| 1 |
298 |
jeremybenn |
/* { dg-require-effective-target vect_int } */
|
| 2 |
|
|
|
| 3 |
|
|
#include <stdarg.h>
|
| 4 |
|
|
#include "tree-vect.h"
|
| 5 |
|
|
|
| 6 |
|
|
#define N 32
|
| 7 |
|
|
|
| 8 |
|
|
unsigned short sa[N];
|
| 9 |
|
|
unsigned short sc[N] = {0,1,2,3,4,5,6,7,8,9,10,11,12,13,14,15,
|
| 10 |
|
|
16,17,18,19,20,21,22,23,24,25,26,27,28,29,30,31};
|
| 11 |
|
|
unsigned short sb[N] = {0,1,2,3,4,5,6,7,8,9,10,11,12,13,14,15,
|
| 12 |
|
|
16,17,18,19,20,21,22,23,24,25,26,27,28,29,30,31};
|
| 13 |
|
|
unsigned int ia[N];
|
| 14 |
|
|
unsigned int ic[N] = {0,3,6,9,12,15,18,21,24,27,30,33,36,39,42,45,
|
| 15 |
|
|
0,1,2,3,4,5,6,7,8,9,10,11,12,13,14,15};
|
| 16 |
|
|
unsigned int ib[N] = {0,3,6,9,12,15,18,21,24,27,30,33,36,39,42,45,
|
| 17 |
|
|
0,1,2,3,4,5,6,7,8,9,10,11,12,13,14,15};
|
| 18 |
|
|
|
| 19 |
|
|
/* Current peeling-for-alignment scheme will consider the 'sa[i+7]'
|
| 20 |
|
|
access for peeling, and therefore will examine the option of
|
| 21 |
|
|
using a peeling factor = VF-7%VF. This will result in a peeling factor 1,
|
| 22 |
|
|
which will also align the access to 'ia[i+3]', and the loop could be
|
| 23 |
|
|
vectorized on all targets that support unaligned loads. */
|
| 24 |
|
|
|
| 25 |
|
|
__attribute__ ((noinline))
|
| 26 |
|
|
int main1 (int n)
|
| 27 |
|
|
{
|
| 28 |
|
|
int i;
|
| 29 |
|
|
|
| 30 |
|
|
/* Multiple types with different sizes, used in independent
|
| 31 |
|
|
copmutations. Vectorizable. */
|
| 32 |
|
|
for (i = 0; i < n; i++)
|
| 33 |
|
|
{
|
| 34 |
|
|
sa[i+7] = sb[i] + sc[i];
|
| 35 |
|
|
ia[i+3] = ib[i] + ic[i];
|
| 36 |
|
|
}
|
| 37 |
|
|
|
| 38 |
|
|
/* check results: */
|
| 39 |
|
|
for (i = 0; i < n; i++)
|
| 40 |
|
|
{
|
| 41 |
|
|
if (sa[i+7] != sb[i] + sc[i] || ia[i+3] != ib[i] + ic[i])
|
| 42 |
|
|
abort ();
|
| 43 |
|
|
}
|
| 44 |
|
|
|
| 45 |
|
|
return 0;
|
| 46 |
|
|
}
|
| 47 |
|
|
|
| 48 |
|
|
/* Current peeling-for-alignment scheme will consider the 'ia[i+3]'
|
| 49 |
|
|
access for peeling, and therefore will examine the option of
|
| 50 |
|
|
using a peeling factor = VF-3%VF. This will result in a peeling factor
|
| 51 |
|
|
1 if VF=4,2. This will not align the access to 'sa[i+3]', for which we
|
| 52 |
|
|
need to peel 5,1 iterations for VF=4,2 respectively, so the loop can not
|
| 53 |
|
|
be vectorized. */
|
| 54 |
|
|
|
| 55 |
|
|
__attribute__ ((noinline))
|
| 56 |
|
|
int main2 (int n)
|
| 57 |
|
|
{
|
| 58 |
|
|
int i;
|
| 59 |
|
|
|
| 60 |
|
|
/* Multiple types with different sizes, used in independent
|
| 61 |
|
|
copmutations. Vectorizable. */
|
| 62 |
|
|
for (i = 0; i < n; i++)
|
| 63 |
|
|
{
|
| 64 |
|
|
ia[i+3] = ib[i] + ic[i];
|
| 65 |
|
|
sa[i+3] = sb[i] + sc[i];
|
| 66 |
|
|
}
|
| 67 |
|
|
|
| 68 |
|
|
/* check results: */
|
| 69 |
|
|
for (i = 0; i < n; i++)
|
| 70 |
|
|
{
|
| 71 |
|
|
if (sa[i+3] != sb[i] + sc[i] || ia[i+3] != ib[i] + ic[i])
|
| 72 |
|
|
abort ();
|
| 73 |
|
|
}
|
| 74 |
|
|
|
| 75 |
|
|
return 0;
|
| 76 |
|
|
}
|
| 77 |
|
|
|
| 78 |
|
|
int main (void)
|
| 79 |
|
|
{
|
| 80 |
|
|
check_vect ();
|
| 81 |
|
|
|
| 82 |
|
|
main1 (N-7);
|
| 83 |
|
|
main2 (N-3);
|
| 84 |
|
|
|
| 85 |
|
|
return 0;
|
| 86 |
|
|
}
|
| 87 |
|
|
|
| 88 |
|
|
/* { dg-final { scan-tree-dump-times "vectorized 1 loops" 2 "vect" { xfail {! vect_hw_misalign} } } } */
|
| 89 |
|
|
/* { dg-final { scan-tree-dump-times "vectorized 1 loops" 1 "vect" { xfail { vect_no_align || vect_hw_misalign } } } } */
|
| 90 |
|
|
/* { dg-final { scan-tree-dump-times "Alignment of access forced using peeling" 2 "vect" { xfail {! vect_hw_misalign} } } } */
|
| 91 |
|
|
/* { dg-final { scan-tree-dump-times "Alignment of access forced using peeling" 1 "vect" { xfail { vect_no_align || vect_hw_misalign } } } } */
|
| 92 |
|
|
/* { dg-final { scan-tree-dump-times "Vectorizing an unaligned access" 8 "vect" { xfail *-*-* } } } */
|
| 93 |
|
|
/* { dg-final { scan-tree-dump-times "Vectorizing an unaligned access" 4 "vect" { xfail { vect_no_align || vect_hw_misalign } } } } */
|
| 94 |
|
|
/* { dg-final { cleanup-tree-dump "vect" } } */
|
| 95 |
|
|
|
