Talos has detected a Firefox performance regression from push aa134dc8e0c7. As author of one of the patches included in that push, we need your help to address this regression. Regressions: 35% tcanvasmark summary osx-10-10 opt e10s 9531.08 -> 6212.62 You can find links to graphs and comparison views for each of the above tests at: https://treeherder.mozilla.org/perf.html#/alerts?id=3983 On the page above you can see an alert for each affected platform as well as a link to a graph showing the history of scores for this test. There is also a link to a treeherder page showing the Talos jobs in a pushlog format. To learn more about the regressing test(s), please see: https://wiki.mozilla.org/Buildbot/Talos/Tests For information on reproducing and debugging the regression, either on try or locally, see: https://wiki.mozilla.org/Buildbot/Talos/Running *** Please let us know your plans within 3 business days, or the offending patch(es) will be backed out! *** Our wiki page outlines the common responses and expectations: https://wiki.mozilla.org/Buildbot/Talos/RegressionBugsHandling

here is a subtest comparison of tcanvasmark: https://treeherder.mozilla.org/perf.html#/comparesubtest?originalProject=autoland&originalRevision=bf15eb2ce373&newProject=autoland&newRevision=aa134dc8e0c7&originalSignature=7e30e6e1721024983fc4318b9ea1ba24fcb4d1f3&newSignature=7e30e6e1721024983fc4318b9ea1ba24fcb4d1f3&framework=1 pretty much all the sub tests are regressed with 1 nice win. :milan, I see you are the patch author here, can you look into the root cause and help decide if we should accept this regression, fix it, or backout?

This is the patch that switched us back to using accelerated canvas. To confirm the regression came from that switch, can we can rerun the tests with this patch, but with gfx.canvas.azure.accelerated set to false? If we get the same kind of results, then we have a proof. I expect the accelerated canvas to be slower because of the cost of readback.

you can see the preference causes a 64% win: https://treeherder.mozilla.org/perf.html#/compare?originalProject=try&originalRevision=521751dc5eff&newProject=try&newRevision=45f8cb3310a568b3c257f2e44c1b99f5dc0aacc9&framework=1&showOnlyImportant=0

Yeah, so that's the reality of it then - most of the tests are not friendly to accelerated canvas, and we're just going back to the results we had before the acceleration got turned off (by accident.) So, we can just bless these results, and enjoy the month or so of the better results we had :) Given that we're going back to what we had, I'm not sure what is the correct status of this bug.

thanks for confirming :milan!