| 1 |
/* { dg-do run } */ |
| 2 |
/* { dg-require-effective-target avx2 } */ |
| 3 |
/* { dg-options "-O2 -mavx2" } */ |
| 4 |
|
| 5 |
#include <string.h> |
| 6 |
#include "avx2-check.h" |
| 7 |
|
| 8 |
#define msk0 0xC0 |
| 9 |
#define msk1 0x01 |
| 10 |
#define msk2 0xF2 |
| 11 |
#define msk3 0x03 |
| 12 |
#define msk4 0x84 |
| 13 |
#define msk5 0x05 |
| 14 |
#define msk6 0xE6 |
| 15 |
#define msk7 0x67 |
| 16 |
|
| 17 |
|
| 18 |
static void |
| 19 |
compute_mpsadbw (int *i1, int *i2, int mask, int *r) |
| 20 |
{ |
| 21 |
unsigned char s[4]; |
| 22 |
int i, j; |
| 23 |
int offs1, offs2; |
| 24 |
unsigned char *v1 = (char *) i1; |
| 25 |
unsigned char *v2 = (char *) i2; |
| 26 |
unsigned short *ret = (unsigned short *) r; |
| 27 |
|
| 28 |
memset (ret, 0, 32); |
| 29 |
|
| 30 |
/* Lower part */ |
| 31 |
offs2 = 4 * (mask & 3); |
| 32 |
for (i = 0; i < 4; i++) |
| 33 |
s[i] = v2[offs2 + i]; |
| 34 |
|
| 35 |
offs1 = 4 * ((mask & 4) >> 2); |
| 36 |
for (j = 0; j < 8; j++) |
| 37 |
for (i = 0; i < 4; i++) |
| 38 |
ret[j] += abs (v1[offs1 + j + i] - s[i]); |
| 39 |
|
| 40 |
/* Higher part */ |
| 41 |
offs2 = 4 * ((mask >> 3) & 3) + 16; |
| 42 |
for (i = 0; i < 4; i++) |
| 43 |
s[i] = v2[offs2 + i]; |
| 44 |
|
| 45 |
offs1 = 4 * ((mask & 0x20) >> 5) + 16; |
| 46 |
for (j = 0; j < 8; j++) |
| 47 |
for (i = 0; i < 4; i++) |
| 48 |
ret[j + 8] += abs (v1[offs1 + j + i] - s[i]); |
| 49 |
} |
| 50 |
|
| 51 |
static void |
| 52 |
avx2_test (void) |
| 53 |
{ |
| 54 |
union256i_d val1, val2, val3[8], res[8]; |
| 55 |
int tmp[8]; |
| 56 |
unsigned char masks[8]; |
| 57 |
int i, j; |
| 58 |
|
| 59 |
val1.a[0] = 0x35251505; |
| 60 |
val1.a[1] = 0x75655545; |
| 61 |
val1.a[2] = 0xB5A59585; |
| 62 |
val1.a[3] = 0xF5E5D5C5; |
| 63 |
|
| 64 |
val1.a[4] = 0x35251505; |
| 65 |
val1.a[5] = 0x75655545; |
| 66 |
val1.a[6] = 0xB5A59585; |
| 67 |
val1.a[7] = 0xF5E5D5C5; |
| 68 |
|
| 69 |
val2.a[0] = 0x31211101; |
| 70 |
val2.a[1] = 0x71615141; |
| 71 |
val2.a[2] = 0xB1A19181; |
| 72 |
val2.a[3] = 0xF1E1D1C1; |
| 73 |
|
| 74 |
val2.a[4] = 0x31211101; |
| 75 |
val2.a[5] = 0x71615141; |
| 76 |
val2.a[6] = 0xB1A19181; |
| 77 |
val2.a[7] = 0xF1E1D1C1; |
| 78 |
|
| 79 |
for (i = 0; i < 8; i++) |
| 80 |
switch (i % 3) |
| 81 |
{ |
| 82 |
case 1: |
| 83 |
val3[i].a[0] = 0xF1E1D1C1; |
| 84 |
val3[i].a[1] = 0xB1A19181; |
| 85 |
val3[i].a[2] = 0x71615141; |
| 86 |
val3[i].a[3] = 0x31211101; |
| 87 |
break; |
| 88 |
default: |
| 89 |
val3[i].x = val2.x; |
| 90 |
break; |
| 91 |
} |
| 92 |
|
| 93 |
/* Check mpsadbw imm8, ymm, ymm. */ |
| 94 |
res[0].x = _mm256_mpsadbw_epu8 (val1.x, val2.x, msk0); |
| 95 |
res[1].x = _mm256_mpsadbw_epu8 (val1.x, val2.x, msk1); |
| 96 |
res[2].x = _mm256_mpsadbw_epu8 (val1.x, val2.x, msk2); |
| 97 |
res[3].x = _mm256_mpsadbw_epu8 (val1.x, val2.x, msk3); |
| 98 |
res[4].x = _mm256_mpsadbw_epu8 (val1.x, val2.x, msk4); |
| 99 |
res[5].x = _mm256_mpsadbw_epu8 (val1.x, val2.x, msk5); |
| 100 |
res[6].x = _mm256_mpsadbw_epu8 (val1.x, val2.x, msk6); |
| 101 |
res[7].x = _mm256_mpsadbw_epu8 (val1.x, val2.x, msk7); |
| 102 |
|
| 103 |
masks[0] = msk0; |
| 104 |
masks[1] = msk1; |
| 105 |
masks[2] = msk2; |
| 106 |
masks[3] = msk3; |
| 107 |
masks[4] = msk4; |
| 108 |
masks[5] = msk5; |
| 109 |
masks[6] = msk6; |
| 110 |
masks[7] = msk7; |
| 111 |
|
| 112 |
for (i = 0; i < 8; i++) |
| 113 |
{ |
| 114 |
compute_mpsadbw (val1.a, val2.a, masks[i], tmp); |
| 115 |
if (check_union256i_d (res[i], tmp)) |
| 116 |
abort (); |
| 117 |
} |
| 118 |
|
| 119 |
/* Check mpsadbw imm8, m256, ymm. */ |
| 120 |
for (i = 0; i < 8; i++) |
| 121 |
{ |
| 122 |
res[i].x = _mm256_mpsadbw_epu8 (val1.x, val3[i].x, msk4); |
| 123 |
masks[i] = msk4; |
| 124 |
} |
| 125 |
|
| 126 |
for (i = 0; i < 8; i++) |
| 127 |
{ |
| 128 |
compute_mpsadbw (val1.a, val3[i].a, masks[i], tmp); |
| 129 |
if (check_union256i_d (res[i], tmp)) |
| 130 |
abort (); |
| 131 |
} |
| 132 |
} |