happyz
/
llama.cpp
mirror of https://github.com/ggerganov/llama.cpp.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

make sure there are enough channels for split-k

This commit is contained in:
bssrdf 2025-11-06 10:21:49 -05:00
parent 09e3a5f07d
commit 311213d209
2 changed files with 3 additions and 3 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

5
ggml/src/ggml-cuda/conv2d-implicit.cu
View File

@ -991,9 +991,8 @@ static void conv2d_implicit_cuda_f16(ggml_backend_cuda_context & ctx, const floa
const unsigned int shmem_bytes = (BM_dim * BK_dim + BK_dim * BN_dim) * 2 * sizeof(half);
const int nsm = ggml_cuda_info().devices[ggml_cuda_get_device()].nsm;
if (BlocksM * BlocksN < nsm) {
const unsigned int ksplit = 8;
const unsigned int ksplit = 8;
if (BlocksM * BlocksN < nsm && P.c > 8 * ksplit) {
ggml_cuda_pool_alloc<half> Y_H(ctx.pool(id), ksplit * P.k * P.Oh * P.Ow * P.n);
cudaFuncSetAttribute(conv2d_implicit_kernel<BM_dim, BN_dim, BK_dim, WM_dim, WN_dim, WK_dim, ksplit, NumThreads>,

1
tests/test-conv2d.cpp
View File

@ -324,6 +324,7 @@ int main(void)
std::make_tuple(256,128,832,1216,3,3),
std::make_tuple(256,256,832,1216,3,3),
// std::make_tuple(320,256,1024,1920)
// std::make_tuple(32,64,58,58,3,3)
};
int k = 0;