diff options
author | dyen <dyen@chromium.org> | 2016-02-01 16:55:01 -0800 |
---|---|---|
committer | Commit bot <commit-bot@chromium.org> | 2016-02-02 00:56:27 +0000 |
commit | ea6d840e4d96f6df739f55a89ccf69daa3569a77 (patch) | |
tree | 9d9ef45f0ab9f8b0a38d32e2fb846a7ccbb9a93e /extensions/browser/api | |
parent | e261467104d141089906fd02e1230306cb8d0040 (diff) | |
download | chromium_src-ea6d840e4d96f6df739f55a89ccf69daa3569a77.zip chromium_src-ea6d840e4d96f6df739f55a89ccf69daa3569a77.tar.gz chromium_src-ea6d840e4d96f6df739f55a89ccf69daa3569a77.tar.bz2 |
Fix GPUTiming offset so it is stable within significant digits.
The timer offset between GPU and CPU time were resolved to
microsecond accuracy before. This was causing inaccuracy
flakiness when comparing timestamps.
The issue is that we cannot assume these instructions run in
microsecond accuracy, so the offset will change from query to
query. Instead they should be considered only up to a sensible
accuracy for these instructions (such as milliseconds) and only
updating the offset if it has changed more than 1 millisecond.
The reason we compare with the previous offset is so we catch
drifts from the previous calculation rather than an integer
boundary. For example 0.9 vs 1.1 would be 0 and 1. Note even if
we were rounding it we would still have this problem, IE
0.45 vs 0.55 would be 0 and 1.
BUG=582676
Review URL: https://codereview.chromium.org/1647333002
Cr-Commit-Position: refs/heads/master@{#372843}
Diffstat (limited to 'extensions/browser/api')
0 files changed, 0 insertions, 0 deletions