New issue
Advanced search Search tips
Note: Color blocks (like █ or █) mean that a user may not be available. Tooltip shows the reason.

Issue 847752 link

Starred by 3 users
Status: Untriaged
Owner: ----
Cc:
llozano@chromium.org
g...@chromium.org
agrieve@chromium.org
█ yongnian...@intel.com
p...@chromium.org
thakis@chromium.org
pasko@chromium.org
Components:
EstimatedDays: ----
NextAction: ----
OS: Android
Pri: 3
Type: Bug



Sign in to add a comment

Enable ThinLTO optimizations for Chrome/Android on x86

Project Member Reported by hong.zh...@intel.com, May 30 2018 
we find ThinLTO can boost Chromium for x86 Android performance in many benchmarks, like speedometer2, CanvasMark, meanwhile reduce libchrome.so size. If we can turn ThinLTO optimization on for Chrome on x86 Android, it will be a great binary size + performance win.
This issue is to track the optimization.

Comment 1 by dtapu...@chromium.org, May 30 2018

Cc: p...@chromium.org agrieve@chromium.org
Components: Build
Is this not a duplicate of issue 407544?

Comment 2 by dtapu...@chromium.org, May 30 2018

Cc: -p...@chromium.org pasko@chromium.org

Comment 3 by g...@chromium.org, May 30 2018

Cc: -g...@google.com p...@chromium.org thakis@chromium.org llozano@chromium.org g...@chromium.org
Summary: Enable ThinLTO optimizations for Chrome/Android on x86 (was: Enable ThinLTO for Chrome/Android on x86)
dtapuska - I don't think so. Somewhat confusingly, LTO and ThinLTO are two different methods of doing link-time optimization, and that bug appears to be for the former (at least, we already have some ThinLTO -O0 builds for Android, so I'm not sure that ThinLTO would be blocked by the same things as issue 407544).

Also, this bug appears to be specifically to turn on optimizations for ThinLTO. We have plans to turn ThinLTO on for Android at some point, since it enables us to do things like CFI, but whether or not we turn on optimizations is less 'known'. :)

-----

If we enable this, we'll want ThinLTO optimizations on for Chrome/Android on ARM, as well, which I'm investigating in parallel. Since this sounds like both a size/perf win on x86, I'm hoping that my investigations turn out favorably. :)

hong.zheng - Are you able to share the numbers that you've collected and a bit of methodology about how they were collected? Maybe in a public doc (or similar)?
►

Sign in to add a comment