RFR: 8290531: Loom: Parallelize a few tests more deeply
Alan Bateman
alanb at openjdk.org
Mon Aug 1 07:47:37 UTC 2022
On Tue, 19 Jul 2022 12:52:11 GMT, Aleksey Shipilev <shade at openjdk.org> wrote:
> Some of the newly added Loom tests are long-running, and thus they delay the completion of otherwise very parallel tier1/jdk_loom. We can parallelize them a bit better to avoid these testing stalls.
>
> Improvements on Linux x86_64, TR 3970X, `jdk_loom hotspot_loom`:
>
>
> # release before
> real 4m41.424s
> user 24m18.064s
> sys 1m16.440s
>
> # release after
> real 2m47.769s ; -40%
> user 23m44.622s
> sys 1m15.240s
>
>
> # fastdebug before
> real 5m38.078s
> user 67m23.516s
> sys 1m56.446s
>
> # fastdebug after
> real 4m9.249s ; -26%
> user 67m21.940s
> sys 1m57.625s
No objection to changing these tests. We should probably do test/jdk/java/net/vthread/BlockingSocketOps.java at the same time as it also runs the sockets tests with the Poller in both configurations.
For the test ids then I'd prefer not use two styles (id=useDirectRegister and id=no-continuations). Can you change these to `indirect-register` and `no-vmcontinuations`?
I'd prefer if the `@bug` tag is not copied, only because we add to the bug list when extending tests.
-------------
PR: https://git.openjdk.org/jdk/pull/9554
More information about the nio-dev
mailing list