[webkit-reviews] review denied: [Bug 65622] [Chromium] Canvas 2D getImageData method is slow with GPU acceleration enabled : [Attachment 102792] Patch

[bugzilla-daemon at webkit.org]

James Robinson <jamesr at chromium.org> has denied Justin Novosad
<junov at chromium.org>'s request for review:
Bug 65622: [Chromium] Canvas 2D getImageData method is slow with GPU
acceleration enabled
https://bugs.webkit.org/show_bug.cgi?id=65622

Attachment 102792: Patch
https://bugs.webkit.org/attachment.cgi?id=102792&action=review

------- Additional Comments from James Robinson <jamesr at chromium.org>
If we're going to switch from h/w to software, we should switch back all the
way to software rendering (including removing compositing layers).  With this
patch, a switch back to s/w will make us software render, upload to a texture,
and GPU composite for not much benefit.  With that in mind I think it makes a
lot more sense to do this sort of manipulation up at or closer to the
CanvasRenderingContext2D level.

I also wonder if we can't gather a bit more data.  We have the ability to do
histograms in WebKit code, maybe that would be useful?	I think it'd be really
useful to know how good a proxy having done one readback is for throwing the
context into s/w permanently.  A few things that might be useful to measure:
- for contexts that do at least one readback, how many more readbacks do they
do?
- what's the distribution look like for the ratio of readbacks to
canvas->canvas draws or canvas composites?

R- because this will lead in lots of unnecessary compositing layers being left
around for software-rendered canvases, but please consider the other feedback
as well.