Lumpiasty/llama.cpp
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

CUDA: fix crash on uneven context without FA (#16988)

Browse Source
This commit is contained in:
Johannes Gäßler
2025-11-06 14:05:47 +01:00
committed by GitHub
parent 5b180c3d60
commit aa374175c3
7 changed files with 44 additions and 38 deletions
Download Patch File Download Diff File Expand all files Collapse all files
ggml/src/ggml-cuda/mmvf.cu
+7 -1
View File
@@ -716,10 +716,16 @@ void ggml_cuda_op_mul_mat_vec_f(
GGML_UNUSED_VARS(ctx, src1, dst, src1_ddq_i, src1_ncols, src1_padded_row_size);
}
bool ggml_cuda_should_use_mmvf(enum ggml_type type, int cc, const int64_t * src0_ne, int64_t ne11) {
bool ggml_cuda_should_use_mmvf(enum ggml_type type, int cc, const int64_t * src0_ne, const size_t * src0_nb, int64_t ne11) {
if (src0_ne[0] % 2 != 0) {
return false;
}
const size_t ts = ggml_type_size(type);
for (size_t i = 0; i < GGML_MAX_DIMS; ++i) {
if (src0_nb[i] % (2*ts) != 0) {
return false;
}
}
switch (type) {
case GGML_TYPE_F32:
if (GGML_CUDA_CC_IS_NVIDIA(cc)) {