On 3/10/2025 9:59 PM, Michael Niedermayer wrote:
> On Mon, Mar 10, 2025 at 06:48:29PM +0100, Lynne wrote:
>> On 10/03/2025 01:45, Michael Niedermayer wrote:
>>> Sponsored-by: Sovereign Tech Fund
>>> Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
>>> ---
>>>    libavcodec/ffv1dec.c | 20 ++++++++-----
>>>    libavcodec/ffv1enc.c | 67 ++++++++++++++++++++++++++++++++++++++------
>>>    2 files changed, 71 insertions(+), 16 deletions(-)
>>>
>>> diff --git a/libavcodec/ffv1dec.c b/libavcodec/ffv1dec.c
>>> index 9738e4e9c9f..72c93fcf0cd 100644
>>> --- a/libavcodec/ffv1dec.c
>>> +++ b/libavcodec/ffv1dec.c
>>> @@ -92,7 +92,7 @@ static int is_input_end(RangeCoder *c, GetBitContext *gb, int ac)
>>>    static int decode_plane(FFV1Context *f, FFV1SliceContext *sc,
>>>                            GetBitContext *gb,
>>>                            uint8_t *src, int w, int h, int stride, int plane_index,
>>> -                        int pixel_stride, int ac)
>>> +                        int remap_index, int pixel_stride, int ac)
>>>    {
>>>        int x, y;
>>>        int16_t *sample[2];
>>> @@ -116,12 +116,18 @@ static int decode_plane(FFV1Context *f, FFV1SliceContext *sc,
>>>                int ret = decode_line(f, sc, gb, w, sample, plane_index, 8, ac);
>>>                if (ret < 0)
>>>                    return ret;
>>> +            if (sc->remap)
>>> +                for (x = 0; x < w; x++)
>>> +                    sample[1][x] = sc->fltmap[remap_index][sample[1][x]];
>>>                for (x = 0; x < w; x++)
>>>                    src[x*pixel_stride + stride * y] = sample[1][x];
>>>            } else {
>>>                int ret = decode_line(f, sc, gb, w, sample, plane_index, f->avctx->bits_per_raw_sample, ac);
>>>                if (ret < 0)
>>>                    return ret;
>>> +            if (sc->remap)
>>> +                for (x = 0; x < w; x++)
>>> +                    sample[1][x] = sc->fltmap[remap_index][sample[1][x]];
>>>                if (f->packed_at_lsb) {
>>>                    for (x = 0; x < w; x++) {
>>>                        ((uint16_t*)(src + stride*y))[x*pixel_stride] = sample[1][x];
>>> @@ -347,17 +353,17 @@ static int decode_slice(AVCodecContext *c, void *arg)
>>>        if (f->colorspace == 0 && (f->chroma_planes || !f->transparency)) {
>>>            const int cx            = x >> f->chroma_h_shift;
>>>            const int cy            = y >> f->chroma_v_shift;
>>> -        decode_plane(f, sc, &gb, p->data[0] + ps*x + y*p->linesize[0], width, height, p->linesize[0], 0, 1, ac);
>>> +        decode_plane(f, sc, &gb, p->data[0] + ps*x + y*p->linesize[0], width, height, p->linesize[0], 0, 0, 1, ac);
>>>            if (f->chroma_planes) {
>>> -            decode_plane(f, sc, &gb, p->data[1] + ps*cx+cy*p->linesize[1], chroma_width, chroma_height, p->linesize[1], 1, 1, ac);
>>> -            decode_plane(f, sc, &gb, p->data[2] + ps*cx+cy*p->linesize[2], chroma_width, chroma_height, p->linesize[2], 1, 1, ac);
>>> +            decode_plane(f, sc, &gb, p->data[1] + ps*cx+cy*p->linesize[1], chroma_width, chroma_height, p->linesize[1], 1, 1, 1, ac);
>>> +            decode_plane(f, sc, &gb, p->data[2] + ps*cx+cy*p->linesize[2], chroma_width, chroma_height, p->linesize[2], 1, 2, 1, ac);
>>>            }
>>>            if (f->transparency)
>>> -            decode_plane(f, sc, &gb, p->data[3] + ps*x + y*p->linesize[3], width, height, p->linesize[3], (f->version >= 4 && !f->chroma_planes) ? 1 : 2, 1, ac);
>>> +            decode_plane(f, sc, &gb, p->data[3] + ps*x + y*p->linesize[3], width, height, p->linesize[3], (f->version >= 4 && !f->chroma_planes) ? 1 : 2, 2, 1, ac);
>>>        } else if (f->colorspace == 0) {
>>> -         decode_plane(f, sc, &gb, p->data[0] + ps*x + y*p->linesize[0]    , width, height, p->linesize[0], 0, 2, ac);
>>> -         decode_plane(f, sc, &gb, p->data[0] + ps*x + y*p->linesize[0] + 1, width, height, p->linesize[0], 1, 2, ac);
>>> +         decode_plane(f, sc, &gb, p->data[0] + ps*x + y*p->linesize[0]    , width, height, p->linesize[0], 0, 0, 2, ac);
>>> +         decode_plane(f, sc, &gb, p->data[0] + ps*x + y*p->linesize[0] + 1, width, height, p->linesize[0], 1, 1, 2, ac);
>>>        } else if (f->use32bit) {
>>>            uint8_t *planes[4] = { p->data[0] + ps * x + y * p->linesize[0],
>>>                                   p->data[1] + ps * x + y * p->linesize[1],
>>> diff --git a/libavcodec/ffv1enc.c b/libavcodec/ffv1enc.c
>>> index f3041219b71..4c55c01263b 100644
>>> --- a/libavcodec/ffv1enc.c
>>> +++ b/libavcodec/ffv1enc.c
>>> @@ -272,7 +272,7 @@ static inline void put_vlc_symbol(PutBitContext *pb, VlcState *const state,
>>>    static int encode_plane(FFV1Context *f, FFV1SliceContext *sc,
>>>                            const uint8_t *src, int w, int h,
>>> -                         int stride, int plane_index, int pixel_stride, int ac)
>>> +                        int stride, int plane_index, int remap_index, int pixel_stride, int ac)
>>>    {
>>>        int x, y, i, ret;
>>>        const int pass1 = !!(f->avctx->flags & AV_CODEC_FLAG_PASS1);
>>> @@ -288,9 +288,14 @@ static int encode_plane(FFV1Context *f, FFV1SliceContext *sc,
>>>            sample[0][-1]= sample[1][0  ];
>>>            sample[1][ w]= sample[1][w-1];
>>> +
>>>            if (f->bits_per_raw_sample <= 8) {
>>>                for (x = 0; x < w; x++)
>>>                    sample[0][x] = src[x * pixel_stride + stride * y];
>>> +            if (sc->remap)
>>> +                for (x = 0; x < w; x++)
>>> +                    sample[0][x] = sc->fltmap[remap_index][ sample[0][x] ];
>>> +
>>>                if((ret = encode_line(f, sc, f->avctx, w, sample, plane_index, 8, ac, pass1)) < 0)
>>>                    return ret;
>>>            } else {
>>> @@ -303,6 +308,10 @@ static int encode_plane(FFV1Context *f, FFV1SliceContext *sc,
>>>                        sample[0][x] = ((uint16_t*)(src + stride*y))[x] >> (16 - f->bits_per_raw_sample);
>>>                    }
>>>                }
>>> +            if (sc->remap)
>>> +                for (x = 0; x < w; x++)
>>> +                    sample[0][x] = sc->fltmap[remap_index][ sample[0][x] ];
>>> +
>>>                if((ret = encode_line(f, sc, f->avctx, w, sample, plane_index, f->bits_per_raw_sample, ac, pass1)) < 0)
>>>                    return ret;
>>>            }
>>> @@ -310,6 +319,30 @@ static int encode_plane(FFV1Context *f, FFV1SliceContext *sc,
>>>        return 0;
>>>    }
>>> +static void load_plane(FFV1Context *f, FFV1SliceContext *sc,
>>> +                      const uint8_t *src, int w, int h,
>>> +                      int stride, int remap_index, int pixel_stride)
>>> +{
>>> +    int x, y;
>>> +
>>> +    memset(sc->fltmap[remap_index], 0, sizeof(sc->fltmap[remap_index]));
>>> +
>>> +    for (y = 0; y < h; y++) {
>>> +        if (f->bits_per_raw_sample <= 8) {
>>> +            for (x = 0; x < w; x++)
>>> +                sc->fltmap[remap_index][ src[x * pixel_stride + stride * y] ] = 1;
>>> +        } else {
>>> +            if (f->packed_at_lsb) {
>>> +                for (x = 0; x < w; x++)
>>> +                    sc->fltmap[remap_index][ ((uint16_t*)(src + stride*y))[x] ] = 1;
>>> +            } else {
>>> +                for (x = 0; x < w; x++)
>>> +                    sc->fltmap[remap_index][ ((uint16_t*)(src + stride*y))[x] >> (16 - f->bits_per_raw_sample) ] = 1;
>>> +            }
>>> +        }
>>> +    }
>>> +}
>>> +
>>>    static void write_quant_table(RangeCoder *c, int16_t *quant_table)
>>>    {
>>>        int last = 0;
>>> @@ -1178,8 +1211,24 @@ retry:
>>>        }
>>>        if (sc->remap) {
>>> -        if (f->colorspace == 0) {
>>> -            av_assert0(0);
>>> +        if (f->colorspace == 0 && c->pix_fmt != AV_PIX_FMT_YA8) {
>>> +            const int cx            = x >> f->chroma_h_shift;
>>> +            const int cy            = y >> f->chroma_v_shift;
>>> +
>>> +            //TODO decide on the order for the encoded remaps and loads. with golomb rice it
>>> +            // easier to have all range coded ones together, otherwise it may be nicer to handle each plane as a whole?
>>> +
>>> +            load_plane(f, sc, p->data[0] + ps*x + y*p->linesize[0], width, height, p->linesize[0], 0, 1);
>>> +
>>> +            if (f->chroma_planes) {
>>> +                load_plane(f, sc, p->data[1] + ps*cx+cy*p->linesize[1], chroma_width, chroma_height, p->linesize[1], 1, 1);
>>> +                load_plane(f, sc, p->data[2] + ps*cx+cy*p->linesize[2], chroma_width, chroma_height, p->linesize[2], 2, 1);
>>> +            }
>>> +            if (f->transparency)
>>> +                load_plane(f, sc, p->data[3] + ps*x + y*p->linesize[3], width, height, p->linesize[3], 3, 1);
>>> +        } else if (c->pix_fmt == AV_PIX_FMT_YA8) {
>>> +            load_plane(f, sc, p->data[0] +     ps*x + y*p->linesize[0], width, height, p->linesize[0], 0, 2);
>>> +            load_plane(f, sc, p->data[0] + 1 + ps*x + y*p->linesize[0], width, height, p->linesize[0], 1, 2);
>>>            } else if (f->use32bit) {
>>>                load_rgb_frame32(f, sc, planes, width, height, p->linesize);
>>>            } else
>>> @@ -1206,17 +1255,17 @@ retry:
>>>            const int cx            = x >> f->chroma_h_shift;
>>>            const int cy            = y >> f->chroma_v_shift;
>>> -        ret = encode_plane(f, sc, p->data[0] + ps*x + y*p->linesize[0], width, height, p->linesize[0], 0, 1, ac);
>>> +        ret = encode_plane(f, sc, p->data[0] + ps*x + y*p->linesize[0], width, height, p->linesize[0], 0, 0, 1, ac);
>>>            if (f->chroma_planes) {
>>> -            ret |= encode_plane(f, sc, p->data[1] + ps*cx+cy*p->linesize[1], chroma_width, chroma_height, p->linesize[1], 1, 1, ac);
>>> -            ret |= encode_plane(f, sc, p->data[2] + ps*cx+cy*p->linesize[2], chroma_width, chroma_height, p->linesize[2], 1, 1, ac);
>>> +            ret |= encode_plane(f, sc, p->data[1] + ps*cx+cy*p->linesize[1], chroma_width, chroma_height, p->linesize[1], 1, 1, 1, ac);
>>> +            ret |= encode_plane(f, sc, p->data[2] + ps*cx+cy*p->linesize[2], chroma_width, chroma_height, p->linesize[2], 1, 2, 1, ac);
>>>            }
>>>            if (f->transparency)
>>> -            ret |= encode_plane(f, sc, p->data[3] + ps*x + y*p->linesize[3], width, height, p->linesize[3], 2, 1, ac);
>>> +            ret |= encode_plane(f, sc, p->data[3] + ps*x + y*p->linesize[3], width, height, p->linesize[3], 2, 3, 1, ac);
>>>        } else if (c->pix_fmt == AV_PIX_FMT_YA8) {
>>> -        ret  = encode_plane(f, sc, p->data[0] +     ps*x + y*p->linesize[0], width, height, p->linesize[0], 0, 2, ac);
>>> -        ret |= encode_plane(f, sc, p->data[0] + 1 + ps*x + y*p->linesize[0], width, height, p->linesize[0], 1, 2, ac);
>>> +        ret  = encode_plane(f, sc, p->data[0] +     ps*x + y*p->linesize[0], width, height, p->linesize[0], 0, 0, 2, ac);
>>> +        ret |= encode_plane(f, sc, p->data[0] + 1 + ps*x + y*p->linesize[0], width, height, p->linesize[0], 1, 1, 2, ac);
>>>        } else if (f->use32bit) {
>>>            ret = encode_rgb_frame32(f, sc, planes, width, height, p->linesize, ac);
>>>        } else {
>>
>> Doesn't apply properly.
>> What does it even do? Does the current git master not already do this?
> 
> master implements remap for encode_rgb_frame*
> This implements it for the other cases.
> Most of this is untested though, i have no YUV planar float16 material
There is one very tiny exr sample that's YA float16, the one from 
fate-exr-ya-scanline-zip-half-12x8. Not sure how useful can it be 
considering it's so small and has no chroma planes.