1 /* { dg-do run } */
2 /* { dg-options "-O2 -mavx512f -mavx512vnni" } */
3 /* { dg-require-effective-target avx512f } */
4 /* { dg-require-effective-target avx512vnni } */
5
6 #define AVX512F
7
8 #define AVX512VNNI
9 #include "avx512f-helper.h"
10
11 #define SIZE (AVX512F_LEN / 16)
12 #define SIZE_RES (AVX512F_LEN / 32)
13
14 #include "avx512f-mask-type.h"
15
16 static void
CALC(int * r,int * dst,short * s1,short * s2)17 CALC (int *r, int *dst, short *s1, short *s2)
18 {
19 short tempres[SIZE];
20 for (int i = 0; i < SIZE; i++) {
21 tempres[i] = ((int)(s1[i]) * (int)(s2[i]));
22 }
23 for (int i = 0; i < SIZE_RES; i++) {
24 long long test = (long long)dst[i] + tempres[i*2] + tempres[i*2 + 1];
25 r[i] = test;
26 }
27 }
28
29 void
TEST(void)30 TEST (void)
31 {
32 int i;
33 UNION_TYPE (AVX512F_LEN, i_d) res1, res2, res3;
34 UNION_TYPE (AVX512F_LEN, i_w) src1, src2;
35 MASK_TYPE mask = MASK_VALUE;
36 int res_ref[SIZE_RES];
37 int res_ref2[SIZE_RES];
38
39 for (i = 0; i < SIZE; i++)
40 {
41 src1.a[i] = 1 + i;
42 src2.a[i] = 2 + 2*i;
43 }
44
45 for (i = 0; i < SIZE_RES; i++)
46 {
47 res1.a[i] = 0x7fffffff;
48 res2.a[i] = DEFAULT_VALUE;
49 res3.a[i] = DEFAULT_VALUE;
50 }
51
52 CALC (res_ref, res1.a, src1.a, src2.a);
53 CALC (res_ref2, res2.a, src1.a, src2.a);
54
55 res1.x = INTRINSIC (_dpwssd_epi32) (res1.x, src1.x, src2.x);
56 res2.x = INTRINSIC (_mask_dpwssd_epi32) (res2.x, mask, src1.x, src2.x);
57 res3.x = INTRINSIC (_maskz_dpwssd_epi32) (mask, res3.x, src1.x, src2.x);
58
59 if (UNION_CHECK (AVX512F_LEN, i_d) (res1, res_ref))
60 abort ();
61
62 MASK_MERGE (i_d) (res_ref2, mask, SIZE_RES);
63 if (UNION_CHECK (AVX512F_LEN, i_d) (res2, res_ref2))
64 abort ();
65
66 MASK_ZERO (i_d) (res_ref2, mask, SIZE_RES);
67 if (UNION_CHECK (AVX512F_LEN, i_d) (res3, res_ref2))
68 abort ();
69 }
70