Re: [PATCH v5 09/10] media: Avoid parsing quantization and huffman tables

[Philipp Zabel <p.zabel@pengutronix.de>]

On Wed, 2020-12-02 at 13:12 +0100, Hans Verkuil wrote:
> On 12/11/2020 04:05, Mirela Rabulea (OSS) wrote:
> > From: Mirela Rabulea <mirela.rabulea@nxp.com>
> > 
> > These are optional in struct v4l2_jpeg_header, so do not parse if
> > not requested, save some time.
> > 
> > Signed-off-by: Mirela Rabulea <mirela.rabulea@nxp.com>
> > ---
> >  drivers/media/v4l2-core/v4l2-jpeg.c | 6 ++++++
> >  1 file changed, 6 insertions(+)
> > 
> > diff --git a/drivers/media/v4l2-core/v4l2-jpeg.c b/drivers/media/v4l2-core/v4l2-jpeg.c
> > index d77e04083d57..7576cd0ce6b9 100644
> > --- a/drivers/media/v4l2-core/v4l2-jpeg.c
> > +++ b/drivers/media/v4l2-core/v4l2-jpeg.c
> > @@ -307,6 +307,9 @@ static int jpeg_parse_quantization_tables(struct jpeg_stream *stream,
> >  {
> >  	int len = jpeg_get_word_be(stream);
> >  
> > +	if (!tables)
> > +		return 0;
> > +
> 
> It feels more natural to check for a non-NULL out->quantization_tables
> or non-NULL out->huffman_tables pointer in v4l2_jpeg_parse_header()
> rather than in these low-level functions. It's weird to have this check here.

Ah, now I get it.

Yes, if you want to skip the entire DQT for performance reasons, it is
probably better to just call jpeg_skip_segment() instead of
jpeg_parse_quantization_table(). Otherwise the next jpeg_next_marker has
to scan the whole quantization table for segment markers.

regards
Philipp