Theodore Omtzigt

@touisteur indeed, the elementary functions in Universal do not yet use native implementations. However, extensive research has been done on what is required for correctly rounded math libraries. In particular,...

@touisteur just released v3.34 which has a unified mathlib shim among fixpnt, cfloat, and posit. While adding the fixpnt shim, the pattern of a general math lib shim that uses...

@touisteur thanks for the update. I'll add the mathlib shim on the todo list, and I will use your (this) issue to trigger the first step of implementing a native...

@touisteur I have collected a set of references that we can compare and contrast: gcc's libm's implementation: https://code.woboq.org/userspace/glibc/sysdeps/ieee754/dbl-64/e_log.c.html musl: https://git.musl-libc.org/cgit/musl/tree/src/math/log.c fdlibm: https://www.netlib.org/fdlibm/e_log.c rlibm: https://github.com/rutgers-apl/rlibm-32/blob/main/source/posit32/log.c I'll ping Jay and Santosh to...

Well, not quite: it still uses doubles for compute, and maps back to posit on the final conversion. ``` posit32_t rlibm_log(posit32_t x) { if (x.v >= 0x80000000) { return castP32(0x80000000);...

that is the $64000 question. I just pinged the FPbench.org community with that question. There are several research groups (Rutgers, Utah, and UWash) that are working on math function library...

@touisteur took a look at fpm. If I implement the log/exp functions for you, would you consider using Universal instead of fpm?

Cool, I'll put it on my docket for this week. I'll report here on the progress.

regarding a base class design: we have a new set of number systems, `areal` and `bfloat`, which use constexpr machinery to reduce the need for specialized classes as we have...

@thorsten-ottosen fixed the implementation of operator