x86 validator: Implement rewriting "movntq" to "movq" on x86-64

tests/validator/rewrite_nontemporals.c

 /*
  * Copyright 2016 The Native Client Authors. All rights reserved.
  * Use of this source code is governed by a BSD-style license that can be
  * found in the LICENSE file.
  */
 
 #include <string.h>
 
 #include "native_client/src/include/nacl_assert.h"
 
@@ skipping 22 matching lines @@
   asm("movdqa g_src " MEM_SUFFIX ", %%xmm0\n"
       "movntdq %%xmm0, g_dest " MEM_SUFFIX "\n" : : : "xmm0");
   ASSERT_EQ(memcmp(g_dest, g_src, 16), 0);
 
   /* Test prefetchnta.  This has no side effects that we can test for. */
   asm("prefetchnta g_dest " MEM_SUFFIX "\n" :);
 
   /* This compiles to prefetchnta on x86. */
   __builtin_prefetch(&g_dest, /* rw= */ 0, /* locality= */ 0);
 
-#if defined(__i386__)
   /* Test movntq. */
   reset_test_vars();
-  asm("movq g_src, %%mm0\n"
-      "movntq %%mm0, g_dest\n" : : : "mm0");
+  asm("movq g_src" MEM_SUFFIX ", %%mm0\n"
+      "movntq %%mm0, g_dest" MEM_SUFFIX "\n" : : : "mm0");
   ASSERT_EQ(memcmp(g_dest, g_src, 8), 0);
-#endif
 
 #if defined(__x86_64__)
   /* Test movntps. */
   reset_test_vars();
   asm("movdqa g_src(%%r15), %%xmm0\n"
       "movntps %%xmm0, g_dest(%%r15)\n" : : : "xmm0");
   ASSERT_EQ(memcmp(g_dest, g_src, 16), 0);
 
   /* Test movnti, using 32-bit operand. */
   reset_test_vars();
@@ skipping 23 matching lines @@
   asm("mov g_src(%%r15), %%rax\n"
       "movnti %%rax, g_dest(%%rip)\n" : : : "rax");
   ASSERT_EQ(memcmp(g_dest, g_src, 8), 0);
 
   /* Test prefetchnta using RIP-relative addressing. */
   asm("prefetchnta g_dest(%rip)\n");
 #endif
 
   return 0;
 }