Merge remote-tracking branch 'upstream/dev' into dev

9 files changed, 33 insertions, 27 deletions

diff --git a/.gitignore b/.gitignore
index 53fadb5..887b605 100644
--- a/.gitignore
+++ b/.gitignore
@@ -8,6 +8,7 @@ libcuda/cuobjdump_parser.h
 libcuda/cuobjdump_parser.output
 
 lib/*
+doc/doxygen/html
 
 cuobjdump_to_ptxplus/elf_lexer.cc
 cuobjdump_to_ptxplus/elf_parser.cc
@@ -26,4 +27,5 @@ cuobjdump_to_ptxplus/sass_parser.cc
 cuobjdump_to_ptxplus/sass_parser.hh
 cuobjdump_to_ptxplus/sass_parser.output
 
-build/*
\ No newline at end of file
+build/*
+*.swp
diff --git a/libcuda/cuda_runtime_api.cc b/libcuda/cuda_runtime_api.cc
index cbe8a11..9bdb993 100644
--- a/libcuda/cuda_runtime_api.cc
+++ b/libcuda/cuda_runtime_api.cc
@@ -1406,7 +1406,7 @@ void extract_code_using_cuobjdump(){
 		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 << std::endl;
+			std::cout << "Failed to execute: " << cmd.str() << std::endl;
 			exit(1);
 		}
 		std::ifstream libsf;
@@ -1438,7 +1438,7 @@ void extract_code_using_cuobjdump(){
 			if(result) {printf("ERROR: Failed to execute: %s\n", command); exit(1);}
 			std::cout << "Done" << std::endl;
 
-			std::cout << "Trying to parse " << libcodfn << std::endl;
+			std::cout << "Trying to parse " << libcodfn.str() << std::endl;
 			cuobjdump_in = fopen(libcodfn.str().c_str(), "r");
 			cuobjdump_parse();
 			fclose(cuobjdump_in);
@@ -1540,7 +1540,7 @@ std::list<cuobjdumpSection*> pruneSectionList(std::list<cuobjdumpSection*> cuobj
 
 //! Merge all PTX sections that have a specific identifier into one file
 std::list<cuobjdumpSection*> mergeMatchingSections(std::list<cuobjdumpSection*> cuobjdumpSectionList, std::string identifier){
-	char *ptxcode = "";
+	const char *ptxcode = "";
 	std::list<cuobjdumpSection*>::iterator old_iter;
 	cuobjdumpPTXSection* old_ptxsection = NULL;
 	cuobjdumpPTXSection* ptxsection;
@@ -1689,7 +1689,7 @@ std::map<int, bool>fatbin_registered;
 std::map<std::string, symbol_table*> name_symtab;
 
 //! Keep track of the association between filename and cubin handle
-void cuobjdumpRegisterFatBinary(unsigned int handle, char* filename){
+void cuobjdumpRegisterFatBinary(unsigned int handle, const char* filename){
 	fatbinmap[handle] = filename;
 }
 
@@ -1764,6 +1764,7 @@ void** CUDARTAPI __cudaRegisterFatBinary( void *fatCubin )
 		if (sizeof(void*) == 4) 
 			printf("GPGPU-Sim PTX: FatBin file name extraction has not been tested on 32-bit system.\n"); 
 
+		#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;
@@ -1772,12 +1773,11 @@ void** CUDARTAPI __cudaRegisterFatBinary( void *fatCubin )
 		// - 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. 
-		#if (CUDART_VERSION <= 6000)
 		char * pfatbin = (char*) fatDeviceText->d; 
 		int offset = *((int*)(pfatbin+48)); 
 		char * filename = (pfatbin+16+offset); 
 		#else
-		char * filename = "default";
+		const char * filename = "default";
 		#endif
 		// The extracted file name is associated with a fat_cubin_handle passed
 		// into cudaLaunch().  Inside cudaLaunch(), the associated file name is
@@ -1798,7 +1798,7 @@ void** CUDARTAPI __cudaRegisterFatBinary( void *fatCubin )
 
 		return (void**)fat_cubin_handle;
 	} 
-	#if (CUDART_VERSION < 8000)
+#if (CUDART_VERSION < 8000)
 	else {
 		static unsigned source_num=1;
 		unsigned long long fat_cubin_handle = next_fat_bin_handle++;
@@ -1856,7 +1856,12 @@ void** CUDARTAPI __cudaRegisterFatBinary( void *fatCubin )
 		}
 		return (void**)fat_cubin_handle;
 	}
-	#endif
+#else
+        else {
+		printf("ERROR **  __cudaRegisterFatBinary() needs to be updated\n");
+		abort();
+        }
+#endif
 }
 
 void __cudaUnregisterFatBinary(void **fatCubinHandle)
diff --git a/libcuda/cuobjdump.l b/libcuda/cuobjdump.l
index f63ee73..0953ea1 100644
--- a/libcuda/cuobjdump.l
+++ b/libcuda/cuobjdump.l
@@ -159,8 +159,6 @@ newlines	{newline}+
 %%
 void cuobjdump_error(const char* message)
 {
-	printf("   "); printf(message); printf(" near \""); printf(yytext); printf("\"");
-	printf(" on line "); 
-	char line[5]; sprintf(line, "%i", yylineno); printf(line);
-	printf("\n");
+	printf("   %s near \"%s\"",message, yytext);
+	printf(" on line %i\n",yylineno); 
 }
diff --git a/src/abstract_hardware_model.h b/src/abstract_hardware_model.h
index aaa4b00..7125b6b 100644
--- a/src/abstract_hardware_model.h
+++ b/src/abstract_hardware_model.h
@@ -383,7 +383,7 @@ protected:
     std::deque<simt_stack_entry> m_stack;
 };
 
-#define GLOBAL_HEAP_START 0x703E20000
+#define GLOBAL_HEAP_START 0xC0000000
    // start allocating from this address (lower values used for allocating globals in .ptx file)
 #define SHARED_MEM_SIZE_MAX (64*1024)
 #define LOCAL_MEM_SIZE_MAX (8*1024)
diff --git a/src/cuda-sim/cuda-math.h b/src/cuda-sim/cuda-math.h
index 4721e8a..a3db0df 100644
--- a/src/cuda-sim/cuda-math.h
+++ b/src/cuda-sim/cuda-math.h
@@ -321,7 +321,7 @@ float __internal_accurate_fdividef(float a, float b)
 float __saturatef(float a)
 {
    float b; 
-   if (isnan(a)) b = 0.0f; 
+   if (std::isnan(a)) b = 0.0f; 
    else if (a >= 1.0f) b = 1.0f;
    else if (a <= 0.0f) b = 0.0f; 
    else b = a; 
diff --git a/src/cuda-sim/instructions.cc b/src/cuda-sim/instructions.cc
index 011c285..71286c9 100644
--- a/src/cuda-sim/instructions.cc
+++ b/src/cuda-sim/instructions.cc
@@ -1961,7 +1961,7 @@ ptx_reg_t d2d( ptx_reg_t x, unsigned from_width, unsigned to_width, int to_sign,
       y.f64 = x.f64;
       break; 
    }
-   if (isnan(y.f64)) {
+   if (std::isnan(y.f64)) {
       y.u64 = 0xfff8000000000000ull;
    } else if (saturation_mode) {
       y.f64 = cuda_math::__saturatef(y.f64); 
@@ -2086,7 +2086,7 @@ void ptx_round(ptx_reg_t& data, int rounding_mode, int type)
       }
    }
    if ((type == F64_TYPE)||(type == FF64_TYPE)) {
-      if (isnan(data.f64)) {
+      if (std::isnan(data.f64)) {
          data.u64 = 0xfff8000000000000ull;
       }
    }
@@ -2648,12 +2648,12 @@ void mad_def( const ptx_instruction *pI, ptx_thread_info *thread, bool use_carry
 
 bool isNaN(float x)
 {
-   return isnan(x);
+   return std::isnan(x);
 }
 
 bool isNaN(double x)
 {
-   return isnan(x);
+   return std::isnan(x);
 }
 
 void max_impl( const ptx_instruction *pI, ptx_thread_info *thread ) 
diff --git a/src/cuda-sim/ptx.y b/src/cuda-sim/ptx.y
index e00aa4b..4edae5d 100644
--- a/src/cuda-sim/ptx.y
+++ b/src/cuda-sim/ptx.y
@@ -47,7 +47,6 @@ OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 %token  PTR_DIRECTIVE
 %token  ENTRY_DIRECTIVE
 %token  EXTERN_DIRECTIVE
-%token  WEAK_DIRECTIVE
 %token  FILE_DIRECTIVE
 %token  FUNC_DIRECTIVE
 %token  GLOBAL_DIRECTIVE
@@ -513,7 +512,8 @@ compare_spec:EQ_OPTION { add_option(EQ_OPTION); }
 	| NAN_OPTION { add_option(NAN_OPTION); } 
 	;
 
-operand_list: operand
+operand_list: /* empty*/
+    | operand
 	| operand COMMA operand_list;
 
 operand: IDENTIFIER  { add_scalar_operand( $1 ); }
diff --git a/src/cuda-sim/ptx_parser.cc b/src/cuda-sim/ptx_parser.cc
index baa3bcd..a180da9 100644
--- a/src/cuda-sim/ptx_parser.cc
+++ b/src/cuda-sim/ptx_parser.cc
@@ -433,7 +433,7 @@ void add_identifier( const char *identifier, int array_dim, unsigned array_ident
       assert( (num_bits%8) == 0  );
       addr = g_current_symbol_table->get_shared_next();
       addr_pad = pad_address(addr, num_bits/8, 128);
-      printf("from 0x%x to 0x%lx (shared memory space)\n",
+      printf("from 0x%llx to 0x%llx (shared memory space)\n",
               addr+addr_pad,
               addr+addr_pad + num_bits/8);
          fflush(stdout);
@@ -450,7 +450,7 @@ void add_identifier( const char *identifier, int array_dim, unsigned array_ident
          assert( (num_bits%8) == 0  ); 
          addr = g_current_symbol_table->get_global_next();
          addr_pad = pad_address(addr, num_bits/8, 128);
-         printf("from 0x%x to 0x%lx (global memory space) %u\n",
+         printf("from 0x%llx to 0x%llx (global memory space) %u\n",
               addr+addr_pad,
               addr+addr_pad + num_bits/8,
               g_const_alloc++);
@@ -471,7 +471,7 @@ void add_identifier( const char *identifier, int array_dim, unsigned array_ident
       assert( (num_bits%8) == 0  );
       addr = g_current_symbol_table->get_global_next();
       addr_pad = pad_address(addr, num_bits/8, 128);
-      printf("from 0x%x to 0x%lx (global memory space)\n",
+      printf("from 0x%llx to 0x%llx (global memory space)\n",
               addr+addr_pad,
               addr+addr_pad + num_bits/8);
       fflush(stdout);
@@ -488,7 +488,7 @@ void add_identifier( const char *identifier, int array_dim, unsigned array_ident
          assert( (num_bits%8) == 0  );
          addr = g_current_symbol_table->get_local_next();
          addr_pad = pad_address(addr, num_bits/8, 128);
-         printf("from 0x%x to 0x%lx (local memory space)\n",
+         printf("from 0x%llx to 0x%llx (local memory space)\n",
                  addr+addr_pad,
                  addr+addr_pad + num_bits/8);
          fflush(stdout);
@@ -501,7 +501,7 @@ void add_identifier( const char *identifier, int array_dim, unsigned array_ident
         assert( (num_bits%8) == 0 );
         addr = g_current_symbol_table->get_local_next();
         addr_pad = pad_address(addr, num_bits/8, 128);
-        printf("from 0x%x to 0x%lx\n",
+        printf("from 0x%llx to 0x%llx\n",
                 addr+addr_pad,
                 addr+addr_pad + num_bits/8);
         fflush(stdout);
diff --git a/src/gpgpu-sim/gpu-sim.cc b/src/gpgpu-sim/gpu-sim.cc
index 550c550..bd4c00c 100644
--- a/src/gpgpu-sim/gpu-sim.cc
+++ b/src/gpgpu-sim/gpu-sim.cc
@@ -32,6 +32,7 @@
 #include <stdio.h>
 #include <stdlib.h>
 #include <math.h>
+#include <signal.h>
 #include "zlib.h"
 
 
@@ -1486,7 +1487,7 @@ void gpgpu_sim::cycle()
 
 
       if( g_single_step && ((gpu_sim_cycle+gpu_tot_sim_cycle) >= g_single_step) ) {
-          asm("int $03");
+          raise(SIGTRAP); // Debug breakpoint
       }
       gpu_sim_cycle++;
       if( g_interactive_debugger_enabled )