[compiler-rt] [compiler-rt][builtins]Fix complex division for aarch64 (PR #106664)

[Eric Christopher via llvm-commits]

echristo wrote:

> Side note: divsc3.c appears to be the only place in `builtins` where __compiler_rt_fmaxf is used.
> 
> What do you think - would it possible to move forward here ? (this would make things a bit more consistent across architectures / compilers). Otherwise, I agree with the argument presented above "The behavior of signaling NaNs in Annex G is implementation-defined" (I missed it initially) - can abandon this PR.

While it is implementation defined, if we can make the result the same with little performance penalty in order to smooth portability between both platforms it wouldn't necessarily be a bad thing.

Eli, David - thoughts?

https://github.com/llvm/llvm-project/pull/106664