_LAUNCH_SYNC. When this flag is + specified, each kernel + * will only wait for prior work in the stream corresponding to that GPU to + complete before it begins * execution. * - * Similarly, by default, any subsequent work pushed in any of the specified streams will not begin - * execution until the kernels on all GPUs have completed. This behavior can be overridden by specifying - * the flag ::CUDA_COOPERATIVE_LAUNCH_MULTI_DEVICE_NO_POST_LAUNCH_SYNC. When this flag is specified, - * any subsequent work pushed in any of the specified streams will only wait for the kernel launched - * on the GPU corresponding to that stream to complete before it begins execution. - * - * Calling ::cuLaunchCooperativeKernelMultiDevice() sets persistent function state that is - * the same as function state set through ::cuLaunchKernel API when called individually for each + * Similarly, by default, any subsequent work pushed in any of the specified + streams will not begin + * execution until the kernels on all GPUs have completed. This behavior can be + overridden by specifying + * the flag ::CUDA_COOPERATIVE_LAUNCH_MULTI_DEVICE_NO_POST_LAUNCH_SYNC. When + this flag is specified, + * any subsequent work pushed in any of the specified streams will only wait for + the kernel launched + * on the GPU corresponding to that stream to complete before it begins + execution. + * + * Calling ::cuLaunchCooperativeKernelMultiDevice() sets persistent function + state that is + * the same as function state set through ::cuLaunchKernel API when called + individually for each * element in \p launchParamsList. * - * When kernels are launched via ::cuLaunchCooperativeKernelMultiDevice(), the previous - * block shape, shared size and parameter info associated with each ::CUDA_LAUNCH_PARAMS::function + * When kernels are launched via ::cuLaunchCooperativeKernelMultiDevice(), the + previous + * block shape, shared size and parameter info associated with each + ::CUDA_LAUNCH_PARAMS::function * in \p launchParamsList is overwritten. * - * Note that to use ::cuLaunchCooperativeKernelMultiDevice(), the kernels must either have + * Note that to use ::cuLaunchCooperativeKernelMultiDevice(), the kernels must + either have * been compiled with toolchain version 3.2 or later so that it will * contain kernel parameter information, or have no kernel parameters. - * If either of these conditions is not met, then ::cuLaunchCooperativeKernelMultiDevice() will + * If either of these conditions is not met, then + ::cuLaunchCooperativeKernelMultiDevice() will * return ::CUDA_ERROR_INVALID_IMAGE. * * \param launchParamsList - List of launch parameters, one per device @@ -10975,7 +11593,9 @@ CUresult CUDAAPI cuLaunchCooperativeKernel(CUfunction f, * ::cuLaunchCooperativeKernel, * ::cudaLaunchCooperativeKernelMultiDevice */ -CUresult CUDAAPI cuLaunchCooperativeKernelMultiDevice(CUDA_LAUNCH_PARAMS *launchParamsList, unsigned int numDevices, unsigned int flags); +CUresult CUDAAPI cuLaunchCooperativeKernelMultiDevice( + CUDA_LAUNCH_PARAMS *launchParamsList, unsigned int numDevices, + unsigned int flags); #endif /* __CUDA_API_VERSION >= 9000 */ @@ -11022,8 +11642,8 @@ CUresult CUDAAPI cuLaunchCooperativeKernelMultiDevice(CUDA_LAUNCH_PARAMS *launch * called in the event of an error in the CUDA context. * * \param hStream - Stream to enqueue function call in - * \param fn - The function to call once preceding stream operations are complete - * \param userData - User-specified data to be passed to the function + * \param fn - The function to call once preceding stream operations are + * complete \param userData - User-specified data to be passed to the function * * \return * ::CUDA_SUCCESS, @@ -11044,7 +11664,8 @@ CUresult CUDAAPI cuLaunchCooperativeKernelMultiDevice(CUDA_LAUNCH_PARAMS *launch * ::cuStreamAttachMemAsync, * ::cuStreamAddCallback */ -CUresult CUDAAPI cuLaunchHostFunc(CUstream hStream, CUhostFn fn, void *userData); +CUresult CUDAAPI cuLaunchHostFunc(CUstream hStream, CUhostFn fn, + void *userData); #endif /* __CUDA_API_VERSION >= 10000 */ @@ -11096,7 +11717,8 @@ CUresult CUDAAPI cuLaunchHostFunc(CUstream hStream, CUhostFn fn, void *userData) * ::cuLaunchGridAsync, * ::cuLaunchKernel */ -__CUDA_DEPRECATED CUresult CUDAAPI cuFuncSetBlockShape(CUfunction hfunc, int x, int y, int z); +__CUDA_DEPRECATED CUresult CUDAAPI cuFuncSetBlockShape(CUfunction hfunc, int x, + int y, int z); /** * \brief Sets the dynamic shared-memory size for the function @@ -11130,7 +11752,8 @@ __CUDA_DEPRECATED CUresult CUDAAPI cuFuncSetBlockShape(CUfunction hfunc, int x, * ::cuLaunchGridAsync, * ::cuLaunchKernel */ -__CUDA_DEPRECATED CUresult CUDAAPI cuFuncSetSharedSize(CUfunction hfunc, unsigned int bytes); +__CUDA_DEPRECATED CUresult CUDAAPI cuFuncSetSharedSize(CUfunction hfunc, + unsigned int bytes); /** * \brief Sets the parameter size for the function @@ -11162,7 +11785,8 @@ __CUDA_DEPRECATED CUresult CUDAAPI cuFuncSetSharedSize(CUfunction hfunc, unsigne * ::cuLaunchGridAsync, * ::cuLaunchKernel */ -__CUDA_DEPRECATED CUresult CUDAAPI cuParamSetSize(CUfunction hfunc, unsigned int numbytes); +__CUDA_DEPRECATED CUresult CUDAAPI cuParamSetSize(CUfunction hfunc, + unsigned int numbytes); /** * \brief Adds an integer parameter to the function's argument list @@ -11195,7 +11819,8 @@ __CUDA_DEPRECATED CUresult CUDAAPI cuParamSetSize(CUfunction hfunc, unsigned int * ::cuLaunchGridAsync, * ::cuLaunchKernel */ -__CUDA_DEPRECATED CUresult CUDAAPI cuParamSeti(CUfunction hfunc, int offset, unsigned int value); +__CUDA_DEPRECATED CUresult CUDAAPI cuParamSeti(CUfunction hfunc, int offset, + unsigned int value); /** * \brief Adds a floating-point parameter to the function's argument list @@ -11228,7 +11853,8 @@ __CUDA_DEPRECATED CUresult CUDAAPI cuParamSeti(CUfunction hfunc, int offset, uns * ::cuLaunchGridAsync, * ::cuLaunchKernel */ -__CUDA_DEPRECATED CUresult CUDAAPI cuParamSetf(CUfunction hfunc, int offset, float value); +__CUDA_DEPRECATED CUresult CUDAAPI cuParamSetf(CUfunction hfunc, int offset, + float value); /** * \brief Adds arbitrary data to the function's argument list @@ -11263,7 +11889,9 @@ __CUDA_DEPRECATED CUresult CUDAAPI cuParamSetf(CUfunction hfunc, int offset, flo * ::cuLaunchGridAsync, * ::cuLaunchKernel */ -__CUDA_DEPRECATED CUresult CUDAAPI cuParamSetv(CUfunction hfunc, int offset, void *ptr, unsigned int numbytes); +__CUDA_DEPRECATED CUresult CUDAAPI cuParamSetv(CUfunction hfunc, int offset, + void *ptr, + unsigned int numbytes); /** * \brief Launches a CUDA function @@ -11339,7 +11967,8 @@ __CUDA_DEPRECATED CUresult CUDAAPI cuLaunch(CUfunction f); * ::cuLaunchGridAsync, * ::cuLaunchKernel */ -__CUDA_DEPRECATED CUresult CUDAAPI cuLaunchGrid(CUfunction f, int grid_width, int grid_height); +__CUDA_DEPRECATED CUresult CUDAAPI cuLaunchGrid(CUfunction f, int grid_width, + int grid_height); /** * \brief Launches a CUDA function @@ -11368,9 +11997,10 @@ __CUDA_DEPRECATED CUresult CUDAAPI cuLaunchGrid(CUfunction f, int grid_width, in * ::CUDA_ERROR_LAUNCH_INCOMPATIBLE_TEXTURING, * ::CUDA_ERROR_SHARED_OBJECT_INIT_FAILED * - * \note In certain cases where cubins are created with no ABI (i.e., using \p ptxas \p --abi-compile \p no), - * this function may serialize kernel launches. In order to force the CUDA driver to retain - * asynchronous behavior, set the ::CU_CTX_LMEM_RESIZE_TO_MAX flag during context creation (see ::cuCtxCreate). + * \note In certain cases where cubins are created with no ABI (i.e., using \p + * ptxas \p --abi-compile \p no), this function may serialize kernel launches. + * In order to force the CUDA driver to retain asynchronous behavior, set the + * ::CU_CTX_LMEM_RESIZE_TO_MAX flag during context creation (see ::cuCtxCreate). * * \note_null_stream * \notefnerr @@ -11386,8 +12016,10 @@ __CUDA_DEPRECATED CUresult CUDAAPI cuLaunchGrid(CUfunction f, int grid_width, in * ::cuLaunchGrid, * ::cuLaunchKernel */ -__CUDA_DEPRECATED CUresult CUDAAPI cuLaunchGridAsync(CUfunction f, int grid_width, int grid_height, CUstream hStream); - +__CUDA_DEPRECATED CUresult CUDAAPI cuLaunchGridAsync(CUfunction f, + int grid_width, + int grid_height, + CUstream hStream); /** * \brief Adds a texture-reference to the function's argument list @@ -11411,7 +12043,9 @@ __CUDA_DEPRECATED CUresult CUDAAPI cuLaunchGridAsync(CUfunction f, int grid_widt * ::CUDA_ERROR_INVALID_VALUE * \notefnerr */ -__CUDA_DEPRECATED CUresult CUDAAPI cuParamSetTexRef(CUfunction hfunc, int texunit, CUtexref hTexRef); +__CUDA_DEPRECATED CUresult CUDAAPI cuParamSetTexRef(CUfunction hfunc, + int texunit, + CUtexref hTexRef); /** @} */ /* END CUDA_EXEC_DEPRECATED */ #if __CUDA_API_VERSION >= 10000 @@ -11463,11 +12097,12 @@ CUresult CUDAAPI cuGraphCreate(CUgraph *phGraph, unsigned int flags); /** * \brief Creates a kernel execution node and adds it to a graph * - * Creates a new kernel execution node and adds it to \p hGraph with \p numDependencies - * dependencies specified via \p dependencies and arguments specified in \p nodeParams. - * It is possible for \p numDependencies to be 0, in which case the node will be placed - * at the root of the graph. \p dependencies may not have any duplicate entries. - * A handle to the new node will be returned in \p phGraphNode. + * Creates a new kernel execution node and adds it to \p hGraph with \p + * numDependencies dependencies specified via \p dependencies and arguments + * specified in \p nodeParams. It is possible for \p numDependencies to be 0, in + * which case the node will be placed at the root of the graph. \p dependencies + * may not have any duplicate entries. A handle to the new node will be returned + * in \p phGraphNode. * * The CUDA_KERNEL_NODE_PARAMS structure is defined as: * @@ -11486,8 +12121,8 @@ CUresult CUDAAPI cuGraphCreate(CUgraph *phGraph, unsigned int flags); * } CUDA_KERNEL_NODE_PARAMS; * \endcode * - * When the graph is launched, the node will invoke kernel \p func on a (\p gridDimX x - * \p gridDimY x \p gridDimZ) grid of blocks. Each block contains + * When the graph is launched, the node will invoke kernel \p func on a (\p + * gridDimX x \p gridDimY x \p gridDimZ) grid of blocks. Each block contains * (\p blockDimX x \p blockDimY x \p blockDimZ) threads. * * \p sharedMemBytes sets the amount of dynamic shared memory that will be @@ -11495,19 +12130,21 @@ CUresult CUDAAPI cuGraphCreate(CUgraph *phGraph, unsigned int flags); * * Kernel parameters to \p func can be specified in one of two ways: * - * 1) Kernel parameters can be specified via \p kernelParams. If the kernel has N - * parameters, then \p kernelParams needs to be an array of N pointers. Each pointer, - * from \p kernelParams[0] to \p kernelParams[N-1], points to the region of memory from which the actual - * parameter will be copied. The number of kernel parameters and their offsets and sizes do not need - * to be specified as that information is retrieved directly from the kernel's image. - * - * 2) Kernel parameters can also be packaged by the application into a single buffer that is passed in - * via \p extra. This places the burden on the application of knowing each kernel - * parameter's size and alignment/padding within the buffer. The \p extra parameter exists - * to allow this function to take additional less commonly used arguments. \p extra specifies - * a list of names of extra settings and their corresponding values. Each extra setting name is - * immediately followed by the corresponding value. The list must be terminated with either NULL or - * CU_LAUNCH_PARAM_END. + * 1) Kernel parameters can be specified via \p kernelParams. If the kernel has + * N parameters, then \p kernelParams needs to be an array of N pointers. Each + * pointer, from \p kernelParams[0] to \p kernelParams[N-1], points to the + * region of memory from which the actual parameter will be copied. The number + * of kernel parameters and their offsets and sizes do not need to be specified + * as that information is retrieved directly from the kernel's image. + * + * 2) Kernel parameters can also be packaged by the application into a single + * buffer that is passed in via \p extra. This places the burden on the + * application of knowing each kernel parameter's size and alignment/padding + * within the buffer. The \p extra parameter exists to allow this function to + * take additional less commonly used arguments. \p extra specifies a list of + * names of extra settings and their corresponding values. Each extra setting + * name is immediately followed by the corresponding value. The list must be + * terminated with either NULL or CU_LAUNCH_PARAM_END. * * - ::CU_LAUNCH_PARAM_END, which indicates the end of the \p extra * array; @@ -11520,16 +12157,17 @@ CUresult CUDAAPI cuGraphCreate(CUgraph *phGraph, unsigned int flags); * containing the size of the buffer specified with * ::CU_LAUNCH_PARAM_BUFFER_POINTER; * - * The error ::CUDA_ERROR_INVALID_VALUE will be returned if kernel parameters are specified with both - * \p kernelParams and \p extra (i.e. both \p kernelParams and - * \p extra are non-NULL). + * The error ::CUDA_ERROR_INVALID_VALUE will be returned if kernel parameters + * are specified with both \p kernelParams and \p extra (i.e. both \p + * kernelParams and \p extra are non-NULL). * - * The \p kernelParams or \p extra array, as well as the argument values it points to, - * are copied during this call. + * The \p kernelParams or \p extra array, as well as the argument values it + * points to, are copied during this call. * - * \note Kernels launched using graphs must not use texture and surface references. Reading or - * writing through any texture or surface reference is undefined behavior. - * This restriction does not apply to texture and surface objects. + * \note Kernels launched using graphs must not use texture and surface + * references. Reading or writing through any texture or surface reference is + * undefined behavior. This restriction does not apply to texture and surface + * objects. * * \param phGraphNode - Returns newly created node * \param hGraph - Graph to which to add the node @@ -11557,7 +12195,9 @@ CUresult CUDAAPI cuGraphCreate(CUgraph *phGraph, unsigned int flags); * ::cuGraphAddMemcpyNode, * ::cuGraphAddMemsetNode */ -CUresult CUDAAPI cuGraphAddKernelNode(CUgraphNode *phGraphNode, CUgraph hGraph, const CUgraphNode *dependencies, size_t numDependencies, const CUDA_KERNEL_NODE_PARAMS *nodeParams); +CUresult CUDAAPI cuGraphAddKernelNode( + CUgraphNode *phGraphNode, CUgraph hGraph, const CUgraphNode *dependencies, + size_t numDependencies, const CUDA_KERNEL_NODE_PARAMS *nodeParams); /** * \brief Returns a kernel node's parameters @@ -11589,7 +12229,8 @@ CUresult CUDAAPI cuGraphAddKernelNode(CUgraphNode *phGraphNode, CUgraph hGraph, * ::cuGraphAddKernelNode, * ::cuGraphKernelNodeSetParams */ -CUresult CUDAAPI cuGraphKernelNodeGetParams(CUgraphNode hNode, CUDA_KERNEL_NODE_PARAMS *nodeParams); +CUresult CUDAAPI cuGraphKernelNodeGetParams( + CUgraphNode hNode, CUDA_KERNEL_NODE_PARAMS *nodeParams); /** * \brief Sets a kernel node's parameters @@ -11612,26 +12253,30 @@ CUresult CUDAAPI cuGraphKernelNodeGetParams(CUgraphNode hNode, CUDA_KERNEL_NODE_ * ::cuGraphAddKernelNode, * ::cuGraphKernelNodeGetParams */ -CUresult CUDAAPI cuGraphKernelNodeSetParams(CUgraphNode hNode, const CUDA_KERNEL_NODE_PARAMS *nodeParams); +CUresult CUDAAPI cuGraphKernelNodeSetParams( + CUgraphNode hNode, const CUDA_KERNEL_NODE_PARAMS *nodeParams); /** * \brief Creates a memcpy node and adds it to a graph * * Creates a new memcpy node and adds it to \p hGraph with \p numDependencies * dependencies specified via \p dependencies. - * It is possible for \p numDependencies to be 0, in which case the node will be placed - * at the root of the graph. \p dependencies may not have any duplicate entries. - * A handle to the new node will be returned in \p phGraphNode. - * - * When the graph is launched, the node will perform the memcpy described by \p copyParams. - * See ::cuMemcpy3D() for a description of the structure and its restrictions. - * - * Memcpy nodes have some additional restrictions with regards to managed memory, if the - * system contains at least one device which has a zero value for the device attribute - * ::CU_DEVICE_ATTRIBUTE_CONCURRENT_MANAGED_ACCESS. If one or more of the operands refer - * to managed memory, then using the memory type ::CU_MEMORYTYPE_UNIFIED is disallowed - * for those operand(s). The managed memory will be treated as residing on either the - * host or the device, depending on which memory type is specified. + * It is possible for \p numDependencies to be 0, in which case the node will be + * placed at the root of the graph. \p dependencies may not have any duplicate + * entries. A handle to the new node will be returned in \p phGraphNode. + * + * When the graph is launched, the node will perform the memcpy described by \p + * copyParams. See ::cuMemcpy3D() for a description of the structure and its + * restrictions. + * + * Memcpy nodes have some additional restrictions with regards to managed + * memory, if the system contains at least one device which has a zero value for + * the device attribute + * ::CU_DEVICE_ATTRIBUTE_CONCURRENT_MANAGED_ACCESS. If one or more of the + * operands refer to managed memory, then using the memory type + * ::CU_MEMORYTYPE_UNIFIED is disallowed for those operand(s). The managed + * memory will be treated as residing on either the host or the device, + * depending on which memory type is specified. * * \param phGraphNode - Returns newly created node * \param hGraph - Graph to which to add the node @@ -11660,7 +12305,11 @@ CUresult CUDAAPI cuGraphKernelNodeSetParams(CUgraphNode hNode, const CUDA_KERNEL * ::cuGraphAddHostNode, * ::cuGraphAddMemsetNode */ -CUresult CUDAAPI cuGraphAddMemcpyNode(CUgraphNode *phGraphNode, CUgraph hGraph, const CUgraphNode *dependencies, size_t numDependencies, const CUDA_MEMCPY3D *copyParams, CUcontext ctx); +CUresult CUDAAPI cuGraphAddMemcpyNode(CUgraphNode *phGraphNode, CUgraph hGraph, + const CUgraphNode *dependencies, + size_t numDependencies, + const CUDA_MEMCPY3D *copyParams, + CUcontext ctx); /** * \brief Returns a memcpy node's parameters @@ -11683,7 +12332,8 @@ CUresult CUDAAPI cuGraphAddMemcpyNode(CUgraphNode *phGraphNode, CUgraph hGraph, * ::cuGraphAddMemcpyNode, * ::cuGraphMemcpyNodeSetParams */ -CUresult CUDAAPI cuGraphMemcpyNodeGetParams(CUgraphNode hNode, CUDA_MEMCPY3D *nodeParams); +CUresult CUDAAPI cuGraphMemcpyNodeGetParams(CUgraphNode hNode, + CUDA_MEMCPY3D *nodeParams); /** * \brief Sets a memcpy node's parameters @@ -11706,19 +12356,21 @@ CUresult CUDAAPI cuGraphMemcpyNodeGetParams(CUgraphNode hNode, CUDA_MEMCPY3D *no * ::cuGraphAddMemcpyNode, * ::cuGraphMemcpyNodeGetParams */ -CUresult CUDAAPI cuGraphMemcpyNodeSetParams(CUgraphNode hNode, const CUDA_MEMCPY3D *nodeParams); +CUresult CUDAAPI cuGraphMemcpyNodeSetParams(CUgraphNode hNode, + const CUDA_MEMCPY3D *nodeParams); /** * \brief Creates a memset node and adds it to a graph * * Creates a new memset node and adds it to \p hGraph with \p numDependencies * dependencies specified via \p dependencies. - * It is possible for \p numDependencies to be 0, in which case the node will be placed - * at the root of the graph. \p dependencies may not have any duplicate entries. - * A handle to the new node will be returned in \p phGraphNode. + * It is possible for \p numDependencies to be 0, in which case the node will be + * placed at the root of the graph. \p dependencies may not have any duplicate + * entries. A handle to the new node will be returned in \p phGraphNode. * * The element size must be 1, 2, or 4 bytes. - * When the graph is launched, the node will perform the memset described by \p memsetParams. + * When the graph is launched, the node will perform the memset described by \p + * memsetParams. * * \param phGraphNode - Returns newly created node * \param hGraph - Graph to which to add the node @@ -11748,7 +12400,10 @@ CUresult CUDAAPI cuGraphMemcpyNodeSetParams(CUgraphNode hNode, const CUDA_MEMCPY * ::cuGraphAddHostNode, * ::cuGraphAddMemcpyNode */ -CUresult CUDAAPI cuGraphAddMemsetNode(CUgraphNode *phGraphNode, CUgraph hGraph, const CUgraphNode *dependencies, size_t numDependencies, const CUDA_MEMSET_NODE_PARAMS *memsetParams, CUcontext ctx); +CUresult CUDAAPI cuGraphAddMemsetNode( + CUgraphNode *phGraphNode, CUgraph hGraph, const CUgraphNode *dependencies, + size_t numDependencies, const CUDA_MEMSET_NODE_PARAMS *memsetParams, + CUcontext ctx); /** * \brief Returns a memset node's parameters @@ -11771,7 +12426,8 @@ CUresult CUDAAPI cuGraphAddMemsetNode(CUgraphNode *phGraphNode, CUgraph hGraph, * ::cuGraphAddMemsetNode, * ::cuGraphMemsetNodeSetParams */ -CUresult CUDAAPI cuGraphMemsetNodeGetParams(CUgraphNode hNode, CUDA_MEMSET_NODE_PARAMS *nodeParams); +CUresult CUDAAPI cuGraphMemsetNodeGetParams( + CUgraphNode hNode, CUDA_MEMSET_NODE_PARAMS *nodeParams); /** * \brief Sets a memset node's parameters @@ -11794,16 +12450,18 @@ CUresult CUDAAPI cuGraphMemsetNodeGetParams(CUgraphNode hNode, CUDA_MEMSET_NODE_ * ::cuGraphAddMemsetNode, * ::cuGraphMemsetNodeGetParams */ -CUresult CUDAAPI cuGraphMemsetNodeSetParams(CUgraphNode hNode, const CUDA_MEMSET_NODE_PARAMS *nodeParams); +CUresult CUDAAPI cuGraphMemsetNodeSetParams( + CUgraphNode hNode, const CUDA_MEMSET_NODE_PARAMS *nodeParams); /** * \brief Creates a host execution node and adds it to a graph * - * Creates a new CPU execution node and adds it to \p hGraph with \p numDependencies - * dependencies specified via \p dependencies and arguments specified in \p nodeParams. - * It is possible for \p numDependencies to be 0, in which case the node will be placed - * at the root of the graph. \p dependencies may not have any duplicate entries. - * A handle to the new node will be returned in \p phGraphNode. + * Creates a new CPU execution node and adds it to \p hGraph with \p + * numDependencies dependencies specified via \p dependencies and arguments + * specified in \p nodeParams. It is possible for \p numDependencies to be 0, in + * which case the node will be placed at the root of the graph. \p dependencies + * may not have any duplicate entries. A handle to the new node will be returned + * in \p phGraphNode. * * When the graph is launched, the node will invoke the specified CPU function. * Host nodes are not supported under MPS with pre-Volta GPUs. @@ -11835,7 +12493,10 @@ CUresult CUDAAPI cuGraphMemsetNodeSetParams(CUgraphNode hNode, const CUDA_MEMSET * ::cuGraphAddMemcpyNode, * ::cuGraphAddMemsetNode */ -CUresult CUDAAPI cuGraphAddHostNode(CUgraphNode *phGraphNode, CUgraph hGraph, const CUgraphNode *dependencies, size_t numDependencies, const CUDA_HOST_NODE_PARAMS *nodeParams); +CUresult CUDAAPI cuGraphAddHostNode(CUgraphNode *phGraphNode, CUgraph hGraph, + const CUgraphNode *dependencies, + size_t numDependencies, + const CUDA_HOST_NODE_PARAMS *nodeParams); /** * \brief Returns a host node's parameters @@ -11858,7 +12519,8 @@ CUresult CUDAAPI cuGraphAddHostNode(CUgraphNode *phGraphNode, CUgraph hGraph, co * ::cuGraphAddHostNode, * ::cuGraphHostNodeSetParams */ -CUresult CUDAAPI cuGraphHostNodeGetParams(CUgraphNode hNode, CUDA_HOST_NODE_PARAMS *nodeParams); +CUresult CUDAAPI cuGraphHostNodeGetParams(CUgraphNode hNode, + CUDA_HOST_NODE_PARAMS *nodeParams); /** * \brief Sets a host node's parameters @@ -11881,18 +12543,20 @@ CUresult CUDAAPI cuGraphHostNodeGetParams(CUgraphNode hNode, CUDA_HOST_NODE_PARA * ::cuGraphAddHostNode, * ::cuGraphHostNodeGetParams */ -CUresult CUDAAPI cuGraphHostNodeSetParams(CUgraphNode hNode, const CUDA_HOST_NODE_PARAMS *nodeParams); +CUresult CUDAAPI cuGraphHostNodeSetParams( + CUgraphNode hNode, const CUDA_HOST_NODE_PARAMS *nodeParams); /** * \brief Creates a child graph node and adds it to a graph * - * Creates a new node which executes an embedded graph, and adds it to \p hGraph with - * \p numDependencies dependencies specified via \p dependencies. - * It is possible for \p numDependencies to be 0, in which case the node will be placed - * at the root of the graph. \p dependencies may not have any duplicate entries. - * A handle to the new node will be returned in \p phGraphNode. + * Creates a new node which executes an embedded graph, and adds it to \p hGraph + * with \p numDependencies dependencies specified via \p dependencies. It is + * possible for \p numDependencies to be 0, in which case the node will be + * placed at the root of the graph. \p dependencies may not have any duplicate + * entries. A handle to the new node will be returned in \p phGraphNode. * - * The node executes an embedded child graph. The child graph is cloned in this call. + * The node executes an embedded child graph. The child graph is cloned in this + * call. * * \param phGraphNode - Returns newly created node * \param hGraph - Graph to which to add the node @@ -11919,7 +12583,11 @@ CUresult CUDAAPI cuGraphHostNodeSetParams(CUgraphNode hNode, const CUDA_HOST_NOD * ::cuGraphAddMemsetNode, * ::cuGraphClone */ -CUresult CUDAAPI cuGraphAddChildGraphNode(CUgraphNode *phGraphNode, CUgraph hGraph, const CUgraphNode *dependencies, size_t numDependencies, CUgraph childGraph); +CUresult CUDAAPI cuGraphAddChildGraphNode(CUgraphNode *phGraphNode, + CUgraph hGraph, + const CUgraphNode *dependencies, + size_t numDependencies, + CUgraph childGraph); /** * \brief Gets a handle to the embedded graph of a child graph node @@ -11943,16 +12611,17 @@ CUresult CUDAAPI cuGraphAddChildGraphNode(CUgraphNode *phGraphNode, CUgraph hGra * ::cuGraphAddChildGraphNode, * ::cuGraphNodeFindInClone */ -CUresult CUDAAPI cuGraphChildGraphNodeGetGraph(CUgraphNode hNode, CUgraph *phGraph); +CUresult CUDAAPI cuGraphChildGraphNodeGetGraph(CUgraphNode hNode, + CUgraph *phGraph); /** * \brief Creates an empty node and adds it to a graph * * Creates a new node which performs no operation, and adds it to \p hGraph with * \p numDependencies dependencies specified via \p dependencies. - * It is possible for \p numDependencies to be 0, in which case the node will be placed - * at the root of the graph. \p dependencies may not have any duplicate entries. - * A handle to the new node will be returned in \p phGraphNode. + * It is possible for \p numDependencies to be 0, in which case the node will be + * placed at the root of the graph. \p dependencies may not have any duplicate + * entries. A handle to the new node will be returned in \p phGraphNode. * * An empty node performs no operation during execution, but can be used for * transitive ordering. For example, a phased execution graph with 2 groups of n @@ -11981,16 +12650,19 @@ CUresult CUDAAPI cuGraphChildGraphNodeGetGraph(CUgraphNode hNode, CUgraph *phGra * ::cuGraphAddMemcpyNode, * ::cuGraphAddMemsetNode */ -CUresult CUDAAPI cuGraphAddEmptyNode(CUgraphNode *phGraphNode, CUgraph hGraph, const CUgraphNode *dependencies, size_t numDependencies); +CUresult CUDAAPI cuGraphAddEmptyNode(CUgraphNode *phGraphNode, CUgraph hGraph, + const CUgraphNode *dependencies, + size_t numDependencies); /** * \brief Clones a graph * - * This function creates a copy of \p originalGraph and returns it in \p * phGraphClone. - * All parameters are copied into the cloned graph. The original graph may be modified - * after this call without affecting the clone. + * This function creates a copy of \p originalGraph and returns it in \p * + * phGraphClone. All parameters are copied into the cloned graph. The original + * graph may be modified after this call without affecting the clone. * - * Child graph nodes in the original graph are recursively copied into the clone. + * Child graph nodes in the original graph are recursively copied into the + * clone. * * \param phGraphClone - Returns newly created cloned graph * \param originalGraph - Graph to clone @@ -12011,13 +12683,14 @@ CUresult CUDAAPI cuGraphClone(CUgraph *phGraphClone, CUgraph originalGraph); /** * \brief Finds a cloned version of a node * - * This function returns the node in \p hClonedGraph corresponding to \p hOriginalNode - * in the original graph. + * This function returns the node in \p hClonedGraph corresponding to \p + * hOriginalNode in the original graph. * - * \p hClonedGraph must have been cloned from \p hOriginalGraph via ::cuGraphClone. - * \p hOriginalNode must have been in \p hOriginalGraph at the time of the call to - * ::cuGraphClone, and the corresponding cloned node in \p hClonedGraph must not have - * been removed. The cloned node is then returned via \p phClonedNode. + * \p hClonedGraph must have been cloned from \p hOriginalGraph via + * ::cuGraphClone. \p hOriginalNode must have been in \p hOriginalGraph at the + * time of the call to + * ::cuGraphClone, and the corresponding cloned node in \p hClonedGraph must not + * have been removed. The cloned node is then returned via \p phClonedNode. * * \param phNode - Returns handle to the cloned node * \param hOriginalNode - Handle to the original node @@ -12032,7 +12705,9 @@ CUresult CUDAAPI cuGraphClone(CUgraph *phGraphClone, CUgraph originalGraph); * \sa * ::cuGraphClone */ -CUresult CUDAAPI cuGraphNodeFindInClone(CUgraphNode *phNode, CUgraphNode hOriginalNode, CUgraph hClonedGraph); +CUresult CUDAAPI cuGraphNodeFindInClone(CUgraphNode *phNode, + CUgraphNode hOriginalNode, + CUgraph hClonedGraph); /** * \brief Returns a node's type @@ -12070,9 +12745,10 @@ CUresult CUDAAPI cuGraphNodeGetType(CUgraphNode hNode, CUgraphNodeType *type); * * Returns a list of \p hGraph's nodes. \p nodes may be NULL, in which case this * function will return the number of nodes in \p numNodes. Otherwise, - * \p numNodes entries will be filled in. If \p numNodes is higher than the actual - * number of nodes, the remaining entries in \p nodes will be set to NULL, and the - * number of nodes actually obtained will be returned in \p numNodes. + * \p numNodes entries will be filled in. If \p numNodes is higher than the + * actual number of nodes, the remaining entries in \p nodes will be set to + * NULL, and the number of nodes actually obtained will be returned in \p + * numNodes. * * \param hGraph - Graph to query * \param nodes - Pointer to return the nodes @@ -12094,16 +12770,18 @@ CUresult CUDAAPI cuGraphNodeGetType(CUgraphNode hNode, CUgraphNodeType *type); * ::cuGraphNodeGetDependencies, * ::cuGraphNodeGetDependentNodes */ -CUresult CUDAAPI cuGraphGetNodes(CUgraph hGraph, CUgraphNode *nodes, size_t *numNodes); +CUresult CUDAAPI cuGraphGetNodes(CUgraph hGraph, CUgraphNode *nodes, + size_t *numNodes); /** * \brief Returns a graph's root nodes * - * Returns a list of \p hGraph's root nodes. \p rootNodes may be NULL, in which case this - * function will return the number of root nodes in \p numRootNodes. Otherwise, - * \p numRootNodes entries will be filled in. If \p numRootNodes is higher than the actual - * number of root nodes, the remaining entries in \p rootNodes will be set to NULL, and the - * number of nodes actually obtained will be returned in \p numRootNodes. + * Returns a list of \p hGraph's root nodes. \p rootNodes may be NULL, in which + * case this function will return the number of root nodes in \p numRootNodes. + * Otherwise, \p numRootNodes entries will be filled in. If \p numRootNodes is + * higher than the actual number of root nodes, the remaining entries in \p + * rootNodes will be set to NULL, and the number of nodes actually obtained will + * be returned in \p numRootNodes. * * \param hGraph - Graph to query * \param rootNodes - Pointer to return the root nodes @@ -12125,18 +12803,20 @@ CUresult CUDAAPI cuGraphGetNodes(CUgraph hGraph, CUgraphNode *nodes, size_t *num * ::cuGraphNodeGetDependencies, * ::cuGraphNodeGetDependentNodes */ -CUresult CUDAAPI cuGraphGetRootNodes(CUgraph hGraph, CUgraphNode *rootNodes, size_t *numRootNodes); +CUresult CUDAAPI cuGraphGetRootNodes(CUgraph hGraph, CUgraphNode *rootNodes, + size_t *numRootNodes); /** * \brief Returns a graph's dependency edges * - * Returns a list of \p hGraph's dependency edges. Edges are returned via corresponding - * indices in \p from and \p to; that is, the node in \p to[i] has a dependency on the - * node in \p from[i]. \p from and \p to may both be NULL, in which - * case this function only returns the number of edges in \p numEdges. Otherwise, - * \p numEdges entries will be filled in. If \p numEdges is higher than the actual - * number of edges, the remaining entries in \p from and \p to will be set to NULL, and - * the number of edges actually returned will be written to \p numEdges. + * Returns a list of \p hGraph's dependency edges. Edges are returned via + * corresponding indices in \p from and \p to; that is, the node in \p to[i] has + * a dependency on the node in \p from[i]. \p from and \p to may both be NULL, + * in which case this function only returns the number of edges in \p numEdges. + * Otherwise, \p numEdges entries will be filled in. If \p numEdges is higher + * than the actual number of edges, the remaining entries in \p from and \p to + * will be set to NULL, and the number of edges actually returned will be + * written to \p numEdges. * * \param hGraph - Graph to get the edges from * \param from - Location to return edge endpoints @@ -12159,16 +12839,18 @@ CUresult CUDAAPI cuGraphGetRootNodes(CUgraph hGraph, CUgraphNode *rootNodes, siz * ::cuGraphNodeGetDependencies, * ::cuGraphNodeGetDependentNodes */ -CUresult CUDAAPI cuGraphGetEdges(CUgraph hGraph, CUgraphNode *from, CUgraphNode *to, size_t *numEdges); +CUresult CUDAAPI cuGraphGetEdges(CUgraph hGraph, CUgraphNode *from, + CUgraphNode *to, size_t *numEdges); /** * \brief Returns a node's dependencies * - * Returns a list of \p node's dependencies. \p dependencies may be NULL, in which case this - * function will return the number of dependencies in \p numDependencies. Otherwise, - * \p numDependencies entries will be filled in. If \p numDependencies is higher than the actual - * number of dependencies, the remaining entries in \p dependencies will be set to NULL, and the - * number of nodes actually obtained will be returned in \p numDependencies. + * Returns a list of \p node's dependencies. \p dependencies may be NULL, in + * which case this function will return the number of dependencies in \p + * numDependencies. Otherwise, \p numDependencies entries will be filled in. If + * \p numDependencies is higher than the actual number of dependencies, the + * remaining entries in \p dependencies will be set to NULL, and the number of + * nodes actually obtained will be returned in \p numDependencies. * * \param hNode - Node to query * \param dependencies - Pointer to return the dependencies @@ -12190,17 +12872,19 @@ CUresult CUDAAPI cuGraphGetEdges(CUgraph hGraph, CUgraphNode *from, CUgraphNode * ::cuGraphAddDependencies, * ::cuGraphRemoveDependencies */ -CUresult CUDAAPI cuGraphNodeGetDependencies(CUgraphNode hNode, CUgraphNode *dependencies, size_t *numDependencies); +CUresult CUDAAPI cuGraphNodeGetDependencies(CUgraphNode hNode, + CUgraphNode *dependencies, + size_t *numDependencies); /** * \brief Returns a node's dependent nodes * - * Returns a list of \p node's dependent nodes. \p dependentNodes may be NULL, in which - * case this function will return the number of dependent nodes in \p numDependentNodes. - * Otherwise, \p numDependentNodes entries will be filled in. If \p numDependentNodes is - * higher than the actual number of dependent nodes, the remaining entries in - * \p dependentNodes will be set to NULL, and the number of nodes actually obtained will - * be returned in \p numDependentNodes. + * Returns a list of \p node's dependent nodes. \p dependentNodes may be NULL, + * in which case this function will return the number of dependent nodes in \p + * numDependentNodes. Otherwise, \p numDependentNodes entries will be filled in. + * If \p numDependentNodes is higher than the actual number of dependent nodes, + * the remaining entries in \p dependentNodes will be set to NULL, and the + * number of nodes actually obtained will be returned in \p numDependentNodes. * * \param hNode - Node to query * \param dependentNodes - Pointer to return the dependent nodes @@ -12222,7 +12906,9 @@ CUresult CUDAAPI cuGraphNodeGetDependencies(CUgraphNode hNode, CUgraphNode *depe * ::cuGraphAddDependencies, * ::cuGraphRemoveDependencies */ -CUresult CUDAAPI cuGraphNodeGetDependentNodes(CUgraphNode hNode, CUgraphNode *dependentNodes, size_t *numDependentNodes); +CUresult CUDAAPI cuGraphNodeGetDependentNodes(CUgraphNode hNode, + CUgraphNode *dependentNodes, + size_t *numDependentNodes); /** * \brief Adds dependency edges to a graph @@ -12251,7 +12937,9 @@ CUresult CUDAAPI cuGraphNodeGetDependentNodes(CUgraphNode hNode, CUgraphNode *de * ::cuGraphNodeGetDependencies, * ::cuGraphNodeGetDependentNodes */ -CUresult CUDAAPI cuGraphAddDependencies(CUgraph hGraph, const CUgraphNode *from, const CUgraphNode *to, size_t numDependencies); +CUresult CUDAAPI cuGraphAddDependencies(CUgraph hGraph, const CUgraphNode *from, + const CUgraphNode *to, + size_t numDependencies); /** * \brief Removes dependency edges from a graph @@ -12280,13 +12968,16 @@ CUresult CUDAAPI cuGraphAddDependencies(CUgraph hGraph, const CUgraphNode *from, * ::cuGraphNodeGetDependencies, * ::cuGraphNodeGetDependentNodes */ -CUresult CUDAAPI cuGraphRemoveDependencies(CUgraph hGraph, const CUgraphNode *from, const CUgraphNode *to, size_t numDependencies); +CUresult CUDAAPI cuGraphRemoveDependencies(CUgraph hGraph, + const CUgraphNode *from, + const CUgraphNode *to, + size_t numDependencies); /** * \brief Remove a node from the graph * - * Removes \p hNode from its graph. This operation also severs any dependencies of other nodes - * on \p hNode and vice versa. + * Removes \p hNode from its graph. This operation also severs any dependencies + * of other nodes on \p hNode and vice versa. * * \param hNode - Node to remove * @@ -12314,18 +13005,18 @@ CUresult CUDAAPI cuGraphDestroyNode(CUgraphNode hNode); * validated. If instantiation is successful, a handle to the instantiated graph * is returned in \p graphExec. * - * If there are any errors, diagnostic information may be returned in \p errorNode and - * \p logBuffer. This is the primary way to inspect instantiation errors. The output - * will be null terminated unless the diagnostics overflow + * If there are any errors, diagnostic information may be returned in \p + * errorNode and \p logBuffer. This is the primary way to inspect instantiation + * errors. The output will be null terminated unless the diagnostics overflow * the buffer. In this case, they will be truncated, and the last byte can be * inspected to determine if truncation occurred. * * \param phGraphExec - Returns instantiated graph * \param hGraph - Graph to instantiate - * \param phErrorNode - In case of an instantiation error, this may be modified to - * indicate a node contributing to the error - * \param logBuffer - A character buffer to store diagnostic messages - * \param bufferSize - Size of the log buffer in bytes + * \param phErrorNode - In case of an instantiation error, this may be modified + * to indicate a node contributing to the error \param logBuffer - A character + * buffer to store diagnostic messages \param bufferSize - Size of the log + * buffer in bytes * * \return * ::CUDA_SUCCESS, @@ -12340,8 +13031,9 @@ CUresult CUDAAPI cuGraphDestroyNode(CUgraphNode hNode); * ::cuGraphLaunch, * ::cuGraphExecDestroy */ -CUresult CUDAAPI cuGraphInstantiate(CUgraphExec *phGraphExec, CUgraph hGraph, CUgraphNode *phErrorNode, char *logBuffer, size_t bufferSize); - +CUresult CUDAAPI cuGraphInstantiate(CUgraphExec *phGraphExec, CUgraph hGraph, + CUgraphNode *phErrorNode, char *logBuffer, + size_t bufferSize); #if __CUDA_API_VERSION >= 10010 /** @@ -12351,8 +13043,8 @@ CUresult CUDAAPI cuGraphInstantiate(CUgraphExec *phGraphExec, CUgraph hGraph, CU * The node is identified by the corresponding node \p hNode in the * non-executable graph, from which the executable graph was instantiated. * - * \p hNode must not have been removed from the original graph. The \p func field - * of \p nodeParams cannot be modified and must match the original value. + * \p hNode must not have been removed from the original graph. The \p func + * field of \p nodeParams cannot be modified and must match the original value. * All other values can be modified. * * The modifications take effect at the next launch of \p hGraphExec. Already @@ -12360,8 +13052,8 @@ CUresult CUDAAPI cuGraphInstantiate(CUgraphExec *phGraphExec, CUgraph hGraph, CU * \p hNode is also not modified by this call. * * \param hGraphExec - The executable graph in which to set the specified node - * \param hNode - kernel node from the graph from which graphExec was instantiated - * \param nodeParams - Updated Parameters to set + * \param hNode - kernel node from the graph from which graphExec was + * instantiated \param nodeParams - Updated Parameters to set * * \return * ::CUDA_SUCCESS, @@ -12374,17 +13066,20 @@ CUresult CUDAAPI cuGraphInstantiate(CUgraphExec *phGraphExec, CUgraph hGraph, CU * ::cuGraphKernelNodeSetParams, * ::cuGraphInstantiate */ - CUresult CUDAAPI cuGraphExecKernelNodeSetParams(CUgraphExec hGraphExec, CUgraphNode hNode, const CUDA_KERNEL_NODE_PARAMS *nodeParams); +CUresult CUDAAPI +cuGraphExecKernelNodeSetParams(CUgraphExec hGraphExec, CUgraphNode hNode, + const CUDA_KERNEL_NODE_PARAMS *nodeParams); #endif /* __CUDA_API_VERSION >= 10010 */ /** * \brief Launches an executable graph in a stream * - * Executes \p hGraphExec in \p hStream. Only one instance of \p hGraphExec may be executing - * at a time. Each launch is ordered behind both any previous work in \p hStream - * and any previous launches of \p hGraphExec. To execute a graph concurrently, it must be - * instantiated multiple times into multiple executable graphs. + * Executes \p hGraphExec in \p hStream. Only one instance of \p hGraphExec may + * be executing at a time. Each launch is ordered behind both any previous work + * in \p hStream and any previous launches of \p hGraphExec. To execute a graph + * concurrently, it must be instantiated multiple times into multiple executable + * graphs. * * \param hGraphExec - Executable graph to launch * \param hStream - Stream in which to launch the graph @@ -12447,7 +13142,7 @@ CUresult CUDAAPI cuGraphExecDestroy(CUgraphExec hGraphExec); */ CUresult CUDAAPI cuGraphDestroy(CUgraph hGraph); /** @} */ /* END CUDA_GRAPH */ -#endif /* __CUDA_API_VERSION >= 10000 */ +#endif /* __CUDA_API_VERSION >= 10000 */ #if __CUDA_API_VERSION >= 6050 /** @@ -12456,8 +13151,8 @@ CUresult CUDAAPI cuGraphDestroy(CUgraph hGraph); * ___MANBRIEF___ occupancy calculation functions of the low-level CUDA driver * API (___CURRENT_FILE___) ___ENDMANBRIEF___ * - * This section describes the occupancy calculation functions of the low-level CUDA - * driver application programming interface. + * This section describes the occupancy calculation functions of the low-level + * CUDA driver application programming interface. * * @{ */ @@ -12470,8 +13165,9 @@ CUresult CUDAAPI cuGraphDestroy(CUgraph hGraph); * * \param numBlocks - Returned occupancy * \param func - Kernel for which occupancy is calculated - * \param blockSize - Block size the kernel is intended to be launched with - * \param dynamicSMemSize - Per-block dynamic shared memory usage intended, in bytes + * \param blockSize - Block size the kernel is intended to be launched + * with \param dynamicSMemSize - Per-block dynamic shared memory usage intended, + * in bytes * * \return * ::CUDA_SUCCESS, @@ -12485,7 +13181,8 @@ CUresult CUDAAPI cuGraphDestroy(CUgraph hGraph); * \sa * ::cudaOccupancyMaxActiveBlocksPerMultiprocessor */ -CUresult CUDAAPI cuOccupancyMaxActiveBlocksPerMultiprocessor(int *numBlocks, CUfunction func, int blockSize, size_t dynamicSMemSize); +CUresult CUDAAPI cuOccupancyMaxActiveBlocksPerMultiprocessor( + int *numBlocks, CUfunction func, int blockSize, size_t dynamicSMemSize); /** * \brief Returns occupancy of a function @@ -12511,9 +13208,10 @@ CUresult CUDAAPI cuOccupancyMaxActiveBlocksPerMultiprocessor(int *numBlocks, CUf * * \param numBlocks - Returned occupancy * \param func - Kernel for which occupancy is calculated - * \param blockSize - Block size the kernel is intended to be launched with - * \param dynamicSMemSize - Per-block dynamic shared memory usage intended, in bytes - * \param flags - Requested behavior for the occupancy calculator + * \param blockSize - Block size the kernel is intended to be launched + * with \param dynamicSMemSize - Per-block dynamic shared memory usage intended, + * in bytes \param flags - Requested behavior for the occupancy + * calculator * * \return * ::CUDA_SUCCESS, @@ -12527,7 +13225,9 @@ CUresult CUDAAPI cuOccupancyMaxActiveBlocksPerMultiprocessor(int *numBlocks, CUf * \sa * ::cudaOccupancyMaxActiveBlocksPerMultiprocessorWithFlags */ -CUresult CUDAAPI cuOccupancyMaxActiveBlocksPerMultiprocessorWithFlags(int *numBlocks, CUfunction func, int blockSize, size_t dynamicSMemSize, unsigned int flags); +CUresult CUDAAPI cuOccupancyMaxActiveBlocksPerMultiprocessorWithFlags( + int *numBlocks, CUfunction func, int blockSize, size_t dynamicSMemSize, + unsigned int flags); /** * \brief Suggest a launch configuration with reasonable occupancy @@ -12560,12 +13260,14 @@ CUresult CUDAAPI cuOccupancyMaxActiveBlocksPerMultiprocessorWithFlags(int *numBl * size_t blockToSmem(int blockSize); * \endcode * - * \param minGridSize - Returned minimum grid size needed to achieve the maximum occupancy - * \param blockSize - Returned maximum block size that can achieve the maximum occupancy - * \param func - Kernel for which launch configuration is calculated - * \param blockSizeToDynamicSMemSize - A function that calculates how much per-block dynamic shared memory \p func uses based on the block size - * \param dynamicSMemSize - Dynamic shared memory usage intended, in bytes - * \param blockSizeLimit - The maximum block size \p func is designed to handle + * \param minGridSize - Returned minimum grid size needed to achieve the maximum + * occupancy \param blockSize - Returned maximum block size that can achieve + * the maximum occupancy \param func - Kernel for which launch + * configuration is calculated \param blockSizeToDynamicSMemSize - A function + * that calculates how much per-block dynamic shared memory \p func uses based + * on the block size \param dynamicSMemSize - Dynamic shared memory usage + * intended, in bytes \param blockSizeLimit - The maximum block size \p func is + * designed to handle * * \return * ::CUDA_SUCCESS, @@ -12579,7 +13281,10 @@ CUresult CUDAAPI cuOccupancyMaxActiveBlocksPerMultiprocessorWithFlags(int *numBl * \sa * ::cudaOccupancyMaxPotentialBlockSize */ -CUresult CUDAAPI cuOccupancyMaxPotentialBlockSize(int *minGridSize, int *blockSize, CUfunction func, CUoccupancyB2DSize blockSizeToDynamicSMemSize, size_t dynamicSMemSize, int blockSizeLimit); +CUresult CUDAAPI cuOccupancyMaxPotentialBlockSize( + int *minGridSize, int *blockSize, CUfunction func, + CUoccupancyB2DSize blockSizeToDynamicSMemSize, size_t dynamicSMemSize, + int blockSizeLimit); /** * \brief Suggest a launch configuration with reasonable occupancy @@ -12605,13 +13310,14 @@ CUresult CUDAAPI cuOccupancyMaxPotentialBlockSize(int *minGridSize, int *blockSi * can be found about this feature in the "Unified L1/Texture Cache" * section of the Maxwell tuning guide. * - * \param minGridSize - Returned minimum grid size needed to achieve the maximum occupancy - * \param blockSize - Returned maximum block size that can achieve the maximum occupancy - * \param func - Kernel for which launch configuration is calculated - * \param blockSizeToDynamicSMemSize - A function that calculates how much per-block dynamic shared memory \p func uses based on the block size - * \param dynamicSMemSize - Dynamic shared memory usage intended, in bytes - * \param blockSizeLimit - The maximum block size \p func is designed to handle - * \param flags - Options + * \param minGridSize - Returned minimum grid size needed to achieve the maximum + * occupancy \param blockSize - Returned maximum block size that can achieve + * the maximum occupancy \param func - Kernel for which launch + * configuration is calculated \param blockSizeToDynamicSMemSize - A function + * that calculates how much per-block dynamic shared memory \p func uses based + * on the block size \param dynamicSMemSize - Dynamic shared memory usage + * intended, in bytes \param blockSizeLimit - The maximum block size \p func is + * designed to handle \param flags - Options * * \return * ::CUDA_SUCCESS, @@ -12625,10 +13331,13 @@ CUresult CUDAAPI cuOccupancyMaxPotentialBlockSize(int *minGridSize, int *blockSi * \sa * ::cudaOccupancyMaxPotentialBlockSizeWithFlags */ -CUresult CUDAAPI cuOccupancyMaxPotentialBlockSizeWithFlags(int *minGridSize, int *blockSize, CUfunction func, CUoccupancyB2DSize blockSizeToDynamicSMemSize, size_t dynamicSMemSize, int blockSizeLimit, unsigned int flags); +CUresult CUDAAPI cuOccupancyMaxPotentialBlockSizeWithFlags( + int *minGridSize, int *blockSize, CUfunction func, + CUoccupancyB2DSize blockSizeToDynamicSMemSize, size_t dynamicSMemSize, + int blockSizeLimit, unsigned int flags); /** @} */ /* END CUDA_OCCUPANCY */ -#endif /* __CUDA_API_VERSION >= 6050 */ +#endif /* __CUDA_API_VERSION >= 6050 */ /** * \defgroup CUDA_TEXREF_DEPRECATED Texture Reference Management [DEPRECATED] @@ -12671,17 +13380,19 @@ CUresult CUDAAPI cuOccupancyMaxPotentialBlockSizeWithFlags(int *minGridSize, int * ::cuTexRefGetFilterMode, ::cuTexRefGetFlags, ::cuTexRefGetFormat, * ::cudaBindTextureToArray */ -CUresult CUDAAPI cuTexRefSetArray(CUtexref hTexRef, CUarray hArray, unsigned int Flags); +CUresult CUDAAPI cuTexRefSetArray(CUtexref hTexRef, CUarray hArray, + unsigned int Flags); /** * \brief Binds a mipmapped array to a texture reference * * \deprecated * - * Binds the CUDA mipmapped array \p hMipmappedArray to the texture reference \p hTexRef. - * Any previous address or CUDA array state associated with the texture reference - * is superseded by this function. \p Flags must be set to ::CU_TRSA_OVERRIDE_FORMAT. - * Any CUDA array previously bound to \p hTexRef is unbound. + * Binds the CUDA mipmapped array \p hMipmappedArray to the texture reference \p + * hTexRef. Any previous address or CUDA array state associated with the texture + * reference is superseded by this function. \p Flags must be set to + * ::CU_TRSA_OVERRIDE_FORMAT. Any CUDA array previously bound to \p hTexRef is + * unbound. * * \param hTexRef - Texture reference to bind * \param hMipmappedArray - Mipmapped array to bind @@ -12701,7 +13412,9 @@ CUresult CUDAAPI cuTexRefSetArray(CUtexref hTexRef, CUarray hArray, unsigned int * ::cuTexRefGetFilterMode, ::cuTexRefGetFlags, ::cuTexRefGetFormat, * ::cudaBindTextureToMipmappedArray */ -CUresult CUDAAPI cuTexRefSetMipmappedArray(CUtexref hTexRef, CUmipmappedArray hMipmappedArray, unsigned int Flags); +CUresult CUDAAPI cuTexRefSetMipmappedArray(CUtexref hTexRef, + CUmipmappedArray hMipmappedArray, + unsigned int Flags); #if __CUDA_API_VERSION >= 3020 /** @@ -12748,7 +13461,8 @@ CUresult CUDAAPI cuTexRefSetMipmappedArray(CUtexref hTexRef, CUmipmappedArray hM * ::cuTexRefGetFilterMode, ::cuTexRefGetFlags, ::cuTexRefGetFormat, * ::cudaBindTexture */ -CUresult CUDAAPI cuTexRefSetAddress(size_t *ByteOffset, CUtexref hTexRef, CUdeviceptr dptr, size_t bytes); +CUresult CUDAAPI cuTexRefSetAddress(size_t *ByteOffset, CUtexref hTexRef, + CUdeviceptr dptr, size_t bytes); /** * \brief Binds an address as a 2D texture reference @@ -12803,7 +13517,9 @@ CUresult CUDAAPI cuTexRefSetAddress(size_t *ByteOffset, CUtexref hTexRef, CUdevi * ::cuTexRefGetFilterMode, ::cuTexRefGetFlags, ::cuTexRefGetFormat, * ::cudaBindTexture2D */ -CUresult CUDAAPI cuTexRefSetAddress2D(CUtexref hTexRef, const CUDA_ARRAY_DESCRIPTOR *desc, CUdeviceptr dptr, size_t Pitch); +CUresult CUDAAPI cuTexRefSetAddress2D(CUtexref hTexRef, + const CUDA_ARRAY_DESCRIPTOR *desc, + CUdeviceptr dptr, size_t Pitch); #endif /* __CUDA_API_VERSION >= 3020 */ /** @@ -12839,7 +13555,8 @@ CUresult CUDAAPI cuTexRefSetAddress2D(CUtexref hTexRef, const CUDA_ARRAY_DESCRIP * ::cudaBindTextureToArray, * ::cudaBindTextureToMipmappedArray */ -CUresult CUDAAPI cuTexRefSetFormat(CUtexref hTexRef, CUarray_format fmt, int NumPackedComponents); +CUresult CUDAAPI cuTexRefSetFormat(CUtexref hTexRef, CUarray_format fmt, + int NumPackedComponents); /** * \brief Sets the addressing mode for a texture reference @@ -12885,7 +13602,8 @@ CUresult CUDAAPI cuTexRefSetFormat(CUtexref hTexRef, CUarray_format fmt, int Num * ::cudaBindTextureToArray, * ::cudaBindTextureToMipmappedArray */ -CUresult CUDAAPI cuTexRefSetAddressMode(CUtexref hTexRef, int dim, CUaddress_mode am); +CUresult CUDAAPI cuTexRefSetAddressMode(CUtexref hTexRef, int dim, + CUaddress_mode am); /** * \brief Sets the filtering mode for a texture reference @@ -12928,7 +13646,8 @@ CUresult CUDAAPI cuTexRefSetFilterMode(CUtexref hTexRef, CUfilter_mode fm); * * \deprecated * - * Specifies the mipmap filtering mode \p fm to be used when reading memory through + * Specifies the mipmap filtering mode \p fm to be used when reading memory + through * the texture reference \p hTexRef. ::CUfilter_mode_enum is defined as: * * \code @@ -12938,7 +13657,8 @@ CUresult CUDAAPI cuTexRefSetFilterMode(CUtexref hTexRef, CUfilter_mode fm); } CUfilter_mode; * \endcode * - * Note that this call has no effect if \p hTexRef is not bound to a mipmapped array. + * Note that this call has no effect if \p hTexRef is not bound to a mipmapped + array. * * \param hTexRef - Texture reference * \param fm - Filtering mode to set @@ -12957,17 +13677,19 @@ CUresult CUDAAPI cuTexRefSetFilterMode(CUtexref hTexRef, CUfilter_mode fm); * ::cuTexRefGetFilterMode, ::cuTexRefGetFlags, ::cuTexRefGetFormat, * ::cudaBindTextureToMipmappedArray */ -CUresult CUDAAPI cuTexRefSetMipmapFilterMode(CUtexref hTexRef, CUfilter_mode fm); +CUresult CUDAAPI cuTexRefSetMipmapFilterMode(CUtexref hTexRef, + CUfilter_mode fm); /** * \brief Sets the mipmap level bias for a texture reference * * \deprecated * - * Specifies the mipmap level bias \p bias to be added to the specified mipmap level when - * reading memory through the texture reference \p hTexRef. + * Specifies the mipmap level bias \p bias to be added to the specified mipmap + * level when reading memory through the texture reference \p hTexRef. * - * Note that this call has no effect if \p hTexRef is not bound to a mipmapped array. + * Note that this call has no effect if \p hTexRef is not bound to a mipmapped + * array. * * \param hTexRef - Texture reference * \param bias - Mipmap level bias @@ -12993,11 +13715,12 @@ CUresult CUDAAPI cuTexRefSetMipmapLevelBias(CUtexref hTexRef, float bias); * * \deprecated * - * Specifies the min/max mipmap level clamps, \p minMipmapLevelClamp and \p maxMipmapLevelClamp - * respectively, to be used when reading memory through the texture reference - * \p hTexRef. + * Specifies the min/max mipmap level clamps, \p minMipmapLevelClamp and \p + * maxMipmapLevelClamp respectively, to be used when reading memory through the + * texture reference \p hTexRef. * - * Note that this call has no effect if \p hTexRef is not bound to a mipmapped array. + * Note that this call has no effect if \p hTexRef is not bound to a mipmapped + * array. * * \param hTexRef - Texture reference * \param minMipmapLevelClamp - Mipmap min level clamp @@ -13017,15 +13740,17 @@ CUresult CUDAAPI cuTexRefSetMipmapLevelBias(CUtexref hTexRef, float bias); * ::cuTexRefGetFilterMode, ::cuTexRefGetFlags, ::cuTexRefGetFormat, * ::cudaBindTextureToMipmappedArray */ -CUresult CUDAAPI cuTexRefSetMipmapLevelClamp(CUtexref hTexRef, float minMipmapLevelClamp, float maxMipmapLevelClamp); +CUresult CUDAAPI cuTexRefSetMipmapLevelClamp(CUtexref hTexRef, + float minMipmapLevelClamp, + float maxMipmapLevelClamp); /** * \brief Sets the maximum anisotropy for a texture reference * * \deprecated * - * Specifies the maximum anisotropy \p maxAniso to be used when reading memory through - * the texture reference \p hTexRef. + * Specifies the maximum anisotropy \p maxAniso to be used when reading memory + * through the texture reference \p hTexRef. * * Note that this call has no effect if \p hTexRef is bound to linear memory. * @@ -13047,24 +13772,24 @@ CUresult CUDAAPI cuTexRefSetMipmapLevelClamp(CUtexref hTexRef, float minMipmapLe * ::cudaBindTextureToArray, * ::cudaBindTextureToMipmappedArray */ -CUresult CUDAAPI cuTexRefSetMaxAnisotropy(CUtexref hTexRef, unsigned int maxAniso); +CUresult CUDAAPI cuTexRefSetMaxAnisotropy(CUtexref hTexRef, + unsigned int maxAniso); /** * \brief Sets the border color for a texture reference * * \deprecated * - * Specifies the value of the RGBA color via the \p pBorderColor to the texture reference - * \p hTexRef. The color value supports only float type and holds color components in - * the following sequence: - * pBorderColor[0] holds 'R' component - * pBorderColor[1] holds 'G' component - * pBorderColor[2] holds 'B' component - * pBorderColor[3] holds 'A' component + * Specifies the value of the RGBA color via the \p pBorderColor to the texture + * reference \p hTexRef. The color value supports only float type and holds + * color components in the following sequence: pBorderColor[0] holds 'R' + * component pBorderColor[1] holds 'G' component pBorderColor[2] holds 'B' + * component pBorderColor[3] holds 'A' component * * Note that the color values can be set only when the Address mode is set to * CU_TR_ADDRESS_MODE_BORDER using ::cuTexRefSetAddressMode. - * Applications using integer border color values have to "reinterpret_cast" their values to float. + * Applications using integer border color values have to "reinterpret_cast" + * their values to float. * * \param hTexRef - Texture reference * \param pBorderColor - RGBA color @@ -13188,8 +13913,8 @@ CUresult CUDAAPI cuTexRefGetArray(CUarray *phArray, CUtexref hTexRef); * \deprecated * * Returns in \p *phMipmappedArray the CUDA mipmapped array bound to the texture - * reference \p hTexRef, or returns ::CUDA_ERROR_INVALID_VALUE if the texture reference - * is not bound to any CUDA mipmapped array. + * reference \p hTexRef, or returns ::CUDA_ERROR_INVALID_VALUE if the texture + * reference is not bound to any CUDA mipmapped array. * * \param phMipmappedArray - Returned mipmapped array * \param hTexRef - Texture reference @@ -13207,7 +13932,8 @@ CUresult CUDAAPI cuTexRefGetArray(CUarray *phArray, CUtexref hTexRef); * ::cuTexRefGetAddress, ::cuTexRefGetAddressMode, * ::cuTexRefGetFilterMode, ::cuTexRefGetFlags, ::cuTexRefGetFormat */ -CUresult CUDAAPI cuTexRefGetMipmappedArray(CUmipmappedArray *phMipmappedArray, CUtexref hTexRef); +CUresult CUDAAPI cuTexRefGetMipmappedArray(CUmipmappedArray *phMipmappedArray, + CUtexref hTexRef); /** * \brief Gets the addressing mode used by a texture reference @@ -13235,7 +13961,8 @@ CUresult CUDAAPI cuTexRefGetMipmappedArray(CUmipmappedArray *phMipmappedArray, C * ::cuTexRefGetAddress, ::cuTexRefGetArray, * ::cuTexRefGetFilterMode, ::cuTexRefGetFlags, ::cuTexRefGetFormat */ -CUresult CUDAAPI cuTexRefGetAddressMode(CUaddress_mode *pam, CUtexref hTexRef, int dim); +CUresult CUDAAPI cuTexRefGetAddressMode(CUaddress_mode *pam, CUtexref hTexRef, + int dim); /** * \brief Gets the filter-mode used by a texture reference @@ -13289,15 +14016,16 @@ CUresult CUDAAPI cuTexRefGetFilterMode(CUfilter_mode *pfm, CUtexref hTexRef); * ::cuTexRefGetAddress, ::cuTexRefGetAddressMode, ::cuTexRefGetArray, * ::cuTexRefGetFilterMode, ::cuTexRefGetFlags */ -CUresult CUDAAPI cuTexRefGetFormat(CUarray_format *pFormat, int *pNumChannels, CUtexref hTexRef); +CUresult CUDAAPI cuTexRefGetFormat(CUarray_format *pFormat, int *pNumChannels, + CUtexref hTexRef); /** * \brief Gets the mipmap filtering mode for a texture reference * * \deprecated * - * Returns the mipmap filtering mode in \p pfm that's used when reading memory through - * the texture reference \p hTexRef. + * Returns the mipmap filtering mode in \p pfm that's used when reading memory + * through the texture reference \p hTexRef. * * \param pfm - Returned mipmap filtering mode * \param hTexRef - Texture reference @@ -13315,15 +14043,16 @@ CUresult CUDAAPI cuTexRefGetFormat(CUarray_format *pFormat, int *pNumChannels, C * ::cuTexRefGetAddress, ::cuTexRefGetAddressMode, ::cuTexRefGetArray, * ::cuTexRefGetFilterMode, ::cuTexRefGetFlags, ::cuTexRefGetFormat */ -CUresult CUDAAPI cuTexRefGetMipmapFilterMode(CUfilter_mode *pfm, CUtexref hTexRef); +CUresult CUDAAPI cuTexRefGetMipmapFilterMode(CUfilter_mode *pfm, + CUtexref hTexRef); /** * \brief Gets the mipmap level bias for a texture reference * * \deprecated * - * Returns the mipmap level bias in \p pBias that's added to the specified mipmap - * level when reading memory through the texture reference \p hTexRef. + * Returns the mipmap level bias in \p pBias that's added to the specified + * mipmap level when reading memory through the texture reference \p hTexRef. * * \param pbias - Returned mipmap level bias * \param hTexRef - Texture reference @@ -13348,8 +14077,9 @@ CUresult CUDAAPI cuTexRefGetMipmapLevelBias(float *pbias, CUtexref hTexRef); * * \deprecated * - * Returns the min/max mipmap level clamps in \p pminMipmapLevelClamp and \p pmaxMipmapLevelClamp - * that's used when reading memory through the texture reference \p hTexRef. + * Returns the min/max mipmap level clamps in \p pminMipmapLevelClamp and \p + * pmaxMipmapLevelClamp that's used when reading memory through the texture + * reference \p hTexRef. * * \param pminMipmapLevelClamp - Returned mipmap min level clamp * \param pmaxMipmapLevelClamp - Returned mipmap max level clamp @@ -13368,15 +14098,17 @@ CUresult CUDAAPI cuTexRefGetMipmapLevelBias(float *pbias, CUtexref hTexRef); * ::cuTexRefGetAddress, ::cuTexRefGetAddressMode, ::cuTexRefGetArray, * ::cuTexRefGetFilterMode, ::cuTexRefGetFlags, ::cuTexRefGetFormat */ -CUresult CUDAAPI cuTexRefGetMipmapLevelClamp(float *pminMipmapLevelClamp, float *pmaxMipmapLevelClamp, CUtexref hTexRef); +CUresult CUDAAPI cuTexRefGetMipmapLevelClamp(float *pminMipmapLevelClamp, + float *pmaxMipmapLevelClamp, + CUtexref hTexRef); /** * \brief Gets the maximum anisotropy for a texture reference * * \deprecated * - * Returns the maximum anisotropy in \p pmaxAniso that's used when reading memory through - * the texture reference \p hTexRef. + * Returns the maximum anisotropy in \p pmaxAniso that's used when reading + * memory through the texture reference \p hTexRef. * * \param pmaxAniso - Returned maximum anisotropy * \param hTexRef - Texture reference @@ -13497,7 +14229,6 @@ CUresult CUDAAPI cuTexRefDestroy(CUtexref hTexRef); /** @} */ /* END CUDA_TEXREF_DEPRECATED */ - /** * \defgroup CUDA_SURFREF_DEPRECATED Surface Reference Management [DEPRECATED] * @@ -13537,7 +14268,8 @@ CUresult CUDAAPI cuTexRefDestroy(CUtexref hTexRef); * ::cuSurfRefGetArray, * ::cudaBindSurfaceToArray */ -CUresult CUDAAPI cuSurfRefSetArray(CUsurfref hSurfRef, CUarray hArray, unsigned int Flags); +CUresult CUDAAPI cuSurfRefSetArray(CUsurfref hSurfRef, CUarray hArray, + unsigned int Flags); /** * \brief Passes back the CUDA array bound to a surface reference. @@ -13581,15 +14313,21 @@ CUresult CUDAAPI cuSurfRefGetArray(CUarray *phArray, CUsurfref hSurfRef); /** * \brief Creates a texture object * - * Creates a texture object and returns it in \p pTexObject. \p pResDesc describes - * the data to texture from. \p pTexDesc describes how the data should be sampled. - * \p pResViewDesc is an optional argument that specifies an alternate format for + * Creates a texture object and returns it in \p pTexObject. \p pResDesc + describes + * the data to texture from. \p pTexDesc describes how the data should be + sampled. + * \p pResViewDesc is an optional argument that specifies an alternate format + for * the data described by \p pResDesc, and also describes the subresource region - * to restrict access to when texturing. \p pResViewDesc can only be specified if + * to restrict access to when texturing. \p pResViewDesc can only be specified + if * the type of resource is a CUDA array or a CUDA mipmapped array. * - * Texture objects are only supported on devices of compute capability 3.0 or higher. - * Additionally, a texture object is an opaque value, and, as such, should only be + * Texture objects are only supported on devices of compute capability 3.0 or + higher. + * Additionally, a texture object is an opaque value, and, as such, should only + be * accessed through CUDA API calls. * * The ::CUDA_RESOURCE_DESC structure is defined as: @@ -13626,7 +14364,8 @@ CUresult CUDAAPI cuSurfRefGetArray(CUarray *phArray, CUsurfref hSurfRef); * \endcode * where: - * - ::CUDA_RESOURCE_DESC::resType specifies the type of resource to texture from. + * - ::CUDA_RESOURCE_DESC::resType specifies the type of resource to texture + from. * CUresourceType is defined as: * \code typedef enum CUresourcetype_enum { @@ -13638,30 +14377,47 @@ CUresult CUDAAPI cuSurfRefGetArray(CUarray *phArray, CUsurfref hSurfRef); * \endcode * * \par - * If ::CUDA_RESOURCE_DESC::resType is set to ::CU_RESOURCE_TYPE_ARRAY, ::CUDA_RESOURCE_DESC::res::array::hArray + * If ::CUDA_RESOURCE_DESC::resType is set to ::CU_RESOURCE_TYPE_ARRAY, + ::CUDA_RESOURCE_DESC::res::array::hArray * must be set to a valid CUDA array handle. * * \par - * If ::CUDA_RESOURCE_DESC::resType is set to ::CU_RESOURCE_TYPE_MIPMAPPED_ARRAY, ::CUDA_RESOURCE_DESC::res::mipmap::hMipmappedArray + * If ::CUDA_RESOURCE_DESC::resType is set to + ::CU_RESOURCE_TYPE_MIPMAPPED_ARRAY, + ::CUDA_RESOURCE_DESC::res::mipmap::hMipmappedArray * must be set to a valid CUDA mipmapped array handle. * * \par - * If ::CUDA_RESOURCE_DESC::resType is set to ::CU_RESOURCE_TYPE_LINEAR, ::CUDA_RESOURCE_DESC::res::linear::devPtr - * must be set to a valid device pointer, that is aligned to ::CU_DEVICE_ATTRIBUTE_TEXTURE_ALIGNMENT. - * ::CUDA_RESOURCE_DESC::res::linear::format and ::CUDA_RESOURCE_DESC::res::linear::numChannels - * describe the format of each component and the number of components per array element. ::CUDA_RESOURCE_DESC::res::linear::sizeInBytes - * specifies the size of the array in bytes. The total number of elements in the linear address range cannot exceed - * ::CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE1D_LINEAR_WIDTH. The number of elements is computed as (sizeInBytes / (sizeof(format) * numChannels)). + * If ::CUDA_RESOURCE_DESC::resType is set to ::CU_RESOURCE_TYPE_LINEAR, + ::CUDA_RESOURCE_DESC::res::linear::devPtr + * must be set to a valid device pointer, that is aligned to + ::CU_DEVICE_ATTRIBUTE_TEXTURE_ALIGNMENT. + * ::CUDA_RESOURCE_DESC::res::linear::format and + ::CUDA_RESOURCE_DESC::res::linear::numChannels + * describe the format of each component and the number of components per array + element. ::CUDA_RESOURCE_DESC::res::linear::sizeInBytes + * specifies the size of the array in bytes. The total number of elements in the + linear address range cannot exceed + * ::CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE1D_LINEAR_WIDTH. The number of elements + is computed as (sizeInBytes / (sizeof(format) * numChannels)). * * \par - * If ::CUDA_RESOURCE_DESC::resType is set to ::CU_RESOURCE_TYPE_PITCH2D, ::CUDA_RESOURCE_DESC::res::pitch2D::devPtr - * must be set to a valid device pointer, that is aligned to ::CU_DEVICE_ATTRIBUTE_TEXTURE_ALIGNMENT. - * ::CUDA_RESOURCE_DESC::res::pitch2D::format and ::CUDA_RESOURCE_DESC::res::pitch2D::numChannels - * describe the format of each component and the number of components per array element. ::CUDA_RESOURCE_DESC::res::pitch2D::width - * and ::CUDA_RESOURCE_DESC::res::pitch2D::height specify the width and height of the array in elements, and cannot exceed - * ::CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE2D_LINEAR_WIDTH and ::CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE2D_LINEAR_HEIGHT respectively. - * ::CUDA_RESOURCE_DESC::res::pitch2D::pitchInBytes specifies the pitch between two rows in bytes and has to be aligned to - * ::CU_DEVICE_ATTRIBUTE_TEXTURE_PITCH_ALIGNMENT. Pitch cannot exceed ::CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE2D_LINEAR_PITCH. + * If ::CUDA_RESOURCE_DESC::resType is set to ::CU_RESOURCE_TYPE_PITCH2D, + ::CUDA_RESOURCE_DESC::res::pitch2D::devPtr + * must be set to a valid device pointer, that is aligned to + ::CU_DEVICE_ATTRIBUTE_TEXTURE_ALIGNMENT. + * ::CUDA_RESOURCE_DESC::res::pitch2D::format and + ::CUDA_RESOURCE_DESC::res::pitch2D::numChannels + * describe the format of each component and the number of components per array + element. ::CUDA_RESOURCE_DESC::res::pitch2D::width + * and ::CUDA_RESOURCE_DESC::res::pitch2D::height specify the width and height + of the array in elements, and cannot exceed + * ::CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE2D_LINEAR_WIDTH and + ::CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE2D_LINEAR_HEIGHT respectively. + * ::CUDA_RESOURCE_DESC::res::pitch2D::pitchInBytes specifies the pitch between + two rows in bytes and has to be aligned to + * ::CU_DEVICE_ATTRIBUTE_TEXTURE_PITCH_ALIGNMENT. Pitch cannot exceed + ::CU_DEVICE_ATTRIBUTE_MAXIMUM_TEXTURE2D_LINEAR_PITCH. * * - ::flags must be set to zero. * @@ -13680,7 +14436,8 @@ CUresult CUDAAPI cuSurfRefGetArray(CUarray *phArray, CUsurfref hSurfRef); } CUDA_TEXTURE_DESC; * \endcode * where - * - ::CUDA_TEXTURE_DESC::addressMode specifies the addressing mode for each dimension of the texture data. ::CUaddress_mode is defined as: + * - ::CUDA_TEXTURE_DESC::addressMode specifies the addressing mode for each + dimension of the texture data. ::CUaddress_mode is defined as: * \code typedef enum CUaddress_mode_enum { CU_TR_ADDRESS_MODE_WRAP = 0, @@ -13689,35 +14446,47 @@ CUresult CUDAAPI cuSurfRefGetArray(CUarray *phArray, CUsurfref hSurfRef); CU_TR_ADDRESS_MODE_BORDER = 3 } CUaddress_mode; * \endcode - * This is ignored if ::CUDA_RESOURCE_DESC::resType is ::CU_RESOURCE_TYPE_LINEAR. Also, if the flag, ::CU_TRSF_NORMALIZED_COORDINATES + * This is ignored if ::CUDA_RESOURCE_DESC::resType is + ::CU_RESOURCE_TYPE_LINEAR. Also, if the flag, ::CU_TRSF_NORMALIZED_COORDINATES * is not set, the only supported address mode is ::CU_TR_ADDRESS_MODE_CLAMP. * - * - ::CUDA_TEXTURE_DESC::filterMode specifies the filtering mode to be used when fetching from the texture. CUfilter_mode is defined as: + * - ::CUDA_TEXTURE_DESC::filterMode specifies the filtering mode to be used + when fetching from the texture. CUfilter_mode is defined as: * \code typedef enum CUfilter_mode_enum { CU_TR_FILTER_MODE_POINT = 0, CU_TR_FILTER_MODE_LINEAR = 1 } CUfilter_mode; * \endcode - * This is ignored if ::CUDA_RESOURCE_DESC::resType is ::CU_RESOURCE_TYPE_LINEAR. + * This is ignored if ::CUDA_RESOURCE_DESC::resType is + ::CU_RESOURCE_TYPE_LINEAR. * * - ::CUDA_TEXTURE_DESC::flags can be any combination of the following: - * - ::CU_TRSF_READ_AS_INTEGER, which suppresses the default behavior of having the texture promote integer data to floating point data in the - * range [0, 1]. Note that texture with 32-bit integer format would not be promoted, regardless of whether or not this flag is specified. - * - ::CU_TRSF_NORMALIZED_COORDINATES, which suppresses the default behavior of having the texture coordinates range from [0, Dim) where Dim is - * the width or height of the CUDA array. Instead, the texture coordinates [0, 1.0) reference the entire breadth of the array dimension; Note + * - ::CU_TRSF_READ_AS_INTEGER, which suppresses the default behavior of + having the texture promote integer data to floating point data in the + * range [0, 1]. Note that texture with 32-bit integer format would not be + promoted, regardless of whether or not this flag is specified. + * - ::CU_TRSF_NORMALIZED_COORDINATES, which suppresses the default behavior + of having the texture coordinates range from [0, Dim) where Dim is + * the width or height of the CUDA array. Instead, the texture coordinates + [0, 1.0) reference the entire breadth of the array dimension; Note * that for CUDA mipmapped arrays, this flag has to be set. * - * - ::CUDA_TEXTURE_DESC::maxAnisotropy specifies the maximum anisotropy ratio to be used when doing anisotropic filtering. This value will be + * - ::CUDA_TEXTURE_DESC::maxAnisotropy specifies the maximum anisotropy ratio + to be used when doing anisotropic filtering. This value will be * clamped to the range [1,16]. * - * - ::CUDA_TEXTURE_DESC::mipmapFilterMode specifies the filter mode when the calculated mipmap level lies between two defined mipmap levels. + * - ::CUDA_TEXTURE_DESC::mipmapFilterMode specifies the filter mode when the + calculated mipmap level lies between two defined mipmap levels. * - * - ::CUDA_TEXTURE_DESC::mipmapLevelBias specifies the offset to be applied to the calculated mipmap level. + * - ::CUDA_TEXTURE_DESC::mipmapLevelBias specifies the offset to be applied to + the calculated mipmap level. * - * - ::CUDA_TEXTURE_DESC::minMipmapLevelClamp specifies the lower end of the mipmap level range to clamp access to. + * - ::CUDA_TEXTURE_DESC::minMipmapLevelClamp specifies the lower end of the + mipmap level range to clamp access to. * - * - ::CUDA_TEXTURE_DESC::maxMipmapLevelClamp specifies the upper end of the mipmap level range to clamp access to. + * - ::CUDA_TEXTURE_DESC::maxMipmapLevelClamp specifies the upper end of the + mipmap level range to clamp access to. * * * The ::CUDA_RESOURCE_VIEW_DESC struct is defined as @@ -13735,36 +14504,53 @@ CUresult CUDAAPI cuSurfRefGetArray(CUarray *phArray, CUsurfref hSurfRef); } CUDA_RESOURCE_VIEW_DESC; * \endcode * where: - * - ::CUDA_RESOURCE_VIEW_DESC::format specifies how the data contained in the CUDA array or CUDA mipmapped array should - * be interpreted. Note that this can incur a change in size of the texture data. If the resource view format is a block - * compressed format, then the underlying CUDA array or CUDA mipmapped array has to have a base of format ::CU_AD_FORMAT_UNSIGNED_INT32. - * with 2 or 4 channels, depending on the block compressed format. For ex., BC1 and BC4 require the underlying CUDA array to have - * a format of ::CU_AD_FORMAT_UNSIGNED_INT32 with 2 channels. The other BC formats require the underlying resource to have the same base + * - ::CUDA_RESOURCE_VIEW_DESC::format specifies how the data contained in the + CUDA array or CUDA mipmapped array should + * be interpreted. Note that this can incur a change in size of the texture + data. If the resource view format is a block + * compressed format, then the underlying CUDA array or CUDA mipmapped array + has to have a base of format ::CU_AD_FORMAT_UNSIGNED_INT32. + * with 2 or 4 channels, depending on the block compressed format. For ex., + BC1 and BC4 require the underlying CUDA array to have + * a format of ::CU_AD_FORMAT_UNSIGNED_INT32 with 2 channels. The other BC + formats require the underlying resource to have the same base * format but with 4 channels. * - * - ::CUDA_RESOURCE_VIEW_DESC::width specifies the new width of the texture data. If the resource view format is a block - * compressed format, this value has to be 4 times the original width of the resource. For non block compressed formats, + * - ::CUDA_RESOURCE_VIEW_DESC::width specifies the new width of the texture + data. If the resource view format is a block + * compressed format, this value has to be 4 times the original width of the + resource. For non block compressed formats, * this value has to be equal to that of the original resource. * - * - ::CUDA_RESOURCE_VIEW_DESC::height specifies the new height of the texture data. If the resource view format is a block - * compressed format, this value has to be 4 times the original height of the resource. For non block compressed formats, + * - ::CUDA_RESOURCE_VIEW_DESC::height specifies the new height of the texture + data. If the resource view format is a block + * compressed format, this value has to be 4 times the original height of the + resource. For non block compressed formats, * this value has to be equal to that of the original resource. * - * - ::CUDA_RESOURCE_VIEW_DESC::depth specifies the new depth of the texture data. This value has to be equal to that of the + * - ::CUDA_RESOURCE_VIEW_DESC::depth specifies the new depth of the texture + data. This value has to be equal to that of the * original resource. * - * - ::CUDA_RESOURCE_VIEW_DESC::firstMipmapLevel specifies the most detailed mipmap level. This will be the new mipmap level zero. - * For non-mipmapped resources, this value has to be zero.::CUDA_TEXTURE_DESC::minMipmapLevelClamp and ::CUDA_TEXTURE_DESC::maxMipmapLevelClamp - * will be relative to this value. For ex., if the firstMipmapLevel is set to 2, and a minMipmapLevelClamp of 1.2 is specified, + * - ::CUDA_RESOURCE_VIEW_DESC::firstMipmapLevel specifies the most detailed + mipmap level. This will be the new mipmap level zero. + * For non-mipmapped resources, this value has to be + zero.::CUDA_TEXTURE_DESC::minMipmapLevelClamp and + ::CUDA_TEXTURE_DESC::maxMipmapLevelClamp + * will be relative to this value. For ex., if the firstMipmapLevel is set to + 2, and a minMipmapLevelClamp of 1.2 is specified, * then the actual minimum mipmap level clamp will be 3.2. * - * - ::CUDA_RESOURCE_VIEW_DESC::lastMipmapLevel specifies the least detailed mipmap level. For non-mipmapped resources, this value + * - ::CUDA_RESOURCE_VIEW_DESC::lastMipmapLevel specifies the least detailed + mipmap level. For non-mipmapped resources, this value * has to be zero. * - * - ::CUDA_RESOURCE_VIEW_DESC::firstLayer specifies the first layer index for layered textures. This will be the new layer zero. + * - ::CUDA_RESOURCE_VIEW_DESC::firstLayer specifies the first layer index for + layered textures. This will be the new layer zero. * For non-layered resources, this value has to be zero. * - * - ::CUDA_RESOURCE_VIEW_DESC::lastLayer specifies the last layer index for layered textures. For non-layered resources, + * - ::CUDA_RESOURCE_VIEW_DESC::lastLayer specifies the last layer index for + layered textures. For non-layered resources, * this value has to be zero. * * @@ -13784,7 +14570,10 @@ CUresult CUDAAPI cuSurfRefGetArray(CUarray *phArray, CUsurfref hSurfRef); * ::cuTexObjectDestroy, * ::cudaCreateTextureObject */ -CUresult CUDAAPI cuTexObjectCreate(CUtexObject *pTexObject, const CUDA_RESOURCE_DESC *pResDesc, const CUDA_TEXTURE_DESC *pTexDesc, const CUDA_RESOURCE_VIEW_DESC *pResViewDesc); +CUresult CUDAAPI cuTexObjectCreate(CUtexObject *pTexObject, + const CUDA_RESOURCE_DESC *pResDesc, + const CUDA_TEXTURE_DESC *pTexDesc, + const CUDA_RESOURCE_VIEW_DESC *pResViewDesc); /** * \brief Destroys a texture object @@ -13809,7 +14598,8 @@ CUresult CUDAAPI cuTexObjectDestroy(CUtexObject texObject); /** * \brief Returns a texture object's resource descriptor * - * Returns the resource descriptor for the texture object specified by \p texObject. + * Returns the resource descriptor for the texture object specified by \p + * texObject. * * \param pResDesc - Resource descriptor * \param texObject - Texture object @@ -13825,12 +14615,14 @@ CUresult CUDAAPI cuTexObjectDestroy(CUtexObject texObject); * ::cuTexObjectCreate, * ::cudaGetTextureObjectResourceDesc, */ -CUresult CUDAAPI cuTexObjectGetResourceDesc(CUDA_RESOURCE_DESC *pResDesc, CUtexObject texObject); +CUresult CUDAAPI cuTexObjectGetResourceDesc(CUDA_RESOURCE_DESC *pResDesc, + CUtexObject texObject); /** * \brief Returns a texture object's texture descriptor * - * Returns the texture descriptor for the texture object specified by \p texObject. + * Returns the texture descriptor for the texture object specified by \p + * texObject. * * \param pTexDesc - Texture descriptor * \param texObject - Texture object @@ -13846,13 +14638,15 @@ CUresult CUDAAPI cuTexObjectGetResourceDesc(CUDA_RESOURCE_DESC *pResDesc, CUtexO * ::cuTexObjectCreate, * ::cudaGetTextureObjectTextureDesc */ -CUresult CUDAAPI cuTexObjectGetTextureDesc(CUDA_TEXTURE_DESC *pTexDesc, CUtexObject texObject); +CUresult CUDAAPI cuTexObjectGetTextureDesc(CUDA_TEXTURE_DESC *pTexDesc, + CUtexObject texObject); /** * \brief Returns a texture object's resource view descriptor * - * Returns the resource view descriptor for the texture object specified by \p texObject. - * If no resource view was set for \p texObject, the ::CUDA_ERROR_INVALID_VALUE is returned. + * Returns the resource view descriptor for the texture object specified by \p + * texObject. If no resource view was set for \p texObject, the + * ::CUDA_ERROR_INVALID_VALUE is returned. * * \param pResViewDesc - Resource view descriptor * \param texObject - Texture object @@ -13868,7 +14662,8 @@ CUresult CUDAAPI cuTexObjectGetTextureDesc(CUDA_TEXTURE_DESC *pTexDesc, CUtexObj * ::cuTexObjectCreate, * ::cudaGetTextureObjectResourceViewDesc */ -CUresult CUDAAPI cuTexObjectGetResourceViewDesc(CUDA_RESOURCE_VIEW_DESC *pResViewDesc, CUtexObject texObject); +CUresult CUDAAPI cuTexObjectGetResourceViewDesc( + CUDA_RESOURCE_VIEW_DESC *pResViewDesc, CUtexObject texObject); /** @} */ /* END CUDA_TEXOBJECT */ @@ -13888,14 +14683,16 @@ CUresult CUDAAPI cuTexObjectGetResourceViewDesc(CUDA_RESOURCE_VIEW_DESC *pResVie /** * \brief Creates a surface object * - * Creates a surface object and returns it in \p pSurfObject. \p pResDesc describes - * the data to perform surface load/stores on. ::CUDA_RESOURCE_DESC::resType must be + * Creates a surface object and returns it in \p pSurfObject. \p pResDesc + * describes the data to perform surface load/stores on. + * ::CUDA_RESOURCE_DESC::resType must be * ::CU_RESOURCE_TYPE_ARRAY and ::CUDA_RESOURCE_DESC::res::array::hArray - * must be set to a valid CUDA array handle. ::CUDA_RESOURCE_DESC::flags must be set to zero. + * must be set to a valid CUDA array handle. ::CUDA_RESOURCE_DESC::flags must be + * set to zero. * - * Surface objects are only supported on devices of compute capability 3.0 or higher. - * Additionally, a surface object is an opaque value, and, as such, should only be - * accessed through CUDA API calls. + * Surface objects are only supported on devices of compute capability 3.0 or + * higher. Additionally, a surface object is an opaque value, and, as such, + * should only be accessed through CUDA API calls. * * \param pSurfObject - Surface object to create * \param pResDesc - Resource descriptor @@ -13911,7 +14708,8 @@ CUresult CUDAAPI cuTexObjectGetResourceViewDesc(CUDA_RESOURCE_VIEW_DESC *pResVie * ::cuSurfObjectDestroy, * ::cudaCreateSurfaceObject */ -CUresult CUDAAPI cuSurfObjectCreate(CUsurfObject *pSurfObject, const CUDA_RESOURCE_DESC *pResDesc); +CUresult CUDAAPI cuSurfObjectCreate(CUsurfObject *pSurfObject, + const CUDA_RESOURCE_DESC *pResDesc); /** * \brief Destroys a surface object @@ -13936,7 +14734,8 @@ CUresult CUDAAPI cuSurfObjectDestroy(CUsurfObject surfObject); /** * \brief Returns a surface object's resource descriptor * - * Returns the resource descriptor for the surface object specified by \p surfObject. + * Returns the resource descriptor for the surface object specified by \p + * surfObject. * * \param pResDesc - Resource descriptor * \param surfObject - Surface object @@ -13952,10 +14751,11 @@ CUresult CUDAAPI cuSurfObjectDestroy(CUsurfObject surfObject); * ::cuSurfObjectCreate, * ::cudaGetSurfaceObjectResourceDesc */ -CUresult CUDAAPI cuSurfObjectGetResourceDesc(CUDA_RESOURCE_DESC *pResDesc, CUsurfObject surfObject); +CUresult CUDAAPI cuSurfObjectGetResourceDesc(CUDA_RESOURCE_DESC *pResDesc, + CUsurfObject surfObject); /** @} */ /* END CUDA_SURFOBJECT */ -#endif /* __CUDA_API_VERSION >= 5000 */ +#endif /* __CUDA_API_VERSION >= 5000 */ /** * \defgroup CUDA_PEER_ACCESS Peer Context Memory Access @@ -13974,16 +14774,16 @@ CUresult CUDAAPI cuSurfObjectGetResourceDesc(CUDA_RESOURCE_DESC *pResDesc, CUsur /** * \brief Queries if a device may directly access a peer device's memory. * - * Returns in \p *canAccessPeer a value of 1 if contexts on \p dev are capable of - * directly accessing memory from contexts on \p peerDev and 0 otherwise. - * If direct access of \p peerDev from \p dev is possible, then access may be + * Returns in \p *canAccessPeer a value of 1 if contexts on \p dev are capable + * of directly accessing memory from contexts on \p peerDev and 0 otherwise. If + * direct access of \p peerDev from \p dev is possible, then access may be * enabled on two specific contexts by calling ::cuCtxEnablePeerAccess(). * * \param canAccessPeer - Returned access capability * \param dev - Device from which allocations on \p peerDev are to * be directly accessed. - * \param peerDev - Device on which the allocations to be directly accessed - * by \p dev reside. + * \param peerDev - Device on which the allocations to be directly + * accessed by \p dev reside. * * \return * ::CUDA_SUCCESS, @@ -13997,26 +14797,28 @@ CUresult CUDAAPI cuSurfObjectGetResourceDesc(CUDA_RESOURCE_DESC *pResDesc, CUsur * ::cuCtxDisablePeerAccess, * ::cudaDeviceCanAccessPeer */ -CUresult CUDAAPI cuDeviceCanAccessPeer(int *canAccessPeer, CUdevice dev, CUdevice peerDev); +CUresult CUDAAPI cuDeviceCanAccessPeer(int *canAccessPeer, CUdevice dev, + CUdevice peerDev); /** * \brief Enables direct access to memory allocations in a peer context. * - * If both the current context and \p peerContext are on devices which support unified - * addressing (as may be queried using ::CU_DEVICE_ATTRIBUTE_UNIFIED_ADDRESSING) and same - * major compute capability, then on success all allocations from \p peerContext will - * immediately be accessible by the current context. See \ref CUDA_UNIFIED for additional + * If both the current context and \p peerContext are on devices which support + * unified addressing (as may be queried using + * ::CU_DEVICE_ATTRIBUTE_UNIFIED_ADDRESSING) and same major compute capability, + * then on success all allocations from \p peerContext will immediately be + * accessible by the current context. See \ref CUDA_UNIFIED for additional * details. * - * Note that access granted by this call is unidirectional and that in order to access - * memory from the current context in \p peerContext, a separate symmetric call - * to ::cuCtxEnablePeerAccess() is required. + * Note that access granted by this call is unidirectional and that in order to + * access memory from the current context in \p peerContext, a separate + * symmetric call to ::cuCtxEnablePeerAccess() is required. * * There is a system-wide maximum of eight peer connections per device. * - * Returns ::CUDA_ERROR_PEER_ACCESS_UNSUPPORTED if ::cuDeviceCanAccessPeer() indicates - * that the ::CUdevice of the current context cannot directly access memory - * from the ::CUdevice of \p peerContext. + * Returns ::CUDA_ERROR_PEER_ACCESS_UNSUPPORTED if ::cuDeviceCanAccessPeer() + * indicates that the ::CUdevice of the current context cannot directly access + * memory from the ::CUdevice of \p peerContext. * * Returns ::CUDA_ERROR_PEER_ACCESS_ALREADY_ENABLED if direct access of * \p peerContext from the current context has already been enabled. @@ -14024,13 +14826,14 @@ CUresult CUDAAPI cuDeviceCanAccessPeer(int *canAccessPeer, CUdevice dev, CUdevic * Returns ::CUDA_ERROR_TOO_MANY_PEERS if direct peer access is not possible * because hardware resources required for peer access have been exhausted. * - * Returns ::CUDA_ERROR_INVALID_CONTEXT if there is no current context, \p peerContext - * is not a valid context, or if the current context is \p peerContext. + * Returns ::CUDA_ERROR_INVALID_CONTEXT if there is no current context, \p + * peerContext is not a valid context, or if the current context is \p + * peerContext. * * Returns ::CUDA_ERROR_INVALID_VALUE if \p Flags is not 0. * - * \param peerContext - Peer context to enable direct access to from the current context - * \param Flags - Reserved for future use and must be set to 0 + * \param peerContext - Peer context to enable direct access to from the current + * context \param Flags - Reserved for future use and must be set to 0 * * \return * ::CUDA_SUCCESS, @@ -14048,7 +14851,8 @@ CUresult CUDAAPI cuDeviceCanAccessPeer(int *canAccessPeer, CUdevice dev, CUdevic * ::cuCtxDisablePeerAccess, * ::cudaDeviceEnablePeerAccess */ -CUresult CUDAAPI cuCtxEnablePeerAccess(CUcontext peerContext, unsigned int Flags); +CUresult CUDAAPI cuCtxEnablePeerAccess(CUcontext peerContext, + unsigned int Flags); /** * \brief Disables direct access to memory allocations in a peer context and @@ -14089,21 +14893,22 @@ CUresult CUDAAPI cuCtxDisablePeerAccess(CUcontext peerContext); * - ::CU_DEVICE_P2P_ATTRIBUTE_PERFORMANCE_RANK: A relative value indicating the * performance of the link between two devices. * - ::CU_DEVICE_P2P_ATTRIBUTE_ACCESS_SUPPORTED P2P: 1 if P2P Access is enable. - * - ::CU_DEVICE_P2P_ATTRIBUTE_NATIVE_ATOMIC_SUPPORTED: 1 if Atomic operations over - * the link are supported. + * - ::CU_DEVICE_P2P_ATTRIBUTE_NATIVE_ATOMIC_SUPPORTED: 1 if Atomic operations + * over the link are supported. * - ::CU_DEVICE_P2P_ATTRIBUTE_CUDA_ARRAY_ACCESS_SUPPORTED: 1 if cudaArray can * be accessed over the link. * - * Returns ::CUDA_ERROR_INVALID_DEVICE if \p srcDevice or \p dstDevice are not valid - * or if they represent the same device. + * Returns ::CUDA_ERROR_INVALID_DEVICE if \p srcDevice or \p dstDevice are not + * valid or if they represent the same device. * - * Returns ::CUDA_ERROR_INVALID_VALUE if \p attrib is not valid or if \p value is - * a null pointer. + * Returns ::CUDA_ERROR_INVALID_VALUE if \p attrib is not valid or if \p value + * is a null pointer. * * \param value - Returned value of the requested attribute - * \param attrib - The requested attribute of the link between \p srcDevice and \p dstDevice. - * \param srcDevice - The source device of the target link. - * \param dstDevice - The destination device of the target link. + * \param attrib - The requested attribute of the link between \p + * srcDevice and \p dstDevice. \param srcDevice - The source device of the + * target link. \param dstDevice - The destination device of the target + * link. * * \return * ::CUDA_SUCCESS, @@ -14119,7 +14924,10 @@ CUresult CUDAAPI cuCtxDisablePeerAccess(CUcontext peerContext); * ::cuDeviceCanAccessPeer, * ::cudaDeviceGetP2PAttribute */ -CUresult CUDAAPI cuDeviceGetP2PAttribute(int* value, CUdevice_P2PAttribute attrib, CUdevice srcDevice, CUdevice dstDevice); +CUresult CUDAAPI cuDeviceGetP2PAttribute(int *value, + CUdevice_P2PAttribute attrib, + CUdevice srcDevice, + CUdevice dstDevice); #endif /* __CUDA_API_VERSION >= 8000 */ @@ -14168,12 +14976,13 @@ CUresult CUDAAPI cuDeviceGetP2PAttribute(int* value, CUdevice_P2PAttribute attri CUresult CUDAAPI cuGraphicsUnregisterResource(CUgraphicsResource resource); /** - * \brief Get an array through which to access a subresource of a mapped graphics resource. + * \brief Get an array through which to access a subresource of a mapped + * graphics resource. * * Returns in \p *pArray an array through which the subresource of the mapped * graphics resource \p resource which corresponds to array index \p arrayIndex - * and mipmap level \p mipLevel may be accessed. The value set in \p *pArray may - * change every time that \p resource is mapped. + * and mipmap level \p mipLevel may be accessed. The value set in \p *pArray + * may change every time that \p resource is mapped. * * If \p resource is not a texture then it cannot be accessed via an array and * ::CUDA_ERROR_NOT_MAPPED_AS_ARRAY is returned. @@ -14183,8 +14992,8 @@ CUresult CUDAAPI cuGraphicsUnregisterResource(CUgraphicsResource resource); * ::CUDA_ERROR_INVALID_VALUE is returned. * If \p resource is not mapped then ::CUDA_ERROR_NOT_MAPPED is returned. * - * \param pArray - Returned array through which a subresource of \p resource may be accessed - * \param resource - Mapped resource to access + * \param pArray - Returned array through which a subresource of \p + * resource may be accessed \param resource - Mapped resource to access * \param arrayIndex - Array index for array textures or cubemap face * index as defined by ::CUarray_cubemap_face for * cubemap textures for the subresource to access @@ -14205,23 +15014,27 @@ CUresult CUDAAPI cuGraphicsUnregisterResource(CUgraphicsResource resource); * ::cuGraphicsResourceGetMappedPointer, * ::cudaGraphicsSubResourceGetMappedArray */ -CUresult CUDAAPI cuGraphicsSubResourceGetMappedArray(CUarray *pArray, CUgraphicsResource resource, unsigned int arrayIndex, unsigned int mipLevel); +CUresult CUDAAPI cuGraphicsSubResourceGetMappedArray( + CUarray *pArray, CUgraphicsResource resource, unsigned int arrayIndex, + unsigned int mipLevel); #if __CUDA_API_VERSION >= 5000 /** - * \brief Get a mipmapped array through which to access a mapped graphics resource. + * \brief Get a mipmapped array through which to access a mapped graphics + * resource. * - * Returns in \p *pMipmappedArray a mipmapped array through which the mapped graphics - * resource \p resource. The value set in \p *pMipmappedArray may change every time - * that \p resource is mapped. + * Returns in \p *pMipmappedArray a mipmapped array through which the mapped + * graphics resource \p resource. The value set in \p *pMipmappedArray may + * change every time that \p resource is mapped. * - * If \p resource is not a texture then it cannot be accessed via a mipmapped array and + * If \p resource is not a texture then it cannot be accessed via a mipmapped + * array and * ::CUDA_ERROR_NOT_MAPPED_AS_ARRAY is returned. * If \p resource is not mapped then ::CUDA_ERROR_NOT_MAPPED is returned. * - * \param pMipmappedArray - Returned mipmapped array through which \p resource may be accessed - * \param resource - Mapped resource to access + * \param pMipmappedArray - Returned mipmapped array through which \p resource + * may be accessed \param resource - Mapped resource to access * * \return * ::CUDA_SUCCESS, @@ -14238,26 +15051,29 @@ CUresult CUDAAPI cuGraphicsSubResourceGetMappedArray(CUarray *pArray, CUgraphics * ::cuGraphicsResourceGetMappedPointer, * ::cudaGraphicsResourceGetMappedMipmappedArray */ -CUresult CUDAAPI cuGraphicsResourceGetMappedMipmappedArray(CUmipmappedArray *pMipmappedArray, CUgraphicsResource resource); +CUresult CUDAAPI cuGraphicsResourceGetMappedMipmappedArray( + CUmipmappedArray *pMipmappedArray, CUgraphicsResource resource); #endif /* __CUDA_API_VERSION >= 5000 */ #if __CUDA_API_VERSION >= 3020 /** - * \brief Get a device pointer through which to access a mapped graphics resource. + * \brief Get a device pointer through which to access a mapped graphics + * resource. * * Returns in \p *pDevPtr a pointer through which the mapped graphics resource * \p resource may be accessed. - * Returns in \p pSize the size of the memory in bytes which may be accessed from that pointer. - * The value set in \p pPointer may change every time that \p resource is mapped. + * Returns in \p pSize the size of the memory in bytes which may be accessed + * from that pointer. The value set in \p pPointer may change every time that \p + * resource is mapped. * * If \p resource is not a buffer then it cannot be accessed via a pointer and * ::CUDA_ERROR_NOT_MAPPED_AS_POINTER is returned. * If \p resource is not mapped then ::CUDA_ERROR_NOT_MAPPED is returned. * * - * \param pDevPtr - Returned pointer through which \p resource may be accessed - * \param pSize - Returned size of the buffer accessible starting at \p *pPointer - * \param resource - Mapped resource to access + * \param pDevPtr - Returned pointer through which \p resource may be + * accessed \param pSize - Returned size of the buffer accessible starting + * at \p *pPointer \param resource - Mapped resource to access * * \return * ::CUDA_SUCCESS, @@ -14275,7 +15091,8 @@ CUresult CUDAAPI cuGraphicsResourceGetMappedMipmappedArray(CUmipmappedArray *pMi * ::cuGraphicsSubResourceGetMappedArray, * ::cudaGraphicsResourceGetMappedPointer */ -CUresult CUDAAPI cuGraphicsResourceGetMappedPointer(CUdeviceptr *pDevPtr, size_t *pSize, CUgraphicsResource resource); +CUresult CUDAAPI cuGraphicsResourceGetMappedPointer( + CUdeviceptr *pDevPtr, size_t *pSize, CUgraphicsResource resource); #endif /* __CUDA_API_VERSION >= 3020 */ /** @@ -14289,7 +15106,8 @@ CUresult CUDAAPI cuGraphicsResourceGetMappedPointer(CUdeviceptr *pDevPtr, size_t * - ::CU_GRAPHICS_MAP_RESOURCE_FLAGS_NONE: Specifies no hints about how this * resource will be used. It is therefore assumed that this resource will be * read from and written to by CUDA kernels. This is the default value. - * - ::CU_GRAPHICS_MAP_RESOURCE_FLAGS_READONLY: Specifies that CUDA kernels which + * - ::CU_GRAPHICS_MAP_RESOURCE_FLAGS_READONLY: Specifies that CUDA kernels + which * access this resource will not write to this resource. * - ::CU_GRAPHICS_MAP_RESOURCE_FLAGS_WRITEDISCARD: Specifies that CUDA kernels * which access this resource will not read from this resource and will @@ -14298,7 +15116,8 @@ CUresult CUDAAPI cuGraphicsResourceGetMappedPointer(CUdeviceptr *pDevPtr, size_t * * If \p resource is presently mapped for access by CUDA then * ::CUDA_ERROR_ALREADY_MAPPED is returned. - * If \p flags is not one of the above values then ::CUDA_ERROR_INVALID_VALUE is returned. + * If \p flags is not one of the above values then ::CUDA_ERROR_INVALID_VALUE is + returned. * * \param resource - Registered resource to set flags for * \param flags - Parameters for resource mapping @@ -14317,7 +15136,8 @@ CUresult CUDAAPI cuGraphicsResourceGetMappedPointer(CUdeviceptr *pDevPtr, size_t * ::cuGraphicsMapResources, * ::cudaGraphicsResourceSetMapFlags */ -CUresult CUDAAPI cuGraphicsResourceSetMapFlags(CUgraphicsResource resource, unsigned int flags); +CUresult CUDAAPI cuGraphicsResourceSetMapFlags(CUgraphicsResource resource, + unsigned int flags); /** * \brief Map graphics resources for access by CUDA @@ -14330,11 +15150,12 @@ CUresult CUDAAPI cuGraphicsResourceSetMapFlags(CUgraphicsResource resource, unsi * application does so, the results are undefined. * * This function provides the synchronization guarantee that any graphics calls - * issued before ::cuGraphicsMapResources() will complete before any subsequent CUDA - * work issued in \p stream begins. + * issued before ::cuGraphicsMapResources() will complete before any subsequent + * CUDA work issued in \p stream begins. * - * If \p resources includes any duplicate entries then ::CUDA_ERROR_INVALID_HANDLE is returned. - * If any of \p resources are presently mapped for access by CUDA then ::CUDA_ERROR_ALREADY_MAPPED is returned. + * If \p resources includes any duplicate entries then + * ::CUDA_ERROR_INVALID_HANDLE is returned. If any of \p resources are presently + * mapped for access by CUDA then ::CUDA_ERROR_ALREADY_MAPPED is returned. * * \param count - Number of resources to map * \param resources - Resources to map for CUDA usage @@ -14357,7 +15178,9 @@ CUresult CUDAAPI cuGraphicsResourceSetMapFlags(CUgraphicsResource resource, unsi * ::cuGraphicsUnmapResources, * ::cudaGraphicsMapResources */ -CUresult CUDAAPI cuGraphicsMapResources(unsigned int count, CUgraphicsResource *resources, CUstream hStream); +CUresult CUDAAPI cuGraphicsMapResources(unsigned int count, + CUgraphicsResource *resources, + CUstream hStream); /** * \brief Unmap graphics resources. @@ -14367,13 +15190,14 @@ CUresult CUDAAPI cuGraphicsMapResources(unsigned int count, CUgraphicsResource * * Once unmapped, the resources in \p resources may not be accessed by CUDA * until they are mapped again. * - * This function provides the synchronization guarantee that any CUDA work issued - * in \p stream before ::cuGraphicsUnmapResources() will complete before any - * subsequently issued graphics work begins. + * This function provides the synchronization guarantee that any CUDA work + * issued in \p stream before ::cuGraphicsUnmapResources() will complete before + * any subsequently issued graphics work begins. * * - * If \p resources includes any duplicate entries then ::CUDA_ERROR_INVALID_HANDLE is returned. - * If any of \p resources are not presently mapped for access by CUDA then ::CUDA_ERROR_NOT_MAPPED is returned. + * If \p resources includes any duplicate entries then + * ::CUDA_ERROR_INVALID_HANDLE is returned. If any of \p resources are not + * presently mapped for access by CUDA then ::CUDA_ERROR_NOT_MAPPED is returned. * * \param count - Number of resources to unmap * \param resources - Resources to unmap @@ -14394,264 +15218,318 @@ CUresult CUDAAPI cuGraphicsMapResources(unsigned int count, CUgraphicsResource * * ::cuGraphicsMapResources, * ::cudaGraphicsUnmapResources */ -CUresult CUDAAPI cuGraphicsUnmapResources(unsigned int count, CUgraphicsResource *resources, CUstream hStream); +CUresult CUDAAPI cuGraphicsUnmapResources(unsigned int count, + CUgraphicsResource *resources, + CUstream hStream); /** @} */ /* END CUDA_GRAPHICS */ -CUresult CUDAAPI cuGetExportTable(const void **ppExportTable, const CUuuid *pExportTableId); - +CUresult CUDAAPI cuGetExportTable(const void **ppExportTable, + const CUuuid *pExportTableId); /** * CUDA API versioning support */ #if defined(__CUDA_API_VERSION_INTERNAL) - #undef cuMemHostRegister - #undef cuGraphicsResourceSetMapFlags - #undef cuLinkCreate - #undef cuLinkAddData - #undef cuLinkAddFile - #undef cuDeviceTotalMem - #undef cuCtxCreate - #undef cuModuleGetGlobal - #undef cuMemGetInfo - #undef cuMemAlloc - #undef cuMemAllocPitch - #undef cuMemFree - #undef cuMemGetAddressRange - #undef cuMemAllocHost - #undef cuMemHostGetDevicePointer - #undef cuMemcpyHtoD - #undef cuMemcpyDtoH - #undef cuMemcpyDtoD - #undef cuMemcpyDtoA - #undef cuMemcpyAtoD - #undef cuMemcpyHtoA - #undef cuMemcpyAtoH - #undef cuMemcpyAtoA - #undef cuMemcpyHtoAAsync - #undef cuMemcpyAtoHAsync - #undef cuMemcpy2D - #undef cuMemcpy2DUnaligned - #undef cuMemcpy3D - #undef cuMemcpyHtoDAsync - #undef cuMemcpyDtoHAsync - #undef cuMemcpyDtoDAsync - #undef cuMemcpy2DAsync - #undef cuMemcpy3DAsync - #undef cuMemsetD8 - #undef cuMemsetD16 - #undef cuMemsetD32 - #undef cuMemsetD2D8 - #undef cuMemsetD2D16 - #undef cuMemsetD2D32 - #undef cuArrayCreate - #undef cuArrayGetDescriptor - #undef cuArray3DCreate - #undef cuArray3DGetDescriptor - #undef cuTexRefSetAddress - #undef cuTexRefSetAddress2D - #undef cuTexRefGetAddress - #undef cuGraphicsResourceGetMappedPointer - #undef cuCtxDestroy - #undef cuCtxPopCurrent - #undef cuCtxPushCurrent - #undef cuStreamDestroy - #undef cuEventDestroy - #undef cuMemcpy - #undef cuMemcpyAsync - #undef cuMemcpyPeer - #undef cuMemcpyPeerAsync - #undef cuMemcpy3DPeer - #undef cuMemcpy3DPeerAsync - #undef cuMemsetD8Async - #undef cuMemsetD16Async - #undef cuMemsetD32Async - #undef cuMemsetD2D8Async - #undef cuMemsetD2D16Async - #undef cuMemsetD2D32Async - #undef cuStreamGetPriority - #undef cuStreamGetFlags - #undef cuStreamGetCtx - #undef cuStreamWaitEvent - #undef cuStreamAddCallback - #undef cuStreamAttachMemAsync - #undef cuStreamQuery - #undef cuStreamSynchronize - #undef cuEventRecord - #undef cuLaunchKernel - #undef cuLaunchHostFunc - #undef cuGraphicsMapResources - #undef cuGraphicsUnmapResources - #undef cuStreamWriteValue32 - #undef cuStreamWaitValue32 - #undef cuStreamWriteValue64 - #undef cuStreamWaitValue64 - #undef cuStreamBatchMemOp - #undef cuMemPrefetchAsync - #undef cuLaunchCooperativeKernel - #undef cuSignalExternalSemaphoresAsync - #undef cuWaitExternalSemaphoresAsync - #undef cuStreamBeginCapture - #undef cuStreamEndCapture - #undef cuStreamIsCapturing - #undef cuStreamGetCaptureInfo - #undef cuGraphLaunch +#undef cuMemHostRegister +#undef cuGraphicsResourceSetMapFlags +#undef cuLinkCreate +#undef cuLinkAddData +#undef cuLinkAddFile +#undef cuDeviceTotalMem +#undef cuCtxCreate +#undef cuModuleGetGlobal +#undef cuMemGetInfo +#undef cuMemAlloc +#undef cuMemAllocPitch +#undef cuMemFree +#undef cuMemGetAddressRange +#undef cuMemAllocHost +#undef cuMemHostGetDevicePointer +#undef cuMemcpyHtoD +#undef cuMemcpyDtoH +#undef cuMemcpyDtoD +#undef cuMemcpyDtoA +#undef cuMemcpyAtoD +#undef cuMemcpyHtoA +#undef cuMemcpyAtoH +#undef cuMemcpyAtoA +#undef cuMemcpyHtoAAsync +#undef cuMemcpyAtoHAsync +#undef cuMemcpy2D +#undef cuMemcpy2DUnaligned +#undef cuMemcpy3D +#undef cuMemcpyHtoDAsync +#undef cuMemcpyDtoHAsync +#undef cuMemcpyDtoDAsync +#undef cuMemcpy2DAsync +#undef cuMemcpy3DAsync +#undef cuMemsetD8 +#undef cuMemsetD16 +#undef cuMemsetD32 +#undef cuMemsetD2D8 +#undef cuMemsetD2D16 +#undef cuMemsetD2D32 +#undef cuArrayCreate +#undef cuArrayGetDescriptor +#undef cuArray3DCreate +#undef cuArray3DGetDescriptor +#undef cuTexRefSetAddress +#undef cuTexRefSetAddress2D +#undef cuTexRefGetAddress +#undef cuGraphicsResourceGetMappedPointer +#undef cuCtxDestroy +#undef cuCtxPopCurrent +#undef cuCtxPushCurrent +#undef cuStreamDestroy +#undef cuEventDestroy +#undef cuMemcpy +#undef cuMemcpyAsync +#undef cuMemcpyPeer +#undef cuMemcpyPeerAsync +#undef cuMemcpy3DPeer +#undef cuMemcpy3DPeerAsync +#undef cuMemsetD8Async +#undef cuMemsetD16Async +#undef cuMemsetD32Async +#undef cuMemsetD2D8Async +#undef cuMemsetD2D16Async +#undef cuMemsetD2D32Async +#undef cuStreamGetPriority +#undef cuStreamGetFlags +#undef cuStreamGetCtx +#undef cuStreamWaitEvent +#undef cuStreamAddCallback +#undef cuStreamAttachMemAsync +#undef cuStreamQuery +#undef cuStreamSynchronize +#undef cuEventRecord +#undef cuLaunchKernel +#undef cuLaunchHostFunc +#undef cuGraphicsMapResources +#undef cuGraphicsUnmapResources +#undef cuStreamWriteValue32 +#undef cuStreamWaitValue32 +#undef cuStreamWriteValue64 +#undef cuStreamWaitValue64 +#undef cuStreamBatchMemOp +#undef cuMemPrefetchAsync +#undef cuLaunchCooperativeKernel +#undef cuSignalExternalSemaphoresAsync +#undef cuWaitExternalSemaphoresAsync +#undef cuStreamBeginCapture +#undef cuStreamEndCapture +#undef cuStreamIsCapturing +#undef cuStreamGetCaptureInfo +#undef cuGraphLaunch #endif /* __CUDA_API_VERSION_INTERNAL */ -#if defined(__CUDA_API_VERSION_INTERNAL) || (__CUDA_API_VERSION >= 4000 && __CUDA_API_VERSION < 6050) -CUresult CUDAAPI cuMemHostRegister(void *p, size_t bytesize, unsigned int Flags); -#endif /* defined(__CUDA_API_VERSION_INTERNAL) || (__CUDA_API_VERSION >= 4000 && __CUDA_API_VERSION < 6050) */ +#if defined(__CUDA_API_VERSION_INTERNAL) || \ + (__CUDA_API_VERSION >= 4000 && __CUDA_API_VERSION < 6050) +CUresult CUDAAPI cuMemHostRegister(void *p, size_t bytesize, + unsigned int Flags); +#endif /* defined(__CUDA_API_VERSION_INTERNAL) || (__CUDA_API_VERSION >= 4000 \ + && __CUDA_API_VERSION < 6050) */ #if defined(__CUDA_API_VERSION_INTERNAL) || __CUDA_API_VERSION < 6050 -CUresult CUDAAPI cuGraphicsResourceSetMapFlags(CUgraphicsResource resource, unsigned int flags); +CUresult CUDAAPI cuGraphicsResourceSetMapFlags(CUgraphicsResource resource, + unsigned int flags); #endif /* defined(__CUDA_API_VERSION_INTERNAL) || __CUDA_API_VERSION < 6050 */ -#if defined(__CUDA_API_VERSION_INTERNAL) || (__CUDA_API_VERSION >= 5050 && __CUDA_API_VERSION < 6050) -CUresult CUDAAPI cuLinkCreate(unsigned int numOptions, CUjit_option *options, void **optionValues, CUlinkState *stateOut); -CUresult CUDAAPI cuLinkAddData(CUlinkState state, CUjitInputType type, void *data, size_t size, const char *name, - unsigned int numOptions, CUjit_option *options, void **optionValues); -CUresult CUDAAPI cuLinkAddFile(CUlinkState state, CUjitInputType type, const char *path, - unsigned int numOptions, CUjit_option *options, void **optionValues); -#endif /* __CUDA_API_VERSION_INTERNAL || (__CUDA_API_VERSION >= 5050 && __CUDA_API_VERSION < 6050) */ - -#if defined(__CUDA_API_VERSION_INTERNAL) || (__CUDA_API_VERSION >= 3020 && __CUDA_API_VERSION < 4010) -CUresult CUDAAPI cuTexRefSetAddress2D_v2(CUtexref hTexRef, const CUDA_ARRAY_DESCRIPTOR *desc, CUdeviceptr dptr, size_t Pitch); -#endif /* __CUDA_API_VERSION_INTERNAL || (__CUDA_API_VERSION >= 3020 && __CUDA_API_VERSION < 4010) */ +#if defined(__CUDA_API_VERSION_INTERNAL) || \ + (__CUDA_API_VERSION >= 5050 && __CUDA_API_VERSION < 6050) +CUresult CUDAAPI cuLinkCreate(unsigned int numOptions, CUjit_option *options, + void **optionValues, CUlinkState *stateOut); +CUresult CUDAAPI cuLinkAddData(CUlinkState state, CUjitInputType type, + void *data, size_t size, const char *name, + unsigned int numOptions, CUjit_option *options, + void **optionValues); +CUresult CUDAAPI cuLinkAddFile(CUlinkState state, CUjitInputType type, + const char *path, unsigned int numOptions, + CUjit_option *options, void **optionValues); +#endif /* __CUDA_API_VERSION_INTERNAL || (__CUDA_API_VERSION >= 5050 && \ + __CUDA_API_VERSION < 6050) */ + +#if defined(__CUDA_API_VERSION_INTERNAL) || \ + (__CUDA_API_VERSION >= 3020 && __CUDA_API_VERSION < 4010) +CUresult CUDAAPI cuTexRefSetAddress2D_v2(CUtexref hTexRef, + const CUDA_ARRAY_DESCRIPTOR *desc, + CUdeviceptr dptr, size_t Pitch); +#endif /* __CUDA_API_VERSION_INTERNAL || (__CUDA_API_VERSION >= 3020 && \ + __CUDA_API_VERSION < 4010) */ /** * CUDA API made obselete at API version 3020 */ #if defined(__CUDA_API_VERSION_INTERNAL) - #define CUdeviceptr CUdeviceptr_v1 - #define CUDA_MEMCPY2D_st CUDA_MEMCPY2D_v1_st - #define CUDA_MEMCPY2D CUDA_MEMCPY2D_v1 - #define CUDA_MEMCPY3D_st CUDA_MEMCPY3D_v1_st - #define CUDA_MEMCPY3D CUDA_MEMCPY3D_v1 - #define CUDA_ARRAY_DESCRIPTOR_st CUDA_ARRAY_DESCRIPTOR_v1_st - #define CUDA_ARRAY_DESCRIPTOR CUDA_ARRAY_DESCRIPTOR_v1 - #define CUDA_ARRAY3D_DESCRIPTOR_st CUDA_ARRAY3D_DESCRIPTOR_v1_st - #define CUDA_ARRAY3D_DESCRIPTOR CUDA_ARRAY3D_DESCRIPTOR_v1 +#define CUdeviceptr CUdeviceptr_v1 +#define CUDA_MEMCPY2D_st CUDA_MEMCPY2D_v1_st +#define CUDA_MEMCPY2D CUDA_MEMCPY2D_v1 +#define CUDA_MEMCPY3D_st CUDA_MEMCPY3D_v1_st +#define CUDA_MEMCPY3D CUDA_MEMCPY3D_v1 +#define CUDA_ARRAY_DESCRIPTOR_st CUDA_ARRAY_DESCRIPTOR_v1_st +#define CUDA_ARRAY_DESCRIPTOR CUDA_ARRAY_DESCRIPTOR_v1 +#define CUDA_ARRAY3D_DESCRIPTOR_st CUDA_ARRAY3D_DESCRIPTOR_v1_st +#define CUDA_ARRAY3D_DESCRIPTOR CUDA_ARRAY3D_DESCRIPTOR_v1 #endif /* CUDA_FORCE_LEGACY32_INTERNAL */ #if defined(__CUDA_API_VERSION_INTERNAL) || __CUDA_API_VERSION < 3020 typedef unsigned int CUdeviceptr; -typedef struct CUDA_MEMCPY2D_st -{ - unsigned int srcXInBytes; /**< Source X in bytes */ - unsigned int srcY; /**< Source Y */ - CUmemorytype srcMemoryType; /**< Source memory type (host, device, array) */ - const void *srcHost; /**< Source host pointer */ - CUdeviceptr srcDevice; /**< Source device pointer */ - CUarray srcArray; /**< Source array reference */ - unsigned int srcPitch; /**< Source pitch (ignored when src is array) */ - - unsigned int dstXInBytes; /**< Destination X in bytes */ - unsigned int dstY; /**< Destination Y */ - CUmemorytype dstMemoryType; /**< Destination memory type (host, device, array) */ - void *dstHost; /**< Destination host pointer */ - CUdeviceptr dstDevice; /**< Destination device pointer */ - CUarray dstArray; /**< Destination array reference */ - unsigned int dstPitch; /**< Destination pitch (ignored when dst is array) */ - - unsigned int WidthInBytes; /**< Width of 2D memory copy in bytes */ - unsigned int Height; /**< Height of 2D memory copy */ +typedef struct CUDA_MEMCPY2D_st { + unsigned int srcXInBytes; /**< Source X in bytes */ + unsigned int srcY; /**< Source Y */ + CUmemorytype srcMemoryType; /**< Source memory type (host, device, array) */ + const void *srcHost; /**< Source host pointer */ + CUdeviceptr srcDevice; /**< Source device pointer */ + CUarray srcArray; /**< Source array reference */ + unsigned int srcPitch; /**< Source pitch (ignored when src is array) */ + + unsigned int dstXInBytes; /**< Destination X in bytes */ + unsigned int dstY; /**< Destination Y */ + CUmemorytype + dstMemoryType; /**< Destination memory type (host, device, array) */ + void *dstHost; /**< Destination host pointer */ + CUdeviceptr dstDevice; /**< Destination device pointer */ + CUarray dstArray; /**< Destination array reference */ + unsigned int dstPitch; /**< Destination pitch (ignored when dst is array) */ + + unsigned int WidthInBytes; /**< Width of 2D memory copy in bytes */ + unsigned int Height; /**< Height of 2D memory copy */ } CUDA_MEMCPY2D; -typedef struct CUDA_MEMCPY3D_st -{ - unsigned int srcXInBytes; /**< Source X in bytes */ - unsigned int srcY; /**< Source Y */ - unsigned int srcZ; /**< Source Z */ - unsigned int srcLOD; /**< Source LOD */ - CUmemorytype srcMemoryType; /**< Source memory type (host, device, array) */ - const void *srcHost; /**< Source host pointer */ - CUdeviceptr srcDevice; /**< Source device pointer */ - CUarray srcArray; /**< Source array reference */ - void *reserved0; /**< Must be NULL */ - unsigned int srcPitch; /**< Source pitch (ignored when src is array) */ - unsigned int srcHeight; /**< Source height (ignored when src is array; may be 0 if Depth==1) */ - - unsigned int dstXInBytes; /**< Destination X in bytes */ - unsigned int dstY; /**< Destination Y */ - unsigned int dstZ; /**< Destination Z */ - unsigned int dstLOD; /**< Destination LOD */ - CUmemorytype dstMemoryType; /**< Destination memory type (host, device, array) */ - void *dstHost; /**< Destination host pointer */ - CUdeviceptr dstDevice; /**< Destination device pointer */ - CUarray dstArray; /**< Destination array reference */ - void *reserved1; /**< Must be NULL */ - unsigned int dstPitch; /**< Destination pitch (ignored when dst is array) */ - unsigned int dstHeight; /**< Destination height (ignored when dst is array; may be 0 if Depth==1) */ - - unsigned int WidthInBytes; /**< Width of 3D memory copy in bytes */ - unsigned int Height; /**< Height of 3D memory copy */ - unsigned int Depth; /**< Depth of 3D memory copy */ +typedef struct CUDA_MEMCPY3D_st { + unsigned int srcXInBytes; /**< Source X in bytes */ + unsigned int srcY; /**< Source Y */ + unsigned int srcZ; /**< Source Z */ + unsigned int srcLOD; /**< Source LOD */ + CUmemorytype srcMemoryType; /**< Source memory type (host, device, array) */ + const void *srcHost; /**< Source host pointer */ + CUdeviceptr srcDevice; /**< Source device pointer */ + CUarray srcArray; /**< Source array reference */ + void *reserved0; /**< Must be NULL */ + unsigned int srcPitch; /**< Source pitch (ignored when src is array) */ + unsigned int srcHeight; /**< Source height (ignored when src is array; may be + 0 if Depth==1) */ + + unsigned int dstXInBytes; /**< Destination X in bytes */ + unsigned int dstY; /**< Destination Y */ + unsigned int dstZ; /**< Destination Z */ + unsigned int dstLOD; /**< Destination LOD */ + CUmemorytype + dstMemoryType; /**< Destination memory type (host, device, array) */ + void *dstHost; /**< Destination host pointer */ + CUdeviceptr dstDevice; /**< Destination device pointer */ + CUarray dstArray; /**< Destination array reference */ + void *reserved1; /**< Must be NULL */ + unsigned int dstPitch; /**< Destination pitch (ignored when dst is array) */ + unsigned int dstHeight; /**< Destination height (ignored when dst is array; + may be 0 if Depth==1) */ + + unsigned int WidthInBytes; /**< Width of 3D memory copy in bytes */ + unsigned int Height; /**< Height of 3D memory copy */ + unsigned int Depth; /**< Depth of 3D memory copy */ } CUDA_MEMCPY3D; -typedef struct CUDA_ARRAY_DESCRIPTOR_st -{ - unsigned int Width; /**< Width of array */ - unsigned int Height; /**< Height of array */ +typedef struct CUDA_ARRAY_DESCRIPTOR_st { + unsigned int Width; /**< Width of array */ + unsigned int Height; /**< Height of array */ - CUarray_format Format; /**< Array format */ - unsigned int NumChannels; /**< Channels per array element */ + CUarray_format Format; /**< Array format */ + unsigned int NumChannels; /**< Channels per array element */ } CUDA_ARRAY_DESCRIPTOR; -typedef struct CUDA_ARRAY3D_DESCRIPTOR_st -{ - unsigned int Width; /**< Width of 3D array */ - unsigned int Height; /**< Height of 3D array */ - unsigned int Depth; /**< Depth of 3D array */ +typedef struct CUDA_ARRAY3D_DESCRIPTOR_st { + unsigned int Width; /**< Width of 3D array */ + unsigned int Height; /**< Height of 3D array */ + unsigned int Depth; /**< Depth of 3D array */ - CUarray_format Format; /**< Array format */ - unsigned int NumChannels; /**< Channels per array element */ - unsigned int Flags; /**< Flags */ + CUarray_format Format; /**< Array format */ + unsigned int NumChannels; /**< Channels per array element */ + unsigned int Flags; /**< Flags */ } CUDA_ARRAY3D_DESCRIPTOR; CUresult CUDAAPI cuDeviceTotalMem(unsigned int *bytes, CUdevice dev); CUresult CUDAAPI cuCtxCreate(CUcontext *pctx, unsigned int flags, CUdevice dev); -CUresult CUDAAPI cuModuleGetGlobal(CUdeviceptr *dptr, unsigned int *bytes, CUmodule hmod, const char *name); +CUresult CUDAAPI cuModuleGetGlobal(CUdeviceptr *dptr, unsigned int *bytes, + CUmodule hmod, const char *name); CUresult CUDAAPI cuMemGetInfo(unsigned int *free, unsigned int *total); CUresult CUDAAPI cuMemAlloc(CUdeviceptr *dptr, unsigned int bytesize); -CUresult CUDAAPI cuMemAllocPitch(CUdeviceptr *dptr, unsigned int *pPitch, unsigned int WidthInBytes, unsigned int Height, unsigned int ElementSizeBytes); +CUresult CUDAAPI cuMemAllocPitch(CUdeviceptr *dptr, unsigned int *pPitch, + unsigned int WidthInBytes, unsigned int Height, + unsigned int ElementSizeBytes); CUresult CUDAAPI cuMemFree(CUdeviceptr dptr); -CUresult CUDAAPI cuMemGetAddressRange(CUdeviceptr *pbase, unsigned int *psize, CUdeviceptr dptr); +CUresult CUDAAPI cuMemGetAddressRange(CUdeviceptr *pbase, unsigned int *psize, + CUdeviceptr dptr); CUresult CUDAAPI cuMemAllocHost(void **pp, unsigned int bytesize); -CUresult CUDAAPI cuMemHostGetDevicePointer(CUdeviceptr *pdptr, void *p, unsigned int Flags); -CUresult CUDAAPI cuMemcpyHtoD(CUdeviceptr dstDevice, const void *srcHost, unsigned int ByteCount); -CUresult CUDAAPI cuMemcpyDtoH(void *dstHost, CUdeviceptr srcDevice, unsigned int ByteCount); -CUresult CUDAAPI cuMemcpyDtoD(CUdeviceptr dstDevice, CUdeviceptr srcDevice, unsigned int ByteCount); -CUresult CUDAAPI cuMemcpyDtoA(CUarray dstArray, unsigned int dstOffset, CUdeviceptr srcDevice, unsigned int ByteCount); -CUresult CUDAAPI cuMemcpyAtoD(CUdeviceptr dstDevice, CUarray srcArray, unsigned int srcOffset, unsigned int ByteCount); -CUresult CUDAAPI cuMemcpyHtoA(CUarray dstArray, unsigned int dstOffset, const void *srcHost, unsigned int ByteCount); -CUresult CUDAAPI cuMemcpyAtoH(void *dstHost, CUarray srcArray, unsigned int srcOffset, unsigned int ByteCount); -CUresult CUDAAPI cuMemcpyAtoA(CUarray dstArray, unsigned int dstOffset, CUarray srcArray, unsigned int srcOffset, unsigned int ByteCount); -CUresult CUDAAPI cuMemcpyHtoAAsync(CUarray dstArray, unsigned int dstOffset, const void *srcHost, unsigned int ByteCount, CUstream hStream); -CUresult CUDAAPI cuMemcpyAtoHAsync(void *dstHost, CUarray srcArray, unsigned int srcOffset, unsigned int ByteCount, CUstream hStream); +CUresult CUDAAPI cuMemHostGetDevicePointer(CUdeviceptr *pdptr, void *p, + unsigned int Flags); +CUresult CUDAAPI cuMemcpyHtoD(CUdeviceptr dstDevice, const void *srcHost, + unsigned int ByteCount); +CUresult CUDAAPI cuMemcpyDtoH(void *dstHost, CUdeviceptr srcDevice, + unsigned int ByteCount); +CUresult CUDAAPI cuMemcpyDtoD(CUdeviceptr dstDevice, CUdeviceptr srcDevice, + unsigned int ByteCount); +CUresult CUDAAPI cuMemcpyDtoA(CUarray dstArray, unsigned int dstOffset, + CUdeviceptr srcDevice, unsigned int ByteCount); +CUresult CUDAAPI cuMemcpyAtoD(CUdeviceptr dstDevice, CUarray srcArray, + unsigned int srcOffset, unsigned int ByteCount); +CUresult CUDAAPI cuMemcpyHtoA(CUarray dstArray, unsigned int dstOffset, + const void *srcHost, unsigned int ByteCount); +CUresult CUDAAPI cuMemcpyAtoH(void *dstHost, CUarray srcArray, + unsigned int srcOffset, unsigned int ByteCount); +CUresult CUDAAPI cuMemcpyAtoA(CUarray dstArray, unsigned int dstOffset, + CUarray srcArray, unsigned int srcOffset, + unsigned int ByteCount); +CUresult CUDAAPI cuMemcpyHtoAAsync(CUarray dstArray, unsigned int dstOffset, + const void *srcHost, unsigned int ByteCount, + CUstream hStream); +CUresult CUDAAPI cuMemcpyAtoHAsync(void *dstHost, CUarray srcArray, + unsigned int srcOffset, + unsigned int ByteCount, CUstream hStream); CUresult CUDAAPI cuMemcpy2D(const CUDA_MEMCPY2D *pCopy); CUresult CUDAAPI cuMemcpy2DUnaligned(const CUDA_MEMCPY2D *pCopy); CUresult CUDAAPI cuMemcpy3D(const CUDA_MEMCPY3D *pCopy); -CUresult CUDAAPI cuMemcpyHtoDAsync(CUdeviceptr dstDevice, const void *srcHost, unsigned int ByteCount, CUstream hStream); -CUresult CUDAAPI cuMemcpyDtoHAsync(void *dstHost, CUdeviceptr srcDevice, unsigned int ByteCount, CUstream hStream); -CUresult CUDAAPI cuMemcpyDtoDAsync(CUdeviceptr dstDevice, CUdeviceptr srcDevice, unsigned int ByteCount, CUstream hStream); +CUresult CUDAAPI cuMemcpyHtoDAsync(CUdeviceptr dstDevice, const void *srcHost, + unsigned int ByteCount, CUstream hStream); +CUresult CUDAAPI cuMemcpyDtoHAsync(void *dstHost, CUdeviceptr srcDevice, + unsigned int ByteCount, CUstream hStream); +CUresult CUDAAPI cuMemcpyDtoDAsync(CUdeviceptr dstDevice, CUdeviceptr srcDevice, + unsigned int ByteCount, CUstream hStream); CUresult CUDAAPI cuMemcpy2DAsync(const CUDA_MEMCPY2D *pCopy, CUstream hStream); CUresult CUDAAPI cuMemcpy3DAsync(const CUDA_MEMCPY3D *pCopy, CUstream hStream); -CUresult CUDAAPI cuMemsetD8(CUdeviceptr dstDevice, unsigned char uc, unsigned int N); -CUresult CUDAAPI cuMemsetD16(CUdeviceptr dstDevice, unsigned short us, unsigned int N); -CUresult CUDAAPI cuMemsetD32(CUdeviceptr dstDevice, unsigned int ui, unsigned int N); -CUresult CUDAAPI cuMemsetD2D8(CUdeviceptr dstDevice, unsigned int dstPitch, unsigned char uc, unsigned int Width, unsigned int Height); -CUresult CUDAAPI cuMemsetD2D16(CUdeviceptr dstDevice, unsigned int dstPitch, unsigned short us, unsigned int Width, unsigned int Height); -CUresult CUDAAPI cuMemsetD2D32(CUdeviceptr dstDevice, unsigned int dstPitch, unsigned int ui, unsigned int Width, unsigned int Height); -CUresult CUDAAPI cuArrayCreate(CUarray *pHandle, const CUDA_ARRAY_DESCRIPTOR *pAllocateArray); -CUresult CUDAAPI cuArrayGetDescriptor(CUDA_ARRAY_DESCRIPTOR *pArrayDescriptor, CUarray hArray); -CUresult CUDAAPI cuArray3DCreate(CUarray *pHandle, const CUDA_ARRAY3D_DESCRIPTOR *pAllocateArray); -CUresult CUDAAPI cuArray3DGetDescriptor(CUDA_ARRAY3D_DESCRIPTOR *pArrayDescriptor, CUarray hArray); -CUresult CUDAAPI cuTexRefSetAddress(unsigned int *ByteOffset, CUtexref hTexRef, CUdeviceptr dptr, unsigned int bytes); -CUresult CUDAAPI cuTexRefSetAddress2D(CUtexref hTexRef, const CUDA_ARRAY_DESCRIPTOR *desc, CUdeviceptr dptr, unsigned int Pitch); +CUresult CUDAAPI cuMemsetD8(CUdeviceptr dstDevice, unsigned char uc, + unsigned int N); +CUresult CUDAAPI cuMemsetD16(CUdeviceptr dstDevice, unsigned short us, + unsigned int N); +CUresult CUDAAPI cuMemsetD32(CUdeviceptr dstDevice, unsigned int ui, + unsigned int N); +CUresult CUDAAPI cuMemsetD2D8(CUdeviceptr dstDevice, unsigned int dstPitch, + unsigned char uc, unsigned int Width, + unsigned int Height); +CUresult CUDAAPI cuMemsetD2D16(CUdeviceptr dstDevice, unsigned int dstPitch, + unsigned short us, unsigned int Width, + unsigned int Height); +CUresult CUDAAPI cuMemsetD2D32(CUdeviceptr dstDevice, unsigned int dstPitch, + unsigned int ui, unsigned int Width, + unsigned int Height); +CUresult CUDAAPI cuArrayCreate(CUarray *pHandle, + const CUDA_ARRAY_DESCRIPTOR *pAllocateArray); +CUresult CUDAAPI cuArrayGetDescriptor(CUDA_ARRAY_DESCRIPTOR *pArrayDescriptor, + CUarray hArray); +CUresult CUDAAPI cuArray3DCreate(CUarray *pHandle, + const CUDA_ARRAY3D_DESCRIPTOR *pAllocateArray); +CUresult CUDAAPI cuArray3DGetDescriptor( + CUDA_ARRAY3D_DESCRIPTOR *pArrayDescriptor, CUarray hArray); +CUresult CUDAAPI cuTexRefSetAddress(unsigned int *ByteOffset, CUtexref hTexRef, + CUdeviceptr dptr, unsigned int bytes); +CUresult CUDAAPI cuTexRefSetAddress2D(CUtexref hTexRef, + const CUDA_ARRAY_DESCRIPTOR *desc, + CUdeviceptr dptr, unsigned int Pitch); CUresult CUDAAPI cuTexRefGetAddress(CUdeviceptr *pdptr, CUtexref hTexRef); -CUresult CUDAAPI cuGraphicsResourceGetMappedPointer(CUdeviceptr *pDevPtr, unsigned int *pSize, CUgraphicsResource resource); +CUresult CUDAAPI cuGraphicsResourceGetMappedPointer( + CUdeviceptr *pDevPtr, unsigned int *pSize, CUgraphicsResource resource); #endif /* __CUDA_API_VERSION_INTERNAL || __CUDA_API_VERSION < 3020 */ #if defined(__CUDA_API_VERSION_INTERNAL) || __CUDA_API_VERSION < 4000 CUresult CUDAAPI cuCtxDestroy(CUcontext ctx); @@ -14661,85 +15539,162 @@ CUresult CUDAAPI cuStreamDestroy(CUstream hStream); CUresult CUDAAPI cuEventDestroy(CUevent hEvent); #endif /* __CUDA_API_VERSION_INTERNAL || __CUDA_API_VERSION < 4000 */ #if defined(__CUDA_API_VERSION_INTERNAL) - #undef CUdeviceptr - #undef CUDA_MEMCPY2D_st - #undef CUDA_MEMCPY2D - #undef CUDA_MEMCPY3D_st - #undef CUDA_MEMCPY3D - #undef CUDA_ARRAY_DESCRIPTOR_st - #undef CUDA_ARRAY_DESCRIPTOR - #undef CUDA_ARRAY3D_DESCRIPTOR_st - #undef CUDA_ARRAY3D_DESCRIPTOR +#undef CUdeviceptr +#undef CUDA_MEMCPY2D_st +#undef CUDA_MEMCPY2D +#undef CUDA_MEMCPY3D_st +#undef CUDA_MEMCPY3D +#undef CUDA_ARRAY_DESCRIPTOR_st +#undef CUDA_ARRAY_DESCRIPTOR +#undef CUDA_ARRAY3D_DESCRIPTOR_st +#undef CUDA_ARRAY3D_DESCRIPTOR #endif /* __CUDA_API_VERSION_INTERNAL */ #if defined(__CUDA_API_VERSION_INTERNAL) - CUresult CUDAAPI cuMemcpyHtoD_v2(CUdeviceptr dstDevice, const void *srcHost, size_t ByteCount); - CUresult CUDAAPI cuMemcpyDtoH_v2(void *dstHost, CUdeviceptr srcDevice, size_t ByteCount); - CUresult CUDAAPI cuMemcpyDtoD_v2(CUdeviceptr dstDevice, CUdeviceptr srcDevice, size_t ByteCount); - CUresult CUDAAPI cuMemcpyDtoA_v2(CUarray dstArray, size_t dstOffset, CUdeviceptr srcDevice, size_t ByteCount); - CUresult CUDAAPI cuMemcpyAtoD_v2(CUdeviceptr dstDevice, CUarray srcArray, size_t srcOffset, size_t ByteCount); - CUresult CUDAAPI cuMemcpyHtoA_v2(CUarray dstArray, size_t dstOffset, const void *srcHost, size_t ByteCount); - CUresult CUDAAPI cuMemcpyAtoH_v2(void *dstHost, CUarray srcArray, size_t srcOffset, size_t ByteCount); - CUresult CUDAAPI cuMemcpyAtoA_v2(CUarray dstArray, size_t dstOffset, CUarray srcArray, size_t srcOffset, size_t ByteCount); - CUresult CUDAAPI cuMemcpyHtoAAsync_v2(CUarray dstArray, size_t dstOffset, const void *srcHost, size_t ByteCount, CUstream hStream); - CUresult CUDAAPI cuMemcpyAtoHAsync_v2(void *dstHost, CUarray srcArray, size_t srcOffset, size_t ByteCount, CUstream hStream); - CUresult CUDAAPI cuMemcpy2D_v2(const CUDA_MEMCPY2D *pCopy); - CUresult CUDAAPI cuMemcpy2DUnaligned_v2(const CUDA_MEMCPY2D *pCopy); - CUresult CUDAAPI cuMemcpy3D_v2(const CUDA_MEMCPY3D *pCopy); - CUresult CUDAAPI cuMemcpyHtoDAsync_v2(CUdeviceptr dstDevice, const void *srcHost, size_t ByteCount, CUstream hStream); - CUresult CUDAAPI cuMemcpyDtoHAsync_v2(void *dstHost, CUdeviceptr srcDevice, size_t ByteCount, CUstream hStream); - CUresult CUDAAPI cuMemcpyDtoDAsync_v2(CUdeviceptr dstDevice, CUdeviceptr srcDevice, size_t ByteCount, CUstream hStream); - CUresult CUDAAPI cuMemcpy2DAsync_v2(const CUDA_MEMCPY2D *pCopy, CUstream hStream); - CUresult CUDAAPI cuMemcpy3DAsync_v2(const CUDA_MEMCPY3D *pCopy, CUstream hStream); - CUresult CUDAAPI cuMemsetD8_v2(CUdeviceptr dstDevice, unsigned char uc, size_t N); - CUresult CUDAAPI cuMemsetD16_v2(CUdeviceptr dstDevice, unsigned short us, size_t N); - CUresult CUDAAPI cuMemsetD32_v2(CUdeviceptr dstDevice, unsigned int ui, size_t N); - CUresult CUDAAPI cuMemsetD2D8_v2(CUdeviceptr dstDevice, size_t dstPitch, unsigned char uc, size_t Width, size_t Height); - CUresult CUDAAPI cuMemsetD2D16_v2(CUdeviceptr dstDevice, size_t dstPitch, unsigned short us, size_t Width, size_t Height); - CUresult CUDAAPI cuMemsetD2D32_v2(CUdeviceptr dstDevice, size_t dstPitch, unsigned int ui, size_t Width, size_t Height); - CUresult CUDAAPI cuMemcpy(CUdeviceptr dst, CUdeviceptr src, size_t ByteCount); - CUresult CUDAAPI cuMemcpyAsync(CUdeviceptr dst, CUdeviceptr src, size_t ByteCount, CUstream hStream); - CUresult CUDAAPI cuMemcpyPeer(CUdeviceptr dstDevice, CUcontext dstContext, CUdeviceptr srcDevice, CUcontext srcContext, size_t ByteCount); - CUresult CUDAAPI cuMemcpyPeerAsync(CUdeviceptr dstDevice, CUcontext dstContext, CUdeviceptr srcDevice, CUcontext srcContext, size_t ByteCount, CUstream hStream); - CUresult CUDAAPI cuMemcpy3DPeer(const CUDA_MEMCPY3D_PEER *pCopy); - CUresult CUDAAPI cuMemcpy3DPeerAsync(const CUDA_MEMCPY3D_PEER *pCopy, CUstream hStream); - - CUresult CUDAAPI cuMemsetD8Async(CUdeviceptr dstDevice, unsigned char uc, size_t N, CUstream hStream); - CUresult CUDAAPI cuMemsetD16Async(CUdeviceptr dstDevice, unsigned short us, size_t N, CUstream hStream); - CUresult CUDAAPI cuMemsetD32Async(CUdeviceptr dstDevice, unsigned int ui, size_t N, CUstream hStream); - CUresult CUDAAPI cuMemsetD2D8Async(CUdeviceptr dstDevice, size_t dstPitch, unsigned char uc, size_t Width, size_t Height, CUstream hStream); - CUresult CUDAAPI cuMemsetD2D16Async(CUdeviceptr dstDevice, size_t dstPitch, unsigned short us, size_t Width, size_t Height, CUstream hStream); - CUresult CUDAAPI cuMemsetD2D32Async(CUdeviceptr dstDevice, size_t dstPitch, unsigned int ui, size_t Width, size_t Height, CUstream hStream); - - CUresult CUDAAPI cuStreamGetPriority(CUstream hStream, int *priority); - CUresult CUDAAPI cuStreamGetFlags(CUstream hStream, unsigned int *flags); - CUresult CUDAAPI cuStreamGetCtx(CUstream hStream, CUcontext *pctx); - CUresult CUDAAPI cuStreamWaitEvent(CUstream hStream, CUevent hEvent, unsigned int Flags); - CUresult CUDAAPI cuStreamAddCallback(CUstream hStream, CUstreamCallback callback, void *userData, unsigned int flags); - CUresult CUDAAPI cuStreamAttachMemAsync(CUstream hStream, CUdeviceptr dptr, size_t length, unsigned int flags); - CUresult CUDAAPI cuStreamQuery(CUstream hStream); - CUresult CUDAAPI cuStreamSynchronize(CUstream hStream); - CUresult CUDAAPI cuEventRecord(CUevent hEvent, CUstream hStream); - CUresult CUDAAPI cuLaunchKernel(CUfunction f, unsigned int gridDimX, unsigned int gridDimY, unsigned int gridDimZ, unsigned int blockDimX, unsigned int blockDimY, unsigned int blockDimZ, unsigned int sharedMemBytes, CUstream hStream, void **kernelParams, void **extra); - CUresult CUDAAPI cuLaunchHostFunc(CUstream hStream, CUhostFn fn, void *userData); - CUresult CUDAAPI cuGraphicsMapResources(unsigned int count, CUgraphicsResource *resources, CUstream hStream); - CUresult CUDAAPI cuGraphicsUnmapResources(unsigned int count, CUgraphicsResource *resources, CUstream hStream); - CUresult CUDAAPI cuStreamWriteValue32(CUstream stream, CUdeviceptr addr, cuuint32_t value, unsigned int flags); - CUresult CUDAAPI cuStreamWaitValue32(CUstream stream, CUdeviceptr addr, cuuint32_t value, unsigned int flags); - CUresult CUDAAPI cuStreamWriteValue64(CUstream stream, CUdeviceptr addr, cuuint64_t value, unsigned int flags); - CUresult CUDAAPI cuStreamWaitValue64(CUstream stream, CUdeviceptr addr, cuuint64_t value, unsigned int flags); - CUresult CUDAAPI cuStreamBatchMemOp(CUstream stream, unsigned int count, CUstreamBatchMemOpParams *paramArray, unsigned int flags); - CUresult CUDAAPI cuMemPrefetchAsync(CUdeviceptr devPtr, size_t count, CUdevice dstDevice, CUstream hStream); - CUresult CUDAAPI cuLaunchCooperativeKernel(CUfunction f, unsigned int gridDimX, unsigned int gridDimY, unsigned int gridDimZ, unsigned int blockDimX, unsigned int blockDimY, unsigned int blockDimZ, unsigned int sharedMemBytes, CUstream hStream, void **kernelParams); - CUresult CUDAAPI cuSignalExternalSemaphoresAsync(const CUexternalSemaphore *extSemArray, const CUDA_EXTERNAL_SEMAPHORE_SIGNAL_PARAMS *paramsArray, unsigned int numExtSems, CUstream stream); - CUresult CUDAAPI cuWaitExternalSemaphoresAsync(const CUexternalSemaphore *extSemArray, const CUDA_EXTERNAL_SEMAPHORE_WAIT_PARAMS *paramsArray, unsigned int numExtSems, CUstream stream); - CUresult CUDAAPI cuStreamBeginCapture(CUstream hStream); - CUresult CUDAAPI cuStreamBeginCapture_ptsz(CUstream hStream); - CUresult CUDAAPI cuStreamBeginCapture_v2(CUstream hStream, CUstreamCaptureMode mode); - CUresult CUDAAPI cuStreamEndCapture(CUstream hStream, CUgraph *phGraph); - CUresult CUDAAPI cuStreamIsCapturing(CUstream hStream, CUstreamCaptureStatus *captureStatus); - CUresult CUDAAPI cuStreamGetCaptureInfo(CUstream hStream, CUstreamCaptureStatus *captureStatus, cuuint64_t *id); - CUresult CUDAAPI cuGraphLaunch(CUgraphExec hGraph, CUstream hStream); +CUresult CUDAAPI cuMemcpyHtoD_v2(CUdeviceptr dstDevice, const void *srcHost, + size_t ByteCount); +CUresult CUDAAPI cuMemcpyDtoH_v2(void *dstHost, CUdeviceptr srcDevice, + size_t ByteCount); +CUresult CUDAAPI cuMemcpyDtoD_v2(CUdeviceptr dstDevice, CUdeviceptr srcDevice, + size_t ByteCount); +CUresult CUDAAPI cuMemcpyDtoA_v2(CUarray dstArray, size_t dstOffset, + CUdeviceptr srcDevice, size_t ByteCount); +CUresult CUDAAPI cuMemcpyAtoD_v2(CUdeviceptr dstDevice, CUarray srcArray, + size_t srcOffset, size_t ByteCount); +CUresult CUDAAPI cuMemcpyHtoA_v2(CUarray dstArray, size_t dstOffset, + const void *srcHost, size_t ByteCount); +CUresult CUDAAPI cuMemcpyAtoH_v2(void *dstHost, CUarray srcArray, + size_t srcOffset, size_t ByteCount); +CUresult CUDAAPI cuMemcpyAtoA_v2(CUarray dstArray, size_t dstOffset, + CUarray srcArray, size_t srcOffset, + size_t ByteCount); +CUresult CUDAAPI cuMemcpyHtoAAsync_v2(CUarray dstArray, size_t dstOffset, + const void *srcHost, size_t ByteCount, + CUstream hStream); +CUresult CUDAAPI cuMemcpyAtoHAsync_v2(void *dstHost, CUarray srcArray, + size_t srcOffset, size_t ByteCount, + CUstream hStream); +CUresult CUDAAPI cuMemcpy2D_v2(const CUDA_MEMCPY2D *pCopy); +CUresult CUDAAPI cuMemcpy2DUnaligned_v2(const CUDA_MEMCPY2D *pCopy); +CUresult CUDAAPI cuMemcpy3D_v2(const CUDA_MEMCPY3D *pCopy); +CUresult CUDAAPI cuMemcpyHtoDAsync_v2(CUdeviceptr dstDevice, + const void *srcHost, size_t ByteCount, + CUstream hStream); +CUresult CUDAAPI cuMemcpyDtoHAsync_v2(void *dstHost, CUdeviceptr srcDevice, + size_t ByteCount, CUstream hStream); +CUresult CUDAAPI cuMemcpyDtoDAsync_v2(CUdeviceptr dstDevice, + CUdeviceptr srcDevice, size_t ByteCount, + CUstream hStream); +CUresult CUDAAPI cuMemcpy2DAsync_v2(const CUDA_MEMCPY2D *pCopy, + CUstream hStream); +CUresult CUDAAPI cuMemcpy3DAsync_v2(const CUDA_MEMCPY3D *pCopy, + CUstream hStream); +CUresult CUDAAPI cuMemsetD8_v2(CUdeviceptr dstDevice, unsigned char uc, + size_t N); +CUresult CUDAAPI cuMemsetD16_v2(CUdeviceptr dstDevice, unsigned short us, + size_t N); +CUresult CUDAAPI cuMemsetD32_v2(CUdeviceptr dstDevice, unsigned int ui, + size_t N); +CUresult CUDAAPI cuMemsetD2D8_v2(CUdeviceptr dstDevice, size_t dstPitch, + unsigned char uc, size_t Width, size_t Height); +CUresult CUDAAPI cuMemsetD2D16_v2(CUdeviceptr dstDevice, size_t dstPitch, + unsigned short us, size_t Width, + size_t Height); +CUresult CUDAAPI cuMemsetD2D32_v2(CUdeviceptr dstDevice, size_t dstPitch, + unsigned int ui, size_t Width, size_t Height); +CUresult CUDAAPI cuMemcpy(CUdeviceptr dst, CUdeviceptr src, size_t ByteCount); +CUresult CUDAAPI cuMemcpyAsync(CUdeviceptr dst, CUdeviceptr src, + size_t ByteCount, CUstream hStream); +CUresult CUDAAPI cuMemcpyPeer(CUdeviceptr dstDevice, CUcontext dstContext, + CUdeviceptr srcDevice, CUcontext srcContext, + size_t ByteCount); +CUresult CUDAAPI cuMemcpyPeerAsync(CUdeviceptr dstDevice, CUcontext dstContext, + CUdeviceptr srcDevice, CUcontext srcContext, + size_t ByteCount, CUstream hStream); +CUresult CUDAAPI cuMemcpy3DPeer(const CUDA_MEMCPY3D_PEER *pCopy); +CUresult CUDAAPI cuMemcpy3DPeerAsync(const CUDA_MEMCPY3D_PEER *pCopy, + CUstream hStream); + +CUresult CUDAAPI cuMemsetD8Async(CUdeviceptr dstDevice, unsigned char uc, + size_t N, CUstream hStream); +CUresult CUDAAPI cuMemsetD16Async(CUdeviceptr dstDevice, unsigned short us, + size_t N, CUstream hStream); +CUresult CUDAAPI cuMemsetD32Async(CUdeviceptr dstDevice, unsigned int ui, + size_t N, CUstream hStream); +CUresult CUDAAPI cuMemsetD2D8Async(CUdeviceptr dstDevice, size_t dstPitch, + unsigned char uc, size_t Width, + size_t Height, CUstream hStream); +CUresult CUDAAPI cuMemsetD2D16Async(CUdeviceptr dstDevice, size_t dstPitch, + unsigned short us, size_t Width, + size_t Height, CUstream hStream); +CUresult CUDAAPI cuMemsetD2D32Async(CUdeviceptr dstDevice, size_t dstPitch, + unsigned int ui, size_t Width, + size_t Height, CUstream hStream); + +CUresult CUDAAPI cuStreamGetPriority(CUstream hStream, int *priority); +CUresult CUDAAPI cuStreamGetFlags(CUstream hStream, unsigned int *flags); +CUresult CUDAAPI cuStreamGetCtx(CUstream hStream, CUcontext *pctx); +CUresult CUDAAPI cuStreamWaitEvent(CUstream hStream, CUevent hEvent, + unsigned int Flags); +CUresult CUDAAPI cuStreamAddCallback(CUstream hStream, + CUstreamCallback callback, void *userData, + unsigned int flags); +CUresult CUDAAPI cuStreamAttachMemAsync(CUstream hStream, CUdeviceptr dptr, + size_t length, unsigned int flags); +CUresult CUDAAPI cuStreamQuery(CUstream hStream); +CUresult CUDAAPI cuStreamSynchronize(CUstream hStream); +CUresult CUDAAPI cuEventRecord(CUevent hEvent, CUstream hStream); +CUresult CUDAAPI cuLaunchKernel(CUfunction f, unsigned int gridDimX, + unsigned int gridDimY, unsigned int gridDimZ, + unsigned int blockDimX, unsigned int blockDimY, + unsigned int blockDimZ, + unsigned int sharedMemBytes, CUstream hStream, + void **kernelParams, void **extra); +CUresult CUDAAPI cuLaunchHostFunc(CUstream hStream, CUhostFn fn, + void *userData); +CUresult CUDAAPI cuGraphicsMapResources(unsigned int count, + CUgraphicsResource *resources, + CUstream hStream); +CUresult CUDAAPI cuGraphicsUnmapResources(unsigned int count, + CUgraphicsResource *resources, + CUstream hStream); +CUresult CUDAAPI cuStreamWriteValue32(CUstream stream, CUdeviceptr addr, + cuuint32_t value, unsigned int flags); +CUresult CUDAAPI cuStreamWaitValue32(CUstream stream, CUdeviceptr addr, + cuuint32_t value, unsigned int flags); +CUresult CUDAAPI cuStreamWriteValue64(CUstream stream, CUdeviceptr addr, + cuuint64_t value, unsigned int flags); +CUresult CUDAAPI cuStreamWaitValue64(CUstream stream, CUdeviceptr addr, + cuuint64_t value, unsigned int flags); +CUresult CUDAAPI cuStreamBatchMemOp(CUstream stream, unsigned int count, + CUstreamBatchMemOpParams *paramArray, + unsigned int flags); +CUresult CUDAAPI cuMemPrefetchAsync(CUdeviceptr devPtr, size_t count, + CUdevice dstDevice, CUstream hStream); +CUresult CUDAAPI cuLaunchCooperativeKernel( + CUfunction f, unsigned int gridDimX, unsigned int gridDimY, + unsigned int gridDimZ, unsigned int blockDimX, unsigned int blockDimY, + unsigned int blockDimZ, unsigned int sharedMemBytes, CUstream hStream, + void **kernelParams); +CUresult CUDAAPI cuSignalExternalSemaphoresAsync( + const CUexternalSemaphore *extSemArray, + const CUDA_EXTERNAL_SEMAPHORE_SIGNAL_PARAMS *paramsArray, + unsigned int numExtSems, CUstream stream); +CUresult CUDAAPI cuWaitExternalSemaphoresAsync( + const CUexternalSemaphore *extSemArray, + const CUDA_EXTERNAL_SEMAPHORE_WAIT_PARAMS *paramsArray, + unsigned int numExtSems, CUstream stream); +CUresult CUDAAPI cuStreamBeginCapture(CUstream hStream); +CUresult CUDAAPI cuStreamBeginCapture_ptsz(CUstream hStream); +CUresult CUDAAPI cuStreamBeginCapture_v2(CUstream hStream, + CUstreamCaptureMode mode); +CUresult CUDAAPI cuStreamEndCapture(CUstream hStream, CUgraph *phGraph); +CUresult CUDAAPI cuStreamIsCapturing(CUstream hStream, + CUstreamCaptureStatus *captureStatus); +CUresult CUDAAPI cuStreamGetCaptureInfo(CUstream hStream, + CUstreamCaptureStatus *captureStatus, + cuuint64_t *id); +CUresult CUDAAPI cuGraphLaunch(CUgraphExec hGraph, CUstream hStream); #endif #ifdef __cplusplus diff --git a/libcuda/cuda_api_object.h b/libcuda/cuda_api_object.h index 51416f2..d292e22 100644 --- a/libcuda/cuda_api_object.h +++ b/libcuda/cuda_api_object.h @@ -8,9 +8,9 @@ #include "builtin_types.h" -#include "../src/gpgpu-sim/gpu-sim.h" -#include "../src/cuda-sim/ptx_ir.h" #include "../src/abstract_hardware_model.h" +#include "../src/cuda-sim/ptx_ir.h" +#include "../src/gpgpu-sim/gpu-sim.h" #include "cuobjdump.h" typedef std::list gpgpu_ptx_sim_arg_list_t; @@ -20,194 +20,198 @@ typedef unsigned long GLuint; #endif struct glbmap_entry { - GLuint m_bufferObj; - void *m_devPtr; - size_t m_size; - struct glbmap_entry *m_next; + GLuint m_bufferObj; + void *m_devPtr; + size_t m_size; + struct glbmap_entry *m_next; }; typedef struct glbmap_entry glbmap_entry_t; struct _cuda_device_id { - _cuda_device_id(gpgpu_sim* gpu) {m_id = 0; m_next = NULL; m_gpgpu=gpu;} - struct _cuda_device_id *next() { return m_next; } - unsigned num_shader() const { return m_gpgpu->get_config().num_shader(); } - int num_devices() const { - if( m_next == NULL ) return 1; - else return 1 + m_next->num_devices(); - } - struct _cuda_device_id *get_device( unsigned n ) - { - assert( n < (unsigned)num_devices() ); - struct _cuda_device_id *p=this; - for(unsigned i=0; im_next; - return p; - } - const struct cudaDeviceProp *get_prop() const - { - return m_gpgpu->get_prop(); - } - unsigned get_id() const { return m_id; } - - gpgpu_sim *get_gpgpu() { return m_gpgpu; } -private: - unsigned m_id; - class gpgpu_sim *m_gpgpu; - struct _cuda_device_id *m_next; + _cuda_device_id(gpgpu_sim *gpu) { + m_id = 0; + m_next = NULL; + m_gpgpu = gpu; + } + struct _cuda_device_id *next() { + return m_next; + } + unsigned num_shader() const { return m_gpgpu->get_config().num_shader(); } + int num_devices() const { + if (m_next == NULL) + return 1; + else + return 1 + m_next->num_devices(); + } + struct _cuda_device_id *get_device(unsigned n) { + assert(n < (unsigned)num_devices()); + struct _cuda_device_id *p = this; + for (unsigned i = 0; i < n; i++) p = p->m_next; + return p; + } + const struct cudaDeviceProp *get_prop() const { return m_gpgpu->get_prop(); } + unsigned get_id() const { return m_id; } + + gpgpu_sim *get_gpgpu() { return m_gpgpu; } + + private: + unsigned m_id; + class gpgpu_sim *m_gpgpu; + struct _cuda_device_id *m_next; }; struct CUctx_st { - CUctx_st( _cuda_device_id *gpu ) - { - m_gpu = gpu; - m_binary_info.cmem = 0; - m_binary_info.gmem = 0; - no_of_ptx=0; - } - - _cuda_device_id *get_device() { return m_gpu; } - - void add_binary( symbol_table *symtab, unsigned fat_cubin_handle ) - { - m_code[fat_cubin_handle] = symtab; - m_last_fat_cubin_handle = fat_cubin_handle; - } - - void add_ptxinfo( const char *deviceFun, const struct gpgpu_ptx_sim_info &info ) - { - symbol *s = m_code[m_last_fat_cubin_handle]->lookup(deviceFun); - assert( s != NULL ); - function_info *f = s->get_pc(); - assert( f != NULL ); - f->set_kernel_info(info); - } - - void add_ptxinfo( const struct gpgpu_ptx_sim_info &info ) - { - m_binary_info = info; - } - - void register_function( unsigned fat_cubin_handle, const char *hostFun, const char *deviceFun ) - { - if( m_code.find(fat_cubin_handle) != m_code.end() ) { - symbol *s = m_code[fat_cubin_handle]->lookup(deviceFun); - if(s != NULL) { - function_info *f = s->get_pc(); - assert( f != NULL ); - m_kernel_lookup[hostFun] = f; - } - else { - printf("Warning: cannot find deviceFun %s\n", deviceFun); - m_kernel_lookup[hostFun] = NULL; - } - // assert( s != NULL ); - // function_info *f = s->get_pc(); - // assert( f != NULL ); - // m_kernel_lookup[hostFun] = f; - } else { - m_kernel_lookup[hostFun] = NULL; - } - } - - void register_hostFun_function( const char*hostFun, function_info* f){ + CUctx_st(_cuda_device_id *gpu) { + m_gpu = gpu; + m_binary_info.cmem = 0; + m_binary_info.gmem = 0; + no_of_ptx = 0; + } + + _cuda_device_id *get_device() { return m_gpu; } + + void add_binary(symbol_table *symtab, unsigned fat_cubin_handle) { + m_code[fat_cubin_handle] = symtab; + m_last_fat_cubin_handle = fat_cubin_handle; + } + + void add_ptxinfo(const char *deviceFun, + const struct gpgpu_ptx_sim_info &info) { + symbol *s = m_code[m_last_fat_cubin_handle]->lookup(deviceFun); + assert(s != NULL); + function_info *f = s->get_pc(); + assert(f != NULL); + f->set_kernel_info(info); + } + + void add_ptxinfo(const struct gpgpu_ptx_sim_info &info) { + m_binary_info = info; + } + + void register_function(unsigned fat_cubin_handle, const char *hostFun, + const char *deviceFun) { + if (m_code.find(fat_cubin_handle) != m_code.end()) { + symbol *s = m_code[fat_cubin_handle]->lookup(deviceFun); + if (s != NULL) { + function_info *f = s->get_pc(); + assert(f != NULL); m_kernel_lookup[hostFun] = f; + } else { + printf("Warning: cannot find deviceFun %s\n", deviceFun); + m_kernel_lookup[hostFun] = NULL; + } + // assert( s != NULL ); + // function_info *f = s->get_pc(); + // assert( f != NULL ); + // m_kernel_lookup[hostFun] = f; + } else { + m_kernel_lookup[hostFun] = NULL; } - - function_info *get_kernel(const char *hostFun) - { - std::map::iterator i=m_kernel_lookup.find(hostFun); - assert( i != m_kernel_lookup.end() ); - return i->second; - } - - int no_of_ptx; - -private: - _cuda_device_id *m_gpu; // selected gpu - std::map m_code; // fat binary handle => global symbol table - unsigned m_last_fat_cubin_handle; - std::map m_kernel_lookup; // unique id (CUDA app function address) => kernel entry point - struct gpgpu_ptx_sim_info m_binary_info; - + } + + void register_hostFun_function(const char *hostFun, function_info *f) { + m_kernel_lookup[hostFun] = f; + } + + function_info *get_kernel(const char *hostFun) { + std::map::iterator i = + m_kernel_lookup.find(hostFun); + assert(i != m_kernel_lookup.end()); + return i->second; + } + + int no_of_ptx; + + private: + _cuda_device_id *m_gpu; // selected gpu + std::map + m_code; // fat binary handle => global symbol table + unsigned m_last_fat_cubin_handle; + std::map + m_kernel_lookup; // unique id (CUDA app function address) => kernel entry + // point + struct gpgpu_ptx_sim_info m_binary_info; }; class kernel_config { -public: - kernel_config( dim3 GridDim, dim3 BlockDim, size_t sharedMem, struct CUstream_st *stream ) - { - m_GridDim=GridDim; - m_BlockDim=BlockDim; - m_sharedMem=sharedMem; - m_stream = stream; - } - kernel_config() - { - m_GridDim=dim3(-1,-1,-1); - m_BlockDim=dim3(-1,-1,-1); - m_sharedMem=0; - m_stream =NULL; - } - void set_arg( const void *arg, size_t size, size_t offset ) - { - m_args.push_front( gpgpu_ptx_sim_arg(arg,size,offset) ); - } - dim3 grid_dim() const { return m_GridDim; } - dim3 block_dim() const { return m_BlockDim; } - void set_grid_dim(dim3 *d) { m_GridDim = *d; } - void set_block_dim(dim3 *d) { m_BlockDim = *d; } - gpgpu_ptx_sim_arg_list_t get_args() { return m_args; } - struct CUstream_st *get_stream() { return m_stream; } - -private: - dim3 m_GridDim; - dim3 m_BlockDim; - size_t m_sharedMem; - struct CUstream_st *m_stream; - gpgpu_ptx_sim_arg_list_t m_args; + public: + kernel_config(dim3 GridDim, dim3 BlockDim, size_t sharedMem, + struct CUstream_st *stream) { + m_GridDim = GridDim; + m_BlockDim = BlockDim; + m_sharedMem = sharedMem; + m_stream = stream; + } + kernel_config() { + m_GridDim = dim3(-1, -1, -1); + m_BlockDim = dim3(-1, -1, -1); + m_sharedMem = 0; + m_stream = NULL; + } + void set_arg(const void *arg, size_t size, size_t offset) { + m_args.push_front(gpgpu_ptx_sim_arg(arg, size, offset)); + } + dim3 grid_dim() const { return m_GridDim; } + dim3 block_dim() const { return m_BlockDim; } + void set_grid_dim(dim3 *d) { m_GridDim = *d; } + void set_block_dim(dim3 *d) { m_BlockDim = *d; } + gpgpu_ptx_sim_arg_list_t get_args() { return m_args; } + struct CUstream_st *get_stream() { + return m_stream; + } + + private: + dim3 m_GridDim; + dim3 m_BlockDim; + size_t m_sharedMem; + struct CUstream_st *m_stream; + gpgpu_ptx_sim_arg_list_t m_args; }; class cuda_runtime_api { - public: - cuda_runtime_api( gpgpu_context* ctx ) { - g_glbmap = NULL; - g_active_device = 0; //active gpu that runs the code - gpgpu_ctx = ctx; - } - // global list - std::list cuobjdumpSectionList; - std::list libSectionList; - std::list g_cuda_launch_stack; - std::mapfatbin_registered; - std::map fatbinmap; - std::map name_symtab; - std::map g_mallocPtr_Size; - //maps sm version number to set of filenames - std::map > version_filename; - std::map pinned_memory; //support for pinned memories added - std::map pinned_memory_size; - glbmap_entry_t* g_glbmap; - int g_active_device; //active gpu that runs the code - // backward pointer - class gpgpu_context* gpgpu_ctx; - // member function list - void cuobjdumpInit(); - void extract_code_using_cuobjdump(); - void extract_ptx_files_using_cuobjdump(CUctx_st *context); - std::list pruneSectionList(CUctx_st *context); - std::list mergeMatchingSections(std::string identifier); - std::list mergeSections(); - cuobjdumpELFSection* findELFSection(const std::string identifier); - cuobjdumpPTXSection* findPTXSection(const std::string identifier); - cuobjdumpPTXSection* findPTXSectionInList(std::list §ionlist, const std::string identifier); - void cuobjdumpRegisterFatBinary(unsigned int handle, const char* filename, CUctx_st *context); - kernel_info_t *gpgpu_cuda_ptx_sim_init_grid( const char *kernel_key, - gpgpu_ptx_sim_arg_list_t args, - struct dim3 gridDim, - struct dim3 blockDim, - struct CUctx_st* context ); - int load_static_globals( symbol_table *symtab, unsigned min_gaddr, unsigned max_gaddr, gpgpu_t *gpu ); - int load_constants( symbol_table *symtab, addr_t min_gaddr, gpgpu_t *gpu ); - + public: + cuda_runtime_api(gpgpu_context *ctx) { + g_glbmap = NULL; + g_active_device = 0; // active gpu that runs the code + gpgpu_ctx = ctx; + } + // global list + std::list cuobjdumpSectionList; + std::list libSectionList; + std::list g_cuda_launch_stack; + std::map fatbin_registered; + std::map fatbinmap; + std::map name_symtab; + std::map g_mallocPtr_Size; + // maps sm version number to set of filenames + std::map > version_filename; + std::map pinned_memory; // support for pinned memories added + std::map pinned_memory_size; + glbmap_entry_t *g_glbmap; + int g_active_device; // active gpu that runs the code + // backward pointer + class gpgpu_context *gpgpu_ctx; + // member function list + void cuobjdumpInit(); + void extract_code_using_cuobjdump(); + void extract_ptx_files_using_cuobjdump(CUctx_st *context); + std::list pruneSectionList(CUctx_st *context); + std::list mergeMatchingSections(std::string identifier); + std::list mergeSections(); + cuobjdumpELFSection *findELFSection(const std::string identifier); + cuobjdumpPTXSection *findPTXSection(const std::string identifier); + cuobjdumpPTXSection *findPTXSectionInList( + std::list §ionlist, const std::string identifier); + void cuobjdumpRegisterFatBinary(unsigned int handle, const char *filename, + CUctx_st *context); + kernel_info_t *gpgpu_cuda_ptx_sim_init_grid(const char *kernel_key, + gpgpu_ptx_sim_arg_list_t args, + struct dim3 gridDim, + struct dim3 blockDim, + struct CUctx_st *context); + int load_static_globals(symbol_table *symtab, unsigned min_gaddr, + unsigned max_gaddr, gpgpu_t *gpu); + int load_constants(symbol_table *symtab, addr_t min_gaddr, gpgpu_t *gpu); }; #endif /* __cuda_api_object_h__ */ diff --git a/libcuda/cuda_runtime_api.cc b/libcuda/cuda_runtime_api.cc index 273194e..cc01f12 100644 --- a/libcuda/cuda_runtime_api.cc +++ b/libcuda/cuda_runtime_api.cc @@ -2,16 +2,16 @@ // Changes Copyright 2009, Tor M. Aamodt, Ali Bakhoda and George L. Yuan // University of British Columbia -/* +/* * cuda_runtime_api.cc * - * Copyright © 2009 by Tor M. Aamodt, Wilson W. L. Fung, Ali Bakhoda, - * George L. Yuan and the University of British Columbia, Vancouver, + * Copyright © 2009 by Tor M. Aamodt, Wilson W. L. Fung, Ali Bakhoda, + * George L. Yuan and the University of British Columbia, Vancouver, * BC V6T 1Z4, All Rights Reserved. - * + * * THIS IS A LEGAL DOCUMENT BY DOWNLOADING GPGPU-SIM, YOU ARE AGREEING TO THESE * TERMS AND CONDITIONS. - * + * * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" * AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE @@ -23,99 +23,100 @@ * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE * POSSIBILITY OF SUCH DAMAGE. - * + * * NOTE: The files libcuda/cuda_runtime_api.c and src/cuda-sim/cuda-math.h * are derived from the CUDA Toolset available from http://www.nvidia.com/cuda - * (property of NVIDIA). The files benchmarks/BlackScholes/ and - * benchmarks/template/ are derived from the CUDA SDK available from - * http://www.nvidia.com/cuda (also property of NVIDIA). The files from - * src/intersim/ are derived from Booksim (a simulator provided with the - * textbook "Principles and Practices of Interconnection Networks" available - * from http://cva.stanford.edu/books/ppin/). As such, those files are bound by - * the corresponding legal terms and conditions set forth separately (original - * copyright notices are left in files from these sources and where we have - * modified a file our copyright notice appears before the original copyright - * notice). - * - * Using this version of GPGPU-Sim requires a complete installation of CUDA - * which is distributed seperately by NVIDIA under separate terms and + * (property of NVIDIA). The files benchmarks/BlackScholes/ and + * benchmarks/template/ are derived from the CUDA SDK available from + * http://www.nvidia.com/cuda (also property of NVIDIA). The files from + * src/intersim/ are derived from Booksim (a simulator provided with the + * textbook "Principles and Practices of Interconnection Networks" available + * from http://cva.stanford.edu/books/ppin/). As such, those files are bound by + * the corresponding legal terms and conditions set forth separately (original + * copyright notices are left in files from these sources and where we have + * modified a file our copyright notice appears before the original copyright + * notice). + * + * Using this version of GPGPU-Sim requires a complete installation of CUDA + * which is distributed seperately by NVIDIA under separate terms and * conditions. To use this version of GPGPU-Sim with OpenCL requires a * recent version of NVIDIA's drivers which support OpenCL. - * + * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions are met: - * + * * 1. Redistributions of source code must retain the above copyright notice, * this list of conditions and the following disclaimer. - * + * * 2. Redistributions in binary form must reproduce the above copyright notice, * this list of conditions and the following disclaimer in the documentation * and/or other materials provided with the distribution. - * + * * 3. Neither the name of the University of British Columbia nor the names of * its contributors may be used to endorse or promote products derived from * this software without specific prior written permission. - * - * 4. This version of GPGPU-SIM is distributed freely for non-commercial use only. - * + * + * 4. This version of GPGPU-SIM is distributed freely for non-commercial use + * only. + * * 5. No nonprofit user may place any restrictions on the use of this software, * including as modified by the user, by any other authorized user. - * - * 6. GPGPU-SIM was developed primarily by Tor M. Aamodt, Wilson W. L. Fung, - * Ali Bakhoda, George L. Yuan, at the University of British Columbia, + * + * 6. GPGPU-SIM was developed primarily by Tor M. Aamodt, Wilson W. L. Fung, + * Ali Bakhoda, George L. Yuan, at the University of British Columbia, * Vancouver, BC V6T 1Z4 */ /* * Copyright 1993-2007 NVIDIA Corporation. All rights reserved. * - * NOTICE TO USER: + * NOTICE TO USER: * - * This source code is subject to NVIDIA ownership rights under U.S. and - * international Copyright laws. Users and possessors of this source code - * are hereby granted a nonexclusive, royalty-free license to use this code + * This source code is subject to NVIDIA ownership rights under U.S. and + * international Copyright laws. Users and possessors of this source code + * are hereby granted a nonexclusive, royalty-free license to use this code * in individual and commercial software. * - * NVIDIA MAKES NO REPRESENTATION ABOUT THE SUITABILITY OF THIS SOURCE - * CODE FOR ANY PURPOSE. IT IS PROVIDED "AS IS" WITHOUT EXPRESS OR - * IMPLIED WARRANTY OF ANY KIND. NVIDIA DISCLAIMS ALL WARRANTIES WITH - * REGARD TO THIS SOURCE CODE, INCLUDING ALL IMPLIED WARRANTIES OF + * NVIDIA MAKES NO REPRESENTATION ABOUT THE SUITABILITY OF THIS SOURCE + * CODE FOR ANY PURPOSE. IT IS PROVIDED "AS IS" WITHOUT EXPRESS OR + * IMPLIED WARRANTY OF ANY KIND. NVIDIA DISCLAIMS ALL WARRANTIES WITH + * REGARD TO THIS SOURCE CODE, INCLUDING ALL IMPLIED WARRANTIES OF * MERCHANTABILITY, NONINFRINGEMENT, AND FITNESS FOR A PARTICULAR PURPOSE. - * IN NO EVENT SHALL NVIDIA BE LIABLE FOR ANY SPECIAL, INDIRECT, INCIDENTAL, - * OR CONSEQUENTIAL DAMAGES, OR ANY DAMAGES WHATSOEVER RESULTING FROM LOSS - * OF USE, DATA OR PROFITS, WHETHER IN AN ACTION OF CONTRACT, NEGLIGENCE - * OR OTHER TORTIOUS ACTION, ARISING OUT OF OR IN CONNECTION WITH THE USE - * OR PERFORMANCE OF THIS SOURCE CODE. + * IN NO EVENT SHALL NVIDIA BE LIABLE FOR ANY SPECIAL, INDIRECT, INCIDENTAL, + * OR CONSEQUENTIAL DAMAGES, OR ANY DAMAGES WHATSOEVER RESULTING FROM LOSS + * OF USE, DATA OR PROFITS, WHETHER IN AN ACTION OF CONTRACT, NEGLIGENCE + * OR OTHER TORTIOUS ACTION, ARISING OUT OF OR IN CONNECTION WITH THE USE + * OR PERFORMANCE OF THIS SOURCE CODE. * - * U.S. Government End Users. This source code is a "commercial item" as - * that term is defined at 48 C.F.R. 2.101 (OCT 1995), consisting of - * "commercial computer software" and "commercial computer software - * documentation" as such terms are used in 48 C.F.R. 12.212 (SEPT 1995) - * and is provided to the U.S. Government only as a commercial end item. - * Consistent with 48 C.F.R.12.212 and 48 C.F.R. 227.7202-1 through - * 227.7202-4 (JUNE 1995), all U.S. Government End Users acquire the - * source code with only those rights set forth herein. + * U.S. Government End Users. This source code is a "commercial item" as + * that term is defined at 48 C.F.R. 2.101 (OCT 1995), consisting of + * "commercial computer software" and "commercial computer software + * documentation" as such terms are used in 48 C.F.R. 12.212 (SEPT 1995) + * and is provided to the U.S. Government only as a commercial end item. + * Consistent with 48 C.F.R.12.212 and 48 C.F.R. 227.7202-1 through + * 227.7202-4 (JUNE 1995), all U.S. Government End Users acquire the + * source code with only those rights set forth herein. * - * Any use of this source code in individual and commercial software must + * Any use of this source code in individual and commercial software must * include, in the user documentation and internal comments to the code, * the above Disclaimer and U.S. Government End Users Notice. */ -#include +#include +#include #include +#include #include -#include #include -#include +#include #include -#include #include #include -#include +#include #ifdef OPENGL_SUPPORT #define GL_GLEXT_PROTOTYPES #ifdef __APPLE__ -#include // Apple's version of GLUT is here +#include // Apple's version of GLUT is here #else #include #endif @@ -150,23 +151,20 @@ #include #endif - /*DEVICE_BUILTIN*/ -struct cudaArray -{ - void *devPtr; - int devPtr32; - struct cudaChannelFormatDesc desc; - int width; - int height; - int size; //in bytes - unsigned dimensions; +struct cudaArray { + void *devPtr; + int devPtr32; + struct cudaChannelFormatDesc desc; + int width; + int height; + int size; // in bytes + unsigned dimensions; }; #if !defined(__dv) #if defined(__cplusplus) -#define __dv(v) \ - = v +#define __dv(v) = v #else /* __cplusplus */ #define __dv(v) #endif /* __cplusplus */ @@ -174,1962 +172,2115 @@ struct cudaArray cudaError_t g_last_cudaError = cudaSuccess; -void register_ptx_function( const char *name, function_info *impl ) -{ - // no longer need this +void register_ptx_function(const char *name, function_info *impl) { + // no longer need this } #if defined __APPLE__ -# define __my_func__ __PRETTY_FUNCTION__ +#define __my_func__ __PRETTY_FUNCTION__ +#else +#if defined __cplusplus ? __GNUC_PREREQ(2, 6) : __GNUC_PREREQ(2, 4) +#define __my_func__ __PRETTY_FUNCTION__ #else -# if defined __cplusplus ? __GNUC_PREREQ (2, 6) : __GNUC_PREREQ (2, 4) -# define __my_func__ __PRETTY_FUNCTION__ -# else -# if defined __STDC_VERSION__ && __STDC_VERSION__ >= 199901L -# define __my_func__ __func__ -# else -# define __my_func__ ((__const char *) 0) -# endif -# endif +#if defined __STDC_VERSION__ && __STDC_VERSION__ >= 199901L +#define __my_func__ __func__ +#else +#define __my_func__ ((__const char *)0) +#endif +#endif #endif -struct _cuda_device_id *gpgpu_context::GPGPUSim_Init() -{ - _cuda_device_id *the_device = the_gpgpusim->the_cude_device; - if( !the_device ) { - gpgpu_sim *the_gpu = gpgpu_ptx_sim_init_perf(); - - cudaDeviceProp *prop = (cudaDeviceProp *) calloc(sizeof(cudaDeviceProp),1); - snprintf(prop->name,256,"GPGPU-Sim_v%s", g_gpgpusim_version_string ); - prop->major = the_gpu->compute_capability_major(); - prop->minor = the_gpu->compute_capability_minor(); - prop->totalGlobalMem = 0x80000000 /* 2 GB */; - prop->memPitch = 0; - if(prop->major >= 2) { - prop->maxThreadsPerBlock = 1024; - prop->maxThreadsDim[0] = 1024; - prop->maxThreadsDim[1] = 1024; - } - else - { - prop->maxThreadsPerBlock = 512; - prop->maxThreadsDim[0] = 512; - prop->maxThreadsDim[1] = 512; - } - - prop->maxThreadsDim[2] = 64; - prop->maxGridSize[0] = 0x40000000; - prop->maxGridSize[1] = 0x40000000; - prop->maxGridSize[2] = 0x40000000; - prop->totalConstMem = 0x40000000; - prop->textureAlignment = 0; -// * TODO: Update the .config and xml files of all GPU config files with new value of sharedMemPerBlock and regsPerBlock - prop->sharedMemPerBlock = the_gpu->shared_mem_per_block(); +struct _cuda_device_id *gpgpu_context::GPGPUSim_Init() { + _cuda_device_id *the_device = the_gpgpusim->the_cude_device; + if (!the_device) { + gpgpu_sim *the_gpu = gpgpu_ptx_sim_init_perf(); + + cudaDeviceProp *prop = (cudaDeviceProp *)calloc(sizeof(cudaDeviceProp), 1); + snprintf(prop->name, 256, "GPGPU-Sim_v%s", g_gpgpusim_version_string); + prop->major = the_gpu->compute_capability_major(); + prop->minor = the_gpu->compute_capability_minor(); + prop->totalGlobalMem = 0x80000000 /* 2 GB */; + prop->memPitch = 0; + if (prop->major >= 2) { + prop->maxThreadsPerBlock = 1024; + prop->maxThreadsDim[0] = 1024; + prop->maxThreadsDim[1] = 1024; + } else { + prop->maxThreadsPerBlock = 512; + prop->maxThreadsDim[0] = 512; + prop->maxThreadsDim[1] = 512; + } + + prop->maxThreadsDim[2] = 64; + prop->maxGridSize[0] = 0x40000000; + prop->maxGridSize[1] = 0x40000000; + prop->maxGridSize[2] = 0x40000000; + prop->totalConstMem = 0x40000000; + prop->textureAlignment = 0; + // * TODO: Update the .config and xml files of all GPU config files + // with new value of sharedMemPerBlock and regsPerBlock + prop->sharedMemPerBlock = the_gpu->shared_mem_per_block(); #if (CUDART_VERSION > 5050) - prop->regsPerMultiprocessor = the_gpu->num_registers_per_core(); - prop->sharedMemPerMultiprocessor = the_gpu->shared_mem_size(); -#endif - prop->sharedMemPerBlock = the_gpu->shared_mem_per_block(); - prop->regsPerBlock = the_gpu->num_registers_per_block(); - prop->warpSize = the_gpu->wrp_size(); - prop->clockRate = the_gpu->shader_clock(); + prop->regsPerMultiprocessor = the_gpu->num_registers_per_core(); + prop->sharedMemPerMultiprocessor = the_gpu->shared_mem_size(); +#endif + prop->sharedMemPerBlock = the_gpu->shared_mem_per_block(); + prop->regsPerBlock = the_gpu->num_registers_per_block(); + prop->warpSize = the_gpu->wrp_size(); + prop->clockRate = the_gpu->shader_clock(); #if (CUDART_VERSION >= 2010) - prop->multiProcessorCount = the_gpu->get_config().num_shader(); + prop->multiProcessorCount = the_gpu->get_config().num_shader(); #endif #if (CUDART_VERSION >= 4000) - prop->maxThreadsPerMultiProcessor = the_gpu->threads_per_core(); + prop->maxThreadsPerMultiProcessor = the_gpu->threads_per_core(); #endif - the_gpu->set_prop(prop); - the_gpgpusim->the_cude_device = new _cuda_device_id(the_gpu); - the_device = the_gpgpusim->the_cude_device; - } - start_sim_thread(1); - return the_device; -} - -CUctx_st* GPGPUSim_Context(gpgpu_context * ctx) -{ - //static CUctx_st *the_context = NULL; - CUctx_st *the_context = ctx->the_gpgpusim->the_context; - if( the_context == NULL ) { - _cuda_device_id *the_gpu = ctx->GPGPUSim_Init(); - ctx->the_gpgpusim->the_context = new CUctx_st(the_gpu); - the_context = ctx->the_gpgpusim->the_context; - } - return the_context; -} - -gpgpu_context* GPGPU_Context() -{ - static gpgpu_context *gpgpu_ctx = NULL; - if( gpgpu_ctx == NULL ) { - gpgpu_ctx = new gpgpu_context(); - } - return gpgpu_ctx; + the_gpu->set_prop(prop); + the_gpgpusim->the_cude_device = new _cuda_device_id(the_gpu); + the_device = the_gpgpusim->the_cude_device; + } + start_sim_thread(1); + return the_device; +} + +CUctx_st *GPGPUSim_Context(gpgpu_context *ctx) { + // static CUctx_st *the_context = NULL; + CUctx_st *the_context = ctx->the_gpgpusim->the_context; + if (the_context == NULL) { + _cuda_device_id *the_gpu = ctx->GPGPUSim_Init(); + ctx->the_gpgpusim->the_context = new CUctx_st(the_gpu); + the_context = ctx->the_gpgpusim->the_context; + } + return the_context; +} + +gpgpu_context *GPGPU_Context() { + static gpgpu_context *gpgpu_ctx = NULL; + if (gpgpu_ctx == NULL) { + gpgpu_ctx = new gpgpu_context(); + } + return gpgpu_ctx; +} + +void ptxinfo_data::ptxinfo_addinfo() { + CUctx_st *context = GPGPUSim_Context(gpgpu_ctx); + if (!get_ptxinfo_kname()) { + /* This info is not per kernel (since CUDA 5.0 some info (e.g. gmem, and + * cmem) is added at the beginning for the whole binary ) */ + print_ptxinfo(); + context->add_ptxinfo(get_ptxinfo()); + clear_ptxinfo(); + return; + } + if (!strcmp("__cuda_dummy_entry__", get_ptxinfo_kname())) { + // this string produced by ptxas for empty ptx files (e.g., bandwidth test) + clear_ptxinfo(); + return; + } + print_ptxinfo(); + context->add_ptxinfo(get_ptxinfo_kname(), get_ptxinfo()); + clear_ptxinfo(); } - void ptxinfo_data::ptxinfo_addinfo() -{ - CUctx_st *context = GPGPUSim_Context(gpgpu_ctx); - if(!get_ptxinfo_kname()){ - /* This info is not per kernel (since CUDA 5.0 some info (e.g. gmem, and cmem) is added at the beginning for the whole binary ) */ - print_ptxinfo(); - context->add_ptxinfo(get_ptxinfo()); - clear_ptxinfo(); - return; - } - if( !strcmp("__cuda_dummy_entry__",get_ptxinfo_kname()) ) { - // this string produced by ptxas for empty ptx files (e.g., bandwidth test) - clear_ptxinfo(); - return; - } - print_ptxinfo(); - context->add_ptxinfo( get_ptxinfo_kname(), get_ptxinfo() ); - clear_ptxinfo(); -} - -void cuda_not_implemented( const char* func, unsigned line ) -{ - fflush(stdout); - fflush(stderr); - printf("\n\nGPGPU-Sim PTX: Execution error: CUDA API function \"%s()\" has not been implemented yet.\n" - " [$GPGPUSIM_ROOT/libcuda/%s around line %u]\n\n\n", - func,__FILE__, line ); - fflush(stdout); - abort(); +void cuda_not_implemented(const char *func, unsigned line) { + fflush(stdout); + fflush(stderr); + printf( + "\n\nGPGPU-Sim PTX: Execution error: CUDA API function \"%s()\" has not " + "been implemented yet.\n" + " [$GPGPUSIM_ROOT/libcuda/%s around line %u]\n\n\n", + func, __FILE__, line); + fflush(stdout); + abort(); } -void announce_call( const char* func ) -{ - printf("\n\nGPGPU-Sim PTX: CUDA API function \"%s\" has been called.\n", func); - fflush(stdout); +void announce_call(const char *func) { + printf("\n\nGPGPU-Sim PTX: CUDA API function \"%s\" has been called.\n", + func); + fflush(stdout); } -#define gpgpusim_ptx_error(msg, ...) gpgpusim_ptx_error_impl(__func__, __FILE__,__LINE__, msg, ##__VA_ARGS__) -#define gpgpusim_ptx_assert(cond,msg, ...) gpgpusim_ptx_assert_impl((cond),__func__, __FILE__,__LINE__, msg, ##__VA_ARGS__) +#define gpgpusim_ptx_error(msg, ...) \ + gpgpusim_ptx_error_impl(__func__, __FILE__, __LINE__, msg, ##__VA_ARGS__) +#define gpgpusim_ptx_assert(cond, msg, ...) \ + gpgpusim_ptx_assert_impl((cond), __func__, __FILE__, __LINE__, msg, \ + ##__VA_ARGS__) -void gpgpusim_ptx_error_impl( const char *func, const char *file, unsigned line, const char *msg, ... ) -{ - va_list ap; - char buf[1024]; - va_start(ap,msg); - vsnprintf(buf,1024,msg,ap); - va_end(ap); +void gpgpusim_ptx_error_impl(const char *func, const char *file, unsigned line, + const char *msg, ...) { + va_list ap; + char buf[1024]; + va_start(ap, msg); + vsnprintf(buf, 1024, msg, ap); + va_end(ap); - printf("GPGPU-Sim CUDA API: %s\n", buf); - printf(" [%s:%u : %s]\n", file, line, func ); - abort(); + printf("GPGPU-Sim CUDA API: %s\n", buf); + printf(" [%s:%u : %s]\n", file, line, func); + abort(); } -void gpgpusim_ptx_assert_impl( int test_value, const char *func, const char *file, unsigned line, const char *msg, ... ) -{ - va_list ap; - char buf[1024]; - va_start(ap,msg); - vsnprintf(buf,1024,msg,ap); - va_end(ap); +void gpgpusim_ptx_assert_impl(int test_value, const char *func, + const char *file, unsigned line, const char *msg, + ...) { + va_list ap; + char buf[1024]; + va_start(ap, msg); + vsnprintf(buf, 1024, msg, ap); + va_end(ap); - if ( test_value == 0 ) - gpgpusim_ptx_error_impl(func, file, line, msg); + if (test_value == 0) gpgpusim_ptx_error_impl(func, file, line, msg); } - -typedef std::map event_tracker_t; +typedef std::map event_tracker_t; int CUevent_st::m_next_event_uid; event_tracker_t g_timer_events; -extern int cuobjdump_lex_init(yyscan_t* scanner); -extern void cuobjdump_set_in (FILE * _in_str ,yyscan_t yyscanner ); -extern int cuobjdump_parse(yyscan_t scanner, struct cuobjdump_parser* parser, std::list &cuobjdumpSectionList); +extern int cuobjdump_lex_init(yyscan_t *scanner); +extern void cuobjdump_set_in(FILE *_in_str, yyscan_t yyscanner); +extern int cuobjdump_parse(yyscan_t scanner, struct cuobjdump_parser *parser, + std::list &cuobjdumpSectionList); extern int cuobjdump_lex_destroy(yyscan_t scanner); -enum cuobjdumpSectionType { - PTXSECTION=0, - ELFSECTION -}; - +enum cuobjdumpSectionType { PTXSECTION = 0, ELFSECTION }; // sectiontype: 0 for ptx, 1 for elf -void addCuobjdumpSection(int sectiontype, std::list &cuobjdumpSectionList){ - if (sectiontype) - cuobjdumpSectionList.push_front(new cuobjdumpELFSection()); - else - cuobjdumpSectionList.push_front(new cuobjdumpPTXSection()); - printf("## Adding new section %s\n", sectiontype?"ELF":"PTX"); -} - -void setCuobjdumparch(const char* arch, std::list &cuobjdumpSectionList){ - unsigned archnum; - sscanf(arch, "sm_%u", &archnum); - assert (archnum && "cannot have sm_0"); - printf("Adding arch: %s\n", arch); - cuobjdumpSectionList.front()->setArch(archnum); -} - -void setCuobjdumpidentifier(const char* identifier, std::list &cuobjdumpSectionList){ - printf("Adding identifier: %s\n", identifier); - cuobjdumpSectionList.front()->setIdentifier(identifier); -} - -void setCuobjdumpptxfilename(const char* filename, std::list &cuobjdumpSectionList){ - printf("Adding ptx filename: %s\n", filename); - cuobjdumpSection* x = cuobjdumpSectionList.front(); - if (dynamic_cast(x) == NULL){ - assert (0 && "You shouldn't be trying to add a ptxfilename to an elf section"); - } - (dynamic_cast(x))->setPTXfilename(filename); -} - -void setCuobjdumpelffilename(const char* filename, std::list &cuobjdumpSectionList){ - if (dynamic_cast(cuobjdumpSectionList.front()) == NULL){ - assert (0 && "You shouldn't be trying to add a elffilename to an ptx section"); - } - (dynamic_cast(cuobjdumpSectionList.front()))->setELFfilename(filename); -} - -void setCuobjdumpsassfilename(const char* filename, std::list &cuobjdumpSectionList){ - if (dynamic_cast(cuobjdumpSectionList.front()) == NULL){ - assert (0 && "You shouldn't be trying to add a sassfilename to an ptx section"); - } - (dynamic_cast(cuobjdumpSectionList.front()))->setSASSfilename(filename); -} - -//! Return the executable file of the process containing the PTX/SASS code +void addCuobjdumpSection(int sectiontype, + std::list &cuobjdumpSectionList) { + if (sectiontype) + cuobjdumpSectionList.push_front(new cuobjdumpELFSection()); + else + cuobjdumpSectionList.push_front(new cuobjdumpPTXSection()); + printf("## Adding new section %s\n", sectiontype ? "ELF" : "PTX"); +} + +void setCuobjdumparch(const char *arch, + std::list &cuobjdumpSectionList) { + unsigned archnum; + sscanf(arch, "sm_%u", &archnum); + assert(archnum && "cannot have sm_0"); + printf("Adding arch: %s\n", arch); + cuobjdumpSectionList.front()->setArch(archnum); +} + +void setCuobjdumpidentifier( + const char *identifier, + std::list &cuobjdumpSectionList) { + printf("Adding identifier: %s\n", identifier); + cuobjdumpSectionList.front()->setIdentifier(identifier); +} + +void setCuobjdumpptxfilename( + const char *filename, std::list &cuobjdumpSectionList) { + printf("Adding ptx filename: %s\n", filename); + cuobjdumpSection *x = cuobjdumpSectionList.front(); + if (dynamic_cast(x) == NULL) { + assert(0 && + "You shouldn't be trying to add a ptxfilename to an elf section"); + } + (dynamic_cast(x))->setPTXfilename(filename); +} + +void setCuobjdumpelffilename( + const char *filename, std::list &cuobjdumpSectionList) { + if (dynamic_cast(cuobjdumpSectionList.front()) == + NULL) { + assert(0 && + "You shouldn't be trying to add a elffilename to an ptx section"); + } + (dynamic_cast(cuobjdumpSectionList.front())) + ->setELFfilename(filename); +} + +void setCuobjdumpsassfilename( + const char *filename, std::list &cuobjdumpSectionList) { + if (dynamic_cast(cuobjdumpSectionList.front()) == + NULL) { + assert(0 && + "You shouldn't be trying to add a sassfilename to an ptx section"); + } + (dynamic_cast(cuobjdumpSectionList.front())) + ->setSASSfilename(filename); +} + +//! Return the executable file of the process containing the PTX/SASS code //! //! This Function returns the executable file ran by the process. This //! executable is supposed to contain the PTX/SASS code. It provides workaround -//! for processes running on valgrind by dereferencing /proc//exe within the -//! GPGPU-Sim process before calling cuobjdump to extract PTX/SASS. This is +//! for processes running on valgrind by dereferencing /proc//exe within +//! the GPGPU-Sim process before calling cuobjdump to extract PTX/SASS. This is //! needed because valgrind uses x86 emulation to detect memory leak. Other //! processes (e.g. cuobjdump) reading /proc//exe will see the emulator -//! executable instead of the application binary. -//! -std::string get_app_binary(){ - char self_exe_path[1025]; +//! executable instead of the application binary. +//! +std::string get_app_binary() { + char self_exe_path[1025]; #ifdef __APPLE__ - uint32_t size = sizeof(self_exe_path); - if( _NSGetExecutablePath(self_exe_path,&size) != 0 ) { - printf("GPGPU-Sim ** ERROR: _NSGetExecutablePath input buffer too small\n"); - exit(1); - } + uint32_t size = sizeof(self_exe_path); + if (_NSGetExecutablePath(self_exe_path, &size) != 0) { + printf("GPGPU-Sim ** ERROR: _NSGetExecutablePath input buffer too small\n"); + exit(1); + } #else - std::stringstream exec_link; - exec_link << "/proc/self/exe"; + std::stringstream exec_link; + exec_link << "/proc/self/exe"; - ssize_t path_length = readlink(exec_link.str().c_str(), self_exe_path, 1024); - assert(path_length != -1); - self_exe_path[path_length] = '\0'; + ssize_t path_length = readlink(exec_link.str().c_str(), self_exe_path, 1024); + assert(path_length != -1); + self_exe_path[path_length] = '\0'; #endif - printf("self exe links to: %s\n", self_exe_path); - return self_exe_path; + printf("self exe links to: %s\n", self_exe_path); + return self_exe_path; } -//above func gives abs path whereas this give just the name of application. -char* get_app_binary_name(std::string abs_path){ - char *self_exe_path; +// above func gives abs path whereas this give just the name of application. +char *get_app_binary_name(std::string abs_path) { + char *self_exe_path; #ifdef __APPLE__ - //TODO: get apple device and check the result. - printf("WARNING: not tested for Apple-mac devices \n"); - abort(); + // TODO: get apple device and check the result. + printf("WARNING: not tested for Apple-mac devices \n"); + abort(); #else - char* buf = strdup(abs_path.c_str()); - char *token = strtok(buf, "/"); - while(token !=NULL){ - self_exe_path = token; - token = strtok(NULL,"/"); - } + char *buf = strdup(abs_path.c_str()); + char *token = strtok(buf, "/"); + while (token != NULL) { + self_exe_path = token; + token = strtok(NULL, "/"); + } #endif - self_exe_path = strtok(self_exe_path, "."); - printf("self exe links to: %s\n", self_exe_path); - return self_exe_path; + self_exe_path = strtok(self_exe_path, "."); + printf("self exe links to: %s\n", self_exe_path); + return self_exe_path; } static int get_app_cuda_version() { - int app_cuda_version = 0; - char fname[1024]; - snprintf(fname,1024,"_app_cuda_version_XXXXXX"); - int fd=mkstemp(fname); - close(fd); - std::string app_cuda_version_command = "ldd " + get_app_binary() + " | grep libcudart.so | sed 's/.*libcudart.so.\\(.*\\) =>.*/\\1/' > " + fname; - system(app_cuda_version_command.c_str()); - FILE * cmd = fopen(fname, "r"); - char buf[256]; - while (fgets(buf, sizeof(buf), cmd) != 0) { - std::cout << buf; - app_cuda_version = atoi(buf); - } - fclose(cmd); - if ( app_cuda_version == 0 ) { - printf( "Error - Cannot detect the app's CUDA version.\n" ); - exit(1); - } - return app_cuda_version; + int app_cuda_version = 0; + char fname[1024]; + snprintf(fname, 1024, "_app_cuda_version_XXXXXX"); + int fd = mkstemp(fname); + close(fd); + std::string app_cuda_version_command = + "ldd " + get_app_binary() + + " | grep libcudart.so | sed 's/.*libcudart.so.\\(.*\\) =>.*/\\1/' > " + + fname; + system(app_cuda_version_command.c_str()); + FILE *cmd = fopen(fname, "r"); + char buf[256]; + while (fgets(buf, sizeof(buf), cmd) != 0) { + std::cout << buf; + app_cuda_version = atoi(buf); + } + fclose(cmd); + if (app_cuda_version == 0) { + printf("Error - Cannot detect the app's CUDA version.\n"); + exit(1); + } + return app_cuda_version; } //! Keep track of the association between filename and cubin handle -void cuda_runtime_api::cuobjdumpRegisterFatBinary(unsigned int handle, const char* filename, CUctx_st *context){ - fatbinmap[handle] = filename; +void cuda_runtime_api::cuobjdumpRegisterFatBinary(unsigned int handle, + const char *filename, + CUctx_st *context) { + fatbinmap[handle] = filename; } - - /******************************************************************************* - * Add internal cuda runtime API call to accept gpgpu_context * + * Add internal cuda runtime API call to accept gpgpu_context * *******************************************************************************/ -cudaError_t cudaSetDeviceInternal(int device, gpgpu_context* gpgpu_ctx = NULL) -{ - gpgpu_context *ctx; - if (gpgpu_ctx){ - ctx = gpgpu_ctx; - } else { - ctx = GPGPU_Context(); - } - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - //set the active device to run cuda - if ( device <= ctx->GPGPUSim_Init()->num_devices() ) { - ctx->api->g_active_device = device; - return g_last_cudaError = cudaSuccess; - } else { - return g_last_cudaError = cudaErrorInvalidDevice; - } -} - -cudaError_t cudaGetDeviceInternal(int *device, gpgpu_context* gpgpu_ctx = NULL) -{ - gpgpu_context *ctx; - if (gpgpu_ctx){ - ctx = gpgpu_ctx; - } else { - ctx = GPGPU_Context(); - } - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - *device = ctx->api->g_active_device; - return g_last_cudaError = cudaSuccess; -} - -__host__ cudaError_t CUDARTAPI cudaDeviceGetLimitInternal( size_t* pValue, cudaLimit limit, gpgpu_context* gpgpu_ctx = NULL ) -{ - gpgpu_context *ctx; - if (gpgpu_ctx){ - ctx = gpgpu_ctx; - } else { - ctx = GPGPU_Context(); - } - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - _cuda_device_id *dev = ctx->GPGPUSim_Init(); - const struct cudaDeviceProp *prop = dev->get_prop(); - const gpgpu_sim_config& config=dev->get_gpgpu()->get_config(); - switch(limit) { - case 0: // cudaLimitStackSize - *pValue=config.stack_limit(); - break; - case 2: // cudaLimitMallocHeapSize - *pValue=config.heap_limit(); - break; +cudaError_t cudaSetDeviceInternal(int device, gpgpu_context *gpgpu_ctx = NULL) { + gpgpu_context *ctx; + if (gpgpu_ctx) { + ctx = gpgpu_ctx; + } else { + ctx = GPGPU_Context(); + } + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + // set the active device to run cuda + if (device <= ctx->GPGPUSim_Init()->num_devices()) { + ctx->api->g_active_device = device; + return g_last_cudaError = cudaSuccess; + } else { + return g_last_cudaError = cudaErrorInvalidDevice; + } +} + +cudaError_t cudaGetDeviceInternal(int *device, + gpgpu_context *gpgpu_ctx = NULL) { + gpgpu_context *ctx; + if (gpgpu_ctx) { + ctx = gpgpu_ctx; + } else { + ctx = GPGPU_Context(); + } + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + *device = ctx->api->g_active_device; + return g_last_cudaError = cudaSuccess; +} + +__host__ cudaError_t CUDARTAPI cudaDeviceGetLimitInternal( + size_t *pValue, cudaLimit limit, gpgpu_context *gpgpu_ctx = NULL) { + gpgpu_context *ctx; + if (gpgpu_ctx) { + ctx = gpgpu_ctx; + } else { + ctx = GPGPU_Context(); + } + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + _cuda_device_id *dev = ctx->GPGPUSim_Init(); + const struct cudaDeviceProp *prop = dev->get_prop(); + const gpgpu_sim_config &config = dev->get_gpgpu()->get_config(); + switch (limit) { + case 0: // cudaLimitStackSize + *pValue = config.stack_limit(); + break; + case 2: // cudaLimitMallocHeapSize + *pValue = config.heap_limit(); + break; #if (CUDART_VERSION > 5050) - case 3: // cudaLimitDevRuntimeSyncDepth - if(prop->major > 2){ - *pValue=config.sync_depth_limit(); - break; - } - else{ - printf("ERROR:Limit %d is not supported on this architecture \n", limit); - abort(); - } - case 4: // cudaLimitDevRuntimePendingLaunchCount - if(prop->major > 2){ - *pValue=config.pending_launch_count_limit(); - break; - } - else{ - printf("ERROR:Limit %d is not supported on this architecture \n",limit); - abort(); - } + case 3: // cudaLimitDevRuntimeSyncDepth + if (prop->major > 2) { + *pValue = config.sync_depth_limit(); + break; + } else { + printf("ERROR:Limit %d is not supported on this architecture \n", + limit); + abort(); + } + case 4: // cudaLimitDevRuntimePendingLaunchCount + if (prop->major > 2) { + *pValue = config.pending_launch_count_limit(); + break; + } else { + printf("ERROR:Limit %d is not supported on this architecture \n", + limit); + abort(); + } #endif - default: - printf("ERROR:Limit %d unimplemented \n",limit); - abort(); - } - return g_last_cudaError = cudaSuccess; - -} - - -void** cudaRegisterFatBinaryInternal( void *fatCubin, gpgpu_context* gpgpu_ctx = NULL) -{ - gpgpu_context *ctx; - if (gpgpu_ctx){ - ctx = gpgpu_ctx; - } else { - ctx = GPGPU_Context(); - } - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } + default: + printf("ERROR:Limit %d unimplemented \n", limit); + abort(); + } + return g_last_cudaError = cudaSuccess; +} + +void **cudaRegisterFatBinaryInternal(void *fatCubin, + gpgpu_context *gpgpu_ctx = NULL) { + gpgpu_context *ctx; + if (gpgpu_ctx) { + ctx = gpgpu_ctx; + } else { + ctx = GPGPU_Context(); + } + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } #if (CUDART_VERSION < 2010) - printf("GPGPU-Sim PTX: ERROR ** this version of GPGPU-Sim requires CUDA 2.1 or higher\n"); - exit(1); + printf( + "GPGPU-Sim PTX: ERROR ** this version of GPGPU-Sim requires CUDA 2.1 or " + "higher\n"); + exit(1); #endif - CUctx_st *context = GPGPUSim_Context(ctx); - static unsigned next_fat_bin_handle = 1; - if(context->get_device()->get_gpgpu()->get_config().use_cuobjdump()) { - // The following workaround has only been verified on 64-bit systems. - if (sizeof(void*) == 4) - printf("GPGPU-Sim PTX: FatBin file name extraction has not been tested on 32-bit system.\n"); - - // This code will get the CUDA version the app was compiled with. - // We need this to determine how to handle the parsing of the binary. - // Making this a runtime variable based on the app, enables GPGPU-Sim compiled - // with a newer version of CUDA to run apps compiled with older versions of - // CUDA. This is especially useful for PTXPLUS execution. - //Skip cuda version check for pytorch application - std::string app_binary_path = get_app_binary(); - int pos = app_binary_path.find("python"); - if (pos==std::string::npos){ - // Not pytorch app : checking cuda version - int app_cuda_version = get_app_cuda_version(); - assert( app_cuda_version == CUDART_VERSION / 1000 && "The app must be compiled with same major version as the simulator." ); - } - - //int app_cuda_version = get_app_cuda_version(); - //assert( app_cuda_version == CUDART_VERSION / 1000 && "The app must be compiled with same major version as the simulator." ); - const char* filename; + CUctx_st *context = GPGPUSim_Context(ctx); + static unsigned next_fat_bin_handle = 1; + if (context->get_device()->get_gpgpu()->get_config().use_cuobjdump()) { + // The following workaround has only been verified on 64-bit systems. + if (sizeof(void *) == 4) + printf( + "GPGPU-Sim PTX: FatBin file name extraction has not been tested on " + "32-bit system.\n"); + + // This code will get the CUDA version the app was compiled with. + // We need this to determine how to handle the parsing of the binary. + // Making this a runtime variable based on the app, enables GPGPU-Sim + // compiled with a newer version of CUDA to run apps compiled with older + // versions of CUDA. This is especially useful for PTXPLUS execution. + // Skip cuda version check for pytorch application + std::string app_binary_path = get_app_binary(); + int pos = app_binary_path.find("python"); + if (pos == std::string::npos) { + // Not pytorch app : checking cuda version + int app_cuda_version = get_app_cuda_version(); + assert( + app_cuda_version == CUDART_VERSION / 1000 && + "The app must be compiled with same major version as the simulator."); + } + + // int app_cuda_version = get_app_cuda_version(); + // assert( app_cuda_version == CUDART_VERSION / 1000 && "The app must be + // compiled with same major version as the simulator." ); + const char *filename; #if CUDART_VERSION < 6000 - // FatBin handle from the .fatbin.c file (one of the intermediate files generated by NVCC) - typedef struct {int m; int v; const unsigned long long* d; char* f;} __fatDeviceText __attribute__ ((aligned (8))); - __fatDeviceText * fatDeviceText = (__fatDeviceText *) fatCubin; - - // Extract the source code file name that generate the given FatBin. - // - Obtains the pointer to the actual fatbin structure from the FatBin handle (fatCubin). - // - An integer inside the fatbin structure contains the relative offset to the source code file name. - // - This offset differs among different CUDA and GCC versions. - char * pfatbin = (char*) fatDeviceText->d; - int offset = *((int*)(pfatbin+48)); - filename = (pfatbin+16+offset); + // FatBin handle from the .fatbin.c file (one of the intermediate files + // generated by NVCC) + typedef struct { + int m; + int v; + const unsigned long long *d; + char *f; + } __fatDeviceText __attribute__((aligned(8))); + __fatDeviceText *fatDeviceText = (__fatDeviceText *)fatCubin; + + // Extract the source code file name that generate the given FatBin. + // - Obtains the pointer to the actual fatbin structure from the FatBin + // handle (fatCubin). + // - An integer inside the fatbin structure contains the relative offset to + // the source code file name. + // - This offset differs among different CUDA and GCC versions. + char *pfatbin = (char *)fatDeviceText->d; + int offset = *((int *)(pfatbin + 48)); + filename = (pfatbin + 16 + offset); #else - filename = "default"; + filename = "default"; #endif - // The extracted file name is associated with a fat_cubin_handle passed - // into cudaLaunch(). Inside cudaLaunch(), the associated file name is - // used to find the PTX/SASS section from cuobjdump, which contains the - // PTX/SASS code for the launched kernel function. - // This allows us to work around the fact that cuobjdump only outputs the - // file name associated with each section. - unsigned long long fat_cubin_handle = next_fat_bin_handle; - next_fat_bin_handle++; - printf("GPGPU-Sim PTX: __cudaRegisterFatBinary, fat_cubin_handle = %llu, filename=%s\n", fat_cubin_handle, filename); - /*! - * This function extracts all data from all files in first call - * then for next calls, only returns the appropriate number - */ - assert(fat_cubin_handle >= 1); - if (fat_cubin_handle==1) ctx->api->cuobjdumpInit(); - ctx->api->cuobjdumpRegisterFatBinary(fat_cubin_handle, filename, context); - - return (void**)fat_cubin_handle; - } + // The extracted file name is associated with a fat_cubin_handle passed + // into cudaLaunch(). Inside cudaLaunch(), the associated file name is + // used to find the PTX/SASS section from cuobjdump, which contains the + // PTX/SASS code for the launched kernel function. + // This allows us to work around the fact that cuobjdump only outputs the + // file name associated with each section. + unsigned long long fat_cubin_handle = next_fat_bin_handle; + next_fat_bin_handle++; + printf( + "GPGPU-Sim PTX: __cudaRegisterFatBinary, fat_cubin_handle = %llu, " + "filename=%s\n", + fat_cubin_handle, filename); + /*! + * This function extracts all data from all files in first call + * then for next calls, only returns the appropriate number + */ + assert(fat_cubin_handle >= 1); + if (fat_cubin_handle == 1) ctx->api->cuobjdumpInit(); + ctx->api->cuobjdumpRegisterFatBinary(fat_cubin_handle, filename, context); + + return (void **)fat_cubin_handle; + } #if (CUDART_VERSION < 8000) - else { - static unsigned source_num=1; - unsigned long long fat_cubin_handle = next_fat_bin_handle++; - __cudaFatCudaBinary *info = (__cudaFatCudaBinary *)fatCubin; - assert( info->version >= 3 ); - unsigned num_ptx_versions=0; - unsigned max_capability=0; - unsigned selected_capability=0; - bool found=false; - unsigned forced_max_capability = context->get_device()->get_gpgpu()->get_config().get_forced_max_capability(); - if (!info->ptx){ - printf("ERROR: Cannot find ptx code in cubin file\n" - "\tIf you are using CUDA 4.0 or higher, please enable -gpgpu_ptx_use_cuobjdump or downgrade to CUDA 3.1\n"); - exit(1); - } - while( info->ptx[num_ptx_versions].gpuProfileName != NULL ) { - unsigned capability=0; - sscanf(info->ptx[num_ptx_versions].gpuProfileName,"compute_%u",&capability); - printf("GPGPU-Sim PTX: __cudaRegisterFatBinary found PTX versions for '%s', ", info->ident); - printf("capability = %s\n", info->ptx[num_ptx_versions].gpuProfileName ); - if( forced_max_capability ) { - if( capability > max_capability && capability <= forced_max_capability ) { - found = true; - max_capability=capability; - selected_capability = num_ptx_versions; - } - } else { - if( capability > max_capability ) { - found = true; - max_capability=capability; - selected_capability = num_ptx_versions; - } - } - num_ptx_versions++; - } - if( found ) { - printf("GPGPU-Sim PTX: Loading PTX for %s, capability = %s\n", - info->ident, info->ptx[selected_capability].gpuProfileName ); - symbol_table *symtab; - const char *ptx = info->ptx[selected_capability].ptx; - if(context->get_device()->get_gpgpu()->get_config().convert_to_ptxplus() ) { - printf("GPGPU-Sim PTX: ERROR ** PTXPlus is only supported through cuobjdump\n" - "\tEither enable cuobjdump or disable PTXPlus in your configuration file\n"); - exit(1); - } else { - symtab=ctx->gpgpu_ptx_sim_load_ptx_from_string(ptx,source_num); - context->add_binary(symtab,fat_cubin_handle); - ctx->gpgpu_ptxinfo_load_from_string( ptx, source_num, max_capability, context->no_of_ptx ); - } - source_num++; - ctx->api->load_static_globals(symtab,STATIC_ALLOC_LIMIT,0xFFFFFFFF,context->get_device()->get_gpgpu()); - ctx->api->load_constants(symtab,STATIC_ALLOC_LIMIT,context->get_device()->get_gpgpu()); - } else { - printf("GPGPU-Sim PTX: warning -- did not find an appropriate PTX in cubin\n"); - } - return (void**)fat_cubin_handle; - } -#else - else { - printf("ERROR ** __cudaRegisterFatBinary() needs to be updated\n"); - abort(); + else { + static unsigned source_num = 1; + unsigned long long fat_cubin_handle = next_fat_bin_handle++; + __cudaFatCudaBinary *info = (__cudaFatCudaBinary *)fatCubin; + assert(info->version >= 3); + unsigned num_ptx_versions = 0; + unsigned max_capability = 0; + unsigned selected_capability = 0; + bool found = false; + unsigned forced_max_capability = context->get_device() + ->get_gpgpu() + ->get_config() + .get_forced_max_capability(); + if (!info->ptx) { + printf( + "ERROR: Cannot find ptx code in cubin file\n" + "\tIf you are using CUDA 4.0 or higher, please enable " + "-gpgpu_ptx_use_cuobjdump or downgrade to CUDA 3.1\n"); + exit(1); + } + while (info->ptx[num_ptx_versions].gpuProfileName != NULL) { + unsigned capability = 0; + sscanf(info->ptx[num_ptx_versions].gpuProfileName, "compute_%u", + &capability); + printf( + "GPGPU-Sim PTX: __cudaRegisterFatBinary found PTX versions for " + "'%s', ", + info->ident); + printf("capability = %s\n", info->ptx[num_ptx_versions].gpuProfileName); + if (forced_max_capability) { + if (capability > max_capability && + capability <= forced_max_capability) { + found = true; + max_capability = capability; + selected_capability = num_ptx_versions; } + } else { + if (capability > max_capability) { + found = true; + max_capability = capability; + selected_capability = num_ptx_versions; + } + } + num_ptx_versions++; + } + if (found) { + printf("GPGPU-Sim PTX: Loading PTX for %s, capability = %s\n", + info->ident, info->ptx[selected_capability].gpuProfileName); + symbol_table *symtab; + const char *ptx = info->ptx[selected_capability].ptx; + if (context->get_device() + ->get_gpgpu() + ->get_config() + .convert_to_ptxplus()) { + printf( + "GPGPU-Sim PTX: ERROR ** PTXPlus is only supported through " + "cuobjdump\n" + "\tEither enable cuobjdump or disable PTXPlus in your " + "configuration file\n"); + exit(1); + } else { + symtab = ctx->gpgpu_ptx_sim_load_ptx_from_string(ptx, source_num); + context->add_binary(symtab, fat_cubin_handle); + ctx->gpgpu_ptxinfo_load_from_string(ptx, source_num, max_capability, + context->no_of_ptx); + } + source_num++; + ctx->api->load_static_globals(symtab, STATIC_ALLOC_LIMIT, 0xFFFFFFFF, + context->get_device()->get_gpgpu()); + ctx->api->load_constants(symtab, STATIC_ALLOC_LIMIT, + context->get_device()->get_gpgpu()); + } else { + printf( + "GPGPU-Sim PTX: warning -- did not find an appropriate PTX in " + "cubin\n"); + } + return (void **)fat_cubin_handle; + } +#else + else { + printf("ERROR ** __cudaRegisterFatBinary() needs to be updated\n"); + abort(); + } #endif } -void cudaRegisterFunctionInternal( - void **fatCubinHandle, - const char *hostFun, - char *deviceFun, - const char *deviceName, - int thread_limit, - uint3 *tid, - uint3 *bid, - dim3 *bDim, - dim3 *gDim, - gpgpu_context *gpgpu_ctx = NULL -) -{ - gpgpu_context *ctx; - if (gpgpu_ctx){ - ctx = gpgpu_ctx; - } else { - ctx = GPGPU_Context(); - } - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - CUctx_st *context = GPGPUSim_Context(ctx); - unsigned fat_cubin_handle = (unsigned)(unsigned long long)fatCubinHandle; - printf("GPGPU-Sim PTX: __cudaRegisterFunction %s : hostFun 0x%p, fat_cubin_handle = %u\n", - deviceFun, hostFun, fat_cubin_handle); - if(context->get_device()->get_gpgpu()->get_config().use_cuobjdump()) - ctx->cuobjdumpParseBinary(fat_cubin_handle); - context->register_function( fat_cubin_handle, hostFun, deviceFun ); +void cudaRegisterFunctionInternal(void **fatCubinHandle, const char *hostFun, + char *deviceFun, const char *deviceName, + int thread_limit, uint3 *tid, uint3 *bid, + dim3 *bDim, dim3 *gDim, + gpgpu_context *gpgpu_ctx = NULL) { + gpgpu_context *ctx; + if (gpgpu_ctx) { + ctx = gpgpu_ctx; + } else { + ctx = GPGPU_Context(); + } + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + CUctx_st *context = GPGPUSim_Context(ctx); + unsigned fat_cubin_handle = (unsigned)(unsigned long long)fatCubinHandle; + printf( + "GPGPU-Sim PTX: __cudaRegisterFunction %s : hostFun 0x%p, " + "fat_cubin_handle = %u\n", + deviceFun, hostFun, fat_cubin_handle); + if (context->get_device()->get_gpgpu()->get_config().use_cuobjdump()) + ctx->cuobjdumpParseBinary(fat_cubin_handle); + context->register_function(fat_cubin_handle, hostFun, deviceFun); } void cudaRegisterVarInternal( - void **fatCubinHandle, - char *hostVar, //pointer to...something - char *deviceAddress, //name of variable - const char *deviceName, //name of variable (same as above) - int ext, - int size, - int constant, - int global, - gpgpu_context *gpgpu_ctx = NULL) -{ - gpgpu_context *ctx; - if (gpgpu_ctx){ - ctx = gpgpu_ctx; - } else { - ctx = GPGPU_Context(); - } - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("GPGPU-Sim PTX: __cudaRegisterVar: hostVar = %p; deviceAddress = %s; deviceName = %s\n", hostVar, deviceAddress, deviceName); - printf("GPGPU-Sim PTX: __cudaRegisterVar: Registering const memory space of %d bytes\n", size); - if(GPGPUSim_Context(ctx)->get_device()->get_gpgpu()->get_config().use_cuobjdump()) - ctx->cuobjdumpParseBinary((unsigned)(unsigned long long)fatCubinHandle); - fflush(stdout); - if ( constant && !global && !ext ) { - ctx->func_sim->gpgpu_ptx_sim_register_const_variable(hostVar,deviceName,size); - } else if ( !constant && !global && !ext ) { - ctx->func_sim->gpgpu_ptx_sim_register_global_variable(hostVar,deviceName,size); - } else cuda_not_implemented(__my_func__,__LINE__); -} - -cudaError_t cudaConfigureCallInternal(dim3 gridDim, dim3 blockDim, size_t sharedMem, cudaStream_t stream, gpgpu_context* gpgpu_ctx = NULL) -{ - gpgpu_context *ctx; - if (gpgpu_ctx){ - ctx = gpgpu_ctx; + void **fatCubinHandle, + char *hostVar, // pointer to...something + char *deviceAddress, // name of variable + const char *deviceName, // name of variable (same as above) + int ext, int size, int constant, int global, + gpgpu_context *gpgpu_ctx = NULL) { + gpgpu_context *ctx; + if (gpgpu_ctx) { + ctx = gpgpu_ctx; + } else { + ctx = GPGPU_Context(); + } + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf( + "GPGPU-Sim PTX: __cudaRegisterVar: hostVar = %p; deviceAddress = %s; " + "deviceName = %s\n", + hostVar, deviceAddress, deviceName); + printf( + "GPGPU-Sim PTX: __cudaRegisterVar: Registering const memory space of %d " + "bytes\n", + size); + if (GPGPUSim_Context(ctx) + ->get_device() + ->get_gpgpu() + ->get_config() + .use_cuobjdump()) + ctx->cuobjdumpParseBinary((unsigned)(unsigned long long)fatCubinHandle); + fflush(stdout); + if (constant && !global && !ext) { + ctx->func_sim->gpgpu_ptx_sim_register_const_variable(hostVar, deviceName, + size); + } else if (!constant && !global && !ext) { + ctx->func_sim->gpgpu_ptx_sim_register_global_variable(hostVar, deviceName, + size); + } else + cuda_not_implemented(__my_func__, __LINE__); +} + +cudaError_t cudaConfigureCallInternal(dim3 gridDim, dim3 blockDim, + size_t sharedMem, cudaStream_t stream, + gpgpu_context *gpgpu_ctx = NULL) { + gpgpu_context *ctx; + if (gpgpu_ctx) { + ctx = gpgpu_ctx; + } else { + ctx = GPGPU_Context(); + } + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + struct CUstream_st *s = (struct CUstream_st *)stream; + ctx->api->g_cuda_launch_stack.push_back( + kernel_config(gridDim, blockDim, sharedMem, s)); + return g_last_cudaError = cudaSuccess; +} + +__host__ cudaError_t CUDARTAPI +cudaGetDeviceCountInternal(int *count, gpgpu_context *gpgpu_ctx = NULL) { + gpgpu_context *ctx; + if (gpgpu_ctx) { + ctx = gpgpu_ctx; + } else { + ctx = GPGPU_Context(); + } + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + _cuda_device_id *dev = ctx->GPGPUSim_Init(); + *count = dev->num_devices(); + return g_last_cudaError = cudaSuccess; +} + +__host__ cudaError_t CUDARTAPI cudaGetDevicePropertiesInternal( + struct cudaDeviceProp *prop, int device, gpgpu_context *gpgpu_ctx = NULL) { + gpgpu_context *ctx; + if (gpgpu_ctx) { + ctx = gpgpu_ctx; + } else { + ctx = GPGPU_Context(); + } + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + _cuda_device_id *dev = ctx->GPGPUSim_Init(); + if (device <= dev->num_devices()) { + *prop = *dev->get_prop(); + return g_last_cudaError = cudaSuccess; + } else { + return g_last_cudaError = cudaErrorInvalidDevice; + } +} + +__host__ cudaError_t CUDARTAPI +cudaChooseDeviceInternal(int *device, const struct cudaDeviceProp *prop, + gpgpu_context *gpgpu_ctx = NULL) { + gpgpu_context *ctx; + if (gpgpu_ctx) { + ctx = gpgpu_ctx; + } else { + ctx = GPGPU_Context(); + } + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + _cuda_device_id *dev = ctx->GPGPUSim_Init(); + *device = dev->get_id(); + return g_last_cudaError = cudaSuccess; +} + +cudaError_t cudaSetupArgumentInternal(const void *arg, size_t size, + size_t offset, + gpgpu_context *gpgpu_ctx = NULL) { + gpgpu_context *ctx; + if (gpgpu_ctx) { + ctx = gpgpu_ctx; + } else { + ctx = GPGPU_Context(); + } + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + gpgpusim_ptx_assert(!ctx->api->g_cuda_launch_stack.empty(), + "empty launch stack"); + kernel_config &config = ctx->api->g_cuda_launch_stack.back(); + config.set_arg(arg, size, offset); + printf( + "GPGPU-Sim PTX: Setting up arguments for %zu bytes starting at " + "0x%llx..\n", + size, (unsigned long long)arg); + + return g_last_cudaError = cudaSuccess; +} + +cudaError_t cudaLaunchInternal(const char *hostFun, + gpgpu_context *gpgpu_ctx = NULL) { + gpgpu_context *ctx; + if (gpgpu_ctx) { + ctx = gpgpu_ctx; + } else { + ctx = GPGPU_Context(); + } + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + CUctx_st *context = GPGPUSim_Context(ctx); + char *mode = getenv("PTX_SIM_MODE_FUNC"); + if (mode) sscanf(mode, "%u", &(ctx->func_sim->g_ptx_sim_mode)); + gpgpusim_ptx_assert(!ctx->api->g_cuda_launch_stack.empty(), + "empty launch stack"); + kernel_config config = ctx->api->g_cuda_launch_stack.back(); + { + dim3 gridDim = config.grid_dim(); + dim3 blockDim = config.block_dim(); + if (gridDim.x * gridDim.y * gridDim.z == 0 || + blockDim.x * blockDim.y * blockDim.z == 0) { + // can't launch + printf("can't launch a empty kernel\n"); + ctx->api->g_cuda_launch_stack.pop_back(); + return g_last_cudaError = cudaErrorInvalidConfiguration; + } + } + struct CUstream_st *stream = config.get_stream(); + + printf("\nGPGPU-Sim PTX: cudaLaunch for 0x%p (mode=%s) on stream %u\n", + hostFun, + (ctx->func_sim->g_ptx_sim_mode) ? "functional simulation" + : "performance simulation", + stream ? stream->get_uid() : 0); + kernel_info_t *grid = ctx->api->gpgpu_cuda_ptx_sim_init_grid( + hostFun, config.get_args(), config.grid_dim(), config.block_dim(), + context); + // do dynamic PDOM analysis for performance simulation scenario + std::string kname = grid->name(); + function_info *kernel_func_info = grid->entry(); + if (kernel_func_info->is_pdom_set()) { + printf("GPGPU-Sim PTX: PDOM analysis already done for %s \n", + kname.c_str()); + } else { + printf("GPGPU-Sim PTX: finding reconvergence points for \'%s\'...\n", + kname.c_str()); + kernel_func_info->do_pdom(); + kernel_func_info->set_pdom(); + } + dim3 gridDim = config.grid_dim(); + dim3 blockDim = config.block_dim(); + + gpgpu_t *gpu = context->get_device()->get_gpgpu(); + checkpoint *g_checkpoint; + g_checkpoint = new checkpoint(); + class memory_space *global_mem; + global_mem = gpu->get_global_memory(); + + if (gpu->resume_option == 1 && (grid->get_uid() == gpu->resume_kernel)) { + char f1name[2048]; + snprintf(f1name, 2048, "checkpoint_files/global_mem_%d.txt", + grid->get_uid()); + + g_checkpoint->load_global_mem(global_mem, f1name); + for (int i = 0; i < gpu->resume_CTA; i++) grid->increment_cta_id(); + } + if (gpu->resume_option == 1 && (grid->get_uid() < gpu->resume_kernel)) { + char f1name[2048]; + snprintf(f1name, 2048, "checkpoint_files/global_mem_%d.txt", + grid->get_uid()); + + g_checkpoint->load_global_mem(global_mem, f1name); + printf("Skipping kernel %d as resuming from kernel %d\n", grid->get_uid(), + gpu->resume_kernel); + ctx->api->g_cuda_launch_stack.pop_back(); + return g_last_cudaError = cudaSuccess; + } + if (gpu->checkpoint_option == 1 && + (grid->get_uid() > gpu->checkpoint_kernel)) { + printf("Skipping kernel %d as checkpoint from kernel %d\n", grid->get_uid(), + gpu->checkpoint_kernel); + ctx->api->g_cuda_launch_stack.pop_back(); + return g_last_cudaError = cudaSuccess; + } + printf( + "GPGPU-Sim PTX: pushing kernel \'%s\' to stream %u, gridDim= (%u,%u,%u) " + "blockDim = (%u,%u,%u) \n", + kname.c_str(), stream ? stream->get_uid() : 0, gridDim.x, gridDim.y, + gridDim.z, blockDim.x, blockDim.y, blockDim.z); + stream_operation op(grid, ctx->func_sim->g_ptx_sim_mode, stream); + ctx->the_gpgpusim->g_stream_manager->push(op); + ctx->api->g_cuda_launch_stack.pop_back(); + return g_last_cudaError = cudaSuccess; +} + +cudaError_t cudaMallocInternal(void **devPtr, size_t size, + gpgpu_context *gpgpu_ctx = NULL) { + gpgpu_context *ctx; + if (gpgpu_ctx) { + ctx = gpgpu_ctx; + } else { + ctx = GPGPU_Context(); + } + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + CUctx_st *context = GPGPUSim_Context(ctx); + *devPtr = context->get_device()->get_gpgpu()->gpu_malloc(size); + if (g_debug_execution >= 3) { + printf("GPGPU-Sim PTX: cudaMallocing %zu bytes starting at 0x%llx..\n", + size, (unsigned long long)*devPtr); + ctx->api->g_mallocPtr_Size[(unsigned long long)*devPtr] = size; + } + if (*devPtr) { + return g_last_cudaError = cudaSuccess; + } else { + return g_last_cudaError = cudaErrorMemoryAllocation; + } +} + +cudaError_t cudaMallocHostInternal(void **ptr, size_t size, + gpgpu_context *gpgpu_ctx = NULL) { + gpgpu_context *ctx; + if (gpgpu_ctx) { + ctx = gpgpu_ctx; + } else { + ctx = GPGPU_Context(); + } + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + *ptr = malloc(size); + if (*ptr) { + // track pinned memory size allocated in the host so that same amount of + // memory is also allocated in GPU. + ctx->api->pinned_memory_size[*ptr] = size; + return g_last_cudaError = cudaSuccess; + } else { + return g_last_cudaError = cudaErrorMemoryAllocation; + } +} + +__host__ cudaError_t CUDARTAPI +cudaMallocPitchInternal(void **devPtr, size_t *pitch, size_t width, + size_t height, gpgpu_context *gpgpu_ctx = NULL) { + gpgpu_context *ctx; + if (gpgpu_ctx) { + ctx = gpgpu_ctx; + } else { + ctx = GPGPU_Context(); + } + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + unsigned malloc_width_inbytes = width; + printf("GPGPU-Sim PTX: cudaMallocPitch (width = %d)\n", malloc_width_inbytes); + CUctx_st *context = GPGPUSim_Context(ctx); + *devPtr = context->get_device()->get_gpgpu()->gpu_malloc( + malloc_width_inbytes * height); + pitch[0] = malloc_width_inbytes; + if (*devPtr) { + return g_last_cudaError = cudaSuccess; + } else { + return g_last_cudaError = cudaErrorMemoryAllocation; + } +} + +cudaError_t cudaHostGetDevicePointerInternal(void **pDevice, void *pHost, + unsigned int flags, + gpgpu_context *gpgpu_ctx = NULL) { + gpgpu_context *ctx; + if (gpgpu_ctx) { + ctx = gpgpu_ctx; + } else { + ctx = GPGPU_Context(); + } + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + // only cpu memory allocation happens in cudaHostAlloc. Linking with device + // pointer to pinned memory happens here. + // TODO: once kernel is executed, the contents in global pointer of GPU must + // be copied back to CPU host pointer! + flags = 0; + CUctx_st *context = GPGPUSim_Context(ctx); + gpgpu_t *gpu = context->get_device()->get_gpgpu(); + std::map::const_iterator i = + ctx->api->pinned_memory_size.find(pHost); + assert(i != ctx->api->pinned_memory_size.end()); + size_t size = i->second; + *pDevice = gpu->gpu_malloc(size); + if (g_debug_execution >= 3) { + printf("GPGPU-Sim PTX: cudaMallocing %zu bytes starting at 0x%llx..\n", + size, (unsigned long long)*pDevice); + ctx->api->g_mallocPtr_Size[(unsigned long long)*pDevice] = size; + } + if (*pDevice) { + ctx->api->pinned_memory[pHost] = pDevice; + // Copy contents in cpu to gpu + gpu->memcpy_to_gpu((size_t)*pDevice, pHost, size); + return g_last_cudaError = cudaSuccess; + } else { + return g_last_cudaError = cudaErrorMemoryAllocation; + } +} + +__host__ cudaError_t CUDARTAPI cudaMallocArrayInternal( + struct cudaArray **array, const struct cudaChannelFormatDesc *desc, + size_t width, size_t height __dv(1), gpgpu_context *gpgpu_ctx = NULL) { + gpgpu_context *ctx; + if (gpgpu_ctx) { + ctx = gpgpu_ctx; + } else { + ctx = GPGPU_Context(); + } + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + unsigned size = + width * height * ((desc->x + desc->y + desc->z + desc->w) / 8); + CUctx_st *context = GPGPUSim_Context(ctx); + (*array) = (struct cudaArray *)malloc(sizeof(struct cudaArray)); + (*array)->desc = *desc; + (*array)->width = width; + (*array)->height = height; + (*array)->size = size; + (*array)->dimensions = 2; + ((*array)->devPtr32) = + (int)(long long)context->get_device()->get_gpgpu()->gpu_mallocarray(size); + printf("GPGPU-Sim PTX: cudaMallocArray: devPtr32 = %d\n", + ((*array)->devPtr32)); + ((*array)->devPtr) = (void *)(long long)((*array)->devPtr32); + if (((*array)->devPtr)) { + return g_last_cudaError = cudaSuccess; + } else { + return g_last_cudaError = cudaErrorMemoryAllocation; + } +} + +__host__ cudaError_t CUDARTAPI +cudaMemcpyInternal(void *dst, const void *src, size_t count, + enum cudaMemcpyKind kind, gpgpu_context *gpgpu_ctx = NULL) { + gpgpu_context *ctx; + if (gpgpu_ctx) { + ctx = gpgpu_ctx; + } else { + ctx = GPGPU_Context(); + } + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + // CUctx_st *context = GPGPUSim_Context(); + // gpgpu_t *gpu = context->get_device()->get_gpgpu(); + if (g_debug_execution >= 3) + printf("GPGPU-Sim PTX: cudaMemcpy(): devPtr = %p\n", dst); + if (kind == cudaMemcpyHostToDevice) + ctx->the_gpgpusim->g_stream_manager->push( + stream_operation(src, (size_t)dst, count, 0)); + else if (kind == cudaMemcpyDeviceToHost) + ctx->the_gpgpusim->g_stream_manager->push( + stream_operation((size_t)src, dst, count, 0)); + else if (kind == cudaMemcpyDeviceToDevice) + ctx->the_gpgpusim->g_stream_manager->push( + stream_operation((size_t)src, (size_t)dst, count, 0)); + else if (kind == cudaMemcpyDefault) { + if ((size_t)src >= GLOBAL_HEAP_START) { + if ((size_t)dst >= GLOBAL_HEAP_START) + ctx->the_gpgpusim->g_stream_manager->push(stream_operation( + (size_t)src, (size_t)dst, count, 0)); // device to device + else + ctx->the_gpgpusim->g_stream_manager->push( + stream_operation((size_t)src, dst, count, 0)); // device to host } else { - ctx = GPGPU_Context(); - } - if(g_debug_execution >= 3){ - announce_call(__my_func__); + if ((size_t)dst >= GLOBAL_HEAP_START) + ctx->the_gpgpusim->g_stream_manager->push( + stream_operation(src, (size_t)dst, count, 0)); + else { + printf( + "GPGPU-Sim PTX: cudaMemcpy - ERROR : unsupported transfer: host to " + "host\n"); + abort(); + } } - struct CUstream_st *s = (struct CUstream_st *)stream; - ctx->api->g_cuda_launch_stack.push_back( kernel_config(gridDim,blockDim,sharedMem,s) ); - return g_last_cudaError = cudaSuccess; + } else { + printf("GPGPU-Sim PTX: cudaMemcpy - ERROR : unsupported cudaMemcpyKind\n"); + abort(); + } + return g_last_cudaError = cudaSuccess; +} + +__host__ cudaError_t CUDARTAPI cudaMemcpyToArrayInternal( + struct cudaArray *dst, size_t wOffset, size_t hOffset, const void *src, + size_t count, enum cudaMemcpyKind kind, gpgpu_context *gpgpu_ctx = NULL) { + gpgpu_context *ctx; + if (gpgpu_ctx) { + ctx = gpgpu_ctx; + } else { + ctx = GPGPU_Context(); + } + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + CUctx_st *context = GPGPUSim_Context(ctx); + gpgpu_t *gpu = context->get_device()->get_gpgpu(); + size_t size = count; + printf("GPGPU-Sim PTX: cudaMemcpyToArray\n"); + if (kind == cudaMemcpyHostToDevice) + gpu->memcpy_to_gpu((size_t)(dst->devPtr), src, size); + else if (kind == cudaMemcpyDeviceToHost) + gpu->memcpy_from_gpu(dst->devPtr, (size_t)src, size); + else if (kind == cudaMemcpyDeviceToDevice) + gpu->memcpy_gpu_to_gpu((size_t)(dst->devPtr), (size_t)src, size); + else { + printf( + "GPGPU-Sim PTX: cudaMemcpyToArray - ERROR : unsupported " + "cudaMemcpyKind\n"); + abort(); + } + dst->devPtr32 = (unsigned)(size_t)(dst->devPtr); + return g_last_cudaError = cudaSuccess; +} + +__host__ cudaError_t CUDARTAPI cudaMemcpy2DInternal( + void *dst, size_t dpitch, const void *src, size_t spitch, size_t width, + size_t height, enum cudaMemcpyKind kind, gpgpu_context *gpgpu_ctx = NULL) { + gpgpu_context *ctx; + if (gpgpu_ctx) { + ctx = gpgpu_ctx; + } else { + ctx = GPGPU_Context(); + } + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + CUctx_st *context = GPGPUSim_Context(ctx); + gpgpu_t *gpu = context->get_device()->get_gpgpu(); + size_t size = spitch * height; + gpgpusim_ptx_assert((dpitch == spitch), + "different src and dst pitch not supported yet"); + if (kind == cudaMemcpyHostToDevice) + gpu->memcpy_to_gpu((size_t)dst, src, size); + else if (kind == cudaMemcpyDeviceToHost) + gpu->memcpy_from_gpu(dst, (size_t)src, size); + else if (kind == cudaMemcpyDeviceToDevice) + gpu->memcpy_gpu_to_gpu((size_t)dst, (size_t)src, size); + else { + printf( + "GPGPU-Sim PTX: cudaMemcpy2D - ERROR : unsupported cudaMemcpyKind\n"); + abort(); + } + return g_last_cudaError = cudaSuccess; +} + +__host__ cudaError_t CUDARTAPI cudaMemcpy2DToArrayInternal( + struct cudaArray *dst, size_t wOffset, size_t hOffset, const void *src, + size_t spitch, size_t width, size_t height, enum cudaMemcpyKind kind, + gpgpu_context *gpgpu_ctx = NULL) { + gpgpu_context *ctx; + if (gpgpu_ctx) { + ctx = gpgpu_ctx; + } else { + ctx = GPGPU_Context(); + } + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + CUctx_st *context = GPGPUSim_Context(ctx); + gpgpu_t *gpu = context->get_device()->get_gpgpu(); + size_t size = spitch * height; + size_t channel_size = dst->desc.w + dst->desc.x + dst->desc.y + dst->desc.z; + gpgpusim_ptx_assert( + ((channel_size % 8) == 0), + "none byte multiple destination channel size not supported (sz=%u)", + channel_size); + unsigned elem_size = channel_size / 8; + gpgpusim_ptx_assert((dst->dimensions == 2), + "copy to none 2D array not supported"); + gpgpusim_ptx_assert((wOffset == 0), "non-zero wOffset not yet supported"); + gpgpusim_ptx_assert((hOffset == 0), "non-zero hOffset not yet supported"); + gpgpusim_ptx_assert((dst->height == (int)height), + "partial copy not supported"); + gpgpusim_ptx_assert((elem_size * dst->width == width), + "partial copy not supported"); + gpgpusim_ptx_assert((spitch == width), "spitch != width not supported"); + if (kind == cudaMemcpyHostToDevice) + gpu->memcpy_to_gpu((size_t)(dst->devPtr), src, size); + else if (kind == cudaMemcpyDeviceToHost) + gpu->memcpy_from_gpu(dst->devPtr, (size_t)src, size); + else if (kind == cudaMemcpyDeviceToDevice) + gpu->memcpy_gpu_to_gpu((size_t)dst->devPtr, (size_t)src, size); + else { + printf( + "GPGPU-Sim PTX: cudaMemcpy2D - ERROR : unsupported cudaMemcpyKind\n"); + abort(); + } + dst->devPtr32 = (unsigned)(size_t)(dst->devPtr); + return g_last_cudaError = cudaSuccess; +} + +__host__ cudaError_t CUDARTAPI cudaMemcpyToSymbolInternal( + const char *symbol, const void *src, size_t count, size_t offset __dv(0), + enum cudaMemcpyKind kind __dv(cudaMemcpyHostToDevice), + gpgpu_context *gpgpu_ctx = NULL) { + gpgpu_context *ctx; + if (gpgpu_ctx) { + ctx = gpgpu_ctx; + } else { + ctx = GPGPU_Context(); + } + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + // CUctx_st *context = GPGPUSim_Context(); + assert(kind == cudaMemcpyHostToDevice); + printf("GPGPU-Sim PTX: cudaMemcpyToSymbol: symbol = %p\n", symbol); + // stream_operation( const char *symbol, const void *src, size_t count, size_t + // offset ) + ctx->the_gpgpusim->g_stream_manager->push( + stream_operation(src, symbol, count, offset, 0)); + // gpgpu_ptx_sim_memcpy_symbol(symbol,src,count,offset,1,context->get_device()->get_gpgpu()); + return g_last_cudaError = cudaSuccess; +} + +__host__ cudaError_t CUDARTAPI cudaMemcpyFromSymbolInternal( + void *dst, const char *symbol, size_t count, size_t offset __dv(0), + enum cudaMemcpyKind kind __dv(cudaMemcpyDeviceToHost), + gpgpu_context *gpgpu_ctx = NULL) { + gpgpu_context *ctx; + if (gpgpu_ctx) { + ctx = gpgpu_ctx; + } else { + ctx = GPGPU_Context(); + } + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + // CUctx_st *context = GPGPUSim_Context(); + assert(kind == cudaMemcpyDeviceToHost); + printf("GPGPU-Sim PTX: cudaMemcpyFromSymbol: symbol = %p\n", symbol); + ctx->the_gpgpusim->g_stream_manager->push( + stream_operation(symbol, dst, count, offset, 0)); + // gpgpu_ptx_sim_memcpy_symbol(symbol,dst,count,offset,0,context->get_device()->get_gpgpu()); + return g_last_cudaError = cudaSuccess; +} + +__host__ cudaError_t CUDARTAPI cudaMemcpyAsyncInternal( + void *dst, const void *src, size_t count, enum cudaMemcpyKind kind, + cudaStream_t stream, gpgpu_context *gpgpu_ctx = NULL) { + gpgpu_context *ctx; + if (gpgpu_ctx) { + ctx = gpgpu_ctx; + } else { + ctx = GPGPU_Context(); + } + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + struct CUstream_st *s = (struct CUstream_st *)stream; + switch (kind) { + case cudaMemcpyHostToDevice: + ctx->the_gpgpusim->g_stream_manager->push( + stream_operation(src, (size_t)dst, count, s)); + break; + case cudaMemcpyDeviceToHost: + ctx->the_gpgpusim->g_stream_manager->push( + stream_operation((size_t)src, dst, count, s)); + break; + case cudaMemcpyDeviceToDevice: + ctx->the_gpgpusim->g_stream_manager->push( + stream_operation((size_t)src, (size_t)dst, count, s)); + break; + default: + abort(); + } + return g_last_cudaError = cudaSuccess; } -__host__ cudaError_t CUDARTAPI cudaGetDeviceCountInternal(int *count, gpgpu_context* gpgpu_ctx = NULL) -{ - gpgpu_context *ctx; - if (gpgpu_ctx){ - ctx = gpgpu_ctx; - } else { - ctx = GPGPU_Context(); - } - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - _cuda_device_id *dev = ctx->GPGPUSim_Init(); - *count = dev->num_devices(); +#if (CUDART_VERSION >= 8000) +cudaError_t CUDARTAPI +cudaOccupancyMaxActiveBlocksPerMultiprocessorWithFlagsInternal( + int *numBlocks, const char *hostFunc, int blockSize, size_t dynamicSMemSize, + unsigned int flags, gpgpu_context *gpgpu_ctx = NULL) { + gpgpu_context *ctx; + if (gpgpu_ctx) { + ctx = gpgpu_ctx; + } else { + ctx = GPGPU_Context(); + } + printf( + "GPGPU-Sim PTX: cudaOccupancyMaxActiveBlocksPerMultiprocessorWithFlags " + "%p\n", + hostFunc); + CUctx_st *context = GPGPUSim_Context(ctx); + function_info *entry = context->get_kernel(hostFunc); + printf( + "Calculate Maxium Active Block with function ptr=%p, blockSize=%d, " + "SMemSize=%d\n", + hostFunc, blockSize, dynamicSMemSize); + if (flags == cudaOccupancyDefault) { + // create kernel_info based on entry + dim3 gridDim(context->get_device()->get_gpgpu()->max_cta_per_core() * + context->get_device()->get_gpgpu()->get_config().num_shader()); + dim3 blockDim(blockSize); + kernel_info_t result(gridDim, blockDim, entry); + // if(entry == NULL){ + // *numBlocks = 1; + // return g_last_cudaError = cudaErrorUnknown; + //} + *numBlocks = context->get_device()->get_gpgpu()->get_max_cta(result); + printf("Maximum size is %d with gridDim %d and blockDim %d\n", *numBlocks, + gridDim.x, blockDim.x); return g_last_cudaError = cudaSuccess; + } else { + cuda_not_implemented(__my_func__, __LINE__); + return g_last_cudaError = cudaErrorUnknown; + } } -__host__ cudaError_t CUDARTAPI cudaGetDevicePropertiesInternal(struct cudaDeviceProp *prop, int device, gpgpu_context* gpgpu_ctx = NULL) -{ - gpgpu_context *ctx; - if (gpgpu_ctx){ - ctx = gpgpu_ctx; - } else { - ctx = GPGPU_Context(); - } - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - _cuda_device_id *dev = ctx->GPGPUSim_Init(); - if (device <= dev->num_devices() ) { - *prop= *dev->get_prop(); - return g_last_cudaError = cudaSuccess; - } else { - return g_last_cudaError = cudaErrorInvalidDevice; - } -} +#endif +__host__ cudaError_t CUDARTAPI cudaMemsetInternal( + void *mem, int c, size_t count, gpgpu_context *gpgpu_ctx = NULL) { + gpgpu_context *ctx; + if (gpgpu_ctx) { + ctx = gpgpu_ctx; + } else { + ctx = GPGPU_Context(); + } + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + CUctx_st *context = GPGPUSim_Context(ctx); + gpgpu_t *gpu = context->get_device()->get_gpgpu(); + gpu->gpu_memset((size_t)mem, c, count); + return g_last_cudaError = cudaSuccess; +} + +// memset operation is done but i think its not async? +__host__ cudaError_t CUDARTAPI +cudaMemsetAsyncInternal(void *mem, int c, size_t count, cudaStream_t stream = 0, + gpgpu_context *gpgpu_ctx = NULL) { + gpgpu_context *ctx; + if (gpgpu_ctx) { + ctx = gpgpu_ctx; + } else { + ctx = GPGPU_Context(); + } + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("GPGPU-Sim PTX: WARNING: Asynchronous memset not supported (%s)\n", + __my_func__); + CUctx_st *context = GPGPUSim_Context(ctx); + gpgpu_t *gpu = context->get_device()->get_gpgpu(); + gpu->gpu_memset((size_t)mem, c, count); + return g_last_cudaError = cudaSuccess; +} + +cudaError_t cudaGLMapBufferObjectInternal(void **devPtr, GLuint bufferObj, + gpgpu_context *gpgpu_ctx = NULL) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } +#ifdef OPENGL_SUPPORT + gpgpu_context *ctx; + if (gpgpu_ctx) { + ctx = gpgpu_ctx; + } else { + ctx = GPGPU_Context(); + } + GLint buffer_size = 0; + CUctx_st *context = GPGPUSim_Context(ctx); + + glbmap_entry_t *p = ctx->api->g_glbmap; + while (p && p->m_bufferObj != bufferObj) p = p->m_next; + if (p == NULL) { + glBindBuffer(GL_ARRAY_BUFFER, bufferObj); + glGetBufferParameteriv(GL_ARRAY_BUFFER, GL_BUFFER_SIZE, &buffer_size); + assert(buffer_size != 0); + *devPtr = context->get_device()->get_gpgpu()->gpu_malloc(buffer_size); + + // create entry and insert to front of list + glbmap_entry_t *n = (glbmap_entry_t *)calloc(1, sizeof(glbmap_entry_t)); + n->m_next = ctx->api->g_glbmap; + ctx->api->g_glbmap = n; + + // initialize entry + n->m_bufferObj = bufferObj; + n->m_devPtr = *devPtr; + n->m_size = buffer_size; + + p = n; + } else { + buffer_size = p->m_size; + *devPtr = p->m_devPtr; + } + + if (*devPtr) { + char *data = (char *)calloc(p->m_size, 1); + glGetBufferSubData(GL_ARRAY_BUFFER, 0, buffer_size, data); + memcpy_to_gpu((size_t)*devPtr, data, buffer_size); + free(data); + printf( + "GPGPU-Sim PTX: cudaGLMapBufferObject %zu bytes starting at 0x%llx..\n", + (size_t)buffer_size, (unsigned long long)*devPtr); + return g_last_cudaError = cudaSuccess; + } else { + return g_last_cudaError = cudaErrorMemoryAllocation; + } -__host__ cudaError_t CUDARTAPI cudaChooseDeviceInternal(int *device, const struct cudaDeviceProp *prop, gpgpu_context* gpgpu_ctx = NULL) -{ - gpgpu_context *ctx; - if (gpgpu_ctx){ - ctx = gpgpu_ctx; - } else { - ctx = GPGPU_Context(); - } - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - _cuda_device_id *dev = ctx->GPGPUSim_Init(); - *device = dev->get_id(); - return g_last_cudaError = cudaSuccess; + return g_last_cudaError = cudaSuccess; +#else + fflush(stdout); + fflush(stderr); + printf( + "GPGPU-Sim PTX: GPGPU-Sim support for OpenGL integration disabled -- " + "exiting\n"); + fflush(stdout); + exit(50); +#endif } -cudaError_t cudaSetupArgumentInternal(const void *arg, size_t size, size_t offset, gpgpu_context* gpgpu_ctx = NULL) -{ - gpgpu_context *ctx; - if (gpgpu_ctx){ - ctx = gpgpu_ctx; - } else { - ctx = GPGPU_Context(); - } - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - gpgpusim_ptx_assert( !ctx->api->g_cuda_launch_stack.empty(), "empty launch stack" ); - kernel_config &config = ctx->api->g_cuda_launch_stack.back(); - config.set_arg(arg,size,offset); - printf("GPGPU-Sim PTX: Setting up arguments for %zu bytes starting at 0x%llx..\n",size, (unsigned long long) arg); - - return g_last_cudaError = cudaSuccess; +#if CUDART_VERSION >= 6050 +CUresult cuLinkAddFileInternal(CUlinkState state, CUjitInputType type, + const char *path, unsigned int numOptions, + CUjit_option *options, void **optionValues, + gpgpu_context *gpgpu_ctx = NULL) { + gpgpu_context *ctx; + if (gpgpu_ctx) { + ctx = gpgpu_ctx; + } else { + ctx = GPGPU_Context(); + } + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + static bool addedFile = false; + if (addedFile) { + printf( + "GPGPU-Sim PTX: ERROR: cuLinkAddFile does not support multiple " + "files\n"); + abort(); + } + + // blocking + assert(type == CU_JIT_INPUT_PTX); + CUctx_st *context = GPGPUSim_Context(ctx); + char *file = getenv("PTX_JIT_PATH"); + if (file == NULL) { + printf("GPGPU-Sim PTX: ERROR: PTX_JIT_PATH has not been set\n"); + abort(); + } + strcat(file, "/"); + strcat(file, path); + symbol_table *symtab = ctx->gpgpu_ptx_sim_load_ptx_from_filename(file); + std::string fname(path); + ctx->api->name_symtab[fname] = symtab; + context->add_binary(symtab, 1); + ctx->api->load_static_globals(symtab, STATIC_ALLOC_LIMIT, 0xFFFFFFFF, + context->get_device()->get_gpgpu()); + ctx->api->load_constants(symtab, STATIC_ALLOC_LIMIT, + context->get_device()->get_gpgpu()); + addedFile = true; + return CUDA_SUCCESS; } +#endif -cudaError_t cudaLaunchInternal( const char *hostFun, gpgpu_context* gpgpu_ctx = NULL ) -{ - gpgpu_context *ctx; - if (gpgpu_ctx){ - ctx = gpgpu_ctx; - } else { - ctx = GPGPU_Context(); - } - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - CUctx_st* context = GPGPUSim_Context(ctx); - char *mode = getenv("PTX_SIM_MODE_FUNC"); - if( mode ) - sscanf(mode,"%u", &(ctx->func_sim->g_ptx_sim_mode)); - gpgpusim_ptx_assert( !ctx->api->g_cuda_launch_stack.empty(), "empty launch stack" ); - kernel_config config = ctx->api->g_cuda_launch_stack.back(); - { - dim3 gridDim = config.grid_dim(); - dim3 blockDim = config.block_dim(); - if (gridDim.x * gridDim.y * gridDim.z == 0 || blockDim.x * blockDim.y * blockDim.z == 0) - { - //can't launch - printf("can't launch a empty kernel\n"); - ctx->api->g_cuda_launch_stack.pop_back(); - return g_last_cudaError = cudaErrorInvalidConfiguration; - } - } - struct CUstream_st *stream = config.get_stream(); - - printf("\nGPGPU-Sim PTX: cudaLaunch for 0x%p (mode=%s) on stream %u\n", hostFun, - (ctx->func_sim->g_ptx_sim_mode)?"functional simulation":"performance simulation", stream?stream->get_uid():0 ); - kernel_info_t *grid = ctx->api->gpgpu_cuda_ptx_sim_init_grid(hostFun,config.get_args(),config.grid_dim(),config.block_dim(),context); - //do dynamic PDOM analysis for performance simulation scenario - std::string kname = grid->name(); - function_info *kernel_func_info = grid->entry(); - if (kernel_func_info->is_pdom_set()) { - printf("GPGPU-Sim PTX: PDOM analysis already done for %s \n", kname.c_str() ); - } else { - printf("GPGPU-Sim PTX: finding reconvergence points for \'%s\'...\n", kname.c_str() ); - kernel_func_info->do_pdom(); - kernel_func_info->set_pdom(); - } - dim3 gridDim = config.grid_dim(); - dim3 blockDim = config.block_dim(); - - gpgpu_t *gpu = context->get_device()->get_gpgpu(); - checkpoint *g_checkpoint; - g_checkpoint = new checkpoint(); - class memory_space *global_mem; - global_mem = gpu->get_global_memory(); - - if(gpu->resume_option ==1 && (grid->get_uid()==gpu->resume_kernel)) - { - - char f1name[2048]; - snprintf(f1name,2048,"checkpoint_files/global_mem_%d.txt", grid->get_uid()); - - g_checkpoint->load_global_mem(global_mem, f1name); - for (int i=0;iresume_CTA;i++) - grid->increment_cta_id(); - } - if(gpu->resume_option==1 && (grid->get_uid()resume_kernel)) - { - char f1name[2048]; - snprintf(f1name,2048,"checkpoint_files/global_mem_%d.txt", grid->get_uid()); - - g_checkpoint->load_global_mem(global_mem, f1name); - printf("Skipping kernel %d as resuming from kernel %d\n",grid->get_uid(),gpu->resume_kernel ); - ctx->api->g_cuda_launch_stack.pop_back(); - return g_last_cudaError = cudaSuccess; - - } - if(gpu->checkpoint_option==1 && (grid->get_uid()>gpu->checkpoint_kernel)) - { - printf("Skipping kernel %d as checkpoint from kernel %d\n",grid->get_uid(),gpu->checkpoint_kernel ); - ctx->api->g_cuda_launch_stack.pop_back(); - return g_last_cudaError = cudaSuccess; - - } - printf("GPGPU-Sim PTX: pushing kernel \'%s\' to stream %u, gridDim= (%u,%u,%u) blockDim = (%u,%u,%u) \n", - kname.c_str(), stream?stream->get_uid():0, gridDim.x,gridDim.y,gridDim.z,blockDim.x,blockDim.y,blockDim.z ); - stream_operation op(grid,ctx->func_sim->g_ptx_sim_mode,stream); - ctx->the_gpgpusim->g_stream_manager->push(op); - ctx->api->g_cuda_launch_stack.pop_back(); - return g_last_cudaError = cudaSuccess; -} - -cudaError_t cudaMallocInternal(void **devPtr, size_t size, gpgpu_context* gpgpu_ctx = NULL) -{ - gpgpu_context *ctx; - if (gpgpu_ctx){ - ctx = gpgpu_ctx; - } else { - ctx = GPGPU_Context(); - } - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - CUctx_st* context = GPGPUSim_Context(ctx); - *devPtr = context->get_device()->get_gpgpu()->gpu_malloc(size); - if(g_debug_execution >= 3){ - printf("GPGPU-Sim PTX: cudaMallocing %zu bytes starting at 0x%llx..\n",size, (unsigned long long) *devPtr); - ctx->api->g_mallocPtr_Size[(unsigned long long)*devPtr] = size; - } - if ( *devPtr ) { - return g_last_cudaError = cudaSuccess; - } else { - return g_last_cudaError = cudaErrorMemoryAllocation; - } -} +#if (CUDART_VERSION >= 2010) -cudaError_t cudaMallocHostInternal(void **ptr, size_t size, gpgpu_context* gpgpu_ctx = NULL) -{ - gpgpu_context *ctx; - if (gpgpu_ctx){ - ctx = gpgpu_ctx; - } else { - ctx = GPGPU_Context(); - } - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - *ptr = malloc(size); - if ( *ptr ) { - //track pinned memory size allocated in the host so that same amount of memory is also allocated in GPU. - ctx->api->pinned_memory_size[*ptr]=size; - return g_last_cudaError = cudaSuccess; - } else { - return g_last_cudaError = cudaErrorMemoryAllocation; - } +cudaError_t cudaHostAllocInternal(void **pHost, size_t bytes, + unsigned int flags, + gpgpu_context *gpgpu_ctx = NULL) { + gpgpu_context *ctx; + if (gpgpu_ctx) { + ctx = gpgpu_ctx; + } else { + ctx = GPGPU_Context(); + } + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + *pHost = malloc(bytes); + // need to track the size allocated so that cudaHostGetDevicePointer() can + // function properly. + // TODO: vary this function behavior based on flags value (following nvidia + // documentation) + ctx->api->pinned_memory_size[*pHost] = bytes; + if (*pHost) + return g_last_cudaError = cudaSuccess; + else + return g_last_cudaError = cudaErrorMemoryAllocation; } -__host__ cudaError_t CUDARTAPI cudaMallocPitchInternal(void **devPtr, size_t *pitch, size_t width, size_t height, gpgpu_context* gpgpu_ctx = NULL) -{ - gpgpu_context *ctx; - if (gpgpu_ctx){ - ctx = gpgpu_ctx; - } else { - ctx = GPGPU_Context(); - } - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - unsigned malloc_width_inbytes = width; - printf("GPGPU-Sim PTX: cudaMallocPitch (width = %d)\n", malloc_width_inbytes); - CUctx_st* context = GPGPUSim_Context(ctx); - *devPtr = context->get_device()->get_gpgpu()->gpu_malloc(malloc_width_inbytes*height); - pitch[0] = malloc_width_inbytes; - if ( *devPtr ) { - return g_last_cudaError = cudaSuccess; - } else { - return g_last_cudaError = cudaErrorMemoryAllocation; - } -} - -cudaError_t cudaHostGetDevicePointerInternal(void **pDevice, void *pHost, unsigned int flags, gpgpu_context* gpgpu_ctx = NULL) -{ - gpgpu_context *ctx; - if (gpgpu_ctx){ - ctx = gpgpu_ctx; - } else { - ctx = GPGPU_Context(); - } - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - //only cpu memory allocation happens in cudaHostAlloc. Linking with device pointer to pinned memory happens here. - //TODO: once kernel is executed, the contents in global pointer of GPU must be copied back to CPU host pointer! - flags=0; - CUctx_st* context = GPGPUSim_Context(ctx); - gpgpu_t *gpu = context->get_device()->get_gpgpu(); - std::map::const_iterator i = ctx->api->pinned_memory_size.find(pHost); - assert(i != ctx->api->pinned_memory_size.end()); - size_t size = i->second; - *pDevice = gpu->gpu_malloc(size); - if(g_debug_execution >= 3){ - printf("GPGPU-Sim PTX: cudaMallocing %zu bytes starting at 0x%llx..\n",size, (unsigned long long) *pDevice); - ctx->api->g_mallocPtr_Size[(unsigned long long)*pDevice] = size; - } - if ( *pDevice ) { - ctx->api->pinned_memory[pHost]=pDevice; - //Copy contents in cpu to gpu - gpu->memcpy_to_gpu((size_t)*pDevice,pHost,size); - return g_last_cudaError = cudaSuccess; - } else { - return g_last_cudaError = cudaErrorMemoryAllocation; - } -} +#endif -__host__ cudaError_t CUDARTAPI cudaMallocArrayInternal(struct cudaArray **array, const struct cudaChannelFormatDesc *desc, size_t width, size_t height __dv(1), gpgpu_context* gpgpu_ctx = NULL) -{ - gpgpu_context *ctx; - if (gpgpu_ctx){ - ctx = gpgpu_ctx; - } else { - ctx = GPGPU_Context(); - } - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - unsigned size = width * height * ((desc->x + desc->y + desc->z + desc->w)/8); - CUctx_st* context = GPGPUSim_Context(ctx); - (*array) = (struct cudaArray*) malloc(sizeof(struct cudaArray)); - (*array)->desc = *desc; - (*array)->width = width; - (*array)->height = height; - (*array)->size = size; - (*array)->dimensions = 2; - ((*array)->devPtr32)= (int) (long long)context->get_device()->get_gpgpu()->gpu_mallocarray(size); - printf("GPGPU-Sim PTX: cudaMallocArray: devPtr32 = %d\n", ((*array)->devPtr32)); - ((*array)->devPtr) = (void*) (long long) ((*array)->devPtr32); - if ( ((*array)->devPtr) ) { - return g_last_cudaError = cudaSuccess; - } else { - return g_last_cudaError = cudaErrorMemoryAllocation; - } -} - -__host__ cudaError_t CUDARTAPI cudaMemcpyInternal(void *dst, const void *src, size_t count, enum cudaMemcpyKind kind, gpgpu_context* gpgpu_ctx = NULL) -{ - gpgpu_context *ctx; - if (gpgpu_ctx){ - ctx = gpgpu_ctx; - } else { - ctx = GPGPU_Context(); - } - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - //CUctx_st *context = GPGPUSim_Context(); - //gpgpu_t *gpu = context->get_device()->get_gpgpu(); - if(g_debug_execution >= 3) - printf("GPGPU-Sim PTX: cudaMemcpy(): devPtr = %p\n", dst); - if( kind == cudaMemcpyHostToDevice ) - ctx->the_gpgpusim->g_stream_manager->push( stream_operation(src,(size_t)dst,count,0) ); - else if( kind == cudaMemcpyDeviceToHost ) - ctx->the_gpgpusim->g_stream_manager->push( stream_operation((size_t)src,dst,count,0) ); - else if( kind == cudaMemcpyDeviceToDevice ) - ctx->the_gpgpusim->g_stream_manager->push( stream_operation((size_t)src,(size_t)dst,count,0) ); - else if ( kind == cudaMemcpyDefault ) { - if ((size_t)src >= GLOBAL_HEAP_START) { - if ((size_t)dst >= GLOBAL_HEAP_START) - ctx->the_gpgpusim->g_stream_manager->push( stream_operation((size_t)src,(size_t)dst,count,0) ); // device to device - else - ctx->the_gpgpusim->g_stream_manager->push( stream_operation((size_t)src,dst,count,0) ); // device to host - } - else { - if ((size_t)dst >= GLOBAL_HEAP_START) - ctx->the_gpgpusim->g_stream_manager->push( stream_operation(src,(size_t)dst,count,0) ); - else { - printf("GPGPU-Sim PTX: cudaMemcpy - ERROR : unsupported transfer: host to host\n"); - abort(); - } - } - } - else { - printf("GPGPU-Sim PTX: cudaMemcpy - ERROR : unsupported cudaMemcpyKind\n"); - abort(); - } - return g_last_cudaError = cudaSuccess; -} - -__host__ cudaError_t CUDARTAPI cudaMemcpyToArrayInternal(struct cudaArray *dst, size_t wOffset, size_t hOffset, const void *src, size_t count, enum cudaMemcpyKind kind, gpgpu_context* gpgpu_ctx = NULL) -{ - gpgpu_context *ctx; - if (gpgpu_ctx){ - ctx = gpgpu_ctx; - } else { - ctx = GPGPU_Context(); - } - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - CUctx_st *context = GPGPUSim_Context(ctx); - gpgpu_t *gpu = context->get_device()->get_gpgpu(); - size_t size = count; - printf("GPGPU-Sim PTX: cudaMemcpyToArray\n"); - if( kind == cudaMemcpyHostToDevice ) - gpu->memcpy_to_gpu( (size_t)(dst->devPtr), src, size); - else if( kind == cudaMemcpyDeviceToHost ) - gpu->memcpy_from_gpu( dst->devPtr, (size_t)src, size); - else if( kind == cudaMemcpyDeviceToDevice ) - gpu->memcpy_gpu_to_gpu( (size_t)(dst->devPtr), (size_t)src, size); - else { - printf("GPGPU-Sim PTX: cudaMemcpyToArray - ERROR : unsupported cudaMemcpyKind\n"); - abort(); - } - dst->devPtr32 = (unsigned) (size_t)(dst->devPtr); - return g_last_cudaError = cudaSuccess; -} - -__host__ cudaError_t CUDARTAPI cudaMemcpy2DInternal(void *dst, size_t dpitch, const void *src, size_t spitch, size_t width, size_t height, enum cudaMemcpyKind kind, gpgpu_context* gpgpu_ctx = NULL) -{ - gpgpu_context *ctx; - if (gpgpu_ctx){ - ctx = gpgpu_ctx; - } else { - ctx = GPGPU_Context(); - } - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - CUctx_st *context = GPGPUSim_Context(ctx); - gpgpu_t *gpu = context->get_device()->get_gpgpu(); - size_t size = spitch*height; - gpgpusim_ptx_assert( (dpitch==spitch), "different src and dst pitch not supported yet" ); - if( kind == cudaMemcpyHostToDevice ) - gpu->memcpy_to_gpu( (size_t)dst, src, size ); - else if( kind == cudaMemcpyDeviceToHost ) - gpu->memcpy_from_gpu( dst, (size_t)src, size ); - else if( kind == cudaMemcpyDeviceToDevice ) - gpu->memcpy_gpu_to_gpu( (size_t)dst, (size_t)src, size); - else { - printf("GPGPU-Sim PTX: cudaMemcpy2D - ERROR : unsupported cudaMemcpyKind\n"); - abort(); - } - return g_last_cudaError = cudaSuccess; -} - -__host__ cudaError_t CUDARTAPI cudaMemcpy2DToArrayInternal(struct cudaArray *dst, size_t wOffset, size_t hOffset, const void *src, size_t spitch, size_t width, size_t height, enum cudaMemcpyKind kind, gpgpu_context* gpgpu_ctx = NULL) -{ - gpgpu_context *ctx; - if (gpgpu_ctx){ - ctx = gpgpu_ctx; - } else { - ctx = GPGPU_Context(); - } - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - CUctx_st *context = GPGPUSim_Context(ctx); - gpgpu_t *gpu = context->get_device()->get_gpgpu(); - size_t size = spitch*height; - size_t channel_size = dst->desc.w+dst->desc.x+dst->desc.y+dst->desc.z; - gpgpusim_ptx_assert( ((channel_size%8) == 0), "none byte multiple destination channel size not supported (sz=%u)", channel_size ); - unsigned elem_size = channel_size/8; - gpgpusim_ptx_assert( (dst->dimensions==2), "copy to none 2D array not supported" ); - gpgpusim_ptx_assert( (wOffset==0), "non-zero wOffset not yet supported" ); - gpgpusim_ptx_assert( (hOffset==0), "non-zero hOffset not yet supported" ); - gpgpusim_ptx_assert( (dst->height == (int)height), "partial copy not supported" ); - gpgpusim_ptx_assert( (elem_size*dst->width == width), "partial copy not supported" ); - gpgpusim_ptx_assert( (spitch == width), "spitch != width not supported" ); - if( kind == cudaMemcpyHostToDevice ) - gpu->memcpy_to_gpu( (size_t)(dst->devPtr), src, size); - else if( kind == cudaMemcpyDeviceToHost ) - gpu->memcpy_from_gpu( dst->devPtr, (size_t)src, size); - else if( kind == cudaMemcpyDeviceToDevice ) - gpu->memcpy_gpu_to_gpu( (size_t)dst->devPtr, (size_t)src, size); - else { - printf("GPGPU-Sim PTX: cudaMemcpy2D - ERROR : unsupported cudaMemcpyKind\n"); - abort(); - } - dst->devPtr32 = (unsigned) (size_t)(dst->devPtr); - return g_last_cudaError = cudaSuccess; -} - -__host__ cudaError_t CUDARTAPI cudaMemcpyToSymbolInternal(const char *symbol, const void *src, size_t count, size_t offset __dv(0), enum cudaMemcpyKind kind __dv(cudaMemcpyHostToDevice), gpgpu_context* gpgpu_ctx = NULL) -{ - gpgpu_context *ctx; - if (gpgpu_ctx){ - ctx = gpgpu_ctx; - } else { - ctx = GPGPU_Context(); - } - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - //CUctx_st *context = GPGPUSim_Context(); - assert(kind == cudaMemcpyHostToDevice); - printf("GPGPU-Sim PTX: cudaMemcpyToSymbol: symbol = %p\n", symbol); - //stream_operation( const char *symbol, const void *src, size_t count, size_t offset ) - ctx->the_gpgpusim->g_stream_manager->push( stream_operation(src,symbol,count,offset,0) ); - //gpgpu_ptx_sim_memcpy_symbol(symbol,src,count,offset,1,context->get_device()->get_gpgpu()); - return g_last_cudaError = cudaSuccess; -} +size_t getMaxThreadsPerBlock(struct cudaFuncAttributes *attr, + gpgpu_context *ctx) { + _cuda_device_id *dev = ctx->GPGPUSim_Init(); + struct cudaDeviceProp prop; + prop = *dev->get_prop(); -__host__ cudaError_t CUDARTAPI cudaMemcpyFromSymbolInternal(void *dst, const char *symbol, size_t count, size_t offset __dv(0), enum cudaMemcpyKind kind __dv(cudaMemcpyDeviceToHost), gpgpu_context* gpgpu_ctx = NULL) -{ - gpgpu_context *ctx; - if (gpgpu_ctx){ - ctx = gpgpu_ctx; - } else { - ctx = GPGPU_Context(); - } - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - //CUctx_st *context = GPGPUSim_Context(); - assert(kind == cudaMemcpyDeviceToHost); - printf("GPGPU-Sim PTX: cudaMemcpyFromSymbol: symbol = %p\n", symbol); - ctx->the_gpgpusim->g_stream_manager->push( stream_operation(symbol,dst,count,offset,0) ); - //gpgpu_ptx_sim_memcpy_symbol(symbol,dst,count,offset,0,context->get_device()->get_gpgpu()); - return g_last_cudaError = cudaSuccess; -} + size_t max = prop.maxThreadsPerBlock; -__host__ cudaError_t CUDARTAPI cudaMemcpyAsyncInternal(void *dst, const void *src, size_t count, enum cudaMemcpyKind kind, cudaStream_t stream, gpgpu_context* gpgpu_ctx = NULL) -{ - gpgpu_context *ctx; - if (gpgpu_ctx){ - ctx = gpgpu_ctx; - } else { - ctx = GPGPU_Context(); - } - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - struct CUstream_st *s = (struct CUstream_st *)stream; - switch( kind ) { - case cudaMemcpyHostToDevice: ctx->the_gpgpusim->g_stream_manager->push( stream_operation(src,(size_t)dst,count,s) ); break; - case cudaMemcpyDeviceToHost: ctx->the_gpgpusim->g_stream_manager->push( stream_operation((size_t)src,dst,count,s) ); break; - case cudaMemcpyDeviceToDevice: ctx->the_gpgpusim->g_stream_manager->push( stream_operation((size_t)src,(size_t)dst,count,s) ); break; - default: - abort(); - } - return g_last_cudaError = cudaSuccess; -} + if (attr->numRegs && (prop.regsPerBlock / attr->numRegs) < max) + max = prop.regsPerBlock / attr->numRegs; + if (attr->sharedSizeBytes && + (prop.sharedMemPerBlock / attr->sharedSizeBytes) < max) + max = prop.sharedMemPerBlock / attr->sharedSizeBytes; -#if (CUDART_VERSION >= 8000) -cudaError_t CUDARTAPI cudaOccupancyMaxActiveBlocksPerMultiprocessorWithFlagsInternal(int* numBlocks, const char *hostFunc, int blockSize, size_t dynamicSMemSize, unsigned int flags, gpgpu_context* gpgpu_ctx = NULL) -{ - gpgpu_context *ctx; - if (gpgpu_ctx){ - ctx = gpgpu_ctx; - } else { - ctx = GPGPU_Context(); - } - printf("GPGPU-Sim PTX: cudaOccupancyMaxActiveBlocksPerMultiprocessorWithFlags %p\n", hostFunc); - CUctx_st *context = GPGPUSim_Context(ctx); - function_info *entry = context->get_kernel(hostFunc); - printf("Calculate Maxium Active Block with function ptr=%p, blockSize=%d, SMemSize=%d\n", hostFunc, blockSize, dynamicSMemSize); - if (flags == cudaOccupancyDefault) { - //create kernel_info based on entry - dim3 gridDim(context->get_device()->get_gpgpu()->max_cta_per_core() - * context->get_device()->get_gpgpu()->get_config().num_shader()); - dim3 blockDim(blockSize); - kernel_info_t result(gridDim, blockDim, entry); - //if(entry == NULL){ - // *numBlocks = 1; - // return g_last_cudaError = cudaErrorUnknown; - //} - *numBlocks = context->get_device()->get_gpgpu()->get_max_cta(result); - printf("Maximum size is %d with gridDim %d and blockDim %d\n", *numBlocks, gridDim.x, blockDim.x); - return g_last_cudaError = cudaSuccess; - } else { - cuda_not_implemented(__my_func__,__LINE__); - return g_last_cudaError = cudaErrorUnknown; - } + return max; } +cudaError_t CUDARTAPI cudaFuncGetAttributesInternal( + struct cudaFuncAttributes *attr, const char *hostFun, + gpgpu_context *gpgpu_ctx = NULL) { + gpgpu_context *ctx; + if (gpgpu_ctx) { + ctx = gpgpu_ctx; + } else { + ctx = GPGPU_Context(); + } + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + CUctx_st *context = GPGPUSim_Context(ctx); + function_info *entry = context->get_kernel(hostFun); + if (entry) { + const struct gpgpu_ptx_sim_info *kinfo = entry->get_kernel_info(); + attr->sharedSizeBytes = kinfo->smem; + attr->constSizeBytes = kinfo->cmem; + attr->localSizeBytes = kinfo->lmem; + attr->numRegs = kinfo->regs; + if (kinfo->maxthreads > 0) + attr->maxThreadsPerBlock = kinfo->maxthreads; + else + attr->maxThreadsPerBlock = getMaxThreadsPerBlock(attr, ctx); +#if CUDART_VERSION >= 3000 + attr->ptxVersion = kinfo->ptx_version; + attr->binaryVersion = kinfo->sm_target; #endif - -__host__ cudaError_t CUDARTAPI cudaMemsetInternal(void *mem, int c, size_t count, gpgpu_context* gpgpu_ctx = NULL) -{ - gpgpu_context *ctx; - if (gpgpu_ctx){ - ctx = gpgpu_ctx; - } else { - ctx = GPGPU_Context(); - } - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - CUctx_st *context = GPGPUSim_Context(ctx); - gpgpu_t *gpu = context->get_device()->get_gpgpu(); - gpu->gpu_memset((size_t)mem, c, count); - return g_last_cudaError = cudaSuccess; + } + return g_last_cudaError = cudaSuccess; } -//memset operation is done but i think its not async? -__host__ cudaError_t CUDARTAPI cudaMemsetAsyncInternal(void *mem, int c, size_t count, cudaStream_t stream=0, gpgpu_context* gpgpu_ctx = NULL) -{ - gpgpu_context *ctx; - if (gpgpu_ctx){ - ctx = gpgpu_ctx; - } else { - ctx = GPGPU_Context(); - } - if(g_debug_execution >= 3){ - announce_call(__my_func__); +#if (CUDART_VERSION > 5000) +__host__ cudaError_t CUDARTAPI +cudaDeviceGetAttributeInternal(int *value, enum cudaDeviceAttr attr, int device, + gpgpu_context *gpgpu_ctx = NULL) { + gpgpu_context *ctx; + if (gpgpu_ctx) { + ctx = gpgpu_ctx; + } else { + ctx = GPGPU_Context(); + } + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + + const struct cudaDeviceProp *prop; + _cuda_device_id *dev = ctx->GPGPUSim_Init(); + + if (device <= dev->num_devices()) { + prop = dev->get_prop(); + switch (attr) { + case 1: + *value = prop->maxThreadsPerBlock; + break; + case 2: + *value = prop->maxThreadsDim[0]; + break; + case 3: + *value = prop->maxThreadsDim[1]; + break; + case 4: + *value = prop->maxThreadsDim[2]; + break; + case 5: + *value = prop->maxGridSize[0]; + break; + case 6: + *value = prop->maxGridSize[1]; + break; + case 7: + *value = prop->maxGridSize[2]; + break; + case 8: + *value = prop->sharedMemPerBlock; + break; + case 9: + *value = prop->totalConstMem; + break; + case 10: + *value = prop->warpSize; + break; + case 11: + *value = 16; // dummy value + break; + case 12: + *value = prop->regsPerBlock; + break; + case 13: + *value = 1480000; // for 1080ti + break; + case 14: + *value = prop->textureAlignment; + break; + case 15: + *value = 0; + break; + case 16: + *value = prop->multiProcessorCount; + break; + case 17: + case 18: + case 19: + *value = 0; + break; + case 21: + case 22: + case 23: + case 24: + case 25: + case 26: + case 27: + case 28: + case 42: + case 45: + case 46: + case 47: + case 48: + case 49: + case 52: + case 53: + case 55: + case 56: + case 57: + case 58: + case 59: + case 60: + case 61: + case 62: + case 63: + case 64: + case 66: + case 67: + case 69: + case 70: + case 71: + case 73: + case 74: + case 77: + *value = 1000; // dummy value + break; + case 29: + case 43: + case 54: + case 65: + case 68: + case 72: + *value = 10; // dummy value + break; + case 30: + case 51: + *value = 128; // dummy value + break; + case 31: + *value = 1; + break; + case 32: + *value = 0; + break; + case 33: + case 50: + *value = 0; // dummy value + break; + case 34: + *value = 0; + break; + case 35: + *value = 0; + break; + case 36: + *value = 1250000; // CK value for 1080ti + break; + case 37: + *value = 352; // value for 1080ti + break; + case 38: + *value = 3000000; // value for 1080ti + break; + case 39: + *value = dev->get_gpgpu()->threads_per_core(); + break; + case 40: + *value = 0; + break; + case 41: + *value = 0; + break; + case 75: // cudaDevAttrComputeCapabilityMajor + *value = prop->major; + break; + case 76: // cudaDevAttrComputeCapabilityMinor + *value = prop->minor; + break; + case 78: + *value = 0; // TODO: as of now, we dont support stream priorities. + break; + case 79: + *value = 0; + break; + case 80: + *value = 0; + break; +#if (CUDART_VERSION > 5050) + case 81: + *value = prop->sharedMemPerMultiprocessor; + break; + case 82: + *value = prop->regsPerMultiprocessor; + break; +#endif + case 83: + case 84: + case 85: + case 86: + *value = 0; + break; + case 87: + *value = 4; // dummy value + break; + case 88: + case 89: + *value = 0; + break; + default: + printf("ERROR: Attribute number %d unimplemented \n", attr); + abort(); } - printf("GPGPU-Sim PTX: WARNING: Asynchronous memset not supported (%s)\n", __my_func__); - CUctx_st *context = GPGPUSim_Context(ctx); - gpgpu_t *gpu = context->get_device()->get_gpgpu(); - gpu->gpu_memset((size_t)mem, c, count); - return g_last_cudaError = cudaSuccess; + return g_last_cudaError = cudaSuccess; + } else { + return g_last_cudaError = cudaErrorInvalidDevice; + } } +#endif -cudaError_t cudaGLMapBufferObjectInternal(void** devPtr, GLuint bufferObj, gpgpu_context* gpgpu_ctx = NULL) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } -#ifdef OPENGL_SUPPORT - gpgpu_context *ctx; - if (gpgpu_ctx){ - ctx = gpgpu_ctx; - } else { - ctx = GPGPU_Context(); - } - GLint buffer_size=0; - CUctx_st* context = GPGPUSim_Context(ctx); - - glbmap_entry_t *p = ctx->api->g_glbmap; - while ( p && p->m_bufferObj != bufferObj ) - p = p->m_next; - if ( p == NULL ) { - glBindBuffer(GL_ARRAY_BUFFER,bufferObj); - glGetBufferParameteriv(GL_ARRAY_BUFFER,GL_BUFFER_SIZE,&buffer_size); - assert( buffer_size != 0 ); - *devPtr = context->get_device()->get_gpgpu()->gpu_malloc(buffer_size); - - // create entry and insert to front of list - glbmap_entry_t *n = (glbmap_entry_t *) calloc(1,sizeof(glbmap_entry_t)); - n->m_next = ctx->api->g_glbmap; - ctx->api->g_glbmap = n; - - // initialize entry - n->m_bufferObj = bufferObj; - n->m_devPtr = *devPtr; - n->m_size = buffer_size; - - p = n; - } else { - buffer_size = p->m_size; - *devPtr = p->m_devPtr; - } - - if ( *devPtr ) { - char *data = (char *) calloc(p->m_size,1); - glGetBufferSubData(GL_ARRAY_BUFFER,0,buffer_size,data); - memcpy_to_gpu( (size_t) *devPtr, data, buffer_size ); - free(data); - printf("GPGPU-Sim PTX: cudaGLMapBufferObject %zu bytes starting at 0x%llx..\n", (size_t)buffer_size, - (unsigned long long) *devPtr); - return g_last_cudaError = cudaSuccess; - } else { - return g_last_cudaError = cudaErrorMemoryAllocation; - } - - return g_last_cudaError = cudaSuccess; +__host__ cudaError_t CUDARTAPI cudaBindTextureInternal( + size_t *offset, const struct textureReference *texref, const void *devPtr, + const struct cudaChannelFormatDesc *desc, size_t size __dv(UINT_MAX), + gpgpu_context *gpgpu_ctx = NULL) { + gpgpu_context *ctx; + if (gpgpu_ctx) { + ctx = gpgpu_ctx; + } else { + ctx = GPGPU_Context(); + } + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + CUctx_st *context = GPGPUSim_Context(ctx); + gpgpu_t *gpu = context->get_device()->get_gpgpu(); + printf( + "GPGPU-Sim PTX: in cudaBindTexture: sizeof(struct textureReference) = " + "%zu\n", + sizeof(struct textureReference)); + struct cudaArray *array; + array = (struct cudaArray *)malloc(sizeof(struct cudaArray)); + array->desc = *desc; + array->size = size; + array->width = size; + array->height = 1; + array->dimensions = 1; + array->devPtr = (void *)devPtr; + array->devPtr32 = (int)(long long)devPtr; + offset = 0; + printf("GPGPU-Sim PTX: size = %zu\n", size); + printf("GPGPU-Sim PTX: texref = %p, array = %p\n", texref, array); + printf("GPGPU-Sim PTX: devPtr32 = %x\n", array->devPtr32); + printf("GPGPU-Sim PTX: Name corresponding to textureReference: %s\n", + gpu->gpgpu_ptx_sim_findNamefromTexture(texref)); + printf("GPGPU-Sim PTX: ChannelFormatDesc: x=%d, y=%d, z=%d, w=%d\n", + desc->x, desc->y, desc->z, desc->w); + printf("GPGPU-Sim PTX: Texture Normalized? = %d\n", texref->normalized); + gpu->gpgpu_ptx_sim_bindTextureToArray(texref, array); + devPtr = (void *)(long long)array->devPtr32; + printf("GPGPU-Sim PTX: devPtr = %p\n", devPtr); + return g_last_cudaError = cudaSuccess; +} + +__host__ cudaError_t CUDARTAPI cudaBindTextureToArrayInternal( + const struct textureReference *texref, const struct cudaArray *array, + const struct cudaChannelFormatDesc *desc, gpgpu_context *gpgpu_ctx = NULL) { + gpgpu_context *ctx; + if (gpgpu_ctx) { + ctx = gpgpu_ctx; + } else { + ctx = GPGPU_Context(); + } + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + CUctx_st *context = GPGPUSim_Context(ctx); + gpgpu_t *gpu = context->get_device()->get_gpgpu(); + printf("GPGPU-Sim PTX: in cudaBindTextureToArray: %p %p\n", texref, array); + printf("GPGPU-Sim PTX: devPtr32 = %x\n", array->devPtr32); + printf("GPGPU-Sim PTX: Name corresponding to textureReference: %s\n", + gpu->gpgpu_ptx_sim_findNamefromTexture(texref)); + printf("GPGPU-Sim PTX: Texture Normalized? = %d\n", texref->normalized); + gpu->gpgpu_ptx_sim_bindTextureToArray(texref, array); + return g_last_cudaError = cudaSuccess; +} + +__host__ cudaError_t CUDARTAPI cudaUnbindTextureInternal( + const struct textureReference *texref, gpgpu_context *gpgpu_ctx = NULL) { + gpgpu_context *ctx; + if (gpgpu_ctx) { + ctx = gpgpu_ctx; + } else { + ctx = GPGPU_Context(); + } + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + CUctx_st *context = GPGPUSim_Context(ctx); + gpgpu_t *gpu = context->get_device()->get_gpgpu(); + printf( + "GPGPU-Sim PTX: in cudaUnbindTexture: sizeof(struct textureReference) = " + "%zu\n", + sizeof(struct textureReference)); + printf("GPGPU-Sim PTX: Name corresponding to textureReference: %s\n", + gpu->gpgpu_ptx_sim_findNamefromTexture(texref)); + + gpu->gpgpu_ptx_sim_unbindTexture(texref); + return g_last_cudaError = cudaSuccess; +} + +__host__ cudaError_t CUDARTAPI cudaLaunchKernelInternal( + const char *hostFun, dim3 gridDim, dim3 blockDim, const void **args, + size_t sharedMem, cudaStream_t stream, gpgpu_context *gpgpu_ctx = NULL) { + gpgpu_context *ctx; + if (gpgpu_ctx) { + ctx = gpgpu_ctx; + } else { + ctx = GPGPU_Context(); + } + + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + CUctx_st *context = GPGPUSim_Context(ctx); + function_info *entry = context->get_kernel(hostFun); +#if CUDART_VERSION < 10000 + cudaConfigureCallInternal(gridDim, blockDim, sharedMem, stream, ctx); +#endif + for (unsigned i = 0; i < entry->num_args(); i++) { + std::pair p = entry->get_param_config(i); + cudaSetupArgumentInternal(args[i], p.first, p.second); + } + + cudaLaunchInternal(hostFun); + return g_last_cudaError = cudaSuccess; +} + +__host__ cudaError_t CUDARTAPI cudaStreamCreateInternal( + cudaStream_t *stream, gpgpu_context *gpgpu_ctx = NULL) { + gpgpu_context *ctx; + if (gpgpu_ctx) { + ctx = gpgpu_ctx; + } else { + ctx = GPGPU_Context(); + } + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("GPGPU-Sim PTX: cudaStreamCreate\n"); +#if (CUDART_VERSION >= 3000) + *stream = new struct CUstream_st(); + ctx->the_gpgpusim->g_stream_manager->add_stream(*stream); #else - fflush(stdout); - fflush(stderr); - printf("GPGPU-Sim PTX: GPGPU-Sim support for OpenGL integration disabled -- exiting\n"); - fflush(stdout); - exit(50); + *stream = 0; + printf( + "GPGPU-Sim PTX: WARNING: Asynchronous kernel execution not supported " + "(%s)\n", + __my_func__); #endif -} - -#if CUDART_VERSION >= 6050 -CUresult -cuLinkAddFileInternal(CUlinkState state, CUjitInputType type, const char *path, - unsigned int numOptions, CUjit_option *options, void **optionValues, gpgpu_context* gpgpu_ctx = NULL) -{ - gpgpu_context *ctx; - if (gpgpu_ctx){ - ctx = gpgpu_ctx; - } else { - ctx = GPGPU_Context(); - } - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - static bool addedFile = false; - if (addedFile){ - printf("GPGPU-Sim PTX: ERROR: cuLinkAddFile does not support multiple files\n"); - abort(); - } - - //blocking - assert(type==CU_JIT_INPUT_PTX); - CUctx_st *context = GPGPUSim_Context(ctx); - char *file = getenv("PTX_JIT_PATH"); - if(file==NULL){ - printf("GPGPU-Sim PTX: ERROR: PTX_JIT_PATH has not been set\n"); - abort(); - } - strcat(file,"/"); - strcat(file,path); - symbol_table *symtab = ctx->gpgpu_ptx_sim_load_ptx_from_filename( file ); - std::string fname(path); - ctx->api->name_symtab[fname] = symtab; - context->add_binary(symtab, 1); - ctx->api->load_static_globals(symtab,STATIC_ALLOC_LIMIT,0xFFFFFFFF,context->get_device()->get_gpgpu()); - ctx->api->load_constants(symtab,STATIC_ALLOC_LIMIT,context->get_device()->get_gpgpu()); - addedFile = true; - return CUDA_SUCCESS; -} -#endif - -#if (CUDART_VERSION >= 2010) - -cudaError_t cudaHostAllocInternal(void **pHost, size_t bytes, unsigned int flags, gpgpu_context* gpgpu_ctx = NULL) -{ - gpgpu_context *ctx; - if (gpgpu_ctx){ - ctx = gpgpu_ctx; - } else { - ctx = GPGPU_Context(); - } - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - *pHost = malloc(bytes); - //need to track the size allocated so that cudaHostGetDevicePointer() can function properly. - //TODO: vary this function behavior based on flags value (following nvidia documentation) - ctx->api->pinned_memory_size[*pHost]=bytes; - if( *pHost ) - return g_last_cudaError = cudaSuccess; - else - return g_last_cudaError = cudaErrorMemoryAllocation; -} - -#endif - -size_t getMaxThreadsPerBlock(struct cudaFuncAttributes *attr, gpgpu_context *ctx) { - _cuda_device_id *dev = ctx->GPGPUSim_Init(); - struct cudaDeviceProp prop; - - prop = *dev->get_prop(); - - size_t max = prop.maxThreadsPerBlock; - - if (attr->numRegs && (prop.regsPerBlock / attr->numRegs) < max) - max = prop.regsPerBlock / attr->numRegs; - - if (attr->sharedSizeBytes && (prop.sharedMemPerBlock / attr->sharedSizeBytes) < max) - max = prop.sharedMemPerBlock / attr->sharedSizeBytes; - - return max; -} - -cudaError_t CUDARTAPI cudaFuncGetAttributesInternal(struct cudaFuncAttributes *attr, const char *hostFun, gpgpu_context* gpgpu_ctx = NULL ) -{ - gpgpu_context *ctx; - if (gpgpu_ctx){ - ctx = gpgpu_ctx; - } else { - ctx = GPGPU_Context(); - } - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - CUctx_st *context = GPGPUSim_Context(ctx); - function_info *entry = context->get_kernel(hostFun); - if( entry ) { - const struct gpgpu_ptx_sim_info *kinfo = entry->get_kernel_info(); - attr->sharedSizeBytes = kinfo->smem; - attr->constSizeBytes = kinfo->cmem; - attr->localSizeBytes = kinfo->lmem; - attr->numRegs = kinfo->regs; - if(kinfo->maxthreads > 0) - attr->maxThreadsPerBlock = kinfo->maxthreads; - else - attr->maxThreadsPerBlock = getMaxThreadsPerBlock(attr, ctx); -#if CUDART_VERSION >= 3000 - attr->ptxVersion = kinfo->ptx_version; - attr->binaryVersion = kinfo->sm_target; -#endif - } - return g_last_cudaError = cudaSuccess; -} - -#if (CUDART_VERSION > 5000) -__host__ cudaError_t CUDARTAPI cudaDeviceGetAttributeInternal(int *value, enum cudaDeviceAttr attr, int device, gpgpu_context* gpgpu_ctx = NULL) -{ - gpgpu_context *ctx; - if (gpgpu_ctx){ - ctx = gpgpu_ctx; - } else { - ctx = GPGPU_Context(); - } - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - - const struct cudaDeviceProp *prop; - _cuda_device_id *dev = ctx->GPGPUSim_Init(); - - if (device <= dev->num_devices() ) { - prop = dev->get_prop(); - switch (attr) { - case 1: - *value= prop->maxThreadsPerBlock; - break; - case 2: - *value= prop->maxThreadsDim[0]; - break; - case 3: - *value= prop->maxThreadsDim[1]; - break; - case 4: - *value= prop->maxThreadsDim[2]; - break; - case 5: - *value= prop->maxGridSize[0]; - break; - case 6: - *value= prop->maxGridSize[1]; - break; - case 7: - *value= prop->maxGridSize[2]; - break; - case 8: - *value= prop->sharedMemPerBlock; - break; - case 9: - *value= prop->totalConstMem; - break; - case 10: - *value= prop->warpSize; - break; - case 11: - *value= 16;//dummy value - break; - case 12: - *value= prop->regsPerBlock; - break; - case 13: - *value= 1480000;//for 1080ti - break; - case 14: - *value= prop->textureAlignment ; - break; - case 15: - *value = 0; - break; - case 16: - *value= prop->multiProcessorCount ; - break; - case 17: - case 18: - case 19: - *value = 0; - break; - case 21: - case 22: - case 23: - case 24: - case 25: - case 26: - case 27: - case 28: - case 42: - case 45: - case 46: - case 47: - case 48: - case 49: - case 52: - case 53: - case 55: - case 56: - case 57: - case 58: - case 59: - case 60: - case 61: - case 62: - case 63: - case 64: - case 66: - case 67: - case 69: - case 70: - case 71: - case 73: - case 74: - case 77: - *value = 1000;//dummy value - break; - case 29: - case 43: - case 54: - case 65: - case 68: - case 72: - *value = 10;//dummy value - break; - case 30: - case 51: - *value = 128;//dummy value - break; - case 31: - *value = 1; - break; - case 32: - *value = 0; - break; - case 33: - case 50: - *value = 0;//dummy value - break; - case 34: - *value= 0; - break; - case 35: - *value = 0; - break; - case 36: - *value = 1250000;//CK value for 1080ti - break; - case 37: - *value = 352;//value for 1080ti - break; - case 38: - *value = 3000000;//value for 1080ti - break; - case 39: - *value= dev->get_gpgpu()->threads_per_core(); - break; - case 40: - *value= 0; - break; - case 41: - *value= 0; - break; - case 75://cudaDevAttrComputeCapabilityMajor - *value= prop->major ; - break; - case 76://cudaDevAttrComputeCapabilityMinor - *value= prop->minor ; - break; - case 78: - *value= 0 ; //TODO: as of now, we dont support stream priorities. - break; - case 79: - *value= 0; - break; - case 80: - *value= 0; - break; - #if (CUDART_VERSION > 5050) - case 81: - *value= prop->sharedMemPerMultiprocessor; - break; - case 82: - *value= prop->regsPerMultiprocessor; - break; - #endif - case 83: - case 84: - case 85: - case 86: - *value= 0; - break; - case 87: - *value= 4;//dummy value - break; - case 88: - case 89: - *value= 0; - break; - default: - printf("ERROR: Attribute number %d unimplemented \n",attr); - abort(); - } - return g_last_cudaError = cudaSuccess; - } else { - return g_last_cudaError = cudaErrorInvalidDevice; - } -} -#endif - -__host__ cudaError_t CUDARTAPI cudaBindTextureInternal(size_t *offset, - const struct textureReference *texref, - const void *devPtr, - const struct cudaChannelFormatDesc *desc, - size_t size __dv(UINT_MAX), - gpgpu_context* gpgpu_ctx = NULL) -{ - gpgpu_context *ctx; - if (gpgpu_ctx){ - ctx = gpgpu_ctx; - } else { - ctx = GPGPU_Context(); - } - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - CUctx_st *context = GPGPUSim_Context(ctx); - gpgpu_t *gpu = context->get_device()->get_gpgpu(); - printf("GPGPU-Sim PTX: in cudaBindTexture: sizeof(struct textureReference) = %zu\n", sizeof(struct textureReference)); - struct cudaArray *array; - array = (struct cudaArray*) malloc(sizeof(struct cudaArray)); - array->desc = *desc; - array->size = size; - array->width = size; - array->height = 1; - array->dimensions = 1; - array->devPtr = (void*)devPtr; - array->devPtr32 = (int)(long long)devPtr; - offset = 0; - printf("GPGPU-Sim PTX: size = %zu\n", size); - printf("GPGPU-Sim PTX: texref = %p, array = %p\n", texref, array); - printf("GPGPU-Sim PTX: devPtr32 = %x\n", array->devPtr32); - printf("GPGPU-Sim PTX: Name corresponding to textureReference: %s\n", gpu->gpgpu_ptx_sim_findNamefromTexture(texref)); - printf("GPGPU-Sim PTX: ChannelFormatDesc: x=%d, y=%d, z=%d, w=%d\n", desc->x, desc->y, desc->z, desc->w); - printf("GPGPU-Sim PTX: Texture Normalized? = %d\n", texref->normalized); - gpu->gpgpu_ptx_sim_bindTextureToArray(texref, array); - devPtr = (void*)(long long)array->devPtr32; - printf("GPGPU-Sim PTX: devPtr = %p\n", devPtr); - return g_last_cudaError = cudaSuccess; -} - -__host__ cudaError_t CUDARTAPI cudaBindTextureToArrayInternal(const struct textureReference *texref, const struct cudaArray *array, const struct cudaChannelFormatDesc *desc, gpgpu_context* gpgpu_ctx = NULL) -{ - gpgpu_context *ctx; - if (gpgpu_ctx){ - ctx = gpgpu_ctx; - } else { - ctx = GPGPU_Context(); - } - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - CUctx_st *context = GPGPUSim_Context(ctx); - gpgpu_t *gpu = context->get_device()->get_gpgpu(); - printf("GPGPU-Sim PTX: in cudaBindTextureToArray: %p %p\n", texref, array); - printf("GPGPU-Sim PTX: devPtr32 = %x\n", array->devPtr32); - printf("GPGPU-Sim PTX: Name corresponding to textureReference: %s\n", gpu->gpgpu_ptx_sim_findNamefromTexture(texref)); - printf("GPGPU-Sim PTX: Texture Normalized? = %d\n", texref->normalized); - gpu->gpgpu_ptx_sim_bindTextureToArray(texref, array); - return g_last_cudaError = cudaSuccess; -} - -__host__ cudaError_t CUDARTAPI cudaUnbindTextureInternal(const struct textureReference *texref, gpgpu_context* gpgpu_ctx = NULL) -{ - gpgpu_context *ctx; - if (gpgpu_ctx){ - ctx = gpgpu_ctx; - } else { - ctx = GPGPU_Context(); - } - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - CUctx_st *context = GPGPUSim_Context(ctx); - gpgpu_t *gpu = context->get_device()->get_gpgpu(); - printf("GPGPU-Sim PTX: in cudaUnbindTexture: sizeof(struct textureReference) = %zu\n", sizeof(struct textureReference)); - printf("GPGPU-Sim PTX: Name corresponding to textureReference: %s\n", gpu->gpgpu_ptx_sim_findNamefromTexture(texref)); - - gpu->gpgpu_ptx_sim_unbindTexture(texref); - return g_last_cudaError = cudaSuccess; -} - -__host__ cudaError_t CUDARTAPI cudaLaunchKernelInternal( const char* hostFun, dim3 gridDim, dim3 blockDim, const void** args, size_t sharedMem, cudaStream_t stream, gpgpu_context* gpgpu_ctx = NULL ) -{ - gpgpu_context *ctx; - if (gpgpu_ctx){ - ctx = gpgpu_ctx; - } else { - ctx = GPGPU_Context(); - } - - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - CUctx_st *context = GPGPUSim_Context(ctx); - function_info *entry = context->get_kernel(hostFun); -#if CUDART_VERSION < 10000 - cudaConfigureCallInternal(gridDim, blockDim, sharedMem, stream, ctx); -#endif - for(unsigned i = 0; i < entry->num_args(); i++){ - std::pair p = entry->get_param_config(i); - cudaSetupArgumentInternal(args[i], p.first, p.second); - } - - cudaLaunchInternal(hostFun); - return g_last_cudaError = cudaSuccess; -} - -__host__ cudaError_t CUDARTAPI cudaStreamCreateInternal(cudaStream_t *stream, gpgpu_context* gpgpu_ctx = NULL) -{ - gpgpu_context *ctx; - if (gpgpu_ctx){ - ctx = gpgpu_ctx; - } else { - ctx = GPGPU_Context(); - } - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("GPGPU-Sim PTX: cudaStreamCreate\n"); -#if (CUDART_VERSION >= 3000) - *stream = new struct CUstream_st(); - ctx->the_gpgpusim->g_stream_manager->add_stream(*stream); -#else - *stream = 0; - printf("GPGPU-Sim PTX: WARNING: Asynchronous kernel execution not supported (%s)\n", __my_func__); -#endif - return g_last_cudaError = cudaSuccess; -} - -__host__ cudaError_t CUDARTAPI cudaStreamDestroyInternal(cudaStream_t stream, gpgpu_context* gpgpu_ctx = NULL) -{ - gpgpu_context *ctx; - if (gpgpu_ctx){ - ctx = gpgpu_ctx; - } else { - ctx = GPGPU_Context(); - } - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } + return g_last_cudaError = cudaSuccess; +} + +__host__ cudaError_t CUDARTAPI cudaStreamDestroyInternal( + cudaStream_t stream, gpgpu_context *gpgpu_ctx = NULL) { + gpgpu_context *ctx; + if (gpgpu_ctx) { + ctx = gpgpu_ctx; + } else { + ctx = GPGPU_Context(); + } + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } #if (CUDART_VERSION >= 3000) - //per-stream synchronization required for application using external libraries without explicit synchronization in the code to - //avoid the stream_manager from spinning forever to destroy non-empty streams without making any forward progress. - stream->synchronize(); - ctx->the_gpgpusim->g_stream_manager->destroy_stream(stream); + // per-stream synchronization required for application using external + // libraries without explicit synchronization in the code to avoid the + // stream_manager from spinning forever to destroy non-empty streams without + // making any forward progress. + stream->synchronize(); + ctx->the_gpgpusim->g_stream_manager->destroy_stream(stream); #endif - return g_last_cudaError = cudaSuccess; -} - -__host__ cudaError_t CUDARTAPI cudaStreamSynchronizeInternal(cudaStream_t stream, gpgpu_context* gpgpu_ctx = NULL) -{ - gpgpu_context *ctx; - if (gpgpu_ctx){ - ctx = gpgpu_ctx; - } else { - ctx = GPGPU_Context(); - } - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } + return g_last_cudaError = cudaSuccess; +} + +__host__ cudaError_t CUDARTAPI cudaStreamSynchronizeInternal( + cudaStream_t stream, gpgpu_context *gpgpu_ctx = NULL) { + gpgpu_context *ctx; + if (gpgpu_ctx) { + ctx = gpgpu_ctx; + } else { + ctx = GPGPU_Context(); + } + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } #if (CUDART_VERSION >= 3000) - if( stream == NULL ) - ctx->synchronize(); - return g_last_cudaError = cudaSuccess; - stream->synchronize(); + if (stream == NULL) ctx->synchronize(); + return g_last_cudaError = cudaSuccess; + stream->synchronize(); #else - printf("GPGPU-Sim PTX: WARNING: Asynchronous kernel execution not supported (%s)\n", __my_func__); + printf( + "GPGPU-Sim PTX: WARNING: Asynchronous kernel execution not supported " + "(%s)\n", + __my_func__); #endif - return g_last_cudaError = cudaSuccess; + return g_last_cudaError = cudaSuccess; } void __cudaRegisterTextureInternal( - void **fatCubinHandle, - const struct textureReference *hostVar, - const void **deviceAddress, - const char *deviceName, - int dim, - int norm, - int ext, - gpgpu_context* gpgpu_ctx = NULL -) //passes in a newly created textureReference -{ - gpgpu_context *ctx; - if (gpgpu_ctx){ - ctx = gpgpu_ctx; - } else { - ctx = GPGPU_Context(); - } - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - std::string devStr (deviceName); - #if (CUDART_VERSION > 4020) - if (devStr.size() > 2 && devStr.data()[0] == ':' && devStr.data()[1] == ':') - devStr = devStr.replace(0, 2, ""); - #endif - CUctx_st *context = GPGPUSim_Context(ctx); - gpgpu_t *gpu = context->get_device()->get_gpgpu(); - printf("GPGPU-Sim PTX: in __cudaRegisterTexture:\n"); - gpu->gpgpu_ptx_sim_bindNameToTexture(devStr.data(), hostVar, dim, norm, ext); - printf("GPGPU-Sim PTX: int dim = %d\n", dim); - printf("GPGPU-Sim PTX: int norm = %d\n", norm); - printf("GPGPU-Sim PTX: int ext = %d\n", ext); - printf("GPGPU-Sim PTX: Execution warning: Not finished implementing \"%s\"\n", __my_func__ ); -} - -cudaError_t cudaGLUnmapBufferObjectInternal(GLuint bufferObj, gpgpu_context* gpgpu_ctx = NULL) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } + void **fatCubinHandle, const struct textureReference *hostVar, + const void **deviceAddress, const char *deviceName, int dim, int norm, + int ext, + gpgpu_context *gpgpu_ctx = + NULL) // passes in a newly created textureReference +{ + gpgpu_context *ctx; + if (gpgpu_ctx) { + ctx = gpgpu_ctx; + } else { + ctx = GPGPU_Context(); + } + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + std::string devStr(deviceName); +#if (CUDART_VERSION > 4020) + if (devStr.size() > 2 && devStr.data()[0] == ':' && devStr.data()[1] == ':') + devStr = devStr.replace(0, 2, ""); +#endif + CUctx_st *context = GPGPUSim_Context(ctx); + gpgpu_t *gpu = context->get_device()->get_gpgpu(); + printf("GPGPU-Sim PTX: in __cudaRegisterTexture:\n"); + gpu->gpgpu_ptx_sim_bindNameToTexture(devStr.data(), hostVar, dim, norm, ext); + printf("GPGPU-Sim PTX: int dim = %d\n", dim); + printf("GPGPU-Sim PTX: int norm = %d\n", norm); + printf("GPGPU-Sim PTX: int ext = %d\n", ext); + printf( + "GPGPU-Sim PTX: Execution warning: Not finished implementing \"%s\"\n", + __my_func__); +} + +cudaError_t cudaGLUnmapBufferObjectInternal(GLuint bufferObj, + gpgpu_context *gpgpu_ctx = NULL) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } #ifdef OPENGL_SUPPORT - gpgpu_context *ctx; - if (gpgpu_ctx){ - ctx = gpgpu_ctx; - } else { - ctx = GPGPU_Context(); - } - CUctx_st* ctx = GPGPUSim_Context(ctx); - glbmap_entry_t *p = ctx->api->g_glbmap; - while ( p && p->m_bufferObj != bufferObj ) - p = p->m_next; - if ( p == NULL ) - return g_last_cudaError = cudaErrorUnknown; - - char *data = (char *) calloc(p->m_size,1); - memcpy_from_gpu( data,(size_t)p->m_devPtr,p->m_size ); - glBufferSubData(GL_ARRAY_BUFFER,0,p->m_size,data); - free(data); - - return g_last_cudaError = cudaSuccess; + gpgpu_context *ctx; + if (gpgpu_ctx) { + ctx = gpgpu_ctx; + } else { + ctx = GPGPU_Context(); + } + CUctx_st *ctx = GPGPUSim_Context(ctx); + glbmap_entry_t *p = ctx->api->g_glbmap; + while (p && p->m_bufferObj != bufferObj) p = p->m_next; + if (p == NULL) return g_last_cudaError = cudaErrorUnknown; + + char *data = (char *)calloc(p->m_size, 1); + memcpy_from_gpu(data, (size_t)p->m_devPtr, p->m_size); + glBufferSubData(GL_ARRAY_BUFFER, 0, p->m_size, data); + free(data); + + return g_last_cudaError = cudaSuccess; #else - fflush(stdout); - fflush(stderr); - printf("GPGPU-Sim PTX: support for OpenGL integration disabled -- exiting\n"); - fflush(stdout); - exit(50); + fflush(stdout); + fflush(stderr); + printf("GPGPU-Sim PTX: support for OpenGL integration disabled -- exiting\n"); + fflush(stdout); + exit(50); #endif } #if CUDART_VERSION >= 3000 -__host__ cudaError_t CUDARTAPI cudaFuncSetCacheConfigInternal(const char *func, enum cudaFuncCache cacheConfig, gpgpu_context* gpgpu_ctx = NULL ) -{ - gpgpu_context *ctx; - if (gpgpu_ctx){ - ctx = gpgpu_ctx; - } else { - ctx = GPGPU_Context(); - } - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - CUctx_st *context = GPGPUSim_Context(ctx); - context->get_device()->get_gpgpu()->set_cache_config(context->get_kernel(func)->get_name(), (FuncCache)cacheConfig); - return g_last_cudaError = cudaSuccess; +__host__ cudaError_t CUDARTAPI +cudaFuncSetCacheConfigInternal(const char *func, enum cudaFuncCache cacheConfig, + gpgpu_context *gpgpu_ctx = NULL) { + gpgpu_context *ctx; + if (gpgpu_ctx) { + ctx = gpgpu_ctx; + } else { + ctx = GPGPU_Context(); + } + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + CUctx_st *context = GPGPUSim_Context(ctx); + context->get_device()->get_gpgpu()->set_cache_config( + context->get_kernel(func)->get_name(), (FuncCache)cacheConfig); + return g_last_cudaError = cudaSuccess; } #endif #if CUDART_VERSION >= 4000 -CUresult CUDAAPI cuLaunchKernelInternal(CUfunction f, - unsigned int gridDimX, - unsigned int gridDimY, - unsigned int gridDimZ, - unsigned int blockDimX, - unsigned int blockDimY, - unsigned int blockDimZ, - unsigned int sharedMemBytes, - CUstream hStream, - void **kernelParams, - void **extra, - gpgpu_context* gpgpu_ctx = NULL) -{ - gpgpu_context *ctx; - if (gpgpu_ctx){ - ctx = gpgpu_ctx; - } else { - ctx = GPGPU_Context(); - } - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - if (extra!=NULL){ - printf("GPGPU-Sim CUDA DRIVER API: ERROR: Currently do not support void** extra.\n"); - abort(); - } - const char *hostFun = (const char*) f; - CUctx_st *context = GPGPUSim_Context(ctx); - function_info *entry = context->get_kernel(hostFun); - cudaConfigureCallInternal(dim3(gridDimX, gridDimY, gridDimZ), dim3(blockDimX, blockDimY, blockDimZ), sharedMemBytes, (cudaStream_t) hStream, ctx); - for(unsigned i = 0; i < entry->num_args(); i++){ - std::pair p = entry->get_param_config(i); - cudaSetupArgumentInternal(kernelParams[i], p.first, p.second, ctx); - } - cudaLaunchInternal(hostFun, ctx); - return CUDA_SUCCESS; +CUresult CUDAAPI cuLaunchKernelInternal( + CUfunction f, unsigned int gridDimX, unsigned int gridDimY, + unsigned int gridDimZ, unsigned int blockDimX, unsigned int blockDimY, + unsigned int blockDimZ, unsigned int sharedMemBytes, CUstream hStream, + void **kernelParams, void **extra, gpgpu_context *gpgpu_ctx = NULL) { + gpgpu_context *ctx; + if (gpgpu_ctx) { + ctx = gpgpu_ctx; + } else { + ctx = GPGPU_Context(); + } + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + if (extra != NULL) { + printf( + "GPGPU-Sim CUDA DRIVER API: ERROR: Currently do not support void** " + "extra.\n"); + abort(); + } + const char *hostFun = (const char *)f; + CUctx_st *context = GPGPUSim_Context(ctx); + function_info *entry = context->get_kernel(hostFun); + cudaConfigureCallInternal(dim3(gridDimX, gridDimY, gridDimZ), + dim3(blockDimX, blockDimY, blockDimZ), + sharedMemBytes, (cudaStream_t)hStream, ctx); + for (unsigned i = 0; i < entry->num_args(); i++) { + std::pair p = entry->get_param_config(i); + cudaSetupArgumentInternal(kernelParams[i], p.first, p.second, ctx); + } + cudaLaunchInternal(hostFun, ctx); + return CUDA_SUCCESS; } #endif /* CUDART_VERSION >= 4000 */ -CUevent_st *get_event(cudaEvent_t event) -{ - unsigned event_uid; +CUevent_st *get_event(cudaEvent_t event) { + unsigned event_uid; #if CUDART_VERSION >= 3000 - event_uid = event->get_uid(); + event_uid = event->get_uid(); #else - event_uid = event; + event_uid = event; #endif - event_tracker_t::iterator e = g_timer_events.find(event_uid); - if( e == g_timer_events.end() ) - return NULL; - return e->second; -} - -__host__ cudaError_t CUDARTAPI cudaEventRecordInternal(cudaEvent_t event, cudaStream_t stream, gpgpu_context* gpgpu_ctx = NULL) -{ - gpgpu_context *ctx; - if (gpgpu_ctx){ - ctx = gpgpu_ctx; - } else { - ctx = GPGPU_Context(); - } - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - CUevent_st *e = get_event(event); - if( !e ) return g_last_cudaError = cudaErrorUnknown; - struct CUstream_st *s = (struct CUstream_st *)stream; - stream_operation op(e,s); - ctx->the_gpgpusim->g_stream_manager->push(op); - return g_last_cudaError = cudaSuccess; -} - -__host__ cudaError_t CUDARTAPI cudaStreamWaitEventInternal(cudaStream_t stream, cudaEvent_t event, unsigned int flags, gpgpu_context* gpgpu_ctx = NULL) -{ - gpgpu_context *ctx; - if (gpgpu_ctx){ - ctx = gpgpu_ctx; - } else { - ctx = GPGPU_Context(); - } - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - //reference: https://www.cs.cmu.edu/afs/cs/academic/class/15668-s11/www/cuda-doc/html/group__CUDART__STREAM_gfe68d207dc965685d92d3f03d77b0876.html - CUevent_st *e = get_event(event); - if( !e ){ - printf("GPGPU-Sim API: Warning: cudaEventRecord has not been called on event before calling cudaStreamWaitEvent.\nNothing to be done.\n"); - return g_last_cudaError = cudaSuccess; - } - if (!stream){ - ctx->the_gpgpusim->g_stream_manager->pushCudaStreamWaitEventToAllStreams(e, flags); - } else { - struct CUstream_st *s = (struct CUstream_st *)stream; - stream_operation op(s,e,flags); - ctx->the_gpgpusim->g_stream_manager->push(op); - } - return g_last_cudaError = cudaSuccess; -} - -__host__ cudaError_t CUDARTAPI cudaThreadExitInternal(gpgpu_context* gpgpu_ctx = NULL) -{ - gpgpu_context *ctx; - if (gpgpu_ctx){ - ctx = gpgpu_ctx; - } else { - ctx = GPGPU_Context(); - } - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - ctx->exit_simulation(); - return g_last_cudaError = cudaSuccess; -} - -__host__ cudaError_t CUDARTAPI cudaThreadSynchronizeInternal(gpgpu_context* gpgpu_ctx = NULL) -{ - gpgpu_context *ctx; - if (gpgpu_ctx){ - ctx = gpgpu_ctx; - } else { - ctx = GPGPU_Context(); - } - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - //Called on host side - ctx->synchronize(); - return g_last_cudaError = cudaSuccess; -} - -cudaError_t CUDARTAPI cudaDeviceSynchronizeInternal(gpgpu_context* gpgpu_ctx = NULL) -{ - gpgpu_context *ctx; - if (gpgpu_ctx){ - ctx = gpgpu_ctx; - } else { - ctx = GPGPU_Context(); - } - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - //Blocks until the device has completed all preceding requested tasks - ctx->synchronize(); - return g_last_cudaError = cudaSuccess; + event_tracker_t::iterator e = g_timer_events.find(event_uid); + if (e == g_timer_events.end()) return NULL; + return e->second; +} + +__host__ cudaError_t CUDARTAPI cudaEventRecordInternal( + cudaEvent_t event, cudaStream_t stream, gpgpu_context *gpgpu_ctx = NULL) { + gpgpu_context *ctx; + if (gpgpu_ctx) { + ctx = gpgpu_ctx; + } else { + ctx = GPGPU_Context(); + } + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + CUevent_st *e = get_event(event); + if (!e) return g_last_cudaError = cudaErrorUnknown; + struct CUstream_st *s = (struct CUstream_st *)stream; + stream_operation op(e, s); + ctx->the_gpgpusim->g_stream_manager->push(op); + return g_last_cudaError = cudaSuccess; +} + +__host__ cudaError_t CUDARTAPI cudaStreamWaitEventInternal( + cudaStream_t stream, cudaEvent_t event, unsigned int flags, + gpgpu_context *gpgpu_ctx = NULL) { + gpgpu_context *ctx; + if (gpgpu_ctx) { + ctx = gpgpu_ctx; + } else { + ctx = GPGPU_Context(); + } + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + // reference: + // https://www.cs.cmu.edu/afs/cs/academic/class/15668-s11/www/cuda-doc/html/group__CUDART__STREAM_gfe68d207dc965685d92d3f03d77b0876.html + CUevent_st *e = get_event(event); + if (!e) { + printf( + "GPGPU-Sim API: Warning: cudaEventRecord has not been called on event " + "before calling cudaStreamWaitEvent.\nNothing to be done.\n"); + return g_last_cudaError = cudaSuccess; + } + if (!stream) { + ctx->the_gpgpusim->g_stream_manager->pushCudaStreamWaitEventToAllStreams( + e, flags); + } else { + struct CUstream_st *s = (struct CUstream_st *)stream; + stream_operation op(s, e, flags); + ctx->the_gpgpusim->g_stream_manager->push(op); + } + return g_last_cudaError = cudaSuccess; +} + +__host__ cudaError_t CUDARTAPI +cudaThreadExitInternal(gpgpu_context *gpgpu_ctx = NULL) { + gpgpu_context *ctx; + if (gpgpu_ctx) { + ctx = gpgpu_ctx; + } else { + ctx = GPGPU_Context(); + } + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + ctx->exit_simulation(); + return g_last_cudaError = cudaSuccess; +} + +__host__ cudaError_t CUDARTAPI +cudaThreadSynchronizeInternal(gpgpu_context *gpgpu_ctx = NULL) { + gpgpu_context *ctx; + if (gpgpu_ctx) { + ctx = gpgpu_ctx; + } else { + ctx = GPGPU_Context(); + } + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + // Called on host side + ctx->synchronize(); + return g_last_cudaError = cudaSuccess; +} + +cudaError_t CUDARTAPI +cudaDeviceSynchronizeInternal(gpgpu_context *gpgpu_ctx = NULL) { + gpgpu_context *ctx; + if (gpgpu_ctx) { + ctx = gpgpu_ctx; + } else { + ctx = GPGPU_Context(); + } + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + // Blocks until the device has completed all preceding requested tasks + ctx->synchronize(); + return g_last_cudaError = cudaSuccess; } /******************************************************************************* @@ -2145,142 +2296,148 @@ extern "C" { * * * * *******************************************************************************/ -cudaError_t cudaPeekAtLastError(void) -{ - return g_last_cudaError; -} +cudaError_t cudaPeekAtLastError(void) { return g_last_cudaError; } -__host__ cudaError_t CUDARTAPI cudaMalloc(void **devPtr, size_t size) -{ - return cudaMallocInternal(devPtr, size); +__host__ cudaError_t CUDARTAPI cudaMalloc(void **devPtr, size_t size) { + return cudaMallocInternal(devPtr, size); } -__host__ cudaError_t CUDARTAPI cudaMallocHost(void **ptr, size_t size) -{ - return cudaMallocHostInternal(ptr, size); +__host__ cudaError_t CUDARTAPI cudaMallocHost(void **ptr, size_t size) { + return cudaMallocHostInternal(ptr, size); } -__host__ cudaError_t CUDARTAPI cudaMallocPitch(void **devPtr, size_t *pitch, size_t width, size_t height) -{ - return cudaMallocPitchInternal(devPtr, pitch, width, height); +__host__ cudaError_t CUDARTAPI cudaMallocPitch(void **devPtr, size_t *pitch, + size_t width, size_t height) { + return cudaMallocPitchInternal(devPtr, pitch, width, height); } -__host__ cudaError_t CUDARTAPI cudaMallocArray(struct cudaArray **array, const struct cudaChannelFormatDesc *desc, size_t width, size_t height __dv(1)) -{ - return cudaMallocArrayInternal(array, desc, width, height); +__host__ cudaError_t CUDARTAPI cudaMallocArray( + struct cudaArray **array, const struct cudaChannelFormatDesc *desc, + size_t width, size_t height __dv(1)) { + return cudaMallocArrayInternal(array, desc, width, height); } -__host__ cudaError_t CUDARTAPI cudaFree(void *devPtr) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - // TODO... manage g_global_mem space? - return g_last_cudaError = cudaSuccess; +__host__ cudaError_t CUDARTAPI cudaFree(void *devPtr) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + // TODO... manage g_global_mem space? + return g_last_cudaError = cudaSuccess; } -__host__ cudaError_t CUDARTAPI cudaFreeHost(void *ptr) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - free (ptr); // this will crash the system if called twice - return g_last_cudaError = cudaSuccess; +__host__ cudaError_t CUDARTAPI cudaFreeHost(void *ptr) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + free(ptr); // this will crash the system if called twice + return g_last_cudaError = cudaSuccess; } -__host__ cudaError_t CUDARTAPI cudaFreeArray(struct cudaArray *array) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - // TODO... manage g_global_mem space? - return g_last_cudaError = cudaSuccess; +__host__ cudaError_t CUDARTAPI cudaFreeArray(struct cudaArray *array) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + // TODO... manage g_global_mem space? + return g_last_cudaError = cudaSuccess; }; - /******************************************************************************* * * * * * * *******************************************************************************/ -__host__ cudaError_t CUDARTAPI cudaMemcpy(void *dst, const void *src, size_t count, enum cudaMemcpyKind kind) -{ - return cudaMemcpyInternal(dst, src, count, kind); +__host__ cudaError_t CUDARTAPI cudaMemcpy(void *dst, const void *src, + size_t count, + enum cudaMemcpyKind kind) { + return cudaMemcpyInternal(dst, src, count, kind); } -__host__ cudaError_t CUDARTAPI cudaMemcpyToArray(struct cudaArray *dst, size_t wOffset, size_t hOffset, const void *src, size_t count, enum cudaMemcpyKind kind) -{ - return cudaMemcpyToArrayInternal(dst, wOffset, hOffset, src, count, kind); +__host__ cudaError_t CUDARTAPI cudaMemcpyToArray(struct cudaArray *dst, + size_t wOffset, size_t hOffset, + const void *src, size_t count, + enum cudaMemcpyKind kind) { + return cudaMemcpyToArrayInternal(dst, wOffset, hOffset, src, count, kind); } - -__host__ cudaError_t CUDARTAPI cudaMemcpyFromArray(void *dst, const struct cudaArray *src, size_t wOffset, size_t hOffset, size_t count, enum cudaMemcpyKind kind) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - cuda_not_implemented(__my_func__,__LINE__); - return g_last_cudaError = cudaErrorUnknown; +__host__ cudaError_t CUDARTAPI cudaMemcpyFromArray(void *dst, + const struct cudaArray *src, + size_t wOffset, + size_t hOffset, size_t count, + enum cudaMemcpyKind kind) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + cuda_not_implemented(__my_func__, __LINE__); + return g_last_cudaError = cudaErrorUnknown; } - -__host__ cudaError_t CUDARTAPI cudaMemcpyArrayToArray(struct cudaArray *dst, size_t wOffsetDst, size_t hOffsetDst, const struct cudaArray *src, size_t wOffsetSrc, size_t hOffsetSrc, size_t count, enum cudaMemcpyKind kind __dv(cudaMemcpyDeviceToDevice)) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - cuda_not_implemented(__my_func__,__LINE__); - return g_last_cudaError = cudaErrorUnknown; +__host__ cudaError_t CUDARTAPI cudaMemcpyArrayToArray( + struct cudaArray *dst, size_t wOffsetDst, size_t hOffsetDst, + const struct cudaArray *src, size_t wOffsetSrc, size_t hOffsetSrc, + size_t count, enum cudaMemcpyKind kind __dv(cudaMemcpyDeviceToDevice)) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + cuda_not_implemented(__my_func__, __LINE__); + return g_last_cudaError = cudaErrorUnknown; } - -__host__ cudaError_t CUDARTAPI cudaMemcpy2D(void *dst, size_t dpitch, const void *src, size_t spitch, size_t width, size_t height, enum cudaMemcpyKind kind) -{ - return cudaMemcpy2DInternal(dst, dpitch, src, spitch, width, height, kind); +__host__ cudaError_t CUDARTAPI cudaMemcpy2D(void *dst, size_t dpitch, + const void *src, size_t spitch, + size_t width, size_t height, + enum cudaMemcpyKind kind) { + return cudaMemcpy2DInternal(dst, dpitch, src, spitch, width, height, kind); } -__host__ cudaError_t CUDARTAPI cudaMemcpy2DToArray(struct cudaArray *dst, size_t wOffset, size_t hOffset, const void *src, size_t spitch, size_t width, size_t height, enum cudaMemcpyKind kind) -{ - return cudaMemcpy2DToArrayInternal(dst, wOffset, hOffset, src, spitch, width, height, kind); +__host__ cudaError_t CUDARTAPI cudaMemcpy2DToArray( + struct cudaArray *dst, size_t wOffset, size_t hOffset, const void *src, + size_t spitch, size_t width, size_t height, enum cudaMemcpyKind kind) { + return cudaMemcpy2DToArrayInternal(dst, wOffset, hOffset, src, spitch, width, + height, kind); } -__host__ cudaError_t CUDARTAPI cudaMemcpy2DFromArray(void *dst, size_t dpitch, const struct cudaArray *src, size_t wOffset, size_t hOffset, size_t width, size_t height, enum cudaMemcpyKind kind) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - cuda_not_implemented(__my_func__,__LINE__); - return g_last_cudaError = cudaErrorUnknown; +__host__ cudaError_t CUDARTAPI cudaMemcpy2DFromArray( + void *dst, size_t dpitch, const struct cudaArray *src, size_t wOffset, + size_t hOffset, size_t width, size_t height, enum cudaMemcpyKind kind) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + cuda_not_implemented(__my_func__, __LINE__); + return g_last_cudaError = cudaErrorUnknown; } -__host__ cudaError_t CUDARTAPI cudaMemcpy2DArrayToArray(struct cudaArray *dst, size_t wOffsetDst, size_t hOffsetDst, const struct cudaArray *src, size_t wOffsetSrc, size_t hOffsetSrc, size_t width, size_t height, enum cudaMemcpyKind kind __dv(cudaMemcpyDeviceToDevice)) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - cuda_not_implemented(__my_func__,__LINE__); - return g_last_cudaError = cudaErrorUnknown; +__host__ cudaError_t CUDARTAPI cudaMemcpy2DArrayToArray( + struct cudaArray *dst, size_t wOffsetDst, size_t hOffsetDst, + const struct cudaArray *src, size_t wOffsetSrc, size_t hOffsetSrc, + size_t width, size_t height, + enum cudaMemcpyKind kind __dv(cudaMemcpyDeviceToDevice)) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + cuda_not_implemented(__my_func__, __LINE__); + return g_last_cudaError = cudaErrorUnknown; } -__host__ cudaError_t CUDARTAPI cudaMemcpyToSymbol(const char *symbol, const void *src, size_t count, size_t offset __dv(0), enum cudaMemcpyKind kind __dv(cudaMemcpyHostToDevice)) -{ - return cudaMemcpyToSymbolInternal(symbol, src, count, offset, kind); +__host__ cudaError_t CUDARTAPI cudaMemcpyToSymbol( + const char *symbol, const void *src, size_t count, size_t offset __dv(0), + enum cudaMemcpyKind kind __dv(cudaMemcpyHostToDevice)) { + return cudaMemcpyToSymbolInternal(symbol, src, count, offset, kind); } - -__host__ cudaError_t CUDARTAPI cudaMemcpyFromSymbol(void *dst, const char *symbol, size_t count, size_t offset __dv(0), enum cudaMemcpyKind kind __dv(cudaMemcpyDeviceToHost)) -{ - return cudaMemcpyFromSymbolInternal(dst, symbol, count, offset, kind); +__host__ cudaError_t CUDARTAPI cudaMemcpyFromSymbol( + void *dst, const char *symbol, size_t count, size_t offset __dv(0), + enum cudaMemcpyKind kind __dv(cudaMemcpyDeviceToHost)) { + return cudaMemcpyFromSymbolInternal(dst, symbol, count, offset, kind); } -__host__ cudaError_t CUDARTAPI cudaMemGetInfo (size_t *free, size_t *total){ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - //placeholder; should interact with cudaMalloc and cudaFree? - *free = 10000000000; - *total = 10000000000; +__host__ cudaError_t CUDARTAPI cudaMemGetInfo(size_t *free, size_t *total) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + // placeholder; should interact with cudaMalloc and cudaFree? + *free = 10000000000; + *total = 10000000000; - return g_last_cudaError = cudaSuccess; + return g_last_cudaError = cudaSuccess; } /******************************************************************************* @@ -2289,391 +2446,377 @@ __host__ cudaError_t CUDARTAPI cudaMemGetInfo (size_t *free, size_t *total){ * * *******************************************************************************/ -__host__ cudaError_t CUDARTAPI cudaMemcpyAsync(void *dst, const void *src, size_t count, enum cudaMemcpyKind kind, cudaStream_t stream) -{ - return cudaMemcpyAsyncInternal(dst, src, count, kind, stream); -} - - -__host__ cudaError_t CUDARTAPI cudaMemcpyToArrayAsync(struct cudaArray *dst, size_t wOffset, size_t hOffset, const void *src, size_t count, enum cudaMemcpyKind kind, cudaStream_t stream) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - cuda_not_implemented(__my_func__,__LINE__); - return g_last_cudaError = cudaErrorUnknown; -} - - -__host__ cudaError_t CUDARTAPI cudaMemcpyFromArrayAsync(void *dst, const struct cudaArray *src, size_t wOffset, size_t hOffset, size_t count, enum cudaMemcpyKind kind, cudaStream_t stream) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - cuda_not_implemented(__my_func__,__LINE__); - return g_last_cudaError = cudaErrorUnknown; -} - - -__host__ cudaError_t CUDARTAPI cudaMemcpy2DAsync(void *dst, size_t dpitch, const void *src, size_t spitch, size_t width, size_t height, enum cudaMemcpyKind kind, cudaStream_t stream) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - cuda_not_implemented(__my_func__,__LINE__); - return g_last_cudaError = cudaErrorUnknown; -} - - -__host__ cudaError_t CUDARTAPI cudaMemcpy2DToArrayAsync(struct cudaArray *dst, size_t wOffset, size_t hOffset, const void *src, size_t spitch, size_t width, size_t height, enum cudaMemcpyKind kind, cudaStream_t stream) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - cuda_not_implemented(__my_func__,__LINE__); - return g_last_cudaError = cudaErrorUnknown; -} - - -__host__ cudaError_t CUDARTAPI cudaMemcpy2DFromArrayAsync(void *dst, size_t dpitch, const struct cudaArray *src, size_t wOffset, size_t hOffset, size_t width, size_t height, enum cudaMemcpyKind kind, cudaStream_t stream) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - cuda_not_implemented(__my_func__,__LINE__); - return g_last_cudaError = cudaErrorUnknown; +__host__ cudaError_t CUDARTAPI cudaMemcpyAsync(void *dst, const void *src, + size_t count, + enum cudaMemcpyKind kind, + cudaStream_t stream) { + return cudaMemcpyAsyncInternal(dst, src, count, kind, stream); +} + +__host__ cudaError_t CUDARTAPI cudaMemcpyToArrayAsync( + struct cudaArray *dst, size_t wOffset, size_t hOffset, const void *src, + size_t count, enum cudaMemcpyKind kind, cudaStream_t stream) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + cuda_not_implemented(__my_func__, __LINE__); + return g_last_cudaError = cudaErrorUnknown; +} + +__host__ cudaError_t CUDARTAPI cudaMemcpyFromArrayAsync( + void *dst, const struct cudaArray *src, size_t wOffset, size_t hOffset, + size_t count, enum cudaMemcpyKind kind, cudaStream_t stream) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + cuda_not_implemented(__my_func__, __LINE__); + return g_last_cudaError = cudaErrorUnknown; +} + +__host__ cudaError_t CUDARTAPI cudaMemcpy2DAsync(void *dst, size_t dpitch, + const void *src, size_t spitch, + size_t width, size_t height, + enum cudaMemcpyKind kind, + cudaStream_t stream) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + cuda_not_implemented(__my_func__, __LINE__); + return g_last_cudaError = cudaErrorUnknown; +} + +__host__ cudaError_t CUDARTAPI cudaMemcpy2DToArrayAsync( + struct cudaArray *dst, size_t wOffset, size_t hOffset, const void *src, + size_t spitch, size_t width, size_t height, enum cudaMemcpyKind kind, + cudaStream_t stream) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + cuda_not_implemented(__my_func__, __LINE__); + return g_last_cudaError = cudaErrorUnknown; +} + +__host__ cudaError_t CUDARTAPI cudaMemcpy2DFromArrayAsync( + void *dst, size_t dpitch, const struct cudaArray *src, size_t wOffset, + size_t hOffset, size_t width, size_t height, enum cudaMemcpyKind kind, + cudaStream_t stream) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + cuda_not_implemented(__my_func__, __LINE__); + return g_last_cudaError = cudaErrorUnknown; } #if (CUDART_VERSION >= 8000) -cudaError_t CUDARTAPI cudaOccupancyMaxActiveBlocksPerMultiprocessorWithFlags(int* numBlocks, const char *hostFunc, int blockSize, size_t dynamicSMemSize, unsigned int flags) -{ - return cudaOccupancyMaxActiveBlocksPerMultiprocessorWithFlagsInternal(numBlocks, hostFunc, blockSize, dynamicSMemSize, flags); +cudaError_t CUDARTAPI cudaOccupancyMaxActiveBlocksPerMultiprocessorWithFlags( + int *numBlocks, const char *hostFunc, int blockSize, size_t dynamicSMemSize, + unsigned int flags) { + return cudaOccupancyMaxActiveBlocksPerMultiprocessorWithFlagsInternal( + numBlocks, hostFunc, blockSize, dynamicSMemSize, flags); } #endif - - /******************************************************************************* * * * * * * *******************************************************************************/ -__host__ cudaError_t CUDARTAPI cudaMemset(void *mem, int c, size_t count) -{ - return cudaMemsetInternal(mem, c, count); +__host__ cudaError_t CUDARTAPI cudaMemset(void *mem, int c, size_t count) { + return cudaMemsetInternal(mem, c, count); } -//memset operation is done but i think its not async? -__host__ cudaError_t CUDARTAPI cudaMemsetAsync(void *mem, int c, size_t count, cudaStream_t stream=0) -{ - return cudaMemsetAsyncInternal(mem, c, count, stream=0); +// memset operation is done but i think its not async? +__host__ cudaError_t CUDARTAPI cudaMemsetAsync(void *mem, int c, size_t count, + cudaStream_t stream = 0) { + return cudaMemsetAsyncInternal(mem, c, count, stream = 0); } -__host__ cudaError_t CUDARTAPI cudaMemset2D(void *mem, size_t pitch, int c, size_t width, size_t height) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - cuda_not_implemented(__my_func__,__LINE__); - return g_last_cudaError = cudaErrorUnknown; +__host__ cudaError_t CUDARTAPI cudaMemset2D(void *mem, size_t pitch, int c, + size_t width, size_t height) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + cuda_not_implemented(__my_func__, __LINE__); + return g_last_cudaError = cudaErrorUnknown; } - - /******************************************************************************* * * * * * * *******************************************************************************/ -__host__ cudaError_t CUDARTAPI cudaGetSymbolAddress(void **devPtr, const char *symbol) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - cuda_not_implemented(__my_func__,__LINE__); - return g_last_cudaError = cudaErrorUnknown; +__host__ cudaError_t CUDARTAPI cudaGetSymbolAddress(void **devPtr, + const char *symbol) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + cuda_not_implemented(__my_func__, __LINE__); + return g_last_cudaError = cudaErrorUnknown; } - -__host__ cudaError_t CUDARTAPI cudaGetSymbolSize(size_t *size, const char *symbol) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - cuda_not_implemented(__my_func__,__LINE__); - return g_last_cudaError = cudaErrorUnknown; +__host__ cudaError_t CUDARTAPI cudaGetSymbolSize(size_t *size, + const char *symbol) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + cuda_not_implemented(__my_func__, __LINE__); + return g_last_cudaError = cudaErrorUnknown; } - - /******************************************************************************* * * * * * * *******************************************************************************/ -__host__ cudaError_t CUDARTAPI cudaGetDeviceCount(int *count) -{ - return cudaGetDeviceCountInternal(count); +__host__ cudaError_t CUDARTAPI cudaGetDeviceCount(int *count) { + return cudaGetDeviceCountInternal(count); } -__host__ cudaError_t CUDARTAPI cudaGetDeviceProperties(struct cudaDeviceProp *prop, int device) -{ - return cudaGetDevicePropertiesInternal(prop, device); +__host__ cudaError_t CUDARTAPI +cudaGetDeviceProperties(struct cudaDeviceProp *prop, int device) { + return cudaGetDevicePropertiesInternal(prop, device); } #if (CUDART_VERSION > 5000) -__host__ cudaError_t CUDARTAPI cudaDeviceGetAttribute(int *value, enum cudaDeviceAttr attr, int device) -{ - return cudaDeviceGetAttributeInternal(value, attr, device); +__host__ cudaError_t CUDARTAPI cudaDeviceGetAttribute(int *value, + enum cudaDeviceAttr attr, + int device) { + return cudaDeviceGetAttributeInternal(value, attr, device); } #endif -__host__ cudaError_t CUDARTAPI cudaChooseDevice(int *device, const struct cudaDeviceProp *prop) -{ - return cudaChooseDeviceInternal(device, prop); +__host__ cudaError_t CUDARTAPI +cudaChooseDevice(int *device, const struct cudaDeviceProp *prop) { + return cudaChooseDeviceInternal(device, prop); } -__host__ cudaError_t CUDARTAPI cudaSetDevice(int device) -{ - return cudaSetDeviceInternal(device); +__host__ cudaError_t CUDARTAPI cudaSetDevice(int device) { + return cudaSetDeviceInternal(device); } -__host__ cudaError_t CUDARTAPI cudaGetDevice(int *device) -{ - return cudaGetDeviceInternal(device); +__host__ cudaError_t CUDARTAPI cudaGetDevice(int *device) { + return cudaGetDeviceInternal(device); } -__host__ cudaError_t CUDARTAPI cudaDeviceGetLimit( size_t* pValue, cudaLimit limit ) -{ - return cudaDeviceGetLimitInternal( pValue, limit ); +__host__ cudaError_t CUDARTAPI cudaDeviceGetLimit(size_t *pValue, + cudaLimit limit) { + return cudaDeviceGetLimitInternal(pValue, limit); } -__host__ cudaError_t CUDARTAPI cudaStreamGetPriority ( cudaStream_t hStream, int* priority ) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - cuda_not_implemented(__my_func__,__LINE__); - return g_last_cudaError = cudaSuccess; - +__host__ cudaError_t CUDARTAPI cudaStreamGetPriority(cudaStream_t hStream, + int *priority) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + cuda_not_implemented(__my_func__, __LINE__); + return g_last_cudaError = cudaSuccess; } -__host__ cudaError_t CUDARTAPI cudaDeviceGetPCIBusId ( - char *pciBusId, - int len, - int device -) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - cuda_not_implemented(__my_func__,__LINE__); - return g_last_cudaError = cudaErrorUnknown; +__host__ cudaError_t CUDARTAPI cudaDeviceGetPCIBusId(char *pciBusId, int len, + int device) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + cuda_not_implemented(__my_func__, __LINE__); + return g_last_cudaError = cudaErrorUnknown; } -__host__ cudaError_t CUDARTAPI cudaIpcGetMemHandle( cudaIpcMemHandle_t* handle, void* devPtr ) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - cuda_not_implemented(__my_func__,__LINE__); - return g_last_cudaError = cudaErrorUnknown; +__host__ cudaError_t CUDARTAPI cudaIpcGetMemHandle(cudaIpcMemHandle_t *handle, + void *devPtr) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + cuda_not_implemented(__my_func__, __LINE__); + return g_last_cudaError = cudaErrorUnknown; } -__host__ cudaError_t cudaIpcOpenMemHandle( - void **devPtr, - cudaIpcMemHandle_t handle, - unsigned int flags -) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - cuda_not_implemented(__my_func__,__LINE__); - return g_last_cudaError = cudaErrorUnknown; +__host__ cudaError_t cudaIpcOpenMemHandle(void **devPtr, + cudaIpcMemHandle_t handle, + unsigned int flags) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + cuda_not_implemented(__my_func__, __LINE__); + return g_last_cudaError = cudaErrorUnknown; } -__host__ cudaError_t CUDARTAPI cudaDestroyTextureObject(cudaTextureObject_t texObject) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - cuda_not_implemented(__my_func__,__LINE__); - return g_last_cudaError = cudaErrorUnknown; +__host__ cudaError_t CUDARTAPI +cudaDestroyTextureObject(cudaTextureObject_t texObject) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + cuda_not_implemented(__my_func__, __LINE__); + return g_last_cudaError = cudaErrorUnknown; } - /******************************************************************************* * * * * * * *******************************************************************************/ -__host__ cudaError_t CUDARTAPI cudaBindTexture(size_t *offset, - const struct textureReference *texref, - const void *devPtr, - const struct cudaChannelFormatDesc *desc, - size_t size __dv(UINT_MAX)) -{ - return cudaBindTextureInternal(offset, texref, devPtr, desc, size __dv(UINT_MAX)); +__host__ cudaError_t CUDARTAPI cudaBindTexture( + size_t *offset, const struct textureReference *texref, const void *devPtr, + const struct cudaChannelFormatDesc *desc, size_t size __dv(UINT_MAX)) { + return cudaBindTextureInternal(offset, texref, devPtr, desc, + size __dv(UINT_MAX)); } +__host__ cudaError_t CUDARTAPI cudaBindTextureToArray( + const struct textureReference *texref, const struct cudaArray *array, + const struct cudaChannelFormatDesc *desc) { + return cudaBindTextureToArrayInternal(texref, array, desc); +} -__host__ cudaError_t CUDARTAPI cudaBindTextureToArray(const struct textureReference *texref, const struct cudaArray *array, const struct cudaChannelFormatDesc *desc) -{ - return cudaBindTextureToArrayInternal(texref, array, desc); +__host__ cudaError_t CUDARTAPI +cudaUnbindTexture(const struct textureReference *texref) { + return cudaUnbindTextureInternal(texref); } -__host__ cudaError_t CUDARTAPI cudaUnbindTexture(const struct textureReference *texref) -{ - return cudaUnbindTextureInternal(texref); +__host__ cudaError_t CUDARTAPI cudaGetTextureAlignmentOffset( + size_t *offset, const struct textureReference *texref) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + cuda_not_implemented(__my_func__, __LINE__); + return g_last_cudaError = cudaErrorUnknown; } -__host__ cudaError_t CUDARTAPI cudaGetTextureAlignmentOffset(size_t *offset, const struct textureReference *texref) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - cuda_not_implemented(__my_func__,__LINE__); - return g_last_cudaError = cudaErrorUnknown; +__host__ cudaError_t CUDARTAPI cudaGetTextureReference( + const struct textureReference **texref, const char *symbol) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + cuda_not_implemented(__my_func__, __LINE__); + return g_last_cudaError = cudaErrorUnknown; } -__host__ cudaError_t CUDARTAPI cudaGetTextureReference(const struct textureReference **texref, const char *symbol) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - cuda_not_implemented(__my_func__,__LINE__); - return g_last_cudaError = cudaErrorUnknown; +__host__ cudaError_t CUDARTAPI cudaGetChannelDesc( + struct cudaChannelFormatDesc *desc, const struct cudaArray *array) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + *desc = array->desc; + return g_last_cudaError = cudaSuccess; } -__host__ cudaError_t CUDARTAPI cudaGetChannelDesc(struct cudaChannelFormatDesc *desc, const struct cudaArray *array) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - *desc = array->desc; - return g_last_cudaError = cudaSuccess; +__host__ struct cudaChannelFormatDesc CUDARTAPI cudaCreateChannelDesc( + int x, int y, int z, int w, enum cudaChannelFormatKind f) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + struct cudaChannelFormatDesc dummy; + dummy.x = x; + dummy.y = y; + dummy.z = z; + dummy.w = w; + dummy.f = f; + return dummy; } +__host__ cudaError_t CUDARTAPI cudaGetLastError(void) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + return g_last_cudaError; +} -__host__ struct cudaChannelFormatDesc CUDARTAPI cudaCreateChannelDesc(int x, int y, int z, int w, enum cudaChannelFormatKind f) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - struct cudaChannelFormatDesc dummy; - dummy.x = x; - dummy.y = y; - dummy.z = z; - dummy.w = w; - dummy.f = f; - return dummy; +__host__ const char *cudaGetErrorName(cudaError_t error) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + cuda_not_implemented(__my_func__, __LINE__); + return NULL; } -__host__ cudaError_t CUDARTAPI cudaGetLastError(void) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - return g_last_cudaError; +__host__ const char *CUDARTAPI cudaGetErrorString(cudaError_t error) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + if (g_last_cudaError == cudaSuccess) return "no error"; + char buf[1024]; + snprintf(buf, 1024, "<>", + g_last_cudaError); + return strdup(buf); } -__host__ const char *cudaGetErrorName(cudaError_t error) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - cuda_not_implemented(__my_func__,__LINE__); - return NULL; -} - -__host__ const char* CUDARTAPI cudaGetErrorString(cudaError_t error) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - if( g_last_cudaError == cudaSuccess ) - return "no error"; - char buf[1024]; - snprintf(buf,1024,"<>", g_last_cudaError); - return strdup(buf); -} - -__host__ cudaError_t CUDARTAPI cudaSetupArgument(const void *arg, size_t size, size_t offset) -{ - return cudaSetupArgumentInternal(arg, size, offset); +__host__ cudaError_t CUDARTAPI cudaSetupArgument(const void *arg, size_t size, + size_t offset) { + return cudaSetupArgumentInternal(arg, size, offset); } - -__host__ cudaError_t CUDARTAPI cudaLaunch( const char *hostFun ) -{ - return cudaLaunchInternal( hostFun ); +__host__ cudaError_t CUDARTAPI cudaLaunch(const char *hostFun) { + return cudaLaunchInternal(hostFun); } -__host__ cudaError_t CUDARTAPI cudaLaunchKernel( const char* hostFun, dim3 gridDim, dim3 blockDim, const void** args, size_t sharedMem, cudaStream_t stream ) -{ - return cudaLaunchKernelInternal(hostFun, gridDim, blockDim, args, sharedMem, stream); +__host__ cudaError_t CUDARTAPI cudaLaunchKernel(const char *hostFun, + dim3 gridDim, dim3 blockDim, + const void **args, + size_t sharedMem, + cudaStream_t stream) { + return cudaLaunchKernelInternal(hostFun, gridDim, blockDim, args, sharedMem, + stream); } - /******************************************************************************* * * * * * * *******************************************************************************/ -__host__ cudaError_t CUDARTAPI cudaStreamCreate(cudaStream_t *stream) -{ - return cudaStreamCreateInternal(stream); +__host__ cudaError_t CUDARTAPI cudaStreamCreate(cudaStream_t *stream) { + return cudaStreamCreateInternal(stream); } -//TODO: introduce priorities -__host__ cudaError_t CUDARTAPI cudaStreamCreateWithPriority(cudaStream_t *stream, unsigned int flags, int priority) { - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - return cudaStreamCreate(stream); +// TODO: introduce priorities +__host__ cudaError_t CUDARTAPI cudaStreamCreateWithPriority( + cudaStream_t *stream, unsigned int flags, int priority) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + return cudaStreamCreate(stream); } -__host__ cudaError_t CUDARTAPI cudaDeviceGetStreamPriorityRange(int* leastPriority, int* greatestPriority) { - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - return cudaSuccess; +__host__ cudaError_t CUDARTAPI +cudaDeviceGetStreamPriorityRange(int *leastPriority, int *greatestPriority) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + return cudaSuccess; } -__host__ __device__ cudaError_t CUDARTAPI cudaStreamCreateWithFlags(cudaStream_t *stream, unsigned int flags) { - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - return cudaStreamCreate(stream); +__host__ __device__ cudaError_t CUDARTAPI +cudaStreamCreateWithFlags(cudaStream_t *stream, unsigned int flags) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + return cudaStreamCreate(stream); } -__host__ cudaError_t CUDARTAPI cudaStreamDestroy(cudaStream_t stream) -{ - return cudaStreamDestroyInternal(stream); +__host__ cudaError_t CUDARTAPI cudaStreamDestroy(cudaStream_t stream) { + return cudaStreamDestroyInternal(stream); } -__host__ cudaError_t CUDARTAPI cudaStreamSynchronize(cudaStream_t stream) -{ - return cudaStreamSynchronizeInternal(stream); +__host__ cudaError_t CUDARTAPI cudaStreamSynchronize(cudaStream_t stream) { + return cudaStreamSynchronizeInternal(stream); } -__host__ cudaError_t CUDARTAPI cudaStreamQuery(cudaStream_t stream) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } +__host__ cudaError_t CUDARTAPI cudaStreamQuery(cudaStream_t stream) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } #if (CUDART_VERSION >= 3000) - if( stream == NULL ) - return g_last_cudaError = cudaErrorInvalidResourceHandle; - return g_last_cudaError = stream->empty()?cudaSuccess:cudaErrorNotReady; + if (stream == NULL) return g_last_cudaError = cudaErrorInvalidResourceHandle; + return g_last_cudaError = stream->empty() ? cudaSuccess : cudaErrorNotReady; #else - printf("GPGPU-Sim PTX: WARNING: Asynchronous kernel execution not supported (%s)\n", __my_func__); - return g_last_cudaError = cudaSuccess; // it is always success because all cuda calls are synchronous + printf( + "GPGPU-Sim PTX: WARNING: Asynchronous kernel execution not supported " + "(%s)\n", + __my_func__); + return g_last_cudaError = cudaSuccess; // it is always success because all + // cuda calls are synchronous #endif } @@ -2683,119 +2826,108 @@ __host__ cudaError_t CUDARTAPI cudaStreamQuery(cudaStream_t stream) * * *******************************************************************************/ -__host__ cudaError_t CUDARTAPI cudaEventCreate(cudaEvent_t *event) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - CUevent_st *e = new CUevent_st(false); - g_timer_events[e->get_uid()] = e; +__host__ cudaError_t CUDARTAPI cudaEventCreate(cudaEvent_t *event) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + CUevent_st *e = new CUevent_st(false); + g_timer_events[e->get_uid()] = e; #if CUDART_VERSION >= 3000 - *event = e; + *event = e; #else - *event = e->get_uid(); + *event = e->get_uid(); #endif - return g_last_cudaError = cudaSuccess; -} - -__host__ cudaError_t CUDARTAPI cudaEventRecord(cudaEvent_t event, cudaStream_t stream) -{ - return cudaEventRecordInternal(event, stream); -} - -__host__ cudaError_t CUDARTAPI cudaStreamWaitEvent(cudaStream_t stream, cudaEvent_t event, unsigned int flags) -{ - return cudaStreamWaitEventInternal(stream, event, flags); -} - -__host__ cudaError_t CUDARTAPI cudaEventQuery(cudaEvent_t event) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - CUevent_st *e = get_event(event); - if( e == NULL ) { - return g_last_cudaError = cudaErrorInvalidValue; - } else if( e->done() ) { - return g_last_cudaError = cudaSuccess; - } else { - return g_last_cudaError = cudaErrorNotReady; - } + return g_last_cudaError = cudaSuccess; } -__host__ cudaError_t CUDARTAPI cudaEventSynchronize(cudaEvent_t event) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("GPGPU-Sim API: cudaEventSynchronize ** waiting for event\n"); - fflush(stdout); - CUevent_st *e = (CUevent_st*) event; - while( !e->done() ) - ; - printf("GPGPU-Sim API: cudaEventSynchronize ** event detected\n"); - fflush(stdout); - return g_last_cudaError = cudaSuccess; +__host__ cudaError_t CUDARTAPI cudaEventRecord(cudaEvent_t event, + cudaStream_t stream) { + return cudaEventRecordInternal(event, stream); } -__host__ cudaError_t CUDARTAPI cudaEventDestroy(cudaEvent_t event) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - CUevent_st *e = get_event(event); - unsigned event_uid = e->get_uid(); - event_tracker_t::iterator pe = g_timer_events.find(event_uid); - if( pe == g_timer_events.end() ) - return g_last_cudaError = cudaErrorInvalidValue; - g_timer_events.erase(pe); - return g_last_cudaError = cudaSuccess; +__host__ cudaError_t CUDARTAPI cudaStreamWaitEvent(cudaStream_t stream, + cudaEvent_t event, + unsigned int flags) { + return cudaStreamWaitEventInternal(stream, event, flags); } - -__host__ cudaError_t CUDARTAPI cudaEventElapsedTime(float *ms, cudaEvent_t start, cudaEvent_t end) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - time_t elapsed_time; - CUevent_st *s = get_event(start); - CUevent_st *e = get_event(end); - if( s==NULL || e==NULL ) - return g_last_cudaError = cudaErrorUnknown; - elapsed_time = e->clock() - s->clock(); - *ms = 1000*elapsed_time; - return g_last_cudaError = cudaSuccess; +__host__ cudaError_t CUDARTAPI cudaEventQuery(cudaEvent_t event) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + CUevent_st *e = get_event(event); + if (e == NULL) { + return g_last_cudaError = cudaErrorInvalidValue; + } else if (e->done()) { + return g_last_cudaError = cudaSuccess; + } else { + return g_last_cudaError = cudaErrorNotReady; + } +} + +__host__ cudaError_t CUDARTAPI cudaEventSynchronize(cudaEvent_t event) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("GPGPU-Sim API: cudaEventSynchronize ** waiting for event\n"); + fflush(stdout); + CUevent_st *e = (CUevent_st *)event; + while (!e->done()) + ; + printf("GPGPU-Sim API: cudaEventSynchronize ** event detected\n"); + fflush(stdout); + return g_last_cudaError = cudaSuccess; +} + +__host__ cudaError_t CUDARTAPI cudaEventDestroy(cudaEvent_t event) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + CUevent_st *e = get_event(event); + unsigned event_uid = e->get_uid(); + event_tracker_t::iterator pe = g_timer_events.find(event_uid); + if (pe == g_timer_events.end()) + return g_last_cudaError = cudaErrorInvalidValue; + g_timer_events.erase(pe); + return g_last_cudaError = cudaSuccess; +} + +__host__ cudaError_t CUDARTAPI cudaEventElapsedTime(float *ms, + cudaEvent_t start, + cudaEvent_t end) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + time_t elapsed_time; + CUevent_st *s = get_event(start); + CUevent_st *e = get_event(end); + if (s == NULL || e == NULL) return g_last_cudaError = cudaErrorUnknown; + elapsed_time = e->clock() - s->clock(); + *ms = 1000 * elapsed_time; + return g_last_cudaError = cudaSuccess; } - - /******************************************************************************* * * * * * * *******************************************************************************/ -__host__ cudaError_t CUDARTAPI cudaThreadExit(void) -{ - return cudaThreadExitInternal(); +__host__ cudaError_t CUDARTAPI cudaThreadExit(void) { + return cudaThreadExitInternal(); } -__host__ cudaError_t CUDARTAPI cudaThreadSynchronize(void) -{ - return cudaThreadSynchronizeInternal(); +__host__ cudaError_t CUDARTAPI cudaThreadSynchronize(void) { + return cudaThreadSynchronizeInternal(); } -int CUDARTAPI __cudaSynchronizeThreads(void**, void*) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - return cudaThreadExit(); +int CUDARTAPI __cudaSynchronizeThreads(void **, void *) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + return cudaThreadExit(); } - - /******************************************************************************* * * * * @@ -2804,39 +2936,40 @@ int CUDARTAPI __cudaSynchronizeThreads(void**, void*) #if (CUDART_VERSION >= 3010) int dummy0() { - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } -return 0; } + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + return 0; +} int dummy1() { - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } -return 2 << 20; } + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + return 2 << 20; +} typedef int (*ExportedFunction)(); static ExportedFunction exportTable[3] = {&dummy0, &dummy0, &dummy0}; -__host__ cudaError_t CUDARTAPI cudaGetExportTable(const void **ppExportTable, const cudaUUID_t *pExportTableId) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("cudaGetExportTable: UUID = "); - for (int s = 0; s < 16; s++) { - printf("%#2x ", (unsigned char) (pExportTableId->bytes[s])); - } - *ppExportTable = &exportTable; +__host__ cudaError_t CUDARTAPI cudaGetExportTable( + const void **ppExportTable, const cudaUUID_t *pExportTableId) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("cudaGetExportTable: UUID = "); + for (int s = 0; s < 16; s++) { + printf("%#2x ", (unsigned char)(pExportTableId->bytes[s])); + } + *ppExportTable = &exportTable; - printf("\n"); - return g_last_cudaError = cudaSuccess; + printf("\n"); + return g_last_cudaError = cudaSuccess; } #endif - /******************************************************************************* * * * * @@ -2845,875 +2978,862 @@ __host__ cudaError_t CUDARTAPI cudaGetExportTable(const void **ppExportTable, co //#include "../../cuobjdump_to_ptxplus/cuobjdump_parser.h" -//extracts all ptx files from binary and dumps into prog_name.unique_no.sm_<>.ptx files -void cuda_runtime_api::extract_ptx_files_using_cuobjdump(CUctx_st *context){ - char command[1000]; - char *pytorch_bin = getenv("PYTORCH_BIN"); - std::string app_binary = get_app_binary(); - +// extracts all ptx files from binary and dumps into +// prog_name.unique_no.sm_<>.ptx files +void cuda_runtime_api::extract_ptx_files_using_cuobjdump(CUctx_st *context) { + char command[1000]; + char *pytorch_bin = getenv("PYTORCH_BIN"); + std::string app_binary = get_app_binary(); + + char ptx_list_file_name[1024]; + snprintf(ptx_list_file_name, 1024, "_cuobjdump_list_ptx_XXXXXX"); + int fd2 = mkstemp(ptx_list_file_name); + close(fd2); + + if (pytorch_bin != NULL && strlen(pytorch_bin) != 0) { + app_binary = std::string(pytorch_bin); + } + + // only want file names + snprintf(command, 1000, + "$CUDA_INSTALL_PATH/bin/cuobjdump -lptx %s | cut -d \":\" -f 2 | " + "awk '{$1=$1}1' > %s", + app_binary.c_str(), ptx_list_file_name); + if (system(command) != 0) { + printf("WARNING: Failed to execute cuobjdump to get list of ptx files \n"); + exit(0); + } + if (!gpgpu_ctx->device_runtime->g_cdp_enabled) { + // based on the list above, dump ptx files individually. Format of dumped + // ptx file is prog_name.unique_no.sm_<>.ptx - char ptx_list_file_name[1024]; - snprintf(ptx_list_file_name,1024,"_cuobjdump_list_ptx_XXXXXX"); - int fd2=mkstemp(ptx_list_file_name); - close(fd2); - - if (pytorch_bin!=NULL && strlen(pytorch_bin)!=0){ - app_binary = std::string(pytorch_bin); - } - - //only want file names - snprintf(command,1000,"$CUDA_INSTALL_PATH/bin/cuobjdump -lptx %s | cut -d \":\" -f 2 | awk '{$1=$1}1' > %s", app_binary.c_str(), ptx_list_file_name); - if( system(command) != 0 ) { - printf("WARNING: Failed to execute cuobjdump to get list of ptx files \n"); + std::ifstream infile(ptx_list_file_name); + std::string line; + while (std::getline(infile, line)) { + // int pos = line.find(std::string(get_app_binary_name(app_binary))); + const char *ptx_file = line.c_str(); + printf("Extracting specific PTX file named %s \n", ptx_file); + snprintf(command, 1000, "$CUDA_INSTALL_PATH/bin/cuobjdump -xptx %s %s", + ptx_file, app_binary.c_str()); + if (system(command) != 0) { + printf("ERROR: command: %s failed \n", command); exit(0); - } - if(!gpgpu_ctx->device_runtime->g_cdp_enabled) { - //based on the list above, dump ptx files individually. Format of dumped ptx file is prog_name.unique_no.sm_<>.ptx - - std::ifstream infile(ptx_list_file_name); - std::string line; - while (std::getline(infile, line)) - { - //int pos = line.find(std::string(get_app_binary_name(app_binary))); - const char *ptx_file = line.c_str(); - printf("Extracting specific PTX file named %s \n",ptx_file); - snprintf(command,1000,"$CUDA_INSTALL_PATH/bin/cuobjdump -xptx %s %s", ptx_file, app_binary.c_str()); - if (system(command)!=0) { - printf("ERROR: command: %s failed \n",command); - exit(0); - } - context->no_of_ptx++; - } + } + context->no_of_ptx++; } + } - if(!context->no_of_ptx){ - printf("WARNING: Number of ptx in the executable file are 0. One of the reasons might be\n"); - printf("\t1. CDP is enabled\n"); - printf("\t2. When using PyTorch, PYTORCH_BIN is not set correctly\n"); - } + if (!context->no_of_ptx) { + printf( + "WARNING: Number of ptx in the executable file are 0. One of the " + "reasons might be\n"); + printf("\t1. CDP is enabled\n"); + printf("\t2. When using PyTorch, PYTORCH_BIN is not set correctly\n"); + } - std::ifstream infile(ptx_list_file_name); - std::string line; - while (std::getline(infile, line)) - { - //int pos = line.find(std::string(get_app_binary_name(app_binary))); - int pos1 = line.find("sm_"); - int pos2 = line.find_last_of("."); - if (pos1==std::string::npos&&pos2==std::string::npos){ - printf("ERROR: PTX list is not in correct format"); - exit(0); - } - std::string vstr = line.substr(pos1+3,pos2-pos1-3); - int version = atoi(vstr.c_str()); - if (version_filename.find(version)==version_filename.end()){ - version_filename[version] = std::set(); - } - version_filename[version].insert(line); + std::ifstream infile(ptx_list_file_name); + std::string line; + while (std::getline(infile, line)) { + // int pos = line.find(std::string(get_app_binary_name(app_binary))); + int pos1 = line.find("sm_"); + int pos2 = line.find_last_of("."); + if (pos1 == std::string::npos && pos2 == std::string::npos) { + printf("ERROR: PTX list is not in correct format"); + exit(0); } - + std::string vstr = line.substr(pos1 + 3, pos2 - pos1 - 3); + int version = atoi(vstr.c_str()); + if (version_filename.find(version) == version_filename.end()) { + version_filename[version] = std::set(); + } + version_filename[version].insert(line); + } } - //! Call cuobjdump to extract everything (-elf -sass -ptx) /*! * This Function extract the whole PTX (for all the files) using cuobjdump - * to _cuobjdump_complete_output_XXXXXX then runs a parser to chop it up with each binary in - * its own file - * It is also responsible for extracting the libraries linked to the binary if the option is - * enabled + * to _cuobjdump_complete_output_XXXXXX then runs a parser to chop it up + *with each binary in its own file It is also responsible for extracting the + *libraries linked to the binary if the option is enabled * */ -void cuda_runtime_api::extract_code_using_cuobjdump(){ - CUctx_st *context = GPGPUSim_Context(gpgpu_ctx); - - //prevent the dumping by cuobjdump everytime we execute the code! - const char *override_cuobjdump = getenv("CUOBJDUMP_SIM_FILE"); - char command[1000]; - std::string app_binary = get_app_binary(); - //Running cuobjdump using dynamic link to current process - snprintf(command,1000,"md5sum %s ", app_binary.c_str()); - printf("Running md5sum using \"%s\"\n", command); - if(system(command)){ - std::cout << "Failed to execute: " << command << std::endl; +void cuda_runtime_api::extract_code_using_cuobjdump() { + CUctx_st *context = GPGPUSim_Context(gpgpu_ctx); + + // prevent the dumping by cuobjdump everytime we execute the code! + const char *override_cuobjdump = getenv("CUOBJDUMP_SIM_FILE"); + char command[1000]; + std::string app_binary = get_app_binary(); + // Running cuobjdump using dynamic link to current process + snprintf(command, 1000, "md5sum %s ", app_binary.c_str()); + printf("Running md5sum using \"%s\"\n", command); + if (system(command)) { + std::cout << "Failed to execute: " << command << std::endl; + exit(1); + } + // Running cuobjdump using dynamic link to current process + // Needs the option '-all' to extract PTX from CDP-enabled binary + + // dump ptx for all individial ptx files into sepearte files which is later + // used by ptxas. + int result = 0; +#if (CUDART_VERSION >= 6000) + extract_ptx_files_using_cuobjdump(context); + return; +#endif + // TODO: redundant to dump twice. how can it be prevented? + // dump only for specific arch + char fname[1024]; + if ((override_cuobjdump == NULL) || (strlen(override_cuobjdump) == 0)) { + snprintf(fname, 1024, "_cuobjdump_complete_output_XXXXXX"); + int fd = mkstemp(fname); + close(fd); + if (!gpgpu_ctx->device_runtime->g_cdp_enabled) + snprintf(command, 1000, + "$CUDA_INSTALL_PATH/bin/cuobjdump -ptx -elf -sass %s > %s", + app_binary.c_str(), fname); + else + snprintf(command, 1000, + "$CUDA_INSTALL_PATH/bin/cuobjdump -ptx -elf -sass -all %s > %s", + app_binary.c_str(), fname); + bool parse_output = true; + result = system(command); + if (result) { + if (context->get_device() + ->get_gpgpu() + ->get_config() + .experimental_lib_support() && + (result == 65280)) { + // Some CUDA application may exclusively use kernels provided by CUDA + // libraries (e.g. CUBLAS). Skipping cuobjdump extraction from the + // executable for this case. + // 65280 is the return code from cuobjdump denoting the specific error + // (tested on CUDA 4.0/4.1/4.2) + printf("WARNING: Failed to execute: %s\n", command); + printf(" Executable binary does not contain any GPU kernel.\n"); + parse_output = false; + } else { + printf("ERROR: Failed to execute: %s\n", command); exit(1); + } } - // Running cuobjdump using dynamic link to current process - // Needs the option '-all' to extract PTX from CDP-enabled binary - //dump ptx for all individial ptx files into sepearte files which is later used by ptxas. - int result=0; -#if (CUDART_VERSION >= 6000) - extract_ptx_files_using_cuobjdump(context); - return; -#endif - //TODO: redundant to dump twice. how can it be prevented? - //dump only for specific arch - char fname[1024]; - if ((override_cuobjdump == NULL) || (strlen(override_cuobjdump)==0)) { - snprintf(fname,1024,"_cuobjdump_complete_output_XXXXXX"); - int fd=mkstemp(fname); - close(fd); - if(!gpgpu_ctx->device_runtime->g_cdp_enabled) - snprintf(command,1000,"$CUDA_INSTALL_PATH/bin/cuobjdump -ptx -elf -sass %s > %s", app_binary.c_str(), fname); - else - snprintf(command,1000,"$CUDA_INSTALL_PATH/bin/cuobjdump -ptx -elf -sass -all %s > %s", app_binary.c_str(), fname); - bool parse_output = true; - result = system(command); - if(result) { - if (context->get_device()->get_gpgpu()->get_config().experimental_lib_support() && (result == 65280)) { - // Some CUDA application may exclusively use kernels provided by CUDA - // libraries (e.g. CUBLAS). Skipping cuobjdump extraction from the - // executable for this case. - // 65280 is the return code from cuobjdump denoting the specific error (tested on CUDA 4.0/4.1/4.2) - printf("WARNING: Failed to execute: %s\n", command); - printf(" Executable binary does not contain any GPU kernel.\n"); - parse_output = false; - } else { - printf("ERROR: Failed to execute: %s\n", command); - exit(1); - } - } + if (parse_output) { + printf("Parsing file %s\n", fname); + FILE *cuobjdump_in; + cuobjdump_in = fopen(fname, "r"); + + struct cuobjdump_parser parser; + parser.elfserial = 1; + parser.ptxserial = 1; + cuobjdump_lex_init(&(parser.scanner)); + cuobjdump_set_in(cuobjdump_in, (parser.scanner)); + cuobjdump_parse(parser.scanner, &parser, cuobjdumpSectionList); + cuobjdump_lex_destroy(parser.scanner); + fclose(cuobjdump_in); + printf("Done parsing!!!\n"); + } else { + printf("Parsing skipped for %s\n", fname); + } + + if (context->get_device() + ->get_gpgpu() + ->get_config() + .experimental_lib_support()) { + // Experimental library support + // Currently only for cufft + + std::stringstream cmd; + cmd << "ldd " << app_binary + << " | grep $CUDA_INSTALL_PATH | awk \'{print $3}\' > _tempfile_.txt"; + int result = system(cmd.str().c_str()); + if (result) { + std::cout << "Failed to execute: " << cmd.str() << std::endl; + exit(1); + } + std::ifstream libsf; + libsf.open("_tempfile_.txt"); + if (!libsf.is_open()) { + std::cout << "Failed to open: _tempfile_.txt" << std::endl; + exit(1); + } - if (parse_output) { - printf("Parsing file %s\n", fname); - FILE *cuobjdump_in; - cuobjdump_in = fopen(fname, "r"); - - struct cuobjdump_parser parser; - parser.elfserial = 1; - parser.ptxserial = 1; - cuobjdump_lex_init(&(parser.scanner)); - cuobjdump_set_in(cuobjdump_in, (parser.scanner)); - cuobjdump_parse(parser.scanner, &parser, cuobjdumpSectionList); - cuobjdump_lex_destroy(parser.scanner); - fclose(cuobjdump_in); - printf("Done parsing!!!\n"); - } else { - printf("Parsing skipped for %s\n", fname); + // Save the original section list + std::list tmpsl = cuobjdumpSectionList; + cuobjdumpSectionList.clear(); + + std::string line; + std::getline(libsf, line); + std::cout << "DOING: " << line << std::endl; + int cnt = 1; + while (libsf.good()) { + std::stringstream libcodfn; + libcodfn << "_cuobjdump_complete_lib_" << cnt << "_"; + cmd.str(""); // resetting + cmd << "$CUDA_INSTALL_PATH/bin/cuobjdump -ptx -elf -sass "; + cmd << line; + cmd << " > "; + cmd << libcodfn.str(); + std::cout << "Running cuobjdump on " << line << std::endl; + std::cout << "Using command: " << cmd.str() << std::endl; + result = system(cmd.str().c_str()); + if (result) { + printf("ERROR: Failed to execute: %s\n", command); + exit(1); } + std::cout << "Done" << std::endl; + + std::cout << "Trying to parse " << libcodfn.str() << std::endl; + FILE *cuobjdump_in; + cuobjdump_in = fopen(libcodfn.str().c_str(), "r"); + struct cuobjdump_parser parser; + parser.elfserial = 1; + parser.ptxserial = 1; + cuobjdump_lex_init(&(parser.scanner)); + cuobjdump_set_in(cuobjdump_in, (parser.scanner)); + cuobjdump_parse(parser.scanner, &parser, cuobjdumpSectionList); + cuobjdump_lex_destroy(parser.scanner); + fclose(cuobjdump_in); + std::getline(libsf, line); + } + libSectionList = cuobjdumpSectionList; - if (context->get_device()->get_gpgpu()->get_config().experimental_lib_support()){ - //Experimental library support - //Currently only for cufft - - std::stringstream cmd; - cmd << "ldd " << app_binary << " | grep $CUDA_INSTALL_PATH | awk \'{print $3}\' > _tempfile_.txt"; - int result = system(cmd.str().c_str()); - if(result){ - std::cout << "Failed to execute: " << cmd.str() << std::endl; - exit(1); - } - std::ifstream libsf; - libsf.open("_tempfile_.txt"); - if(!libsf.is_open()) { - std::cout << "Failed to open: _tempfile_.txt" << std::endl; - exit(1); - } - - //Save the original section list - std::list tmpsl = cuobjdumpSectionList; - cuobjdumpSectionList.clear(); - - std::string line; - std::getline(libsf, line); - std::cout << "DOING: " << line << std::endl; - int cnt=1; - while(libsf.good()){ - std::stringstream libcodfn; - libcodfn << "_cuobjdump_complete_lib_" << cnt << "_"; - cmd.str(""); //resetting - cmd << "$CUDA_INSTALL_PATH/bin/cuobjdump -ptx -elf -sass "; - cmd << line; - cmd << " > "; - cmd << libcodfn.str(); - std::cout << "Running cuobjdump on " << line << std::endl; - std::cout << "Using command: " << cmd.str() << std::endl; - result = system(cmd.str().c_str()); - if(result) {printf("ERROR: Failed to execute: %s\n", command); exit(1);} - std::cout << "Done" << std::endl; - - std::cout << "Trying to parse " << libcodfn.str() << std::endl; - FILE *cuobjdump_in; - cuobjdump_in = fopen(libcodfn.str().c_str(), "r"); - struct cuobjdump_parser parser; - parser.elfserial = 1; - parser.ptxserial = 1; - cuobjdump_lex_init(&(parser.scanner)); - cuobjdump_set_in(cuobjdump_in, (parser.scanner)); - cuobjdump_parse(parser.scanner, &parser, cuobjdumpSectionList); - cuobjdump_lex_destroy(parser.scanner); - fclose(cuobjdump_in); - std::getline(libsf, line); - } - libSectionList = cuobjdumpSectionList; - - //Restore the original section list - cuobjdumpSectionList = tmpsl; - } - } else { - printf("GPGPU-Sim PTX: overriding cuobjdump with '%s' (CUOBJDUMP_SIM_FILE is set)\n", override_cuobjdump); - snprintf(fname,1024, "%s",override_cuobjdump); + // Restore the original section list + cuobjdumpSectionList = tmpsl; } + } else { + printf( + "GPGPU-Sim PTX: overriding cuobjdump with '%s' (CUOBJDUMP_SIM_FILE is " + "set)\n", + override_cuobjdump); + snprintf(fname, 1024, "%s", override_cuobjdump); + } } //! Read file into char* -//TODO: convert this to C++ streams, will be way cleaner -char* readfile (const std::string filename){ - assert (filename != ""); - FILE* fp = fopen(filename.c_str(),"r"); - if (!fp) { - std::cout << "ERROR: Could not open file %s for reading\n" << filename << std::endl; - assert (0); - } - // finding size of the file - int filesize= 0; - fseek (fp , 0 , SEEK_END); - - filesize = ftell (fp); - fseek (fp, 0, SEEK_SET); - // allocate and copy the entire ptx - char* ret = (char*)malloc((filesize +1)* sizeof(char)); - fread(ret,1,filesize,fp); - ret[filesize]='\0'; - fclose(fp); - return ret; +// TODO: convert this to C++ streams, will be way cleaner +char *readfile(const std::string filename) { + assert(filename != ""); + FILE *fp = fopen(filename.c_str(), "r"); + if (!fp) { + std::cout << "ERROR: Could not open file %s for reading\n" + << filename << std::endl; + assert(0); + } + // finding size of the file + int filesize = 0; + fseek(fp, 0, SEEK_END); + + filesize = ftell(fp); + fseek(fp, 0, SEEK_SET); + // allocate and copy the entire ptx + char *ret = (char *)malloc((filesize + 1) * sizeof(char)); + fread(ret, 1, filesize, fp); + ret[filesize] = '\0'; + fclose(fp); + return ret; } //! Function that helps debugging -void printSectionList(std::list sl) { - std::list::iterator iter; - for ( iter = sl.begin(); - iter != sl.end(); - iter++ - ){ - (*iter)->print(); - } +void printSectionList(std::list sl) { + std::list::iterator iter; + for (iter = sl.begin(); iter != sl.end(); iter++) { + (*iter)->print(); + } } //! Remove unecessary sm versions from the section list -std::list cuda_runtime_api::pruneSectionList(CUctx_st *context) { - unsigned forced_max_capability = context->get_device()->get_gpgpu()->get_config().get_forced_max_capability(); - - //For ptxplus, force the max capability to 19 if it's higher or unspecified(0) - if (context->get_device()->get_gpgpu()->get_config().convert_to_ptxplus()){ - if ( (forced_max_capability == 0) || - (forced_max_capability >= 20)){ - printf("GPGPU-Sim: WARNING: Capability >= 20 are not supported in PTXPlus\n\tSetting forced_max_capability to 19\n"); - forced_max_capability = 19; - } - } - - std::list prunedList; - - //Find the highest capability (that is lower than the forced maximum) for each cubin file - //and set it in cuobjdumpSectionMap. Do this only for ptx sections - std::map cuobjdumpSectionMap; - int min_ptx_capability_found=0; - for ( std::list::iterator iter = cuobjdumpSectionList.begin(); - iter != cuobjdumpSectionList.end(); - iter++){ - unsigned capability = (*iter)->getArch(); - if(dynamic_cast(*iter) != NULL){ - if(capabilitygetIdentifier())==cuobjdumpSectionMap.end()) - || (cuobjdumpSectionMap[(*iter)->getIdentifier()] < capability)) - cuobjdumpSectionMap[(*iter)->getIdentifier()] = capability; - } - } - } - - //Throw away the sections with the lower capabilites and push those with the highest in - //the pruned list - for ( std::list::iterator iter = cuobjdumpSectionList.begin(); - iter != cuobjdumpSectionList.end(); - iter++){ - unsigned capability = (*iter)->getArch(); - if(capability == cuobjdumpSectionMap[(*iter)->getIdentifier()]){ - prunedList.push_back(*iter); - } else { - delete *iter; - } - } - if(prunedList.empty()){ - printf("Error: No PTX sections found with sm capability that is lower than current forced maximum capability \n minimum ptx capability found = %u, maximum forced ptx capability = %u \n User might want to change either the forced maximum capability from gpgpusim configuration or update the compilation to generate the required PTX version\n",min_ptx_capability_found,forced_max_capability); - abort(); - } - return prunedList; +std::list cuda_runtime_api::pruneSectionList( + CUctx_st *context) { + unsigned forced_max_capability = context->get_device() + ->get_gpgpu() + ->get_config() + .get_forced_max_capability(); + + // For ptxplus, force the max capability to 19 if it's higher or + // unspecified(0) + if (context->get_device()->get_gpgpu()->get_config().convert_to_ptxplus()) { + if ((forced_max_capability == 0) || (forced_max_capability >= 20)) { + printf( + "GPGPU-Sim: WARNING: Capability >= 20 are not supported in " + "PTXPlus\n\tSetting forced_max_capability to 19\n"); + forced_max_capability = 19; + } + } + + std::list prunedList; + + // Find the highest capability (that is lower than the forced maximum) for + // each cubin file and set it in cuobjdumpSectionMap. Do this only for ptx + // sections + std::map cuobjdumpSectionMap; + int min_ptx_capability_found = 0; + for (std::list::iterator iter = + cuobjdumpSectionList.begin(); + iter != cuobjdumpSectionList.end(); iter++) { + unsigned capability = (*iter)->getArch(); + if (dynamic_cast(*iter) != NULL) { + if (capability < min_ptx_capability_found || + min_ptx_capability_found == 0) + min_ptx_capability_found = capability; + if (capability <= forced_max_capability || forced_max_capability == 0) { + if ((cuobjdumpSectionMap.find((*iter)->getIdentifier()) == + cuobjdumpSectionMap.end()) || + (cuobjdumpSectionMap[(*iter)->getIdentifier()] < capability)) + cuobjdumpSectionMap[(*iter)->getIdentifier()] = capability; + } + } + } + + // Throw away the sections with the lower capabilites and push those with the + // highest in the pruned list + for (std::list::iterator iter = + cuobjdumpSectionList.begin(); + iter != cuobjdumpSectionList.end(); iter++) { + unsigned capability = (*iter)->getArch(); + if (capability == cuobjdumpSectionMap[(*iter)->getIdentifier()]) { + prunedList.push_back(*iter); + } else { + delete *iter; + } + } + if (prunedList.empty()) { + printf( + "Error: No PTX sections found with sm capability that is lower than " + "current forced maximum capability \n minimum ptx capability found = " + "%u, maximum forced ptx capability = %u \n User might want to change " + "either the forced maximum capability from gpgpusim configuration or " + "update the compilation to generate the required PTX version\n", + min_ptx_capability_found, forced_max_capability); + abort(); + } + return prunedList; } //! Merge all PTX sections that have a specific identifier into one file -std::list cuda_runtime_api::mergeMatchingSections(std::string identifier){ - const char *ptxcode = ""; - std::list::iterator old_iter; - cuobjdumpPTXSection* old_ptxsection = NULL; - cuobjdumpPTXSection* ptxsection; - std::list mergedList; - - for ( std::list::iterator iter = cuobjdumpSectionList.begin(); - iter != cuobjdumpSectionList.end(); - iter++){ - if((ptxsection=dynamic_cast(*iter)) != NULL && - strcmp(ptxsection->getIdentifier().c_str(), identifier.c_str()) == 0){ - // Read and remove the last PTX section - if (old_ptxsection != NULL) { - ptxcode = readfile(old_ptxsection->getPTXfilename()); - // remove ptx file? - delete *old_iter; - } - - // Append all the PTX from the last PTX section into the current PTX section - // Add 50 to ptxcode to ignore the information regarding version/target/address_size - if (strlen(ptxcode) >= 50) { - FILE *ptxfile = fopen((ptxsection->getPTXfilename()).c_str(), "a"); - fprintf(ptxfile, "%s", ptxcode + 50); - fclose(ptxfile); - } - - old_iter = iter; - old_ptxsection = ptxsection; - } - // Store all non-PTX sections and PTX sections with non-matching identifiers - else { - mergedList.push_back(*iter); - } - } - - // Store the final PTX section - mergedList.push_back(*old_iter); - - return mergedList; +std::list cuda_runtime_api::mergeMatchingSections( + std::string identifier) { + const char *ptxcode = ""; + std::list::iterator old_iter; + cuobjdumpPTXSection *old_ptxsection = NULL; + cuobjdumpPTXSection *ptxsection; + std::list mergedList; + + for (std::list::iterator iter = + cuobjdumpSectionList.begin(); + iter != cuobjdumpSectionList.end(); iter++) { + if ((ptxsection = dynamic_cast(*iter)) != NULL && + strcmp(ptxsection->getIdentifier().c_str(), identifier.c_str()) == 0) { + // Read and remove the last PTX section + if (old_ptxsection != NULL) { + ptxcode = readfile(old_ptxsection->getPTXfilename()); + // remove ptx file? + delete *old_iter; + } + + // Append all the PTX from the last PTX section into the current PTX + // section Add 50 to ptxcode to ignore the information regarding + // version/target/address_size + if (strlen(ptxcode) >= 50) { + FILE *ptxfile = fopen((ptxsection->getPTXfilename()).c_str(), "a"); + fprintf(ptxfile, "%s", ptxcode + 50); + fclose(ptxfile); + } + + old_iter = iter; + old_ptxsection = ptxsection; + } + // Store all non-PTX sections and PTX sections with non-matching identifiers + else { + mergedList.push_back(*iter); + } + } + + // Store the final PTX section + mergedList.push_back(*old_iter); + + return mergedList; } //! Merge any PTX sections with matching identifiers -std::list cuda_runtime_api::mergeSections(){ - std::vector identifier; - cuobjdumpPTXSection* ptxsection; - - // Add all identifiers present in PTX sections to a vector - for ( std::list::iterator iter = cuobjdumpSectionList.begin(); - iter != cuobjdumpSectionList.end(); - iter++){ - if((ptxsection=dynamic_cast(*iter)) != NULL){ - std::string current_id = ptxsection->getIdentifier(); - - // If we haven't yet seen a given identifier, add it to the vector - if (std::find(identifier.begin(), identifier.end(), current_id) == identifier.end()) { - identifier.push_back(current_id); - } - } - } - - // Call mergeMatchingSections on all identifiers in the vector - for ( std::vector::iterator iter = identifier.begin(); - iter != identifier.end(); - iter++) { - cuobjdumpSectionList = mergeMatchingSections(*iter); - } - - return cuobjdumpSectionList; -} - - -//! Within the section list, find the ELF section corresponding to a given identifier -cuobjdumpELFSection* findELFSectionInList(std::list sectionlist, const std::string identifier){ - - std::list::iterator iter; - for ( iter = sectionlist.begin(); - iter != sectionlist.end(); - iter++ - ){ - cuobjdumpELFSection* elfsection; - if((elfsection=dynamic_cast(*iter)) != NULL){ - if(elfsection->getIdentifier() == identifier) - return elfsection; - } - } - return NULL; +std::list cuda_runtime_api::mergeSections() { + std::vector identifier; + cuobjdumpPTXSection *ptxsection; + + // Add all identifiers present in PTX sections to a vector + for (std::list::iterator iter = + cuobjdumpSectionList.begin(); + iter != cuobjdumpSectionList.end(); iter++) { + if ((ptxsection = dynamic_cast(*iter)) != NULL) { + std::string current_id = ptxsection->getIdentifier(); + + // If we haven't yet seen a given identifier, add it to the vector + if (std::find(identifier.begin(), identifier.end(), current_id) == + identifier.end()) { + identifier.push_back(current_id); + } + } + } + + // Call mergeMatchingSections on all identifiers in the vector + for (std::vector::iterator iter = identifier.begin(); + iter != identifier.end(); iter++) { + cuobjdumpSectionList = mergeMatchingSections(*iter); + } + + return cuobjdumpSectionList; +} + +//! Within the section list, find the ELF section corresponding to a given +//! identifier +cuobjdumpELFSection *findELFSectionInList( + std::list sectionlist, const std::string identifier) { + std::list::iterator iter; + for (iter = sectionlist.begin(); iter != sectionlist.end(); iter++) { + cuobjdumpELFSection *elfsection; + if ((elfsection = dynamic_cast(*iter)) != NULL) { + if (elfsection->getIdentifier() == identifier) return elfsection; + } + } + return NULL; } //! Find an ELF section in all the known lists -cuobjdumpELFSection* cuda_runtime_api::findELFSection(const std::string identifier){ - cuobjdumpELFSection* sec = findELFSectionInList(cuobjdumpSectionList, identifier); - if (sec!=NULL)return sec; - sec = findELFSectionInList(libSectionList, identifier); - if (sec!=NULL)return sec; - std::cout << "Could not find " << identifier << std::endl; - assert(0 && "Could not find the required ELF section"); - return NULL; -} - -//! Within the section list, find the PTX section corresponding to a given identifier -cuobjdumpPTXSection* cuda_runtime_api::findPTXSectionInList(std::list §ionlist, const std::string identifier){ - std::list::iterator iter; - for ( iter = sectionlist.begin(); - iter != sectionlist.end(); - iter++ - ){ - cuobjdumpPTXSection* ptxsection; - if((ptxsection=dynamic_cast(*iter)) != NULL){ - if(ptxsection->getIdentifier() == identifier) - return ptxsection; - else { - if(gpgpu_ctx->device_runtime->g_cdp_enabled) { - printf("Warning: __cudaRegisterFatBinary needs %s, but find PTX section with %s\n", - identifier.c_str(), ptxsection->getIdentifier().c_str()); - return ptxsection; - } - } - } - } - return NULL; +cuobjdumpELFSection *cuda_runtime_api::findELFSection( + const std::string identifier) { + cuobjdumpELFSection *sec = + findELFSectionInList(cuobjdumpSectionList, identifier); + if (sec != NULL) return sec; + sec = findELFSectionInList(libSectionList, identifier); + if (sec != NULL) return sec; + std::cout << "Could not find " << identifier << std::endl; + assert(0 && "Could not find the required ELF section"); + return NULL; +} + +//! Within the section list, find the PTX section corresponding to a given +//! identifier +cuobjdumpPTXSection *cuda_runtime_api::findPTXSectionInList( + std::list §ionlist, const std::string identifier) { + std::list::iterator iter; + for (iter = sectionlist.begin(); iter != sectionlist.end(); iter++) { + cuobjdumpPTXSection *ptxsection; + if ((ptxsection = dynamic_cast(*iter)) != NULL) { + if (ptxsection->getIdentifier() == identifier) + return ptxsection; + else { + if (gpgpu_ctx->device_runtime->g_cdp_enabled) { + printf( + "Warning: __cudaRegisterFatBinary needs %s, but find PTX section " + "with %s\n", + identifier.c_str(), ptxsection->getIdentifier().c_str()); + return ptxsection; + } + } + } + } + return NULL; } //! Find an PTX section in all the known lists -cuobjdumpPTXSection* cuda_runtime_api::findPTXSection(const std::string identifier){ - cuobjdumpPTXSection* sec = findPTXSectionInList(cuobjdumpSectionList, identifier); - if (sec!=NULL)return sec; - sec = findPTXSectionInList(libSectionList, identifier); - if (sec!=NULL)return sec; - std::cout << "Could not find " << identifier << std::endl; - assert(0 && "Could not find the required PTX section"); - return NULL; +cuobjdumpPTXSection *cuda_runtime_api::findPTXSection( + const std::string identifier) { + cuobjdumpPTXSection *sec = + findPTXSectionInList(cuobjdumpSectionList, identifier); + if (sec != NULL) return sec; + sec = findPTXSectionInList(libSectionList, identifier); + if (sec != NULL) return sec; + std::cout << "Could not find " << identifier << std::endl; + assert(0 && "Could not find the required PTX section"); + return NULL; } - - //! Extract the code using cuobjdump and remove unnecessary sections -void cuda_runtime_api::cuobjdumpInit(){ - CUctx_st *context = GPGPUSim_Context(gpgpu_ctx); - extract_code_using_cuobjdump(); //extract all the output of cuobjdump to _cuobjdump_*.* - const char* pre_load = getenv("CUOBJDUMP_SIM_FILE"); - if (pre_load ==NULL || strlen(pre_load)==0){ - cuobjdumpSectionList = pruneSectionList(context); - cuobjdumpSectionList = mergeSections(); - } +void cuda_runtime_api::cuobjdumpInit() { + CUctx_st *context = GPGPUSim_Context(gpgpu_ctx); + extract_code_using_cuobjdump(); // extract all the output of cuobjdump to + // _cuobjdump_*.* + const char *pre_load = getenv("CUOBJDUMP_SIM_FILE"); + if (pre_load == NULL || strlen(pre_load) == 0) { + cuobjdumpSectionList = pruneSectionList(context); + cuobjdumpSectionList = mergeSections(); + } } - //! Either submit PTX for simulation or convert SASS to PTXPlus and submit it -void gpgpu_context::cuobjdumpParseBinary(unsigned int handle){ - - CUctx_st *context = GPGPUSim_Context(this); - if(api->fatbin_registered[handle]) return; - api->fatbin_registered[handle] = true; - std::string fname = api->fatbinmap[handle]; - - if (api->name_symtab.find(fname) != api->name_symtab.end()) { - symbol_table *symtab = api->name_symtab[fname]; - context->add_binary(symtab, handle); - return; - } - symbol_table *symtab; +void gpgpu_context::cuobjdumpParseBinary(unsigned int handle) { + CUctx_st *context = GPGPUSim_Context(this); + if (api->fatbin_registered[handle]) return; + api->fatbin_registered[handle] = true; + std::string fname = api->fatbinmap[handle]; + + if (api->name_symtab.find(fname) != api->name_symtab.end()) { + symbol_table *symtab = api->name_symtab[fname]; + context->add_binary(symtab, handle); + return; + } + symbol_table *symtab; #if (CUDART_VERSION >= 6000) - //loops through all ptx files from smallest sm version to largest - std::map >::iterator itr_m; - for (itr_m = api->version_filename.begin(); itr_m!=api->version_filename.end(); itr_m++){ - std::set::iterator itr_s; - for (itr_s = itr_m->second.begin(); itr_s!=itr_m->second.end(); itr_s++){ - std::string ptx_filename = *itr_s; - printf("GPGPU-Sim PTX: Parsing %s\n",ptx_filename.c_str()); - symtab = gpgpu_ptx_sim_load_ptx_from_filename( ptx_filename.c_str() ); - } - } - api->name_symtab[fname] = symtab; - context->add_binary(symtab, handle); - api->load_static_globals(symtab,STATIC_ALLOC_LIMIT,0xFFFFFFFF,context->get_device()->get_gpgpu()); - api->load_constants(symtab,STATIC_ALLOC_LIMIT,context->get_device()->get_gpgpu()); - for (itr_m = api->version_filename.begin(); itr_m!=api->version_filename.end(); itr_m++){ - std::set::iterator itr_s; - for (itr_s = itr_m->second.begin(); itr_s!=itr_m->second.end(); itr_s++){ - std::string ptx_filename = *itr_s; - printf("GPGPU-Sim PTX: Loading PTXInfo from %s\n",ptx_filename.c_str()); - gpgpu_ptx_info_load_from_filename( ptx_filename.c_str(), itr_m->first ); - } - } - return; + // loops through all ptx files from smallest sm version to largest + std::map >::iterator itr_m; + for (itr_m = api->version_filename.begin(); + itr_m != api->version_filename.end(); itr_m++) { + std::set::iterator itr_s; + for (itr_s = itr_m->second.begin(); itr_s != itr_m->second.end(); itr_s++) { + std::string ptx_filename = *itr_s; + printf("GPGPU-Sim PTX: Parsing %s\n", ptx_filename.c_str()); + symtab = gpgpu_ptx_sim_load_ptx_from_filename(ptx_filename.c_str()); + } + } + api->name_symtab[fname] = symtab; + context->add_binary(symtab, handle); + api->load_static_globals(symtab, STATIC_ALLOC_LIMIT, 0xFFFFFFFF, + context->get_device()->get_gpgpu()); + api->load_constants(symtab, STATIC_ALLOC_LIMIT, + context->get_device()->get_gpgpu()); + for (itr_m = api->version_filename.begin(); + itr_m != api->version_filename.end(); itr_m++) { + std::set::iterator itr_s; + for (itr_s = itr_m->second.begin(); itr_s != itr_m->second.end(); itr_s++) { + std::string ptx_filename = *itr_s; + printf("GPGPU-Sim PTX: Loading PTXInfo from %s\n", ptx_filename.c_str()); + gpgpu_ptx_info_load_from_filename(ptx_filename.c_str(), itr_m->first); + } + } + return; #endif - unsigned max_capability = 0; - for ( std::list::iterator iter = api->cuobjdumpSectionList.begin(); - iter != api->cuobjdumpSectionList.end(); - iter++){ - unsigned capability = (*iter)->getArch(); - if (capability > max_capability) max_capability = capability; - } - if (max_capability > 20) printf("WARNING: No guarantee that PTX will be parsed for SM version %u\n", max_capability); - if (max_capability == 0) max_capability=context->get_device()->get_gpgpu()->get_config().get_forced_max_capability(); - - cuobjdumpPTXSection* ptx = NULL; - const char* pre_load = getenv("CUOBJDUMP_SIM_FILE"); - if(pre_load==NULL || strlen(pre_load)==0) - ptx = api->findPTXSection(fname); - char *ptxcode; - const char *override_ptx_name = getenv("PTX_SIM_KERNELFILE"); - if (override_ptx_name == NULL or getenv("PTX_SIM_USE_PTX_FILE") == NULL or strlen(getenv("PTX_SIM_USE_PTX_FILE"))==0) { - ptxcode = readfile(ptx->getPTXfilename()); - } else { - printf("GPGPU-Sim PTX: overriding embedded ptx with '%s' (PTX_SIM_USE_PTX_FILE is set)\n", override_ptx_name); - ptxcode = readfile(override_ptx_name); - } - if(context->get_device()->get_gpgpu()->get_config().convert_to_ptxplus() ) { - cuobjdumpELFSection* elfsection = api->findELFSection(ptx->getIdentifier()); - assert (elfsection!= NULL); - char *ptxplus_str = ptxinfo->gpgpu_ptx_sim_convert_ptx_and_sass_to_ptxplus( - ptx->getPTXfilename(), - elfsection->getELFfilename(), - elfsection->getSASSfilename()); - symtab=gpgpu_ptx_sim_load_ptx_from_string(ptxplus_str, handle); - printf("Adding %s with cubin handle %u\n", ptx->getPTXfilename().c_str(), handle); - context->add_binary(symtab, handle); - gpgpu_ptxinfo_load_from_string( ptxcode, handle, max_capability, context->no_of_ptx ); - delete[] ptxplus_str; - } else { - symtab=gpgpu_ptx_sim_load_ptx_from_string(ptxcode, handle); - //if CUOBJDUMP_SIM_FILE is not set, ptx is NULL. So comment below. - //printf("Adding %s with cubin handle %u\n", ptx->getPTXfilename().c_str(), handle); - context->add_binary(symtab, handle); - gpgpu_ptxinfo_load_from_string( ptxcode, handle, max_capability, context->no_of_ptx ); - } - api->load_static_globals(symtab,STATIC_ALLOC_LIMIT,0xFFFFFFFF,context->get_device()->get_gpgpu()); - api->load_constants(symtab,STATIC_ALLOC_LIMIT,context->get_device()->get_gpgpu()); - api->name_symtab[fname] = symtab; - - //TODO: Remove temporarily files as per configurations + unsigned max_capability = 0; + for (std::list::iterator iter = + api->cuobjdumpSectionList.begin(); + iter != api->cuobjdumpSectionList.end(); iter++) { + unsigned capability = (*iter)->getArch(); + if (capability > max_capability) max_capability = capability; + } + if (max_capability > 20) + printf("WARNING: No guarantee that PTX will be parsed for SM version %u\n", + max_capability); + if (max_capability == 0) + max_capability = context->get_device() + ->get_gpgpu() + ->get_config() + .get_forced_max_capability(); + + cuobjdumpPTXSection *ptx = NULL; + const char *pre_load = getenv("CUOBJDUMP_SIM_FILE"); + if (pre_load == NULL || strlen(pre_load) == 0) + ptx = api->findPTXSection(fname); + char *ptxcode; + const char *override_ptx_name = getenv("PTX_SIM_KERNELFILE"); + if (override_ptx_name == NULL or getenv("PTX_SIM_USE_PTX_FILE") == NULL or + strlen(getenv("PTX_SIM_USE_PTX_FILE")) == 0) { + ptxcode = readfile(ptx->getPTXfilename()); + } else { + printf( + "GPGPU-Sim PTX: overriding embedded ptx with '%s' " + "(PTX_SIM_USE_PTX_FILE is set)\n", + override_ptx_name); + ptxcode = readfile(override_ptx_name); + } + if (context->get_device()->get_gpgpu()->get_config().convert_to_ptxplus()) { + cuobjdumpELFSection *elfsection = api->findELFSection(ptx->getIdentifier()); + assert(elfsection != NULL); + char *ptxplus_str = ptxinfo->gpgpu_ptx_sim_convert_ptx_and_sass_to_ptxplus( + ptx->getPTXfilename(), elfsection->getELFfilename(), + elfsection->getSASSfilename()); + symtab = gpgpu_ptx_sim_load_ptx_from_string(ptxplus_str, handle); + printf("Adding %s with cubin handle %u\n", ptx->getPTXfilename().c_str(), + handle); + context->add_binary(symtab, handle); + gpgpu_ptxinfo_load_from_string(ptxcode, handle, max_capability, + context->no_of_ptx); + delete[] ptxplus_str; + } else { + symtab = gpgpu_ptx_sim_load_ptx_from_string(ptxcode, handle); + // if CUOBJDUMP_SIM_FILE is not set, ptx is NULL. So comment below. + // printf("Adding %s with cubin handle %u\n", ptx->getPTXfilename().c_str(), + // handle); + context->add_binary(symtab, handle); + gpgpu_ptxinfo_load_from_string(ptxcode, handle, max_capability, + context->no_of_ptx); + } + api->load_static_globals(symtab, STATIC_ALLOC_LIMIT, 0xFFFFFFFF, + context->get_device()->get_gpgpu()); + api->load_constants(symtab, STATIC_ALLOC_LIMIT, + context->get_device()->get_gpgpu()); + api->name_symtab[fname] = symtab; + + // TODO: Remove temporarily files as per configurations } } extern "C" { -void** CUDARTAPI __cudaRegisterFatBinary( void *fatCubin ) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - return cudaRegisterFatBinaryInternal(fatCubin); +void **CUDARTAPI __cudaRegisterFatBinary(void *fatCubin) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + return cudaRegisterFatBinaryInternal(fatCubin); } -void CUDARTAPI __cudaRegisterFatBinaryEnd( void **fatCubinHandle ) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } +void CUDARTAPI __cudaRegisterFatBinaryEnd(void **fatCubinHandle) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } } -unsigned CUDARTAPI __cudaPushCallConfiguration(dim3 gridDim, - dim3 blockDim, - size_t sharedMem = 0, - struct CUstream_st *stream = 0) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - cudaConfigureCallInternal(gridDim, blockDim, sharedMem, stream); +unsigned CUDARTAPI __cudaPushCallConfiguration(dim3 gridDim, dim3 blockDim, + size_t sharedMem = 0, + struct CUstream_st *stream = 0) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + cudaConfigureCallInternal(gridDim, blockDim, sharedMem, stream); } -cudaError_t CUDARTAPI __cudaPopCallConfiguration( - dim3 *gridDim, - dim3 *blockDim, - size_t *sharedMem, - void *stream -) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - return g_last_cudaError = cudaSuccess; +cudaError_t CUDARTAPI __cudaPopCallConfiguration(dim3 *gridDim, dim3 *blockDim, + size_t *sharedMem, + void *stream) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + return g_last_cudaError = cudaSuccess; } - -void CUDARTAPI __cudaRegisterFunction( - void **fatCubinHandle, - const char *hostFun, - char *deviceFun, - const char *deviceName, - int thread_limit, - uint3 *tid, - uint3 *bid, - dim3 *bDim, - dim3 *gDim -) { - cudaRegisterFunctionInternal( - fatCubinHandle, - hostFun, - deviceFun, - deviceName, - thread_limit, - tid, - bid, - bDim, - gDim - ); - +void CUDARTAPI __cudaRegisterFunction(void **fatCubinHandle, + const char *hostFun, char *deviceFun, + const char *deviceName, int thread_limit, + uint3 *tid, uint3 *bid, dim3 *bDim, + dim3 *gDim) { + cudaRegisterFunctionInternal(fatCubinHandle, hostFun, deviceFun, deviceName, + thread_limit, tid, bid, bDim, gDim); } extern void __cudaRegisterVar( - void **fatCubinHandle, - char *hostVar, //pointer to...something - char *deviceAddress, //name of variable - const char *deviceName, //name of variable (same as above) - int ext, - int size, - int constant, - int global ) -{ - cudaRegisterVarInternal( - fatCubinHandle, - hostVar, - deviceAddress, - deviceName, - ext, - size, - constant, - global ); -} - -__host__ cudaError_t CUDARTAPI cudaConfigureCall(dim3 gridDim, dim3 blockDim, size_t sharedMem, cudaStream_t stream) -{ - return cudaConfigureCallInternal(gridDim, blockDim, sharedMem, stream); + void **fatCubinHandle, + char *hostVar, // pointer to...something + char *deviceAddress, // name of variable + const char *deviceName, // name of variable (same as above) + int ext, int size, int constant, int global) { + cudaRegisterVarInternal(fatCubinHandle, hostVar, deviceAddress, deviceName, + ext, size, constant, global); } -void __cudaUnregisterFatBinary(void **fatCubinHandle) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } +__host__ cudaError_t CUDARTAPI cudaConfigureCall(dim3 gridDim, dim3 blockDim, + size_t sharedMem, + cudaStream_t stream) { + return cudaConfigureCallInternal(gridDim, blockDim, sharedMem, stream); } -cudaError_t cudaDeviceReset ( void ) { - // Should reset the simulated GPU - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - return g_last_cudaError = cudaSuccess; +void __cudaUnregisterFatBinary(void **fatCubinHandle) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } } -cudaError_t CUDARTAPI cudaDeviceSynchronize(void) -{ - return cudaDeviceSynchronizeInternal(); +cudaError_t cudaDeviceReset(void) { + // Should reset the simulated GPU + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + return g_last_cudaError = cudaSuccess; } -void __cudaRegisterShared( - void **fatCubinHandle, - void **devicePtr -) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - // we don't do anything here - printf("GPGPU-Sim PTX: __cudaRegisterShared\n" ); +cudaError_t CUDARTAPI cudaDeviceSynchronize(void) { + return cudaDeviceSynchronizeInternal(); } -void CUDARTAPI __cudaRegisterSharedVar( - void **fatCubinHandle, - void **devicePtr, - size_t size, - size_t alignment, - int storage -) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - // we don't do anything here - printf("GPGPU-Sim PTX: __cudaRegisterSharedVar\n" ); +void __cudaRegisterShared(void **fatCubinHandle, void **devicePtr) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + // we don't do anything here + printf("GPGPU-Sim PTX: __cudaRegisterShared\n"); } -void __cudaRegisterTexture( - void **fatCubinHandle, - const struct textureReference *hostVar, - const void **deviceAddress, - const char *deviceName, - int dim, - int norm, - int ext -) //passes in a newly created textureReference -{ - __cudaRegisterTextureInternal(fatCubinHandle, hostVar, deviceAddress, deviceName, dim, norm, ext); +void CUDARTAPI __cudaRegisterSharedVar(void **fatCubinHandle, void **devicePtr, + size_t size, size_t alignment, + int storage) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + // we don't do anything here + printf("GPGPU-Sim PTX: __cudaRegisterSharedVar\n"); } - -char __cudaInitModule( - void **fatCubinHandle -) +void __cudaRegisterTexture( + void **fatCubinHandle, const struct textureReference *hostVar, + const void **deviceAddress, const char *deviceName, int dim, int norm, + int ext) // passes in a newly created textureReference { - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - cuda_not_implemented(__my_func__,__LINE__); - return g_last_cudaError = cudaErrorUnknown; + __cudaRegisterTextureInternal(fatCubinHandle, hostVar, deviceAddress, + deviceName, dim, norm, ext); } +char __cudaInitModule(void **fatCubinHandle) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + cuda_not_implemented(__my_func__, __LINE__); + return g_last_cudaError = cudaErrorUnknown; +} -cudaError_t cudaGLRegisterBufferObject(GLuint bufferObj) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("GPGPU-Sim PTX: Execution warning: ignoring call to \"%s\"\n", __my_func__ ); - return g_last_cudaError = cudaSuccess; +cudaError_t cudaGLRegisterBufferObject(GLuint bufferObj) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("GPGPU-Sim PTX: Execution warning: ignoring call to \"%s\"\n", + __my_func__); + return g_last_cudaError = cudaSuccess; } -cudaError_t cudaGLMapBufferObject(void** devPtr, GLuint bufferObj) -{ - return cudaGLMapBufferObjectInternal(devPtr, bufferObj); +cudaError_t cudaGLMapBufferObject(void **devPtr, GLuint bufferObj) { + return cudaGLMapBufferObjectInternal(devPtr, bufferObj); } -cudaError_t cudaGLUnmapBufferObject(GLuint bufferObj) -{ - return cudaGLUnmapBufferObjectInternal(bufferObj); +cudaError_t cudaGLUnmapBufferObject(GLuint bufferObj) { + return cudaGLUnmapBufferObjectInternal(bufferObj); } -cudaError_t cudaGLUnregisterBufferObject(GLuint bufferObj) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("GPGPU-Sim PTX: Execution warning: ignoring call to \"%s\"\n", __my_func__ ); - return g_last_cudaError = cudaSuccess; +cudaError_t cudaGLUnregisterBufferObject(GLuint bufferObj) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("GPGPU-Sim PTX: Execution warning: ignoring call to \"%s\"\n", + __my_func__); + return g_last_cudaError = cudaSuccess; } #if (CUDART_VERSION >= 2010) -cudaError_t CUDARTAPI cudaHostAlloc(void **pHost, size_t bytes, unsigned int flags) -{ - return cudaHostAllocInternal(pHost, bytes, flags); +cudaError_t CUDARTAPI cudaHostAlloc(void **pHost, size_t bytes, + unsigned int flags) { + return cudaHostAllocInternal(pHost, bytes, flags); } -cudaError_t CUDARTAPI cudaHostGetDevicePointer(void **pDevice, void *pHost, unsigned int flags) -{ - return cudaHostGetDevicePointerInternal(pDevice, pHost, flags); +cudaError_t CUDARTAPI cudaHostGetDevicePointer(void **pDevice, void *pHost, + unsigned int flags) { + return cudaHostGetDevicePointerInternal(pDevice, pHost, flags); } -__host__ cudaError_t CUDARTAPI cudaPointerGetAttributes( - cudaPointerAttributes *attributes, - const void *ptr -) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - cuda_not_implemented(__my_func__,__LINE__); - return g_last_cudaError = cudaErrorUnknown; +__host__ cudaError_t CUDARTAPI +cudaPointerGetAttributes(cudaPointerAttributes *attributes, const void *ptr) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + cuda_not_implemented(__my_func__, __LINE__); + return g_last_cudaError = cudaErrorUnknown; } -__host__ cudaError_t CUDARTAPI cudaDeviceCanAccessPeer( - int *canAccessPeer, - int device, - int peerDevice -) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - cuda_not_implemented(__my_func__,__LINE__); - return g_last_cudaError = cudaErrorUnknown; +__host__ cudaError_t CUDARTAPI cudaDeviceCanAccessPeer(int *canAccessPeer, + int device, + int peerDevice) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + cuda_not_implemented(__my_func__, __LINE__); + return g_last_cudaError = cudaErrorUnknown; } -__host__ cudaError_t CUDARTAPI cudaDeviceEnablePeerAccess( - int peerDevice, - unsigned int flags -) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - cuda_not_implemented(__my_func__,__LINE__); - return g_last_cudaError = cudaErrorUnknown; +__host__ cudaError_t CUDARTAPI cudaDeviceEnablePeerAccess(int peerDevice, + unsigned int flags) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + cuda_not_implemented(__my_func__, __LINE__); + return g_last_cudaError = cudaErrorUnknown; } -cudaError_t CUDARTAPI cudaSetValidDevices(int *device_arr, int len) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - cuda_not_implemented(__my_func__,__LINE__); - return g_last_cudaError = cudaErrorUnknown; +cudaError_t CUDARTAPI cudaSetValidDevices(int *device_arr, int len) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + cuda_not_implemented(__my_func__, __LINE__); + return g_last_cudaError = cudaErrorUnknown; } -cudaError_t CUDARTAPI cudaSetDeviceFlags( int flags ) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - // This flag is implicitly always on (unless you are using the driver API). It is safe for GPGPU-Sim to - // just ignore it. - if ( cudaDeviceMapHost == flags ) { - return g_last_cudaError = cudaSuccess; - } else { - cuda_not_implemented(__my_func__,__LINE__); - return g_last_cudaError = cudaErrorUnknown; - } +cudaError_t CUDARTAPI cudaSetDeviceFlags(int flags) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + // This flag is implicitly always on (unless you are using the driver API). It + // is safe for GPGPU-Sim to just ignore it. + if (cudaDeviceMapHost == flags) { + return g_last_cudaError = cudaSuccess; + } else { + cuda_not_implemented(__my_func__, __LINE__); + return g_last_cudaError = cudaErrorUnknown; + } } - -cudaError_t CUDARTAPI cudaFuncGetAttributes(struct cudaFuncAttributes *attr, const char *hostFun ) -{ - return cudaFuncGetAttributesInternal(attr, hostFun ); +cudaError_t CUDARTAPI cudaFuncGetAttributes(struct cudaFuncAttributes *attr, + const char *hostFun) { + return cudaFuncGetAttributesInternal(attr, hostFun); } -cudaError_t CUDARTAPI cudaDriverGetVersion(int *driverVersion) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - *driverVersion = CUDART_VERSION; - return g_last_cudaError = cudaSuccess; +cudaError_t CUDARTAPI cudaDriverGetVersion(int *driverVersion) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + *driverVersion = CUDART_VERSION; + return g_last_cudaError = cudaSuccess; } -cudaError_t CUDARTAPI cudaRuntimeGetVersion(int *runtimeVersion) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - *runtimeVersion = CUDART_VERSION; - return g_last_cudaError = cudaSuccess; +cudaError_t CUDARTAPI cudaRuntimeGetVersion(int *runtimeVersion) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + *runtimeVersion = CUDART_VERSION; + return g_last_cudaError = cudaSuccess; } #if CUDART_VERSION >= 3000 -__host__ cudaError_t CUDARTAPI cudaFuncSetCacheConfig(const char *func, enum cudaFuncCache cacheConfig ) -{ - return cudaFuncSetCacheConfigInternal(func, cacheConfig); +__host__ cudaError_t CUDARTAPI +cudaFuncSetCacheConfig(const char *func, enum cudaFuncCache cacheConfig) { + return cudaFuncSetCacheConfigInternal(func, cacheConfig); } -//Jin: hack for cdp -__host__ cudaError_t CUDARTAPI cudaDeviceSetLimit(enum cudaLimit limit, size_t value) { - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - return g_last_cudaError = cudaSuccess; +// Jin: hack for cdp +__host__ cudaError_t CUDARTAPI cudaDeviceSetLimit(enum cudaLimit limit, + size_t value) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + return g_last_cudaError = cudaSuccess; } //#if CUDART_VERSION >= 9000 -//__host__ cudaError_t cudaFuncSetAttribute ( const void* func, enum cudaFuncAttribute attr, int value ) { +//__host__ cudaError_t cudaFuncSetAttribute ( const void* func, enum +// cudaFuncAttribute attr, int value ) { - //ignore this Attribute for now, and the default is that carveout = cudaSharedmemCarveoutDefault; // (-1) +// ignore this Attribute for now, and the default is that carveout = +// cudaSharedmemCarveoutDefault; // (-1) // return g_last_cudaError = cudaSuccess; //} - #endif #endif - #if CUDART_VERSION >= 9000 /** * \brief Set attributes for a given function * * This function sets the attributes of a function specified via \p entry. * The parameter \p entry must be a pointer to a function that executes - * on the device. The parameter specified by \p entry must be declared as a \p __global__ - * function. The enumeration defined by \p attr is set to the value defined by \p value - * If the specified function does not exist, then ::cudaErrorInvalidDeviceFunction is returned. - * If the specified attribute cannot be written, or if the value is incorrect, - * then ::cudaErrorInvalidValue is returned. + * on the device. The parameter specified by \p entry must be declared as a \p + * __global__ function. The enumeration defined by \p attr is set to the value + * defined by \p value If the specified function does not exist, then + * ::cudaErrorInvalidDeviceFunction is returned. If the specified attribute + * cannot be written, or if the value is incorrect, then ::cudaErrorInvalidValue + * is returned. * * Valid values for \p attr are: - * ::cuFuncAttrMaxDynamicSharedMem - Maximum size of dynamic shared memory per block - * ::cudaFuncAttributePreferredSharedMemoryCarveout - Preferred shared memory-L1 cache split ratio + * ::cuFuncAttrMaxDynamicSharedMem - Maximum size of dynamic shared memory per + * block + * ::cudaFuncAttributePreferredSharedMemoryCarveout - Preferred shared memory-L1 + * cache split ratio * * \param entry - Function to get attributes of * \param attr - Attribute to set @@ -3726,216 +3846,235 @@ __host__ cudaError_t CUDARTAPI cudaDeviceSetLimit(enum cudaLimit limit, size_t v * ::cudaErrorInvalidValue * \notefnerr * - * \ref ::cudaLaunchKernel(const T *func, dim3 gridDim, dim3 blockDim, void **args, size_t sharedMem, cudaStream_t stream) "cudaLaunchKernel (C++ API)", - * \ref ::cudaFuncSetCacheConfig(T*, enum cudaFuncCache) "cudaFuncSetCacheConfig (C++ API)", - * \ref ::cudaFuncGetAttributes(struct cudaFuncAttributes*, const void*) "cudaFuncGetAttributes (C API)", + * \ref ::cudaLaunchKernel(const T *func, dim3 gridDim, dim3 blockDim, void + * **args, size_t sharedMem, cudaStream_t stream) "cudaLaunchKernel (C++ API)", + * \ref ::cudaFuncSetCacheConfig(T*, enum cudaFuncCache) "cudaFuncSetCacheConfig + * (C++ API)", \ref ::cudaFuncGetAttributes(struct cudaFuncAttributes*, const + * void*) "cudaFuncGetAttributes (C API)", * ::cudaSetDoubleForDevice, * ::cudaSetDoubleForHost, * \ref ::cudaSetupArgument(T, size_t) "cudaSetupArgument (C++ API)" */ -cudaError_t CUDARTAPI cudaFuncSetAttribute(const void *func, enum cudaFuncAttribute attr, int value) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("GPGPU-Sim PTX: Execution warning: ignoring call to \"%s ( func=%p, attr=%d, value=%d )\"\n", - __my_func__, func, attr, value ); - return g_last_cudaError = cudaSuccess; +cudaError_t CUDARTAPI cudaFuncSetAttribute(const void *func, + enum cudaFuncAttribute attr, + int value) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf( + "GPGPU-Sim PTX: Execution warning: ignoring call to \"%s ( func=%p, " + "attr=%d, value=%d )\"\n", + __my_func__, func, attr, value); + return g_last_cudaError = cudaSuccess; } #endif -cudaError_t CUDARTAPI cudaGLSetGLDevice(int device) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("GPGPU-Sim PTX: Execution warning: ignoring call to \"%s\"\n", __my_func__ ); - return g_last_cudaError = cudaErrorUnknown; +cudaError_t CUDARTAPI cudaGLSetGLDevice(int device) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("GPGPU-Sim PTX: Execution warning: ignoring call to \"%s\"\n", + __my_func__); + return g_last_cudaError = cudaErrorUnknown; } -typedef void* HGPUNV; +typedef void *HGPUNV; -cudaError_t CUDARTAPI cudaWGLGetDevice(int *device, HGPUNV hGpu) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - cuda_not_implemented(__my_func__,__LINE__); - return g_last_cudaError = cudaErrorUnknown; +cudaError_t CUDARTAPI cudaWGLGetDevice(int *device, HGPUNV hGpu) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + cuda_not_implemented(__my_func__, __LINE__); + return g_last_cudaError = cudaErrorUnknown; } -void CUDARTAPI __cudaMutexOperation(int lock) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - cuda_not_implemented(__my_func__,__LINE__); +void CUDARTAPI __cudaMutexOperation(int lock) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + cuda_not_implemented(__my_func__, __LINE__); } -void CUDARTAPI __cudaTextureFetch(const void *tex, void *index, int integer, void *val) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - cuda_not_implemented(__my_func__,__LINE__); +void CUDARTAPI __cudaTextureFetch(const void *tex, void *index, int integer, + void *val) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + cuda_not_implemented(__my_func__, __LINE__); } - } namespace cuda_math { -void CUDARTAPI __cudaMutexOperation(int lock) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - cuda_not_implemented(__my_func__,__LINE__); +void CUDARTAPI __cudaMutexOperation(int lock) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + cuda_not_implemented(__my_func__, __LINE__); } -void CUDARTAPI __cudaTextureFetch(const void *tex, void *index, int integer, void *val) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - cuda_not_implemented(__my_func__,__LINE__); +void CUDARTAPI __cudaTextureFetch(const void *tex, void *index, int integer, + void *val) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + cuda_not_implemented(__my_func__, __LINE__); } -int CUDARTAPI __cudaSynchronizeThreads(void**, void*) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - //TODO This function should syncronize if we support Asyn kernel calls - return g_last_cudaError = cudaSuccess; +int CUDARTAPI __cudaSynchronizeThreads(void **, void *) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + // TODO This function should syncronize if we support Asyn kernel calls + return g_last_cudaError = cudaSuccess; } -} +} // namespace cuda_math //////// /// static functions -int cuda_runtime_api::load_static_globals( symbol_table *symtab, unsigned min_gaddr, unsigned max_gaddr, gpgpu_t *gpu ) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf( "GPGPU-Sim PTX: loading globals with explicit initializers... \n" ); - fflush(stdout); - int ng_bytes=0; - symbol_table::iterator g=symtab->global_iterator_begin(); - - for ( ; g!=symtab->global_iterator_end(); g++) { - symbol *global = *g; - if ( global->has_initializer() ) { - printf( "GPGPU-Sim PTX: initializing '%s' ... ", global->name().c_str() ); - unsigned addr=global->get_address(); - const type_info *type = global->type(); - type_info_key ti=type->get_key(); - size_t size; - int t; - ti.type_decode(size,t); - int nbytes = size/8; - int offset=0; - std::list init_list = global->get_initializer(); - for ( std::list::iterator i=init_list.begin(); i!=init_list.end(); i++ ) { - operand_info op = *i; - ptx_reg_t value = op.get_literal_value(); - assert( (addr+offset+nbytes) < min_gaddr ); // min_gaddr is start of "heap" for cudaMalloc - gpu->get_global_memory()->write(addr+offset,nbytes,&value,NULL,NULL); // assuming little endian here - offset+=nbytes; - ng_bytes+=nbytes; - } - printf(" wrote %u bytes\n", offset ); - } - } - printf( "GPGPU-Sim PTX: finished loading globals (%u bytes total).\n", ng_bytes ); - fflush(stdout); - return ng_bytes; -} - -int cuda_runtime_api::load_constants( symbol_table *symtab, addr_t min_gaddr, gpgpu_t *gpu ) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf( "GPGPU-Sim PTX: loading constants with explicit initializers... " ); - fflush(stdout); - int nc_bytes = 0; - symbol_table::iterator g=symtab->const_iterator_begin(); - - for ( ; g!=symtab->const_iterator_end(); g++) { - symbol *constant = *g; - if ( constant->is_const() && constant->has_initializer() ) { - - // get the constant element data size - int basic_type; - size_t num_bits; - constant->type()->get_key().type_decode(num_bits,basic_type); - - std::list init_list = constant->get_initializer(); - int nbytes_written = 0; - for ( std::list::iterator i=init_list.begin(); i!=init_list.end(); i++ ) { - operand_info op = *i; - ptx_reg_t value = op.get_literal_value(); - int nbytes = num_bits/8; - switch ( op.get_type() ) { - case int_t: assert(nbytes >= 1); break; - case float_op_t: assert(nbytes == 4); break; - case double_op_t: assert(nbytes >= 4); break; // account for double DEMOTING - default: - abort(); - } - unsigned addr=constant->get_address() + nbytes_written; - assert( addr+nbytes < min_gaddr ); - - gpu->get_global_memory()->write(addr,nbytes,&value,NULL,NULL); // assume little endian (so u8 is the first byte in u32) - nc_bytes+=nbytes; - nbytes_written += nbytes; - } - } - } - printf( " done.\n"); - fflush(stdout); - return nc_bytes; -} - -kernel_info_t * cuda_runtime_api::gpgpu_cuda_ptx_sim_init_grid( const char *hostFun, - gpgpu_ptx_sim_arg_list_t args, - struct dim3 gridDim, - struct dim3 blockDim, - CUctx_st* context ) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - function_info *entry = context->get_kernel(hostFun); - gpgpu_t* gpu= context->get_device()->get_gpgpu(); - /* - Passing a snapshot of the GPU's current texture mapping to the kernel's info - as kernels should use texture bindings present at the time of their launch. - */ - kernel_info_t *result = new kernel_info_t(gridDim,blockDim,entry,gpu->getNameArrayMapping(),gpu->getNameInfoMapping()); - if( entry == NULL ) { - printf("GPGPU-Sim PTX: ERROR launching kernel -- no PTX implementation found for %p\n", hostFun); - abort(); - } - unsigned argcount=args.size(); - unsigned argn=1; - for( gpgpu_ptx_sim_arg_list_t::iterator a = args.begin(); a != args.end(); a++ ) { - entry->add_param_data(argcount-argn,&(*a)); - argn++; - } - - entry->finalize(result->get_param_memory()); - gpgpu_ctx->func_sim->g_ptx_kernel_count++; - fflush(stdout); - - if(g_debug_execution >= 4){ - entry->ptx_jit_config(g_mallocPtr_Size, result->get_param_memory(), (gpgpu_t *) context->get_device()->get_gpgpu(), gridDim, blockDim); +int cuda_runtime_api::load_static_globals(symbol_table *symtab, + unsigned min_gaddr, + unsigned max_gaddr, gpgpu_t *gpu) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("GPGPU-Sim PTX: loading globals with explicit initializers... \n"); + fflush(stdout); + int ng_bytes = 0; + symbol_table::iterator g = symtab->global_iterator_begin(); + + for (; g != symtab->global_iterator_end(); g++) { + symbol *global = *g; + if (global->has_initializer()) { + printf("GPGPU-Sim PTX: initializing '%s' ... ", + global->name().c_str()); + unsigned addr = global->get_address(); + const type_info *type = global->type(); + type_info_key ti = type->get_key(); + size_t size; + int t; + ti.type_decode(size, t); + int nbytes = size / 8; + int offset = 0; + std::list init_list = global->get_initializer(); + for (std::list::iterator i = init_list.begin(); + i != init_list.end(); i++) { + operand_info op = *i; + ptx_reg_t value = op.get_literal_value(); + assert((addr + offset + nbytes) < + min_gaddr); // min_gaddr is start of "heap" for cudaMalloc + gpu->get_global_memory()->write(addr + offset, nbytes, &value, NULL, + NULL); // assuming little endian here + offset += nbytes; + ng_bytes += nbytes; + } + printf(" wrote %u bytes\n", offset); + } + } + printf("GPGPU-Sim PTX: finished loading globals (%u bytes total).\n", + ng_bytes); + fflush(stdout); + return ng_bytes; +} + +int cuda_runtime_api::load_constants(symbol_table *symtab, addr_t min_gaddr, + gpgpu_t *gpu) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("GPGPU-Sim PTX: loading constants with explicit initializers... "); + fflush(stdout); + int nc_bytes = 0; + symbol_table::iterator g = symtab->const_iterator_begin(); + + for (; g != symtab->const_iterator_end(); g++) { + symbol *constant = *g; + if (constant->is_const() && constant->has_initializer()) { + // get the constant element data size + int basic_type; + size_t num_bits; + constant->type()->get_key().type_decode(num_bits, basic_type); + + std::list init_list = constant->get_initializer(); + int nbytes_written = 0; + for (std::list::iterator i = init_list.begin(); + i != init_list.end(); i++) { + operand_info op = *i; + ptx_reg_t value = op.get_literal_value(); + int nbytes = num_bits / 8; + switch (op.get_type()) { + case int_t: + assert(nbytes >= 1); + break; + case float_op_t: + assert(nbytes == 4); + break; + case double_op_t: + assert(nbytes >= 4); + break; // account for double DEMOTING + default: + abort(); + } + unsigned addr = constant->get_address() + nbytes_written; + assert(addr + nbytes < min_gaddr); + + gpu->get_global_memory()->write( + addr, nbytes, &value, NULL, + NULL); // assume little endian (so u8 is the first byte in u32) + nc_bytes += nbytes; + nbytes_written += nbytes; + } } - - return result; + } + printf(" done.\n"); + fflush(stdout); + return nc_bytes; +} + +kernel_info_t *cuda_runtime_api::gpgpu_cuda_ptx_sim_init_grid( + const char *hostFun, gpgpu_ptx_sim_arg_list_t args, struct dim3 gridDim, + struct dim3 blockDim, CUctx_st *context) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + function_info *entry = context->get_kernel(hostFun); + gpgpu_t *gpu = context->get_device()->get_gpgpu(); + /* + Passing a snapshot of the GPU's current texture mapping to the kernel's info + as kernels should use texture bindings present at the time of their launch. + */ + kernel_info_t *result = + new kernel_info_t(gridDim, blockDim, entry, gpu->getNameArrayMapping(), + gpu->getNameInfoMapping()); + if (entry == NULL) { + printf( + "GPGPU-Sim PTX: ERROR launching kernel -- no PTX implementation found " + "for %p\n", + hostFun); + abort(); + } + unsigned argcount = args.size(); + unsigned argn = 1; + for (gpgpu_ptx_sim_arg_list_t::iterator a = args.begin(); a != args.end(); + a++) { + entry->add_param_data(argcount - argn, &(*a)); + argn++; + } + + entry->finalize(result->get_param_memory()); + gpgpu_ctx->func_sim->g_ptx_kernel_count++; + fflush(stdout); + + if (g_debug_execution >= 4) { + entry->ptx_jit_config(g_mallocPtr_Size, result->get_param_memory(), + (gpgpu_t *)context->get_device()->get_gpgpu(), + gridDim, blockDim); + } + + return result; } /******************************************************************************* @@ -3945,2934 +4084,2912 @@ kernel_info_t * cuda_runtime_api::gpgpu_cuda_ptx_sim_init_grid( const char *host *******************************************************************************/ //***extra api for pytorch*** -CUresult CUDAAPI cuGetErrorString(CUresult error, const char **pStr) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; -} - -CUresult CUDAAPI cuGetErrorName(CUresult error, const char **pStr) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; -} - -CUresult CUDAAPI cuInit(unsigned int Flags) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; -} - -CUresult CUDAAPI cuDriverGetVersion(int *driverVersion) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - cudaError_t e = cudaDriverGetVersion(driverVersion); - assert(e == cudaSuccess); - return CUDA_SUCCESS; -} - -CUresult CUDAAPI cuDeviceGet(CUdevice *device, int ordinal) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - int deviceI = -1; - cudaError_t e = cudaGetDevice(&deviceI); - assert(e == cudaSuccess); - assert(deviceI!=-1); - *device = deviceI; - return CUDA_SUCCESS; -} - -CUresult CUDAAPI cuDeviceGetCount(int *count) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - cudaError_t e = cudaGetDeviceCount(count); - assert(e == cudaSuccess); - return CUDA_SUCCESS; -} - -CUresult CUDAAPI cuDeviceGetName(char *name, int len, CUdevice dev) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - assert(len>=10); - strcpy(name, "GPGPU-Sim"); - return CUDA_SUCCESS; +CUresult CUDAAPI cuGetErrorString(CUresult error, const char **pStr) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; +} + +CUresult CUDAAPI cuGetErrorName(CUresult error, const char **pStr) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; +} + +CUresult CUDAAPI cuInit(unsigned int Flags) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; +} + +CUresult CUDAAPI cuDriverGetVersion(int *driverVersion) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + cudaError_t e = cudaDriverGetVersion(driverVersion); + assert(e == cudaSuccess); + return CUDA_SUCCESS; +} + +CUresult CUDAAPI cuDeviceGet(CUdevice *device, int ordinal) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + int deviceI = -1; + cudaError_t e = cudaGetDevice(&deviceI); + assert(e == cudaSuccess); + assert(deviceI != -1); + *device = deviceI; + return CUDA_SUCCESS; +} + +CUresult CUDAAPI cuDeviceGetCount(int *count) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + cudaError_t e = cudaGetDeviceCount(count); + assert(e == cudaSuccess); + return CUDA_SUCCESS; +} + +CUresult CUDAAPI cuDeviceGetName(char *name, int len, CUdevice dev) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + assert(len >= 10); + strcpy(name, "GPGPU-Sim"); + return CUDA_SUCCESS; } #if CUDART_VERSION >= 3020 -CUresult CUDAAPI cuDeviceTotalMem(size_t *bytes, CUdevice dev) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - *bytes = 20000000000;//dummy value - return CUDA_SUCCESS; +CUresult CUDAAPI cuDeviceTotalMem(size_t *bytes, CUdevice dev) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + *bytes = 20000000000; // dummy value + return CUDA_SUCCESS; } #endif /* CUDART_VERSION >= 3020 */ #if (CUDART_VERSION > 5000) -CUresult CUDAAPI cuDeviceGetAttribute(int *pi, CUdevice_attribute attrib, CUdevice dev) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - cudaError_t e = cudaDeviceGetAttribute(pi, (cudaDeviceAttr)attrib, dev); - assert(e == cudaSuccess); +CUresult CUDAAPI cuDeviceGetAttribute(int *pi, CUdevice_attribute attrib, + CUdevice dev) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + cudaError_t e = cudaDeviceGetAttribute(pi, (cudaDeviceAttr)attrib, dev); + assert(e == cudaSuccess); - return CUDA_SUCCESS; + return CUDA_SUCCESS; } #endif -CUresult CUDAAPI cuDeviceGetProperties(CUdevprop *prop, CUdevice dev) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; +CUresult CUDAAPI cuDeviceGetProperties(CUdevprop *prop, CUdevice dev) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; } -CUresult CUDAAPI cuDeviceComputeCapability(int *major, int *minor, CUdevice dev) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; +CUresult CUDAAPI cuDeviceComputeCapability(int *major, int *minor, + CUdevice dev) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; } #if CUDART_VERSION >= 7000 -CUresult CUDAAPI cuDevicePrimaryCtxRetain(CUcontext *pctx, CUdevice dev) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; +CUresult CUDAAPI cuDevicePrimaryCtxRetain(CUcontext *pctx, CUdevice dev) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; } -CUresult CUDAAPI cuDevicePrimaryCtxRelease(CUdevice dev) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; +CUresult CUDAAPI cuDevicePrimaryCtxRelease(CUdevice dev) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; } -CUresult CUDAAPI cuDevicePrimaryCtxSetFlags(CUdevice dev, unsigned int flags) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; +CUresult CUDAAPI cuDevicePrimaryCtxSetFlags(CUdevice dev, unsigned int flags) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; } -CUresult CUDAAPI cuDevicePrimaryCtxGetState(CUdevice dev, unsigned int *flags, int *active) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; +CUresult CUDAAPI cuDevicePrimaryCtxGetState(CUdevice dev, unsigned int *flags, + int *active) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; } -CUresult CUDAAPI cuDevicePrimaryCtxReset(CUdevice dev) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; +CUresult CUDAAPI cuDevicePrimaryCtxReset(CUdevice dev) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; } #endif /* CUDART_VERSION >= 7000 */ #if CUDART_VERSION >= 3020 -CUresult CUDAAPI cuCtxCreate(CUcontext *pctx, unsigned int flags, CUdevice dev) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; +CUresult CUDAAPI cuCtxCreate(CUcontext *pctx, unsigned int flags, + CUdevice dev) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; } #endif /* CUDART_VERSION >= 3020 */ #if CUDART_VERSION >= 4000 -CUresult CUDAAPI cuCtxDestroy(CUcontext ctx) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; +CUresult CUDAAPI cuCtxDestroy(CUcontext ctx) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; } #endif /* CUDART_VERSION >= 4000 */ #if CUDART_VERSION >= 4000 -CUresult CUDAAPI cuCtxPushCurrent(CUcontext ctx) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; +CUresult CUDAAPI cuCtxPushCurrent(CUcontext ctx) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; } -CUresult CUDAAPI cuCtxPopCurrent(CUcontext *pctx) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; +CUresult CUDAAPI cuCtxPopCurrent(CUcontext *pctx) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; } -CUresult CUDAAPI cuCtxSetCurrent(CUcontext ctx) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; +CUresult CUDAAPI cuCtxSetCurrent(CUcontext ctx) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; } -CUresult CUDAAPI cuCtxGetCurrent(CUcontext *pctx) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; +CUresult CUDAAPI cuCtxGetCurrent(CUcontext *pctx) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; } #endif /* CUDART_VERSION >= 4000 */ -CUresult CUDAAPI cuCtxGetDevice(CUdevice *device) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; +CUresult CUDAAPI cuCtxGetDevice(CUdevice *device) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; } #if CUDART_VERSION >= 7000 -CUresult CUDAAPI cuCtxGetFlags(unsigned int *flags) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; +CUresult CUDAAPI cuCtxGetFlags(unsigned int *flags) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; } #endif /* CUDART_VERSION >= 7000 */ -CUresult CUDAAPI cuCtxSynchronize(void) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; +CUresult CUDAAPI cuCtxSynchronize(void) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; } -CUresult CUDAAPI cuCtxSetLimit(CUlimit limit, size_t value) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; +CUresult CUDAAPI cuCtxSetLimit(CUlimit limit, size_t value) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; } -CUresult CUDAAPI cuCtxGetLimit(size_t *pvalue, CUlimit limit) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; +CUresult CUDAAPI cuCtxGetLimit(size_t *pvalue, CUlimit limit) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; } -CUresult CUDAAPI cuCtxGetCacheConfig(CUfunc_cache *pconfig) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; +CUresult CUDAAPI cuCtxGetCacheConfig(CUfunc_cache *pconfig) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; } -CUresult CUDAAPI cuCtxSetCacheConfig(CUfunc_cache config) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; +CUresult CUDAAPI cuCtxSetCacheConfig(CUfunc_cache config) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; } #if CUDART_VERSION >= 4020 -CUresult CUDAAPI cuCtxGetSharedMemConfig(CUsharedconfig *pConfig) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; +CUresult CUDAAPI cuCtxGetSharedMemConfig(CUsharedconfig *pConfig) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; } -CUresult CUDAAPI cuCtxSetSharedMemConfig(CUsharedconfig config) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; +CUresult CUDAAPI cuCtxSetSharedMemConfig(CUsharedconfig config) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; } #endif -CUresult CUDAAPI cuCtxGetApiVersion(CUcontext ctx, unsigned int *version) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; +CUresult CUDAAPI cuCtxGetApiVersion(CUcontext ctx, unsigned int *version) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; } -CUresult CUDAAPI cuCtxGetStreamPriorityRange(int *leastPriority, int *greatestPriority) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; +CUresult CUDAAPI cuCtxGetStreamPriorityRange(int *leastPriority, + int *greatestPriority) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; } -CUresult CUDAAPI cuCtxAttach(CUcontext *pctx, unsigned int flags) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; +CUresult CUDAAPI cuCtxAttach(CUcontext *pctx, unsigned int flags) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; } -CUresult CUDAAPI cuCtxDetach(CUcontext ctx) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; +CUresult CUDAAPI cuCtxDetach(CUcontext ctx) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; } -CUresult CUDAAPI cuModuleLoad(CUmodule *module, const char *fname) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; +CUresult CUDAAPI cuModuleLoad(CUmodule *module, const char *fname) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; } -CUresult CUDAAPI cuModuleLoadData(CUmodule *module, const void *image) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; +CUresult CUDAAPI cuModuleLoadData(CUmodule *module, const void *image) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; } -CUresult CUDAAPI cuModuleLoadDataEx(CUmodule *module, const void *image, unsigned int numOptions, CUjit_option *options, void **optionValues) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; +CUresult CUDAAPI cuModuleLoadDataEx(CUmodule *module, const void *image, + unsigned int numOptions, + CUjit_option *options, + void **optionValues) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; } -CUresult CUDAAPI cuModuleLoadFatBinary(CUmodule *module, const void *fatCubin) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; +CUresult CUDAAPI cuModuleLoadFatBinary(CUmodule *module, const void *fatCubin) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; } -CUresult CUDAAPI cuModuleUnload(CUmodule hmod) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; +CUresult CUDAAPI cuModuleUnload(CUmodule hmod) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; } -CUresult CUDAAPI cuModuleGetFunction(CUfunction *hfunc, CUmodule hmod, const char *name) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; +CUresult CUDAAPI cuModuleGetFunction(CUfunction *hfunc, CUmodule hmod, + const char *name) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; } #if CUDART_VERSION >= 3020 -CUresult CUDAAPI cuModuleGetGlobal(CUdeviceptr *dptr, size_t *bytes, CUmodule hmod, const char *name) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; +CUresult CUDAAPI cuModuleGetGlobal(CUdeviceptr *dptr, size_t *bytes, + CUmodule hmod, const char *name) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; } #endif /* CUDART_VERSION >= 3020 */ -CUresult CUDAAPI cuModuleGetTexRef(CUtexref *pTexRef, CUmodule hmod, const char *name) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; +CUresult CUDAAPI cuModuleGetTexRef(CUtexref *pTexRef, CUmodule hmod, + const char *name) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; } -CUresult CUDAAPI cuModuleGetSurfRef(CUsurfref *pSurfRef, CUmodule hmod, const char *name) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; +CUresult CUDAAPI cuModuleGetSurfRef(CUsurfref *pSurfRef, CUmodule hmod, + const char *name) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; } #if CUDART_VERSION >= 6050 -CUresult CUDAAPI -cuLinkCreate(unsigned int numOptions, CUjit_option *options, void **optionValues, CUlinkState *stateOut) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - //currently do not support options or multiple CUlinkStates - return CUDA_SUCCESS; -} - -CUresult CUDAAPI -cuLinkAddData(CUlinkState state, CUjitInputType type, void *data, size_t size, const char *name, - unsigned int numOptions, CUjit_option *options, void **optionValues) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - assert(type==CU_JIT_INPUT_PTX); - cuda_not_implemented(__my_func__,__LINE__); - return CUDA_ERROR_UNKNOWN; -} - -CUresult CUDAAPI -cuLinkAddFile(CUlinkState state, CUjitInputType type, const char *path, - unsigned int numOptions, CUjit_option *options, void **optionValues) -{ - return cuLinkAddFileInternal(state, type, path, - numOptions, options, optionValues); +CUresult CUDAAPI cuLinkCreate(unsigned int numOptions, CUjit_option *options, + void **optionValues, CUlinkState *stateOut) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + // currently do not support options or multiple CUlinkStates + return CUDA_SUCCESS; +} + +CUresult CUDAAPI cuLinkAddData(CUlinkState state, CUjitInputType type, + void *data, size_t size, const char *name, + unsigned int numOptions, CUjit_option *options, + void **optionValues) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + assert(type == CU_JIT_INPUT_PTX); + cuda_not_implemented(__my_func__, __LINE__); + return CUDA_ERROR_UNKNOWN; +} + +CUresult CUDAAPI cuLinkAddFile(CUlinkState state, CUjitInputType type, + const char *path, unsigned int numOptions, + CUjit_option *options, void **optionValues) { + return cuLinkAddFileInternal(state, type, path, numOptions, options, + optionValues); } #endif #if CUDART_VERSION >= 5050 -CUresult CUDAAPI -cuLinkComplete(CUlinkState state, void **cubinOut, size_t *sizeOut) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - //all cuLink* function are implemented to block until completion so nothing to do here - return CUDA_SUCCESS; +CUresult CUDAAPI cuLinkComplete(CUlinkState state, void **cubinOut, + size_t *sizeOut) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + // all cuLink* function are implemented to block until completion so nothing + // to do here + return CUDA_SUCCESS; } -CUresult CUDAAPI -cuLinkDestroy(CUlinkState state) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - //currently do not support options or multiple CUlinkStates - return CUDA_SUCCESS; +CUresult CUDAAPI cuLinkDestroy(CUlinkState state) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + // currently do not support options or multiple CUlinkStates + return CUDA_SUCCESS; } #endif /* CUDART_VERSION >= 5050 */ #if CUDART_VERSION >= 3020 -CUresult CUDAAPI cuMemGetInfo(size_t *free, size_t *total) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; -} - -CUresult CUDAAPI cuMemAlloc(CUdeviceptr *dptr, size_t bytesize) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; -} - -CUresult CUDAAPI cuMemAllocPitch(CUdeviceptr *dptr, size_t *pPitch, size_t WidthInBytes, size_t Height, unsigned int ElementSizeBytes) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; -} - -CUresult CUDAAPI cuMemFree(CUdeviceptr dptr) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; -} - -CUresult CUDAAPI cuMemGetAddressRange(CUdeviceptr *pbase, size_t *psize, CUdeviceptr dptr) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; -} - -CUresult CUDAAPI cuMemAllocHost(void **pp, size_t bytesize) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; +CUresult CUDAAPI cuMemGetInfo(size_t *free, size_t *total) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; +} + +CUresult CUDAAPI cuMemAlloc(CUdeviceptr *dptr, size_t bytesize) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; +} + +CUresult CUDAAPI cuMemAllocPitch(CUdeviceptr *dptr, size_t *pPitch, + size_t WidthInBytes, size_t Height, + unsigned int ElementSizeBytes) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; +} + +CUresult CUDAAPI cuMemFree(CUdeviceptr dptr) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; +} + +CUresult CUDAAPI cuMemGetAddressRange(CUdeviceptr *pbase, size_t *psize, + CUdeviceptr dptr) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; +} + +CUresult CUDAAPI cuMemAllocHost(void **pp, size_t bytesize) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; } #endif /* CUDART_VERSION >= 3020 */ -CUresult CUDAAPI cuMemFreeHost(void *p) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; +CUresult CUDAAPI cuMemFreeHost(void *p) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; } -CUresult CUDAAPI cuMemHostAlloc(void **pp, size_t bytesize, unsigned int Flags) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; +CUresult CUDAAPI cuMemHostAlloc(void **pp, size_t bytesize, + unsigned int Flags) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; } #if CUDART_VERSION >= 3020 -CUresult CUDAAPI cuMemHostGetDevicePointer(CUdeviceptr *pdptr, void *p, unsigned int Flags) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; +CUresult CUDAAPI cuMemHostGetDevicePointer(CUdeviceptr *pdptr, void *p, + unsigned int Flags) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; } #endif /* CUDART_VERSION >= 3020 */ -CUresult CUDAAPI cuMemHostGetFlags(unsigned int *pFlags, void *p) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; +CUresult CUDAAPI cuMemHostGetFlags(unsigned int *pFlags, void *p) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; } #if CUDART_VERSION >= 6000 -CUresult CUDAAPI cuMemAllocManaged(CUdeviceptr *dptr, size_t bytesize, unsigned int flags) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; +CUresult CUDAAPI cuMemAllocManaged(CUdeviceptr *dptr, size_t bytesize, + unsigned int flags) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; } #endif /* CUDART_VERSION >= 6000 */ #if CUDART_VERSION >= 4010 -CUresult CUDAAPI cuDeviceGetByPCIBusId(CUdevice *dev, const char *pciBusId) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; +CUresult CUDAAPI cuDeviceGetByPCIBusId(CUdevice *dev, const char *pciBusId) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; } -CUresult CUDAAPI cuDeviceGetPCIBusId(char *pciBusId, int len, CUdevice dev) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; +CUresult CUDAAPI cuDeviceGetPCIBusId(char *pciBusId, int len, CUdevice dev) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; } -CUresult CUDAAPI cuIpcGetEventHandle(CUipcEventHandle *pHandle, CUevent event) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; +CUresult CUDAAPI cuIpcGetEventHandle(CUipcEventHandle *pHandle, CUevent event) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; } -CUresult CUDAAPI cuIpcOpenEventHandle(CUevent *phEvent, CUipcEventHandle handle) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; +CUresult CUDAAPI cuIpcOpenEventHandle(CUevent *phEvent, + CUipcEventHandle handle) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; } -CUresult CUDAAPI cuIpcGetMemHandle(CUipcMemHandle *pHandle, CUdeviceptr dptr) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; +CUresult CUDAAPI cuIpcGetMemHandle(CUipcMemHandle *pHandle, CUdeviceptr dptr) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; } -CUresult CUDAAPI cuIpcOpenMemHandle(CUdeviceptr *pdptr, CUipcMemHandle handle, unsigned int Flags) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; +CUresult CUDAAPI cuIpcOpenMemHandle(CUdeviceptr *pdptr, CUipcMemHandle handle, + unsigned int Flags) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; } -CUresult CUDAAPI cuIpcCloseMemHandle(CUdeviceptr dptr) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; +CUresult CUDAAPI cuIpcCloseMemHandle(CUdeviceptr dptr) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; } #endif /* CUDART_VERSION >= 4010 */ #if CUDART_VERSION >= 6050 -CUresult CUDAAPI cuMemHostRegister(void *p, size_t bytesize, unsigned int Flags) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; -} -__host__ cudaError_t cudaHostRegister(void* ptr, size_t size, unsigned int flags) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return g_last_cudaError = cudaSuccess; -} - -__host__ cudaError_t cudaProfilerStart ( ) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return g_last_cudaError = cudaSuccess; -} - -__host__ cudaError_t cudaProfilerStop ( ) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return g_last_cudaError = cudaSuccess; +CUresult CUDAAPI cuMemHostRegister(void *p, size_t bytesize, + unsigned int Flags) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; +} +__host__ cudaError_t cudaHostRegister(void *ptr, size_t size, + unsigned int flags) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return g_last_cudaError = cudaSuccess; +} + +__host__ cudaError_t cudaProfilerStart() { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return g_last_cudaError = cudaSuccess; +} + +__host__ cudaError_t cudaProfilerStop() { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return g_last_cudaError = cudaSuccess; } #endif #if CUDART_VERSION >= 4000 -CUresult CUDAAPI cuMemHostUnregister(void *p) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; +CUresult CUDAAPI cuMemHostUnregister(void *p) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; } -CUresult CUDAAPI cuMemcpy(CUdeviceptr dst, CUdeviceptr src, size_t ByteCount) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; +CUresult CUDAAPI cuMemcpy(CUdeviceptr dst, CUdeviceptr src, size_t ByteCount) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; } -CUresult CUDAAPI cuMemcpyPeer(CUdeviceptr dstDevice, CUcontext dstContext, CUdeviceptr srcDevice, CUcontext srcContext, size_t ByteCount) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; +CUresult CUDAAPI cuMemcpyPeer(CUdeviceptr dstDevice, CUcontext dstContext, + CUdeviceptr srcDevice, CUcontext srcContext, + size_t ByteCount) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; } #endif /* CUDART_VERSION >= 4000 */ #if CUDART_VERSION >= 3020 -CUresult CUDAAPI cuMemcpyHtoD(CUdeviceptr dstDevice, const void *srcHost, size_t ByteCount) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; +CUresult CUDAAPI cuMemcpyHtoD(CUdeviceptr dstDevice, const void *srcHost, + size_t ByteCount) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; +} + +CUresult CUDAAPI cuMemcpyDtoH(void *dstHost, CUdeviceptr srcDevice, + size_t ByteCount) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; +} + +CUresult CUDAAPI cuMemcpyDtoD(CUdeviceptr dstDevice, CUdeviceptr srcDevice, + size_t ByteCount) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; +} + +CUresult CUDAAPI cuMemcpyDtoA(CUarray dstArray, size_t dstOffset, + CUdeviceptr srcDevice, size_t ByteCount) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; +} + +CUresult CUDAAPI cuMemcpyAtoD(CUdeviceptr dstDevice, CUarray srcArray, + size_t srcOffset, size_t ByteCount) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; +} + +CUresult CUDAAPI cuMemcpyHtoA(CUarray dstArray, size_t dstOffset, + const void *srcHost, size_t ByteCount) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; +} + +CUresult CUDAAPI cuMemcpyAtoH(void *dstHost, CUarray srcArray, size_t srcOffset, + size_t ByteCount) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; +} + +CUresult CUDAAPI cuMemcpyAtoA(CUarray dstArray, size_t dstOffset, + CUarray srcArray, size_t srcOffset, + size_t ByteCount) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; +} + +CUresult CUDAAPI cuMemcpy2D(const CUDA_MEMCPY2D *pCopy) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; +} + +CUresult CUDAAPI cuMemcpy2DUnaligned(const CUDA_MEMCPY2D *pCopy) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; +} + +CUresult CUDAAPI cuMemcpy3D(const CUDA_MEMCPY3D *pCopy) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; } +#endif /* CUDART_VERSION >= 3020 */ -CUresult CUDAAPI cuMemcpyDtoH(void *dstHost, CUdeviceptr srcDevice, size_t ByteCount) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; +#if CUDART_VERSION >= 4000 +CUresult CUDAAPI cuMemcpy3DPeer(const CUDA_MEMCPY3D_PEER *pCopy) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; +} + +CUresult CUDAAPI cuMemcpyAsync(CUdeviceptr dst, CUdeviceptr src, + size_t ByteCount, CUstream hStream) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; +} + +CUresult CUDAAPI cuMemcpyPeerAsync(CUdeviceptr dstDevice, CUcontext dstContext, + CUdeviceptr srcDevice, CUcontext srcContext, + size_t ByteCount, CUstream hStream) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; } +#endif /* CUDART_VERSION >= 4000 */ -CUresult CUDAAPI cuMemcpyDtoD(CUdeviceptr dstDevice, CUdeviceptr srcDevice, size_t ByteCount) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; +#if CUDART_VERSION >= 3020 +CUresult CUDAAPI cuMemcpyHtoDAsync(CUdeviceptr dstDevice, const void *srcHost, + size_t ByteCount, CUstream hStream) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; +} + +CUresult CUDAAPI cuMemcpyDtoHAsync(void *dstHost, CUdeviceptr srcDevice, + size_t ByteCount, CUstream hStream) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; +} + +CUresult CUDAAPI cuMemcpyDtoDAsync(CUdeviceptr dstDevice, CUdeviceptr srcDevice, + size_t ByteCount, CUstream hStream) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; +} + +CUresult CUDAAPI cuMemcpyHtoAAsync(CUarray dstArray, size_t dstOffset, + const void *srcHost, size_t ByteCount, + CUstream hStream) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; +} + +CUresult CUDAAPI cuMemcpyAtoHAsync(void *dstHost, CUarray srcArray, + size_t srcOffset, size_t ByteCount, + CUstream hStream) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; +} + +CUresult CUDAAPI cuMemcpy2DAsync(const CUDA_MEMCPY2D *pCopy, CUstream hStream) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; +} + +CUresult CUDAAPI cuMemcpy3DAsync(const CUDA_MEMCPY3D *pCopy, CUstream hStream) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; } +#endif /* CUDART_VERSION >= 3020 */ -CUresult CUDAAPI cuMemcpyDtoA(CUarray dstArray, size_t dstOffset, CUdeviceptr srcDevice, size_t ByteCount) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; +#if CUDART_VERSION >= 4000 +CUresult CUDAAPI cuMemcpy3DPeerAsync(const CUDA_MEMCPY3D_PEER *pCopy, + CUstream hStream) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; } +#endif /* CUDART_VERSION >= 4000 */ -CUresult CUDAAPI cuMemcpyAtoD(CUdeviceptr dstDevice, CUarray srcArray, size_t srcOffset, size_t ByteCount) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; +#if CUDART_VERSION >= 3020 +CUresult CUDAAPI cuMemsetD8(CUdeviceptr dstDevice, unsigned char uc, size_t N) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; +} + +CUresult CUDAAPI cuMemsetD16(CUdeviceptr dstDevice, unsigned short us, + size_t N) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; +} + +CUresult CUDAAPI cuMemsetD32(CUdeviceptr dstDevice, unsigned int ui, size_t N) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; +} + +CUresult CUDAAPI cuMemsetD2D8(CUdeviceptr dstDevice, size_t dstPitch, + unsigned char uc, size_t Width, size_t Height) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; +} + +CUresult CUDAAPI cuMemsetD2D16(CUdeviceptr dstDevice, size_t dstPitch, + unsigned short us, size_t Width, size_t Height) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; +} + +CUresult CUDAAPI cuMemsetD2D32(CUdeviceptr dstDevice, size_t dstPitch, + unsigned int ui, size_t Width, size_t Height) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; +} + +CUresult CUDAAPI cuMemsetD8Async(CUdeviceptr dstDevice, unsigned char uc, + size_t N, CUstream hStream) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; +} + +CUresult CUDAAPI cuMemsetD16Async(CUdeviceptr dstDevice, unsigned short us, + size_t N, CUstream hStream) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; +} + +CUresult CUDAAPI cuMemsetD32Async(CUdeviceptr dstDevice, unsigned int ui, + size_t N, CUstream hStream) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; +} + +CUresult CUDAAPI cuMemsetD2D8Async(CUdeviceptr dstDevice, size_t dstPitch, + unsigned char uc, size_t Width, + size_t Height, CUstream hStream) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; +} + +CUresult CUDAAPI cuMemsetD2D16Async(CUdeviceptr dstDevice, size_t dstPitch, + unsigned short us, size_t Width, + size_t Height, CUstream hStream) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; +} + +CUresult CUDAAPI cuMemsetD2D32Async(CUdeviceptr dstDevice, size_t dstPitch, + unsigned int ui, size_t Width, + size_t Height, CUstream hStream) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; +} + +CUresult CUDAAPI cuArrayCreate(CUarray *pHandle, + const CUDA_ARRAY_DESCRIPTOR *pAllocateArray) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; +} + +CUresult CUDAAPI cuArrayGetDescriptor(CUDA_ARRAY_DESCRIPTOR *pArrayDescriptor, + CUarray hArray) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; } +#endif /* CUDART_VERSION >= 3020 */ -CUresult CUDAAPI cuMemcpyHtoA(CUarray dstArray, size_t dstOffset, const void *srcHost, size_t ByteCount) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; +CUresult CUDAAPI cuArrayDestroy(CUarray hArray) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; } -CUresult CUDAAPI cuMemcpyAtoH(void *dstHost, CUarray srcArray, size_t srcOffset, size_t ByteCount) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; +#if CUDART_VERSION >= 3020 +CUresult CUDAAPI cuArray3DCreate( + CUarray *pHandle, const CUDA_ARRAY3D_DESCRIPTOR *pAllocateArray) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; +} + +CUresult CUDAAPI cuArray3DGetDescriptor( + CUDA_ARRAY3D_DESCRIPTOR *pArrayDescriptor, CUarray hArray) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; } +#endif /* CUDART_VERSION >= 3020 */ -CUresult CUDAAPI cuMemcpyAtoA(CUarray dstArray, size_t dstOffset, CUarray srcArray, size_t srcOffset, size_t ByteCount) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; +#if CUDART_VERSION >= 5000 + +CUresult CUDAAPI +cuMipmappedArrayCreate(CUmipmappedArray *pHandle, + const CUDA_ARRAY3D_DESCRIPTOR *pMipmappedArrayDesc, + unsigned int numMipmapLevels) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; +} + +CUresult CUDAAPI cuMipmappedArrayGetLevel(CUarray *pLevelArray, + CUmipmappedArray hMipmappedArray, + unsigned int level) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; +} + +CUresult CUDAAPI cuMipmappedArrayDestroy(CUmipmappedArray hMipmappedArray) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; } -CUresult CUDAAPI cuMemcpy2D(const CUDA_MEMCPY2D *pCopy) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; +#endif /* CUDART_VERSION >= 5000 */ + +/** @} */ /* END CUDA_MEM */ + +#if CUDART_VERSION >= 4000 +CUresult CUDAAPI cuPointerGetAttribute(void *data, + CUpointer_attribute attribute, + CUdeviceptr ptr) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; } +#endif /* CUDART_VERSION >= 4000 */ -CUresult CUDAAPI cuMemcpy2DUnaligned(const CUDA_MEMCPY2D *pCopy) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; +#if CUDART_VERSION >= 8000 +__host__ cudaError_t CUDARTAPI cudaCreateTextureObject( + cudaTextureObject_t *pTexObject, const cudaResourceDesc *pResDesc, + const cudaTextureDesc *pTexDesc, const cudaResourceViewDesc *pResViewDesc) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + cuda_not_implemented(__my_func__, __LINE__); + return g_last_cudaError = cudaSuccess; +} + +CUresult CUDAAPI cuMemPrefetchAsync(CUdeviceptr devPtr, size_t count, + CUdevice dstDevice, CUstream hStream) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; +} + +CUresult CUDAAPI cuMemAdvise(CUdeviceptr devPtr, size_t count, + CUmem_advise advice, CUdevice device) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; +} + +CUresult CUDAAPI cuMemRangeGetAttribute(void *data, size_t dataSize, + CUmem_range_attribute attribute, + CUdeviceptr devPtr, size_t count) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; +} + +CUresult CUDAAPI cuMemRangeGetAttributes(void **data, size_t *dataSizes, + CUmem_range_attribute *attributes, + size_t numAttributes, + CUdeviceptr devPtr, size_t count) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; } +#endif /* CUDART_VERSION >= 8000 */ -CUresult CUDAAPI cuMemcpy3D(const CUDA_MEMCPY3D *pCopy) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; +#if CUDART_VERSION >= 6000 +CUresult CUDAAPI cuPointerSetAttribute(const void *value, + CUpointer_attribute attribute, + CUdeviceptr ptr) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; } -#endif /* CUDART_VERSION >= 3020 */ +#endif /* CUDART_VERSION >= 6000 */ -#if CUDART_VERSION >= 4000 -CUresult CUDAAPI cuMemcpy3DPeer(const CUDA_MEMCPY3D_PEER *pCopy) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; +#if CUDART_VERSION >= 7000 +CUresult CUDAAPI cuPointerGetAttributes(unsigned int numAttributes, + CUpointer_attribute *attributes, + void **data, CUdeviceptr ptr) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; } +#endif /* CUDART_VERSION >= 7000 */ -CUresult CUDAAPI cuMemcpyAsync(CUdeviceptr dst, CUdeviceptr src, size_t ByteCount, CUstream hStream) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; +/** @} */ /* END CUDA_UNIFIED */ + +CUresult CUDAAPI cuStreamCreate(CUstream *phStream, unsigned int Flags) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; } -CUresult CUDAAPI cuMemcpyPeerAsync(CUdeviceptr dstDevice, CUcontext dstContext, CUdeviceptr srcDevice, CUcontext srcContext, size_t ByteCount, CUstream hStream) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; +CUresult CUDAAPI cuStreamCreateWithPriority(CUstream *phStream, + unsigned int flags, int priority) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; } -#endif /* CUDART_VERSION >= 4000 */ -#if CUDART_VERSION >= 3020 -CUresult CUDAAPI cuMemcpyHtoDAsync(CUdeviceptr dstDevice, const void *srcHost, size_t ByteCount, CUstream hStream) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; +CUresult CUDAAPI cuStreamGetPriority(CUstream hStream, int *priority) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; } -CUresult CUDAAPI cuMemcpyDtoHAsync(void *dstHost, CUdeviceptr srcDevice, size_t ByteCount, CUstream hStream) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; +CUresult CUDAAPI cuStreamGetFlags(CUstream hStream, unsigned int *flags) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; } -CUresult CUDAAPI cuMemcpyDtoDAsync(CUdeviceptr dstDevice, CUdeviceptr srcDevice, size_t ByteCount, CUstream hStream) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; +CUresult CUDAAPI cuStreamWaitEvent(CUstream hStream, CUevent hEvent, + unsigned int Flags) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; } -CUresult CUDAAPI cuMemcpyHtoAAsync(CUarray dstArray, size_t dstOffset, const void *srcHost, size_t ByteCount, CUstream hStream) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; +CUresult CUDAAPI cuStreamAddCallback(CUstream hStream, + CUstreamCallback callback, void *userData, + unsigned int flags) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; } -CUresult CUDAAPI cuMemcpyAtoHAsync(void *dstHost, CUarray srcArray, size_t srcOffset, size_t ByteCount, CUstream hStream) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; +#if CUDART_VERSION >= 6000 + +CUresult CUDAAPI cuStreamAttachMemAsync(CUstream hStream, CUdeviceptr dptr, + size_t length, unsigned int flags) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; } -CUresult CUDAAPI cuMemcpy2DAsync(const CUDA_MEMCPY2D *pCopy, CUstream hStream) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; +#endif /* CUDART_VERSION >= 6000 */ + +CUresult CUDAAPI cuStreamQuery(CUstream hStream) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; } -CUresult CUDAAPI cuMemcpy3DAsync(const CUDA_MEMCPY3D *pCopy, CUstream hStream) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; +CUresult CUDAAPI cuStreamSynchronize(CUstream hStream) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; } -#endif /* CUDART_VERSION >= 3020 */ #if CUDART_VERSION >= 4000 -CUresult CUDAAPI cuMemcpy3DPeerAsync(const CUDA_MEMCPY3D_PEER *pCopy, CUstream hStream) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; +CUresult CUDAAPI cuStreamDestroy(CUstream hStream) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; } #endif /* CUDART_VERSION >= 4000 */ -#if CUDART_VERSION >= 3020 -CUresult CUDAAPI cuMemsetD8(CUdeviceptr dstDevice, unsigned char uc, size_t N) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; +/** @} */ /* END CUDA_STREAM */ + +CUresult CUDAAPI cuEventCreate(CUevent *phEvent, unsigned int Flags) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; } -CUresult CUDAAPI cuMemsetD16(CUdeviceptr dstDevice, unsigned short us, size_t N) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; +CUresult CUDAAPI cuEventRecord(CUevent hEvent, CUstream hStream) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; } -CUresult CUDAAPI cuMemsetD32(CUdeviceptr dstDevice, unsigned int ui, size_t N) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; +CUresult CUDAAPI cuEventQuery(CUevent hEvent) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; } -CUresult CUDAAPI cuMemsetD2D8(CUdeviceptr dstDevice, size_t dstPitch, unsigned char uc, size_t Width, size_t Height) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; +CUresult CUDAAPI cuEventSynchronize(CUevent hEvent) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; } -CUresult CUDAAPI cuMemsetD2D16(CUdeviceptr dstDevice, size_t dstPitch, unsigned short us, size_t Width, size_t Height) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; +#if CUDART_VERSION >= 4000 +CUresult CUDAAPI cuEventDestroy(CUevent hEvent) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; } +#endif /* CUDART_VERSION >= 4000 */ -CUresult CUDAAPI cuMemsetD2D32(CUdeviceptr dstDevice, size_t dstPitch, unsigned int ui, size_t Width, size_t Height) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; +CUresult CUDAAPI cuEventElapsedTime(float *pMilliseconds, CUevent hStart, + CUevent hEnd) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; } -CUresult CUDAAPI cuMemsetD8Async(CUdeviceptr dstDevice, unsigned char uc, size_t N, CUstream hStream) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; +#if CUDART_VERSION >= 8000 +CUresult CUDAAPI cuStreamWaitValue32(CUstream stream, CUdeviceptr addr, + cuuint32_t value, unsigned int flags) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; +} + +CUresult CUDAAPI cuStreamWriteValue32(CUstream stream, CUdeviceptr addr, + cuuint32_t value, unsigned int flags) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; +} + +CUresult CUDAAPI cuStreamBatchMemOp(CUstream stream, unsigned int count, + CUstreamBatchMemOpParams *paramArray, + unsigned int flags) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; } +#endif /* CUDART_VERSION >= 8000 */ -CUresult CUDAAPI cuMemsetD16Async(CUdeviceptr dstDevice, unsigned short us, size_t N, CUstream hStream) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; +/** @} */ /* END CUDA_EVENT */ + +CUresult CUDAAPI cuFuncGetAttribute(int *pi, CUfunction_attribute attrib, + CUfunction hfunc) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; } -CUresult CUDAAPI cuMemsetD32Async(CUdeviceptr dstDevice, unsigned int ui, size_t N, CUstream hStream) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; +CUresult CUDAAPI cuFuncSetCacheConfig(CUfunction hfunc, CUfunc_cache config) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; } -CUresult CUDAAPI cuMemsetD2D8Async(CUdeviceptr dstDevice, size_t dstPitch, unsigned char uc, size_t Width, size_t Height, CUstream hStream) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; +#if CUDART_VERSION >= 4020 +CUresult CUDAAPI cuFuncSetSharedMemConfig(CUfunction hfunc, + CUsharedconfig config) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; } +#endif -CUresult CUDAAPI cuMemsetD2D16Async(CUdeviceptr dstDevice, size_t dstPitch, unsigned short us, size_t Width, size_t Height, CUstream hStream) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; +#if CUDART_VERSION >= 4000 +CUresult CUDAAPI cuLaunchKernel(CUfunction f, unsigned int gridDimX, + unsigned int gridDimY, unsigned int gridDimZ, + unsigned int blockDimX, unsigned int blockDimY, + unsigned int blockDimZ, + unsigned int sharedMemBytes, CUstream hStream, + void **kernelParams, void **extra) { + return cuLaunchKernelInternal(f, gridDimX, gridDimY, gridDimZ, blockDimX, + blockDimY, blockDimZ, sharedMemBytes, hStream, + kernelParams, extra); } +#endif /* CUDART_VERSION >= 4000 */ -CUresult CUDAAPI cuMemsetD2D32Async(CUdeviceptr dstDevice, size_t dstPitch, unsigned int ui, size_t Width, size_t Height, CUstream hStream) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; +/** @} */ /* END CUDA_EXEC */ + +CUresult CUDAAPI cuFuncSetBlockShape(CUfunction hfunc, int x, int y, int z) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; } -CUresult CUDAAPI cuArrayCreate(CUarray *pHandle, const CUDA_ARRAY_DESCRIPTOR *pAllocateArray) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; +CUresult CUDAAPI cuFuncSetSharedSize(CUfunction hfunc, unsigned int bytes) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; } -CUresult CUDAAPI cuArrayGetDescriptor(CUDA_ARRAY_DESCRIPTOR *pArrayDescriptor, CUarray hArray) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; +CUresult CUDAAPI cuParamSetSize(CUfunction hfunc, unsigned int numbytes) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; } -#endif /* CUDART_VERSION >= 3020 */ +CUresult CUDAAPI cuParamSeti(CUfunction hfunc, int offset, unsigned int value) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; +} -CUresult CUDAAPI cuArrayDestroy(CUarray hArray) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; -} - -#if CUDART_VERSION >= 3020 -CUresult CUDAAPI cuArray3DCreate(CUarray *pHandle, const CUDA_ARRAY3D_DESCRIPTOR *pAllocateArray) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; -} - -CUresult CUDAAPI cuArray3DGetDescriptor(CUDA_ARRAY3D_DESCRIPTOR *pArrayDescriptor, CUarray hArray) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; -} -#endif /* CUDART_VERSION >= 3020 */ - -#if CUDART_VERSION >= 5000 - -CUresult CUDAAPI cuMipmappedArrayCreate(CUmipmappedArray *pHandle, const CUDA_ARRAY3D_DESCRIPTOR *pMipmappedArrayDesc, unsigned int numMipmapLevels) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; -} - -CUresult CUDAAPI cuMipmappedArrayGetLevel(CUarray *pLevelArray, CUmipmappedArray hMipmappedArray, unsigned int level) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; -} - -CUresult CUDAAPI cuMipmappedArrayDestroy(CUmipmappedArray hMipmappedArray) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; -} - -#endif /* CUDART_VERSION >= 5000 */ - -/** @} */ /* END CUDA_MEM */ - - -#if CUDART_VERSION >= 4000 -CUresult CUDAAPI cuPointerGetAttribute(void *data, CUpointer_attribute attribute, CUdeviceptr ptr) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; +CUresult CUDAAPI cuParamSetf(CUfunction hfunc, int offset, float value) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; } -#endif /* CUDART_VERSION >= 4000 */ - -#if CUDART_VERSION >= 8000 -__host__ cudaError_t CUDARTAPI cudaCreateTextureObject ( cudaTextureObject_t* pTexObject, const cudaResourceDesc* pResDesc, const cudaTextureDesc* pTexDesc, const cudaResourceViewDesc* pResViewDesc ) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - cuda_not_implemented(__my_func__,__LINE__); - return g_last_cudaError = cudaSuccess; +CUresult CUDAAPI cuParamSetv(CUfunction hfunc, int offset, void *ptr, + unsigned int numbytes) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; } -CUresult CUDAAPI cuMemPrefetchAsync(CUdeviceptr devPtr, size_t count, CUdevice dstDevice, CUstream hStream) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; +CUresult CUDAAPI cuLaunch(CUfunction f) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; } -CUresult CUDAAPI cuMemAdvise(CUdeviceptr devPtr, size_t count, CUmem_advise advice, CUdevice device) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; +CUresult CUDAAPI cuLaunchGrid(CUfunction f, int grid_width, int grid_height) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; } -CUresult CUDAAPI cuMemRangeGetAttribute(void *data, size_t dataSize, CUmem_range_attribute attribute, CUdeviceptr devPtr, size_t count) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; +CUresult CUDAAPI cuLaunchGridAsync(CUfunction f, int grid_width, + int grid_height, CUstream hStream) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; } -CUresult CUDAAPI cuMemRangeGetAttributes(void **data, size_t *dataSizes, CUmem_range_attribute *attributes, size_t numAttributes, CUdeviceptr devPtr, size_t count) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; +CUresult CUDAAPI cuParamSetTexRef(CUfunction hfunc, int texunit, + CUtexref hTexRef) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; } -#endif /* CUDART_VERSION >= 8000 */ +/** @} */ /* END CUDA_EXEC_DEPRECATED */ -#if CUDART_VERSION >= 6000 -CUresult CUDAAPI cuPointerSetAttribute(const void *value, CUpointer_attribute attribute, CUdeviceptr ptr) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; -} -#endif /* CUDART_VERSION >= 6000 */ +#if CUDART_VERSION >= 6050 -#if CUDART_VERSION >= 7000 -CUresult CUDAAPI cuPointerGetAttributes(unsigned int numAttributes, CUpointer_attribute *attributes, void **data, CUdeviceptr ptr) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; +CUresult CUDAAPI cuOccupancyMaxActiveBlocksPerMultiprocessor( + int *numBlocks, CUfunction func, int blockSize, size_t dynamicSMemSize) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; +} + +CUresult CUDAAPI cuOccupancyMaxActiveBlocksPerMultiprocessorWithFlags( + int *numBlocks, CUfunction func, int blockSize, size_t dynamicSMemSize, + unsigned int flags) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; +} + +CUresult CUDAAPI cuOccupancyMaxPotentialBlockSize( + int *minGridSize, int *blockSize, CUfunction func, + CUoccupancyB2DSize blockSizeToDynamicSMemSize, size_t dynamicSMemSize, + int blockSizeLimit) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; +} + +CUresult CUDAAPI cuOccupancyMaxPotentialBlockSizeWithFlags( + int *minGridSize, int *blockSize, CUfunction func, + CUoccupancyB2DSize blockSizeToDynamicSMemSize, size_t dynamicSMemSize, + int blockSizeLimit, unsigned int flags) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; } -#endif /* CUDART_VERSION >= 7000 */ -/** @} */ /* END CUDA_UNIFIED */ - - -CUresult CUDAAPI cuStreamCreate(CUstream *phStream, unsigned int Flags) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; -} +/** @} */ /* END CUDA_OCCUPANCY */ +#endif /* CUDART_VERSION >= 6050 */ -CUresult CUDAAPI cuStreamCreateWithPriority(CUstream *phStream, unsigned int flags, int priority) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; +CUresult CUDAAPI cuTexRefSetArray(CUtexref hTexRef, CUarray hArray, + unsigned int Flags) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; } - -CUresult CUDAAPI cuStreamGetPriority(CUstream hStream, int *priority) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; +CUresult CUDAAPI cuTexRefSetMipmappedArray(CUtexref hTexRef, + CUmipmappedArray hMipmappedArray, + unsigned int Flags) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; } -CUresult CUDAAPI cuStreamGetFlags(CUstream hStream, unsigned int *flags) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; +#if CUDART_VERSION >= 3020 +CUresult CUDAAPI cuTexRefSetAddress(size_t *ByteOffset, CUtexref hTexRef, + CUdeviceptr dptr, size_t bytes) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; +} + +CUresult CUDAAPI cuTexRefSetAddress2D(CUtexref hTexRef, + const CUDA_ARRAY_DESCRIPTOR *desc, + CUdeviceptr dptr, size_t Pitch) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; } +#endif /* CUDART_VERSION >= 3020 */ - -CUresult CUDAAPI cuStreamWaitEvent(CUstream hStream, CUevent hEvent, unsigned int Flags) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; +CUresult CUDAAPI cuTexRefSetFormat(CUtexref hTexRef, CUarray_format fmt, + int NumPackedComponents) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; } -CUresult CUDAAPI cuStreamAddCallback(CUstream hStream, CUstreamCallback callback, void *userData, unsigned int flags) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; +CUresult CUDAAPI cuTexRefSetAddressMode(CUtexref hTexRef, int dim, + CUaddress_mode am) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; } -#if CUDART_VERSION >= 6000 - -CUresult CUDAAPI cuStreamAttachMemAsync(CUstream hStream, CUdeviceptr dptr, size_t length, unsigned int flags) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; +CUresult CUDAAPI cuTexRefSetFilterMode(CUtexref hTexRef, CUfilter_mode fm) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; } -#endif /* CUDART_VERSION >= 6000 */ - -CUresult CUDAAPI cuStreamQuery(CUstream hStream) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; +CUresult CUDAAPI cuTexRefSetMipmapFilterMode(CUtexref hTexRef, + CUfilter_mode fm) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; } -CUresult CUDAAPI cuStreamSynchronize(CUstream hStream) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; +CUresult CUDAAPI cuTexRefSetMipmapLevelBias(CUtexref hTexRef, float bias) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; } -#if CUDART_VERSION >= 4000 -CUresult CUDAAPI cuStreamDestroy(CUstream hStream) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; +CUresult CUDAAPI cuTexRefSetMipmapLevelClamp(CUtexref hTexRef, + float minMipmapLevelClamp, + float maxMipmapLevelClamp) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; } -#endif /* CUDART_VERSION >= 4000 */ - -/** @} */ /* END CUDA_STREAM */ - - -CUresult CUDAAPI cuEventCreate(CUevent *phEvent, unsigned int Flags) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; +CUresult CUDAAPI cuTexRefSetMaxAnisotropy(CUtexref hTexRef, + unsigned int maxAniso) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; } -CUresult CUDAAPI cuEventRecord(CUevent hEvent, CUstream hStream) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; +CUresult CUDAAPI cuTexRefSetBorderColor(CUtexref hTexRef, float *pBorderColor) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; } -CUresult CUDAAPI cuEventQuery(CUevent hEvent) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; +CUresult CUDAAPI cuTexRefSetFlags(CUtexref hTexRef, unsigned int Flags) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; } -CUresult CUDAAPI cuEventSynchronize(CUevent hEvent) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; +#if CUDART_VERSION >= 3020 +CUresult CUDAAPI cuTexRefGetAddress(CUdeviceptr *pdptr, CUtexref hTexRef) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; } +#endif /* CUDART_VERSION >= 3020 */ -#if CUDART_VERSION >= 4000 -CUresult CUDAAPI cuEventDestroy(CUevent hEvent) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; +CUresult CUDAAPI cuTexRefGetArray(CUarray *phArray, CUtexref hTexRef) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; +} + +CUresult CUDAAPI cuTexRefGetMipmappedArray(CUmipmappedArray *phMipmappedArray, + CUtexref hTexRef) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; +} + +CUresult CUDAAPI cuTexRefGetAddressMode(CUaddress_mode *pam, CUtexref hTexRef, + int dim) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; +} + +CUresult CUDAAPI cuTexRefGetFilterMode(CUfilter_mode *pfm, CUtexref hTexRef) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; +} + +CUresult CUDAAPI cuTexRefGetFormat(CUarray_format *pFormat, int *pNumChannels, + CUtexref hTexRef) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; +} + +CUresult CUDAAPI cuTexRefGetMipmapFilterMode(CUfilter_mode *pfm, + CUtexref hTexRef) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; +} + +CUresult CUDAAPI cuTexRefGetMipmapLevelBias(float *pbias, CUtexref hTexRef) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; +} + +CUresult CUDAAPI cuTexRefGetMipmapLevelClamp(float *pminMipmapLevelClamp, + float *pmaxMipmapLevelClamp, + CUtexref hTexRef) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; +} + +CUresult CUDAAPI cuTexRefGetMaxAnisotropy(int *pmaxAniso, CUtexref hTexRef) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; +} + +CUresult CUDAAPI cuTexRefGetBorderColor(float *pBorderColor, CUtexref hTexRef) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; +} + +CUresult CUDAAPI cuTexRefGetFlags(unsigned int *pFlags, CUtexref hTexRef) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; +} + +CUresult CUDAAPI cuTexRefCreate(CUtexref *pTexRef) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; +} + +CUresult CUDAAPI cuTexRefDestroy(CUtexref hTexRef) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; +} + +CUresult CUDAAPI cuSurfRefSetArray(CUsurfref hSurfRef, CUarray hArray, + unsigned int Flags) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; +} + +CUresult CUDAAPI cuSurfRefGetArray(CUarray *phArray, CUsurfref hSurfRef) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; } -#endif /* CUDART_VERSION >= 4000 */ -CUresult CUDAAPI cuEventElapsedTime(float *pMilliseconds, CUevent hStart, CUevent hEnd) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; -} +/** @} */ /* END CUDA_SURFREF */ -#if CUDART_VERSION >= 8000 -CUresult CUDAAPI cuStreamWaitValue32(CUstream stream, CUdeviceptr addr, cuuint32_t value, unsigned int flags) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; +#if CUDART_VERSION >= 5000 +CUresult CUDAAPI +cuTexObjectCreate(CUtexObject *pTexObject, const CUDA_RESOURCE_DESC *pResDesc, + const CUDA_TEXTURE_DESC *pTexDesc, + const CUDA_RESOURCE_VIEW_DESC *pResViewDesc) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; +} + +CUresult CUDAAPI cuTexObjectDestroy(CUtexObject texObject) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; +} + +CUresult CUDAAPI cuTexObjectGetResourceDesc(CUDA_RESOURCE_DESC *pResDesc, + CUtexObject texObject) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; +} + +CUresult CUDAAPI cuTexObjectGetTextureDesc(CUDA_TEXTURE_DESC *pTexDesc, + CUtexObject texObject) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; +} + +CUresult CUDAAPI cuTexObjectGetResourceViewDesc( + CUDA_RESOURCE_VIEW_DESC *pResViewDesc, CUtexObject texObject) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; } -CUresult CUDAAPI cuStreamWriteValue32(CUstream stream, CUdeviceptr addr, cuuint32_t value, unsigned int flags) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; -} +/** @} */ /* END CUDA_TEXOBJECT */ -CUresult CUDAAPI cuStreamBatchMemOp(CUstream stream, unsigned int count, CUstreamBatchMemOpParams *paramArray, unsigned int flags) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; +CUresult CUDAAPI cuSurfObjectCreate(CUsurfObject *pSurfObject, + const CUDA_RESOURCE_DESC *pResDesc) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; } -#endif /* CUDART_VERSION >= 8000 */ - -/** @} */ /* END CUDA_EVENT */ - -CUresult CUDAAPI cuFuncGetAttribute(int *pi, CUfunction_attribute attrib, CUfunction hfunc) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; +CUresult CUDAAPI cuSurfObjectDestroy(CUsurfObject surfObject) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; } -CUresult CUDAAPI cuFuncSetCacheConfig(CUfunction hfunc, CUfunc_cache config) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; +CUresult CUDAAPI cuSurfObjectGetResourceDesc(CUDA_RESOURCE_DESC *pResDesc, + CUsurfObject surfObject) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; } -#if CUDART_VERSION >= 4020 -CUresult CUDAAPI cuFuncSetSharedMemConfig(CUfunction hfunc, CUsharedconfig config) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; -} -#endif +#endif /* CUDART_VERSION >= 5000 */ #if CUDART_VERSION >= 4000 -CUresult CUDAAPI cuLaunchKernel(CUfunction f, - unsigned int gridDimX, - unsigned int gridDimY, - unsigned int gridDimZ, - unsigned int blockDimX, - unsigned int blockDimY, - unsigned int blockDimZ, - unsigned int sharedMemBytes, - CUstream hStream, - void **kernelParams, - void **extra) -{ - return cuLaunchKernelInternal(f, gridDimX, gridDimY, gridDimZ, blockDimX, blockDimY, blockDimZ, sharedMemBytes, hStream, kernelParams, extra); -} -#endif /* CUDART_VERSION >= 4000 */ - -/** @} */ /* END CUDA_EXEC */ - - -CUresult CUDAAPI cuFuncSetBlockShape(CUfunction hfunc, int x, int y, int z) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; -} - -CUresult CUDAAPI cuFuncSetSharedSize(CUfunction hfunc, unsigned int bytes) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; -} - -CUresult CUDAAPI cuParamSetSize(CUfunction hfunc, unsigned int numbytes) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; -} - -CUresult CUDAAPI cuParamSeti(CUfunction hfunc, int offset, unsigned int value) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; -} - -CUresult CUDAAPI cuParamSetf(CUfunction hfunc, int offset, float value) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; -} - -CUresult CUDAAPI cuParamSetv(CUfunction hfunc, int offset, void *ptr, unsigned int numbytes) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; -} - -CUresult CUDAAPI cuLaunch(CUfunction f) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; -} - -CUresult CUDAAPI cuLaunchGrid(CUfunction f, int grid_width, int grid_height) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; -} - -CUresult CUDAAPI cuLaunchGridAsync(CUfunction f, int grid_width, int grid_height, CUstream hStream) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; -} - - -CUresult CUDAAPI cuParamSetTexRef(CUfunction hfunc, int texunit, CUtexref hTexRef) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; -} -/** @} */ /* END CUDA_EXEC_DEPRECATED */ - - -#if CUDART_VERSION >= 6050 - -CUresult CUDAAPI cuOccupancyMaxActiveBlocksPerMultiprocessor(int *numBlocks, CUfunction func, int blockSize, size_t dynamicSMemSize) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; +CUresult CUDAAPI cuDeviceCanAccessPeer(int *canAccessPeer, CUdevice dev, + CUdevice peerDev) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; +} + +CUresult CUDAAPI cuDeviceGetP2PAttribute(int *value, + CUdevice_P2PAttribute attrib, + CUdevice srcDevice, + CUdevice dstDevice) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; +} + +CUresult CUDAAPI cuCtxEnablePeerAccess(CUcontext peerContext, + unsigned int Flags) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; +} + +CUresult CUDAAPI cuCtxDisablePeerAccess(CUcontext peerContext) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; } -CUresult CUDAAPI cuOccupancyMaxActiveBlocksPerMultiprocessorWithFlags(int *numBlocks, CUfunction func, int blockSize, size_t dynamicSMemSize, unsigned int flags) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; -} +/** @} */ /* END CUDA_PEER_ACCESS */ +#endif /* CUDART_VERSION >= 4000 */ -CUresult CUDAAPI cuOccupancyMaxPotentialBlockSize(int *minGridSize, int *blockSize, CUfunction func, CUoccupancyB2DSize blockSizeToDynamicSMemSize, size_t dynamicSMemSize, int blockSizeLimit) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; +CUresult CUDAAPI cuGraphicsUnregisterResource(CUgraphicsResource resource) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; } -CUresult CUDAAPI cuOccupancyMaxPotentialBlockSizeWithFlags(int *minGridSize, int *blockSize, CUfunction func, CUoccupancyB2DSize blockSizeToDynamicSMemSize, size_t dynamicSMemSize, int blockSizeLimit, unsigned int flags) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; +CUresult CUDAAPI cuGraphicsSubResourceGetMappedArray( + CUarray *pArray, CUgraphicsResource resource, unsigned int arrayIndex, + unsigned int mipLevel) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; } -/** @} */ /* END CUDA_OCCUPANCY */ -#endif /* CUDART_VERSION >= 6050 */ +#if CUDART_VERSION >= 5000 -CUresult CUDAAPI cuTexRefSetArray(CUtexref hTexRef, CUarray hArray, unsigned int Flags) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; +CUresult CUDAAPI cuGraphicsResourceGetMappedMipmappedArray( + CUmipmappedArray *pMipmappedArray, CUgraphicsResource resource) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; } -CUresult CUDAAPI cuTexRefSetMipmappedArray(CUtexref hTexRef, CUmipmappedArray hMipmappedArray, unsigned int Flags) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; -} +#endif /* CUDART_VERSION >= 5000 */ #if CUDART_VERSION >= 3020 -CUresult CUDAAPI cuTexRefSetAddress(size_t *ByteOffset, CUtexref hTexRef, CUdeviceptr dptr, size_t bytes) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; -} - -CUresult CUDAAPI cuTexRefSetAddress2D(CUtexref hTexRef, const CUDA_ARRAY_DESCRIPTOR *desc, CUdeviceptr dptr, size_t Pitch) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; +CUresult CUDAAPI cuGraphicsResourceGetMappedPointer( + CUdeviceptr *pDevPtr, size_t *pSize, CUgraphicsResource resource) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; } #endif /* CUDART_VERSION >= 3020 */ -CUresult CUDAAPI cuTexRefSetFormat(CUtexref hTexRef, CUarray_format fmt, int NumPackedComponents) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; -} - -CUresult CUDAAPI cuTexRefSetAddressMode(CUtexref hTexRef, int dim, CUaddress_mode am) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; +CUresult CUDAAPI cuGraphicsResourceSetMapFlags(CUgraphicsResource resource, + unsigned int flags) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; } -CUresult CUDAAPI cuTexRefSetFilterMode(CUtexref hTexRef, CUfilter_mode fm) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; +CUresult CUDAAPI cuGraphicsMapResources(unsigned int count, + CUgraphicsResource *resources, + CUstream hStream) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; } -CUresult CUDAAPI cuTexRefSetMipmapFilterMode(CUtexref hTexRef, CUfilter_mode fm) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; +CUresult CUDAAPI cuGraphicsUnmapResources(unsigned int count, + CUgraphicsResource *resources, + CUstream hStream) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; } -CUresult CUDAAPI cuTexRefSetMipmapLevelBias(CUtexref hTexRef, float bias) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; -} +/** @} */ /* END CUDA_GRAPHICS */ -CUresult CUDAAPI cuTexRefSetMipmapLevelClamp(CUtexref hTexRef, float minMipmapLevelClamp, float maxMipmapLevelClamp) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; -} +CUresult CUDAAPI cuGetExportTable(const void **ppExportTable, + const CUuuid *pExportTableId) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + cudaError_t e = cudaGetExportTable(ppExportTable, pExportTableId); + assert(e == cudaSuccess); + return CUDA_SUCCESS; +} + +#if defined(CUDART_VERSION_INTERNAL) || \ + (CUDART_VERSION >= 4000 && CUDART_VERSION < 6050) +CUresult CUDAAPI cuMemHostRegister(void *p, size_t bytesize, + unsigned int Flags) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; +} +#endif /* defined(CUDART_VERSION_INTERNAL) || (CUDART_VERSION >= 4000 && \ + CUDART_VERSION < 6050) */ + +#if defined(CUDART_VERSION_INTERNAL) || \ + (CUDART_VERSION >= 5050 && CUDART_VERSION < 6050) +CUresult CUDAAPI cuLinkCreate(unsigned int numOptions, CUjit_option *options, + void **optionValues, CUlinkState *stateOut) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; +} +CUresult CUDAAPI cuLinkAddData(CUlinkState state, CUjitInputType type, + void *data, size_t size, const char *name, + unsigned int numOptions, CUjit_option *options, + void **optionValues) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; +} +CUresult CUDAAPI cuLinkAddFile(CUlinkState state, CUjitInputType type, + const char *path, unsigned int numOptions, + CUjit_option *options, void **optionValues) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; +} +#endif /* CUDART_VERSION_INTERNAL || (CUDART_VERSION >= 5050 && CUDART_VERSION \ + < 6050) */ + +#if defined(CUDART_VERSION_INTERNAL) || \ + (CUDART_VERSION >= 3020 && CUDART_VERSION < 4010) +CUresult CUDAAPI cuTexRefSetAddress2D_v2(CUtexref hTexRef, + const CUDA_ARRAY_DESCRIPTOR *desc, + CUdeviceptr dptr, size_t Pitch) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; +} +#endif /* CUDART_VERSION_INTERNAL || (CUDART_VERSION >= 3020 && CUDART_VERSION \ + < 4010) */ -CUresult CUDAAPI cuTexRefSetMaxAnisotropy(CUtexref hTexRef, unsigned int maxAniso) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; +#if defined(CUDART_VERSION_INTERNAL) || CUDART_VERSION < 4000 +CUresult CUDAAPI cuCtxDestroy(CUcontext ctx) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; +} +CUresult CUDAAPI cuCtxPopCurrent(CUcontext *pctx) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; +} +CUresult CUDAAPI cuCtxPushCurrent(CUcontext ctx) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; +} +CUresult CUDAAPI cuStreamDestroy(CUstream hStream) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; +} +CUresult CUDAAPI cuEventDestroy(CUevent hEvent) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; } +#endif /* CUDART_VERSION_INTERNAL || CUDART_VERSION < 4000 */ -CUresult CUDAAPI cuTexRefSetBorderColor(CUtexref hTexRef, float *pBorderColor) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; +#if defined(CUDART_VERSION_INTERNAL) +CUresult CUDAAPI cuMemcpyHtoD_v2(CUdeviceptr dstDevice, const void *srcHost, + size_t ByteCount) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; +} +CUresult CUDAAPI cuMemcpyDtoH_v2(void *dstHost, CUdeviceptr srcDevice, + size_t ByteCount) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; +} +CUresult CUDAAPI cuMemcpyDtoD_v2(CUdeviceptr dstDevice, CUdeviceptr srcDevice, + size_t ByteCount) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; +} +CUresult CUDAAPI cuMemcpyDtoA_v2(CUarray dstArray, size_t dstOffset, + CUdeviceptr srcDevice, size_t ByteCount) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; +} +CUresult CUDAAPI cuMemcpyAtoD_v2(CUdeviceptr dstDevice, CUarray srcArray, + size_t srcOffset, size_t ByteCount) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; +} +CUresult CUDAAPI cuMemcpyHtoA_v2(CUarray dstArray, size_t dstOffset, + const void *srcHost, size_t ByteCount) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; +} +CUresult CUDAAPI cuMemcpyAtoH_v2(void *dstHost, CUarray srcArray, + size_t srcOffset, size_t ByteCount) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; +} +CUresult CUDAAPI cuMemcpyAtoA_v2(CUarray dstArray, size_t dstOffset, + CUarray srcArray, size_t srcOffset, + size_t ByteCount) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; +} +CUresult CUDAAPI cuMemcpyHtoAAsync_v2(CUarray dstArray, size_t dstOffset, + const void *srcHost, size_t ByteCount, + CUstream hStream) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; +} +CUresult CUDAAPI cuMemcpyAtoHAsync_v2(void *dstHost, CUarray srcArray, + size_t srcOffset, size_t ByteCount, + CUstream hStream) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; +} +CUresult CUDAAPI cuMemcpy2D_v2(const CUDA_MEMCPY2D *pCopy) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; +} +CUresult CUDAAPI cuMemcpy2DUnaligned_v2(const CUDA_MEMCPY2D *pCopy) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; +} +CUresult CUDAAPI cuMemcpy3D_v2(const CUDA_MEMCPY3D *pCopy) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; +} +CUresult CUDAAPI cuMemcpyHtoDAsync_v2(CUdeviceptr dstDevice, + const void *srcHost, size_t ByteCount, + CUstream hStream) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; +} +CUresult CUDAAPI cuMemcpyDtoHAsync_v2(void *dstHost, CUdeviceptr srcDevice, + size_t ByteCount, CUstream hStream) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; +} +CUresult CUDAAPI cuMemcpyDtoDAsync_v2(CUdeviceptr dstDevice, + CUdeviceptr srcDevice, size_t ByteCount, + CUstream hStream) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; +} +CUresult CUDAAPI cuMemcpy2DAsync_v2(const CUDA_MEMCPY2D *pCopy, + CUstream hStream) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; +} +CUresult CUDAAPI cuMemcpy3DAsync_v2(const CUDA_MEMCPY3D *pCopy, + CUstream hStream) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; +} +CUresult CUDAAPI cuMemsetD8_v2(CUdeviceptr dstDevice, unsigned char uc, + size_t N) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; +} +CUresult CUDAAPI cuMemsetD16_v2(CUdeviceptr dstDevice, unsigned short us, + size_t N) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; +} +CUresult CUDAAPI cuMemsetD32_v2(CUdeviceptr dstDevice, unsigned int ui, + size_t N) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; +} +CUresult CUDAAPI cuMemsetD2D8_v2(CUdeviceptr dstDevice, size_t dstPitch, + unsigned char uc, size_t Width, + size_t Height) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; +} +CUresult CUDAAPI cuMemsetD2D16_v2(CUdeviceptr dstDevice, size_t dstPitch, + unsigned short us, size_t Width, + size_t Height) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; +} +CUresult CUDAAPI cuMemsetD2D32_v2(CUdeviceptr dstDevice, size_t dstPitch, + unsigned int ui, size_t Width, + size_t Height) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; +} +CUresult CUDAAPI cuMemcpy(CUdeviceptr dst, CUdeviceptr src, size_t ByteCount) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; +} +CUresult CUDAAPI cuMemcpyAsync(CUdeviceptr dst, CUdeviceptr src, + size_t ByteCount, CUstream hStream) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; +} +CUresult CUDAAPI cuMemcpyPeer(CUdeviceptr dstDevice, CUcontext dstContext, + CUdeviceptr srcDevice, CUcontext srcContext, + size_t ByteCount) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; +} +CUresult CUDAAPI cuMemcpyPeerAsync(CUdeviceptr dstDevice, CUcontext dstContext, + CUdeviceptr srcDevice, CUcontext srcContext, + size_t ByteCount, CUstream hStream) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; +} +CUresult CUDAAPI cuMemcpy3DPeer(const CUDA_MEMCPY3D_PEER *pCopy) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; +} +CUresult CUDAAPI cuMemcpy3DPeerAsync(const CUDA_MEMCPY3D_PEER *pCopy, + CUstream hStream) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; +} + +CUresult CUDAAPI cuMemsetD8Async(CUdeviceptr dstDevice, unsigned char uc, + size_t N, CUstream hStream) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; +} +CUresult CUDAAPI cuMemsetD16Async(CUdeviceptr dstDevice, unsigned short us, + size_t N, CUstream hStream) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; +} +CUresult CUDAAPI cuMemsetD32Async(CUdeviceptr dstDevice, unsigned int ui, + size_t N, CUstream hStream) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; +} +CUresult CUDAAPI cuMemsetD2D8Async(CUdeviceptr dstDevice, size_t dstPitch, + unsigned char uc, size_t Width, + size_t Height, CUstream hStream) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; +} +CUresult CUDAAPI cuMemsetD2D16Async(CUdeviceptr dstDevice, size_t dstPitch, + unsigned short us, size_t Width, + size_t Height, CUstream hStream) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; +} +CUresult CUDAAPI cuMemsetD2D32Async(CUdeviceptr dstDevice, size_t dstPitch, + unsigned int ui, size_t Width, + size_t Height, CUstream hStream) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; +} + +CUresult CUDAAPI cuStreamGetPriority(CUstream hStream, int *priority) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; +} +CUresult CUDAAPI cuStreamGetFlags(CUstream hStream, unsigned int *flags) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; +} +CUresult CUDAAPI cuStreamWaitEvent(CUstream hStream, CUevent hEvent, + unsigned int Flags) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; +} +CUresult CUDAAPI cuStreamAddCallback(CUstream hStream, + CUstreamCallback callback, void *userData, + unsigned int flags) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; +} +CUresult CUDAAPI cuStreamAttachMemAsync(CUstream hStream, CUdeviceptr dptr, + size_t length, unsigned int flags) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; +} +CUresult CUDAAPI cuStreamQuery(CUstream hStream) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; +} +CUresult CUDAAPI cuStreamSynchronize(CUstream hStream) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; +} +CUresult CUDAAPI cuEventRecord(CUevent hEvent, CUstream hStream) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; +} +CUresult CUDAAPI cuLaunchKernel(CUfunction f, unsigned int gridDimX, + unsigned int gridDimY, unsigned int gridDimZ, + unsigned int blockDimX, unsigned int blockDimY, + unsigned int blockDimZ, + unsigned int sharedMemBytes, CUstream hStream, + void **kernelParams, void **extra) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; +} +CUresult CUDAAPI cuGraphicsMapResources(unsigned int count, + CUgraphicsResource *resources, + CUstream hStream) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; +} +CUresult CUDAAPI cuGraphicsUnmapResources(unsigned int count, + CUgraphicsResource *resources, + CUstream hStream) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; +} +CUresult CUDAAPI cuMemPrefetchAsync(CUdeviceptr devPtr, size_t count, + CUdevice dstDevice, CUstream hStream) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; +} +CUresult CUDAAPI cuStreamWriteValue32(CUstream stream, CUdeviceptr addr, + cuuint32_t value, unsigned int flags) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; +} +CUresult CUDAAPI cuStreamWaitValue32(CUstream stream, CUdeviceptr addr, + cuuint32_t value, unsigned int flags) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; +} +CUresult CUDAAPI cuStreamBatchMemOp(CUstream stream, unsigned int count, + CUstreamBatchMemOpParams *paramArray, + unsigned int flags) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; } +#endif -CUresult CUDAAPI cuTexRefSetFlags(CUtexref hTexRef, unsigned int Flags) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; +CUresult cuProfilerInitialize(const char *configFile, const char *outputFile, + CUoutput_mode outputMode) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; +} +CUresult cuProfilerStart(void) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; +} +CUresult cuProfilerStop(void) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; } -#if CUDART_VERSION >= 3020 -CUresult CUDAAPI cuTexRefGetAddress(CUdeviceptr *pdptr, CUtexref hTexRef) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; -} -#endif /* CUDART_VERSION >= 3020 */ +//_ptds -CUresult CUDAAPI cuTexRefGetArray(CUarray *phArray, CUtexref hTexRef) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; -} - -CUresult CUDAAPI cuTexRefGetMipmappedArray(CUmipmappedArray *phMipmappedArray, CUtexref hTexRef) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; -} - -CUresult CUDAAPI cuTexRefGetAddressMode(CUaddress_mode *pam, CUtexref hTexRef, int dim) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; -} - -CUresult CUDAAPI cuTexRefGetFilterMode(CUfilter_mode *pfm, CUtexref hTexRef) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; -} - -CUresult CUDAAPI cuTexRefGetFormat(CUarray_format *pFormat, int *pNumChannels, CUtexref hTexRef) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; -} - -CUresult CUDAAPI cuTexRefGetMipmapFilterMode(CUfilter_mode *pfm, CUtexref hTexRef) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; -} - -CUresult CUDAAPI cuTexRefGetMipmapLevelBias(float *pbias, CUtexref hTexRef) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; -} - -CUresult CUDAAPI cuTexRefGetMipmapLevelClamp(float *pminMipmapLevelClamp, float *pmaxMipmapLevelClamp, CUtexref hTexRef) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; -} - -CUresult CUDAAPI cuTexRefGetMaxAnisotropy(int *pmaxAniso, CUtexref hTexRef) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; -} - -CUresult CUDAAPI cuTexRefGetBorderColor(float *pBorderColor, CUtexref hTexRef) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; -} - -CUresult CUDAAPI cuTexRefGetFlags(unsigned int *pFlags, CUtexref hTexRef) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; -} - -CUresult CUDAAPI cuTexRefCreate(CUtexref *pTexRef) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; -} - -CUresult CUDAAPI cuTexRefDestroy(CUtexref hTexRef) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; -} - -CUresult CUDAAPI cuSurfRefSetArray(CUsurfref hSurfRef, CUarray hArray, unsigned int Flags) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; -} - -CUresult CUDAAPI cuSurfRefGetArray(CUarray *phArray, CUsurfref hSurfRef) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; -} - -/** @} */ /* END CUDA_SURFREF */ - -#if CUDART_VERSION >= 5000 -CUresult CUDAAPI cuTexObjectCreate(CUtexObject *pTexObject, const CUDA_RESOURCE_DESC *pResDesc, const CUDA_TEXTURE_DESC *pTexDesc, const CUDA_RESOURCE_VIEW_DESC *pResViewDesc) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; -} - -CUresult CUDAAPI cuTexObjectDestroy(CUtexObject texObject) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; -} - -CUresult CUDAAPI cuTexObjectGetResourceDesc(CUDA_RESOURCE_DESC *pResDesc, CUtexObject texObject) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; -} - -CUresult CUDAAPI cuTexObjectGetTextureDesc(CUDA_TEXTURE_DESC *pTexDesc, CUtexObject texObject) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; -} - -CUresult CUDAAPI cuTexObjectGetResourceViewDesc(CUDA_RESOURCE_VIEW_DESC *pResViewDesc, CUtexObject texObject) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; -} - -/** @} */ /* END CUDA_TEXOBJECT */ - - -CUresult CUDAAPI cuSurfObjectCreate(CUsurfObject *pSurfObject, const CUDA_RESOURCE_DESC *pResDesc) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; -} - -CUresult CUDAAPI cuSurfObjectDestroy(CUsurfObject surfObject) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; -} - -CUresult CUDAAPI cuSurfObjectGetResourceDesc(CUDA_RESOURCE_DESC *pResDesc, CUsurfObject surfObject) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; -} - -#endif /* CUDART_VERSION >= 5000 */ - -#if CUDART_VERSION >= 4000 -CUresult CUDAAPI cuDeviceCanAccessPeer(int *canAccessPeer, CUdevice dev, CUdevice peerDev) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; -} - -CUresult CUDAAPI cuDeviceGetP2PAttribute(int* value, CUdevice_P2PAttribute attrib, CUdevice srcDevice, CUdevice dstDevice) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; -} - -CUresult CUDAAPI cuCtxEnablePeerAccess(CUcontext peerContext, unsigned int Flags) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; -} - -CUresult CUDAAPI cuCtxDisablePeerAccess(CUcontext peerContext) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; -} - -/** @} */ /* END CUDA_PEER_ACCESS */ -#endif /* CUDART_VERSION >= 4000 */ - - -CUresult CUDAAPI cuGraphicsUnregisterResource(CUgraphicsResource resource) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; -} - -CUresult CUDAAPI cuGraphicsSubResourceGetMappedArray(CUarray *pArray, CUgraphicsResource resource, unsigned int arrayIndex, unsigned int mipLevel) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; -} - -#if CUDART_VERSION >= 5000 - -CUresult CUDAAPI cuGraphicsResourceGetMappedMipmappedArray(CUmipmappedArray *pMipmappedArray, CUgraphicsResource resource) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; -} - -#endif /* CUDART_VERSION >= 5000 */ - -#if CUDART_VERSION >= 3020 -CUresult CUDAAPI cuGraphicsResourceGetMappedPointer(CUdeviceptr *pDevPtr, size_t *pSize, CUgraphicsResource resource) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; -} -#endif /* CUDART_VERSION >= 3020 */ - -CUresult CUDAAPI cuGraphicsResourceSetMapFlags(CUgraphicsResource resource, unsigned int flags) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; -} - -CUresult CUDAAPI cuGraphicsMapResources(unsigned int count, CUgraphicsResource *resources, CUstream hStream) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; -} - -CUresult CUDAAPI cuGraphicsUnmapResources(unsigned int count, CUgraphicsResource *resources, CUstream hStream) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; -} - -/** @} */ /* END CUDA_GRAPHICS */ - -CUresult CUDAAPI cuGetExportTable(const void **ppExportTable, const CUuuid *pExportTableId) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - cudaError_t e = cudaGetExportTable(ppExportTable, pExportTableId); - assert(e == cudaSuccess); - return CUDA_SUCCESS; -} - - -#if defined(CUDART_VERSION_INTERNAL) || (CUDART_VERSION >= 4000 && CUDART_VERSION < 6050) -CUresult CUDAAPI cuMemHostRegister(void *p, size_t bytesize, unsigned int Flags) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; -} -#endif /* defined(CUDART_VERSION_INTERNAL) || (CUDART_VERSION >= 4000 && CUDART_VERSION < 6050) */ - -#if defined(CUDART_VERSION_INTERNAL) || (CUDART_VERSION >= 5050 && CUDART_VERSION < 6050) -CUresult CUDAAPI cuLinkCreate(unsigned int numOptions, CUjit_option *options, void **optionValues, CUlinkState *stateOut) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; -} -CUresult CUDAAPI cuLinkAddData(CUlinkState state, CUjitInputType type, void *data, size_t size, const char *name, - unsigned int numOptions, CUjit_option *options, void **optionValues) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; -} -CUresult CUDAAPI cuLinkAddFile(CUlinkState state, CUjitInputType type, const char *path, - unsigned int numOptions, CUjit_option *options, void **optionValues) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; -} -#endif /* CUDART_VERSION_INTERNAL || (CUDART_VERSION >= 5050 && CUDART_VERSION < 6050) */ - -#if defined(CUDART_VERSION_INTERNAL) || (CUDART_VERSION >= 3020 && CUDART_VERSION < 4010) -CUresult CUDAAPI cuTexRefSetAddress2D_v2(CUtexref hTexRef, const CUDA_ARRAY_DESCRIPTOR *desc, CUdeviceptr dptr, size_t Pitch) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; -} -#endif /* CUDART_VERSION_INTERNAL || (CUDART_VERSION >= 3020 && CUDART_VERSION < 4010) */ - -#if defined(CUDART_VERSION_INTERNAL) || CUDART_VERSION < 4000 -CUresult CUDAAPI cuCtxDestroy(CUcontext ctx) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; -} -CUresult CUDAAPI cuCtxPopCurrent(CUcontext *pctx) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; -} -CUresult CUDAAPI cuCtxPushCurrent(CUcontext ctx) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; -} -CUresult CUDAAPI cuStreamDestroy(CUstream hStream) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; -} -CUresult CUDAAPI cuEventDestroy(CUevent hEvent) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; -} -#endif /* CUDART_VERSION_INTERNAL || CUDART_VERSION < 4000 */ - -#if defined(CUDART_VERSION_INTERNAL) - CUresult CUDAAPI cuMemcpyHtoD_v2(CUdeviceptr dstDevice, const void *srcHost, size_t ByteCount) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; -} - CUresult CUDAAPI cuMemcpyDtoH_v2(void *dstHost, CUdeviceptr srcDevice, size_t ByteCount) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; -} - CUresult CUDAAPI cuMemcpyDtoD_v2(CUdeviceptr dstDevice, CUdeviceptr srcDevice, size_t ByteCount) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; -} - CUresult CUDAAPI cuMemcpyDtoA_v2(CUarray dstArray, size_t dstOffset, CUdeviceptr srcDevice, size_t ByteCount) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; -} - CUresult CUDAAPI cuMemcpyAtoD_v2(CUdeviceptr dstDevice, CUarray srcArray, size_t srcOffset, size_t ByteCount) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; -} - CUresult CUDAAPI cuMemcpyHtoA_v2(CUarray dstArray, size_t dstOffset, const void *srcHost, size_t ByteCount) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; -} - CUresult CUDAAPI cuMemcpyAtoH_v2(void *dstHost, CUarray srcArray, size_t srcOffset, size_t ByteCount) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; -} - CUresult CUDAAPI cuMemcpyAtoA_v2(CUarray dstArray, size_t dstOffset, CUarray srcArray, size_t srcOffset, size_t ByteCount) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; -} - CUresult CUDAAPI cuMemcpyHtoAAsync_v2(CUarray dstArray, size_t dstOffset, const void *srcHost, size_t ByteCount, CUstream hStream) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; -} - CUresult CUDAAPI cuMemcpyAtoHAsync_v2(void *dstHost, CUarray srcArray, size_t srcOffset, size_t ByteCount, CUstream hStream) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; -} - CUresult CUDAAPI cuMemcpy2D_v2(const CUDA_MEMCPY2D *pCopy) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; -} - CUresult CUDAAPI cuMemcpy2DUnaligned_v2(const CUDA_MEMCPY2D *pCopy) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; -} - CUresult CUDAAPI cuMemcpy3D_v2(const CUDA_MEMCPY3D *pCopy) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; -} - CUresult CUDAAPI cuMemcpyHtoDAsync_v2(CUdeviceptr dstDevice, const void *srcHost, size_t ByteCount, CUstream hStream) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; -} - CUresult CUDAAPI cuMemcpyDtoHAsync_v2(void *dstHost, CUdeviceptr srcDevice, size_t ByteCount, CUstream hStream) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; -} - CUresult CUDAAPI cuMemcpyDtoDAsync_v2(CUdeviceptr dstDevice, CUdeviceptr srcDevice, size_t ByteCount, CUstream hStream) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; -} - CUresult CUDAAPI cuMemcpy2DAsync_v2(const CUDA_MEMCPY2D *pCopy, CUstream hStream) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; -} - CUresult CUDAAPI cuMemcpy3DAsync_v2(const CUDA_MEMCPY3D *pCopy, CUstream hStream) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; -} - CUresult CUDAAPI cuMemsetD8_v2(CUdeviceptr dstDevice, unsigned char uc, size_t N) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; -} - CUresult CUDAAPI cuMemsetD16_v2(CUdeviceptr dstDevice, unsigned short us, size_t N) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; -} - CUresult CUDAAPI cuMemsetD32_v2(CUdeviceptr dstDevice, unsigned int ui, size_t N) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; -} - CUresult CUDAAPI cuMemsetD2D8_v2(CUdeviceptr dstDevice, size_t dstPitch, unsigned char uc, size_t Width, size_t Height) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; -} - CUresult CUDAAPI cuMemsetD2D16_v2(CUdeviceptr dstDevice, size_t dstPitch, unsigned short us, size_t Width, size_t Height) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; -} - CUresult CUDAAPI cuMemsetD2D32_v2(CUdeviceptr dstDevice, size_t dstPitch, unsigned int ui, size_t Width, size_t Height) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; -} - CUresult CUDAAPI cuMemcpy(CUdeviceptr dst, CUdeviceptr src, size_t ByteCount) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; -} - CUresult CUDAAPI cuMemcpyAsync(CUdeviceptr dst, CUdeviceptr src, size_t ByteCount, CUstream hStream) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; -} - CUresult CUDAAPI cuMemcpyPeer(CUdeviceptr dstDevice, CUcontext dstContext, CUdeviceptr srcDevice, CUcontext srcContext, size_t ByteCount) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; -} - CUresult CUDAAPI cuMemcpyPeerAsync(CUdeviceptr dstDevice, CUcontext dstContext, CUdeviceptr srcDevice, CUcontext srcContext, size_t ByteCount, CUstream hStream) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; -} - CUresult CUDAAPI cuMemcpy3DPeer(const CUDA_MEMCPY3D_PEER *pCopy) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; -} - CUresult CUDAAPI cuMemcpy3DPeerAsync(const CUDA_MEMCPY3D_PEER *pCopy, CUstream hStream) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; -} - - CUresult CUDAAPI cuMemsetD8Async(CUdeviceptr dstDevice, unsigned char uc, size_t N, CUstream hStream) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; -} - CUresult CUDAAPI cuMemsetD16Async(CUdeviceptr dstDevice, unsigned short us, size_t N, CUstream hStream) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; -} - CUresult CUDAAPI cuMemsetD32Async(CUdeviceptr dstDevice, unsigned int ui, size_t N, CUstream hStream) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; -} - CUresult CUDAAPI cuMemsetD2D8Async(CUdeviceptr dstDevice, size_t dstPitch, unsigned char uc, size_t Width, size_t Height, CUstream hStream) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; -} - CUresult CUDAAPI cuMemsetD2D16Async(CUdeviceptr dstDevice, size_t dstPitch, unsigned short us, size_t Width, size_t Height, CUstream hStream) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; -} - CUresult CUDAAPI cuMemsetD2D32Async(CUdeviceptr dstDevice, size_t dstPitch, unsigned int ui, size_t Width, size_t Height, CUstream hStream) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; -} - - CUresult CUDAAPI cuStreamGetPriority(CUstream hStream, int *priority) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; -} - CUresult CUDAAPI cuStreamGetFlags(CUstream hStream, unsigned int *flags) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; -} - CUresult CUDAAPI cuStreamWaitEvent(CUstream hStream, CUevent hEvent, unsigned int Flags) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; -} - CUresult CUDAAPI cuStreamAddCallback(CUstream hStream, CUstreamCallback callback, void *userData, unsigned int flags) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; -} - CUresult CUDAAPI cuStreamAttachMemAsync(CUstream hStream, CUdeviceptr dptr, size_t length, unsigned int flags) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; -} - CUresult CUDAAPI cuStreamQuery(CUstream hStream) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; -} - CUresult CUDAAPI cuStreamSynchronize(CUstream hStream) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; -} - CUresult CUDAAPI cuEventRecord(CUevent hEvent, CUstream hStream) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; -} - CUresult CUDAAPI cuLaunchKernel(CUfunction f, unsigned int gridDimX, unsigned int gridDimY, unsigned int gridDimZ, unsigned int blockDimX, unsigned int blockDimY, unsigned int blockDimZ, unsigned int sharedMemBytes, CUstream hStream, void **kernelParams, void **extra) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; -} - CUresult CUDAAPI cuGraphicsMapResources(unsigned int count, CUgraphicsResource *resources, CUstream hStream) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; -} - CUresult CUDAAPI cuGraphicsUnmapResources(unsigned int count, CUgraphicsResource *resources, CUstream hStream) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; -} - CUresult CUDAAPI cuMemPrefetchAsync(CUdeviceptr devPtr, size_t count, CUdevice dstDevice, CUstream hStream) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; -} - CUresult CUDAAPI cuStreamWriteValue32(CUstream stream, CUdeviceptr addr, cuuint32_t value, unsigned int flags) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; -} - CUresult CUDAAPI cuStreamWaitValue32(CUstream stream, CUdeviceptr addr, cuuint32_t value, unsigned int flags) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; -} - CUresult CUDAAPI cuStreamBatchMemOp(CUstream stream, unsigned int count, CUstreamBatchMemOpParams *paramArray, unsigned int flags) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; -} -#endif - -CUresult cuProfilerInitialize ( const char* configFile, const char* outputFile, CUoutput_mode outputMode ) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; -} -CUresult cuProfilerStart ( void ) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; -} -CUresult cuProfilerStop ( void ) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; -} - -//_ptds - -extern "C" CUresult CUDAAPI cuMemcpy_ptds(CUdeviceptr dst, CUdeviceptr src, size_t ByteCount) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; -} - -extern "C" CUresult CUDAAPI cuMemcpyPeer_ptds(CUdeviceptr dstDevice, CUcontext dstContext, CUdeviceptr srcDevice, CUcontext srcContext, size_t ByteCount) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; -} - - extern "C" CUresult CUDAAPI cuMemcpyHtoD_v2_ptds(CUdeviceptr dstDevice, const void *srcHost, size_t ByteCount) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; -} - extern "C" CUresult CUDAAPI cuMemcpyDtoH_v2_ptds(void *dstHost, CUdeviceptr srcDevice, size_t ByteCount) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; -} - extern "C" CUresult CUDAAPI cuMemcpyDtoD_v2_ptds(CUdeviceptr dstDevice, CUdeviceptr srcDevice, size_t ByteCount) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; -} - extern "C" CUresult CUDAAPI cuMemcpy2DUnaligned_v2_ptds(const CUDA_MEMCPY2D *pCopy) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; -} - extern "C" CUresult CUDAAPI cuMemcpy3D_v2_ptds(const CUDA_MEMCPY3D *pCopy) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; -} - extern "C" CUresult CUDAAPI cuMemcpy3DPeer_ptds(const CUDA_MEMCPY3D_PEER *pCopy) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; -} -extern "C" CUresult CUDAAPI cuMemsetD8_v2_ptds(CUdeviceptr dstDevice, unsigned char uc, unsigned int N) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; -} -extern "C" CUresult CUDAAPI cuMemsetD16_v2_ptds(CUdeviceptr dstDevice, unsigned short us, unsigned int N) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; -} -extern "C" CUresult CUDAAPI cuMemsetD32_v2_ptds(CUdeviceptr dstDevice, unsigned int ui, unsigned int N) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; -} -extern "C" CUresult CUDAAPI cuMemsetD2D8_v2_ptds(CUdeviceptr dstDevice, unsigned int dstPitch, unsigned char uc, unsigned int Width, unsigned int Height) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; -} -extern "C" CUresult CUDAAPI cuMemsetD2D16_v2_ptds(CUdeviceptr dstDevice, unsigned int dstPitch, unsigned short us, unsigned int Width, unsigned int Height) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; -} -extern "C" CUresult CUDAAPI cuMemsetD2D32_v2_ptds(CUdeviceptr dstDevice, unsigned int dstPitch, unsigned int ui, unsigned int Width, unsigned int Height) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; +extern "C" CUresult CUDAAPI cuMemcpy_ptds(CUdeviceptr dst, CUdeviceptr src, + size_t ByteCount) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; +} + +extern "C" CUresult CUDAAPI cuMemcpyPeer_ptds(CUdeviceptr dstDevice, + CUcontext dstContext, + CUdeviceptr srcDevice, + CUcontext srcContext, + size_t ByteCount) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; +} + +extern "C" CUresult CUDAAPI cuMemcpyHtoD_v2_ptds(CUdeviceptr dstDevice, + const void *srcHost, + size_t ByteCount) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; +} +extern "C" CUresult CUDAAPI cuMemcpyDtoH_v2_ptds(void *dstHost, + CUdeviceptr srcDevice, + size_t ByteCount) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; +} +extern "C" CUresult CUDAAPI cuMemcpyDtoD_v2_ptds(CUdeviceptr dstDevice, + CUdeviceptr srcDevice, + size_t ByteCount) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; +} +extern "C" CUresult CUDAAPI +cuMemcpy2DUnaligned_v2_ptds(const CUDA_MEMCPY2D *pCopy) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; +} +extern "C" CUresult CUDAAPI cuMemcpy3D_v2_ptds(const CUDA_MEMCPY3D *pCopy) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; +} +extern "C" CUresult CUDAAPI +cuMemcpy3DPeer_ptds(const CUDA_MEMCPY3D_PEER *pCopy) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; +} +extern "C" CUresult CUDAAPI cuMemsetD8_v2_ptds(CUdeviceptr dstDevice, + unsigned char uc, + unsigned int N) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; +} +extern "C" CUresult CUDAAPI cuMemsetD16_v2_ptds(CUdeviceptr dstDevice, + unsigned short us, + unsigned int N) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; +} +extern "C" CUresult CUDAAPI cuMemsetD32_v2_ptds(CUdeviceptr dstDevice, + unsigned int ui, + unsigned int N) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; +} +extern "C" CUresult CUDAAPI cuMemsetD2D8_v2_ptds(CUdeviceptr dstDevice, + unsigned int dstPitch, + unsigned char uc, + unsigned int Width, + unsigned int Height) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; +} +extern "C" CUresult CUDAAPI cuMemsetD2D16_v2_ptds(CUdeviceptr dstDevice, + unsigned int dstPitch, + unsigned short us, + unsigned int Width, + unsigned int Height) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; +} +extern "C" CUresult CUDAAPI cuMemsetD2D32_v2_ptds(CUdeviceptr dstDevice, + unsigned int dstPitch, + unsigned int ui, + unsigned int Width, + unsigned int Height) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; } //_ptsz -extern "C" CUresult CUDAAPI cuMemcpy3DPeer_ptsz(const CUDA_MEMCPY3D_PEER *pCopy) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; -} - -extern "C" CUresult CUDAAPI cuMemcpyAsync_ptsz(CUdeviceptr dst, CUdeviceptr src, size_t ByteCount, CUstream hStream) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; -} - -extern "C" CUresult CUDAAPI cuMemcpyPeerAsync_ptsz(CUdeviceptr dstDevice, CUcontext dstContext, CUdeviceptr srcDevice, CUcontext srcContext, size_t ByteCount, CUstream hStream) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; -} - extern "C" CUresult CUDAAPI cuMemcpyHtoAAsync_v2_ptsz(CUarray dstArray, size_t dstOffset, const void *srcHost, size_t ByteCount, CUstream hStream) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; -} - extern "C" CUresult CUDAAPI cuMemcpyAtoHAsync_v2_ptsz(void *dstHost, CUarray srcArray, size_t srcOffset, size_t ByteCount, CUstream hStream) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; -} - extern "C" CUresult CUDAAPI cuMemcpyHtoDAsync_v2_ptsz(CUdeviceptr dstDevice, const void *srcHost, size_t ByteCount, CUstream hStream) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; -} - extern "C" CUresult CUDAAPI cuMemcpyDtoHAsync_v2_ptsz(void *dstHost, CUdeviceptr srcDevice, size_t ByteCount, CUstream hStream) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; -} - extern "C" CUresult CUDAAPI cuMemcpyDtoDAsync_v2_ptsz(CUdeviceptr dstDevice, CUdeviceptr srcDevice, size_t ByteCount, CUstream hStream) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; -} - extern "C" CUresult CUDAAPI cuMemcpy2DAsync_v2_ptsz(const CUDA_MEMCPY2D *pCopy, CUstream hStream) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; -} - extern "C" CUresult CUDAAPI cuMemcpy3DAsync_v2_ptsz(const CUDA_MEMCPY3D *pCopy, CUstream hStream) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; -} - extern "C" CUresult CUDAAPI cuMemcpy3DPeerAsync_ptsz(const CUDA_MEMCPY3D_PEER *pCopy, CUstream hStream) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; -} - - extern "C" CUresult CUDAAPI cuMemsetD8Async_ptsz(CUdeviceptr dstDevice, unsigned char uc, size_t N, CUstream hStream) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; -} - extern "C" CUresult CUDAAPI cuMemsetD2D8Async_ptsz(CUdeviceptr dstDevice, size_t dstPitch, unsigned char uc, size_t Width, size_t Height, CUstream hStream) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; -} - extern "C" CUresult CUDAAPI cuLaunchKernel_ptsz(CUfunction f, unsigned int gridDimX, unsigned int gridDimY, unsigned int gridDimZ, unsigned int blockDimX, unsigned int blockDimY, unsigned int blockDimZ, unsigned int sharedMemBytes, CUstream hStream, void **kernelParams, void **extra) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; -} - extern "C" CUresult CUDAAPI cuEventRecord_ptsz(CUevent hEvent, CUstream hStream) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; -} - extern "C" CUresult CUDAAPI cuStreamWriteValue32_ptsz(CUstream stream, CUdeviceptr addr, cuuint32_t value, unsigned int flags) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; -} - extern "C" CUresult CUDAAPI cuStreamWaitValue32_ptsz(CUstream stream, CUdeviceptr addr, cuuint32_t value, unsigned int flags) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; -} - extern "C" CUresult CUDAAPI cuStreamBatchMemOp_ptsz(CUstream stream, unsigned int count, CUstreamBatchMemOpParams *paramArray, unsigned int flags) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; -} -extern "C" CUresult CUDAAPI cuStreamGetPriority_ptsz(CUstream hStream, int *priority) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; -} -extern "C" CUresult CUDAAPI cuStreamGetFlags_ptsz(CUstream hStream, unsigned int *flags) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; -} - - -extern "C" CUresult CUDAAPI cuStreamWaitEvent_ptsz(CUstream hStream, CUevent hEvent, unsigned int Flags) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; -} - -extern "C" CUresult CUDAAPI cuStreamAddCallback_ptsz(CUstream hStream, CUstreamCallback callback, void *userData, unsigned int flags) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; -} - -extern "C" CUresult CUDAAPI cuStreamSynchronize_ptsz(CUstream hStream) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; -} - - extern "C" CUresult CUDAAPI cuStreamQuery_ptsz(CUstream hStream) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; -} -extern "C" CUresult CUDAAPI cuStreamAttachMemAsync_ptsz(CUstream hStream, CUdeviceptr dptr, size_t length, unsigned int flags) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; -} - -extern "C" CUresult CUDAAPI cuGraphicsMapResources_ptsz(unsigned int count, CUgraphicsResource *resources, CUstream hStream) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; -} - - -extern "C" CUresult CUDAAPI cuGraphicsUnmapResources_ptsz(unsigned int count, CUgraphicsResource *resources, CUstream hStream) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; -} - - extern "C" CUresult CUDAAPI cuMemPrefetchAsync_ptsz(CUdeviceptr devPtr, size_t count, CUdevice dstDevice, CUstream hStream) -{ - if(g_debug_execution >= 3){ - announce_call(__my_func__); - } - printf("WARNING: this function has not been implemented yet."); - return CUDA_SUCCESS; +extern "C" CUresult CUDAAPI +cuMemcpy3DPeer_ptsz(const CUDA_MEMCPY3D_PEER *pCopy) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; +} + +extern "C" CUresult CUDAAPI cuMemcpyAsync_ptsz(CUdeviceptr dst, CUdeviceptr src, + size_t ByteCount, + CUstream hStream) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; +} + +extern "C" CUresult CUDAAPI cuMemcpyPeerAsync_ptsz( + CUdeviceptr dstDevice, CUcontext dstContext, CUdeviceptr srcDevice, + CUcontext srcContext, size_t ByteCount, CUstream hStream) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; +} +extern "C" CUresult CUDAAPI cuMemcpyHtoAAsync_v2_ptsz(CUarray dstArray, + size_t dstOffset, + const void *srcHost, + size_t ByteCount, + CUstream hStream) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; +} +extern "C" CUresult CUDAAPI cuMemcpyAtoHAsync_v2_ptsz(void *dstHost, + CUarray srcArray, + size_t srcOffset, + size_t ByteCount, + CUstream hStream) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; +} +extern "C" CUresult CUDAAPI cuMemcpyHtoDAsync_v2_ptsz(CUdeviceptr dstDevice, + const void *srcHost, + size_t ByteCount, + CUstream hStream) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; +} +extern "C" CUresult CUDAAPI cuMemcpyDtoHAsync_v2_ptsz(void *dstHost, + CUdeviceptr srcDevice, + size_t ByteCount, + CUstream hStream) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; +} +extern "C" CUresult CUDAAPI cuMemcpyDtoDAsync_v2_ptsz(CUdeviceptr dstDevice, + CUdeviceptr srcDevice, + size_t ByteCount, + CUstream hStream) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; +} +extern "C" CUresult CUDAAPI cuMemcpy2DAsync_v2_ptsz(const CUDA_MEMCPY2D *pCopy, + CUstream hStream) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; +} +extern "C" CUresult CUDAAPI cuMemcpy3DAsync_v2_ptsz(const CUDA_MEMCPY3D *pCopy, + CUstream hStream) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; +} +extern "C" CUresult CUDAAPI +cuMemcpy3DPeerAsync_ptsz(const CUDA_MEMCPY3D_PEER *pCopy, CUstream hStream) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; +} + +extern "C" CUresult CUDAAPI cuMemsetD8Async_ptsz(CUdeviceptr dstDevice, + unsigned char uc, size_t N, + CUstream hStream) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; +} +extern "C" CUresult CUDAAPI cuMemsetD2D8Async_ptsz(CUdeviceptr dstDevice, + size_t dstPitch, + unsigned char uc, + size_t Width, size_t Height, + CUstream hStream) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; +} +extern "C" CUresult CUDAAPI cuLaunchKernel_ptsz( + CUfunction f, unsigned int gridDimX, unsigned int gridDimY, + unsigned int gridDimZ, unsigned int blockDimX, unsigned int blockDimY, + unsigned int blockDimZ, unsigned int sharedMemBytes, CUstream hStream, + void **kernelParams, void **extra) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; +} +extern "C" CUresult CUDAAPI cuEventRecord_ptsz(CUevent hEvent, + CUstream hStream) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; +} +extern "C" CUresult CUDAAPI cuStreamWriteValue32_ptsz(CUstream stream, + CUdeviceptr addr, + cuuint32_t value, + unsigned int flags) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; +} +extern "C" CUresult CUDAAPI cuStreamWaitValue32_ptsz(CUstream stream, + CUdeviceptr addr, + cuuint32_t value, + unsigned int flags) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; +} +extern "C" CUresult CUDAAPI cuStreamBatchMemOp_ptsz( + CUstream stream, unsigned int count, CUstreamBatchMemOpParams *paramArray, + unsigned int flags) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; +} +extern "C" CUresult CUDAAPI cuStreamGetPriority_ptsz(CUstream hStream, + int *priority) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; +} +extern "C" CUresult CUDAAPI cuStreamGetFlags_ptsz(CUstream hStream, + unsigned int *flags) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; +} + +extern "C" CUresult CUDAAPI cuStreamWaitEvent_ptsz(CUstream hStream, + CUevent hEvent, + unsigned int Flags) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; +} + +extern "C" CUresult CUDAAPI cuStreamAddCallback_ptsz(CUstream hStream, + CUstreamCallback callback, + void *userData, + unsigned int flags) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; +} + +extern "C" CUresult CUDAAPI cuStreamSynchronize_ptsz(CUstream hStream) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; +} + +extern "C" CUresult CUDAAPI cuStreamQuery_ptsz(CUstream hStream) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; +} +extern "C" CUresult CUDAAPI cuStreamAttachMemAsync_ptsz(CUstream hStream, + CUdeviceptr dptr, + size_t length, + unsigned int flags) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; +} + +extern "C" CUresult CUDAAPI cuGraphicsMapResources_ptsz( + unsigned int count, CUgraphicsResource *resources, CUstream hStream) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; +} + +extern "C" CUresult CUDAAPI cuGraphicsUnmapResources_ptsz( + unsigned int count, CUgraphicsResource *resources, CUstream hStream) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; +} + +extern "C" CUresult CUDAAPI cuMemPrefetchAsync_ptsz(CUdeviceptr devPtr, + size_t count, + CUdevice dstDevice, + CUstream hStream) { + if (g_debug_execution >= 3) { + announce_call(__my_func__); + } + printf("WARNING: this function has not been implemented yet."); + return CUDA_SUCCESS; } diff --git a/libcuda/cuobjdump.h b/libcuda/cuobjdump.h index 6ab6778..38afa4c 100644 --- a/libcuda/cuobjdump.h +++ b/libcuda/cuobjdump.h @@ -1,80 +1,81 @@ #ifndef __cuobjdump_h__ #define __cuobjdump_h__ -#include -#include #include +#include +#include -typedef void * yyscan_t; +typedef void *yyscan_t; struct cuobjdump_parser { - yyscan_t scanner; - int elfserial; - int ptxserial; - FILE *ptxfile; - FILE *elffile; - FILE *sassfile; - char filename [1024]; + yyscan_t scanner; + int elfserial; + int ptxserial; + FILE *ptxfile; + FILE *elffile; + FILE *sassfile; + char filename[1024]; }; class cuobjdumpSection { -public: - //Constructor - cuobjdumpSection() { - arch = 0; - identifier = ""; - } - virtual ~cuobjdumpSection() {} - unsigned getArch() {return arch;} - void setArch(unsigned a) {arch = a;} - std::string getIdentifier() {return identifier;} - void setIdentifier(std::string i) {identifier = i;} - virtual void print(){std::cout << "cuobjdump Section: unknown type" << std::endl;} -private: - unsigned arch; - std::string identifier; + public: + // Constructor + cuobjdumpSection() { + arch = 0; + identifier = ""; + } + virtual ~cuobjdumpSection() {} + unsigned getArch() { return arch; } + void setArch(unsigned a) { arch = a; } + std::string getIdentifier() { return identifier; } + void setIdentifier(std::string i) { identifier = i; } + virtual void print() { + std::cout << "cuobjdump Section: unknown type" << std::endl; + } + + private: + unsigned arch; + std::string identifier; }; -class cuobjdumpELFSection : public cuobjdumpSection -{ -public: - cuobjdumpELFSection() {} - virtual ~cuobjdumpELFSection() { - elffilename = ""; - sassfilename = ""; - } - std::string getELFfilename() {return elffilename;} - void setELFfilename(std::string f) {elffilename = f;} - std::string getSASSfilename() {return sassfilename;} - void setSASSfilename(std::string f) {sassfilename = f;} - virtual void print() { - std::cout << "ELF Section:" << std::endl; - std::cout << "arch: sm_" << getArch() << std::endl; - std::cout << "identifier: " << getIdentifier() << std::endl; - std::cout << "elf filename: " << getELFfilename() << std::endl; - std::cout << "sass filename: " << getSASSfilename() << std::endl; - std::cout << std::endl; - } -private: - std::string elffilename; - std::string sassfilename; +class cuobjdumpELFSection : public cuobjdumpSection { + public: + cuobjdumpELFSection() {} + virtual ~cuobjdumpELFSection() { + elffilename = ""; + sassfilename = ""; + } + std::string getELFfilename() { return elffilename; } + void setELFfilename(std::string f) { elffilename = f; } + std::string getSASSfilename() { return sassfilename; } + void setSASSfilename(std::string f) { sassfilename = f; } + virtual void print() { + std::cout << "ELF Section:" << std::endl; + std::cout << "arch: sm_" << getArch() << std::endl; + std::cout << "identifier: " << getIdentifier() << std::endl; + std::cout << "elf filename: " << getELFfilename() << std::endl; + std::cout << "sass filename: " << getSASSfilename() << std::endl; + std::cout << std::endl; + } + + private: + std::string elffilename; + std::string sassfilename; }; -class cuobjdumpPTXSection : public cuobjdumpSection -{ -public: - cuobjdumpPTXSection(){ - ptxfilename = ""; - } - std::string getPTXfilename() {return ptxfilename;} - void setPTXfilename(std::string f) {ptxfilename = f;} - virtual void print() { - std::cout << "PTX Section:" << std::endl; - std::cout << "arch: sm_" << getArch() << std::endl; - std::cout << "identifier: " << getIdentifier() << std::endl; - std::cout << "ptx filename: " << getPTXfilename() << std::endl; - std::cout << std::endl; - } -private: - std::string ptxfilename; +class cuobjdumpPTXSection : public cuobjdumpSection { + public: + cuobjdumpPTXSection() { ptxfilename = ""; } + std::string getPTXfilename() { return ptxfilename; } + void setPTXfilename(std::string f) { ptxfilename = f; } + virtual void print() { + std::cout << "PTX Section:" << std::endl; + std::cout << "arch: sm_" << getArch() << std::endl; + std::cout << "identifier: " << getIdentifier() << std::endl; + std::cout << "ptx filename: " << getPTXfilename() << std::endl; + std::cout << std::endl; + } + + private: + std::string ptxfilename; }; #endif /* __cuobjdump_h__ */ diff --git a/libcuda/gpgpu_context.h b/libcuda/gpgpu_context.h index 61d7507..d0cd7c4 100644 --- a/libcuda/gpgpu_context.h +++ b/libcuda/gpgpu_context.h @@ -1,76 +1,83 @@ #ifndef __gpgpu_context_h__ #define __gpgpu_context_h__ -#include "cuda_api_object.h" -#include "../src/cuda-sim/ptx_loader.h" -#include "../src/cuda-sim/ptx_parser.h" -#include "../src/gpgpusim_entrypoint.h" #include "../src/cuda-sim/cuda-sim.h" #include "../src/cuda-sim/cuda_device_runtime.h" #include "../src/cuda-sim/ptx-stats.h" +#include "../src/cuda-sim/ptx_loader.h" +#include "../src/cuda-sim/ptx_parser.h" +#include "../src/gpgpusim_entrypoint.h" +#include "cuda_api_object.h" class gpgpu_context { - public: - gpgpu_context() { - g_global_allfiles_symbol_table = NULL; - sm_next_access_uid=0; - warp_inst_sm_next_uid=0; - operand_info_sm_next_uid = 1; - kernel_info_m_next_uid = 1; - g_num_ptx_inst_uid = 0; - g_ptx_cta_info_uid = 1; - symbol_sm_next_uid = 1; - function_info_sm_next_uid = 1; - debug_tensorcore = 0; - api = new cuda_runtime_api(this); - ptxinfo = new ptxinfo_data(this); - ptx_parser = new ptx_recognizer(this); - the_gpgpusim = new GPGPUsim_ctx(this); - func_sim = new cuda_sim(this); - device_runtime = new cuda_device_runtime(this); - stats = new ptx_stats(this); - } - // global list - symbol_table *g_global_allfiles_symbol_table; - const char *g_filename; - unsigned sm_next_access_uid; - unsigned warp_inst_sm_next_uid; - unsigned operand_info_sm_next_uid;//uid for operand_info - unsigned kernel_info_m_next_uid;//uid for kernel_info_t - unsigned g_num_ptx_inst_uid; //uid for ptx inst inside ptx_instruction - unsigned long long g_ptx_cta_info_uid; - unsigned symbol_sm_next_uid; //uid for symbol - unsigned function_info_sm_next_uid; - std::vector s_g_pc_to_insn; // a direct mapping from PC to instruction - bool debug_tensorcore; + public: + gpgpu_context() { + g_global_allfiles_symbol_table = NULL; + sm_next_access_uid = 0; + warp_inst_sm_next_uid = 0; + operand_info_sm_next_uid = 1; + kernel_info_m_next_uid = 1; + g_num_ptx_inst_uid = 0; + g_ptx_cta_info_uid = 1; + symbol_sm_next_uid = 1; + function_info_sm_next_uid = 1; + debug_tensorcore = 0; + api = new cuda_runtime_api(this); + ptxinfo = new ptxinfo_data(this); + ptx_parser = new ptx_recognizer(this); + the_gpgpusim = new GPGPUsim_ctx(this); + func_sim = new cuda_sim(this); + device_runtime = new cuda_device_runtime(this); + stats = new ptx_stats(this); + } + // global list + symbol_table *g_global_allfiles_symbol_table; + const char *g_filename; + unsigned sm_next_access_uid; + unsigned warp_inst_sm_next_uid; + unsigned operand_info_sm_next_uid; // uid for operand_info + unsigned kernel_info_m_next_uid; // uid for kernel_info_t + unsigned g_num_ptx_inst_uid; // uid for ptx inst inside ptx_instruction + unsigned long long g_ptx_cta_info_uid; + unsigned symbol_sm_next_uid; // uid for symbol + unsigned function_info_sm_next_uid; + std::vector + s_g_pc_to_insn; // a direct mapping from PC to instruction + bool debug_tensorcore; - // objects pointers for each file - cuda_runtime_api* api; - ptxinfo_data* ptxinfo; - ptx_recognizer* ptx_parser; - GPGPUsim_ctx* the_gpgpusim; - cuda_sim* func_sim; - cuda_device_runtime* device_runtime; - ptx_stats* stats; - // member function list - void synchronize(); - void exit_simulation(); - void print_simulation_time(); - int gpgpu_opencl_ptx_sim_main_perf( kernel_info_t *grid ); - void cuobjdumpParseBinary(unsigned int handle); - class symbol_table *gpgpu_ptx_sim_load_ptx_from_string( const char *p, unsigned source_num ); - class symbol_table *gpgpu_ptx_sim_load_ptx_from_filename( const char *filename ); - void gpgpu_ptx_info_load_from_filename( const char *filename, unsigned sm_version); - void gpgpu_ptxinfo_load_from_string( const char *p_for_info, unsigned source_num, unsigned sm_version=20, int no_of_ptx=0 ); - void print_ptx_file( const char *p, unsigned source_num, const char *filename ); - class symbol_table* init_parser(const char*); - class gpgpu_sim *gpgpu_ptx_sim_init_perf(); - void start_sim_thread(int api); - struct _cuda_device_id *GPGPUSim_Init(); - void ptx_reg_options(option_parser_t opp); - const ptx_instruction* pc_to_instruction(unsigned pc); - const warp_inst_t *ptx_fetch_inst( address_type pc ); - unsigned translate_pc_to_ptxlineno(unsigned pc); + // objects pointers for each file + cuda_runtime_api *api; + ptxinfo_data *ptxinfo; + ptx_recognizer *ptx_parser; + GPGPUsim_ctx *the_gpgpusim; + cuda_sim *func_sim; + cuda_device_runtime *device_runtime; + ptx_stats *stats; + // member function list + void synchronize(); + void exit_simulation(); + void print_simulation_time(); + int gpgpu_opencl_ptx_sim_main_perf(kernel_info_t *grid); + void cuobjdumpParseBinary(unsigned int handle); + class symbol_table *gpgpu_ptx_sim_load_ptx_from_string(const char *p, + unsigned source_num); + class symbol_table *gpgpu_ptx_sim_load_ptx_from_filename( + const char *filename); + void gpgpu_ptx_info_load_from_filename(const char *filename, + unsigned sm_version); + void gpgpu_ptxinfo_load_from_string(const char *p_for_info, + unsigned source_num, + unsigned sm_version = 20, + int no_of_ptx = 0); + void print_ptx_file(const char *p, unsigned source_num, const char *filename); + class symbol_table *init_parser(const char *); + class gpgpu_sim *gpgpu_ptx_sim_init_perf(); + void start_sim_thread(int api); + struct _cuda_device_id *GPGPUSim_Init(); + void ptx_reg_options(option_parser_t opp); + const ptx_instruction *pc_to_instruction(unsigned pc); + const warp_inst_t *ptx_fetch_inst(address_type pc); + unsigned translate_pc_to_ptxlineno(unsigned pc); }; -gpgpu_context* GPGPU_Context(); +gpgpu_context *GPGPU_Context(); #endif /* __gpgpu_context_h__ */ -- cgit v1.3 From 5053bb5f7dbc0e43dc1cd08f8a82f0fe04f63283 Mon Sep 17 00:00:00 2001 From: tgrogers Date: Thu, 9 Apr 2020 13:39:44 -0400 Subject: Missed a line in the merge that was causing livelock on the asyncAPI app --- libcuda/cuda_runtime_api.cc | 9 ++++++--- 1 file changed, 6 insertions(+), 3 deletions(-) (limited to 'libcuda/cuda_runtime_api.cc') diff --git a/libcuda/cuda_runtime_api.cc b/libcuda/cuda_runtime_api.cc index 12f9636..add0a65 100644 --- a/libcuda/cuda_runtime_api.cc +++ b/libcuda/cuda_runtime_api.cc @@ -2200,6 +2200,7 @@ __host__ cudaError_t CUDARTAPI cudaEventRecordInternal( if (!e) return g_last_cudaError = cudaErrorUnknown; struct CUstream_st *s = (struct CUstream_st *)stream; stream_operation op(e, s); + e->issue(); ctx->the_gpgpusim->g_stream_manager->push(op); return g_last_cudaError = cudaSuccess; } @@ -2220,9 +2221,11 @@ __host__ cudaError_t CUDARTAPI cudaStreamWaitEventInternal( // https://www.cs.cmu.edu/afs/cs/academic/class/15668-s11/www/cuda-doc/html/group__CUDART__STREAM_gfe68d207dc965685d92d3f03d77b0876.html CUevent_st *e = get_event(event); if (!e) { - printf( - "GPGPU-Sim API: Warning: cudaEventRecord has not been called on event " - "before calling cudaStreamWaitEvent.\nNothing to be done.\n"); + printf("GPGPU-Sim API: Error at cudaStreamWaitEvent. Event is not created .\n"); + return g_last_cudaError = cudaErrorInvalidResourceHandle; + } + else if(e->num_issued() == 0) { + printf("GPGPU-Sim API: Warning: cudaEventRecord has not been called on event before calling cudaStreamWaitEvent.\nNothin g to be done.\n"); return g_last_cudaError = cudaSuccess; } if (!stream) { -- cgit v1.3 From 5168397c85fedab857b5ede85d1ab5708c243bb3 Mon Sep 17 00:00:00 2001 From: tgrogers Date: Tue, 21 Apr 2020 14:37:45 -0400 Subject: Small change from the auto-formatter --- libcuda/cuda_runtime_api.cc | 24 +++++++++++++----------- 1 file changed, 13 insertions(+), 11 deletions(-) (limited to 'libcuda/cuda_runtime_api.cc') diff --git a/libcuda/cuda_runtime_api.cc b/libcuda/cuda_runtime_api.cc index add0a65..fd05f55 100644 --- a/libcuda/cuda_runtime_api.cc +++ b/libcuda/cuda_runtime_api.cc @@ -2221,11 +2221,14 @@ __host__ cudaError_t CUDARTAPI cudaStreamWaitEventInternal( // https://www.cs.cmu.edu/afs/cs/academic/class/15668-s11/www/cuda-doc/html/group__CUDART__STREAM_gfe68d207dc965685d92d3f03d77b0876.html CUevent_st *e = get_event(event); if (!e) { - printf("GPGPU-Sim API: Error at cudaStreamWaitEvent. Event is not created .\n"); + printf( + "GPGPU-Sim API: Error at cudaStreamWaitEvent. Event is not created " + ".\n"); return g_last_cudaError = cudaErrorInvalidResourceHandle; - } - else if(e->num_issued() == 0) { - printf("GPGPU-Sim API: Warning: cudaEventRecord has not been called on event before calling cudaStreamWaitEvent.\nNothin g to be done.\n"); + } else if (e->num_issued() == 0) { + printf( + "GPGPU-Sim API: Warning: cudaEventRecord has not been called on event " + "before calling cudaStreamWaitEvent.\nNothin g to be done.\n"); return g_last_cudaError = cudaSuccess; } if (!stream) { @@ -3775,16 +3778,15 @@ cudaError_t CUDARTAPI cudaFuncGetAttributes(struct cudaFuncAttributes *attr, return cudaFuncGetAttributesInternal(attr, hostFun); } -cudaError_t CUDARTAPI cudaEventCreateWithFlags(cudaEvent_t *event, int flags) -{ - CUevent_st *e = new CUevent_st(flags==cudaEventBlockingSync); - g_timer_events[e->get_uid()] = e; +cudaError_t CUDARTAPI cudaEventCreateWithFlags(cudaEvent_t *event, int flags) { + CUevent_st *e = new CUevent_st(flags == cudaEventBlockingSync); + g_timer_events[e->get_uid()] = e; #if CUDART_VERSION >= 3000 - *event = e; + *event = e; #else - *event = e->get_uid(); + *event = e->get_uid(); #endif - return g_last_cudaError = cudaSuccess; + return g_last_cudaError = cudaSuccess; } cudaError_t CUDARTAPI cudaDriverGetVersion(int *driverVersion) { -- cgit v1.3