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 #include <stdlib.h>
  10
  11 /* mingw runtime don't provide random().  */
  12 #ifdef __MINGW32__
  13 #define random rand
  14 #endif
  15
  16 #define NUM 20
  17
  18 #undef MASK
  19 #define MASK 0xe
  20
  21 static void
  22 init_blendps (float *src1, float *src2)
  23 {
  24   int i, sign = 1;
  25
  26   for (i = 0; i < NUM * 4; i++)
  27     {
  28       src1[i] = i * i * sign;
  29       src2[i] = (i + 20) * sign;
  30       sign = -sign;
  31     }
  32 }
  33
  34 static int
  35 check_blendps (__m128 *dst, float *src1, float *src2)
  36 {
  37   float tmp[4];
  38   int j;
  39
  40   memcpy (&tmp[0], src1, sizeof (tmp));
  41   for (j = 0; j < 4; j++)
  42     if ((MASK & (1 << j)))
  43       tmp[j] = src2[j];
  44
  45   return memcmp (dst, &tmp[0], sizeof (tmp));
  46 }
  47
  48 static void
  49 sse4_1_test (void)
  50 {
  51   __m128 x, y;
  52   union
  53     {
  54       __m128 x[NUM];
  55       float f[NUM * 4];
  56     } dst, src1, src2;
  57   union
  58     {
  59       __m128 x;
  60       float f[4];
  61     } src3;
  62   int i;
  63
  64   init_blendps (src1.f, src2.f);
  65
  66   for (i = 0; i < 4; i++)
  67     src3.f[i] = (int) random ();
  68
  69   /* Check blendps imm8, m128, xmm */
  70   for (i = 0; i < NUM; i++)
  71     {
  72       dst.x[i] = _mm_blend_ps (src1.x[i], src2.x[i], MASK);
  73       if (check_blendps (&dst.x[i], &src1.f[i * 4], &src2.f[i * 4]))
  74         abort ();
  75     }
  76
  77    /* Check blendps imm8, xmm, xmm */
  78   x = _mm_blend_ps (dst.x[2], src3.x, MASK);
  79   y = _mm_blend_ps (src3.x, dst.x[2], MASK);
  80
  81   if (check_blendps (&x, &dst.f[8], &src3.f[0]))
  82     abort ();
  83
  84   if (check_blendps (&y, &src3.f[0], &dst.f[8]))
  85     abort ();
  86 }