[FFmpeg-devel] [PATCH 2/3] avcodec/fmvc: buffer size is stride based not 4*width

[Michael Niedermayer]

On Mon, Jun 13, 2022 at 12:10:44PM +0200, Paul B Mahol wrote:
> On Mon, Jun 13, 2022 at 11:48 AM Anton Khirnov <anton at khirnov.net> wrote:
> 
> > Quoting Paul B Mahol (2022-06-13 11:34:44)
> > > On Mon, Jun 13, 2022 at 11:10 AM Anton Khirnov <anton at khirnov.net>
> > wrote:
> > >
> > > > Quoting Paul B Mahol (2022-06-13 10:04:04)
> > > > > On Sat, Jun 11, 2022 at 4:55 PM Michael Niedermayer <
> > > > michael at niedermayer.cc>
> > > > > wrote:
> > > > >
> > > > > > On Sat, Jun 11, 2022 at 10:47:57AM +0200, Paul B Mahol wrote:
> > > > > > > Have you actually tested this "change" ?
> > > > > >
> > > > > > On every file i found
> > > > > > 6-methyl-5-hepten-2-one-CC-db_small.avi
> > > > > > fmvcVirtualDub_small.avi
> > > > > > skrzyzowanie4.avi
> > > > > > fmvc-poc.avi
> > > > > >
> > > > > > are there any other files i should test it on ?
> > > > > >
> > > > >
> > > > > Yes, the ones where stride != width.
> > > >
> > > > Give examples of such files then. And add more tests.
> > > >
> > > > You really should try to be more helpful if you care about this code
> > > > working.
> > >
> > >
> > > Code works perfectly from start. There are always attempts to break it.
> > > Your attempts to belittle my work are futile.
> >
> > Perfect code should live in an external repository that is locked
> > against modification.
> >
> > The ffmpeg repository is only for imperfect code that evolves with time,
> > and so requires changes.
> >
> >
> I dunno what Michael attempts to fix. Decoder works fine with valid files.
> I doubt that encoder would encode random bytes or padding into valid file
> bitstream.

the stride*4 / width*4 change was because of 2 things.
first with AV_PIX_FMT_BGR24 the data stored is not width*4

stride is in units of 4 bytes for some reason, so stride*4
fixes this
The 2nd issue is that the code addresses it by "s->stride * 4"
so the buffer allocation should be stride*4 if we belive the
other code is correct

        src = s->buffer;
...
        for (y = 0; y < avctx->height; y++) {
...
            src += s->stride * 4;

width*4 works because its bigger than stride*4 for BGR24 which is what all
samples i have use.

also
        ssrc = s->buffer;
        ...
        for (y = 0; y < avctx->height; y++) {
            ...
            ssrc += s->stride * 4;
and
        dst = (uint32_t *)s->buffer;

        for (block = 0, y = 0; y < s->yb; y++) {
            int block_h = s->blocks[block].h;
            uint32_t *rect = dst;

            for (x = 0; x < s->xb; x++) {
                int block_w = s->blocks[block].w;
                uint32_t *row = dst;

                block_h = s->blocks[block].h;
                if (s->blocks[block].xor) {
                    for (k = 0; k < block_h; k++) {
                        uint32_t *column = dst;
                        for (l = 0; l < block_w; l++)
                            *dst++ ^= *src++;
                        dst = &column[s->stride];
                    }
                }
                dst = &row[block_w];
                ++block;
            }
            dst = &rect[block_h * s->stride];
        }

Again, if you have fmvc files with more odd widths or other pixel formats
these would be very welcome. I can just say the code as is in git is wrong
and the buffer size as is in git is wrong. I noticed this when i added 
a check to see if the buffer is only partly filled and realized its
always partly filled even when the whole image is actually touched

thx

[...]
-- 
Michael     GnuPG fingerprint: 9FF2128B147EF6730BADF133611EC787040B0FAB

Rewriting code that is poorly written but fully understood is good.
Rewriting code that one doesnt understand is a sign that one is less smart
than the original author, trying to rewrite it will not make it better.
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 195 bytes
Desc: not available
URL: <https://ffmpeg.org/pipermail/ffmpeg-devel/attachments/20220613/a27ef498/attachment.sig>