All graphs for this bug:
  https://chromeperf.appspot.com/group_report?bug_id=841807

(For debugging:) Original alerts at time of bug-filing:
  https://chromeperf.appspot.com/group_report?sid=b29ef7f1478a5e919e15de7dd6ddb4dbabed0826cd7ee30a65c50990df7d1952


Bot(s) for this bug's original alert(s):

chromium-rel-mac11-air
chromium-rel-mac11-pro

📍 Pinpoint job started.
https://pinpoint-dot-chromeperf.appspot.com/job/16c80f6fc40000

📍 Found significant differences after each of 2 commits.
https://pinpoint-dot-chromeperf.appspot.com/job/16c80f6fc40000

Switch to using skcms for ICC profile parsing by brianosman@google.com
https://chromium.googlesource.com/chromium/src/+/e6ae8f4c3bf66e5bf6641711062c15ea61df9cd2

[autofill] Suppress the use of invalid profile data. by rogerm@chromium.org
https://chromium.googlesource.com/chromium/src/+/c8d6aa1656c546f1bbef7273af2662d611bcd68c

Understanding performance regressions:
  http://g.co/ChromePerformanceRegressions

We think this is the "Switch to using skcms... " CL.

That's unexpected.  I'm looking into this now.

We weren't able to reproduce this on my Mac, so we're going to revert that CL and investigate offline.  We think it's got something to do with the Mac-specific ICC profile handling used to skip a transform, but we're not quite sure we understand that enough to roll forward.

I've created the revert at https://chromium-review.googlesource.com/c/chromium/src/+/1054509, but I don't have committer status, so I can't actually +2 it. If anyone else wants to land the revert, that would be great.

The following revision refers to this bug:
  https://chromium.googlesource.com/chromium/src.git/+/6c332b23418691dc7d4a8cc09ea95eeb6274e748

commit 6c332b23418691dc7d4a8cc09ea95eeb6274e748
Author: Brian Osman <brianosman@google.com>
Date: Fri May 11 14:22:21 2018

Revert "Switch to using skcms for ICC profile parsing"

This reverts commit e6ae8f4c3bf66e5bf6641711062c15ea61df9cd2.

Reason for revert: Perf regression

Bug:  chromium:841807 

Original change's description:
> Switch to using skcms for ICC profile parsing
> 
> Change-Id: I8bafe64117d7247b7806e1d64d6e7e793078877c
> Reviewed-on: https://chromium-review.googlesource.com/1028452
> Reviewed-by: Brian White <bcwhite@chromium.org>
> Reviewed-by: ccameron <ccameron@chromium.org>
> Reviewed-by: Mike Klein <mtklein@chromium.org>
> Commit-Queue: Mike Klein <mtklein@chromium.org>
> Cr-Commit-Position: refs/heads/master@{#556861}

TBR=ccameron@chromium.org,bcwhite@chromium.org,mtklein@google.com,brianosman@google.com,mtklein@chromium.org,bcwhite@google.com

# Not skipping CQ checks because original CL landed > 1 day ago.

Change-Id: I2cc981f9fcce42e94a0cc74240d51fc7dbb0f1db
Reviewed-on: https://chromium-review.googlesource.com/1054509
Commit-Queue: Mike Klein <mtklein@chromium.org>
Reviewed-by: Mike Klein <mtklein@chromium.org>
Cr-Commit-Position: refs/heads/master@{#557869}
[modify] https://crrev.com/6c332b23418691dc7d4a8cc09ea95eeb6274e748/skia/BUILD.gn
[modify] https://crrev.com/6c332b23418691dc7d4a8cc09ea95eeb6274e748/tools/metrics/histograms/enums.xml
[modify] https://crrev.com/6c332b23418691dc7d4a8cc09ea95eeb6274e748/tools/metrics/histograms/histograms.xml
[modify] https://crrev.com/6c332b23418691dc7d4a8cc09ea95eeb6274e748/ui/gfx/BUILD.gn
[modify] https://crrev.com/6c332b23418691dc7d4a8cc09ea95eeb6274e748/ui/gfx/color_space_unittest.cc
[modify] https://crrev.com/6c332b23418691dc7d4a8cc09ea95eeb6274e748/ui/gfx/color_transform_unittest.cc
[modify] https://crrev.com/6c332b23418691dc7d4a8cc09ea95eeb6274e748/ui/gfx/icc_profile.cc
[modify] https://crrev.com/6c332b23418691dc7d4a8cc09ea95eeb6274e748/ui/gfx/icc_profile.h
[modify] https://crrev.com/6c332b23418691dc7d4a8cc09ea95eeb6274e748/ui/gfx/icc_profile_unittest.cc
[modify] https://crrev.com/6c332b23418691dc7d4a8cc09ea95eeb6274e748/ui/gfx/skia_color_space_util.cc
[modify] https://crrev.com/6c332b23418691dc7d4a8cc09ea95eeb6274e748/ui/gfx/skia_color_space_util.h

We've noticed since last week that these benchmarks are running with only GPU compositing, no Canvas2D acceleration.  Presumably this is some sort of OS-too-old veto for 10.11?

Even still, comparing with --disable-accelerated-2d-canvas isn't helping us reproduce the regression.  We're now seeing frame times that look plausible (versus the 60fps when running with GPU acceleration), but not able to see any significant difference between this CL and the one before it, even running on the same hardware as one of the two regressed bots, a MacBookPro 11,5.

Our inclination here is to ignore these software-only Mac10.11 bots and reland.  Are we bad people?

 Issue 845917  has been merged into this issue.

Just to confirm, the CL was relanded in https://chromium-review.googlesource.com/c/chromium/src/+/1064172


I am a bit worried that this is regressing the power 


> Even still, comparing with --disable-accelerated-2d-canvas isn't helping us reproduce the regression.  We're now seeing frame times that look plausible (versus the 60fps when running with GPU acceleration), but not able to see any significant difference between this CL and the one before it, even running on the same hardware as one of the two regressed bots, a MacBookPro 11,5.

One of the largest regressions (40%) is in frame_times/tough_canvas_cases_canvas-animation-no-clear.html [1]. Trying that one may help see a difference.

You can also take a look at the trace that was recorded on the bot after [2] and before [3] the regression. Looking at those I see an increase in main thread frames going from ~90ms to ~115ms.
Also interestingly previously the main thread was doing a much larger paint (~20ms) vs (~1ms) after. 

I am not sure how to interpret these. But it seems the before the change we where doing more
software painting but we have now moved these off main-threads which non-intuitively is producing
a worse outcome.

Also, you can ask the bot to re-run the particular build with custom tracing flags. See "Request Debug Trace" in the dashboard.

> Our inclination here is to ignore these software-only Mac10.11 bots and reland.  Are we bad people?

I will leave it up to benchmark owner's junov@ to determine if and how much of a user impact
this is. 


[1] https://chromeperf.appspot.com/group_report?sid=732254d3aa3efd2abcc685f9e37ffa2b56722ca418f88f26f07084dfcc901a49


[2] https://00e9e64bac50ef3272fad7b2d748b256e8f952f55502bcf4ff-apidata.googleusercontent.com/download/storage/v1/b/chrome-telemetry-output/o/tough_canvas_cases_canvas_animation_no_clear_html_2018-05-08_14-19-53_21149.html?qk=AD5uMEv9weLQLDhaI2H1cy-HwWWZ-vLi2dLj-23_aSPT4D3BXNJ53s8PPSU8JBSMvD8nizA47DNFtiEP7VkxpW3sMHRUWNTRqY_xM34mSHKl7B16R5V710rpgHfsxaa1cFBVoLPrL5oIWD-Wu5rAdAtrygmVwK8aJeBY7gIKn5lUjzPU5gaaG_a5OYJxY4QgCnQVTG-oqgTLEfwjpD-3C1zyouzJYWaFVV5vH9Fo_tvXSIGTknAEOF2rcKaVd8QmFfkjRjFwcXKMQvHPabVqm94RyWYSo80xxbXB4wcqkJSIvzmMdCzumvbV1u3btQTkuyBzZZHHdmDHS6sQ5yESKGK9QZTLXq4YyBDeNW6kCAcNYyy8fphd_7AzwN-6YfWJeDSWkvY3bK61kF--3tQY_7SyxkmYDZTjISyGI9J-PYbblo57FZrGe6UoTU9M5znQMqhJlFrFjyibojo4GaQU9bMyPS-d9uTEIfzGeKoyC0UAPFo8zOL22X-g0icve5t-jbU0rjg314MrZwEP3nLXEGVJYgh8ybqVyrdbB8faJ4VOrkcKGh2qKXQXqLNoN4nC5weuYwtdY9GQsbhxDe2wggUN6gm_HbZhJyFK65Mly9JSkOXl_vcJLpnhjI20Cy-EXuXaYX-7Hrs0RtSP2-r8oBF45sEVeF2I_t2BJLFv4cIVsCJ_d_-YUlp5q6BcnofB3P7wyKYgIjdXt7x_1dFNmMLgFFY1Pn4FrooLdYopNMpctGzmM8K817-RJ-Jx8dSswa_W8CC965uPoOwkLFfBVxnmGSOF5iTz3SXZgFQM8l193jLUwAQzho9K-eXgXlKogdG_Nqv-iWTsDIu5f31fcWICy9AFCDq_9Q

[3] https://00e9e64bac77ce52c96c9f4fd96be9ac24e405de56d8d4dc44-apidata.googleusercontent.com/download/storage/v1/b/chrome-telemetry-output/o/tough_canvas_cases_canvas_animation_no_clear_html_2018-05-08_06-56-11_20223.html?qk=AD5uMEvitl8PkNS7ims4VvZY9_TH8fEzXBaklc9bDfSnqFsA3t0tOYBZL3bi06X7ziWZkwAiEwSH-73nUuvAlUM1l23GDHZ7_mFmbEhP8AGD4BV4aGkbCxObVn6sPNIeGs9ysB5nm_q0kwg_3fpf4i-JLpNvwgYR4XqaEK4Y5Hx63V1TdvX3u3dtPuog4Vu4F9Ga6tZrv0QMUdPRMcC2_ftmVpFmlf_cS3RCgRpZF52ABRqnoWnGWAo-I7SsYJQ4iaoayAj2RNU0ZvS91V3IdA_MvbFPHhPiZRmcBA9pWQu6jxxKGjFu-agpm612fjHqQllQJsvt6K1h6szJyCz3M1F6pfaUzApBxzWY5L2n3uSzLkfPsFBzyM0BBIkcoUmZP5a8JB4gxHgLHO96HQc5pzAeI-gKiFY-fAyVErea0XkRfrweJuRXJ1Ri-621o1LxO3MnldGwXMMa68GbGW6yzBZy6XRvx_6C23f5D8-Z67oZ6J_ooethGoL1AaAttj4PJ2g4bMGJjgDx35J04IhxrR8juRHeEJ4ZhKE9Uvj6_TeO9Gq_crwwyBN_hkQ1bRMyAtJcuxyn_Kcy-lwmPlKKM7v9NwbNXWMa3iUEJYdmUS3YbfWq3adAxjBBTJweOHaw7woR2d80nY91h-jvXLbgUgUTQzbTbmknbdFUJaEYl--qZkrD59QSqgjmlsP7OQRfBk-OHsQisECSkoJh4KETx-5M9M8L1XVQMAttDKivTPrkyrQI9OFce1g7QuMmPrqJ0REvRaVCyIp20p_L7VDEgCQ2PYIni941ZvpdVxcHOZjyoK9eHYzLVMMGpgFhV1lwyVXsTPxZgYRDmui0Xd_wX3ndec-CjahxjQ

> One of the largest regressions (40%) is in frame_times/tough_canvas_cases_canvas-animation-no-clear.html [1]. Trying that one may help see a difference.

Yes, this is the case we used to try to reproduce the problem.

> You can also take a look at the trace...

I don't think the traces go interestingly deep enough for us to make any use of.  If we can reproduce this locally, it'd be fairly easy to profile in perf/Instruments.  We just can't repro.