BUG: check alignment before loading the data

some cpus don't support loads that are not aligned to the elementsize.
regression introduced in 99cb95f7379a9.

1 files changed, 30 insertions, 26 deletions

diff --git a/numpy/core/src/multiarray/ctors.c b/numpy/core/src/multiarray/ctors.c
index b1a9d9859..5c692bd02 100644
--- a/numpy/core/src/multiarray/ctors.c
+++ b/numpy/core/src/multiarray/ctors.c
@@ -311,38 +311,42 @@ _strided_byte_swap(void *p, npy_intp stride, npy_intp n, int size)
     case 1: /* no byteswap necessary */
         break;
     case 4:
-        for (a = (char*)p; n > 0; n--, a += stride) {
-            npy_uint32 * a_ = (npy_uint32 *)a;
-#ifdef HAVE___BUILTIN_BSWAP32
-            *a_ = __builtin_bswap32(*a_);
-#else
-            /* a decent compiler can convert this to bswap too */
-            *a_ = ((*a_ & 0xff000000u) >> 24) | ((*a_ & 0x00ff0000u) >>  8) |
-                  ((*a_ & 0x0000ff00u) <<  8) | ((*a_ & 0x000000ffu) << 24);
-#endif
+        if (npy_is_aligned(p, sizeof(npy_uint32))) {
+            for (a = (char*)p; n > 0; n--, a += stride) {
+                npy_uint32 * a_ = (npy_uint32 *)a;
+                *a_ = npy_bswap4(*a_);
+            }
+        }
+        else {
+            for (a = (char*)p; n > 0; n--, a += stride) {
+                npy_bswap4_unaligned(a);
+            }
         }
         break;
     case 8:
-        for (a = (char*)p; n > 0; n--) {
-#ifdef HAVE___BUILTIN_BSWAP64
-            npy_uint64 * a_ = (npy_uint64 *)a;
-            *a_ = __builtin_bswap64(*a_);
-            a += stride;
-#else
-            /* mask version would be faster but requires C99 */
-            b = a + 7;
-            c = *a; *a++ = *b; *b-- = c;
-            c = *a; *a++ = *b; *b-- = c;
-            c = *a; *a++ = *b; *b-- = c;
-            c = *a; *a = *b; *b   = c;
-            a += stride - 3;
-#endif
+        if (npy_is_aligned(p, sizeof(npy_uint64))) {
+            for (a = (char*)p; n > 0; n--, a += stride) {
+                npy_uint64 * a_ = (npy_uint64 *)a;
+                *a_ = npy_bswap8(*a_);
+            }
+        }
+        else {
+            for (a = (char*)p; n > 0; n--, a += stride) {
+                npy_bswap8_unaligned(a);
+            }
         }
         break;
     case 2:
-        for (a = (char*)p; n > 0; n--, a += stride) {
-            npy_uint16 * a_ = (npy_uint16 *)a;
-            *a_ = (((*a_ >> 8) & 0xffu) | ((*a_ & 0xffu) << 8));
+        if (npy_is_aligned(p, sizeof(npy_uint16))) {
+            for (a = (char*)p; n > 0; n--, a += stride) {
+                npy_uint16 * a_ = (npy_uint16 *)a;
+                *a_ = npy_bswap2(*a_);
+            }
+        }
+        else {
+            for (a = (char*)p; n > 0; n--, a += stride) {
+                npy_bswap2_unaligned(a);
+            }
         }
         break;
     default: