Skip to content

CUDA: avoid mul + bias fusion when buffers are split #16935

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 1 commit into from
Nov 4, 2025
Merged

CUDA: avoid mul + bias fusion when buffers are split #16935

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
17 changes: 17 additions & 0 deletions ggml/src/ggml-cuda/ggml-cuda.cu
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -2115,6 +2115,14 @@ static bool ggml_cuda_should_fuse_mul_mat_vec_f(const ggml_tensor * tensor) {
const int cc = ggml_cuda_info().devices[ggml_cuda_get_device()].cc;
use_mul_mat_vec_f = use_mul_mat_vec_f && ggml_cuda_should_use_mmvf(src0->type, cc, src0->ne, is_mul_mat_id ? src1->ne[2] : src1->ne[1]);

const bool split = ggml_backend_buft_is_cuda_split(src0->buffer->buft) ||
ggml_backend_buft_is_cuda_split(src1->buffer->buft);

//TODO: add support for fusion for split buffers
if (split) {
return false;
}

//we only support fusion for ncols_dst = 1
if (tensor->op == GGML_OP_MUL_MAT && dst->ne[1] != 1) {
return false;
Expand Down Expand Up @@ -2154,6 +2162,15 @@ static bool ggml_cuda_should_fuse_mul_mat_vec_q(const ggml_tensor * tensor) {
return false;
}


const bool split = ggml_backend_buft_is_cuda_split(src0->buffer->buft) ||
ggml_backend_buft_is_cuda_split(src1->buffer->buft);

//TODO: add support for fusion for split buffers
if (split) {
return false;
}

return use_mul_mat_vec_q;
}

Expand Down
Loading