1 /*
2 Copyright (C) 2002 by Red Hat, Incorporated. All rights reserved.
3
4 Permission to use, copy, modify, and distribute this software
5 is freely granted, provided that this notice is preserved.
6 */
7 /*
8 FUNCTION
9 <<fma>>, <<fmaf>>---floating multiply add
10 INDEX
11 fma
12 INDEX
13 fmaf
14
15 SYNOPSIS
16 #include <math.h>
17 double fma(double <[x]>, double <[y]>, double <[z]>);
18 float fmaf(float <[x]>, float <[y]>, float <[z]>);
19
20 DESCRIPTION
21 The <<fma>> functions compute (<[x]> * <[y]>) + <[z]>, rounded as one ternary
22 operation: they compute the value (as if) to infinite precision and round once
23 to the result format, according to the rounding mode characterized by the value
24 of FLT_ROUNDS. That is, they are supposed to do this: see below.
25
26 RETURNS
27 The <<fma>> functions return (<[x]> * <[y]>) + <[z]>, rounded as one ternary
28 operation.
29
30 BUGS
31 This implementation does not provide the function that it should, purely
32 returning "(<[x]> * <[y]>) + <[z]>;" with no attempt at all to provide the
33 simulated infinite precision intermediates which are required. DO NOT USE THEM.
34
35 If double has enough more precision than float, then <<fmaf>> should provide
36 the expected numeric results, as it does use double for the calculation. But
37 since this is not the case for all platforms, this manual cannot determine
38 if it is so for your case.
39
40 PORTABILITY
41 ANSI C, POSIX.
42
43 */
44
45 #include "fdlibm.h"
46
47 #if !_HAVE_FAST_FMA
48
49 #ifdef _NEED_FLOAT64
50
51 #ifdef __clang__
52 #pragma STDC FP_CONTRACT OFF
53 #endif
54
55 __float64
fma64(__float64 x,__float64 y,__float64 z)56 fma64(__float64 x, __float64 y, __float64 z)
57 {
58 /* Implementation defined. */
59 return (x * y) + z;
60 }
61
62 _MATH_ALIAS_d_ddd(fma)
63
64 #endif /* _NEED_FLOAT64 */
65
66 #endif /* !_HAVE_FAST_FMA */
67
