All graphs for this bug:
  https://chromeperf.appspot.com/group_report?bug_id=849051

(For debugging:) Original alerts at time of bug-filing:
  https://chromeperf.appspot.com/group_report?sid=79bf9064e4eb84bf712fc6ba28b92abea09e462a613c86fa6a7a215da4e4fc23


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

android-nexus5X

📍 Pinpoint job started.
https://pinpoint-dot-chromeperf.appspot.com/job/14cf8bd6240000

📍 Found a significant difference after 1 commit.
https://pinpoint-dot-chromeperf.appspot.com/job/14cf8bd6240000

GPU: Remove transfer buffer autoflushing. by jdarpinian@chromium.org
https://chromium.googlesource.com/chromium/src/+/dcfaac01164977c0e9862880ba33ba6edb5639c2
4.528 → 4.978 (+0.4498)

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

Hmm, we are doing 10% more work per frame, but the benchmark takes 10% less time to run. I am not sure this is a regression. I'll need to understand what the benchmark is measuring a little better.

After a closer look, total CPU time per frame is unchanged or slightly down. The benchmark is running slightly differently after my change: when I look at the traces I see different numbers of CSS animations running, and the wall clock time taken is different. This change in workload accounts for the rebalancing of work between different threads (IO is down, GPU is up).

Importantly, I see no regression in the smoothness benchmarks for this test. Given that, and the fact that the increase in GPU thread work is balanced by a reduction in IO thread work, I believe that this is not a performance regression.