[Ffmpeg-cvslog] r6096 - trunk/libavcodec/bitstream.h

Michael Niedermayer michaelni
Wed Aug 30 23:33:30 CEST 2006


Hi

On Wed, Aug 30, 2006 at 08:13:41PM +0200, Aurelien Jacobs wrote:
> On Wed, 30 Aug 2006 15:31:21 +0200
> Michael Niedermayer <michaelni at gmx.at> wrote:
> 
> > Hi
> > 
> > On Wed, Aug 30, 2006 at 01:46:41AM +0200, Aurelien Jacobs wrote:
> > > On Sat, 26 Aug 2006 13:37:34 +0200 (CEST)
> > > michael <subversion at mplayerhq.hu> wrote:
> > > 
> > > > Author: michael
> > > > Date: Sat Aug 26 13:37:33 2006
> > > > New Revision: 6096
> > > > 
> > > > Modified:
> > > >    trunk/libavcodec/bitstream.h
> > > > 
> > > > Log:
> > > > make A32 reader align its ptr during init no matter what missaligned mess is given to it
> > > 
> > > Indeed this fixed the segfault happening on ARM since r6062.
> > > But this unfortunatly ruined the performance of A32_BITSTREM_READER :-(
> > > 
> > > r6095
> > > BENCHMARKs: VC:   0.000s VO:   0.000s A:  68.007s Sys:   0.365s =   68.372s
> > > BENCHMARK%: VC:  0.0000% VO:  0.0000% A: 99.4666% Sys:  0.5334% = 100.0000%
> > > 
> > > r6096
> > > BENCHMARKs: VC:   0.000s VO:   0.000s A:  77.729s Sys:   0.424s =   78.153s
> > > BENCHMARK%: VC:  0.0000% VO:  0.0000% A: 99.4573% Sys:  0.5427% = 100.0000%
> > > 
> > > It is now as slow as the ALT_BITSTREAM_READER.
> > > I don't understand exactly why, but I've double checked that this is
> > > this revision which introduced this regression.
> > 
> > did the fast A32_BITSTREAM_READER produce non currupted sound at all?
> 
> Yes, it produces perfect sound and passes all the regtests !

very good, do you have any clue what caused this speed loss? if no
is there a obvious difference in the nm -S libavcodec/mpegaudiodec.o
output?
is there any difference in the state of the bitstream reader after 
init_get_bits() in the 2 revisions?
and whats the -S (compiled but not assembled) output from
mpegaudiodec.c of the 2 revisions (fast working and slow working)?
any obvious nonsense in there, especially in the huffman_decode() function?

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

In the past you could go to a library and read, borrow or copy any book
Today you'd get arrested for mere telling someone where the library is




More information about the ffmpeg-cvslog mailing list