[FFmpeg-devel,v6,19/22] cbs_h265: Add functions to turn HDR metadata into SEI

---
 libavcodec/Makefile   |   2 +-
 libavcodec/cbs_h265.c | 100 ++++++++++++++++++++++++++++++++++++++++++
 libavcodec/cbs_h265.h |  18 ++++++++
 3 files changed, 119 insertions(+), 1 deletion(-)
 create mode 100644 libavcodec/cbs_h265.c

Mark Thompson:
> ---
>  libavcodec/Makefile   |   2 +-
>  libavcodec/cbs_h265.c | 100 ++++++++++++++++++++++++++++++++++++++++++
>  libavcodec/cbs_h265.h |  18 ++++++++
>  3 files changed, 119 insertions(+), 1 deletion(-)
>  create mode 100644 libavcodec/cbs_h265.c
> 
> diff --git a/libavcodec/Makefile b/libavcodec/Makefile
> index 6394694617..96dc1df343 100644
> --- a/libavcodec/Makefile
> +++ b/libavcodec/Makefile
> @@ -71,7 +71,7 @@ OBJS-$(CONFIG_CABAC)                   += cabac.o
>  OBJS-$(CONFIG_CBS)                     += cbs.o
>  OBJS-$(CONFIG_CBS_AV1)                 += cbs_av1.o
>  OBJS-$(CONFIG_CBS_H264)                += cbs_h2645.o cbs_h264.o h2645_parse.o
> -OBJS-$(CONFIG_CBS_H265)                += cbs_h2645.o h2645_parse.o
> +OBJS-$(CONFIG_CBS_H265)                += cbs_h2645.o cbs_h265.o h2645_parse.o
>  OBJS-$(CONFIG_CBS_JPEG)                += cbs_jpeg.o
>  OBJS-$(CONFIG_CBS_MPEG2)               += cbs_mpeg2.o
>  OBJS-$(CONFIG_CBS_VP9)                 += cbs_vp9.o
> diff --git a/libavcodec/cbs_h265.c b/libavcodec/cbs_h265.c
> new file mode 100644
> index 0000000000..afbdd37896
> --- /dev/null
> +++ b/libavcodec/cbs_h265.c
> @@ -0,0 +1,100 @@
> +/*
> + * This file is part of FFmpeg.
> + *
> + * FFmpeg is free software; you can redistribute it and/or
> + * modify it under the terms of the GNU Lesser General Public
> + * License as published by the Free Software Foundation; either
> + * version 2.1 of the License, or (at your option) any later version.
> + *
> + * FFmpeg is distributed in the hope that it will be useful,
> + * but WITHOUT ANY WARRANTY; without even the implied warranty of
> + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
> + * Lesser General Public License for more details.
> + *
> + * You should have received a copy of the GNU Lesser General Public
> + * License along with FFmpeg; if not, write to the Free Software
> + * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
> + */
> +
> +#include "libavutil/mathematics.h"
> +#include "libavutil/mastering_display_metadata.h"
> +
> +#include "cbs_h265.h"
> +
> +
> +static uint32_t rescale_clip(AVRational value, uint32_t scale, uint32_t max)
> +{
> +    int64_t scaled = av_rescale(scale, value.num, value.den);
> +    return av_clip64(scaled, 0, max);
> +}
> +
> +static uint32_t rescale_fraction(AVRational value, uint32_t max)
> +{
> +    return rescale_clip(value, max, max);
> +}
> +
> +void ff_cbs_h265_fill_sei_mastering_display(H265RawSEIMasteringDisplayColourVolume *mdcv,
> +                                            const AVMasteringDisplayMetadata *mdm)
> +{
> +    memset(mdcv, 0, sizeof(*mdcv));
> +
> +    if (mdm->has_primaries) {
> +        // The values in the metadata structure are fractions between 0 and 1,
> +        // while the SEI message contains fixed-point values with an increment
> +        // of 0.00002.  So, scale up by 50000 to convert between them.
> +
> +        for (int a = 0; a < 3; a++) {
> +            // The metadata structure stores this in RGB order, but the SEI
> +            // wants it in GBR order.
> +            static const uint8_t mapping[] = { 1, 2, 0 };
> +            int b = mapping[a];
> +            mdcv->display_primaries_x[a] =
> +                rescale_fraction(mdm->display_primaries[b][0], 50000);
> +            mdcv->display_primaries_y[a] =
> +                rescale_fraction(mdm->display_primaries[b][1], 50000);

Unfortunately the Mastering display colour volume SEI message semantics
have been rewritten in the revision 06/19 of the H.265 standard. The old
semantics for the x coordinate is only retained when the value is in the
range 5..37000, for y it is 5..42000. Values outside these ranges are
unknown/unspecified/specified by other specifications. So I suggest to
clip it the values into these ranges.

> +        }
> +
> +        mdcv->white_point_x = rescale_fraction(mdm->white_point[0], 50000);
> +        mdcv->white_point_y = rescale_fraction(mdm->white_point[1], 50000);

The 5..37000/42000 ranges also apply to white_point_x/y.

> +    }
> +
> +    if (mdm->has_luminance) {
> +        // Metadata are rational values in candelas per square metre, SEI
> +        // contains fixed point in units of 0.0001 candelas per square
> +        // metre.  So scale up by 10000 to convert between them, and clip to
> +        // ensure that we don't overflow.
> +
> +        mdcv->max_display_mastering_luminance =
> +            rescale_clip(mdm->max_luminance, 10000, UINT32_MAX);
> +        mdcv->min_display_mastering_luminance =
> +            rescale_clip(mdm->min_luminance, 10000, UINT32_MAX);
> +

The old semantics of these fields is only retained in the ranges
50000..100000000 and 1..50000. min == max == 50000 is forbidden.

> +        // The spec has a hard requirement that min is less than the max,
> +        // and the SEI-writing code enforces that.
> +        if (!(mdcv->min_display_mastering_luminance <
> +              mdcv->max_display_mastering_luminance)) {
> +            if (mdcv->max_display_mastering_luminance == UINT32_MAX)
> +                mdcv->min_display_mastering_luminance =
> +                    mdcv->max_display_mastering_luminance - 1;
> +            else
> +                mdcv->max_display_mastering_luminance =
> +                    mdcv->min_display_mastering_luminance + 1;
> +        }
> +    } else {
> +        mdcv->max_display_mastering_luminance = 1;

max_display_mastering_luminance is no longer forbidden to be zero. In
fact the spec mentions: "An example of the use of values outside the
range for which semantics are specified in this Specification is that
ANSI/CTA 861-G (2016) uses the value 0 for the nominal maximum display
luminance of the mastering display to indicate that the nominal maximum
display luminance of the mastering display is unknown."

Obviously, the code for reading/writing the mastering display colour
volume SEI needs to be updated, too.

> +        mdcv->min_display_mastering_luminance = 0;
> +    }
> +}
> +
> +void ff_cbs_h265_fill_sei_content_light_level(H265RawSEIContentLightLevelInfo *cll,
> +                                              const AVContentLightMetadata *clm)
> +{
> +    memset(cll, 0, sizeof(*cll));
> +
> +    // Both the metadata and the SEI are in units of candelas per square
> +    // metre, so we only need to clip to ensure that they are in the valid
> +    // range.
> +
> +    cll->max_content_light_level     = av_clip_uintp2(clm->MaxCLL,  16);
> +    cll->max_pic_average_light_level = av_clip_uintp2(clm->MaxFALL, 16);
> +}

This one seems to be fine.

> diff --git a/libavcodec/cbs_h265.h b/libavcodec/cbs_h265.h
> index c14cd4cd66..999353f607 100644
> --- a/libavcodec/cbs_h265.h
> +++ b/libavcodec/cbs_h265.h
> @@ -747,4 +747,22 @@ typedef struct CodedBitstreamH265Context {
>  } CodedBitstreamH265Context;
>  
>  
> +struct AVMasteringDisplayMetadata;
> +struct AVContentLightMetadata;
> +
> +/**
> + * Fill an SEI Mastering Display Colour Volume structure with values derived
> + * from the AVMasteringDisplayMetadata side-data structure.
> + */
> +void ff_cbs_h265_fill_sei_mastering_display(H265RawSEIMasteringDisplayColourVolume *mdcv,
> +                                            const struct AVMasteringDisplayMetadata *mdm);
> +
> +/**
> + * Fill an SEI Content Light Level Info structure with values derived from
> + * the AVContentLightMetadata side-data structure.
> + */
> +void ff_cbs_h265_fill_sei_content_light_level(H265RawSEIContentLightLevelInfo *cll,
> +                                              const struct AVContentLightMetadata *clm);
> +
> +
>  #endif /* AVCODEC_CBS_H265_H */
>