frozenleaves/pytorch
1
0
Fork 0
mirror of https://github.com/pytorch/pytorch.git synced 2025-10-20 21:14:14 +08:00
Code Issues Packages Projects Releases Wiki Activity

Fix CUDA_MAX_THREADS_PER_SM for sm_87 (#88644)

Browse Source 
#88326
CC @ngimel @ptrblck

Pull Request resolved: https://github.com/pytorch/pytorch/pull/88644
Approved by: https://github.com/ngimel
This commit is contained in:
Eddie Yan
2022-11-08 19:44:23 +00:00
committed by PyTorch MergeBot
parent 6bb7f4f29f
commit 3e30a9ea1c
1 changed files with 5 additions and 5 deletions
Download Patch File Download Diff File Expand all files Collapse all files

10
c10/macros/Macros.h
View File

@ -255,13 +255,13 @@ using namespace c10::hip;
// constants from
// (https://docs.nvidia.com/cuda/cuda-c-programming-guide/index.html#features-and-technical-specifications)
// The maximum number of threads per multiprocessor is 1024 for Turing
// architecture (7.5), 1536 for Geforce Ampere (8.6), and 2048 for all other
// architectures. You'll get warnings if you exceed these constants. Hence, the
// following macros adjust the input values from the user to resolve potential
// warnings.
// architecture (7.5), 1536 for Geforce Ampere (8.6)/Jetson Orin (8.7), and
// 2048 for all other architectures. You'll get warnings if you exceed these
// constants. Hence, the following macros adjust the input values from the user
// to resolve potential warnings.
#if __CUDA_ARCH__ == 750
constexpr uint32_t CUDA_MAX_THREADS_PER_SM = 1024;
#elif __CUDA_ARCH__ == 860
#elif __CUDA_ARCH__ == 860 || __CUDA_ARCH__ == 870
constexpr uint32_t CUDA_MAX_THREADS_PER_SM = 1536;
#else
constexpr uint32_t CUDA_MAX_THREADS_PER_SM = 2048;