Merge pull request #11615 from seberg/clip_order

BUG: Clip uses wrong memory order in output 

2 files changed, 39 insertions, 14 deletions

diff --git a/numpy/core/src/multiarray/calculation.c b/numpy/core/src/multiarray/calculation.c
index e47dd81b9..90ee2c5b2 100644
--- a/numpy/core/src/multiarray/calculation.c
+++ b/numpy/core/src/multiarray/calculation.c
@@ -5,6 +5,7 @@
 #define NPY_NO_DEPRECATED_API NPY_API_VERSION
 #define _MULTIARRAYMODULE
 #include "numpy/arrayobject.h"
+#include "lowlevel_strided_loops.h"
 
 #include "npy_config.h"
 
@@ -1102,7 +1103,18 @@ PyArray_Clip(PyArrayObject *self, PyObject *min, PyObject *max, PyArrayObject *o
     if (out == newin) {
         outgood = 1;
     }
-    if (!outgood && PyArray_ISONESEGMENT(out) &&
+
+
+    /* make sure the shape of the output array is the same */
+    if (!PyArray_SAMESHAPE(newin, out)) {
+        PyErr_SetString(PyExc_ValueError, "clip: Output array must have the"
+                        "same shape as the input.");
+        goto fail;
+    }
+
+    if (!outgood && PyArray_EQUIVALENTLY_ITERABLE(
+                            self, out, PyArray_TRIVIALLY_ITERABLE_OP_READ,
+                            PyArray_TRIVIALLY_ITERABLE_OP_NOREAD) &&
                         PyArray_CHKFLAGS(out, NPY_ARRAY_ALIGNED) &&
                         PyArray_ISNOTSWAPPED(out) &&
                         PyArray_EquivTypes(PyArray_DESCR(out), indescr)) {
@@ -1111,15 +1123,19 @@ PyArray_Clip(PyArrayObject *self, PyObject *min, PyObject *max, PyArrayObject *o
 
     /*
      * Do we still not have a suitable output array?
-     * Create one, now
+     * Create one, now. No matter why the array is not suitable a copy has
+     * to be made. This may be just to avoid memory overlap though.
      */
     if (!outgood) {
         int oflags;
-        if (PyArray_ISFORTRAN(out))
+        if (PyArray_ISFORTRAN(self)) {
             oflags = NPY_ARRAY_FARRAY;
-        else
+        }
+        else {
             oflags = NPY_ARRAY_CARRAY;
-        oflags |= NPY_ARRAY_WRITEBACKIFCOPY | NPY_ARRAY_FORCECAST;
+        }
+        oflags |= (NPY_ARRAY_WRITEBACKIFCOPY | NPY_ARRAY_FORCECAST |
+                   NPY_ARRAY_ENSURECOPY);
         Py_INCREF(indescr);
         newout = (PyArrayObject*)PyArray_FromArray(out, indescr, oflags);
         if (newout == NULL) {
@@ -1131,13 +1147,6 @@ PyArray_Clip(PyArrayObject *self, PyObject *min, PyObject *max, PyArrayObject *o
         Py_INCREF(newout);
     }
 
-    /* make sure the shape of the output array is the same */
-    if (!PyArray_SAMESHAPE(newin, newout)) {
-        PyErr_SetString(PyExc_ValueError, "clip: Output array must have the"
-                        "same shape as the input.");
-        goto fail;
-    }
-
     /* Now we can call the fast-clip function */
     min_data = max_data = NULL;
     if (mina != NULL) {
diff --git a/numpy/core/tests/test_numeric.py b/numpy/core/tests/test_numeric.py
index 53486dc51..a1a92ef32 100644
--- a/numpy/core/tests/test_numeric.py
+++ b/numpy/core/tests/test_numeric.py
@@ -1530,7 +1530,7 @@ class TestClip(object):
         m = -0.5
         M = 0.6
         self.fastclip(a, m, M, a)
-        self.clip(a, m, M, ac)
+        self.clip(ac, m, M, ac)
         assert_array_strict_equal(a, ac)
 
     def test_noncontig_inplace(self):
@@ -1543,7 +1543,7 @@ class TestClip(object):
         m = -0.5
         M = 0.6
         self.fastclip(a, m, M, a)
-        self.clip(a, m, M, ac)
+        self.clip(ac, m, M, ac)
         assert_array_equal(a, ac)
 
     def test_type_cast_01(self):
@@ -1722,6 +1722,22 @@ class TestClip(object):
         self.clip(a, m, M, act)
         assert_array_strict_equal(ac, act)
 
+    def test_clip_with_out_transposed(self):
+        # Test that the out argument works when tranposed
+        a = np.arange(16).reshape(4, 4)
+        out = np.empty_like(a).T
+        a.clip(4, 10, out=out)
+        expected = self.clip(a, 4, 10)
+        assert_array_equal(out, expected)
+
+    def test_clip_with_out_memory_overlap(self):
+        # Test that the out argument works when it has memory overlap
+        a = np.arange(16).reshape(4, 4)
+        ac = a.copy()
+        a[:-1].clip(4, 10, out=a[1:])
+        expected = self.clip(ac[:-1], 4, 10)
+        assert_array_equal(a[1:], expected)
+
     def test_clip_inplace_array(self):
         # Test native double input with array min/max
         a = self._generate_data(self.nr, self.nc)