kargl

This isn't a gfortran problem (except possibly for REAL(16) which uses libquadmath). Your math library is broken, which is trivial to prove. real(4): gfcx -c c.f90 && nm c.o |...

I already told you that tgammaf, tgamma, and tgammal came from libm and tgammaq came from libquadpack. You need to complain to your operating system vendor. I think this suggestion...

Here's the test code. Sorry about poor formatting, but github interface does not seem to hone fixed width fonts with whitespace. ```c #include #include #include int main(void) { fexcept_t flags;...

Note, lgammal(-n) correctly raises the FE_DIVBYZERO exception. lgammaf and lgamma have issues.

What does MPS mean? I usually see speed comparisons reported as microsecond per call or cycles per call. Also, what are the compiler options used in building system libm and...

> > What does MPS mean? I usually see speed comparisons reported as microsecond per call or cycles per call. Also, what are the compiler options used in building system...

> For example, there's a recent new `log` implementation: https://git.musl-libc.org/cgit/musl/tree/src/math/log.c If you're going to update log, you should consider using the table-driven methods of PTP Tang (ACM Trans Math Soft,...

> @kargl Looks like that gives us a better `logl`, but `log` remains the same. Yeah, when logl was committed (and the same with expl), the old Sun fdlibm float...

> @kargl Would you be able to provide fast `log` implementations here? That would be fantastic if possible. Just discovered that github sets noreply@ addresses in email headers, so my...

While the max ULP is an important metric, two additional metrics may be more important: speed and distribution of ULP (if max ULP > 1). Consider `j0f(x)`. It is well-known...