New issue
Advanced search Search tips

Issue 811513 link

Starred by 1 user

Issue metadata

Status: Available
Owner: ----
Cc:
Components:
EstimatedDays: ----
NextAction: ----
OS: ----
Pri: 2
Type: Bug

Blocking:
issue 809147



Sign in to add a comment

oop raster + swiftshader clip path visual bugs

Project Member Reported by enne@chromium.org, Feb 12 2018

Issue description

Run: chrome third_party/WebKit/LayoutTests/compositing/overflow/accelerated-scrolling-with-clip-path.html --enable-gpu-rasterization --use-gl=swiftshader --enable-oop-rasterization

See: some extra clip path garbage (at least on debug linux builds).

This only appears with *both* swiftshader and oop rasterization.  The clip path appears correctly when only one (or none) is turned on.

It doesn't appear to be due to serialization of paths.  Serializing and deserializing the SkPath in the ClipPathOp constructor and then running through the gpu rasterization path in the renderer does not reproduce the problem either.

It's interesting to me that it doesn't reproduce when using gpu rasterization, because that could indicate that it's maybe an angle or command buffer workaround that's preventing the bug for gpu rasterization, as oop rasterization does not include these workarounds.

However, running with --disable-gpu-driver-bug-workarounds + gpu raster + swiftshader doesn't reproduce the bug.  It's possible that maybe there's some angle workarounds that are the problem here.

I don't know for certain if this is a swiftshader bug or a missing workaround or something else at this point.
 
accelerated-scrolling-with-clip-path-actual.png
10.0 KB View Download
Components: Internals>GPU>Rasterization

Comment 2 by xing...@intel.com, Jun 4 2018

I test this on Ubuntu 16.0. I didn't see any clip garbage, but color differs.

OOPR_Only.png: Without --use-gl=swiftshader.
OOPR_sf.png: With --use-gl=swiftshader
OOPR_Only.png
17.7 KB View Download
OOPR_sf.png
17.4 KB View Download

Comment 3 by enne@chromium.org, Jun 4 2018

Cc: backer@chromium.org
I think something has changed here and I'm not sure if the same codepath is being hit.  Running this at ToT today hits the error that backer just added, in gpu_info_collector.cc:135 that oop raster isn't supported because the GrContext doesn't get created.

It seems like this was not the case a few months ago, and maybe Skia's GrContext creation restrictions have changed.

Comment 4 by capn@chromium.org, Jun 5 2018

Fixing  Issue skia:7846  most likely also fixed the artifacts on the edges here.

The darker color seems like an entirely different recent regression.

Sign in to add a comment