The following revision refers to this bug:
  https://chromium.googlesource.com/chromium/src.git/+/a1035286c2bb86ae648e2d54c12f81033b6b1049

commit a1035286c2bb86ae648e2d54c12f81033b6b1049
Author: Dirk Pranke <dpranke@chromium.org>
Date: Tue Jun 06 17:24:59 2017

Markk css3/blending/svg-blend-multiply-alpha as failing.

I think this is probably failing due to a change in skia, but
I need to look into it further.

TBR=chrishtr@chromium.org
NOTRY=true
NOPRESUBMIT=true
NOTREECHECKS=true
BUG= 730071 

Change-Id: Ia7482709bce0949aba857a5bcaa87193e2bec21f
Reviewed-on: https://chromium-review.googlesource.com/526000
Reviewed-by: Dirk Pranke <dpranke@chromium.org>
Commit-Queue: Dirk Pranke <dpranke@chromium.org>
Cr-Commit-Position: refs/heads/master@{#477319}
[modify] https://crrev.com/a1035286c2bb86ae648e2d54c12f81033b6b1049/third_party/WebKit/LayoutTests/TestExpectations

Mike, Florin, is this reftest failure possibly due to your CL that forces all compose shaders through SkRasterPipelineBlitter (https://skia-review.googlesource.com/c/18529/)?

Here's the HTML... does this lead to a compose shader?

https://chromium.googlesource.com/chromium/src/+/7c35bee6935777ecc39bc419325703fb391d0159/third_party/WebKit/LayoutTests/css3/blending/svg-blend-multiply-alpha.html

I don't see a compose shader in the mix there, most likely just a savelayer.

The first failing build appears to be https://build.chromium.org/p/chromium.webkit/builders/WebKit%20Mac10.10/builds/34479, which contains this Skia roll:

https://skia.googlesource.com/skia.git/+log/0e022297fee8..adff9dc98200

Do you think your SkJumper change might have something to do with this?  It's also possible it's not Skia-related...

Nah, the SkJumper change is a behavior no-op.  It just makes profilers work better.

So there's no reason to suspect we're using SkRasterPipelineBlitter at all here?

Attaching the SKP dump, since I can't find a way to share from debugger.skia.org.

Just saveLayer/drawRects in there.  The debugger apparently doesn't print the blend mode, so not 100% sure (since old debugger is borked), but I'm betting on saveLayer(kMultiply).

Would that hit SkRasterPipelineBlitter?

Deleted: layer_0.skp 773 bytes

layer_0.skp 773 bytes Download

I don't think so, though maybe it'd hit the new sprite blitter?

restore from a layer should definitely hit the sprite blits

Ah, that explains things.  We're hitting the lowp stages on Mac but not Linux.  We disable SkCpu runtime detection on both, but on Mac we've got SSSE3 at compile time, where on Linux it's just SSE2.  Having SSSE3 is just enough to trigger the lowp paths.

Once we added multiply in "more lowp blend modes", that sprite blit would start working in lowp.

Looks like now that we've rolled there are nine non-ref tests that'll want rebaselining on all the Mac bots.  Do I do that like this?

 crbug.com/730071  [Mac] fast/borders/border-radius-mask-canvas-with-shadow.html [NeedsRebaseline]
...

?

Yeah, that should work.  You could also do a manual rebaseline with rebaseline-cl.

The following revision refers to this bug:
  https://chromium.googlesource.com/chromium/src.git/+/22af497bd72d37d4da5c7e3398af8a8bf0211cc9

commit 22af497bd72d37d4da5c7e3398af8a8bf0211cc9
Author: reed <reed@google.com>
Date: Wed Jun 07 15:47:24 2017

rebaseline images after spriteblitter optimizations

relevant skia changes have already landed...

BUG= 730071 

Review-Url: https://codereview.chromium.org/2924983004
Cr-Commit-Position: refs/heads/master@{#477660}

[modify] https://crrev.com/22af497bd72d37d4da5c7e3398af8a8bf0211cc9/third_party/WebKit/LayoutTests/TestExpectations

The following revision refers to this bug:
  https://chromium.googlesource.com/chromium/src.git/+/615a5e37d56eec7fc90a93f7ce650c5c345ab167

commit 615a5e37d56eec7fc90a93f7ce650c5c345ab167
Author: fmalita <fmalita@chromium.org>
Date: Wed Jun 07 18:46:25 2017

Convert css3/blending/svg-blend-multiply-alpha to a pixel test

1) we're now rendering differently on Mac

2) the ref has hard-coded colors, so there's little/no value over
   a pixel test

BUG= 730071 
TBR=mtklein@chromium.org,dpranke@chromium.org

Review-Url: https://codereview.chromium.org/2926873002
Cr-Commit-Position: refs/heads/master@{#477712}

[modify] https://crrev.com/615a5e37d56eec7fc90a93f7ce650c5c345ab167/third_party/WebKit/LayoutTests/TestExpectations
[delete] https://crrev.com/bd06835c3db030fb9ffeff153df5d75006192d10/third_party/WebKit/LayoutTests/css3/blending/svg-blend-multiply-alpha-expected.html

I forget what this is but it looks like it's fixed.