fma(), fmaf(), fmal()

Multiply two floating-point numbers and then add a third number

Synopsis:

#include <math.h>

double fma( double x,
            double y,
            double z );

float fmaf( float x,
            float y,
            float z );

long double fmal( long double x,
                  long double y,
                  long double z );

Arguments:

x, y: The numbers that you want to multiply together.
z: The number you want to add to the product.

Library:

libm: The general-purpose math library.
libm-sve: (QNX Neutrino 7.1 or later) A library that optimizes the code for ARMv8.2 chips that have Scalable Vector Extension hardware.

Your system requirements will determine how you should work with these libraries:

If you want only selected processes to run with the SVE version, you can include both libraries in your OS image and use the -l m or -l m-sve option to qcc to link explicitly against the appropriate one.
If you want all processes to use the SVE version, include libm-sve.so in your OS image and set up a symbolic link from libm.so to libm-sve.so. Use the -l m option to qcc to link against the library.

Note: Compile your program with the -fno-builtin option to prevent the compiler from using a built-in version of the function.

Description:

The fma(), fmaf(), and fmal() (“fused multiply-add”) functions calculate (x * y) + z, rounded as one ternary operation. That is, they compute the value to infinite precision and then round the result once, according to the current rounding mode as specified by FLT_ROUNDS.

To check for error situations, use feclearexcept() and fetestexcept(). For example:

Call feclearexcept(FE_ALL_EXCEPT) before calling fma(), fmaf(), or fmal().
On return, if fetestexcept(FE_ALL_EXCEPT) is nonzero, then an error has occurred.

Returns:

(x * y) + z, rounded as one ternary operation.

If:	These functions return:	Errors:
The result overflows	Inf	FE_OVERFLOW
The result underflows	The correct result, after rounding	FE_UNDERFLOW
`x` or `y` is NaN	NaN	—
`x` * `y` is an exact infinity, and `z` is also an infinity but with the opposite sign	NaN	FE_INVALID
One of `x` and `y` is infinite, the other is zero, and `z` isn't a NaN	NaN	FE_INVALID
One of `x` and `y` is infinite, the other is zero, and `z` is NaN	NaN	FE_INVALID
`x` * `y` isn't 0.0 * Inf or Inf * 0.0, and `z` is NaN	NaN	—

These functions raise FE_INEXACT if the FPU reports that the result can't be exactly represented as a floating-point number.

Classification:

C11, POSIX 1003.1

Safety:
Cancellation point	No
Interrupt handler	Yes
Signal handler	Yes
Thread	Yes