On Fri, 24 Jul 2026 15:55:45 GMT, Manuel Hässig <[email protected]> wrote:

> The test `compiler/igvn/TestMaskedStoreIdealization.java` uses a custom node 
> regex to match very specific stores. Because the aryptr memory dump was 
> amended with atomicity and flatness information, the regexes need to match 
> that as well.
> 
> ---------
> - [x] I confirm that I make this contribution in accordance with the [OpenJDK 
> Interim AI Policy](https://openjdk.org/legal/ai).

That part does not seem to be the problem, but testing is revealing 
intermittently different code shapes. Why can it never be simple?

I had to additionally exclude the subtest `testInteger4Mask` specifically from 
IR validation. The code generated in the case where the mask is set to all ones 
is different from mainline and any other subtest in Valhalla. I do not yet 
understand exactly why, but I filed 
[JDK-8389125](https://bugs.openjdk.org/browse/JDK-8389125) to follow up.

Sigh, it seems [JDK-8388490](https://bugs.openjdk.org/browse/JDK-8388490) also 
triggers in Valhalla. In addition to the necessary test fixes, let's also 
problemlist `TestMaskedStoreIdealization.java`.

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

PR Comment: https://git.openjdk.org/valhalla/pull/2668#issuecomment-5071980342
PR Comment: https://git.openjdk.org/valhalla/pull/2668#issuecomment-5088967246
PR Comment: https://git.openjdk.org/valhalla/pull/2668#issuecomment-5089019724

Reply via email to