1/* { dg-do run } */
2/* { dg-require-effective-target fma4 } */
3/* { dg-options "-O2 -mfma4" } */
4
5#include "fma4-check.h"
6
7#include <x86intrin.h>
8#include <string.h>
9
10#define NUM 20
11
12union
13{
14  __m256 x[NUM];
15  float f[NUM * 8];
16  __m256d y[NUM];
17  double d[NUM * 4];
18} dst, res, src1, src2, src3;
19
20/* Note that in macc*,msub*,mnmacc* and mnsub* instructions, the intermdediate
21   product is not rounded, only the addition is rounded. */
22
23static void
24init_nmaccps ()
25{
26  int i;
27  for (i = 0; i < NUM * 8; i++)
28    {
29      src1.f[i] = i;
30      src2.f[i] = i + 10;
31      src3.f[i] = i + 20;
32    }
33}
34
35static void
36init_nmaccpd ()
37{
38  int i;
39  for (i = 0; i < NUM * 4; i++)
40    {
41      src1.d[i] = i;
42      src2.d[i] = i + 10;
43      src3.d[i] = i + 20;
44    }
45}
46
47static int
48check_nmaccps ()
49{
50  int i, j, check_fails = 0;
51  for (i = 0; i < NUM * 8; i = i + 8)
52    for (j = 0; j < 8; j++)
53      {
54	res.f[i + j] = - (src1.f[i + j] * src2.f[i + j]) + src3.f[i + j];
55	if (dst.f[i + j] != res.f[i + j])
56	  check_fails++;
57      }
58  return check_fails++;
59}
60
61static int
62check_nmaccpd ()
63{
64  int i, j, check_fails = 0;
65  for (i = 0; i < NUM * 4; i = i + 4)
66    for (j = 0; j < 4; j++)
67      {
68	res.d[i + j] = - (src1.d[i + j] * src2.d[i + j]) + src3.d[i + j];
69	if (dst.d[i + j] != res.d[i + j])
70	  check_fails++;
71      }
72  return check_fails++;
73}
74
75static void
76fma4_test (void)
77{
78  int i;
79
80  init_nmaccps ();
81
82  for (i = 0; i < NUM; i++)
83    dst.x[i] = _mm256_nmacc_ps (src1.x[i], src2.x[i], src3.x[i]);
84
85  if (check_nmaccps ())
86    abort ();
87
88  init_nmaccpd ();
89
90  for (i = 0; i < NUM; i++)
91    dst.y[i] = _mm256_nmacc_pd (src1.y[i], src2.y[i], src3.y[i]);
92
93  if (check_nmaccpd ())
94    abort ();
95
96}
97