[llvm] Enable logf128 constant folding for hosts with 128bit long double (PR #104929)

[Matthew Devereau via llvm-commits]

MDevereau wrote:

It is known that clang on x86-64 doesn't include logf128 via #include <cmath> (See https://github.com/llvm/llvm-project/pull/96287#discussion_r1674413381). This was planned to be implemented once this patch landed and will need some extra care to get working. This feature is still a work in progress and is not yet exhaustive for all targets. Thank you for the information about the contents of floatn.h, that's useful to know.

ppc is not compatible due to usage of the ibm128 floating point representation rather than IEE754 floats. Thus, we have disabled this for ppc (and the tests now properly do not run now thanks to https://github.com/llvm/llvm-project/commit/001e423ac6261283f0289a774bf5e7577adb1ea6). I personally do not have plans to implement a ppc compatible solution for this and will disable it for this host type.



https://github.com/llvm/llvm-project/pull/104929