Re: [mpg123-devel] fixed point decoders

[Thomas O. <tho...@or...>]

Am Mon, 01 Jun 2009 14:48:28 +0900
schrieb Taihei Monma <tm...@ma...>: 

> On 2009/06/01, at 7:28, Patrick Dehne wrote:
> 
> > I just wanted to say that you are doing amazing work.
> >
> > Kudos!
> 
> Thanks. And with the latest commit, finally we have a fully compliant  
> fixed point decoder!

Awesome! It's now nearly as accurate as the fpu decoder with --enable-int-quality!
Though, one could call using 64bit numbers as temporaries cheating;-) 
(MAD does that, too, if possible... they have lots of variants for integer operations, but one fixed representation of fixed point numbers, with 28 bits fractional part, 4 bits (sign + 3 bits) whole part.)

But, looking at the performance front, I got very confused: On an Athlon XP, the int-quality fpu code is _a_lot_ faster than the normal truncating fpu code!
I got 2.9s runtime vs. 3.9s . On my core2duo, the truncation code as a bit of a lead as we knew before.
But I don't quite understand what's happening on the AthlonXP. Really strange... but I really cannot work on this today. Things to do.


Alrighty then,

Thomas.