cferry-AMD wrote: @simon-camp Thanks for your review and documentation, you're right. There is no need to check the rounding mode for cast operations, so we can just drop the extra flag. https://github.com/llvm/llvm-project/pull/87614