1/* Test that the compiler properly optimizes floating point multiply
2   and add instructions into vfmaddss, vfmsubss, vfnmaddss,
3   vfnmsubss on FMA4 systems.  */
4
5/* { dg-do compile } */
6/* { dg-require-effective-target lp64 } */
7/* { dg-options "-O2 -mfma4" } */
8
9extern void exit (int);
10
11float
12flt_mul_add (float a, float b, float c)
13{
14  return (a * b) + c;
15}
16
17double
18dbl_mul_add (double a, double b, double c)
19{
20  return (a * b) + c;
21}
22
23float
24flt_mul_sub (float a, float b, float c)
25{
26  return (a * b) - c;
27}
28
29double
30dbl_mul_sub (double a, double b, double c)
31{
32  return (a * b) - c;
33}
34
35float
36flt_neg_mul_add (float a, float b, float c)
37{
38  return (-(a * b)) + c;
39}
40
41double
42dbl_neg_mul_add (double a, double b, double c)
43{
44  return (-(a * b)) + c;
45}
46
47float
48flt_neg_mul_sub (float a, float b, float c)
49{
50  return (-(a * b)) - c;
51}
52
53double
54dbl_neg_mul_sub (double a, double b, double c)
55{
56  return (-(a * b)) - c;
57}
58
59float  f[10] = { 2, 3, 4 };
60double d[10] = { 2, 3, 4 };
61
62int main ()
63{
64  f[3] = flt_mul_add (f[0], f[1], f[2]);
65  f[4] = flt_mul_sub (f[0], f[1], f[2]);
66  f[5] = flt_neg_mul_add (f[0], f[1], f[2]);
67  f[6] = flt_neg_mul_sub (f[0], f[1], f[2]);
68
69  d[3] = dbl_mul_add (d[0], d[1], d[2]);
70  d[4] = dbl_mul_sub (d[0], d[1], d[2]);
71  d[5] = dbl_neg_mul_add (d[0], d[1], d[2]);
72  d[6] = dbl_neg_mul_sub (d[0], d[1], d[2]);
73  exit (0);
74}
75
76/* { dg-final { scan-assembler "vfmaddss" } } */
77/* { dg-final { scan-assembler "vfmaddsd" } } */
78/* { dg-final { scan-assembler "vfmsubss" } } */
79/* { dg-final { scan-assembler "vfmsubsd" } } */
80/* { dg-final { scan-assembler "vfnmaddss" } } */
81/* { dg-final { scan-assembler "vfnmaddsd" } } */
82/* { dg-final { scan-assembler "vfnmsubss" } } */
83/* { dg-final { scan-assembler "vfnmsubsd" } } */
84