https://github.com/antoniofrighetto commented: @pratlucas It seems like there is no equivalent fine-grained option as in Clang, but `-ftrapping-math` should be the one that let GCC assume that FP operations can trap. https://github.com/llvm/llvm-project/pull/136139