rjmccall wrote: Oh, that's interesting. I'd been assuming this was a cross-function issue or something like that, but that's a great example of where we need more than that. I agree that it feels like having some kind of barrier instruction is the right way to go — basically a unary operator that can have fast-math restrictions on its operand/result just like all the binary operators and intrinsics do, and then we can decide the right places to emit that.
https://github.com/llvm/llvm-project/pull/105912 _______________________________________________ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
