[lworld] RFR: 8367245: [lworld] C2 compilation fails with "Missed optimization opportunity in PhaseIterGVN"

[Christian Hagedorn]

On Wed, 12 Nov 2025 17:14:36 GMT, Marc Chevalier <mchevalier at openjdk.org> wrote:

> Let's just disable the verification for now. The issue will be fixed in mainline, then reach Valhalla.
> 
> I've changed the 3 files in which I've observed some clashing speculative types at least once.
> 
> This will be reverted by https://bugs.openjdk.org/browse/JDK-8371726 once https://bugs.openjdk.org/browse/JDK-8371716 is solved.
> 
> Thanks,
> Marc

Marked as reviewed by chagedorn (Committer).

test/hotspot/jtreg/compiler/valhalla/inlinetypes/TestArrays.java line 141:

> 139:  *          java.base/jdk.internal.vm.annotation
> 140:  * @requires (os.simpleArch == "x64" | os.simpleArch == "aarch64")
> 141:  * @run main/othervm -XX:+IgnoreUnrecognizedVMOptions -XX:VerifyIterativeGVN=000 compiler.valhalla.inlinetypes.TestArrays 6

That looks reasonable to do.

If we are only seeing the failures in the test VMs, then an easier solution would be to directly add the flag to the framework:

 InlineTypes.getFramework()
                   .addScenarios(scenarios[Integer.parseInt(args[0])])
                   .addFlags("-XX:+IgnoreUnrecognizedVMOptions -XX:VerifyIterativeGVN=000")
                   .addHelperClasses(MyValue1.class, MyValue2.class, MyValue2Inline.class)
                   .start();

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

PR Review: https://git.openjdk.org/valhalla/pull/1732#pullrequestreview-3458204421
PR Review Comment: https://git.openjdk.org/valhalla/pull/1732#discussion_r2522160881