[Intel-gfx] [PATCH] drm/i915/display: Prevent double YUV range correction on HDR planes

Ville Syrjälä ville.syrjala at linux.intel.com
Wed Dec 16 11:18:29 UTC 2020 

On Wed, Dec 16, 2020 at 01:46:58AM -0500, Andres Calderon Jaramillo wrote:
> On Tue, Dec 15, 2020 at 3:13 PM Ville Syrjälä
> <ville.syrjala at linux.intel.com> wrote:
> >
> > On Tue, Dec 15, 2020 at 03:06:30PM -0500, Andres Calderon Jaramillo wrote:
> > > On Tue, Dec 15, 2020 at 1:01 PM Ville Syrjälä
> > > <ville.syrjala at linux.intel.com> wrote:
> > > >
> > > > On Mon, Dec 14, 2020 at 10:57:03PM +0000, Shankar, Uma wrote:
> > > > >
> > > > >
> > > > > > -----Original Message-----
> > > > > > From: andrescj via sendgmr <andrescj at andrescj-cros-specialist.c.googlers.com>
> > > > > > On Behalf Of Andres Calderon Jaramillo
> > > > > > Sent: Tuesday, December 15, 2020 3:50 AM
> > > > > > To: intel-gfx at lists.freedesktop.org
> > > > > > Cc: Shankar, Uma <uma.shankar at intel.com>; Venkatesh Reddy, Sushma
> > > > > > <sushma.venkatesh.reddy at intel.com>; seanpaul at chromium.org; Andres
> > > > > > Calderon Jaramillo <andrescj at chromium.org>
> > > > > > Subject: [PATCH] drm/i915/display: Prevent double YUV range correction on HDR
> > > > > > planes
> > > > > >
> > > > > > From: Andres Calderon Jaramillo <andrescj at chromium.org>
> > > > > >
> > > > > > Prevent the ICL HDR plane pipeline from performing YUV color range correction
> > > > > > twice when the input is in limited range.
> > > > > >
> > > > > > Before this patch the following could happen: user space gives us a YUV buffer in
> > > > > > limited range; per the pipeline in [1], the plane would first go through a "YUV
> > > > > > Range correct" stage that expands the range; the plane would then go through
> > > > > > the "Input CSC" stage which would also expand the range because
> > > > > > icl_program_input_csc() would use a matrix and an offset that assume limited-
> > > > > > range input; this would ultimately cause dark and light colors to appear darker
> > > > > > and lighter than they should respectively.
> > > > > >
> > > > > > This is an issue because if a buffer switches between being scanned out and
> > > > > > being composited with the GPU, the user will see a color difference.
> > > > > > If this switching happens quickly and frequently, the user will perceive this as a
> > > > > > flickering.
> > > > > >
> > > > > > [1] https://01.org/sites/default/files/documentation/intel-gfx-prm-osrc-icllp-
> > > > > > vol12-displayengine_0.pdf#page=281
> > > > >
> > > > > Change looks good to me, double conversion should not be done.
> > > > > Plane input csc coefficients take care of the limited range.
> > > >
> > > > Might make sense to actually use the hw range correction instead.
> > > > Would avoid having to keep around two sets of coefficients.
> > > >
> > > > Also the question now becomes: How can our tests be passing if
> > > > we're doing the range correction twice?
> > > >
> > >
> > > I also considered just removing the limited-range matrix/offsets from
> > > icl_program_input_csc(). However, I figured that since the "Input CSC"
> > > stage must happen regardless of range correction, maybe it would be a
> > > gain to disable the "YUV Range Correction" stage. However, I'm not
> > > familiar with the hardware details, so I don't know for sure. I don't
> > > feel strongly either way; let me know what you'd prefer.
> >
> > IIRC we use the fixed function range compression + full range csc
> > approach on chv as well. Wouldn't hurt to do the same thing on
> > icl+ IMO.
> 
> Sounds good! Thanks for the review.
> 
> >
> > >
> > > I'm also curious about the testing. How do the tests work? I assume
> > > they are in Intel's CI and not open sourced? Do they use the DRM
> > > writeback connector (I didn't think this was implemented for i915
> > > yet)?
> >
> > Tests are here:
> > git://anongit.freedesktop.org/xorg/app/intel-gpu-tools
> >
> > In particular kms_plane/pixel-format tests should rather be failing I
> > think. Unless we've relaxed things a bit too much to get the crcs to
> > match when the results are close enough.
> >
> 
> Nice. I took a look at the tests you referenced. It seems that we may
> have just gotten unlucky with the color choice, but I'm not sure why
> (I kind of expected to still get a color difference with the current
> colors).

Hmm. I guess we need to think a bit more about the colors used.

> I tried changing the color in [1] to {0.0f, 0.2f, 0.2f}, and
> I got "CRC mismatches" failures for multiple formats including NV12,
> C8, and YUYV. Interestingly, I had tried something like this multiple
> times, and I couldn't get the CRC mismatch at first. I don't know what
> changed. Maybe the mode was in a bad state and when I rebooted the
> device, it was good.
> 
> Note, with that color change and with this patch, I still get "CRC
> mismatches" failures but only for C8. I'm unfamiliar with that format,
> but it seems like a different problem.

We render the image as RGB332 for C8 so it can only have 2^3 or 2^2
levels. Whereas for the 8bpc reference image we only chop off the
results to 5 msbs with the LUT, which gets us 2^5 levels. So the
choice of color would have to be constrained to fit the RGB332
limits to get the same output for both.

> 
> [1] https://source.chromium.org/chromiumos/chromiumos/codesearch/+/main:src/third_party/igt-gpu-tools/tests/kms_plane.c;l=383;drc=c8a9aa95e2c5c8d9d39f4f9388a7d602a2e64311
> 
> > >
> > > > >
> > > > > Reviewed-by: Uma Shankar <uma.shankar at intel.com>
> > > > >
> > > > > > Signed-off-by: Andres Calderon Jaramillo <andrescj at chromium.org>
> > > > > > ---
> > > > > >  drivers/gpu/drm/i915/display/intel_display.c | 7 +++++++
> > > > > >  1 file changed, 7 insertions(+)
> > > > > >
> > > > > > diff --git a/drivers/gpu/drm/i915/display/intel_display.c
> > > > > > b/drivers/gpu/drm/i915/display/intel_display.c
> > > > > > index 761be8deaa9b..aeea344b06ad 100644
> > > > > > --- a/drivers/gpu/drm/i915/display/intel_display.c
> > > > > > +++ b/drivers/gpu/drm/i915/display/intel_display.c
> > > > > > @@ -4811,6 +4811,13 @@ u32 glk_plane_color_ctl(const struct intel_crtc_state
> > > > > > *crtc_state,
> > > > > >                     plane_color_ctl |=
> > > > > > PLANE_COLOR_YUV_RANGE_CORRECTION_DISABLE;
> > > > > >     } else if (fb->format->is_yuv) {
> > > > > >             plane_color_ctl |= PLANE_COLOR_INPUT_CSC_ENABLE;
> > > > > > +
> > > > > > +           /*
> > > > > > +            * Disable the YUV range correction stage because the input CSC
> > > > > > +            * stage already takes care of range conversion by using separate
> > > > > > +            * matrices and offsets depending on the color range.
> > > > > > +            */
> > > > > > +           plane_color_ctl |=
> > > > > > PLANE_COLOR_YUV_RANGE_CORRECTION_DISABLE;
> > > > > >     }
> > > > > >
> > > > > >     return plane_color_ctl;
> > > > > > --
> > > > > > 2.29.2.684.gfbc64c5ab5-goog
> > > > >
> > > > > _______________________________________________
> > > > > Intel-gfx mailing list
> > > > > Intel-gfx at lists.freedesktop.org
> > > > > https://lists.freedesktop.org/mailman/listinfo/intel-gfx
> > > >
> > > > --
> > > > Ville Syrjälä
> > > > Intel
> >
> > --
> > Ville Syrjälä
> > Intel

-- 
Ville Syrjälä
Intel

More information about the Intel-gfx mailing list