ver4a/llama.cpp
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity

CUDA: fix race condition in FA vector kernels (#13742)

Browse source
This commit is contained in:
Johannes Gäßler 2025-05-24 11:46:19 +02:00 • committed by GitHub
parent b775345d78
commit ffd0eae60b
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
2 changed files with 2 additions and 0 deletions
Download patch file Download diff file Expand all files Collapse all files

1
ggml/src/ggml-cuda/fattn-vec-f16.cuh
View file

@ -212,6 +212,7 @@ static __global__ void flash_attn_vec_ext_f16(
} }
} }
if (__all_sync(0xFFFFFFFF, skip)) { if (__all_sync(0xFFFFFFFF, skip)) {
__syncthreads();
continue; continue;
} }
#endif // GGML_USE_HIP #endif // GGML_USE_HIP

1
ggml/src/ggml-cuda/fattn-vec-f32.cuh
View file

@ -217,6 +217,7 @@ static __global__ void flash_attn_vec_ext_f32(
} }
} }
if (__all_sync(0xFFFFFFFF, skip)) { if (__all_sync(0xFFFFFFFF, skip)) {
__syncthreads();
continue; continue;
} }
#endif // GGML_USE_HIP #endif // GGML_USE_HIP