| 1 |
/* { dg-do run } */ |
| 2 |
/* { dg-require-effective-target avx2 } */ |
| 3 |
/* { dg-options "-O2 -mavx2" } */ |
| 4 |
|
| 5 |
#include <string.h> |
| 6 |
#include "ssse3-vals.h" |
| 7 |
#include "avx2-check.h" |
| 8 |
|
| 9 |
/* Test the 256-bit form */ |
| 10 |
static void |
| 11 |
avx2_test_palignr256 (__m256i t1, __m256i t2, unsigned int imm, __m256i * r) |
| 12 |
{ |
| 13 |
switch (imm) |
| 14 |
{ |
| 15 |
case 0: |
| 16 |
*r = _mm256_alignr_epi8 (t1, t2, 0); |
| 17 |
break; |
| 18 |
case 1: |
| 19 |
*r = _mm256_alignr_epi8 (t1, t2, 1); |
| 20 |
break; |
| 21 |
case 2: |
| 22 |
*r = _mm256_alignr_epi8 (t1, t2, 2); |
| 23 |
break; |
| 24 |
case 3: |
| 25 |
*r = _mm256_alignr_epi8 (t1, t2, 3); |
| 26 |
break; |
| 27 |
case 4: |
| 28 |
*r = _mm256_alignr_epi8 (t1, t2, 4); |
| 29 |
break; |
| 30 |
case 5: |
| 31 |
*r = _mm256_alignr_epi8 (t1, t2, 5); |
| 32 |
break; |
| 33 |
case 6: |
| 34 |
*r = _mm256_alignr_epi8 (t1, t2, 6); |
| 35 |
break; |
| 36 |
case 7: |
| 37 |
*r = _mm256_alignr_epi8 (t1, t2, 7); |
| 38 |
break; |
| 39 |
case 8: |
| 40 |
*r = _mm256_alignr_epi8 (t1, t2, 8); |
| 41 |
break; |
| 42 |
case 9: |
| 43 |
*r = _mm256_alignr_epi8 (t1, t2, 9); |
| 44 |
break; |
| 45 |
case 10: |
| 46 |
*r = _mm256_alignr_epi8 (t1, t2, 10); |
| 47 |
break; |
| 48 |
case 11: |
| 49 |
*r = _mm256_alignr_epi8 (t1, t2, 11); |
| 50 |
break; |
| 51 |
case 12: |
| 52 |
*r = _mm256_alignr_epi8 (t1, t2, 12); |
| 53 |
break; |
| 54 |
case 13: |
| 55 |
*r = _mm256_alignr_epi8 (t1, t2, 13); |
| 56 |
break; |
| 57 |
case 14: |
| 58 |
*r = _mm256_alignr_epi8 (t1, t2, 14); |
| 59 |
break; |
| 60 |
case 15: |
| 61 |
*r = _mm256_alignr_epi8 (t1, t2, 15); |
| 62 |
break; |
| 63 |
case 16: |
| 64 |
*r = _mm256_alignr_epi8 (t1, t2, 16); |
| 65 |
break; |
| 66 |
case 17: |
| 67 |
*r = _mm256_alignr_epi8 (t1, t2, 17); |
| 68 |
break; |
| 69 |
case 18: |
| 70 |
*r = _mm256_alignr_epi8 (t1, t2, 18); |
| 71 |
break; |
| 72 |
case 19: |
| 73 |
*r = _mm256_alignr_epi8 (t1, t2, 19); |
| 74 |
break; |
| 75 |
case 20: |
| 76 |
*r = _mm256_alignr_epi8 (t1, t2, 20); |
| 77 |
break; |
| 78 |
case 21: |
| 79 |
*r = _mm256_alignr_epi8 (t1, t2, 21); |
| 80 |
break; |
| 81 |
case 22: |
| 82 |
*r = _mm256_alignr_epi8 (t1, t2, 22); |
| 83 |
break; |
| 84 |
case 23: |
| 85 |
*r = _mm256_alignr_epi8 (t1, t2, 23); |
| 86 |
break; |
| 87 |
case 24: |
| 88 |
*r = _mm256_alignr_epi8 (t1, t2, 24); |
| 89 |
break; |
| 90 |
case 25: |
| 91 |
*r = _mm256_alignr_epi8 (t1, t2, 25); |
| 92 |
break; |
| 93 |
case 26: |
| 94 |
*r = _mm256_alignr_epi8 (t1, t2, 26); |
| 95 |
break; |
| 96 |
case 27: |
| 97 |
*r = _mm256_alignr_epi8 (t1, t2, 27); |
| 98 |
break; |
| 99 |
case 28: |
| 100 |
*r = _mm256_alignr_epi8 (t1, t2, 28); |
| 101 |
break; |
| 102 |
case 29: |
| 103 |
*r = _mm256_alignr_epi8 (t1, t2, 29); |
| 104 |
break; |
| 105 |
case 30: |
| 106 |
*r = _mm256_alignr_epi8 (t1, t2, 30); |
| 107 |
break; |
| 108 |
case 31: |
| 109 |
*r = _mm256_alignr_epi8 (t1, t2, 31); |
| 110 |
break; |
| 111 |
default: |
| 112 |
*r = _mm256_alignr_epi8 (t1, t2, 32); |
| 113 |
break; |
| 114 |
} |
| 115 |
} |
| 116 |
|
| 117 |
/* Routine to manually compute the results */ |
| 118 |
static void |
| 119 |
compute_correct_result_256 (int *i1, int *i2, unsigned int imm, int *r) |
| 120 |
{ |
| 121 |
char buf[32]; |
| 122 |
char *bout = (char *) r; |
| 123 |
int i; |
| 124 |
|
| 125 |
/* Fill lowers 128 bit of ymm */ |
| 126 |
memcpy (&buf[0], i2, 16); |
| 127 |
memcpy (&buf[16], i1, 16); |
| 128 |
|
| 129 |
for (i = 0; i < 16; i++) |
| 130 |
if (imm >= 32 || imm + i >= 32) |
| 131 |
bout[i] = 0; |
| 132 |
else |
| 133 |
bout[i] = buf[imm + i]; |
| 134 |
|
| 135 |
/* Fill higher 128 bit of ymm */ |
| 136 |
bout += 16; |
| 137 |
memcpy (&buf[0], i2 + 4, 16); |
| 138 |
memcpy (&buf[16], i1 + 4, 16); |
| 139 |
|
| 140 |
for (i = 0; i < 16; i++) |
| 141 |
if (imm >= 32 || imm + i >= 32) |
| 142 |
bout[i] = 0; |
| 143 |
else |
| 144 |
bout[i] = buf[imm + i]; |
| 145 |
} |
| 146 |
|
| 147 |
static void |
| 148 |
avx2_test (void) |
| 149 |
{ |
| 150 |
int i; |
| 151 |
int ck[8]; |
| 152 |
int r[8]; |
| 153 |
unsigned int imm; |
| 154 |
int fail = 0; |
| 155 |
|
| 156 |
union256i_q s1, s2, d; |
| 157 |
|
| 158 |
for (i = 0; i < 256; i += 16) |
| 159 |
for (imm = 0; imm < 100; imm++) |
| 160 |
{ |
| 161 |
/* Recompute the results for 256-bits */ |
| 162 |
compute_correct_result_256 (&vals[i + 0], &vals[i + 8], imm, ck); |
| 163 |
|
| 164 |
s1.x = _mm256_loadu_si256 ((__m256i *) & vals[i + 0]); |
| 165 |
s2.x = _mm256_loadu_si256 ((__m256i *) & vals[i + 8]); |
| 166 |
|
| 167 |
/* Run the 256-bit tests */ |
| 168 |
avx2_test_palignr256 (s1.x, s2.x, imm, &d.x); |
| 169 |
|
| 170 |
_mm256_storeu_si256 ((__m256i *) r, d.x); |
| 171 |
|
| 172 |
fail += checkVi (r, ck, 8); |
| 173 |
} |
| 174 |
|
| 175 |
if (fail != 0) |
| 176 |
abort (); |
| 177 |
} |