1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
|
/* { dg-require-effective-target vect_simd_clones } */
/* { dg-additional-options "-fopenmp-simd -ffast-math" } */
#include <stdlib.h>
#include "tree-vect.h"
#define N 64
float *px, *py;
float *tx, *ty;
float *x1, *z1, *t1, *t2;
static void inline bar (const float cx, float cy,
float *vx, float *vy)
{
int j;
for (j = 0; j < N; ++j)
{
const float dx = cx - px[j];
const float dy = cy - py[j];
*vx -= dx * tx[j];
*vy -= dy * ty[j];
}
}
__attribute__((noinline, noclone)) void foo1 (int n)
{
int i;
#pragma omp simd
for (i=0; i<n; i++)
bar (px[i], py[i], x1+i, z1+i);
}
__attribute__((noinline, noclone)) void foo2 (int n)
{
volatile int i;
for (i=0; i<n; i++)
bar (px[i], py[i], x1+i, z1+i);
}
int main ()
{
float *X = (float*)malloc (N * 8 * sizeof (float));
int i;
int n = N - 1;
check_vect ();
px = &X[0];
py = &X[N * 1];
tx = &X[N * 2];
ty = &X[N * 3];
x1 = &X[N * 4];
z1 = &X[N * 5];
t1 = &X[N * 6];
t2 = &X[N * 7];
for (i=0; i<N; i++)
{
px[i] = (float) (i+2);
tx[i] = (float) (i+1);
py[i] = (float) (i+4);
ty[i] = (float) (i+3);
x1[i] = z1[i] = 1.0f;
}
foo1 (n); /* vector variant. */
for (i=0; i<N;i++)
{
t1[i] = x1[i]; x1[i] = 1.0f;
t2[i] = z1[i]; z1[i] = 1.0f;
}
foo2 (n); /* scalar variant. */
for (i=0; i<N; i++)
if (x1[i] != t1[i] || z1[i] != t2[i])
abort ();
return 0;
}
/* { dg-final { scan-tree-dump "OUTER LOOP VECTORIZED" "vect" } } */
|
