[FFmpeg-devel] [PATCH 2/4] lavc/mpegvideo: use H263DSP dequant function

[Rémi Denis-Courmont]

Le sunnuntaina 9. kesäkuuta 2024, 19.13.54 EEST Andreas Rheinhardt a écrit :
> Rémi Denis-Courmont:
> > ---
> > 
> >  libavcodec/mpegvideo.c | 30 +++++-------------------------
> >  1 file changed, 5 insertions(+), 25 deletions(-)
> > 
> > diff --git a/libavcodec/mpegvideo.c b/libavcodec/mpegvideo.c
> > index 7af823b8bd..fa25d14970 100644
> > --- a/libavcodec/mpegvideo.c
> > +++ b/libavcodec/mpegvideo.c
> > @@ -201,7 +201,7 @@ static void
> > dct_unquantize_mpeg2_inter_c(MpegEncContext *s,> 
> >  static void dct_unquantize_h263_intra_c(MpegEncContext *s,
> >  
> >                                    int16_t *block, int n, int qscale)
> >  
> >  {
> > 
> > -    int i, level, qmul, qadd;
> > +    int qmul, qadd;
> > 
> >      int nCoeffs;
> >      
> >      av_assert2(s->block_last_index[n]>=0 || s->h263_aic);
> > 
> > @@ -219,23 +219,13 @@ static void
> > dct_unquantize_h263_intra_c(MpegEncContext *s,> 
> >      else
> >      
> >          nCoeffs= s->intra_scantable.raster_end[ s->block_last_index[n] ];
> > 
> > -    for(i=1; i<=nCoeffs; i++) {
> > -        level = block[i];
> > -        if (level) {
> > -            if (level < 0) {
> > -                level = level * qmul - qadd;
> > -            } else {
> > -                level = level * qmul + qadd;
> > -            }
> > -            block[i] = level;
> > -        }
> > -    }
> > +    s->h263dsp.h263_dct_unquantize(block, 1, nCoeffs, qmul, qadd);
> > 
> >  }
> >  
> >  static void dct_unquantize_h263_inter_c(MpegEncContext *s,
> >  
> >                                    int16_t *block, int n, int qscale)
> >  
> >  {
> > 
> > -    int i, level, qmul, qadd;
> > +    int qmul, qadd;
> > 
> >      int nCoeffs;
> >      
> >      av_assert2(s->block_last_index[n]>=0);
> > 
> > @@ -244,18 +234,7 @@ static void
> > dct_unquantize_h263_inter_c(MpegEncContext *s,> 
> >      qmul = qscale << 1;
> >      
> >      nCoeffs= s->inter_scantable.raster_end[ s->block_last_index[n] ];
> > 
> > -
> > -    for(i=0; i<=nCoeffs; i++) {
> > -        level = block[i];
> > -        if (level) {
> > -            if (level < 0) {
> > -                level = level * qmul - qadd;
> > -            } else {
> > -                level = level * qmul + qadd;
> > -            }
> > -            block[i] = level;
> > -        }
> > -    }
> > +    s->h263dsp.h263_dct_unquantize(block, 0, nCoeffs, qmul, qadd);
> > 
> >  }
> > 
> > @@ -275,6 +254,7 @@ static void gray8(uint8_t *dst, const uint8_t *src,
> > ptrdiff_t linesize, int h)> 
> >  static av_cold int dct_init(MpegEncContext *s)
> >  {
> >  
> >      ff_blockdsp_init(&s->bdsp);
> > 
> > +    ff_h263dsp_init(&s->h263dsp);
> > 
> >      ff_hpeldsp_init(&s->hdsp, s->avctx->flags);
> >      ff_videodsp_init(&s->vdsp, s->avctx->bits_per_raw_sample);
> 
> This approach will make H.261 use a h263dsp function which is a misnomer

In reality, that is an existing problem: the *existing* pair of DCT functions 
for H.263 is provisioned regardless of the codec. Fixing the god object and 
spaghetti anti-patterns in MpegEncContext is way outside the scope of this MR.

> and will lead to undefined references if no H.263 decoder or encoder is
> enabled.

Fair enough, *that* is easy to fix.

-- 
Rémi Denis-Courmont
http://www.remlab.net/