[libc-commits] [libc] [libc] Float80 Emulation in LLVM libc (PR #211350)

[Oliver Hunt via libc-commits]

ojhunt wrote:

Having implemented this before I'd make sure you simply reject the bad parts (pseudo-{NaN, infinities, denormals}, unnormals, etc). It might be convenient to use float128 for some of the operations but it's likely you'll hit double rounding in some cases, I'm unsure if that matters for transcendental functions in libc, but it would be good to include test cases that would cover such failures so they're at least known.

https://github.com/llvm/llvm-project/pull/211350