math: fix two fma issues (only affects non-nearest rounding mode, x86)

[musl] / src / math / __signbit.c

#include "libm.h"

// FIXME: macro in math.h
int __signbit(double x)
{
        union {
                double d;
                uint64_t i;
        } y = { x };
        return y.i>>63;
}