The ZeroCopyRasterBuffer::Playback means that the webpage is drawing something (like, not just animating something -- drawing new content).

The way we draw stuff is by issuing commands that are consumed by the GPU process (so the GPU process isn't the root cause and can be ignored -- it's just following orders).

The root of the problem is "what is it that we're being asked to draw so often, which is taking so long to raster?"

Ooops, actually, ZeroCopyRasterBuffer means it's doing CPU raster.

So the GPU probably still doesn't matter because it's drawing what it rasterized.

And, to make things even more strange, the GPU process is spending lots of time doing TexImage2D (which is characteristic of one-copy, not zero-copy).

For the TexImage2Ds mentioned in #3, it looks like all these requests are coming from the Offscreen-MainThread context, which is used for Canvas and Video. I wasn't able to repro this pattern locally (although I did get overall slow raster), but my guess is that there was an ad on screen which was really hammering Canvas2D (on the order of 228 texture uploads a frame).

Other than throttling the Canvas for content we think is less important, not sure there's too much we can do about this issue. I'll continue looking into the long raster times, which may be unrelated.

Something on the page is vetoing us out of using Ganesh, and the software raster time for whatever was invalidating was high, plus as ericrk@ mentioned there was some heavy 2D Canvas work happening in an Ad.

senorblanco@, Ganesh is vetoed on this page.  I'd be curious what is causing the veto.

I don't see the Ganesh veto happening here on Mac or Windows -- GPU raster is on for me. Perhaps the page content changed? Or a particular ad had some complex path content?

At any rate, if the uploads are caused by canvas and video, Ganesh may not make a huge difference.

This issue has been Available for over a year. If it's no longer important or seems unlikely to be fixed, please consider closing it out. If it is important, please re-triage the issue.

Sorry for the inconvenience if the bug really should have been left as Available.

For more details visit https://www.chromium.org/issue-tracking/autotriage - Your friendly Sheriffbot

Since so much has changed since this was filed (e.g. the path veto no longer vetoes to software raster), I'm going to close this. If you can still repro, feel free to re-open.