Screenshots by Chris Gannon (the reporter):

https://twitter.com/ChrisGannon/status/929289653726973952

And more info:

https://twitter.com/ChrisGannon/status/929303839706054658
https://twitter.com/ChrisGannon/status/930009317859831810

"Able to reproduce this issue on windows 10 and Linux Ubuntu 14.04 with Chrome reported version 62.0.3202.89, latest Canary 64.0.3267.0, latest beta 63.0.3239.40.
Note: The issue is not seen on Mac.
Manual Bisect:
-------------
Good build: 61.0.3142.0 -- Revision number(482491)
Bad build: 61.0.3143.0 -- Revision number(482834)

Bisect Tool Info:
----------------
You are probably looking for a change made after 482710 (known good), but no later than 482711 (first known bad).
CHANGELOG URL:

  https://chromium.googlesource.com/chromium/src/+log/4fb9b934954dc5483bbbf78801216599ea3b8df1..669931271c542198ad5e4db8f64d6d4d4e63de47

Possible suspect:
----------------
https://chromium-review.googlesource.com/548536

reed@,Kindly take a look and please help us to reassign this issue to a right owner if not with respect to this change.

Thanks.!"

What's the difference between SK_SUPPORT_LEGACY_COLORFILTER_FILTERSPAN and not?

SK_SUPPORT_LEGACY_COLORFILTER_FILTERSPAN controls whether the *software* rasterizer uses the SIMD pipeline for color filters or not (removing the flag activates the SIMD pipeline).

I don't see how this could affect GPU rasterization.  I'll re-run the bisect.

This is confusing because it conflates two separate issues.

1) distorted snail antennae geometry with GPU rasterization *enabled* (https://codepen.io/chrisgannon/pen/6c932fda8d8fca3e253b655033d86645)

-- regressed in https://chromium.googlesource.com/chromium/src/+log/0c534f203cbe5e049a3d533ceb9dc0d7f403c79a..3123fbf5a3fb31ec55e7076b6689159a2d72c2fe

(skia roll: https://skia.googlesource.com/skia.git/+log/72859cbb149b..267641a90cb1)

Most likely one of the GrTessellator changes.

-- fixed in M63 https://chromium.googlesource.com/chromium/src/+log/8ea0d42b631cfb977760b985d3935f5152897724..1d751dba764894204f2b07a6408619b53eb7a146
(skia roll: https://skia.googlesource.com/skia.git/+log/dd9468b9600c..eafa64b4018a)

Most likely fixed by brianosman's change.


2) distorted filter colors with GPU rasterization *disabled* (https://codepen.io/chrisgannon/pen/OReZrP , hardware-acceleration-svg-bug.zip, )

-- regressed in https://chromium-review.googlesource.com/548536


So #1 works correctly with software rasterization, and is fixed in M63.  #2 works correctly with GPU rasterization and is not fixed yet -- I'll minimize and open a Skia bug.

Deleted: filters_sw.png 6.6 KB

	filters_sw.png 6.6 KB View Download

Deleted: filters_gpu.png 8.1 KB

	filters_gpu.png 8.1 KB View Download

The distorted filter colors with GPU rasterization *disabled* seem to be fixed in 66.0.3341.0

This is just a guess, but from looking at the resulting colors is is very likely that during alpha-premultiplication alpha was not limited to 1.

adjusting priority for the potential gpu correctness issue remaining

Just checked against ToT: I believe at this point both issues are fixed for both CPU/GPU rasterization.

I can confirm that all issues are fixed on 71.0.3578.98 (64Bit) running on Windows 7.

The result of the feSpecularLighting filter is different on Firefox, but I assume this is a Firefox problem. In any case, this would be a different issue because it is not related to hardware acceleration.