[Mplayer-cvslog] CVS: main/libfaad2 mdct_tab.h, NONE, 1.1 neaacdec.h, NONE, 1.1 ps_dec.c, NONE, 1.1 ps_dec.h, NONE, 1.1 ps_syntax.c, NONE, 1.1 ps_tables.h, NONE, 1.1 Makefile, 1.4, 1.5 README, 1.2, 1.3 cfft.c, 1.4, 1.5 common.c, 1.4, 1.5 common.h, 1.5, 1.6 decoder.c, 1.4, 1.5 decoder.h, 1.4, 1.5 error.c, 1.3, 1.4 error.h, 1.3, 1.4 faad.h, 1.4, 1.5 filtbank.c, 1.4, 1.5 fixed.h, 1.4, 1.5 hcr.c, 1.4, 1.5 huffman.c, 1.3, 1.4 iq_table.h, 1.4, 1.5 is.c, 1.4, 1.5 mdct.c, 1.4, 1.5 mp4.c, 1.4, 1.5 mp4.h, 1.4, 1.5 output.c, 1.4, 1.5 output.h, 1.4, 1.5 pns.c, 1.4, 1.5 pns.h, 1.4, 1.5 sbr_dct.c, 1.4, 1.5 sbr_dct.h, 1.4, 1.5 sbr_dec.c, 1.4, 1.5 sbr_dec.h, 1.4, 1.5 sbr_e_nf.c, 1.4, 1.5 sbr_e_nf.h, 1.4, 1.5 sbr_fbt.c, 1.4, 1.5 sbr_hfadj.c, 1.4, 1.5 sbr_hfadj.h, 1.4, 1.5 sbr_hfgen.c, 1.4, 1.5 sbr_hfgen.h, 1.4, 1.5 sbr_huff.c, 1.4, 1.5 sbr_qmf.c, 1.4, 1.5 sbr_qmf.h, 1.4, 1.5 sbr_qmf_c.h, 1.3, 1.4 sbr_syntax.c, 1.4, 1.5 sbr_tf_grid.c, 1.4, 1.5 specrec.c, 1.4, 1.5 specrec.h, 1.4, 1.5 structs.h, 1.4, 1.5 syntax.c, 1.4, 1.5 syntax.h, 1.4, 1.5 tns.c, 1.4, 1.5

[Diego Biurrun]

D Richard Felker III writes:
> On Fri, Sep 24, 2004 at 07:31:41PM +0200, Diego Biurrun CVS wrote:
> > 
> > Modified Files:
> > 	Makefile README cfft.c common.c common.h decoder.c decoder.h 
> > 	error.c error.h faad.h filtbank.c fixed.h hcr.c huffman.c 
> > 	iq_table.h is.c mdct.c mp4.c mp4.h output.c output.h pns.c 
> > 	pns.h sbr_dct.c sbr_dct.h sbr_dec.c sbr_dec.h sbr_e_nf.c 
> > 	sbr_e_nf.h sbr_fbt.c sbr_hfadj.c sbr_hfadj.h sbr_hfgen.c 
> > 	sbr_hfgen.h sbr_huff.c sbr_qmf.c sbr_qmf.h sbr_qmf_c.h 
> > 	sbr_syntax.c sbr_tf_grid.c specrec.c specrec.h structs.h 
> > 	syntax.c syntax.h tns.c 
> > Added Files:
> > 	mdct_tab.h neaacdec.h ps_dec.c ps_dec.h ps_syntax.c 
> > 	ps_tables.h 
> > Log Message:
> > Update FAAD to a 2.1 beta CVS snapshot from 2004.07.12.
> > patch by adland <adland123 at yahoo dot com>
> 
> arrg! this is the new all-time cola record, with
> 100000000000000000000l!
> 
> after this commit, faad goes from using 14-15% cpu for audio decoding
> to 90% or more! both figured are with the makefile edited to define
> FIXED_POINT.
> 
> diego, please either find the cause of this and fix it ASAP, or
> reverse this stupid nonsense! and from now on don't commit code unless
> you're willing to be a coder and test it...

This code has been sitting in my local tree for months and played
every sample I have thrown at it.  There were several bug reports
about the previous version failing for samples that work fine with the
current updated version.

I have checked that it compiles fine and runs all my samples under
Linux, Cygwin and MinGW, it has been benchmarked extensively by adland
and I have also run some benchmarks, the results of which showed clear
performance improvements.

I just added -DFIXED_POINT to CFLAGS in libfaad2/Makefile, recompiled
and checked against the samples I have on this machine (some more
tomorrow).  I do not see any performance degradation.  We both have a
K6-III 500 and the rest of our machines is similar as well IIRC.

I have announced/threatened that I would apply this for eons and Alex
gave his OK back at Linuxtag (IIRC, probably before, it was ages ago).
Please provide me some samples to reproduce this and explain what else
I am supposed to test/check/verify since - for the life of me - I
cannot figure it out myself.

Diego