furyhawk/ik_llama.cpp
1
0
Fork 0
mirror of https://github.com/ikawrakow/ik_llama.cpp.git synced 2026-07-21 02:05:35 +00:00
Code Issues Packages Projects Releases Wiki Activity

Add missing contiguity check (#2080)

Browse Source
This commit is contained in:
Farmadupe
2026-07-06 17:55:14 +02:00
committed by GitHub
parent b569706eae
commit 2992537ca3
1 changed files with 5 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files
ggml/src/ggml-cuda/binbcast.cu
+5 -2
View File
@@ -434,6 +434,7 @@ void ggml_cuda_op_add(ggml_backend_cuda_context & ctx, ggml_tensor * dst) {
}
if (ggml_nrows(dst->src[1]) == 1 && dst->src[0]->ne[0] == dst->src[1]->ne[0] &&
dst->type == GGML_TYPE_F32 && dst->src[0]->type == GGML_TYPE_F32 && dst->src[1]->type == GGML_TYPE_F32 &&
ggml_is_contiguous(dst->src[0]) && ggml_is_contiguous(dst->src[1]) &&
ggml_are_same_shape(dst, dst->src[0]) && ggml_is_contiguous(dst)) {
constexpr int kBlockSize = 256;
auto nelem = ggml_nelements(dst);
@@ -442,7 +443,8 @@ void ggml_cuda_op_add(ggml_backend_cuda_context & ctx, ggml_tensor * dst) {
(const float *)dst->src[0]->data, (const float *)dst->src[1]->data, (float *)dst->data);
return;
}
if (ggml_is_contiguous(dst->src[0]) && ggml_are_same_shape(dst->src[0], dst->src[1]) && ggml_is_contiguous(dst)) {
if (ggml_is_contiguous(dst->src[0]) && ggml_is_contiguous(dst->src[1]) &&
ggml_are_same_shape(dst->src[0], dst->src[1]) && ggml_is_contiguous(dst)) {
constexpr int kBlockSize = 256;
auto nelem = ggml_nelements(dst);
int nblocks = (nelem + kBlockSize - 1)/kBlockSize;
@@ -585,7 +587,8 @@ void ggml_cuda_op_mul(ggml_backend_cuda_context & ctx, ggml_tensor * dst) {
}
auto src0 = dst->src[0];
auto src1 = dst->src[1];
if (ggml_is_contiguous(src0) && ggml_is_contiguous(src1) && src0->type == GGML_TYPE_F32 && src1->type == GGML_TYPE_F32 && dst->type == GGML_TYPE_F32 &&
if (ggml_is_contiguous(src0) && ggml_is_contiguous(src1) && ggml_is_contiguous(dst) &&
src0->type == GGML_TYPE_F32 && src1->type == GGML_TYPE_F32 && dst->type == GGML_TYPE_F32 &&
src1->ne[0] == 1 && src0->ne[1] == src1->ne[1] && src0->ne[2] == src1->ne[2] && src0->ne[3] == src1->ne[3]) {
constexpr int kBlockSize = 256;
int nelem = ggml_nelements(src0);