| 1 |
/* { dg-do run } */ |
| 2 |
/* { dg-options "-O3 -save-temps" } */ |
| 3 |
|
| 4 |
#pragma GCC target "+nosve" |
| 5 |
|
| 6 |
#include <stdint.h> |
| 7 |
#include <string.h> |
| 8 |
|
| 9 |
#define DSIZE 16 |
| 10 |
#define PIXSIZE 64 |
| 11 |
|
| 12 |
extern void |
| 13 |
wplus (uint16_t *d, uint8_t *restrict pix1, uint8_t *restrict pix2 ) |
| 14 |
{ |
| 15 |
for (int y = 0; y < 4; y++ ) |
| 16 |
{ |
| 17 |
for (int x = 0; x < 4; x++ ) |
| 18 |
d[x + y*4] = pix1[x] + pix2[x]; |
| 19 |
pix1 += 16; |
| 20 |
pix2 += 16; |
| 21 |
} |
| 22 |
} |
| 23 |
extern void __attribute__((optimize (0))) |
| 24 |
wplus_no_opt (uint16_t *d, uint8_t *restrict pix1, uint8_t *restrict pix2 ) |
| 25 |
{ |
| 26 |
for (int y = 0; y < 4; y++ ) |
| 27 |
{ |
| 28 |
for (int x = 0; x < 4; x++ ) |
| 29 |
d[x + y*4] = pix1[x] + pix2[x]; |
| 30 |
pix1 += 16; |
| 31 |
pix2 += 16; |
| 32 |
} |
| 33 |
} |
| 34 |
|
| 35 |
extern void |
| 36 |
wminus (uint16_t *d, uint8_t *restrict pix1, uint8_t *restrict pix2 ) |
| 37 |
{ |
| 38 |
for (int y = 0; y < 4; y++ ) |
| 39 |
{ |
| 40 |
for (int x = 0; x < 4; x++ ) |
| 41 |
d[x + y*4] = pix1[x] - pix2[x]; |
| 42 |
pix1 += 16; |
| 43 |
pix2 += 16; |
| 44 |
} |
| 45 |
} |
| 46 |
extern void __attribute__((optimize (0))) |
| 47 |
wminus_no_opt (uint16_t *d, uint8_t *restrict pix1, uint8_t *restrict pix2 ) |
| 48 |
{ |
| 49 |
for (int y = 0; y < 4; y++ ) |
| 50 |
{ |
| 51 |
for (int x = 0; x < 4; x++ ) |
| 52 |
d[x + y*4] = pix1[x] - pix2[x]; |
| 53 |
pix1 += 16; |
| 54 |
pix2 += 16; |
| 55 |
} |
| 56 |
} |
| 57 |
|
| 58 |
extern void |
| 59 |
wmult (uint16_t *d, uint8_t *restrict pix1, uint8_t *restrict pix2 ) |
| 60 |
{ |
| 61 |
for (int y = 0; y < 4; y++ ) |
| 62 |
{ |
| 63 |
for (int x = 0; x < 4; x++ ) |
| 64 |
d[x + y*4] = pix1[x] * pix2[x]; |
| 65 |
pix1 += 16; |
| 66 |
pix2 += 16; |
| 67 |
} |
| 68 |
} |
| 69 |
extern void __attribute__((optimize (0))) |
| 70 |
wmult_no_opt (uint16_t *d, uint8_t *restrict pix1, uint8_t *restrict pix2 ) |
| 71 |
{ |
| 72 |
for (int y = 0; y < 4; y++ ) |
| 73 |
{ |
| 74 |
for (int x = 0; x < 4; x++ ) |
| 75 |
d[x + y*4] = pix1[x] * pix2[x]; |
| 76 |
pix1 += 16; |
| 77 |
pix2 += 16; |
| 78 |
} |
| 79 |
} |
| 80 |
|
| 81 |
extern void |
| 82 |
wlshift (uint16_t *d, uint8_t *restrict pix1) |
| 83 |
|
| 84 |
{ |
| 85 |
for (int y = 0; y < 4; y++ ) |
| 86 |
{ |
| 87 |
for (int x = 0; x < 4; x++ ) |
| 88 |
d[x + y*4] = pix1[x] << 8; |
| 89 |
pix1 += 16; |
| 90 |
} |
| 91 |
} |
| 92 |
extern void __attribute__((optimize (0))) |
| 93 |
wlshift_no_opt (uint16_t *d, uint8_t *restrict pix1) |
| 94 |
|
| 95 |
{ |
| 96 |
for (int y = 0; y < 4; y++ ) |
| 97 |
{ |
| 98 |
for (int x = 0; x < 4; x++ ) |
| 99 |
d[x + y*4] = pix1[x] << 8; |
| 100 |
pix1 += 16; |
| 101 |
} |
| 102 |
} |
| 103 |
|
| 104 |
void __attribute__((optimize (0))) |
| 105 |
init_arrays (uint16_t *d_a, uint16_t *d_b, uint8_t *pix1, uint8_t *pix2) |
| 106 |
{ |
| 107 |
for (int i = 0; i < DSIZE; i++) |
| 108 |
{ |
| 109 |
d_a[i] = (1074 * i)%17; |
| 110 |
d_b[i] = (1074 * i)%17; |
| 111 |
} |
| 112 |
for (int i = 0; i < PIXSIZE; i++) |
| 113 |
{ |
| 114 |
pix1[i] = (1024 * i)%17; |
| 115 |
pix2[i] = (1024 * i)%17; |
| 116 |
} |
| 117 |
} |
| 118 |
|
| 119 |
/* Don't optimize main so we don't get confused over where the vector |
| 120 |
instructions are generated. */ |
| 121 |
__attribute__((optimize (0))) |
| 122 |
int main () |
| 123 |
{ |
| 124 |
uint16_t d_a[DSIZE]; |
| 125 |
uint16_t d_b[DSIZE]; |
| 126 |
uint8_t pix1[PIXSIZE]; |
| 127 |
uint8_t pix2[PIXSIZE]; |
| 128 |
|
| 129 |
init_arrays (d_a, d_b, pix1, pix2); |
| 130 |
wplus (d_a, pix1, pix2); |
| 131 |
wplus_no_opt (d_b, pix1, pix2); |
| 132 |
if (memcmp (d_a,d_b, DSIZE) != 0) |
| 133 |
return 1; |
| 134 |
|
| 135 |
init_arrays (d_a, d_b, pix1, pix2); |
| 136 |
wminus (d_a, pix1, pix2); |
| 137 |
wminus_no_opt (d_b, pix1, pix2); |
| 138 |
if (memcmp (d_a,d_b, DSIZE) != 0) |
| 139 |
return 2; |
| 140 |
|
| 141 |
init_arrays (d_a, d_b, pix1, pix2); |
| 142 |
wmult (d_a, pix1, pix2); |
| 143 |
wmult_no_opt (d_b, pix1, pix2); |
| 144 |
if (memcmp (d_a,d_b, DSIZE) != 0) |
| 145 |
return 3; |
| 146 |
|
| 147 |
init_arrays (d_a, d_b, pix1, pix2); |
| 148 |
wlshift (d_a, pix1); |
| 149 |
wlshift_no_opt (d_b, pix1); |
| 150 |
if (memcmp (d_a,d_b, DSIZE) != 0) |
| 151 |
return 4; |
| 152 |
|
| 153 |
} |
| 154 |
|
| 155 |
/* { dg-final { scan-assembler-times "uaddl\\tv" 2 } } */ |
| 156 |
/* { dg-final { scan-assembler-times "usubl\\tv" 2 } } */ |
| 157 |
/* { dg-final { scan-assembler-times "umull\\tv" 2 } } */ |
| 158 |
/* { dg-final { scan-assembler-times "shll\\tv" 2 } } */ |