Issue metadata
Sign in to add a comment
|
Raster from animated toolbar on https://blog.google/topics/education/celebrating-our-teachers/ take 20ms |
||||||||||||||||||||||||
Issue descriptionGoogle Chrome 57.0.2987.146 (Official Build) (64-bit) Revision 0 Platform 9202.64.0 (Official Build) stable-channel samus What steps will reproduce the problem? (1) Load https://blog.google/topics/education/celebrating-our-teachers/ (2) Scroll up and down a little bit to see the toolbar animate. What is the expected result? Should raster fast, the page only has 252 elements. What happens instead? Raster takes a very long time. On my Pixel 2 the raster is taking 10ms eating up the entire frame. This is also with GPU Raster set to "Enabled", and about:gpu shows "Rasterization: Hardware accelerated", but the trace makes it look like we're CPU rastering.
,
May 10 2017
Note that the veto-to-software for concave paths was removed in 8e75e9c587155933fe3e775605e18111a6cc2b9c, and will ship in M60 if all goes well. Could this be the non-mobile content veto kicking in? That said, scrolling the first bit with the animated menus is equally janky on my Nexus 6P, with or without --force-gpu-rasterization.
,
May 10 2017
This is actually 20ms, it's 10ms when we're using 2 raster threads, but most Chromebooks these days only have one raster thread...
,
May 10 2017
On my Linux box, I see a lot of time during the janky scroll spent in GLImageMemory::CopyTexSubImage(). Not sure what's triggering that. (That's w/GPU rasterization off; will check w/on).
,
May 10 2017
With GPU raster on, the GPU process (PutChanged) on Linux is taking 14ms/frame during the janky part. DevTools is telling me there's a full-screen repaint (possibly also relayout?) on every frame going on during that part, so it's not too surprising to me that raster cost would be expensive. Not that we can't (always) do better at raster, but I wonder if the animation could be reworked to avoid doing full-screen repaints? Or maybe we could do better layerization? Slimming paint vNext?
,
May 11 2017
|
|||||||||||||||||||||||||
►
Sign in to add a comment |
|||||||||||||||||||||||||
Comment 1 by esprehn@chromium.org
, May 10 2017