diff options
| author | Suchita Pati <[email protected]> | 2018-04-10 14:52:49 -0500 |
|---|---|---|
| committer | Suchita Pati <[email protected]> | 2018-04-10 14:52:49 -0500 |
| commit | 960a1492fbbf2a4d098c8d060bc22782211ca60b (patch) | |
| tree | 4b4f02d73c427ccfbac8fd372cb9e759f5a2bdc5 /libcuda/cuda_runtime_api.cc | |
| parent | 6c9555d42f3ab7f11dd003a9271d0c7e0ad21031 (diff) | |
added config -gpgpu_shmem_per_block and fixed cudaGetDeviceAttributes function
Diffstat (limited to 'libcuda/cuda_runtime_api.cc')
| -rw-r--r-- | libcuda/cuda_runtime_api.cc | 31 |
1 files changed, 27 insertions, 4 deletions
diff --git a/libcuda/cuda_runtime_api.cc b/libcuda/cuda_runtime_api.cc index bcc3e14..d6ce4e0 100644 --- a/libcuda/cuda_runtime_api.cc +++ b/libcuda/cuda_runtime_api.cc @@ -349,7 +349,24 @@ class _cuda_device_id *GPGPUSim_Init() prop->maxGridSize[2] = 0x40000000; prop->totalConstMem = 0x40000000; prop->textureAlignment = 0; - prop->sharedMemPerBlock = the_gpu->shared_mem_size(); + /* + + * 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 + prop->regsPerBlock = the_gpu->num_registers_per_core(); prop->warpSize = the_gpu->wrp_size(); prop->clockRate = the_gpu->shader_clock(); @@ -861,6 +878,9 @@ __host__ cudaError_t CUDARTAPI cudaDeviceGetAttribute(int *value, enum cudaDevic case 8: *value= prop->sharedMemPerBlock; break; + case 9: + *value= prop->totalConstMem; + break; case 10: *value= prop->warpSize; break; @@ -873,11 +893,14 @@ __host__ cudaError_t CUDARTAPI cudaDeviceGetAttribute(int *value, enum cudaDevic case 16: *value= prop->multiProcessorCount ; break; + case 34: + *value= 0; + break; case 39: *value= dev->get_gpgpu()->threads_per_core(); break; case 75: - *value= 8 ; + *value= 9 ; break; case 76: *value= 3 ; @@ -886,13 +909,13 @@ __host__ cudaError_t CUDARTAPI cudaDeviceGetAttribute(int *value, enum cudaDevic *value= 0 ; //TODO: as of now, we dont support stream priorities. break; case 81: - *value= prop->sharedMemPerBlock; + *value= prop->sharedMemPerMultiprocessor; break; case 82: *value= prop->regsPerBlock; break; default: - printf("ERROR: implement the attribute numbered %d \n",attr); + printf("ERROR: Attribute number %d unimplemented \n",attr); abort(); } return g_last_cudaError = cudaSuccess; |