Am Fri, 02 Mar 2007 23:35:27 +0100 schrieb "Robert Leiking" <[EMAIL PROTECTED]>:
> Hi Thomas, > > > There is some code / macros in mpg123 for fixed point but they don't seem > > to work atm or are not complete. > > Just because I'm curious about it, which macros do you mean, in which files?? grep for REAL_IS_FIXED ... it's not that much, really ... mpg123.h essentially I didn't work with these in my time as maintainer, I have to admit. > The controller has got a floating point unit, but there are library > functions, which I want to integrate for optimization, those are designed for > fixed point arithmetic. E.g. to replace the inverse DCT part(I assume there's > the most workload) would be easier using fixed point parameters. So not every > part has to be in fixed point. Well... MMX computes fixed point values in the floating point registers ... so the mmx asm optimization of mpg123 does something like that already. Look at decode_i386.c, decode_mmx.s and dct64_mmx.s ... best in current svn, since I reorganized the structure of the optmization code. > > I didn't benchmark how well this performs (yet... I have a 486 and a 386 > > system to play with - on the 486 mpg123 is clearly faster than mpg321, which > > uses mad). > > So I assume 486 and 386 are fixed point as well, right? Yes and no. both machines have a fpu. Questions is if this one is really better than perhaps mad or float emulation. I've heard from ppl that they played cd quality mp3s on 386DX40 (one said even 386DX25 - but he admits that he is not sure anymore). Having the fpu on the 386 machine is not common (I bought an extra i387 chip). On the 486 it's included and seems really to do a better job than fixed point. Out of curiosity I'll test how fast emulated float math is, how this compares to mad and mpg123. Could be interesting to continue that discussion, but I guess we should take it off the mp3encoder list at some point... mpg123 specific stuff is better at [EMAIL PROTECTED] . Alrighty then, Thomas. _______________________________________________ mp3encoder mailing list [email protected] https://minnie.tuhs.org/mailman/listinfo/mp3encoder
