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
12 union
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
21 /* Note that in macc*,msub*,mnmacc* and mnsub* instructions, the intermdediate
22 product is not rounded, only the addition is rounded. */
23
24 static void
init_maccps()25 init_maccps ()
26 {
27 int i;
28 for (i = 0; i < NUM * 8; i++)
29 {
30 src1.f[i] = i;
31 src2.f[i] = i + 10;
32 src3.f[i] = i + 20;
33 }
34 }
35
36 static void
init_maccpd()37 init_maccpd ()
38 {
39 int i;
40 for (i = 0; i < NUM * 4; i++)
41 {
42 src1.d[i] = i;
43 src2.d[i] = i + 10;
44 src3.d[i] = i + 20;
45 }
46 }
47
48 static int
check_maccps()49 check_maccps ()
50 {
51 int i, j, check_fails = 0;
52 for (i = 0; i < NUM * 8; i = i + 8)
53 for (j = 0; j < 8; j++)
54 {
55 res.f[i + j] = (src1.f[i + j] * src2.f[i + j]) + src3.f[i + j];
56 if (dst.f[i + j] != res.f[i + j])
57 check_fails++;
58 }
59 return check_fails++;
60 }
61
62 static int
check_maccpd()63 check_maccpd ()
64 {
65 int i, j, check_fails = 0;
66 for (i = 0; i < NUM * 4; i = i + 4)
67 for (j = 0; j < 4; j++)
68 {
69 res.d[i + j] = (src1.d[i + j] * src2.d[i + j]) + src3.d[i + j];
70 if (dst.d[i + j] != res.d[i + j])
71 check_fails++;
72 }
73 return check_fails++;
74 }
75
76 static void
fma4_test(void)77 fma4_test (void)
78 {
79 int i;
80
81 init_maccps ();
82
83 for (i = 0; i < NUM; i++)
84 dst.x[i] = _mm256_macc_ps (src1.x[i], src2.x[i], src3.x[i]);
85
86 if (check_maccps ())
87 abort ();
88
89 init_maccpd ();
90
91 for (i = 0; i < NUM; i++)
92 dst.y[i] = _mm256_macc_pd (src1.y[i], src2.y[i], src3.y[i]);
93
94 if (check_maccpd ())
95 abort ();
96 }
97