[PATCH] D70096: [strictfp] Replace dangling strictfp attrs with nobuiltin

[Andy Kaylor via Phabricator via llvm-commits]

andrew.w.kaylor added a comment.

In D70096#1741213 <https://reviews.llvm.org/D70096#1741213>, @kpn wrote:

> Do we want to restrict this so it doesn't get upgraded if found in a future bitcode version?


How would we do that?

> Do we want to not upgrade if a constrained intrinsic is used anywhere in the function? We do want to be able to distinguish between pre-constrained intrinsic uses and modern constrained uses that are wrong. I am assuming that the upgrade happens before the IR verifier runs.

Honestly, I'd be very surprised if anyone is using this in a product currently. After the verifier support is added we'll be able to catch incorrect uses as they happen. Incorrect shouldn't get to this point. After the verifier support is complete, functions containing constrained intrinsics without the strictfp attribute on the function definition will fail verification, so I don't think checking for that case here really adds anything.

I'm also assuming that the upgrade happens before the verifier runs. I've chosen to only add this auto-upgrade in the BitcodeReader so that we can still write tests in the text format when the verifier support is added.


CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D70096/new/

https://reviews.llvm.org/D70096