2017-08-29 18:39 GMT+02:00 Martin Flöser <mgraess...@kde.org>:
> Am 2017-08-29 15:28, schrieb Matthias Klumpp:
>>
>> What we would be interested in is knowing about the future plans for
>> Plasma Mobile, especially in the area of design and performance
>> improvements. I didn't have the opportunity yet to test Plasma Mobile
>> on one of our development boards, but I hope I can get to that
>> soonish.
>
> Speaking for the KWin (Wayland compositor) part. Ideal would be drm support
> in the phone directly, but that is probably just wishful thinking.
The Vivante GPU on the i.mx6 is supported natively by the Linux
Etnaviv driver, which should make Wayland work instantly without too
many troubles (I would need to verify that though, maybe we need to do
a bit of kernel work to make the driver ready).
> KWin does
> have a working Android hwcomposer backend through libhybris. That works well
> and good for the limited use cases we had so far. The code base of this
> plugin is extremely small (< 1000 lines of code), which shows that most of
> the code is shared and it's relatively easy to extend and adjust this code
> base for new needs.
>
> Having done the initial port of KWin to the Nexus 5 I can proudly say that
> KWin performs very well on such kind of hardware. We benefit a lot from
> performance improvements done in the general KWin stack. And we can bring
> improvements from other platforms also to the phone stack. For example Roman
> did a lot of work to get layered compositing working in the DRM platform and
> that would be an ideal candidate to bring also in the hwcomposer plugin.
> Thus we have all the heavy lifting done to do things like rendering videos
> directly in a hardware plane.
Very nice :) I am fully convinced of the quality of the KWin
compositor, and I do not think we will run into troubles with it on
the Librem 5 hardware, as long as the Linux drivers work as expected
(as usual, there is a chance that we might need some fiddling to get
this to work properly).
> One of the things currently being worked on for the phone is to get rid of
> XWayland. This could give you a nice performance boost as you have a Wayland
> only system (of course it's also nice to have X support, but one cannot get
> it with GL support, so the gain is not worth it IMHO).
I am relatively sure that Zlatan (or CTO) would make pure Wayland a
pretty high priority ;-) Personally, I would not even bother with X on
the phone and just require everything running there to use Wayland. I
briefly thought that having X is useful for the convergence usecase
(if that becomes a reality), but even for that the apps need some
level of adjustment for a phone already, and in that case walking the
extra mile to get things to work with Wayland makes sense.
Cheers,
Matthias
