diff options
| -rw-r--r-- | libcuda/cuda_runtime_api.cc | 98 |
1 files changed, 49 insertions, 49 deletions
diff --git a/libcuda/cuda_runtime_api.cc b/libcuda/cuda_runtime_api.cc index 6941b26..2b0f944 100644 --- a/libcuda/cuda_runtime_api.cc +++ b/libcuda/cuda_runtime_api.cc @@ -1696,60 +1696,60 @@ 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){ - if(fatbin_registered[handle]) return; - fatbin_registered[handle] = true; - CUctx_st *context = GPGPUSim_Context(); - std::string fname = fatbinmap[handle]; + if(fatbin_registered[handle]) return; + fatbin_registered[handle] = true; + CUctx_st *context = GPGPUSim_Context(); + std::string fname = fatbinmap[handle]; - if (name_symtab.find(fname) != name_symtab.end()) { - symbol_table *symtab = name_symtab[fname]; - context->add_binary(symtab, handle); - return; - } + if (name_symtab.find(fname) != name_symtab.end()) { + symbol_table *symtab = name_symtab[fname]; + context->add_binary(symtab, handle); + return; + } - unsigned max_capability = 0; - for ( std::list<cuobjdumpSection*>::iterator iter = cuobjdumpSectionList.begin(); - iter != cuobjdumpSectionList.end(); - iter++){ - unsigned capability = (*iter)->getArch(); - if (capability > max_capability) max_capability = capability; - } + unsigned max_capability = 0; + for ( std::list<cuobjdumpSection*>::iterator iter = cuobjdumpSectionList.begin(); + iter != cuobjdumpSectionList.end(); + iter++ ){ + unsigned capability = (*iter)->getArch(); + if (capability > max_capability) max_capability = capability; + } printf("Using PTX version = %u\n", max_capability); - if (max_capability > 20) printf("WARNING: No guarantee that PTX will be parsed for SM version %u\n", max_capability); + if (max_capability > 20) printf("WARNING: No guarantee that PTX will be parsed for SM version %u\n", max_capability); - cuobjdumpPTXSection* ptx = findPTXSection(fname); - symbol_table *symtab; - char *ptxcode; - const char *override_ptx_name = getenv("PTX_SIM_KERNELFILE"); - if (override_ptx_name == NULL or getenv("PTX_SIM_USE_PTX_FILE") == NULL) { - 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 = findELFSection(ptx->getIdentifier()); - assert (elfsection!= NULL); - char *ptxplus_str = 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 ); - delete[] ptxplus_str; - } else { - symtab=gpgpu_ptx_sim_load_ptx_from_string(ptxcode, 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 ); - } - 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; + cuobjdumpPTXSection* ptx = findPTXSection(fname); + symbol_table *symtab; + char *ptxcode; + const char *override_ptx_name = getenv("PTX_SIM_KERNELFILE"); + if (override_ptx_name == NULL or getenv("PTX_SIM_USE_PTX_FILE") == NULL) { + 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 = findELFSection(ptx->getIdentifier()); + assert (elfsection!= NULL); + char *ptxplus_str = 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 ); + delete[] ptxplus_str; + } else { + symtab=gpgpu_ptx_sim_load_ptx_from_string(ptxcode, 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 ); + } + 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; - //TODO: Remove temporarily files as per configurations + //TODO: Remove temporarily files as per configurations } void** CUDARTAPI __cudaRegisterFatBinary( void *fatCubin ) |