| 1 |
/* PR rtl-optimization/21239 */ |
| 2 |
/* { dg-do run } */ |
| 3 |
/* { dg-options "-O2 -msse2" } */ |
| 4 |
/* { dg-require-effective-target sse2 } */ |
| 5 |
|
| 6 |
#include "sse2-check.h" |
| 7 |
|
| 8 |
#include <emmintrin.h> |
| 9 |
|
| 10 |
void |
| 11 |
foo (unsigned int x, double *y, const double *z) |
| 12 |
{ |
| 13 |
__m128d tmp; |
| 14 |
while (x) |
| 15 |
{ |
| 16 |
tmp = _mm_load_sd (z); |
| 17 |
_mm_store_sd (y, tmp); |
| 18 |
--x; ++z; ++y; |
| 19 |
} |
| 20 |
} |
| 21 |
|
| 22 |
void |
| 23 |
bar (unsigned int x, float *y, const float *z) |
| 24 |
{ |
| 25 |
__m128 tmp; |
| 26 |
unsigned int i; |
| 27 |
for (i = 0; i < x; ++i) |
| 28 |
{ |
| 29 |
tmp = (__m128) { *z, 0, 0, 0 }; |
| 30 |
*y = __builtin_ia32_vec_ext_v4sf (tmp, 0); |
| 31 |
++z; ++y; |
| 32 |
} |
| 33 |
for (i = 0; i < x; ++i) |
| 34 |
{ |
| 35 |
tmp = (__m128) { 0, *z, 0, 0 }; |
| 36 |
*y = __builtin_ia32_vec_ext_v4sf (tmp, 1); |
| 37 |
++z; ++y; |
| 38 |
} |
| 39 |
for (i = 0; i < x; ++i) |
| 40 |
{ |
| 41 |
tmp = (__m128) { 0, 0, *z, 0 }; |
| 42 |
*y = __builtin_ia32_vec_ext_v4sf (tmp, 2); |
| 43 |
++z; ++y; |
| 44 |
} |
| 45 |
for (i = 0; i < x; ++i) |
| 46 |
{ |
| 47 |
tmp = (__m128) { 0, 0, 0, *z }; |
| 48 |
*y = __builtin_ia32_vec_ext_v4sf (tmp, 3); |
| 49 |
++z; ++y; |
| 50 |
} |
| 51 |
} |
| 52 |
|
| 53 |
static void |
| 54 |
sse2_test (void) |
| 55 |
{ |
| 56 |
unsigned int i; |
| 57 |
double a[16], b[16]; |
| 58 |
float c[16], d[16]; |
| 59 |
for (i = 0; i < 16; ++i) |
| 60 |
{ |
| 61 |
a[i] = 1; |
| 62 |
b[i] = 2; |
| 63 |
c[i] = 3; |
| 64 |
d[i] = 4; |
| 65 |
} |
| 66 |
foo (16, a, b); |
| 67 |
bar (4, c, d); |
| 68 |
for (i = 0; i < 16; ++i) |
| 69 |
{ |
| 70 |
if (a[i] != 2) |
| 71 |
abort (); |
| 72 |
if (c[i] != 4) |
| 73 |
abort (); |
| 74 |
} |
| 75 |
} |