[jdk20] RFR: 8298217: Regressions 30-110% in SwingMark on MacOS, more so on aarch64

[Alexey Ushakov]

On Fri, 16 Dec 2022 15:20:14 GMT, Ajit Ghaisas <aghaisas at openjdk.org> wrote:

> The nature of data plays an important part in rendering performance as noted in the comment made on [JDK-8288948](https://bugs.openjdk.org/browse/JDK-8288948) -
> https://bugs.openjdk.org/browse/JDK-8288948?focusedCommentId=14504772&page=com.atlassian.jira.plugin.system.issuetabpanels%3Acomment-tabpanel#comment-14504772
> 
> Although the fix [JDK-8288948](https://bugs.openjdk.org/browse/JDK-8288948) is logical, it seems to have an adverse impact on SwingMark test numbers especially on aarch64.
> 
> The SwingMark tests closely depict the UI operations in real-world applications as compared to the J2DBench and RenderPerfTest tests. Hence, it is decided to revert the regression causing commits [JDK-8288948](https://bugs.openjdk.org/browse/JDK-8288948) (and dependent [JDK-8291266](https://bugs.openjdk.org/browse/JDK-8291266)).
> 
> Metal rendering pipeline performance numbers before and after this PR are attached to the JBS.
> 
> Note : The fix for [JDK-8291266](https://bugs.openjdk.org/browse/JDK-8291266) has a good regression test - DrawPrimitivesTest.java. It passes even after reverting the said changes. This test will be added back once this PR is committed.

The rest of the code looks good

src/java.desktop/macosx/native/libawt_lwawt/java2d/metal/MTLRenderQueue.m line 2:

> 1: /*
> 2:  * Copyright (c) 2019, 2021, Oracle and/or its affiliates. All rights reserved.

I suppose we should leave 2022 here.

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

Marked as reviewed by avu (Committer).

PR: https://git.openjdk.org/jdk20/pull/47