Sampling profile from z840.

Complexity set to 100. Profiling build with extensions disabled.

https://drive.google.com/file/d/0B9rpPoIDv3vTbmhLUDQ2ejk2RGs/view?usp=sharing

Complexity 1000 with img { will-change: transform; }

Deleted: trace.json.gz 1012 KB

trace.json.gz 1012 KB Download

Complexity 1000, without will-change.

Deleted: trace.json.gz 1.8 MB

trace.json.gz 1.8 MB Download

Looking at the traces that doug attached, a lot of time is being spent in rastering

39.27 SkRasterClip::setPath
>>>27.87 SuperBlitter::flush
28.51 SkAAClip::op

so we tried it with will-change to turn off raster. Visually, this improves the framerate dramatically (from around 2 to something much smoother). In this case, it spends 65.43% in SkRasterClip::op. It also then spends 7.37% in blink::FrameView::synchronizedPaint, 5.65 in blink::FrameView::updateStyleAndLayoutIfNeededRecursive, 2.52 in blink::FrameView::invalidateTreeIfNeededRecursive




Out of interest, I also tried running the basic benchmark on "Ramp" on various browsers on my MacBook Air (13-inch, Mid 2012)

Chrome canary (52.0.27180.0): 23.77, 20.53, 17.35, 22.18, 16.55 (avg: 20.08)
Firefox (45.0): 140.13, 179.0, 140.79, 171.33, 111.77 (avg: 148.60)
Safari: 13.45, 11.43, 11.72, 8.78, 11.26 (avg: 11.33)

I then checked out a copy of WebKit and modified the benchmark to use will-change: transform on img as above, and the scores were

Chrome canary: 393.60, 401.56, 363.93, 426.39. 383.22 (avg: 393.74)
Firefox: 155.35, 174.75, 168.39, 175.67, 153.08 (avg: 165.45)
Safari: 451.14, 515.60, 432.27, 476.33, 535.57 (avg: 482.18)

So all browsers show an improvement with will-change, but Safari manages to come out on top.

FYI, as far as we can see, one of the main aims of these Animometer benchmarks
is to exercise the raster system, so comment 6 is somewhat beside the point.

Re data on rastering:

39.27 SkRasterClip::setPath
>>>27.87 SuperBlitter::flush
28.51 SkAAClip::op

Is there a Skia bug to be filed here? Did you try with Ganesh?

Do we have any idea why our rastering is so much slower here? (ignoring the speed up from will-change)

I've been busy, then away for the past 3 weeks, so I haven't had a chance to look further at this. Als I don't really know what I'm doing, and I'm about to go traveling again, so could you please find a more appropriate owner for this Dru?

assigning to Chris in case he sees any additional opportunities here, although I think we've largely reached the point of diminishing returns on these investigations. Correct me if I'm wrong Chris :)

Nthing more from me, though I have been slammed with other work recently and
have not dug in a lot of detail.

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

Time to retire this one.