gcc/testsuite/gcc.target/i386/sse4_1-blendps-2.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 #undef MASK
  13 #define MASK 0xe
  14
  15 static void
  16 init_blendps (float *src1, float *src2)
  17 {
  18   int i, sign = 1;
  19
  20   for (i = 0; i < NUM * 4; i++)
  21     {
  22       src1[i] = i * i * sign;
  23       src2[i] = (i + 20) * sign;
  24       sign = -sign;
  25     }
  26 }
  27
  28 static int
  29 check_blendps (__m128 *dst, float *src1, float *src2)
  30 {
  31   float tmp[4];
  32   int j;
  33
  34   memcpy (&tmp[0], src1, sizeof (tmp));
  35   for (j = 0; j < 4; j++)
  36     if ((MASK & (1 << j)))
  37       tmp[j] = src2[j];
  38
  39   return memcmp (dst, &tmp[0], sizeof (tmp));
  40 }
  41
  42 static void
  43 sse4_1_test (void)
  44 {
  45   __m128 x, y;
  46   union
  47     {
  48       __m128 x[NUM];
  49       float f[NUM * 4];
  50     } dst, src1, src2;
  51   union
  52     {
  53       __m128 x;
  54       float f[4];
  55     } src3;
  56   int i;
  57
  58   init_blendps (src1.f, src2.f);
  59
  60   /* Check blendps imm8, m128, xmm */
  61   for (i = 0; i < NUM; i++)
  62     {
  63       dst.x[i] = _mm_blend_ps (src1.x[i], src2.x[i], MASK);
  64       if (check_blendps (&dst.x[i], &src1.f[i * 4], &src2.f[i * 4]))
  65         abort ();
  66     }
  67
  68    /* Check blendps imm8, xmm, xmm */
  69   x = _mm_blend_ps (dst.x[2], src3.x, MASK);
  70   y = _mm_blend_ps (src3.x, dst.x[2], MASK);
  71
  72   if (check_blendps (&x, &dst.f[8], &src3.f[0]))
  73     abort ();
  74
  75   if (check_blendps (&y, &src3.f[0], &dst.f[8]))
  76     abort ();
  77 }