RFR: 8323220: Reassociate loop invariants involved in Cmps and Add/Subs [v6]

Joshua Cao duke at openjdk.org
Thu Feb 8 21:39:35 UTC 2024


On Thu, 8 Feb 2024 17:06:28 GMT, Joshua Cao <duke at openjdk.org> wrote:

>> In fact, that would probably be smart. Otherwise you only have result verification for constant values. And who knows, there the code might constant-fold in a way that we don't expect.
>
> I am with you that I am not 100% happy with the tests. I do not mind waiting for that PR to get integrated.
> 
> If I understand correctly, that PR is more for convenience of setting up arguments. How will that help testing correctness for random inputs?

I added correctness tests for two of the tests with random inputs. For the rest, it messes up the IR matching. This goes back to the fragility of matching add/sub nodes.

-------------

PR Review Comment: https://git.openjdk.org/jdk/pull/17375#discussion_r1483617747


More information about the hotspot-compiler-dev mailing list