gcc/testsuite/gcc.target/i386/avx512vbmi-vpermt2b-2.c

   1 /* { dg-do run } */
   2 /* { dg-options "-O2 -mavx512vbmi" } */
   3 /* { dg-require-effective-target avx512vbmi } */
   4
   5 #define AVX512VBMI
   6
   7 #include "avx512f-helper.h"
   8
   9 #define SIZE (AVX512F_LEN / 8)
  10 #include <math.h>
  11 #include <limits.h>
  12 #include <float.h>
  13 #include "avx512f-mask-type.h"
  14
  15 #define NUM 32
  16
  17 void
  18 CALC (char *dst, char *src1, char *ind, char *src2)
  19 {
  20   int i;
  21
  22   for (i = 0; i < SIZE; i++)
  23     {
  24       unsigned long long offset = ind[i] & (SIZE - 1);
  25       unsigned long long cond = ind[i] & SIZE;
  26
  27       dst[i] = cond ? src2[offset] : src1[offset];
  28     }
  29 }
  30
  31 void
  32 TEST (void)
  33 {
  34   int i, j;
  35   UNION_TYPE (AVX512F_LEN, i_b) s1, s2, res1, res2, res3, ind;
  36   char res_ref[SIZE];
  37
  38   MASK_TYPE mask = MASK_VALUE;
  39
  40   for (i = 0; i < NUM; i++)
  41     {
  42       for (j = 0; j < SIZE; j++)
  43         {
  44           ind.a[j] = i * (j << 1);
  45           s1.a[j] = DEFAULT_VALUE;
  46           s2.a[j] = 1.5 * i * 2 * j;
  47
  48           res1.a[j] = DEFAULT_VALUE;
  49           res2.a[j] = DEFAULT_VALUE;
  50           res3.a[j] = DEFAULT_VALUE;
  51         }
  52
  53       CALC (res_ref, s1.a, ind.a, s2.a);
  54
  55       res1.x = INTRINSIC (_permutex2var_epi8) (s1.x, ind.x, s2.x);
  56       res2.x =
  57         INTRINSIC (_mask_permutex2var_epi8) (s1.x, mask, ind.x, s2.x);
  58       res3.x =
  59         INTRINSIC (_maskz_permutex2var_epi8) (mask, s1.x, ind.x,
  60                                                s2.x);
  61
  62       if (UNION_CHECK (AVX512F_LEN, i_b) (res1, res_ref))
  63         abort ();
  64
  65       MASK_MERGE (i_b) (res_ref, mask, SIZE);
  66       if (UNION_CHECK (AVX512F_LEN, i_b) (res2, res_ref))
  67         abort ();
  68
  69       MASK_ZERO (i_b) (res_ref, mask, SIZE);
  70       if (UNION_CHECK (AVX512F_LEN, i_b) (res3, res_ref))
  71         abort ();
  72     }
  73 }