CS348Project/llama.cpp
5
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2025-10-28 08:31:25 +00:00
Code Issues Packages Projects Releases Wiki Activity

CUDA: better error for FA kernel with 0 occupancy (#16643)

Browse Source
This commit is contained in:
Johannes Gäßler
2025-10-21 15:27:53 +02:00
committed by GitHub
parent 4926419c4d
commit 51d1a8c997
1 changed files with 1 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files

1
ggml/src/ggml-cuda/fattn-common.cuh
View File

@@ -895,6 +895,7 @@ void launch_fattn(
const dim3 block_dim(warp_size, nwarps, 1); const dim3 block_dim(warp_size, nwarps, 1);
int max_blocks_per_sm = 1; // Max. number of active blocks limited by occupancy. int max_blocks_per_sm = 1; // Max. number of active blocks limited by occupancy.
CUDA_CHECK(cudaOccupancyMaxActiveBlocksPerMultiprocessor(&max_blocks_per_sm, fattn_kernel, block_dim.x * block_dim.y * block_dim.z, nbytes_shared)); CUDA_CHECK(cudaOccupancyMaxActiveBlocksPerMultiprocessor(&max_blocks_per_sm, fattn_kernel, block_dim.x * block_dim.y * block_dim.z, nbytes_shared));
GGML_ASSERT(max_blocks_per_sm > 0);
int parallel_blocks = max_blocks_per_sm; int parallel_blocks = max_blocks_per_sm;
dim3 blocks_num; dim3 blocks_num;