111
|
1 /* { dg-require-effective-target vect_int } */
|
|
2 /* { dg-require-effective-target arm_v8_2a_dotprod_neon_hw { target { aarch64*-*-* || arm*-*-* } } } */
|
|
3 /* { dg-additional-options "-march=armv8.2-a+dotprod" { target { aarch64*-*-* } } } */
|
|
4 /* { dg-add-options arm_v8_2a_dotprod_neon } */
|
|
5
|
|
6 #include <stdarg.h>
|
|
7 #include "tree-vect.h"
|
|
8
|
|
9 #define N 64
|
|
10
|
|
11 #define DOT1 43680
|
|
12
|
|
13 signed char X[N] __attribute__ ((__aligned__(__BIGGEST_ALIGNMENT__)));
|
|
14 signed char Y[N] __attribute__ ((__aligned__(__BIGGEST_ALIGNMENT__)));
|
|
15
|
|
16 /* char->short->int dot product.
|
|
17 The dot-product pattern should be detected.
|
|
18 Vectorizable on vect_sdot_qi targets (targets that support dot-product of
|
|
19 signed chars).
|
|
20
|
|
21 In the future could also be vectorized as widening-mult + widening-summation,
|
|
22 or with type-conversion support.
|
|
23 */
|
|
24 __attribute__ ((noinline)) int
|
|
25 foo1(int len) {
|
|
26 int i;
|
|
27 int result = 0;
|
|
28 short prod;
|
|
29
|
|
30 for (i=0; i<len; i++) {
|
|
31 prod = X[i] * Y[i];
|
|
32 result += prod;
|
|
33 }
|
|
34 return result;
|
|
35 }
|
|
36
|
|
37 int main (void)
|
|
38 {
|
|
39 int i, dot1;
|
|
40
|
|
41 check_vect ();
|
|
42
|
|
43 for (i=0; i<N; i++) {
|
|
44 X[i] = i;
|
|
45 Y[i] = 64-i;
|
|
46 __asm__ volatile ("");
|
|
47 }
|
|
48
|
|
49 dot1 = foo1 (N);
|
|
50 if (dot1 != DOT1)
|
|
51 abort ();
|
|
52
|
|
53 return 0;
|
|
54 }
|
|
55
|
|
56 /* { dg-final { scan-tree-dump-times "vect_recog_dot_prod_pattern: detected" 1 "vect" } } */
|
|
57 /* { dg-final { scan-tree-dump-times "vect_recog_widen_mult_pattern: detected" 1 "vect" } } */
|
|
58 /* { dg-final { scan-tree-dump-times "vectorized 1 loops" 1 "vect" { target vect_sdot_qi } } } */
|
|
59 /* { dg-final { scan-tree-dump-times "vectorized 1 loops" 1 "vect" { target { vect_widen_mult_qi_to_hi && vect_widen_sum_hi_to_si } } } } */
|
|
60
|