[libc-commits] [libc] [libc][math][c23] Add exp2m1f16 C23 math function (PR #105690)

[Matt Arsenault via libc-commits]

arsenm wrote:

> The usage seems to be with the rounding mode check, so it's important that nothing gets constant propagated. I actually don't know how the rounding mode i handled when

They won't be if you are using strictfp (as you must be if you depend on the rounding mode). If you do know the rounding mode, they will correctly fold.


> These try to get the rounding mode through math operations so it's really important that it's not constant propagated. If you make it a constant then it will go through the LLVM optimizer which just assumes it's to-nearest, see https://godbolt.org/z/ehejrv1zK. We could probably avoid the volatile with some inline assembly or something.

You need to enable fenv access

https://github.com/llvm/llvm-project/pull/105690