Bugzilla

(In reply to Chris Peterson [:cpeterson] from comment #14)

> I was looking at Android Speedometer scores for autoland and m-c (for changes from AArch64 PGO bug 1543215) and see this bug's change does appear to have impacted Speedometer results. The autoland results, in particular, look noisier than before.
> 
> https://treeherder.mozilla.org/perf.html#/graphs?timerange=604800&series=mozilla-central,1776166,1,10&series=mozilla-central,2008594,1,10&series=mozilla-central,1955058,1,10&series=autoland,1795984,1,10&series=autoland,2008165,1,10&selected=autoland,1795984,471903,805392654,10

Interesting. In my testing, some tests had execution time and/or variance go up slightly, but your perfherder data shows otherwise.

I'm in the process of executing 25 runs of each tp6m test on this revision: https://treeherder.mozilla.org/#/jobs?repo=try&group_state=expanded&searchStr=tp6m%2Candroid&revision=e63ac3728fe5c14ba4e092408e7a144911228b8c&selectedJob=245678138 which is on `try`. This is the final revision in this series, which bundles tweaks for both Google Pixel 2 and Motorola G5. The results from this revision should be directly comparable to your perfherder data.

(In reply to Chris Peterson [:cpeterson] from comment #14)

> I was looking at Android Speedometer scores for autoland and m-c (for changes from AArch64 PGO bug 1543215) and see this bug's change does appear to have impacted Speedometer results. The autoland results, in particular, look noisier than before.
> 
> https://treeherder.mozilla.org/perf.html#/graphs?timerange=604800&series=mozilla-central,1776166,1,10&series=mozilla-central,2008594,1,10&series=mozilla-central,1955058,1,10&series=autoland,1795984,1,10&series=autoland,2008165,1,10&selected=autoland,1795984,471903,805392654,10

Interesting. In my testing from prior to this patch landing (https://treeherder.mozilla.org/perf.html#/compare?originalProject=try&originalRevision=3b746ffb5d132ce80e442ac59a48d4ebd0776b0d&newProject=try&newRevision=24358abe9cc59a6283229dbcc95415bfaf586ca7&framework=10), some tests had execution time and/or variance go up slightly, but your perfherder data shows otherwise.

I'm in the process of executing 25 runs of each tp6m test on this revision: https://treeherder.mozilla.org/#/jobs?repo=try&group_state=expanded&searchStr=tp6m%2Candroid&revision=e63ac3728fe5c14ba4e092408e7a144911228b8c&selectedJob=245678138 which is on `try`. This is the final revision in this series, which bundles tweaks for both Google Pixel 2 and Motorola G5. The results from this revision should be directly comparable to your perfherder data.