Advertising supported Dmabuf modifiers of DRM device

[Daniel Stone]

Hi Philipp,

On Fri, 27 Jul 2018 at 14:10, Philipp Zabel <p.zabel at pengutronix.de> wrote:
> On Fri, 2018-07-27 at 11:23 +0100, Daniel Stone wrote:
> > Taking i.MX6 as an example - if we know we can never show at least one
> > view on an overlay (because it's transformed, or too many views, etc),
> > then we should not waste any time and energy having that surface do a
> > linear resolve. We can just skip this step and have the GPU output
> > tiled and then consume tiled during composition.
>
> Since you mention i.MX6, it can support an alpha transparent overlay.
> I have attached a patch to test this locally below, but that doesn't
> check for alpha plane support. Would it be enough to check whether pixel
> formats with alpha channel are advertised as supported by the plane?

To be honest, I'm not entirely sure. What I would want to check is the
damage processing: that the area under the alpha-blended plane still
got updated through the renderer if required. It should be pretty easy
to check by putting simple-shm under an alpha-blended plane, and
seeing if that got updated. Probably the tri-fan debug mode would help
here as well.

> > > Furthermore, we don’t have information in weston about content producing device.
> > > If an application wants to render with gpu, it should be responsibility of the application to query and use gpu friendly modifiers.
> > > We can provide them a list of scanout friendly modifiers which application can use. Then, application should choose a format which would work for rendering and scanout.
> >
> > Right, but again this may lead us to lose performance if the
> > application tries to find the lowest-common-denominator layout, when
> > it doesn't need to. Hence the suggestion for the compositor suggesting
> > the most appropriate sets of modifiers to clients depending on the
> > situation.
>
> Could this be used to notify Vulkan clients that they should reallocate
> their swapchain for better performance, by setting swapchain status to
> VK_SUBOPTIMAL_KHR?

Yeah, exactly!

Cheers,
Daniel