Skip to content

ggml: fix cuda kernel launch configuration for k_compute_batched_ptrs to support large batch #16744

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 4 commits into from
Oct 26, 2025
+12 −2
Merged

ggml: fix cuda kernel launch configuration for k_compute_batched_ptrs to support large batch #16744

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
c2cbb16
fix k_compute_batched_ptrs
leejet Oct 23, 2025
e707662
add backend ops test
leejet Oct 23, 2025
96e2320
Update ggml/src/ggml-cuda/ggml-cuda.cu
leejet Oct 24, 2025
1da9854
reduce the batch size
leejet Oct 24, 2025
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
11 changes: 9 additions & 2 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 @@ -1957,8 +1957,15 @@ static void ggml_cuda_mul_mat_batched_cublas_impl(ggml_backend_cuda_context & ct

size_t src1_stride_size = sizeof(cuda_t);

dim3 block_dims(ne13, ne12);
k_compute_batched_ptrs<<<1, block_dims, 0, main_stream>>>(
const int threads_x = 16;
const int threads_y = 16;
dim3 block_dims(threads_x, threads_y);

dim3 grid_dims(
(ne13 + threads_x - 1) / threads_x,
(ne12 + threads_y - 1) / threads_y
);
k_compute_batched_ptrs<<<grid_dims, block_dims, 0, main_stream>>>(
src0_ptr, src1_ptr, dst_t,
ptrs_src.get(), ptrs_dst.get(),
ne12, ne13,
Expand Down
3 changes: 3 additions & 0 deletions tests/test-backend-ops.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -6562,6 +6562,9 @@ static std::vector<std::unique_ptr<test_case>> make_test_cases_eval() {
test_cases.emplace_back(new test_mul_mat(type_a, type_b, 16, 1, 1024, {3, 2}, {1, 1}));
test_cases.emplace_back(new test_mul_mat(type_a, type_b, 16, 8, 1024, {3, 2}, {1, 1}));
test_cases.emplace_back(new test_mul_mat(type_a, type_b, 16, 16, 1024, {3, 2}, {1, 1}));

// test cases with large batch size
test_cases.emplace_back(new test_mul_mat(type_a, type_b, 16, 8, 256, {1536, 1}, {1, 1}));
}
}
for (ggml_type type_a : other_types) {
Expand Down
Loading