The changelog does include GBM changes to Intel tiling, which could be the root cause. If we end up with the wrong or no tiling, that'd be the kind of thing that would cause this kind of regression.

https://chromium.googlesource.com/chromiumos/platform/minigbm/+log/9e5d12a9d5133cd1709af821680fe6161b7593ba..aff29fdf276d3f565ab16ba88e4386930fb6dd99?n=10000

Kristian, anything here? The perf regression looks like it's still here.

We've had these regression for a while - since July 2017 and the chromeos changes in question are:

https://crosland.corp.google.com/log/9732.0.0..9735.0.0

suggesting that the minigbm changes regressed something. Looking at the first datapoint after CL:1008867 it looks like we're back to the level we had before the July regression.  Without investigating further, my guess would be that we lost tiling for some ARGB buffers that were allocated as scanout but not actually used for scanout.