RFR: 8287453: RenderPerfTest incorrectly measures performance
Alexey Ushakov
avu at openjdk.java.net
Wed Jun 1 09:48:33 UTC 2022
On Wed, 1 Jun 2022 06:46:40 GMT, Ajit Ghaisas <aghaisas at openjdk.org> wrote:
> > > Is it only a test stabilization fix? or were you able to uncover any performance issue (either with OpenGL or with Metal pipeline) with this change? Can you please provide before/after numbers?
> >
> >
> > Initially it was a stabilization fix. However, I've found couple of issues: missing frames (I have minimal test case and here is the issue [JDK-8287600](https://bugs.openjdk.java.net/browse/JDK-8287600) ) and low performance in some test cases comparing with OGL. I've attached the results to this issue.
>
> Thanks for providing the performance numbers in JBS and filing the issue for missing frame.
>
> Another bug needs to be filed for tests having lower performance with the Metal pipeline as compared to OpenGL pipeline. It's a surprise that the modified RenderPerf test shows considerable performance difference (whereas J2DBench tests show difference only in some cases)
Yes, it was a surprise. We've enabled metal in EAP of our IDEs and got some complains about the performance. So, I've decided to revise our measurements. I have fixes for the problems above, but they are not complete yet. (still fighting with some regressions)
-------------
PR: https://git.openjdk.java.net/jdk/pull/8928
More information about the client-libs-dev
mailing list