Wayland and GPU hotswapping [RESEND]

[Jerome Glisse]

The design described here allow such things to happen modulo the
restriction there is from the wayland server point of view (wayland
need to be able to texture from the buffer of all GPU involved).
Thought there is no policy in what we describe and i think it's better
to stay out of this. For instance toolkit can add a properties that
allow user to asign each app to a different gpu, toolkit can then even
define some standard way of doing this.

Cheers,
Jerome Glisse

On 11/10/10, nerdopolis <bluescreen_avenger at verizon.net> wrote:
> On Tuesday, November 09, 2010 05:42:16 pm Jesse Barnes wrote:
>> On Tue, 9 Nov 2010 17:34:21 -0500
>>
>> Jerome Glisse <j.glisse at gmail.com> wrote:
>> > On Tue, Nov 9, 2010 at 8:53 AM, nerdopolis
>> >
>> > <bluescreen_avenger at verizon.net> wrote:
>> > > A feature that seems to be in some laptops, and is being worked into
>> > > the Linux kernel.
>> > >
>> > > Would Wayland ever be able to handle this, when the feature comes into
>> > > Linux, or would the display server always need to be reset?
>> >
>> > Outcome of discussion between Dave, Kristian, Jesse and me while ridding
>> > a bus is that we should add a way for wayland to ask its client to
>> > redraw their surface
>> > using a another EGL driver and that wayland server should be able to
>> > handle client reporting failure doing so. Maybe i did get this wrong,
>> > feel free to correct me :)
>> >
>> > I think to cover all use case Dave presented we should have somethings
>> > like this:
>> > -wayland provide a list of EGL driver it can texture from (optimus case
>> > where we can migrate texture from nvidia to intel)
>> > -wayland can ask to it's client to switch GL driver and report if they
>> > can switch or
>> > not (i think this should happen in 2 step first ask and gather answer
>> > from all client
>> > if one client says it can't then forbid the switch, otherwise ask to
>> > all client to redraw
>> > with new driver)
>> >
>> > All this wouldn't need restart from wayland.
>>
>> Yeah, the big advantage of this approach is that it would allow apps to
>> create a new context, check for extensions, etc when they receive a
>> "hey time to switch" request.  The alternative (shimming all the GL
>> calls through a wrapper and advertising the intersection of the
>> driver feature sets) is way more work and you end up with something
>> that hurts performance and functionality.
>
> Sorry, I forgot to hit "Reply to All"...
>
> Thats pretty slick...
>
>
> What about multiple GPU's and graphics cards?
>
> Would Wayland be able to spread across screens connected to different
> graphics
> cards?
>
> Would Wayland be able to load balance across multple GPU's? Like rendering
> one
> client on one GPU, and another client on another GPU before drawing it on
> the
> display? I know some computers have multiple GPUs that do this...
>