HDR support in Wayland/Weston

[Pekka Paalanen]

On Thu, 10 Jan 2019 20:32:18 +0530
"Sharma, Shashank" <shashank.sharma at intel.com> wrote:

> Hello All,
> 
> This mail is to propose a design for enabling HDR support in 
> Wayland/Weston stack, using display engine capabilities, and get more 
> feedback and input from community.
> Here are few points (you might already know these), about HDR 
> framebuffers, videos and displays:
> - HDR content/buffers are composed in REC2020 colorspace, with bit depth 
> 10/12/16 BPC. Some of the popular formats are P010,P012,P016.
> - HDR content come with their own Metadata to be applied to get the 
> right luminance at the display device.
>      - The metadata can be of two type 1. static 2. dynamic . For 
> simplicity, this solution is focusing on static HDR only (HDR10 standard)
> - HDR content also provide its supported EOTF (electro optical transfer 
> function) information, which is a curve (like SRGB gamma curve). One 
> popular EOTF is PQ(ST2084).
> - HDR capable displays mention their EOTF and HDR metadata support 
> information in EDID CEA-861-G blocks.
> - Normal SRGB buffers are composed in SRGB color space following REC709 
> specifications.
> - For accurate blending in display engines, we need to make sure following:
>      - All the buffers are in same colorspace (Rec 709 or Rec 2020)
>      - All the buffers are liner (gamma/EOTF removed)
>      - All the buffers are tone mapped in same zone (HDR or SDR)
> 
> Please refer to the block diagram below, which presents a simple case of 
> a HDR P010 movie playback, with HDR buffers as video buffers, and SDR 
> buffers as subtitles. The subsystem looks and works like this:
> - A client decodes the buffer (using FFMpeg for example) and gets the 
> two buffers, one with video (HDR) and one subtitles (SDR)
> - Client passes following information to the compositor:
>       - The actual buffers
>       - Their colorspace infromation, BT2020 for HDR buffer, REC709 for 
> SDR buffer (planning to add a new protocol extension for this)
>       - The HDR metadata of the content (planning to add new protocol 
> for this)
> 
> - Compositors actions:
>     - Reads the End display's HDR capabilities from display EDID. Assume 
> its an HDR HDMI monitor.
>     - Compositor tone maps every view's framebuffer to match tone of end 
> display, applying a libVA filter. In this example:
>          - The SDR subtitles frame will go through SDR to HDR tone 
> mapping (called S2H)
>          - The HDR video frame will go through HDR to HDR tone mapping 
> (called H2H) if the HDR capabilities of monitor and content are different.
>          - Now both the buffers and the monitor are in the same tone 
> mapped range.
>      - As the end display is HDR capable, and one of the content frame 
> is HDR, the compositor will prepare all other planes for color space 
> conversion (CSC) from REC709->REC2020 using plane CSC property.
>      - As the CSC and blending should be done in liner space, compositor 
> will also use plane level degamma to make the buffers linear.
>      - These actions will make sure that, during blending:
>          - All the buffers are in same colorspace (REC2020)
>          - All the buffers are linear
>          - All the buffers are tone mapped (HDR)
>          - The plane level color properties patch, for DRM can be found 
> here: https://patchwork.freedesktop.org/series/30875/
>      - Now, in order to re-apply the HDR curve, compositor will apply 
> CRTC level gamma, so that the output buffer is non-linear again.
>      - To pass the output HDR information to kernel, so that it can 
> create and send AVI-info-frames to HDMI, compositor will set Connector 
> HDR metadata property.
>          - Code for the same can be found here: 
> https://patchwork.freedesktop.org/series/25091/
>      - And they will ever live happily after :).
> 
> Please provide inputs, feedbacks and suggestions for this design and 
> plan, so that we can improve out half cooked solution, and start sending 
> the patches.

Hi Shashank,

this is a major feature that would be awesome to have in Weston, but
it's also a big effort to design, implement, and maintain. To ease the
maintenance, we will need some serious work on the test suite, which
currently cannot even run the GL-renderer.

I understand your aim is to leverage display hardware capabilities to
the fullest, but we must also consider hardware that lacks some or all
of the conversion/mapping/other features while the monitor is well
HDR-capable. We also need to consider what happens when a monitor is
not HDR-capable or is somehow lacking. OTOH, whether a compositor
implements HDR support at all would be obvious in the advertised
Wayland globals and pixel formats.

Do we want to support HDR output in some way even if the display engine
(in the graphics card) does not support all the features you want to
use?

That question is on some parts already answered by whether we can
always rely on display engine hardware features: no, we cannot. The
reason is that Weston cannot generally guarantee that it can use a
hardware plane for a specific wl_surface. Hardware planes are used
opportunistically, and we must always have an equivalent fallback path
which usually means the GL-renderer. Users should not be able to see a
difference between a GL composited and a display engine composited
scene.

Needing the fallback from hardware planes determines the minimum of
what GL-renderer needs to implement. Whether there is anything more is
an open question. Using hardware planes in the test suite is difficult,
so having a renderer implementation will help maintenance too.

What if we have only SDR monitors? Or a mixed set of SDR and HDR
monitors?

I suppose it would be good for apps to know if HDR content is
preferable or not. You could hook up to the wl_output and
wl_surface.enter/leave mechanism that is already used to control e.g.
HiDPI preference (buffer/output scale). It means clients know which
outputs a wl_surface is present on, and somehow choose how to draw the
content. The compositor then automatically converts, as necessary, to
any outputs that do not exactly match the app provided content.

If switching the content type is too costly for apps to do during
window moves, then there could be a global recommendation as well,
either with explicit protocol extension or intended to be inferred from
the properties of all the wl_outputs exposed by the compositor.

**

What we need for this is a plan that can be implemented and merged in
small steps, while never knowingly regressing anything.

I've been talking about GL-renderer, because I don't think implementing
any of this with Pixman-renderer would be useful. I think you can just
forget about Pixman-renderer aside from honouring renderer capabilities
when advertising support to Wayland clients.

Before we can use hardware plane capabilities, we probably need the
fallback path, GL-renderer, to be up to par. I hope the HDR support can
be split into smaller bits somehow, where we could iterate between
extending GL-renderer and taking more hardware plane features into use,
if that makes sense.

Since you need a proving vehicle for new kernel UABI, this may be
difficult.

Maybe it would be possible to start with Wayland extensions that are
specific to Weston. You can cut corners there, so that you get content
to feed to into Weston. At some point, the Wayland extensions need to be
promoted to wayland-protocols and that is when you need more buy-in
from the community, and where the protocol design needs to be
well-thought. OTOH, the unstable protocols in wayland-protocols are
supposed to be similarly WIP, so maybe there is no need merge
weston-specific extensions.

Personal experience says that with something this big, you end up
needing to restructure and clean up Weston's current code before you
get to the main matter. The earlier we get any patch merged, the less
work you need carrying them forward until the kernel bits have landed.

Will be interesting to see how this unravels.


Thanks,
pq
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 833 bytes
Desc: OpenPGP digital signature
URL: <https://lists.freedesktop.org/archives/wayland-devel/attachments/20190114/08809d3a/attachment.sig>