[lworld] RFR: 8354981: [lworld] Atomic flat stores need preceding CPUMemBar

Tue Apr 22 17:17:38 UTC 2025

On Thu, 17 Apr 2025 16:58:19 GMT, Quan Anh Mai <qamai at openjdk.org> wrote:

> Hi,
> 
> Atomic flat stores cover many memory slices. As a result, it needs a preceding CPUMemBar or else it only takes the slice of the first field as its memory input, which is incorrect.
> 
> I don't know how to make a regression test for this. I encounter this crash while working on JDK-8335256. The crash was due to the memory input of the flat store gets split during escape analysis. The compiler then tries looking at the flat stores, which is vastly different from what it expects.
> 
> Please take a look and leave your reviews, thanks a lot.

Thanks, I have merged the PR with latest lworld

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

PR Comment: https://git.openjdk.org/valhalla/pull/1437#issuecomment-2821976104