gcc/testsuite/gcc.target/i386/ssse3-phsubd.c

   1 /* { dg-do run } */
   2 /* { dg-require-effective-target ssse3 } */
   3 /* { dg-options "-O2 -fno-strict-aliasing -mssse3" } */
   4
   5 #ifndef CHECK_H
   6 #define CHECK_H "ssse3-check.h"
   7 #endif
   8
   9 #ifndef TEST
  10 #define TEST ssse3_test
  11 #endif
  12
  13 #include CHECK_H
  14
  15 #include "ssse3-vals.h"
  16
  17 #include <tmmintrin.h>
  18
  19 #ifndef __AVX__
  20 /* Test the 64-bit form */
  21 static void
  22 ssse3_test_phsubd (int *i1, int *i2, int *r)
  23 {
  24   __m64 t1 = *(__m64 *) i1;
  25   __m64 t2 = *(__m64 *) i2;
  26   *(__m64 *) r = _mm_hsub_pi32(t1, t2);
  27   _mm_empty ();
  28 }
  29 #endif
  30
  31 /* Test the 128-bit form */
  32 static void
  33 ssse3_test_phsubd128 (int *i1, int *i2, int *r)
  34 {
  35   /* Assumes incoming pointers are 16-byte aligned */
  36   __m128i t1 = *(__m128i *) i1;
  37   __m128i t2 = *(__m128i *) i2;
  38   *(__m128i *) r = _mm_hsub_epi32 (t1, t2);
  39 }
  40
  41 /* Routine to manually compute the results */
  42 static void
  43 compute_correct_result (int *i1, int *i2, int *r)
  44 {
  45   int i;
  46
  47   for (i = 0; i < 2; i++)
  48     r[i] = i1[2 * i] - i1[2 * i + 1];
  49   for (i = 0; i < 2; i++)
  50     r[i + 2] = i2[2 * i] - i2[2 * i + 1];
  51 }
  52
  53 static void
  54 TEST (void)
  55 {
  56   int i;
  57   int r [4] __attribute__ ((aligned(16)));
  58   int ck [4];
  59   int fail = 0;
  60
  61   for (i = 0; i < 256; i += 8)
  62     {
  63       /* Manually compute the result */
  64       compute_correct_result (&vals[i + 0], &vals[i + 4], ck);
  65
  66 #ifndef __AVX__
  67       /* Run the 64-bit tests */
  68       ssse3_test_phsubd (&vals[i + 0], &vals[i + 2], &r[0]);
  69       ssse3_test_phsubd (&vals[i + 4], &vals[i + 6], &r[2]);
  70       fail += chk_128 (ck, r);
  71 #endif
  72
  73       /* Run the 128-bit tests */
  74       ssse3_test_phsubd128 (&vals[i + 0], &vals[i + 4], r);
  75       fail += chk_128 (ck, r);
  76     }
  77
  78   if (fail != 0)
  79     abort ();
  80 }