ikawrakow/ik_llama.cpp
1
0
Fork 0
mirror of https://github.com/ikawrakow/ik_llama.cpp.git synced 2026-04-30 19:31:48 +00:00
Code Issues Packages Projects Releases Wiki Activity

Fix new CUDA FA on Touring (#413)

Browse Source 
Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
This commit is contained in:
Kawrakow
2025-05-12 15:09:33 +03:00
committed by GitHub
parent 1d2da7feae
commit 627f406437
1 changed files with 2 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
ggml/src/ggml-cuda/fattn-new-mma.cu
View File

@@ -273,11 +273,11 @@ struct fattn_mma_f16_config<576, 512> {
} }
static constexpr __device__ int get_nbatch_V2_device(int ncols) { static constexpr __device__ int get_nbatch_V2_device(int ncols) {
#if __CUDA_ARCH__ == GML_CUDA_CC_TURING #if __CUDA_ARCH__ == CC_TURING
return ncols <= 16 ? 64 : 128; return ncols <= 16 ? 64 : 128;
#else #else
return ncols <= 16 ? 256 : 128; return ncols <= 16 ? 256 : 128;
#endif // __CUDA_ARCH__ == GML_CUDA_CC_TURING #endif // __CUDA_ARCH__ == CC_TURING
} }
static int get_nbatch_combine_host(const int /*cc*/, const int /*ncols*/) { static int get_nbatch_combine_host(const int /*cc*/, const int /*ncols*/) {