[FFmpeg-devel] [PATCH] lavc/mjpegdec: fix VA-API MJPEG decoding uninitialized huffman table

[Eoff, Ullysses A]

> -----Original Message-----
> From: ffmpeg-devel [mailto:ffmpeg-devel-bounces at ffmpeg.org] On Behalf Of mypopy at gmail.com
> Sent: Monday, October 29, 2018 8:10 PM
> To: FFmpeg development discussions and patches <ffmpeg-devel at ffmpeg.org>
> Cc: Zhao, Jun <jun.zhao at intel.com>; Lin, Decai <decai.lin at intel.com>
> Subject: Re: [FFmpeg-devel] [PATCH] lavc/mjpegdec: fix VA-API MJPEG decoding uninitialized huffman table
> 
> On Tue, Oct 30, 2018 at 10:41 AM Li, Zhong <zhong.li at intel.com> wrote:
> >
> > > From: ffmpeg-devel [mailto:ffmpeg-devel-bounces at ffmpeg.org] On Behalf
> > > Of mypopy at gmail.com
> > > Sent: Tuesday, October 30, 2018 9:02 AM
> > > To: FFmpeg development discussions and patches
> > > <ffmpeg-devel at ffmpeg.org>
> > > Cc: Zhao, Jun <jun.zhao at intel.com>; Lin, Decai <decai.lin at intel.com>
> > > Subject: Re: [FFmpeg-devel] [PATCH] lavc/mjpegdec: fix VA-API MJPEG
> > > decoding uninitialized huffman table
> > >
> > > On Mon, Oct 29, 2018 at 6:39 PM Li, Zhong <zhong.li at intel.com> wrote:
> > > >
> > > > > From: ffmpeg-devel [mailto:ffmpeg-devel-bounces at ffmpeg.org] On
> > > > > Behalf Of Jun Zhao
> > > > > Sent: Monday, October 29, 2018 6:26 PM
> > > > > To: ffmpeg-devel at ffmpeg.org
> > > > > Cc: Zhao, Jun <jun.zhao at intel.com>; Lin, Decai <decai.lin at intel.com>
> > > > > Subject: [FFmpeg-devel] [PATCH] lavc/mjpegdec: fix VA-API MJPEG
> > > > > decoding uninitialized huffman table
> > > > >
> > > > > From: Jun Zhao <jun.zhao at intel.com>
> > > > >
> > > > > Now VA-API HWAccel MJPEG decoding uninitialized huffman table, it's
> > > > > will lead to the decoding error like"Failed to sync surface 0x5: 23
> > > > > (internal decoding error)." in iHD open source driver.
> > > > >
> > > > > Signed-off-by: dlin2 <decai.lin at intel.com>
> > > > > Signed-off-by: Jun Zhao <jun.zhao at intel.com>
> > > > > ---
> > > > >  libavcodec/mjpegdec.c |   19 +++++++++++++++++++
> > > > >  1 files changed, 19 insertions(+), 0 deletions(-)
> > > > >
> > > > > diff --git a/libavcodec/mjpegdec.c b/libavcodec/mjpegdec.c index
> > > > > b0cb3ff..89effb6 100644
> > > > > --- a/libavcodec/mjpegdec.c
> > > > > +++ b/libavcodec/mjpegdec.c
> > > > > @@ -75,6 +75,25 @@ static int build_vlc(VLC *vlc, const uint8_t
> > > > > *bits_table, static int build_basic_mjpeg_vlc(MJpegDecodeContext *s)
> > > {
> > > > >      int ret;
> > > > > +    int i;
> > > > > +
> > > > > +    /* initialize default huffman tables */
> > > > > +    for (i = 0; i < 16; i++)
> > > > > +        s->raw_huffman_lengths[0][0][i] =
> > > > > avpriv_mjpeg_bits_dc_luminance[i + 1];
> > > > > +    for (i = 0; i < 12; i++)
> > > > > +        s->raw_huffman_values[0][0][i] = avpriv_mjpeg_val_dc[i];
> > > > > +    for (i = 0; i < 16; i++)
> > > > > +        s->raw_huffman_lengths[0][1][i] =
> > > > > avpriv_mjpeg_bits_dc_chrominance[i + 1];
> > > > > +    for (i = 0; i < 12; i++)
> > > > > +        s->raw_huffman_values[0][1][i] = avpriv_mjpeg_val_dc[i];
> > > > > +    for (i = 0; i < 16; i++)
> > > > > +        s->raw_huffman_lengths[1][0][i] =
> > > > > avpriv_mjpeg_bits_ac_luminance[i + 1];
> > > > > +    for (i = 0; i < 162; i++)
> > > > > +        s->raw_huffman_values[1][0][i] =
> > > > > avpriv_mjpeg_val_ac_luminance[i];
> > > > > +    for (i = 0; i < 16; i++)
> > > > > +        s->raw_huffman_lengths[1][1][i] =
> > > > > avpriv_mjpeg_bits_ac_chrominance[i + 1];
> > > > > +    for (i = 0; i < 162; i++)
> > > > > +        s->raw_huffman_values[1][1][i] =
> > > > > + avpriv_mjpeg_val_ac_chrominance[i];
> > > > >
> > > > >      if ((ret = build_vlc(&s->vlcs[0][0],
> > > avpriv_mjpeg_bits_dc_luminance,
> > > > >                           avpriv_mjpeg_val_dc, 12, 0, 0)) < 0)
> > > > > --
> > > > > 1.7.1
> > > >
> > > > Should this be fixed in iHD driver instead of ffmpeg?
> > > No, I don't think driver is a good place to initialize the huffman table.
> >
> > For the default Huffman table, thus should be initialized by driver itself.
> > For non-default case, thus should be passed from ffmpeg.
> > So, what is the case you want to fix? Both (if so, no need to mention iHD driver)?
> Both, now HWAccel MJPEG always setting the huffman table and
> can't distinguish the HWaccel JPEG whether default Huffman table.

If your VA-API HWAccel MJPEG decoding case(s) work with i965 driver
and not iHD then I would assume iHD needs fixing.

> _______________________________________________
> ffmpeg-devel mailing list
> ffmpeg-devel at ffmpeg.org
> http://ffmpeg.org/mailman/listinfo/ffmpeg-devel