diff options
| author | Mengchi Zhang <[email protected]> | 2019-07-08 14:19:32 -0400 |
|---|---|---|
| committer | GitHub <[email protected]> | 2019-07-08 14:19:32 -0400 |
| commit | 109a03655326ed23d26a3496589b21be51089bbb (patch) | |
| tree | 3331bc8b8159411d2ebf9203ea99235cba9baa92 /src/cuda-sim/cuda-sim.cc | |
| parent | 47d18d2dea9a27061193218f1e70ba50cf7ff46f (diff) | |
| parent | b3e786e3d8d720217f36a214e9b5be9a19ab9dd2 (diff) | |
Merge pull request #22 from echoedit/dev
g_keep_intermediate_files
Diffstat (limited to 'src/cuda-sim/cuda-sim.cc')
| -rw-r--r-- | src/cuda-sim/cuda-sim.cc | 7 |
1 files changed, 4 insertions, 3 deletions
diff --git a/src/cuda-sim/cuda-sim.cc b/src/cuda-sim/cuda-sim.cc index 261d605..df0bbd7 100644 --- a/src/cuda-sim/cuda-sim.cc +++ b/src/cuda-sim/cuda-sim.cc @@ -51,6 +51,7 @@ typedef void * yyscan_t; #include "decuda_pred_table/decuda_pred_table.h" #include "../stream_manager.h" #include "cuda_device_runtime.h" +#include "../../libcuda/gpgpu_context.h" int gpgpu_ptx_instruction_classification; void ** g_inst_classification_stat = NULL; @@ -66,12 +67,12 @@ int cp_cta_resume; unsigned g_ptx_sim_num_insn = 0; unsigned gpgpu_param_num_shaders = 0; -char *opcode_latency_int, *opcode_latency_fp, *opcode_latency_dp,*opcode_latency_sfu,*opcode_latency_tensor; +char *opcode_latency_fp, *opcode_latency_dp,*opcode_latency_sfu,*opcode_latency_tensor; char *opcode_initiation_int, *opcode_initiation_fp, *opcode_initiation_dp,*opcode_initiation_sfu,*opcode_initiation_tensor; char *cdp_latency_str; unsigned cdp_latency[5]; -void ptx_opcocde_latency_options (option_parser_t opp) { +void cuda_sim::ptx_opcocde_latency_options (option_parser_t opp) { option_parser_register(opp, "-ptx_opcode_latency_int", OPT_CSTR, &opcode_latency_int, "Opcode latencies for integers <ADD,MAX,MUL,MAD,DIV>" "Default 1,1,19,25,145", @@ -667,7 +668,7 @@ void ptx_instruction::set_opcode_and_latency() * [3] MAD * [4] DIV */ - sscanf(opcode_latency_int, "%u,%u,%u,%u,%u", + sscanf(gpgpu_ctx->func_sim->opcode_latency_int, "%u,%u,%u,%u,%u", &int_latency[0],&int_latency[1],&int_latency[2], &int_latency[3],&int_latency[4]); sscanf(opcode_latency_fp, "%u,%u,%u,%u,%u", |
