| 1 |
/* { dg-do run } */ |
| 2 |
/* { dg-require-effective-target p8vector_hw } */ |
| 3 |
/* { dg-options "-mdejagnu-cpu=power8 -O3" } */ |
| 4 |
|
| 5 |
#include <altivec.h> |
| 6 |
void abort (); |
| 7 |
|
| 8 |
#define N 4096 |
| 9 |
long long ca[N] __attribute__((aligned(16))); |
| 10 |
long long cb[N] __attribute__((aligned(16))); |
| 11 |
long long cc[N] __attribute__((aligned(16))); |
| 12 |
long long cd[N] __attribute__((aligned(16))); |
| 13 |
long long x; |
| 14 |
|
| 15 |
__attribute__((noinline)) void foo () |
| 16 |
{ |
| 17 |
int i; |
| 18 |
vector long long va, vb, vc, vd, tmp; |
| 19 |
volatile unsigned long long three = 3; |
| 20 |
vector unsigned long long threes = vec_splats (three); |
| 21 |
for (i = 0; i < N; i+=2) { |
| 22 |
vb = vec_vsx_ld (0, (vector long long *)&cb[i]); |
| 23 |
vc = vec_vsx_ld (0, (vector long long *)&cc[i]); |
| 24 |
vd = vec_vsx_ld (0, (vector long long *)&cd[i]); |
| 25 |
tmp = vec_add (vb, vc); |
| 26 |
tmp = vec_sub (tmp, vd); |
| 27 |
tmp = vec_sra (tmp, threes); |
| 28 |
x = vec_extract (tmp, 0); |
| 29 |
vec_vsx_st (tmp, 0, (vector long long *)&ca[i]); |
| 30 |
} |
| 31 |
} |
| 32 |
|
| 33 |
__attribute__((noinline)) void init () |
| 34 |
{ |
| 35 |
int i; |
| 36 |
for (i = 0; i < N; ++i) { |
| 37 |
cb[i] = 3 * i - 2048; |
| 38 |
cc[i] = -5 * i + 93; |
| 39 |
cd[i] = i + 14; |
| 40 |
} |
| 41 |
} |
| 42 |
|
| 43 |
int main () |
| 44 |
{ |
| 45 |
int i; |
| 46 |
init (); |
| 47 |
foo (); |
| 48 |
for (i = 0; i < N; ++i) |
| 49 |
if (ca[i] != (-3 * i - 1969) >> 3) |
| 50 |
abort (); |
| 51 |
if (x != ca[N-1]) |
| 52 |
abort (); |
| 53 |
return 0; |
| 54 |
} |