[FFmpeg-devel] [PATCH 1/4] avcodec/j2kenc: Merge dwt_norm into lambda

[Sean McGovern]

Hi,


On Fri, Jun 21, 2024, 05:39 Andreas Rheinhardt <
andreas.rheinhardt at outlook.com> wrote:

> Michael Niedermayer:
> > This moves computations out of a loop
> >
> > Fixes: signed integer overflow: 31665934879948800 * 9998 cannot be
> represented in type 'long'
> > Fixes:
> 69024/clusterfuzz-testcase-minimized-ffmpeg_AV_CODEC_ID_JPEG2000_fuzzer-5949662967169024
> >
> > Found-by: continuous fuzzing process
> https://github.com/google/oss-fuzz/tree/master/projects/ffmpeg
> > Signed-off-by: Michael Niedermayer <michael at niedermayer.cc>
> > ---
> >  libavcodec/j2kenc.c | 9 +++++----
> >  1 file changed, 5 insertions(+), 4 deletions(-)
> >
> > diff --git a/libavcodec/j2kenc.c b/libavcodec/j2kenc.c
> > index 8cf82f7216c..91e66d81048 100644
> > --- a/libavcodec/j2kenc.c
> > +++ b/libavcodec/j2kenc.c
> > @@ -1349,7 +1349,7 @@ static void makelayers(Jpeg2000EncoderContext *s,
> Jpeg2000Tile *tile)
> >      }
> >  }
> >
> > -static int getcut(Jpeg2000Cblk *cblk, uint64_t lambda, int dwt_norm)
> > +static int getcut(Jpeg2000Cblk *cblk, uint64_t lambda)
> >  {
> >      int passno, res = 0;
> >      for (passno = 0; passno < cblk->npasses; passno++){
> > @@ -1361,7 +1361,7 @@ static int getcut(Jpeg2000Cblk *cblk, uint64_t
> lambda, int dwt_norm)
> >          dd = cblk->passes[passno].disto
> >             - (res ? cblk->passes[res-1].disto : 0);
> >
> > -        if (((dd * dwt_norm) >> WMSEDEC_SHIFT) * dwt_norm >= dr *
> lambda)
> > +        if (dd  >= dr * lambda)
> >              res = passno+1;
> >      }
> >      return res;
> > @@ -1384,11 +1384,12 @@ static void truncpasses(Jpeg2000EncoderContext
> *s, Jpeg2000Tile *tile)
> >                      Jpeg2000Band *band = reslevel->band + bandno;
> >                      Jpeg2000Prec *prec = band->prec + precno;
> >
> > +                    int64_t dwt_norm = dwt_norms[codsty->transform ==
> FF_DWT53][bandpos][lev] * (int64_t)band->i_stepsize >> 15;
> > +                    int64_t lambda_prime = av_rescale(s->lambda, 1 <<
> WMSEDEC_SHIFT, dwt_norm * dwt_norm);
> >                      for (cblkno = 0; cblkno <
> prec->nb_codeblocks_height * prec->nb_codeblocks_width; cblkno++){
> >                          Jpeg2000Cblk *cblk = prec->cblk + cblkno;
> >
> > -                        cblk->ninclpasses = getcut(cblk, s->lambda,
> > -                                (int64_t)dwt_norms[codsty->transform ==
> FF_DWT53][bandpos][lev] * (int64_t)band->i_stepsize >> 15);
> > +                        cblk->ninclpasses = getcut(cblk, lambda_prime);
> >                          cblk->layers[0].data_start = cblk->data;
> >                          cblk->layers[0].cum_passes = cblk->ninclpasses;
> >                          cblk->layers[0].npasses = cblk->ninclpasses;
>
> Does this also fix the UB in the vsynth*-jpeg2000-yuva444p16 tests?
>
> - Andreas
>
> _______________________________________________
> ffmpeg-devel mailing list
> ffmpeg-devel at ffmpeg.org
> https://ffmpeg.org/mailman/listinfo/ffmpeg-devel
>
> To unsubscribe, visit link above, or email
> ffmpeg-devel-request at ffmpeg.org with subject "unsubscribe".
>

I'm on staycation for my local jazz festival all week.
Was anyone able to test if this fixed the stuff Andreas mentioned?

-- Sean McGovern

>