Performance cost of ::fpe_check()

[risc26z]

I've found that a large amount of CPU time is spent in the fpe_check() function (wrspice/include/fpe_check.h). In some tests I've made, this function uses more than a third of the total CPU time. Most of the time is spent in feclearexcept().

There's little point in clearing an exception that hasn't happened, so I rewrote the code so that feclearexcept() is called only if an exception is discovered. I also modified the code so that fetestexcept() is only called once.

This is essentially what I ended up with:

        if (!noerrret && Sp.GetFPEmode() == FPEcheck) {
            const int exflags_tested = FE_DIVBYZERO | FE_OVERFLOW | FE_INVALID;
            // read FP exception status into a temporary, which
            // can then be tested with minimal overhead
            int exflags = fetestexcept(exflags_tested);
            if (exflags != 0) {
                if (exflags & FE_DIVBYZERO)
                    err = E_MATHDBZ;
                else if (exflags & FE_OVERFLOW)
                    err = E_MATHOVF;
                // Ignore underflow, these really shouldn't be a problem.
                // else if (fetestexcept(FE_UNDERFLOW))
                //     err = E_MATHUNF;
                else if (exflags & FE_INVALID)
                    err = E_MATHINV;

                // feclearexcept() is slow (at least on x86-64), so try to
                // make sure we call it only when necessary.
                feclearexcept(FE_ALL_EXCEPT);
            }
        }

I haven't tested it thoroughly, but it seems to be considerably quicker. Any thoughts?