[uClibc] 603e performance

Marc Leeman marc.leeman at barco.com
Fri Aug 20 06:46:09 UTC 2004


Hi, 

I am working on an embedded linux solution for about 6 months now. Up
until now, I was using a commercial solution (the largest player as they
claim), but a number of factors have made me question that solution:

  - semi-monopolist behaviour
  - extravagant prices
  - claiming credit where none is due
  - faster and better answers from cummunity then from 'support'
  - lock-in attempts
  - ...

Obviously, this is just me personally :) I feel in danger of losing my
freedom of choice.

Anywho, I've been quite impressed with the busybox/uclibc solution and
toolchain (kudos to all the ppl behind it). I managed to reduce the
cramfs size by 50% without reducing/removing functionality: from 4.5
Megs to 2 Megs.

However, there is one thing which is still bugging me, and though I am
investigating it, I still haven't found anything that might explain the
behaviour.

When using the stock gcc compiler for cross compilation on a ppc glibc
solution (or the cross powerpc-linux-uclib-gcc compiler), there is an
efficiency drop of > 20% in an network/mpeg stream parsing application;
which seems to be entirely in userspace (the apps).

I've been looking through the patches and at first sight, haven't really
identified anything that might be added in the ppc_82xx compiler; that
is not in the uclibc cross compiler (or the debian cross compiler).

Is anyone aware of this and/or does anyone have pointers as to what
might explain this (uclibc 3..3.4 <-> commercial: 3.3.1).

-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 189 bytes
Desc: Digital signature
Url : http://lists.busybox.net/pipermail/uclibc/attachments/20040820/05f32293/attachment-0002.pgp 


More information about the uClibc mailing list