RFR: 8338197: [ubsan] ad_x86.hpp:6417:11: runtime error: shift exponent 100 is too large for 32-bit type 'unsigned int' [v5]

[Andrew Dinn]

On Tue, 30 Sep 2025 16:32:35 GMT, Boris Ulasevich <bulasevich at openjdk.org> wrote:

> Proposal for this PR:
>
>     land the shift-capping at the source (inside the operator) as a strict safety fix.
>     optionally omit the 100->30 tweak.

Yes let's do shift capping and drop the 100->30 latency tweak.

> Let’s keep this PR as a minimal sanitizer fix and move the model/weighting discussion to a follow-up RFE. I’m open to doing some research in this area, though I’m not entirely sure what the best starting point would be. That said, if you feel strongly against this change, I don’t mind dropping it and leaving the rest as is.

Yes, I agree we need a separate RFE for that. I filed https://bugs.openjdk.org/browse/JDK-8368971.

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

PR Comment: https://git.openjdk.org/jdk/pull/26890#issuecomment-3353173210