[PD] OS X performance

[chris clepper]

On Jul 26, 2004, at 10:47 PM, Mirko Petrovich wrote:

> Hi all,
>
> I'm a Linux user an I'm very happy with pd performance in my desktop 
> (PIII 800MHz). Yesterday I tried pd 0.37-1 from Miller's site in a 
> Powerbook G4 (1GHz). I compared the cpu usage with patch 
> 08.convobros.pd and I saw a big difference. In the Linux system it 
> uses about 0.1% and in the G4 about 20%. Why this big difference ?

I just looked at the patch and profiled it.  My 1Ghz laptop used about 
13% CPU as measured by top.  The 0.1% number looks pretty bogus to me 
as the P3 is hardly a floating point monster and roughly equivalent to 
a PPC 7450 clock for clock.  I have noted discrepancies between top and 
the real-world on x86 chips in the past (tasks that obviously take lots 
of time to complete show up as near zero CPU load).

> Should I use a G4 optimized pd

The main code for Pd has little to no work for OSX and the PPC apart 
from making it run.  Thomas Grill had some Altivec code for the really 
basic stuff like the arithmetic objects that I looked at a while ago, 
but I don't know the status of that (devel branch?).  If someone did 
convert Pd to use fftw then the situation will be much improved as long 
as the Altivec code is properly built.

> or should I tweak the OSX system ?

Contrary to popular internet claims, there are not really any OSX 
system 'tweaks' to improve performance.  The best, and really only, way 
to make applications run better is to write better code.

For example:

> I tried some Gem patches too, and G4 performed a lot better


> (excepting the nVidia bug).

Maybe that will get fixed at some point.

cgc