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

4 files changed, 19 insertions, 15 deletions

diff --git a/libcuda/cuda_runtime_api.cc b/libcuda/cuda_runtime_api.cc
index 5452ed0..e4ae251 100644
--- a/libcuda/cuda_runtime_api.cc
+++ b/libcuda/cuda_runtime_api.cc
@@ -1565,7 +1565,7 @@ 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;
@@ -1597,7 +1597,7 @@ 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);
@@ -1703,7 +1703,7 @@ 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;
@@ -1855,7 +1855,7 @@ 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;
 }
 
@@ -1938,6 +1938,7 @@ 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;
@@ -1946,12 +1947,11 @@ 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
@@ -1972,7 +1972,7 @@ 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++;
@@ -2030,8 +2030,12 @@ void** CUDARTAPI __cudaRegisterFatBinary( void *fatCubin )
 		}
 		return (void**)fat_cubin_handle;
 	}
-	#endif
-        return 0;
+#else
+        else {
+		printf("ERROR **  __cudaRegisterFatBinary() needs to be updated\n");
+		abort();
+        }
+#endif
 }
 
 void __cudaUnregisterFatBinary(void **fatCubinHandle)
diff --git a/src/abstract_hardware_model.h b/src/abstract_hardware_model.h
index 67b36c7..f5708bc 100644
--- a/src/abstract_hardware_model.h
+++ b/src/abstract_hardware_model.h
@@ -386,7 +386,7 @@ 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 )