[FFmpeg-devel] [PATCH] Is gray8 a paletted format?

[Måns Rullgård]

Stefano Sabatini <stefano.sabatini-lala at poste.it> writes:

> On date Monday 2010-02-01 02:28:17 +0100, Michael Niedermayer encoded:
>> On Sun, Jan 31, 2010 at 06:04:16PM +0100, Stefano Sabatini wrote:
>> > On date Monday 2009-11-16 01:07:09 +0100, Michael Niedermayer encoded:
>> > > On Sun, Nov 15, 2009 at 11:36:03PM +0100, Stefano Sabatini wrote:
>> > > > Hi all,
>> > > > 
>> > > > I see that in imgconvert.c gray8 is treated like a paletted format:
>> > > > check for example ff_set_systematic_pal() and ff_fill_pointer(), where
>> > > > it is allocated space for its palette.
>> > > > 
>> > > > If gray8 can be considered a paletted format, then I suggest to apply
>> > > > the attached patch, updating both the pixdesc array and the lsws
>> > > > macro.
>> > > > 
>> > > > Regards.
>> > > > -- 
>> > > > FFmpeg = Fundamental and Fantastic Most Porno Erudite Gadget
>> > > 
>> > > >  libavutil/pixdesc.c  |    1 +
>> > > >  libswscale/swscale.c |    1 +
>> > > >  2 files changed, 2 insertions(+)
>> > > > d4e555c7f6b51930b736b49221f546314265ec41  gray-is-pal.patch
>> > > 
>> > > This is ok in principle but it has some potential to uncover bugs
>> > > id like you to make sure that this works before comiting.
>> > > Also please make sure by benchmark that this causes no slowdown
>> > 
>> > Can you suggest what I should benchmark?
>> 
>> id put some START/STOP_TIMER in swscale-test and look at the gray8 cases
>> and see if one changes by a large amount
>
> I used my lite version swscale-test and an ad-hoc script, the results
> are in attachment.
>
> Conversion in some cases is ~10% slower.
>
> An alternative solution would be to make gray *non*-paletted, and
> eventually add a gray8-pal variant.

Why *would* gray8 be a paletted?

-- 
M?ns Rullg?rd
mans at mansr.com