Skip to content

Fix col_reduce_longcolumn for negative-stride views #4529

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

Open
mateuuszzzzz wants to merge 2 commits into
base: main
Choose a base branch
from
+29 −1
Open

Fix col_reduce_longcolumn for negative-stride views #4529

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
5b8c7dc
Fix Metal col_reduce_longcolumn for negative-stride views
mateuuszzzzz Sep 17, 2026
54043df
Add tests covering col_reduce_* for negative-stride views
mateuuszzzzz Sep 17, 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/reduction/reduce_col.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -117,7 +117,7 @@ template <typename T, typename U, typename Op, typename IdxT, int NDIMS>

IdxT out_idx = gid.x + gsize.x * IdxT(gid.y);
IdxT in_idx = elem_to_loc<IdxT>(out_idx, shape, strides, ndim);
in += in_idx + lid.x;
in += in_idx + IdxT(lid.x);

U total = Op::init;
IdxT total_rows = IdxT(non_col_reductions) * IdxT(reduction_size);
Expand Down
28 changes: 28 additions & 0 deletions python/tests/test_reduce.py
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,5 +1,6 @@
# Copyright © 2023 Apple Inc.

import math
from itertools import combinations, permutations

import mlx.core as mx
Expand Down Expand Up @@ -56,6 +57,33 @@ def test_row_reduce_negative_stride(self):
actual = getattr(mx, op)(x_mlx, axis=-1)
self.assertTrue(np.allclose(expected, actual))

def test_col_reduce_negative_stride(self):
# Exercises each Metal column reduction kernel on a negative-stride view.
cases = [
((2, 1024, 16), 1),
((2, 2, 1024, 16), 2),
((2, 512, 2, 2, 16), (1, 3)),
((2, 512, 64), 1),
((2, 64, 512), 1),
((2, 16, 16), 1),
]
for shape, axis in cases:
size = math.prod(shape)
x_npy = np.arange(1, size + 1).reshape(shape)[::-1]
x_mlx = mx.arange(1, size + 1).reshape(shape)[::-1]
for op in ["sum", "max", "min", "mean", "var"]:
with self.subTest(shape=shape, axis=axis, op=op):
expected = getattr(np, op)(x_npy, axis=axis)
actual = getattr(mx, op)(x_mlx, axis=axis)
self.assertTrue(np.allclose(expected, actual))
x_ones = mx.ones(shape, dtype=mx.float32)[::-1]
with self.subTest(shape=shape, axis=axis, op="prod"):
self.assertTrue(mx.all(mx.prod(x_ones, axis=axis) == 1).item())
x_bool = mx.ones(shape, dtype=mx.bool_)[::-1]
for op in ["all", "any"]:
with self.subTest(shape=shape, axis=axis, op=op):
self.assertTrue(mx.all(getattr(mx, op)(x_bool, axis=axis)).item())

def test_dtypes(self):
int_dtypes = [
"int8",
Expand Down