[libc-commits] [libc] [llvm] [libc] Make exponent decomposition f128 functions use emulated Float128 (PR #229855)

[Mohammed Sohail via libc-commits]

msohail22 wrote:

@lntue @overmighty ping — this is ready for a look.

The review from https://github.com/llvm/llvm-project/pull/221540 has been applied (missing `llogbf128` entrypoints added for `darwin/x86_64` and `windows`, `\file` header comments, deps normalised), and that thread is resolved.

The two red checks on the previous SHA were CI infrastructure, not this patch:

- FreeBSD VM `build` — `pkg update` fails with "repository FreeBSD-ports contains packages for wrong OS version: FreeBSD:15:amd64" before any compilation happens.
- `libc-fullbuild on linux-x86_64-MinSizeRel` — built and installed successfully, then `connect ETIMEDOUT 34.118.224.1:443` to the k8s API.

I have pushed an empty commit to re-run them, so the board should settle shortly. Every other config was green (gcc 7-11, MSVC, Bazel, aarch64, qemu, the other fullbuilds).

Two open questions for you:

- There is no non-smoke `llogbf128` test, because `ILogbTest.h` hardcodes `int (*Func)(T)` and `llogbf128` returns `long`. Worth templating the return type, or leave it?
- I left the headers of `shared_math_test.cpp`, `shared_math_constexpr_test.cpp`, `ILogbTest.h` and `LogbTest.h` untouched. Tell me if you want those converted too.


https://github.com/llvm/llvm-project/pull/229855