gcc/testsuite/gcc.target/i386/ssse3-pmulhrsw.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_pmulhrsw (int *i1, int *i2, int *r)
  23 {
  24   __m64 t1 = *(__m64 *) i1;
  25   __m64 t2 = *(__m64 *) i2;
  26   *(__m64 *) r = _mm_mulhrs_pi16 (t1, t2);
  27   _mm_empty ();
  28 }
  29 #endif
  30
  31 /* Test the 128-bit form */
  32 static void
  33 ssse3_test_pmulhrsw128 (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_mulhrs_epi16 (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   short *s1 = (short *) i1;
  46   short *s2 = (short *) i2;
  47   short *sout = (short *) r;
  48   int t0;
  49   int i;
  50
  51   for (i = 0; i < 8; i++)
  52     {
  53       t0 = (((int) s1[i] * (int) s2[i]) >> 14) + 1;
  54       sout[i] = (short) (t0 >> 1);
  55     }
  56 }
  57
  58 static void
  59 TEST (void)
  60 {
  61   int i;
  62   int r [4] __attribute__ ((aligned(16)));
  63   int ck [4];
  64   int fail = 0;
  65
  66   for (i = 0; i < 256; i += 8)
  67     {
  68       /* Manually compute the result */
  69       compute_correct_result (&vals[i + 0], &vals[i + 4], ck);
  70
  71 #ifndef __AVX__
  72       /* Run the 64-bit tests */
  73       ssse3_test_pmulhrsw (&vals[i + 0], &vals[i + 4], &r[0]);
  74       ssse3_test_pmulhrsw (&vals[i + 2], &vals[i + 6], &r[2]);
  75       fail += chk_128 (ck, r);
  76 #endif
  77
  78       /* Run the 128-bit tests */
  79       ssse3_test_pmulhrsw128 (&vals[i + 0], &vals[i + 4], r);
  80       fail += chk_128 (ck, r);
  81     }
  82
  83   if (fail != 0)
  84     abort ();
  85 }