gcc/testsuite/gcc.target/i386/sse4_1-pblendw.c

   1 /* { dg-do run } */
   2 /* { dg-require-effective-target sse4 } */
   3 /* { dg-options "-O2 -msse4.1" } */
   4
   5 #include "sse4_1-check.h"
   6
   7 #include <smmintrin.h>
   8 #include <string.h>
   9
  10 #define NUM 20
  11
  12 #ifndef MASK
  13 #define MASK 0x0f
  14 #endif
  15
  16 static void
  17 init_pblendw (short *src1, short *src2)
  18 {
  19   int i, sign = 1;
  20
  21   for (i = 0; i < NUM * 8; i++)
  22     {
  23       src1[i] = i * i * sign;
  24       src2[i] = (i + 20) * sign;
  25       sign = -sign;
  26     }
  27 }
  28
  29 static int
  30 check_pblendw (__m128i *dst, short *src1, short *src2)
  31 {
  32   short tmp[8];
  33   int j;
  34
  35   memcpy (&tmp[0], src1, sizeof (tmp));
  36   for (j = 0; j < 8; j++)
  37     if ((MASK & (1 << j)))
  38       tmp[j] = src2[j];
  39
  40   return memcmp (dst, &tmp[0], sizeof (tmp));
  41 }
  42
  43 static void
  44 sse4_1_test (void)
  45 {
  46   __m128i x, y;
  47   union
  48     {
  49       __m128i x[NUM];
  50       short s[NUM * 8];
  51     } dst, src1, src2;
  52   union
  53     {
  54       __m128i x;
  55       short s[8];
  56     } src3;
  57   int i;
  58
  59   init_pblendw (src1.s, src2.s);
  60
  61   /* Check pblendw imm8, m128, xmm */
  62   for (i = 0; i < NUM; i++)
  63     {
  64       dst.x[i] = _mm_blend_epi16 (src1.x[i], src2.x[i], MASK);
  65       if (check_pblendw (&dst.x[i], &src1.s[i * 8], &src2.s[i * 8]))
  66         abort ();
  67     }
  68
  69    /* Check pblendw imm8, xmm, xmm */
  70   src3.x = _mm_setzero_si128 ();
  71
  72   x = _mm_blend_epi16 (dst.x[2], src3.x, MASK);
  73   y = _mm_blend_epi16 (src3.x, dst.x[2], MASK);
  74
  75   if (check_pblendw (&x, &dst.s[16], &src3.s[0]))
  76     abort ();
  77
  78   if (check_pblendw (&y, &src3.s[0], &dst.s[16]))
  79     abort ();
  80 }