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

[Alexander Shaposhnikov via llvm-commits]

alexander-shaposhnikov wrote:

@davemgreen - to the best of my knowledge this came up during debugging an issue in our internal project (but I can ask the reporter for more details, though even without it I see references to signaling nans in other projects e.g. V8). Honestly i don't see how this affects our decision making - the code above (https://godbolt.org/z/vGTzc7dPh) is a valid c++ code, yet without the fix it produces the wrong numerical result on aarch64 (without any exceptions involved).
Regarding the performance - the proposed change only affects  __divsc3 (the implementation of __compiler_rt_fmaxf has not changed), but the performance implications there should be negligble (if any at all) (__divsc3 is a quite complex operation). Am I missing something ?



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