[MPlayer-dev-eng] Re: Vorbis decoding performance

[Michael Niedermayer]

Hi

On Tue, Aug 08, 2006 at 12:28:38AM +0200, Balatoni Denes wrote:
> Hi!
> 
> hétf?? 07 augusztus 2006 23.09-kor Michael Niedermayer ezeket a bolcs 
> gondolatokat fogalmazta meg:
> > hmm floating point emulation is slow sure but it shouldnt be that slow ...
> >
> > [...]
> 
> Actually, tremor being at least 20 times faster than floating point whatever 
> is what I would have expected on arm. 
> 
> (wma - floating point - on a 206 mhz arm cpu needed 300% cpu time, compared to 
> tremor or libmad's 10-20%, when I tested these a few years ago on a strongarm 
> 1100 206 mhz cpu).

maybe you want to look at libavutil/softfloat.h
its some simple software floating point thingy, benchmark code is in 
libavutil/softfloat.c, if its fast on arm then maybe changing the speed
relevant float code in lavc to a macro based system which can either use 
real floats or soft floats would be interresting, allthough iam not sure
yet if it would be clean/readable enough for actually being accepted in svn

[...]

-- 
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