diff options
| author | Tim Rogers <[email protected]> | 2019-08-23 15:33:38 -0400 |
|---|---|---|
| committer | GitHub <[email protected]> | 2019-08-23 15:33:38 -0400 |
| commit | 0df569774615120c76fb5f139f8698175a722293 (patch) | |
| tree | b8a1ca8e57bffc08db5ed2378990ec53353af11e /src/abstract_hardware_model.h | |
| parent | 2f5b3332c9b9b3fa9fea43d61276bddb24aa7df2 (diff) | |
| parent | 963947b33c99143afbb477a6d897245e56695b0b (diff) | |
Merge pull request #26 from mkhairy/dev
Merging Mahmoud's CUTLASS support and correlation fixes
Diffstat (limited to 'src/abstract_hardware_model.h')
| -rw-r--r-- | src/abstract_hardware_model.h | 7 |
1 files changed, 4 insertions, 3 deletions
diff --git a/src/abstract_hardware_model.h b/src/abstract_hardware_model.h index d13b8c6..8c19e33 100644 --- a/src/abstract_hardware_model.h +++ b/src/abstract_hardware_model.h @@ -212,6 +212,7 @@ public: // m_num_cores_running=0; // m_param_mem=NULL; // } + kernel_info_t( dim3 gridDim, dim3 blockDim, class function_info *entry); kernel_info_t( dim3 gridDim, dim3 blockDim, class function_info *entry, std::map<std::string, const struct cudaArray*> nameToCudaArray, std::map<std::string, const struct textureInfo*> nameToTextureInfo); ~kernel_info_t(); @@ -443,9 +444,9 @@ protected: #define GLOBAL_HEAP_START 0xC0000000 // start allocating from this address (lower values used for allocating globals in .ptx file) -#define SHARED_MEM_SIZE_MAX (64*1024) -#define LOCAL_MEM_SIZE_MAX (8*1024) -#define MAX_STREAMING_MULTIPROCESSORS 64 +#define SHARED_MEM_SIZE_MAX (96*1024) +#define LOCAL_MEM_SIZE_MAX (16*1024) +#define MAX_STREAMING_MULTIPROCESSORS 80 //scale it to Volta #define MAX_THREAD_PER_SM 2048 #define MAX_WARP_PER_SM 64 #define TOTAL_LOCAL_MEM_PER_SM (MAX_THREAD_PER_SM*LOCAL_MEM_SIZE_MAX) |