GitLab

CPU stack sampling profiler jank analysis is hinting at these methods
is this a bias or is there really jank in here? Add trace events to
balance stack sampling with slow reports.

Added trace events to both MessagePumpForUI::ProcessNextWindowsMessage
and MessagePumpForUI::WaitForWork's lookups. The stack sampling
analysis seems to be hinting more at ProcessNextWindowsMessage for
now, but a lot of this gets inlined in Release builds so it's hard to
tell for sure.

This CL also highlights a discrepancy between the checks in
ProcessNextWindowsMessage and WaitForWork. WaitForWork uses
MSG msg = {0} whereas ProcessNextWindowsMessage uses an uninitialized
MSG object. It doesn't seem necessary for this structure to be
initialized, will make it consistent in a follow-up CL.

R=fdoray@chromium.org

Bug: 899897
Change-Id: Ibd8a43e0346e6eb7e09432582a76235ed350ae6f
Reviewed-on: https://chromium-review.googlesource.com/c/chromium/src/+/2153321
Commit-Queue: Gabriel Charette <gab@chromium.org>
Auto-Submit: Gabriel Charette <gab@chromium.org>
Reviewed-by: oysteine <oysteine@chromium.org>
Reviewed-by: François Doray <fdoray@chromium.org>
Cr-Commit-Position: refs/heads/master@{#760262}