diff options
| author | Tim Rogers <[email protected]> | 2019-05-13 19:42:43 -0400 |
|---|---|---|
| committer | GitHub <[email protected]> | 2019-05-13 19:42:43 -0400 |
| commit | cbb929b6f15f63f0b104e2acac7a17f02c0208fe (patch) | |
| tree | ed2bf825d27848c818a9293075cf9b9d73a26864 /src/cuda-sim | |
| parent | f507979bcf0f14d1e5843c9b08613d6b0a4bb7a2 (diff) | |
| parent | 059dabc1af44b8eb60f0cacc8d5c2d06f1e85a00 (diff) | |
Merge pull request #118 from mkhairy/dev
Integrating Mahmoud's changes that improve simulation speed by 2-3x.
Also includes a change that mitigates memory channel conflicts by using a random hashing function.
Diffstat (limited to 'src/cuda-sim')
| -rw-r--r-- | src/cuda-sim/cuda-sim.h | 2 |
1 files changed, 2 insertions, 0 deletions
diff --git a/src/cuda-sim/cuda-sim.h b/src/cuda-sim/cuda-sim.h index abd32f9..e690356 100644 --- a/src/cuda-sim/cuda-sim.h +++ b/src/cuda-sim/cuda-sim.h @@ -47,6 +47,8 @@ extern int g_debug_thread_uid; extern void ** g_inst_classification_stat; extern void ** g_inst_op_classification_stat; extern int g_ptx_kernel_count; // used for classification stat collection purposes +extern char *opcode_latency_int, *opcode_latency_fp, *opcode_latency_dp,*opcode_latency_sfu,*opcode_latency_tensor; + void ptx_opcocde_latency_options (option_parser_t opp); extern class kernel_info_t *gpgpu_opencl_ptx_sim_init_grid(class function_info *entry, |