RFR: 8346664: C2: Optimize mask check with constant offset [v24]

[Matthias Ernst]

On Tue, 25 Feb 2025 07:22:56 GMT, Emanuel Peter <epeter at openjdk.org> wrote:

>> Matthias Ernst has updated the pull request incrementally with one additional commit since the last revision:
>> 
>>   incorporate @eme64's comment suggestions
>
> About severity: As long as we find and integrate a fix during `JDK25` it is fine (the issue does not break the CI that badly at the moment). If we get close to rampdown, then we have to consider if we want to backout 8346664 or if we defer the bug to `JDK26`.

The good news: it's easy to reproduce the issue by adding a loop to CheckProp, and I have a fix up for CCP that appears(!) straightforward. I do not quite understand the interplay with IGVN, the set of "push" rules does not look consistent between the two phases to me. But re-reading @eme64's response, CCP is necessary for correctness, IGVN is optimization potential(?).

That said, I have taken up new employment and need to sort out the contribution framework first before I can contribute the fix.

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

PR Comment: https://git.openjdk.org/jdk/pull/22856#issuecomment-2689878767