REGRESSION(r170529): [GTK] Performance test Animation/balls:FrameRate got ~10% worse

[Carlos Alberto Lopez Perez]

On Platform GTK the revision r170529 <http://trac.webkit.org/r170529> caused a performance regression around ~10% on the test Animation/balls.

You can check the performance graph at https://perf.webkit.org/#mode=charts&chartList=%5B%5B%22gtk%22%2C%22Animation%2Fballs%3AFrameRate%22%5D%5D&days=44 (we identified r170529 as the commit causing the regression on the interval r170529-r170540 after a manual bisect)

The overall performance is still very poor in this test: Around 5 FPS. Comparing with the other ports: EFL gets around ~10FPS, and MAC ports around ~30FPS.

Zan Dobersek (CC'ed) identified an issue with the GTK port and this test: DrawingArea is not entering the accelerated compositing codepaths. He tried to force-enable the accelerated compositing path, and the animations in the benchmark became much more fluent, but the test is still only reporting ~10FPS.

Then he tried to revert r170529 and the FPS jumped to 20FPS.

So, it seems that r170529 is causing a significative performance regression on this test.

You can run the test as follows:

Tools/Scripts/run-perf-tests --release -2 Animation/balls.html

If you want to see the animation:

USE_NATIVE_XDISPLAY=1 Tools/Scripts/run-perf-tests --release -2 Animation/balls.html

[Carlos Alberto Lopez Perez]

BTW, the test is self-contained. You can run it in any browser by opening: http://trac.webkit.org/export/171143/trunk/PerformanceTests/Animation/balls.html or file://${path-to-webkit}/PerformanceTests/Animation/balls.html

[Andre Moreira Magalhaes]

See https://bugs.webkit.org/show_bug.cgi?id=134109#c8.

[Carlos Alberto Lopez Perez]

Created attachment 235283 [details]
Patch

[WebKit Commit Bot]

Comment on attachment 235283 [details]
Patch

Clearing flags on attachment: 235283

Committed r171343: <http://trac.webkit.org/changeset/171343>

[WebKit Commit Bot]

All reviewed patches have been landed.  Closing bug.