Merge pull request #29191 from charris/backport-29179

charris · web-flow · commit 270eef72d823 · 2025-06-12T08:44:04.000-06:00
BUG: fix matmul with transposed out arg (#29179)
diff --git a/doc/release/upcoming_changes/29179.change.rst b/doc/release/upcoming_changes/29179.change.rst
@@ -0,0 +1,4 @@
+Fix bug in ``matmul`` for non-contiguous out kwarg parameter
+------------------------------------------------------------
+In some cases, if ``out`` was non-contiguous, ``np.matmul`` would cause
+memory corruption or a c-level assert. This was new to v2.3.0 and fixed in v2.3.1.
diff --git a/doc/source/release/2.3.0-notes.rst b/doc/source/release/2.3.0-notes.rst
@@ -414,6 +414,12 @@ the best performance.
 
 (`gh-28769 <https://github.com/numpy/numpy/pull/28769>`__)
 
+Performance improvements for ``np.matmul``
+------------------------------------------
+Enable using BLAS for ``matmul`` even when operands are non-contiguous by copying
+if needed.
+
+(`gh-23752 <https://github.com/numpy/numpy/pull/23752>`__)
 
 Changes
 =======
diff --git a/numpy/_core/src/umath/matmul.c.src b/numpy/_core/src/umath/matmul.c.src
@@ -596,7 +596,7 @@ NPY_NO_EXPORT void
              * Use transpose equivalence:
              * matmul(a, b, o) == matmul(b.T, a.T, o.T)
              */
-            if (o_f_blasable) {
+            if (o_transpose) {
                 @TYPE@_matmul_matrixmatrix(
                     ip2_, is2_p_, is2_n_,
                     ip1_, is1_n_, is1_m_,
diff --git a/numpy/_core/tests/test_multiarray.py b/numpy/_core/tests/test_multiarray.py
@@ -7317,6 +7317,12 @@ def test_dot_equivalent(self, args):
         r3 = np.matmul(args[0].copy(), args[1].copy())
         assert_equal(r1, r3)
 
+        # matrix matrix, issue 29164
+        if [len(args[0].shape), len(args[1].shape)] == [2, 2]:
+            out_f = np.zeros((r2.shape[0] * 2, r2.shape[1] * 2), order='F')
+            r4 = np.matmul(*args, out=out_f[::2, ::2])
+            assert_equal(r2, r4)
+
     def test_matmul_object(self):
         import fractions
 
