bpo-27987: pymalloc: align by 16bytes on 64bit platform (GH-12850) (GH-13319)

(cherry picked from commit f0be4bb)

Author: vstinner

Misc/NEWS.d/next/Core and Builtins/2019-04-16-11-52-21.bpo-27987.n2_DcQ.rst

@@ -0,0 +1,3 @@
+pymalloc returns memory blocks aligned by 16 bytes, instead of 8 bytes, on
+64-bit platforms to conform x86-64 ABI. Recent compilers assume this alignment
+more often. Patch by Inada Naoki.

Objects/obmalloc.c

@@ -154,8 +154,14 @@ static int running_on_valgrind = -1;
  *
  * You shouldn't change this unless you know what you are doing.
  */
+
+#if SIZEOF_VOID_P > 4
+#define ALIGNMENT              16               /* must be 2^N */
+#define ALIGNMENT_SHIFT         4
+#else
 #define ALIGNMENT               8               /* must be 2^N */
 #define ALIGNMENT_SHIFT         3
+#endif
 #define ALIGNMENT_MASK          (ALIGNMENT - 1)
 
 /* Return the number of bytes in size class I, as a uint. */