unity8-dash 100% CPU usage when scrolling

[dinamic]

i'm on meizu mx4 (stable channel -r4), when i scroll up and down on the apps scope, unity8-dash is at 100%CPU usage. i have 60 apps installed and the apps scope is laggy. another annoying thing is that random icons are missing when i scroll fast up and down. i can see the random icons load.

[Launchpad Janitor (janitor)]

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

[Daniel van Vugt (vanvugt)]

Confirmed unity8-dash is weirdly inefficient. The good news is the CPU usage is so pronounced that some simple profiling should explain what the problem is.

[Daniel van Vugt (vanvugt)]

Please keep this bug about the 100% CPU. The lag is actually a combination of multiple different issues, in QtMir/Mir/Unity8/USC.

[Daniel van Vugt (vanvugt)]

Confirmed this bug is still occurring in the latest xenial devel-proposed images.

On a related note, I've been experimenting to observe how the phone behaves when you remove vsync (not for production use, but to take measurments). One interesting finding from that is we're still spending a very high amount of time rendering the unity8 dash, so much so that it's barely faster than with vsync enabled (mako), or in some cases still not fast enough to reach 60Hz (arale). So optimizing the dash, QML, Qt and fixing bugs like this looks like it might be on the critical path for phone performance.

[Daniel van Vugt (vanvugt)]

Possibly related: bug 1564721

[Albert Astals Cid (aacid)]

There's various reasons for this, one of the reasons for some scopes is https://bugs.launchpad.net/ubuntu/+source/unity8/+bug/1604025

[Albert Astals Cid (aacid)]

Another problem is UITK's image path interception https://bugs.launchpad.net/ubuntu/+source/ubuntu-ui-toolkit/+bug/1604029

[Daniel van Vugt (vanvugt)]

Unless mobile and desktop have dramatically different code paths, we can possibly merge this with bug 1604029 now, which has a fix. Because 23% CPU on a desktop is likely the same thing as 100% on a phone.

If you assume both mobile and desktop share the same code then probably one of the following has to be true:
  1. Bug 1604029 was a duplicate of this one and has solved it; or
  2. This bug has gone away mysteriously.

[Daniel van Vugt (vanvugt)]

or
   3. There was something using much more than 23% CPU which we ignored :)

The answer will become clearer after the fix for bug 1604029 hits devices.

[Albert Astals Cid (aacid)]

Ok, you're misunderstanding the 23% CPU time number i wrote there, it meant "23% of the CPU time spent by the program", not that it is using 23% of the CPU itself.

There's a bit more detailed description of where the CPU usage of vertically scrolling comes from at https://docs.google.com/document/d/1Gf2bRxNc_8R_620o_9J1EZ6GWLGOmNBrvNfLRTLwrqc/edit

[Daniel van Vugt (vanvugt)]

Ah OK, thanks. You need to be very careful with wording in those cases. I've changed the title of bug 1604029 to clarify.