[FFmpeg-devel] [PATCH 5/5] libavcodec/ffv1: Support storing LSB raw

[Michael Niedermayer]

On Mon, Oct 21, 2024 at 04:40:41PM +0200, Jerome Martinez wrote:
> Le 16/10/2024 à 22:53, Michael Niedermayer a écrit :
> > what are you testing?
> > the new code is faster than the old code.
> > There is something not right here, the range coder based implementation
> > i posted now is 5% faster then the range coder based one i posted earlier today
> > thats overall speed meassured with "time ./ffmpeg"
> > if you see no difference there is something fishy
> 
> I tested your latest "8/8" and "5/5" patches, I don't see a big difference
> on ~50 frames.
> ~20% but the "bitfield" version is ~27%.
> I wonder if it is due to the fact that you test with 1 frame only.
> 
> 
> > i simply tested this:
> > ./ffmpeg -i rawsamples/16/01.dpx -threads 1  -c:v ffv1 -context 1 -coder 1 -strict -2 -level 4 -rawlsb 4   -y /tmp/speedtest4.nut
> 
> I use this command too except -threads 1, I use all threads.
> GCC 11, Ubuntu on Windows.
> 

> > It uses 1 thread as the speed with more threads was very unstable between runs
> > and we want to know how fast ffv1 is not how multithreading behaves
> 
> Less unstable with more frames :).

The instability was found with more frames IIRC
but 1 frames is faster to test


> Maybe the impact is different between using 1 thread only and all threads,
> but at the end we want to have the best speed compared to compression when
> all threads are used, not with only one thread.
> 
> I tested with 1 thread on 1 frame, FFmpeg + master + put_rac_raw/get_rac_raw
> from a previous patch + the latest patch sent (8/8 Oct 16)
> - 0m9.754s without -strict -2 -level 4 -rawlsb 4
> - 0m8.542s with -strict -2 -level 4 -rawlsb 4
> 
> Worse, only 13% improvement instead of 25% from bitfield patch.

hmm, do you have an idea why we are seeing different results ?

also i see you didnt explicitly mentioned -context 1 -coder 1
that i used.

ATM it seems there are quite a few differences
-context 1 ?
cpu ?
gcc 11 vs gcc 9
different sample videos for testing

thx

[...]
-- 
Michael     GnuPG fingerprint: 9FF2128B147EF6730BADF133611EC787040B0FAB

When you are offended at any man's fault, turn to yourself and study your
own failings. Then you will forget your anger. -- Epictetus
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 195 bytes
Desc: not available
URL: <https://ffmpeg.org/pipermail/ffmpeg-devel/attachments/20241022/65a8e809/attachment.sig>