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 { target { ! { ia32 } } } } */
6/* { dg-options "-O2 -mfma4 -mno-fma" } */
7
8extern void exit (int);
9
10float
11flt_mul_add (float a, float b, float c)
12{
13  return (a * b) + c;
14}
15
16double
17dbl_mul_add (double a, double b, double c)
18{
19  return (a * b) + c;
20}
21
22float
23flt_mul_sub (float a, float b, float c)
24{
25  return (a * b) - c;
26}
27
28double
29dbl_mul_sub (double a, double b, double c)
30{
31  return (a * b) - c;
32}
33
34float
35flt_neg_mul_add (float a, float b, float c)
36{
37  return (-(a * b)) + c;
38}
39
40double
41dbl_neg_mul_add (double a, double b, double c)
42{
43  return (-(a * b)) + c;
44}
45
46float
47flt_neg_mul_sub (float a, float b, float c)
48{
49  return (-(a * b)) - c;
50}
51
52double
53dbl_neg_mul_sub (double a, double b, double c)
54{
55  return (-(a * b)) - c;
56}
57
58float  f[10] = { 2, 3, 4 };
59double d[10] = { 2, 3, 4 };
60
61int main ()
62{
63  f[3] = flt_mul_add (f[0], f[1], f[2]);
64  f[4] = flt_mul_sub (f[0], f[1], f[2]);
65  f[5] = flt_neg_mul_add (f[0], f[1], f[2]);
66  f[6] = flt_neg_mul_sub (f[0], f[1], f[2]);
67
68  d[3] = dbl_mul_add (d[0], d[1], d[2]);
69  d[4] = dbl_mul_sub (d[0], d[1], d[2]);
70  d[5] = dbl_neg_mul_add (d[0], d[1], d[2]);
71  d[6] = dbl_neg_mul_sub (d[0], d[1], d[2]);
72  exit (0);
73}
74
75/* { dg-final { scan-assembler "vfmaddss" } } */
76/* { dg-final { scan-assembler "vfmaddsd" } } */
77/* { dg-final { scan-assembler "vfmsubss" } } */
78/* { dg-final { scan-assembler "vfmsubsd" } } */
79/* { dg-final { scan-assembler "vfnmaddss" } } */
80/* { dg-final { scan-assembler "vfnmaddsd" } } */
81/* { dg-final { scan-assembler "vfnmsubss" } } */
82/* { dg-final { scan-assembler "vfnmsubsd" } } */
83