summaryrefslogtreecommitdiff
diff options
context:
space:
mode:
Diffstat
-rw-r--r--Makefile4
-rw-r--r--getDecuda/decuda.patch37
-rwxr-xr-xgetDecuda/getDecuda.sh13
-rw-r--r--libcuda/cuda_runtime_api.cc66
-rw-r--r--src/cuda-sim/ptx_loader.cc148
-rw-r--r--src/cuda-sim/ptx_loader.h1
6 files changed, 8 insertions, 261 deletions
diff --git a/Makefile b/Makefile
index a146436..8abe321 100644
--- a/Makefile
+++ b/Makefile
@@ -61,7 +61,6 @@ else
TARGETS += $(SIM_LIB_DIR)/libOpenCL.so
endif
TARGETS += decuda_to_ptxplus/decuda_to_ptxplus
- TARGETS += decuda
gpgpusim: makedirs $(TARGETS)
@@ -135,9 +134,6 @@ decuda_to_ptxplus/decuda_to_ptxplus:
$(MAKE) -C ./decuda_to_ptxplus/ depend
$(MAKE) -C ./decuda_to_ptxplus/
-decuda:
- ./getDecuda/getDecuda.sh
-
makedirs:
if [ ! -d $(SIM_LIB_DIR) ]; then mkdir -p $(SIM_LIB_DIR); fi;
if [ ! -d $(SIM_OBJ_FILES_DIR)/libcuda ]; then mkdir -p $(SIM_OBJ_FILES_DIR)/libcuda; fi;
diff --git a/getDecuda/decuda.patch b/getDecuda/decuda.patch
deleted file mode 100644
index eab6c50..0000000
--- a/getDecuda/decuda.patch
+++ /dev/null
@@ -1,37 +0,0 @@
-diff -crB decuda/Instruction.py decudaplus/Instruction.py
-*** decuda/Instruction.py 2010-07-24 05:28:03.000000000 -0700
---- decudaplus/Instruction.py 2010-09-27 12:52:24.000000000 -0700
-***************
-*** 401,406 ****
---- 401,409 ----
- value = self.bits(0,0x001F0000)
- else:
- value = self.bits(0,0x007F0000)
-+ if len(self.inst)==2:
-+ if self.bits(1,0x00200000):
-+ t_offset = 0
-
- if opt_imm and self.fullinst and self.bits(1,0x00100000):
- # Operand 3 is immediate
-diff -crB decuda/Operand.py decudaplus/Operand.py
-*** decuda/Operand.py 2010-07-24 05:28:03.000000000 -0700
---- decudaplus/Operand.py 2010-09-27 12:52:24.000000000 -0700
-***************
-*** 176,183 ****
- rv.write(("0x%06x") % value)
- elif self.offset != None:
- rv.write(("%s[$ofs%i%s%s0x%0"+width+"x]") % (_indir[self.indirection], self.offset, _ofs_inc[self.offset_inc], times, value))
-! elif self.indirection == OP_INDIRECTION_SHARED and self.size==16 and value in _param_space:
-! rv.write(_param_space[value])
- elif self.indirection == OP_INDIRECTION_INTERNAL and value in _ldgpu_ops:
- rv.write(_ldgpu_ops[value])
- else:
---- 176,183 ----
- rv.write(("0x%06x") % value)
- elif self.offset != None:
- rv.write(("%s[$ofs%i%s%s0x%0"+width+"x]") % (_indir[self.indirection], self.offset, _ofs_inc[self.offset_inc], times, value))
-! elif self.indirection == OP_INDIRECTION_SHARED and self.size==16 and value/2 in _param_space:
-! rv.write(_param_space[value/2])
- elif self.indirection == OP_INDIRECTION_INTERNAL and value in _ldgpu_ops:
- rv.write(_ldgpu_ops[value])
- else:
diff --git a/getDecuda/getDecuda.sh b/getDecuda/getDecuda.sh
deleted file mode 100755
index a63a2ec..0000000
--- a/getDecuda/getDecuda.sh
+++ /dev/null
@@ -1,13 +0,0 @@
-# This script downloads the latest version of decuda from http://github.com/laanwj/decuda, untars it and patches it.
-# Only works with decuda 0.4.2.
-
-# decuda.patch must be in the same folder as getDecuda.sh
-
-cd ./getDecuda
-wget -O decuda.tgz --no-check-certificate http://github.com/laanwj/decuda/tarball/master
-tar -xf ./decuda.tgz
-mv laanwj-decuda-c30bd17 decuda # if this fails, we need to check what has changed in decuda, or get a perminant link to decuda 0.4.2
-patch -d ./decuda < ./decuda.patch
-rm ./decuda.tgz
-mv decuda ../decuda
-cd ..
diff --git a/libcuda/cuda_runtime_api.cc b/libcuda/cuda_runtime_api.cc
index c0d7be3..aca7167 100644
--- a/libcuda/cuda_runtime_api.cc
+++ b/libcuda/cuda_runtime_api.cc
@@ -1154,7 +1154,8 @@ void setCuobjdumpsassfilename(const char* filename){
}
extern "C" int cuobjdump_parse();
extern "C" FILE *cuobjdump_in;
-//! Return PTX file
+
+//! Call cuobjdump to extract everything (-elf -sass -ptx) to a file
/*!
* This Function extract the whole PTX (for all the files) using cuobjdump
* */
@@ -1163,7 +1164,7 @@ void extract_ptx(){
char* whole_code;
char fname[1024]="_ptx_whole_code_WESWW";
//! Running CUobjdump using dynamic link to current process
- snprintf(command,1000,"cuobjdump -ptx -elf -sass /proc/%d/exe > %s",getpid(),fname);
+ snprintf(command,1000,"$CUDA_INSTALL_PATH/bin/cuobjdump -ptx -elf -sass /proc/%d/exe > %s",getpid(),fname);
printf("Running cuobjdump using \"%s\"\n", command);
int result = system(command);
if(result) {printf("ERROR: Failed to execute: %s\n", command); exit(1);}
@@ -1174,23 +1175,6 @@ void extract_ptx(){
cuobjdump_parse();
fclose(cuobjdump_in);
printf("Done parsing!!!\n");
- // Just to stop things here
-
- /* Hadi's old code
- FILE* fp = fopen(fname,"r");
- //! finding size of the file
- int lSize= 0;
- fseek (fp , 0 , SEEK_END);
- lSize = ftell (fp);
- rewind (fp);
- //! allocate and copy the entire ptx
- whole_code = (char*)malloc(lSize * sizeof(char));
- fread(whole_code,1,lSize,fp);
- fclose(fp);
- snprintf(command,1000,"rm %s",fname);
- //system(command);
- return whole_code;
- */
}
//! Return proper ptx version
@@ -1217,23 +1201,8 @@ unsigned get_best_version(std::list<cuobjdumpSection> sectionlist, CUctx_st *con
}
}
return max_capability;
-
- /* Hadi's old code
- char* start_entry;
- char* arch_pointer;
- start_entry = strstr(ptx,"Fatbin ptx code:");
- while(start_entry){
- arch_pointer = strstr (start_entry,"arch = ");
- unsigned int capability = 0;
- sscanf(arch_pointer,"arch = sm_%u",&capability);
- if(capability > max_capability && (forced_max_capability == 0 || capability <= forced_max_capability)){
- max_capability = capability;
- }
- start_entry = strstr(arch_pointer,"Fatbin ptx code:");
- }
- return max_capability;
- */
}
+
/*!
* Return number of diffrent ptx files generated in first argument
* which have sm version equal to selected_capability
@@ -1254,20 +1223,6 @@ unsigned get_number_of_ptx(std::list<cuobjdumpSection> sectionlist, unsigned sel
}
}
return result;
- /* Hadi's old code
- char* start_entry;
- char* arch_pointer;
- start_entry = strstr(ptx,"Fatbin ptx code:");
- while(start_entry){
- arch_pointer = strstr (start_entry,"arch = ");
- unsigned int capability = 0;
- sscanf(arch_pointer,"arch = sm_%u",&capability);
- if(capability == selected_capability)
- result++;
- start_entry = strstr(arch_pointer,"Fatbin ptx code:");
- }
- */
-
}
char* readfile (const char* filename){
@@ -1415,16 +1370,9 @@ void** CUDARTAPI __cudaRegisterFatBinary( void *fatCubin )
symbol_table *symtab;
const char *ptx = info->ptx[selected_capability].ptx;
if(context->get_device()->get_gpgpu()->get_config().convert_to_ptxplus() ) {
- if (info->cubin[selected_capability].cubin ==NULL) {
- printf("GPGPU-Sim PTX: Cannot convert to ptxplus no cubin found, probably because it was compiled using newer version of cuda (>=3.0)\nGPGPU-Sim PTX: Exiting ...\n");
- exit(1);
- }
- char *ptxplus_str = gpgpu_ptx_sim_convert_ptx_to_ptxplus(ptx, info->cubin[selected_capability].cubin, source_num++,
- context->get_device()->get_gpgpu()->get_config().saved_converted_ptxplus(), false);
- symtab=gpgpu_ptx_sim_load_ptx_from_string(ptxplus_str,source_num);
- context->add_binary(symtab,fat_cubin_handle);
- gpgpu_ptxinfo_load_from_string(ptx,source_num);
- delete[] ptxplus_str;
+ 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);
diff --git a/src/cuda-sim/ptx_loader.cc b/src/cuda-sim/ptx_loader.cc
index 0c34c99..2879251 100644
--- a/src/cuda-sim/ptx_loader.cc
+++ b/src/cuda-sim/ptx_loader.cc
@@ -90,157 +90,11 @@ void print_ptx_file( const char *p, unsigned source_num, const char *filename )
fflush(stdout);
}
-char* gpgpu_ptx_sim_convert_ptx_to_ptxplus(const char *ptx_str, const char *cubin_str, unsigned source_num, bool save_converted, bool cuobjdump)
-{
-
- printf("GPGPU-Sim PTX: converting EMBEDDED .ptx file to ptxplus \n");
-
- // Extract ptx to a file
- char fname_ptx[1024];
- snprintf(fname_ptx,1024,"_ptx_XXXXXX");
- int fd=mkstemp(fname_ptx);
- close(fd);
-
- printf("GPGPU-Sim PTX: extracting embedded .ptx to temporary file \"%s\"\n", fname_ptx);
- FILE *ptxfile = fopen(fname_ptx,"w");
- fprintf(ptxfile,"%s",ptx_str);
- fclose(ptxfile);
-
- char fname_ptxplus[1024];
- snprintf(fname_ptxplus,1024,"_ptxplus_XXXXXX");
- int fd4=mkstemp(fname_ptxplus);
- close(fd4);
-
- // Extract cubin to a file
- char fname_decuda[1024];
- char fname_cubin[1024];
- if( !cuobjdump){
-
- snprintf(fname_cubin,1024,"_cubin_XXXXXX");
- int fd2=mkstemp(fname_cubin);
- close(fd2);
-
- printf("GPGPU-Sim PTX: extracting embedded cubin to temporary file \"%s\"\n", fname_cubin);
- FILE *cubinfile = fopen(fname_cubin,"w");
- fprintf(cubinfile,"%s",cubin_str);
- fclose(cubinfile);
-
-
- snprintf(fname_decuda,1024,"_decuda_XXXXXX");
- int fd3=mkstemp(fname_decuda);
- close(fd3);
-
- // Run decuda
- char decuda_commandline[1024];
- snprintf(decuda_commandline,1024,"$DECUDA_INSTALL_PATH/decuda.py -o %s %s", fname_decuda, fname_cubin);
- printf("GPGPU-Sim PTX: calling decuda on cubin file, decuda output file = \"%s\"\n", fname_decuda);
- int decuda_result = system(decuda_commandline);
- if( decuda_result != 0 ) {
- printf("GPGPU-Sim PTX: ERROR ** while calling decuda (b) %d\n", decuda_result);
- printf(" Ensure env variable DECUDA_INSTALL_PATH is set and points to decuda base directory.\n");
- exit(1);
- }
- // Run decuda_to_ptxplus
- char d2pp_commandline[1024];
- snprintf(d2pp_commandline,1024,"$D2PP_INSTALL_PATH/decuda_to_ptxplus %s %s %s %s > /dev/null", fname_decuda, fname_ptx, fname_cubin, fname_ptxplus);
- printf("GPGPU-Sim PTX: calling decuda_to_ptxplus, ptxplus output file = \"%s\"\n", fname_ptxplus);
- int d2pp_result = system(d2pp_commandline);
- if( d2pp_result != 0 ) {
- printf("GPGPU-Sim PTX: ERROR ** while calling decuda_to_ptxplus %d\n", d2pp_result);
- printf(" Ensure env variable D2PP_INSTALL_PATH is set and points to decuda_to_ptxplus base directory.\n");
- exit(1);
- }
- }
- else {
- // Run cuobjdump_to_ptxplus
- char commandline[1024];
- int result;
- snprintf(commandline, 1024, "$GPGPUSIM_ROOT/cuobjdump_to_ptxplus/cuobjdump_to_ptxplus /proc/%d/exe %s", getpid(), fname_ptxplus);
- fflush(stdout);
- printf("GPGPU-Sim PTX: calling cuobjdump_to_ptxplus on exe file, output file = \"%s\"\n", fname_ptxplus);
- result = system(commandline);
- if(result){printf("GPGPU-Sim PTX: ERROR ** could not execute %s\n", commandline); exit(1);}
- }
-
- // Get ptxplus from file
- std::ifstream fileStream(fname_ptxplus, std::ios::in);
- std::string text, line;
- while(getline(fileStream,line)) {
- text += (line + "\n");
- }
- fileStream.close();
-
- char* ptxplus_str = new char [strlen(text.c_str())+1];
- strcpy(ptxplus_str, text.c_str());
-
- // Save ptxplus to file if specified
- if(save_converted) {
- char fname_ptxplus_save[1024];
- snprintf(fname_ptxplus_save,1024,"_%u.ptxplus", source_num );
- printf("GPGPU-Sim PTX: saving converted ptxplus to file \"%s\"\n", fname_ptxplus_save);
-
- FILE *file_ptxplus_save = fopen(fname_ptxplus_save,"w");
- fprintf(file_ptxplus_save,"%s",ptxplus_str);
- fclose(file_ptxplus_save);
- }
-
- // Remove temporary files
- char rm_commandline[1024];
- if (!cuobjdump) {
- snprintf(rm_commandline,1024,"rm -f %s %s %s %s", fname_ptx, fname_cubin, fname_decuda, fname_ptxplus);
- }
- else {
- snprintf(rm_commandline,1024,"rm -f %s %s", fname_ptx, fname_ptxplus);
- }
- printf("GPGPU-Sim PTX: removing temporary files using \"%s\"\n", rm_commandline);
- int rm_result = system(rm_commandline);
- if( rm_result != 0 ) {
- printf("GPGPU-Sim PTX: ERROR ** while removing temporary files %d\n", rm_result);
- exit(1);
- }
-
- printf("GPGPU-Sim PTX: DONE converting EMBEDDED .ptx file to ptxplus \n");
-
- return ptxplus_str;
-}
-
char* gpgpu_ptx_sim_convert_ptx_and_sass_to_ptxplus(const char *ptxfilename, const char *elffilename, const char *sassfilename)
{
printf("GPGPU-Sim PTX: converting EMBEDDED .ptx file to ptxplus \n");
-/*
- // Extract ptx to a file
- char fname_ptx[1024];
- snprintf(fname_ptx,1024,"_ptx_XXXXXX");
- int fd1=mkstemp(fname_ptx);
- close(fd1);
-
- printf("GPGPU-Sim PTX: extracting embedded .ptx to temporary file \"%s\"\n", fname_ptx);
- FILE *ptxfile = fopen(fname_ptx,"w");
- fprintf(ptxfile,"%s",ptx_str);
- fclose(ptxfile);
-
- char fname_sass[1024];
- snprintf(fname_sass,1024,"_sass_XXXXXX");
- int fd2=mkstemp(fname_sass);
- close(fd2);
-
- printf("GPGPU-Sim PTX: extracting .sass to temporary file \"%s\"\n", fname_sass);
- FILE *sassfile = fopen(fname_sass,"w");
- fprintf(sassfile,"%s",sass_str);
- fclose(sassfile);
-
- char fname_elf[1024];
- snprintf(fname_elf,1024,"_elf_XXXXXX");
- int fd3=mkstemp(fname_elf);
- close(fd3);
-
- printf("GPGPU-Sim PTX: extracting .elf to temporary file \"%s\"\n", fname_elf);
- FILE *elffile = fopen(fname_elf,"w");
- fprintf(elffile,"%s",elf_str);
- fclose(elffile);
-*/
char fname_ptxplus[1024];
snprintf(fname_ptxplus,1024,"_ptxplus_XXXXXX");
int fd4=mkstemp(fname_ptxplus);
@@ -250,7 +104,6 @@ char* gpgpu_ptx_sim_convert_ptx_and_sass_to_ptxplus(const char *ptxfilename, con
char fname_decuda[1024];
char fname_cubin[1024];
-
// Run cuobjdump_to_ptxplus
char commandline[1024];
int result;
@@ -273,6 +126,7 @@ char* gpgpu_ptx_sim_convert_ptx_and_sass_to_ptxplus(const char *ptxfilename, con
strcpy(ptxplus_str, text.c_str());
// Remove temporary files
+ // TODO: Fix deleting/keeping PTXPlus files
/*
char rm_commandline[1024];
diff --git a/src/cuda-sim/ptx_loader.h b/src/cuda-sim/ptx_loader.h
index aa98a91..9227d85 100644
--- a/src/cuda-sim/ptx_loader.h
+++ b/src/cuda-sim/ptx_loader.h
@@ -32,7 +32,6 @@ extern bool g_override_embedded_ptx;
class symbol_table *gpgpu_ptx_sim_load_ptx_from_string( const char *p, unsigned source_num );
void gpgpu_ptxinfo_load_from_string( const char *p_for_info, unsigned source_num );
-char* gpgpu_ptx_sim_convert_ptx_to_ptxplus(const char *ptx_str, const char *cubin_str, unsigned source_num, bool save_converted, bool cuobjdump );
char* gpgpu_ptx_sim_convert_ptx_and_sass_to_ptxplus(const char *ptx_str, const char *sass_str, const char *elf_str);
#endif
generated by cgit v1.3 (git 2.53.0) at 2026-07-11 06:39:37 +0000