All graphs for this bug:
  https://chromeperf.appspot.com/group_report?bug_id=905495

(For debugging:) Original alerts at time of bug-filing:
  https://chromeperf.appspot.com/group_report?sid=952a647653ffd6ddfdeb28a730275b7da88ec6a8ae952002cfa232b061c791ce


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

linux-perf

media.desktop - Benchmark documentation link:
  None

📍 Pinpoint job started.
https://pinpoint-dot-chromeperf.appspot.com/job/17e5c5d3e40000

📍 Found a significant difference after 1 commit.
https://pinpoint-dot-chromeperf.appspot.com/job/17e5c5d3e40000

Use SharedImageInterface for one-copy staging buffers by piman@chromium.org
https://chromium.googlesource.com/chromium/src/+/9058f7866e74b92eeef1c30780522fba0740d2f3
memory:chrome:all_processes:reported_by_chrome:gpu:effective_size: 1.029e+07 → 1.051e+07 (+2.228e+05)

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

Benchmark documentation link:
  None

 Issue 905447  has been merged into this issue.

Issue 905482 has been merged into this issue.

Issue 905483 has been merged into this issue.

Issue 905484 has been merged into this issue.

Issue 905485 has been merged into this issue.

Issue 905486 has been merged into this issue.

 Issue 905488  has been merged into this issue.

 Issue 905489  has been merged into this issue.

 Issue 905490  has been merged into this issue.

 Issue 905493  has been merged into this issue.

 Issue 905492  has been merged into this issue.

 Issue 905494  has been merged into this issue.

Issue 905748 has been merged into this issue.

It's an accounting issue, as the GPU size is the same (or extremely close), but the staging buffer memory gets counted to the GPU process instead of the renderer.

https://chromium-review.googlesource.com/c/chromium/src/+/1347355/ should do it.

The following revision refers to this bug:
  https://chromium.googlesource.com/chromium/src.git/+/337f41126e2c9d833e259c73119fb8a4d02ea8e8

commit 337f41126e2c9d833e259c73119fb8a4d02ea8e8
Author: Antoine Labour <piman@chromium.org>
Date: Tue Nov 27 21:06:50 2018

Flush after destroying staging buffers

This is to ensure the memory is properly released on the GPU process side.

Bug:  905495 
Change-Id: I152494878a602e748eefb29f0f91f364e6752e42
Reviewed-on: https://chromium-review.googlesource.com/c/1347355
Commit-Queue: Jonathan Backer <backer@chromium.org>
Reviewed-by: Jonathan Backer <backer@chromium.org>
Cr-Commit-Position: refs/heads/master@{#611318}
[modify] https://crrev.com/337f41126e2c9d833e259c73119fb8a4d02ea8e8/cc/raster/staging_buffer_pool.cc

Several graphs have recovered but not all, so I'll poke at this more.

With backer@'s fixes for  bug 911410  (thanks!), all graphs have recovered now.