[lworld] RFR: 8369530: [lworld] Improve execution time of compiler tests
Damon Fenacci
dfenacci at openjdk.org
Wed Oct 29 12:12:59 UTC 2025
On Tue, 28 Oct 2025 14:52:40 GMT, Christian Hagedorn <chagedorn at openjdk.org> wrote:
>> # Issue
>>
>> Several Valhalla compiler specific tests are slow and this affects the overall execution time of lower tiers.
>> [JDK-8369437](https://bugs.openjdk.org/browse/JDK-8369437) has already improved the situation by splitting multiple @run statements in compiler tests into separate @test blocks for several tests. There are nevertheless still a handful of tests that take up to a dozen minutes to finish.
>>
>> # Cause
>>
>> Most tests affected are using multiple IR-Framework scenarios to start tests with different flags. The tests are started sequentially, adding the execution time of each test. The problem is similar to [JDK-8369437](https://bugs.openjdk.org/browse/JDK-8369437) but at IR-Framework level.
>>
>> # Fix
>>
>> We can parallelize the execution of scenarios similar to what has been done in the [fix](https://github.com/openjdk/valhalla/pull/1681/files) for [JDK-8369437](https://bugs.openjdk.org/browse/JDK-8369437). Unfortunately the IR-Framework doesn't allow tests to be run on different machines and there is no simple way to make it interact with JTReg in that way. So, as a provisional and minimalistic solution, scenarios are added and started one by one and run by individual JTReg `@test` entries.
>>
>> # Testing
>>
>> Tier 1-3
>> Worst-time compiler/valhalla JTReg tests went from 16min to <2min
>
> That's an impressive improvement! Thanks for having a closer look at what we could do. Have you also played around with using Java threads instead (i.e. a `TestFramework` instance per thread that adds the scenario flags)?
Thanks @chhagedorn @mhaessig @TobiHartmann for your reviews!
-------------
PR Comment: https://git.openjdk.org/valhalla/pull/1708#issuecomment-3461184391
More information about the valhalla-dev
mailing list