ActivityIndicators cause very high cpu usage for unity8/mir even when visible: false

Some observations:

* For me, this happens only when the Bluetooth page in systemsettings is opened.
* Doesn't matter if windowed or staged mode
* Happens on all devices

turns out the Bluetooth Page in systemsettings is animating 3 ActivityIndicators even when they are not visible.

Not sure how that causes unity8's CPU usage to be that high, but the linked ubuntu-system-settings branch improves the situation a lot.

However, given that the ActivityIndicators are invisible, we probably should investigate why they keep on triggering rendering updates.

Adding ubuntu-system-settings, qtmir and qtubuntu.

Perhaps the UITK should be added to. The ActivityIndicator should not calculate the animations all the when visible is false.

Marking unity8 as invalid as this is what it takes to constantly render an animated app surface. It might be improved in QtMir, but not much that unity8 can do about it.

Testing mir_demo_server_minimal with mir_demo_client_egltriangle, I'm seeing 30% CPU usage by the mir server just drawing a single spinny triangle.

Testing a simple QML demo shell, we get same CPU usage as mir. So punting to Mir team for investigation.

Nexus 4 perf flamegraph, 1 mir_demo_server, 5 mir_demo_client_egltriangle instances

I'm trying to get perf to return a more complete picture, but from the attached flame graph (one mir_demo_server, one mir_demo_client_egltriangle in Nexus 4):

Of the time spent in mir (29%) roughly 50-60% of that is spent in the driver/kernel code (related to gpu/display)

@Alberto - so last week we were pretty well focused on n4 & n7when i logged this bug. Which are both code aurora afiak, so I'm wondering if we're chasing a gpu specific issues. maybe it's worth a quick check on mx4 or bq devices for comparison.

Also worth noting that the 40% figure is with the default "ondemand" governor, which changes cpu clock "ondemand" and can potentially also change gpu clocks.

Using the "performance" governor (which fixes the cpu clock and potentially the gpu clock) the figure is closer to 10% cpu load on an N7.

On a vegetahd, with the default cpu governor, the server cpu load is around 20%. With the performance governor, the cpu load is around 11%.

Looking at callgrind output on the desktop, there are no obvious mir hotspots we could single out and optimize. It's similar to the output of perf, in that quite a bit time is just spent in the GL + dri driver code.

One more point, using the interactive governor (which we should be defaulting to anyway) the N4 cpu usage reported by top is around 23%.

adding projects linux-mako & linux-flo, is it possible to change to the interactive governor by default (instead of OnDemand)

also, not just for mako/flo but this is a question that should be asked/answered for all devices (arale, krillin,... as well)

Status changed to 'Confirmed' because the bug affects multiple users.

This bug was fixed in the package ubuntu-system-settings - 0.3+16.04.20151110-0ubuntu1

---------------
ubuntu-system-settings (0.3+16.04.20151110-0ubuntu1) xenial; urgency=medium

  [ Alejandro J. Cura ]
  * If a system package gets updated by the user, only show the user's
    copy (LP: #1265250)

  [ Bartosz Kosiorek ]
  * Allow translating "Unavailable" string (LP: #1511384) (LP: #1511384)

  [ Ken VanDine ]
  * Don't check for updates if not online (LP: #1505663)

  [ Michael Zanetti ]
  * Only animate ActivityIndicators when they are visible (LP: #1513450)

 -- Ken VanDine <email address hidden> Tue, 10 Nov 2015 15:27:33 +0000

spawned bug 1518418 for the governor topic