On Sat, Jul 19, 2014 at 09:26:48AM +0200, Christophe Gisquet wrote: > Hi, > > 2014-07-10 19:35 GMT+02:00 Michael Niedermayer <michae...@gmx.at>: > > AV_COPY, AV_SWAP, AV_ZERO > > AV_[RW]N[8-64]A need aligned memory > > > > AV_[RW][BLN][8-64] > > AV_COPY*U doesnt need aligned memory, but might be faster if its > > aligned. > > These might be slower than the "aligned only" variants > > Documentation sent in another patch > > If I have to assume I need addresses aligned on 8 to use AV_ZERO64, > then unfortunately, forcing that alignment on the MvField struct bumps > its size from 11 to 16. Maybe it's worth it performance-wise, but I'm > not going to make us spend more time on this part. > > So here's a patch where I align various things to 4 to allow using > AV_ZERO32/AV_RN32A. > > -- > Christophe
> hevc.c | 3 +-- > hevc.h | 2 +- > hevc_mvs.c | 26 +++++++++++--------------- > 3 files changed, 13 insertions(+), 18 deletions(-) > 369100bcf2ed3e276061334a977f360ab26421f0 0002-hevc-use-intreadwrite.patch > From 86fcce0a680e799eff3df86eb28ee77a88736811 Mon Sep 17 00:00:00 2001 > From: Christophe Gisquet <christophe.gisq...@gmail.com> > Date: Sat, 12 Jul 2014 16:51:09 +0200 > Subject: [PATCH 2/4] hevc: use intreadwrite > > When dealing with MVs, both components may be processed at a time. > > On Win64, 560 to 539 cycles for derive_spatial_merge_candidates. applied thanks [...] -- Michael GnuPG fingerprint: 9FF2128B147EF6730BADF133611EC787040B0FAB DNS cache poisoning attacks, popular search engine, Google internet authority dont be evil, please
signature.asc
Description: Digital signature
_______________________________________________ ffmpeg-devel mailing list ffmpeg-devel@ffmpeg.org http://ffmpeg.org/mailman/listinfo/ffmpeg-devel
