Re: [PATCH v4 3/4] gpu: ipu-v3: ipu-ic: Add support for BT.709 encoding

[Steve Longerbeam]

On 2/11/19 2:12 AM, Philipp Zabel wrote:
> On Fri, 2019-02-08 at 17:47 -0800, Steve Longerbeam wrote:
> > Pass v4l2 encoding enum to the ipu_ic task init functions, and add
> > support for the BT.709 encoding and inverse encoding matrices.
> >
> > Reported-by: Tim Harvey <tharvey@xxxxxxxxxxxxx>
> > Signed-off-by: Steve Longerbeam <slongerbeam@xxxxxxxxx>
> > ---
> > Changes in v4:
> > - fix compile error.
> > Chnges in v3:
> > - none.
> > Changes in v2:
> > - only return "Unsupported YCbCr encoding" error if inf != outf,
> >    since if inf == outf, the identity matrix can be used. Reported
> >    by Tim Harvey.
> > ---
> >   drivers/gpu/ipu-v3/ipu-ic.c                 | 71 +++++++++++++++++++--
> >   drivers/gpu/ipu-v3/ipu-image-convert.c      |  1 +
> >   drivers/staging/media/imx/imx-ic-prpencvf.c |  4 +-
> >   include/video/imx-ipu-v3.h                  |  5 +-
> >   4 files changed, 71 insertions(+), 10 deletions(-)
> >
> > diff --git a/drivers/gpu/ipu-v3/ipu-ic.c b/drivers/gpu/ipu-v3/ipu-ic.c
> > index e459615a49a1..c5f83d7e357f 100644
> > --- a/drivers/gpu/ipu-v3/ipu-ic.c
> > +++ b/drivers/gpu/ipu-v3/ipu-ic.c
> > @@ -212,6 +212,23 @@ static const struct ic_csc_params ic_csc_identity = {
> >   	.scale = 2,
> >   };
> >   
> > +/*
> > + * BT.709 encoding from RGB full range to YUV limited range:
> > + *
> > + * Y = R *  .2126 + G *  .7152 + B *  .0722;
> > + * U = R * -.1146 + G * -.3854 + B *  .5000 + 128.;
> > + * V = R *  .5000 + G * -.4542 + B * -.0458 + 128.;
> This is a conversion to YUV full range. Limited range should be:
>
> YÂÂÂR *  .1826Â+ G *Â .6142 + B *ÂÂ.0620 + 16;
> U = R * -.1007Â+ G * -.3385 + B *ÂÂ.4392 + 128;
> VÂÂÂR *  .4392Â+ G * -.3990 + B * -.0402 + 128;

Yep, I fixed these to encode to limited range YUV, and ...

> > + */
> > +static const struct ic_csc_params ic_csc_rgb2ycbcr_bt709 = {
> > +	.coeff = {
> > +		{ 54, 183, 18 },
> > +		{ 483, 413, 128 },
> > +		{ 128, 396, 500 },
> > +	},
> > +	.offset = { 0, 512, 512 },
> > +	.scale = 1,
> > +};
> > +
> >   /*
> >    * Inverse BT.601 encoding from YUV limited range to RGB full range:
> >    *
> > @@ -229,12 +246,31 @@ static const struct ic_csc_params ic_csc_ycbcr2rgb_bt601 = {
> >   	.scale = 2,
> >   };
> >   
> > +/*
> > + * Inverse BT.709 encoding from YUV limited range to RGB full range:
> > + *
> > + * R = (1. * (Y - 16)) + (1.5748 * (Cr - 128));
> > + * G = (1. * (Y - 16)) - (0.1873 * (Cb - 128)) - (0.4681 * (Cr - 128));
> > + * B = (1. * (Y - 16)) + (1.8556 * (Cb - 128);
> The coefficients look like full range again, conversion from limited
> range YUV should look like:
>
>    R = (1.1644 * (Y - 16)) + (1.7927 * (Cr - 128));
>    G = (1.1644 * (Y - 16)) - (0.2132 * (Cb - 128)) - (0.5329 * (Cr - 128));
>    B = (1.1644 * (Y - 16)) + (2.1124 * (Cb - 128);

fixed these to inverse encode from limited range YUV.

> > + */
> > +static const struct ic_csc_params ic_csc_ycbcr2rgb_bt709 = {
> > +	.coeff = {
> > +		{ 128, 0, 202 },
> > +		{ 128, 488, 452 },
> > +		{ 128, 238, 0 },
> > +	},
> > +	.offset = { -435, 136, -507 },
> > +	.scale = 2,
> > +};
> > +
> >   static int init_csc(struct ipu_ic *ic,
> >   		    enum ipu_color_space inf,
> >   		    enum ipu_color_space outf,
> > +		    enum v4l2_ycbcr_encoding encoding,
> Should we support YUV BT.601 <-> YUV REC.709 conversions? That would
> require separate encodings for input and output.

How about if we pass the input and output encodings to the init ic task 
functions, but for now require they be the same? We can support 
transcoding in a later series.

>   Also, this might be a
> good time to think about adding quantization range parameters as well.

Again, I think for now, just include input/output quantization but 
require full range for RGB and limited range for YUV.

But that really balloons the arguments to ipu_ic_task_init_*(). Should 
we create an ipu_ic_task_init structure?

Steve