After bisecting changes between Chrome 57.0.2949.0 and 57.0.2948.0 with ChromeOS R57-9079.0.0. The root cause is from this change: https://codereview.chromium.org/2493223002

chromium > git checkout 39fe426a5bec5352ddacab96ac10c6c33f9d5ac1
chromium > ninja -C out_${SDK_BOARD}/Release -j1000 chrome chrome_sandbox nacl_helper
chromium > deploy_chrome --build-dir=out_${SDK_BOARD}/Release --to=10.10.10.1
chromiumos > test_that minnie cheets_GFXBenchTest

This gets 1293.0

chromium > git checkout d8f4f642d72fc5fdb428f5a90ca207c2143a6f79
chromium > ninja -C out_${SDK_BOARD}/Release -j1000 chrome chrome_sandbox nacl_helper
chromium > deploy_chrome --build-dir=out_${SDK_BOARD}/Release --to=10.10.10.1
chromiumos > test_that minnie cheets_GFXBenchTest

This gets 1037.0

chromium> git log -2 d8f4f642d72fc5fdb428f5a90ca207c2143a6f79

commit d8f4f642d72fc5fdb428f5a90ca207c2143a6f79
Author: staraz <staraz@chromium.org>
Date:   Sat Dec 10 15:49:20 2016 -0800

    Moved exo::SurfaceFactoryOwner to its own file and renamed it to exo::CompositorFrameSink.
    
    CompositorFrameSink implements cc::mojom::MojoCompositorFrameSink.
    
    CompositorFrameSink is no longer a friend class of exo::Surface.
    
    Added exo::CompositorFrameSinkHolder class that implements cc::mojom::MojoCompositorFrameSinkClient.
    
    BUG= 659601 
    CQ_INCLUDE_TRYBOTS=master.tryserver.blink:linux_trusty_blink_rel
    
    Review-Url: https://codereview.chromium.org/2493223002
    Review-Url: https://codereview.chromium.org/2493223002
    Cr-Commit-Position: refs/heads/master@{#437780}

commit 39fe426a5bec5352ddacab96ac10c6c33f9d5ac1
Author: lpromero <lpromero@chromium.org>
Date:   Sat Dec 10 15:48:09 2016 -0800

    Handle storyboards in GN.
    
    BUG=none
    R=sdefresne@chromium.org,rohitrao@chromium.org
    
    Review-Url: https://codereview.chromium.org/2564023002
    Cr-Commit-Position: refs/heads/master@{#437779}

+staraz

Hi staraz, the mentioned code change seems to lower the score of GFXBench on several cheets platforms. Could you please verify if this could be caused by the change https://codereview.chromium.org/2493223002 ? Thanks !

I ran the benchmark the newest chrome revision (score: 5228, 5867) against 
39fe426a5bec5352ddacab96ac10c6c33f9d5ac1 (the revision before my exo refactor) 
(score: 4880, 5854) on my samus device and didn't see the regression.

3 CLs have landed to address some issues introduced in my refactor. They could 
also solve the performance regression. 

samus is x86, so is cyan. I'll find a minnie or other arm devices to make sure. 

Ok, let's keep an eye on these perf tests to make sure they recover.

The regression is still present on veyron-minnie (1287 vs 1040 old vs new)

The following revision refers to this bug:
  https://chromium.googlesource.com/chromium/src.git/+/c43626b432c288cc3b08e46b3a7f813e95d6d773

commit c43626b432c288cc3b08e46b3a7f813e95d6d773
Author: staraz <staraz@chromium.org>
Date: Fri Jan 13 23:05:01 2017

exo::CompositorFrameSink uses CompositorFrameSinkHolder* instead of mojo interface pointer

This fixes the performance regression on arm devices.

BUG= 680066 

Review-Url: https://codereview.chromium.org/2626293005
Cr-Commit-Position: refs/heads/master@{#443712}

[modify] https://crrev.com/c43626b432c288cc3b08e46b3a7f813e95d6d773/components/exo/buffer_unittest.cc
[modify] https://crrev.com/c43626b432c288cc3b08e46b3a7f813e95d6d773/components/exo/compositor_frame_sink.cc
[modify] https://crrev.com/c43626b432c288cc3b08e46b3a7f813e95d6d773/components/exo/compositor_frame_sink.h
[modify] https://crrev.com/c43626b432c288cc3b08e46b3a7f813e95d6d773/components/exo/compositor_frame_sink_holder.cc
[modify] https://crrev.com/c43626b432c288cc3b08e46b3a7f813e95d6d773/components/exo/compositor_frame_sink_holder.h
[modify] https://crrev.com/c43626b432c288cc3b08e46b3a7f813e95d6d773/components/exo/surface.cc

The graph shows that the performance score comes back. Thanks !