diff options
Diffstat (limited to 'src/cuda-sim')
| -rw-r--r-- | src/cuda-sim/ptx_loader.cc | 18 | ||||
| -rw-r--r-- | src/cuda-sim/ptx_loader.h | 2 |
2 files changed, 15 insertions, 5 deletions
diff --git a/src/cuda-sim/ptx_loader.cc b/src/cuda-sim/ptx_loader.cc index 6c1b595..99174cf 100644 --- a/src/cuda-sim/ptx_loader.cc +++ b/src/cuda-sim/ptx_loader.cc @@ -54,6 +54,7 @@ extern int ptxinfo_debug; extern FILE *ptxinfo_in; static bool g_save_embedded_ptx; +static int g_occupancy_sm_number; bool g_keep_intermediate_files; bool m_ptx_save_converted_ptxplus; @@ -71,6 +72,10 @@ void ptx_reg_options(option_parser_t opp) &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, + "The SM number to pass to ptxas when getting register usage for computing GPU occupancy." + "This parameter is required in the config.", + "0"); } void print_ptx_file( const char *p, unsigned source_num, const char *filename ) @@ -287,7 +292,7 @@ void fix_duplicate_errors(char fname2[1024]) { } } -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 ) { char fname[1024]; snprintf(fname,1024,"_ptx_XXXXXX"); @@ -321,12 +326,17 @@ void gpgpu_ptxinfo_load_from_string( const char *p_for_info, unsigned source_num extra_flags[0]=0; #if CUDART_VERSION >= 3000 - if (sm_version == 0) sm_version = 20; + if ( g_occupancy_sm_number == 0 ) { + printf( "gpgpusim.config must specify the sm version for the GPU that you use to compute occupancy.\n" + "The register file size is specifically tied to the sm version used to querry ptxas for register usage." + "A register size/SM mismatch may result in occupancy differences." ); + exit(1); + } extern bool g_cdp_enabled; if(!g_cdp_enabled) - snprintf(extra_flags,1024,"--gpu-name=sm_%u",sm_version); + snprintf(extra_flags,1024,"--gpu-name=sm_%u", g_occupancy_sm_number); else - snprintf(extra_flags,1024,"--compile-only --gpu-name=sm_%u",sm_version); + snprintf(extra_flags,1024,"--compile-only --gpu-name=sm_%u",g_occupancy_sm_number); #endif snprintf(commandline,1024,"$CUDA_INSTALL_PATH/bin/ptxas %s -v %s --output-file /dev/null 2> %s", diff --git a/src/cuda-sim/ptx_loader.h b/src/cuda-sim/ptx_loader.h index d3d0c92..cb02eda 100644 --- a/src/cuda-sim/ptx_loader.h +++ b/src/cuda-sim/ptx_loader.h @@ -32,7 +32,7 @@ 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, unsigned sm_version=20 ); +void gpgpu_ptxinfo_load_from_string( const char *p_for_info, unsigned source_num ); 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(); |