DRM COLOR_RANGE property

[Daniel Vetter]

On Tue, Jul 03, 2018 at 05:18:57PM +0100, Russell King - ARM Linux wrote:
> Can someone provide a deeper explanation about exactly what this
> property represents please?
> 
> Does this represent the range of YCbCr values _into_ a YCbCr-to-RGB
> conversion (in other words, the range of values in the framebuffer),
> or the expected output range from the conversion?
> 
> This matters, because a "limited", (iow, eg, BT601 compliant YCbCr)
> framebuffer where the Y signal is between 16..235 being displayed
> on a full-range RGB output would need different conversion from
> needing a limited-range RGB output.
> 
> If it is indeed the output, then why is this a property of the plane?
> Is that not a property of:
> 
> (a) whether the plane is being blended or overlaid onto a graphics
>     plane which uses full-range RGB
> (b) the properties of the output(s) to which the plane is being
>     displayed.
> 
> IOW, it seems that the output of the CSC is more to do with what's
> downstream of the plane than with the plane itself.
> 
> For example, take this situation:
> 
> plane 0 - graphics, full range RGB
>                                   >-- CRTC --> HDMI sink only supporting
> plane 1 - video, limited range YUV              limited range RGB
> 
> In order to display the graphics correctly in that scenario, the HDMI
> output needs to compress the RGB 0-255 range down to 16..236 to be
> compliant.  If the video is limited range, and the CSC produces a
> limited range RGB output, then plane 1 gets its range further
> compressed at the HDMI output, which surely is undesirable.
> 
> It would surely be better, if it's not possible to map the range of
> plane 0 to limited range, to instead expand the YUV range and then
> recompress it at the HDMI output to match the capabilities of the
> attached source.
> 
> It also seems logical that describing the range of the RGB plane would
> also be sane - if the application is limiting graphics RGB to 16..235,
> then you'd want the CSC output to do the same and there'd be no need
> for any range expansion or compression.
> 
> I'd personally like drm_plane_create_color_properties() to allow
> creation of COLOR_ENCODING without COLOR_RANGE (iow, supported_ranges
> being zero) until COLOR_RANGE is better defined than it is at present.
> 
> Thoughts?
> 
> I'm bringing this up, because the hardware I have has a CSC that
> accepts BT601 and BT709 formats, controlled by a single bit.  Another
> bit controls whether the CSC produces 0..255 output or 16..235 output.
> That is then blended/overlaid with the graphics plane (0..255) and
> sent to the output.  Having a "limited range" YUV plane produce
> 16..235 range output makes it look low-contrast compared to the
> graphics, which is what would be expected - "16" is not black
> compared to the black of the graphics in the same way that "235" is
> not white compared to the graphics.

Drivers are supposed to automatically figure this out by looking at the
edid. In i915 we also allow userspace to override this with the "Broadcast
RGB" property on the connector. Unfortunately we haven't polished that
property yet (not sure what other drivers are doing tbh), so it's only
listed in the Documentation/gpu/kms-properties.csv graveyard :-/

For the blending in-between right now the expectation is that the kernel
driver figures out a blending path that wastes the least amount of
precision. Depending upon hw that might mean blending in full RGB and
limiting later on, or converting all planes to limited RGB first.
Interactons with the CSC on the crtc for color corrections are also not
all that well defined.
-Daniel
-- 
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch