[FFmpeg-devel] [PATCH 2/3] libavcodec: ac3 floating point decoder fixes

Hendrik Leppkes h.leppkes at gmail.com
Mon Dec 3 17:13:07 CET 2012


On Mon, Dec 3, 2012 at 5:08 PM, Babic, Nedeljko <nbabic at mips.com> wrote:
>>> Handling of blocks with different length in downmix is erroneous in current code and can cause incorrect
>>> decoding and our patch is fixing this.
>>
>>I have no clue about AC3 but I suspect this can be solved by adding a
>>condition to not use this optimization in that special case.
>
> That approach is already taken in current code and it isn't working correctly.
>
> Bottom line is that there is a problem in ac3 decoder and our patch provides
> solution.
>
> On the other hand, If the error made is acceptable by FFmpeg we will drop the patch.

No one said that the error is acceptable.
It is just the patch which appears to not be acceptable without at
least looking for better alteratives, if it means that it'll result in
significant slowdown for all AC3 files when using downmixing just to
fix some corner cases.

So the real question is, why can this not be detected and only use the
slower approach on files where its necessary?

- Hendrik


More information about the ffmpeg-devel mailing list