[x264-devel] [ISSUE] r2145 frame broken ( 10-bit, x86_64 )

Harfe Leier astrataro at gmail.com
Mon Jan 16 12:53:36 CET 2012


FFV1 source sample(275MB): http://www.megaupload.com/?d=JJNA8RVB

On 16 January 2012 10:05, Harfe Leier <astrataro at gmail.com> wrote:

> I built x264 ( 10-bit, x86_64 ) with MSYS+GCC 4.6.2, and used it to encode
> some clips and find some broken frames. Re-encoding with same parameters
> re-produced the same result, not random broken frames.
>
> My build parameters:
> PKG_CONFIG_PATH="/mingw/x86_64-w64-mingw32/lib/pkgconfig/" \
> ./configure --prefix=/mingw/x86_64-w64-mingw32 --host=x86_64-w64-mingw32
> --cross-prefix=x86_64-w64-mingw32- --enable-win32thread \
> --extra-cflags="-Ofast -fno-strict-aliasing -fexcess-precision=fast"
> --enable-strip --enable-static \
>  --disable-swscale --disable-lavf --disable-ffms --disable-gpac
> --bit-depth=10
>
> And my encoding settings:
> --crf 18.0000 --profile high10 --level 5.1 --deblock 0:0 --ref 6 --bframes
> 8 --b-adapt 2 --b-pyramid normal --weightp 2 --keyint infinite --min-keyint
> 1 --scenecut 40 --qcomp 0.6 --rc-lookahead 48 --ipratio 1.40 --qpmin 0
> --qpmax 81 --qpstep 4 --stats "E:\MediaCoder\M03-SP-M11-rev\M03-SP-M11-rev.stats"
> --aq-mode 1 --aq-strength 1.0 --subme 11 --merange 32 --me tesa
> --partitions i8x8,p8x8,b8x8,i4x4,p4x4 --direct auto --trellis 0 --psy-rd
> 0.40:0.00 --no-dct-decimate --no-fast-pskip --sar 8:9 --videoformat ntsc
> --colorprim smpte170m --transfer smpte170m --colormatrix smpte170m --range
> tv --input-range tv --pass 1 --slow-firstpass --output
> "E:\MediaCoder\M03-SP-M11-rev\M03-SP-M11-rev_1pass.mkv"
>
> I re-encoded the same source with rev2140 I built several days ago from
> devel git on GitHub, those frames were not broken any more. So it might be
> a bug of CABAC trellis. Then I tried to encoded it with rev2145 but used
> --trellis 0 to turn trellis off ( subme reduced to 9 ), the result was also
> fine. Also, encoding the same source with rev2145 8-bit build gave fine
> result, but I'm not sure 8-bit builds are safe for other sources or not. I
> didn't test with 32-bit binary.
>
> Broken result sample: http://www.megaupload.com/?d=54TBB3OR
> The broken frames start from 00:00:22.
>
> If you need some source clips to re-produce the issue, I would like to
> upload an ffv1 file.
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.videolan.org/pipermail/x264-devel/attachments/20120116/a30de1b3/attachment.html>


More information about the x264-devel mailing list