Skip to content

ggml: fix CUDA grid launch condition for large block_nums.y in binbcast #16742

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 3 commits into from
Oct 24, 2025
Merged

ggml: fix CUDA grid launch condition for large block_nums.y in binbcast #16742

Show file tree
Hide file tree
Changes from 2 commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
c0c1a0e
Fix CUDA grid launch condition for large block_nums.y
leejet Oct 23, 2025
d28bb87
add backend ops test
leejet Oct 23, 2025
62ae943
reduce test repetitions
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
2 changes: 1 addition & 1 deletion ggml/src/ggml-cuda/binbcast.cu
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -272,7 +272,7 @@ static void launch_bin_bcast_pack(const ggml_tensor * src0, const ggml_tensor *
const uint3 ne12 = init_fastdiv_values((uint32_t) cne1[2]);
const uint3 ne13 = init_fastdiv_values((uint32_t) cne1[3]);

if (block_nums.z > 65535) {
if (block_nums.z > 65535 || block_nums.y > 65535) {
int block_num = (ne0 * ne1 * ne2 * ne3 + block_size - 1) / block_size;
const uint3 prod_012 = init_fastdiv_values((uint32_t) (ne0 * ne1 * ne2));
const uint3 prod_01 = init_fastdiv_values((uint32_t) (ne0 * ne1));
Expand Down
1 change: 1 addition & 0 deletions tests/test-backend-ops.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -6407,6 +6407,7 @@ static std::vector<std::unique_ptr<test_case>> make_test_cases_eval() {
add_test_bin_bcast(type, {1, 1, 640, 1}, {32, 32, 1, 1});
add_test_bin_bcast(type, {5120, 1, 1, 1}, {1, 256, 1, 1});
add_test_bin_bcast(type, {640, 1, 1, 1}, {1, 1, 1, 1});
add_test_bin_bcast(type, {64, 262144, 1, 1}, {64, 1, 1, 1});
Copy link
Collaborator

@jeffbolznv jeffbolznv Oct 23, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Thanks. This one is kind of slow. Would it hit the same bug with smaller ne[0]?

Copy link
Contributor Author

@leejet leejet Oct 23, 2025 •
edited
Loading

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

The smaller ne[0] requires a larger ne[1]. block_nums.y = ne1*ne0 / 256 when ne0 <= 128

        const int block_size = 128;

        int64_t hne0 = std::max(ne0 / 2LL, 1LL);

        dim3 block_dims;
        block_dims.x = std::min<unsigned int>(hne0, block_size);
        block_dims.y = std::min<unsigned int>(ne1, block_size / block_dims.x);
        block_dims.z = std::min(std::min<unsigned int>(ne2 * ne3, block_size / block_dims.x / block_dims.y), 64U);

        dim3 block_nums((hne0 + block_dims.x - 1) / block_dims.x, (ne1 + block_dims.y - 1) / block_dims.y,
                        (ne2 * ne3 + block_dims.z - 1) / block_dims.z);

Copy link
Collaborator

@jeffbolznv jeffbolznv Oct 23, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Ah, OK. I'm fine with it then.

Copy link
Member

@slaren slaren Oct 23, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

The second parameter of add_test_bin_bcast is the number of repetitions, so this results in a 4 GB tensor, and it is causing the CI to fail. Using {1,1,1,1} should be enough to test this.

Copy link
Contributor Author

@leejet leejet Oct 24, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Thanks for your review. I’ve now reduced the repetitions.

//add_test_bin_bcast(type, {3, 3, 2560, 1280}, {1, 1, 1, 1});
//add_test_bin_bcast(type, {3, 3, 2560, 1280}, {2, 1, 1, 1});
}
Expand Down
Loading