[bisected] 

b642ce1b14a2b5d9ef78fd1b9ae2b5d6ae1c9cb9 is the first bad commit
commit b642ce1b14a2b5d9ef78fd1b9ae2b5d6ae1c9cb9
Author: Kenneth Russell <kbr@chromium.org>
Date:   Thu Jul 13 05:33:44 2017 +0000

    Support LUMINANCE, ALPHA, and LUMINANCE_ALPHA in CopyTextureCHROMIUM.
    
    Use the luminance/alpha emulation blitter for CopyTex{Sub}Image2D on
    the Core Profile to support these.
    
    Simplified the copy shader and fixed bugs in handling of premul/unmul
    which are tested by the new tests in
    https://github.com/KhronosGroup/WebGL/pull/2451 .
    
    BUG= 733599 
    
    Change-Id: Ic52b318d917dd95b8eb2c8fa3ce7b2aaebc24a63
    Reviewed-on: https://chromium-review.googlesource.com/567306
    Commit-Queue: Kenneth Russell <kbr@chromium.org>
    Reviewed-by: Zhenyao Mo <zmo@chromium.org>
    Cr-Commit-Position: refs/heads/master@{#486281}

The time_geom_mean metric of graphics_WebGLPerformance regressed by following amounts:

cros-bannon: 14.1%
cros-celes: 12.8%
cros-cyan: 14.4%
cros-nyan-blaze: 19.5%
cros-elm: 15.5%
cros-lars: 16.0%
 
kbr@, zmo@ -- can you investigate this regression and try to improve the performance of this CL?  To build ChomeOS, please follow this guide:

http://www.chromium.org/chromium-os/developer-guide

After you've built you can run this test with the following command:

test_that -b ${BOARD} $IP graphics_WebGLPerformance

It will report the time_geom_mean number after a test run.  Let me know if you have any questions.

Could we please get some help from the ChromeOS team on this? This CL was intended to fix support for these texture formats on macOS, and I'm surprised it had an effect on ChromeOS.

Could we set up a time where I could come over and sit down with one of the developers who already has a build set up on the affected hardware?

Sure, you can come over and we can sit and debug this if you want. I'll be around all day.

marcheu@ took the time to sit down and walk me through these tests (thanks!). Their source code, FYI, is here:

https://chromium.googlesource.com/chromiumos/third_party/autotest/+/master/client/deps/webgl_perf/

They're a fairly old but still useful set of microbenchmarks.

In sum: the following four tests were previously silently failing with errors:

convert-Canvas-to-a8.html
convert-Canvas-to-a8-premultiplied.html
convert-Canvas-to-l8.html
convert-Canvas-to-l8-premultiplied.html

and with the fix for  Issue 733599 , they are now running, and producing valid scores which are factored into the geomean for the entire benchmark.

Therefore this regression is unfortunately actually expected, and is a positive improvement. Closing as WontFix.