Hi,

On Mon, Feb 13, 2012 at 6:28 AM, Kostya Shishkov
<[email protected]> wrote:
> On Mon, Feb 13, 2012 at 03:13:53PM +0100, Anton Khirnov wrote:
>>
>> On Sun, 12 Feb 2012 19:22:37 +0100, Kostya Shishkov 
>> <[email protected]> wrote:
>> > On Sun, Feb 12, 2012 at 06:15:59PM +0100, Kostya Shishkov wrote:
>> > [...]
>> > >
>> > > New version will be sent with interface changed to encode2()
>> >
>> > and here it is
>> > From 93afd2d5e913f653e9029906890ac66d3bcc49d7 Mon Sep 17 00:00:00 2001
>> > From: Kostya Shishkov <[email protected]>
>> > Date: Thu, 2 Feb 2012 20:54:53 +0100
>> > Subject: [PATCH 2/3] prores encoder
>> >
>> > ---
>> >  Changelog              |    1 +
>> >  doc/general.texi       |    2 +-
>> >  libavcodec/Makefile    |    1 +
>> >  libavcodec/allcodecs.c |    2 +-
>> >  libavcodec/proresdsp.c |   17 +
>> >  libavcodec/proresdsp.h |    3 +
>> >  libavcodec/proresenc.c |  836 
>> > ++++++++++++++++++++++++++++++++++++++++++++++++
>> >  libavcodec/version.h   |    2 +-
>> >  8 files changed, 861 insertions(+), 3 deletions(-)
>> >  create mode 100644 libavcodec/proresenc.c
>> >
>> > +
>> > +    pkt_size = ctx->mb_width * ctx->mb_height * 64 * 3 * 12
>> > +               + ctx->num_slices * 2 + 200 + FF_MIN_BUFFER_SIZE;
>>
>> What a nice number you have there.
>>
>> > +
>> > +    if ((ret = ff_alloc_packet(pkt, pkt_size)) < 0) {
>> > +        av_log(avctx, AV_LOG_ERROR, "Error getting output packet.\n");
>> > +        return ret;
>> > +    }
>>
>> ff_alloc_packet() means 'ensure that we have at least that much space',
>> it's meant to be used when you know exactly how much space you need. I
>> suppose that pretty number above is the upper bound, so encoding might
>> work even if a user provides a smaller buffer.
>
> Indeed, in theory it should need about
> (ctx->mb_width * ctx->mb_height * ctx->bits_per_mb) / 8 + headers_size
> bytes, but currently encoder is limited by quantisers and may produce bigger
> frame than that.
>
>> And it'd be really nice if you actually checked that you don't write
>> outside of it.
>
> The funny thing is that put_bits should do that in theory but buffer size
> doesn't seem to be used anywhere. At least it shouldn't overflow that frame
> size even theoretically.

That's outside the scope of this patch, and I doubt it's useful. I
wonder if we shouldn't just keep put_bits() unchecked and add checks
to individual encoders. After all, corrupt input isn't an issue.

Ronald
_______________________________________________
libav-devel mailing list
[email protected]
https://lists.libav.org/mailman/listinfo/libav-devel

Reply via email to