[Ffmpeg-devel] [Ffmpeg-devel-old] [patch] tighter snow slicing

Michael Niedermayer michaelni
Mon Apr 25 12:03:21 CEST 2005


On Wednesday 13 April 2005 21:32, Yartrebo wrote:
> I've improved the way that slicing bands are calculated so that idwt,
> decode_subband, and predict all are aligned. Speed-up is about 1%-8%
> depending on resolution for my (for my P4, and with both my submitted
> and my non-submitted asm code in use - AMDs will get better benefits,
> while using the C code will mask much of the benefit)
> I would consider this patch risky because the underlying allocation
> algorithm is fairly complex; however, I have run tests on plenty of
> resolutions (200x150 up to 1024x768, including odd-ball resolutions) and
> it hasn't thrown any assertions in the slice_buffer collection of
> functions (class if this were C++) yet. Lossless + 9/7 fails my tests,
> but it fails identically to cvs version (Michael seems to be quite aware
> of this fact - and it appears that 9/7 for lossless will likely be
> disabled).



"nothing is evil in the beginning. Even Sauron was not so." -- Elrond

More information about the ffmpeg-devel mailing list