111
|
1 /* { dg-require-effective-target vect_int } */
|
|
2 /* { dg-require-effective-target vect_shift } */
|
|
3
|
|
4 #include <stdarg.h>
|
|
5 #include "tree-vect.h"
|
|
6
|
|
7 #define N 512
|
|
8
|
|
9 /* Modified rgb to rgb conversion from FFmpeg. */
|
|
10 __attribute__ ((noinline)) void
|
|
11 foo (unsigned char *src, unsigned char *dst)
|
|
12 {
|
|
13 unsigned char *s = src;
|
|
14 int *d = (int *)dst;
|
|
15 int i;
|
|
16
|
|
17 for (i = 0; i < N/4; i++)
|
|
18 {
|
|
19 const int b = *s++;
|
|
20 const int g = *s++;
|
|
21 const int r = *s++;
|
|
22 const int a = *s++;
|
|
23 *d = ((b>>3) | ((g&0xFC)<<3) | ((r&0xF8)<<8) | (a>>5));
|
|
24 d++;
|
|
25 }
|
|
26
|
|
27 s = src;
|
|
28 d = (int *)dst;
|
|
29 for (i = 0; i < N/4; i++)
|
|
30 {
|
|
31 const int b = *s++;
|
|
32 const int g = *s++;
|
|
33 const int r = *s++;
|
|
34 const int a = *s++;
|
|
35 if (*d != ((b>>3) | ((g&0xFC)<<3) | ((r&0xF8)<<8) | (a>>5)))
|
|
36 abort ();
|
|
37 d++;
|
|
38 }
|
|
39 }
|
|
40
|
|
41 int main (void)
|
|
42 {
|
|
43 int i;
|
|
44 unsigned char in[N], out[N];
|
|
45
|
|
46 check_vect ();
|
|
47
|
|
48 for (i = 0; i < N; i++)
|
|
49 {
|
|
50 in[i] = i;
|
|
51 out[i] = 255;
|
|
52 __asm__ volatile ("");
|
|
53 }
|
|
54
|
|
55 foo (in, out);
|
|
56
|
|
57 return 0;
|
|
58 }
|
|
59
|
131
|
60 /* This is an over-widening even though the final result is still an int.
|
|
61 It's better to do one vector of ops on chars and then widen than to
|
|
62 widen and then do 4 vectors of ops on ints. */
|
|
63 /* { dg-final { scan-tree-dump {vect_recog_over_widening_pattern: detected:[^\n]* << 3} "vect" } } */
|
|
64 /* { dg-final { scan-tree-dump {vect_recog_over_widening_pattern: detected:[^\n]* >> 3} "vect" } } */
|
|
65 /* { dg-final { scan-tree-dump {vect_recog_over_widening_pattern: detected:[^\n]* << 8} "vect" } } */
|
|
66 /* { dg-final { scan-tree-dump {vect_recog_over_widening_pattern: detected:[^\n]* >> 5} "vect" } } */
|
111
|
67 /* { dg-final { scan-tree-dump-times "vectorized 1 loops" 1 "vect" } } */
|
|
68
|