All graphs for this bug:
  https://chromeperf.appspot.com/group_report?bug_id=896281

(For debugging:) Original alerts at time of bug-filing:
  https://chromeperf.appspot.com/group_report?sid=75a66a6af2f7234687c2aec7e4169e8c9003e898ba53aa7a492dc481a8e0995d


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

linux-perf
mac-10_12_laptop_low_end-perf
mac-10_13_laptop_high_end-perf

system_health.memory_desktop - Benchmark documentation link:
  https://bit.ly/system-health-benchmarks

📍 Pinpoint job started.
https://pinpoint-dot-chromeperf.appspot.com/job/17420996e40000

📍 Found a significant difference after 1 commit.
https://pinpoint-dot-chromeperf.appspot.com/job/17420996e40000

Reland gpu: Add UMA metric for measuring total raster task duration. by khushalsagar@chromium.org
https://chromium.googlesource.com/chromium/src/+/cd8fbb77cfa164eef66378204a588443af9bb875
memory:chrome:all_processes:reported_by_chrome:malloc:effective_size: 3.923e+07 → 4.156e+07 (+2.334e+06)

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

Benchmark documentation link:
  https://bit.ly/system-health-benchmarks

The entire increase is in malloc memory reported by the GPU process (9.7 -> 12.5). While the skia cache in the GPU at the time of the dump is empty, the gpu memory is still up by 2M which is completely attributed to mapped memory. In the context cache purging, this is the memory that's cleared. So I think this is another case of change in timing of the purging affecting whether it was processed in the GPU at the time of the memory dump.

I've still requested a debug trace for this to verify: https://pinpoint-dot-chromeperf.appspot.com/job/130b134ce40000