[FFmpeg-devel] [PATCH] fix uninitialized data in dnxhd encoder

[Baptiste Coudurier]

On 09/18/2009 12:34 PM, Reimar D?ffinger wrote:
> On Fri, Sep 18, 2009 at 12:04:55PM -0700, Baptiste Coudurier wrote:
>> On 09/18/2009 06:37 AM, Reimar D?ffinger wrote:
>>> Index: libavcodec/dnxhdenc.c
>>> ===================================================================
>>> --- libavcodec/dnxhdenc.c       (revision 19894)
>>> +++ libavcodec/dnxhdenc.c       (working copy)
>>> @@ -237,6 +237,7 @@
>>>        DNXHDEncContext *ctx = avctx->priv_data;
>>>        const uint8_t header_prefix[5] = { 0x00,0x00,0x02,0x80,0x01 };
>>>
>>> +    memset(buf, 0, 0x170);
>>
>> According to specs header is always 640 and in 720p mode, the slices
>> indexes area must be padded.
>
> Indeed.
>
>>>        memcpy(buf, header_prefix, 5);
>>>        buf[5] = ctx->interlaced ? ctx->cur_field+2 : 0x01;
>>>        buf[6] = 0x80; // crc flag off
>>> @@ -483,7 +484,7 @@
>>>    static void dnxhd_setup_threads_slices(DNXHDEncContext *ctx, uint8_t *buf)
>>>    {
>>>        int mb_y, mb_x;
>>> -    int i, offset = 0;
>>> +    int i, offset = 640;
>>>        for (i = 0; i<   ctx->m.avctx->thread_count; i++) {
>>>            int thread_size = 0;
>>>            for (mb_y = ctx->thread[i]->m.start_mb_y; mb_y<   ctx->thread[i]->m.end_mb_y; mb_y++) {
>>> @@ -496,9 +497,11 @@
>>>                ctx->slice_size[mb_y]>>= 3;
>>>                thread_size += ctx->slice_size[mb_y];
>>>            }
>>> -        init_put_bits(&ctx->thread[i]->m.pb, buf + 640 + offset, thread_size);
>>> +        init_put_bits(&ctx->thread[i]->m.pb, buf + offset, thread_size);
>>>            offset += thread_size;
>>>        }
>>> +    if (offset<   ctx->cid_table->frame_size)
>>> +        memset(buf + offset, 0, ctx->cid_table->frame_size - offset);
>>>    }
>>
>> This will not work for interlaced more.
>
> Hm. How about this patch to test 1080i and 720p instead for better test
> coverage?
>
> Index: tests/codec-regression.sh
> ===================================================================
> --- tests/codec-regression.sh   (revision 19912)
> +++ tests/codec-regression.sh   (working copy)
> @@ -371,8 +371,11 @@
>   fi
>
>   if [ -n "$do_dnxhd" ] ; then
> -do_video_encoding dnxhd.dnxhd "" "-s hd1080 -b 120Mb -pix_fmt yuv422p -an -sws_flags neighbor+bitexact"
> +do_video_encoding 1080i.dnxhd "" "-flags +ildct -s hd1080 -b 120Mb -pix_fmt yuv422p -an -sws_flags neighbor+bitexact"
>   do_video_decoding "-r 25" "-s cif -pix_fmt yuv420p -sws_flags neighbor+bitexact"
> +
> +do_video_encoding 720p.dnxhd "" "-s hd720 -b 220Mb -pix_fmt yuv422p -an -sws_flags neighbor+bitexact"
> +do_video_decoding "-r 25" "-s cif -pix_fmt yuv420p -sws_flags neighbor+bitexact"
>   fi
>
>   if [ -n "$do_svq1" ] ; then
>
>
> Of course the regression test checksums need to be update, but for me
> decoding of the file encoded with
> ./ffmpeg_g -y -flags +bitexact -dct fastint -idct simple -sws_flags +accurate_rnd+bitexact -f image2 -vcodec pgmyuv -i ./tests/vsynth2/%02d.pgm -flags +ildct -s hd1080 -b 120Mb -pix_fmt yuv422p -an -sws_flags neighbor+bitexact ././tests/data/a-1080i.dnxhd
>
> does not work, it seem to be due to the "-flags +ildct" mode...
> Or is it a know issue that the dnxhd decoder can not handle interlaced
> data?

You can use .mov output. It seems parser as problems with raw interlaced 
dnxhd.

Also bitrate of 90Mb for 720p make more sense, even if 220Mb works.

Patch ok except that.

-- 
Baptiste COUDURIER
Key fingerprint                 8D77134D20CC9220201FC5DB0AC9325C5C1ABAAA
FFmpeg maintainer                                  http://www.ffmpeg.org