All graphs for this bug:
  https://chromeperf.appspot.com/group_report?bug_id=793466

(For debugging:) Original alerts at time of bug-filing:
  https://chromeperf.appspot.com/group_report?sid=6f1997cef3bfdc84ff070730baf0518b9c56ba174a8b207ab7e396e13a527acc


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

chromium-rel-mac-retina
chromium-rel-mac11-pro

📍 Pinpoint job started.
https://pinpoint-dot-chromeperf.appspot.com/job/12f77d34040000

📍 Couldn't reproduce a difference.
https://pinpoint-dot-chromeperf.appspot.com/job/12f77d34040000

📍 Pinpoint job started.
https://pinpoint-dot-chromeperf.appspot.com/job/12d5ae82040000

📍 Found a significant difference after 1 commit.
https://pinpoint-dot-chromeperf.appspot.com/job/12d5ae82040000

gpu: Don't clear anonymous IO surfaces.
By sunnyps@chromium.org · Thu Dec 07 03:04:36 2017
chromium @ ea6e12b3635b8f77966cd4bacc02c31b914e95c4

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

I couldn't reproduce the regression locally on my macbook pro mid 2014 with nvidia gt 750m. In fact, I saw a performance improvement with my CL over two runs.

I ran the benchmark like this:
out/Release/performance_browser_tests --gtest_filter="CastV2PerformanceTest.Performance/2" --enable-gpu

This corresponds to CastV2Performance_gpu_60fps.

GN args:
is_debug = false
is_component_build = true
use_goma = true
enable_nacl = false
proprietary_codecs = true
ffmpeg_branding = "Chrome"

See attached files for benchmark output. To clear IO surface I set should_clear to true in GpuMemoryBufferFactoryIOSurface::CreateGpuMemoryBuffer and *is_cleared to true in GpuMemoryBufferFactoryIOSurface::CreateAnonymousImage. Doing this effectively reverts the patch.

Deleted: ToT.txt 4.2 KB

ToT.txt 4.2 KB View Download

Deleted: clear io surface.txt 4.2 KB

clear io surface.txt 4.2 KB View Download

+miu, owner of performance_browser_tests, any ideas on #6?

Looks like both the machines this regressed on are mac 10.11, is that the OS version on the macbook pro you tried to repro on?

Given that:

1. All graphs have since improved a ton (better than pre-regression point).

2. It only affects an older verion of Mac OS.

3. It's faster on other machines (per sunnyps's comment, #c6)

4. I just landed a change that replaces the whole tab capture implementation. :)

...I feel comfortable resolving this as a WontFix.