diff options
| author | Suchita Pati <[email protected]> | 2018-05-21 00:26:04 +0000 |
|---|---|---|
| committer | tgrogers-purdue <[email protected]> | 2018-05-21 00:26:04 +0000 |
| commit | 1b6389ab70211b622f02b149bfb20ae68a2e2552 (patch) | |
| tree | 9e75d6d0474d1cfbf27501e2461022c593b6f740 /libcuda | |
| parent | 0f8c407eea99e1622486127912a83ef3e0dcf397 (diff) | |
| parent | f405cb9484a8b0f961bd7c143bebf1fcb17546da (diff) | |
Merged in suchitapati/gpgpusim-cudnn-05-01-18 (pull request #7)
Support for -gpgpu_registers_per_block and addition of -save_embedded_ptx 1 to config files for mnistCUDNN
Approved-by: tgrogers-purdue <[email protected]>
Diffstat (limited to 'libcuda')
| -rw-r--r-- | libcuda/cuda_runtime_api.cc | 27 |
1 files changed, 8 insertions, 19 deletions
diff --git a/libcuda/cuda_runtime_api.cc b/libcuda/cuda_runtime_api.cc index ef46f00..e3c2542 100644 --- a/libcuda/cuda_runtime_api.cc +++ b/libcuda/cuda_runtime_api.cc @@ -349,24 +349,13 @@ class _cuda_device_id *GPGPUSim_Init() prop->maxGridSize[2] = 0x40000000; prop->totalConstMem = 0x40000000; prop->textureAlignment = 0; - /* - - * TODO: Update the .config and xml files of all GPU config files with new value of sharedMemPerBlock. - - * Previously, this was thought as sharedMemPerMultiprocessor and is being used in many places. - - * Check whether all the instances of shared_mem_size(), gpgpu_shmem_size or sharedMemPerBlock are meant to use sharedMemPerBlock or sharedMemPerMultiprocessor. - - */ - - prop->sharedMemPerBlock = the_gpu->shared_mem_per_block(); - - #if (CUDART_VERSION > 5000) - - prop->sharedMemPerMultiprocessor = the_gpu->shared_mem_size(); - - #endif - +// * TODO: Update the .config and xml files of all GPU config files with new value of sharedMemPerBlock and regsPerBlock + prop->sharedMemPerBlock = the_gpu->shared_mem_per_block(); +#if (CUDART_VERSION > 5000) + prop->regsPerMultiprocessor = the_gpu->num_registers_per_core(); + prop->sharedMemPerMultiprocessor = the_gpu->shared_mem_size(); +#endif + prop->sharedMemPerBlock = the_gpu->shared_mem_per_block(); prop->regsPerBlock = the_gpu->num_registers_per_core(); prop->warpSize = the_gpu->wrp_size(); prop->clockRate = the_gpu->shader_clock(); @@ -912,7 +901,7 @@ __host__ cudaError_t CUDARTAPI cudaDeviceGetAttribute(int *value, enum cudaDevic *value= prop->sharedMemPerMultiprocessor; break; case 82: - *value= prop->regsPerBlock; + *value= prop->regsPerMultiprocessor; break; default: printf("ERROR: Attribute number %d unimplemented \n",attr); |