Skip to content

Fix incorrect output mask computation in mx.block_masked_mm #4652

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
zcbenz merged 2 commits into from
Oct 9, 2026
Merged

Fix incorrect output mask computation in mx.block_masked_mm #4652

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
36d88a2
Fix incorrect output mask computation in `mx.block_masked_mm`
michalk8 Sep 30, 2026
4faf3f6
Merge branch 'main' into fix-gemv-masked-out-mask-stride
michalk8 Oct 8, 2026
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 mlx/backend/metal/kernels/gemv_masked.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -452,7 +452,7 @@ struct GEMVTKernel {
const int n_block_idx = blockM > blockN ? out_col / blockM : int(tid.x);

const int out_mask_offset =
!has_output_mask ? 0 : n_block_idx; // * out_mask_strides[0];
!has_output_mask ? 0 : n_block_idx * out_mask_strides[0];

int mat_mask_offset =
!has_operand_mask ? 0 : n_block_idx * mat_mask_strides[0];
Expand Down
9 changes: 9 additions & 0 deletions python/tests/test_blas.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1241,6 +1241,15 @@ def test_shape(
self.assertTrue(np.allclose(c_mx, c_np, atol=1e-5))
self.assertTrue(np.allclose(e_mx, e_np, atol=1e-5))

# Strided output mask with a vector lhs
a = mx.random.normal((1, 64))
b = mx.random.normal((64, 128))
mask = mx.array([[True, False, False, True, True, False, False, True]])
out_mask = mask[:, ::2]
out = mx.block_masked_mm(a, b, 32, out_mask)
expected = ref_block_masked_mm(a, b, 32, out_mask)
self.assertTrue(mx.allclose(out, expected, atol=1e-5))

def test_gather_matmul(self):
def np_gather_mm(a, b, lhs_indices=None, rhs_indices=None):
a = a.reshape((-1, a.shape[-2], a.shape[-1]))
Expand Down
Loading