| 1 |
/* { dg-do run } */ |
| 2 |
/* { dg-require-effective-target sse4 } */ |
| 3 |
/* { dg-options "-O2 -msse4.1" } */ |
| 4 |
|
| 5 |
#ifndef CHECK_H |
| 6 |
#define CHECK_H "sse4_1-check.h" |
| 7 |
#endif |
| 8 |
|
| 9 |
#ifndef TEST |
| 10 |
#define TEST sse4_1_test |
| 11 |
#endif |
| 12 |
|
| 13 |
#include CHECK_H |
| 14 |
|
| 15 |
#include <smmintrin.h> |
| 16 |
#include <string.h> |
| 17 |
|
| 18 |
#define msk0 0xC0 |
| 19 |
#define msk1 0x01 |
| 20 |
#define msk2 0xF2 |
| 21 |
#define msk3 0x03 |
| 22 |
#define msk4 0x84 |
| 23 |
#define msk5 0x05 |
| 24 |
#define msk6 0xE6 |
| 25 |
#define msk7 0x67 |
| 26 |
|
| 27 |
static __m128i |
| 28 |
compute_mpsadbw (unsigned char *v1, unsigned char *v2, int mask) |
| 29 |
{ |
| 30 |
union |
| 31 |
{ |
| 32 |
__m128i x; |
| 33 |
unsigned short s[8]; |
| 34 |
} ret; |
| 35 |
unsigned char s[4]; |
| 36 |
int i, j; |
| 37 |
int offs1, offs2; |
| 38 |
|
| 39 |
offs2 = 4 * (mask & 3); |
| 40 |
for (i = 0; i < 4; i++) |
| 41 |
s[i] = v2[offs2 + i]; |
| 42 |
|
| 43 |
offs1 = 4 * ((mask & 4) >> 2); |
| 44 |
for (j = 0; j < 8; j++) |
| 45 |
{ |
| 46 |
ret.s[j] = 0; |
| 47 |
for (i = 0; i < 4; i++) |
| 48 |
ret.s[j] += abs (v1[offs1 + j + i] - s[i]); |
| 49 |
} |
| 50 |
|
| 51 |
return ret.x; |
| 52 |
} |
| 53 |
|
| 54 |
static void |
| 55 |
TEST (void) |
| 56 |
{ |
| 57 |
union |
| 58 |
{ |
| 59 |
__m128i x; |
| 60 |
unsigned int i[4]; |
| 61 |
unsigned char c[16]; |
| 62 |
} val1, val2, val3 [8]; |
| 63 |
__m128i res[8], tmp; |
| 64 |
unsigned char masks[8]; |
| 65 |
int i; |
| 66 |
|
| 67 |
val1.i[0] = 0x35251505; |
| 68 |
val1.i[1] = 0x75655545; |
| 69 |
val1.i[2] = 0xB5A59585; |
| 70 |
val1.i[3] = 0xF5E5D5C5; |
| 71 |
|
| 72 |
val2.i[0] = 0x31211101; |
| 73 |
val2.i[1] = 0x71615141; |
| 74 |
val2.i[2] = 0xB1A19181; |
| 75 |
val2.i[3] = 0xF1E1D1C1; |
| 76 |
|
| 77 |
for (i=0; i < 8; i++) |
| 78 |
switch (i % 3) |
| 79 |
{ |
| 80 |
case 1: |
| 81 |
val3[i].i[0] = 0xF1E1D1C1; |
| 82 |
val3[i].i[1] = 0xB1A19181; |
| 83 |
val3[i].i[2] = 0x71615141; |
| 84 |
val3[i].i[3] = 0x31211101; |
| 85 |
break; |
| 86 |
default: |
| 87 |
val3[i].x = val2.x; |
| 88 |
break; |
| 89 |
} |
| 90 |
|
| 91 |
/* Check mpsadbw imm8, xmm, xmm. */ |
| 92 |
res[0] = _mm_mpsadbw_epu8 (val1.x, val2.x, msk0); |
| 93 |
res[1] = _mm_mpsadbw_epu8 (val1.x, val2.x, msk1); |
| 94 |
res[2] = _mm_mpsadbw_epu8 (val1.x, val2.x, msk2); |
| 95 |
res[3] = _mm_mpsadbw_epu8 (val1.x, val2.x, msk3); |
| 96 |
res[4] = _mm_mpsadbw_epu8 (val1.x, val2.x, msk4); |
| 97 |
res[5] = _mm_mpsadbw_epu8 (val1.x, val2.x, msk5); |
| 98 |
res[6] = _mm_mpsadbw_epu8 (val1.x, val2.x, msk6); |
| 99 |
res[7] = _mm_mpsadbw_epu8 (val1.x, val2.x, msk7); |
| 100 |
|
| 101 |
masks[0] = msk0; |
| 102 |
masks[1] = msk1; |
| 103 |
masks[2] = msk2; |
| 104 |
masks[3] = msk3; |
| 105 |
masks[4] = msk4; |
| 106 |
masks[5] = msk5; |
| 107 |
masks[6] = msk6; |
| 108 |
masks[7] = msk7; |
| 109 |
|
| 110 |
for (i=0; i < 8; i++) |
| 111 |
{ |
| 112 |
tmp = compute_mpsadbw (val1.c, val2.c, masks[i]); |
| 113 |
if (memcmp (&tmp, &res[i], sizeof (tmp))) |
| 114 |
abort (); |
| 115 |
} |
| 116 |
|
| 117 |
/* Check mpsadbw imm8, m128, xmm. */ |
| 118 |
for (i=0; i < 8; i++) |
| 119 |
{ |
| 120 |
res[i] = _mm_mpsadbw_epu8 (val1.x, val3[i].x, msk4); |
| 121 |
masks[i] = msk4; |
| 122 |
} |
| 123 |
|
| 124 |
for (i=0; i < 8; i++) |
| 125 |
{ |
| 126 |
tmp = compute_mpsadbw (val1.c, val3[i].c, masks[i]); |
| 127 |
if (memcmp (&tmp, &res[i], sizeof (tmp))) |
| 128 |
abort (); |
| 129 |
} |
| 130 |
} |