[FFmpeg-devel] Decoding AVCHD using multi-threaded code

Alexander Strange astrange
Thu Dec 18 22:09:09 CET 2008


On Dec 17, 2008, at 9:20 AM, Michael Niedermayer wrote:

> On Wed, Dec 17, 2008 at 11:11:09AM +0100, Jean-Michel Pour? wrote:
>> Dear Friends,
>>
>> Are there any chances to parallelize decoding of this AVCHD file:
>> http://www.kdenlive.org/videos/avchd-test-1.mts
>>
>> This is a typical camcorder file.
>>
>> Do you see any hope to parallelize or should be drop official AVCHD
>> support in Kdenlive until multi-threading comes along?
>
> The file only contains 1 slice per field, thus there is no hope to  
> decode any
> slices in parallel, unless 2 fields where decoded at once.
> Decoding several fields and frames at once is being worked on.
> Sadly ive not seem much activity there. If you care about it,
> contact Alexander Strange. (if you provide some manpower (developers),
> patches or (money to alexander) maybe things will move quicker)

I tried to work on it during school, but I never had enough free time.  
I'm on break now and working on it again.
ffdshow-mt testers found a few decoding errors which I've fixed, but I  
could still use more AVCHD streams for testing - I only have x264 to  
generate stuff with.
Beyond that, I just need more time to work on it.

In the meantime, what are you doing the decoding for? For rendering,  
it would definitely help, but not if you're seeking around a lot  
during editing.
And you should definitely look into proxy editing; AVCHD isn't an  
intra format, so transcoding into something else will be more pleasant  
to work with no matter how fast it is.



> The second thing is as has been mentioned that there is a problem with
> timestamps and AFAIK everyone who looked at it had not been seen  
> again.
> If you are brave enough look at H.264 & H.222 (both freely available)
> and implement some parser that fills in the timestamps that are not
> stored in the source file.
>
> also try -flags2 +fast which should provide a little more speed at the
> expense of possible artifacts, and keep in mind you can maybe skip B
> frames to speed things up further.
>
> [...]
> -- 
> Michael     GnuPG fingerprint:  
> 9FF2128B147EF6730BADF133611EC787040B0FAB
>
> There will always be a question for which you do not know the  
> correct awnser.
> _______________________________________________
> ffmpeg-devel mailing list
> ffmpeg-devel at mplayerhq.hu
> https://lists.mplayerhq.hu/mailman/listinfo/ffmpeg-devel





More information about the ffmpeg-devel mailing list