why gcc-3.1?

Michael Brömer mib at telebel.de
Mon May 27 04:05:06 PDT 2002


Am Montag, 27. Mai 2002 06:37 schrieben Sie:
> Other than just being first-on-the-block, what are the advantages of
> gcc-3.1 over gcc-2.95.3?  
We had long discussion about this, have a look at the archives.
 
> Is gcc-3.1 planned for the next LFS? 
It is already in CVS

> Sounds spooky to me. 
Why? 

> I notice the guru's here all catching errors and making
> patches just trying to get a base system running.
Out of all the packages just ncurses,vim and perl need a little
"assistence" compiling and patches are already available.

> Is a system built with gcc-3.1 going to be a safe production system?
No problems here, as solid as a system build with 2.95.3, meaning
"rock-solid."

> What about those of us who don't know how to turn a compile error into a 
> patch and yet find ourselves needing packages that no one here has any need 
> for or experience with?
I guess that all actively maintained projects will make their source
compile with gcc-3.1 sooner or later, i suppose it will be sooner...

And by the way you are by no means forced to use the same compiler as
the book does. Just stick with 2.95.3 if it is more to your liking.

I use gcc-3.1 exclusively since the day it come out and i think its
great, "-O3 -march=athlon-mp" anyone? 

I also postulate (of course without any authority and with just a
single test i ran) that gcc-3.1 gives better performance. 
I did the following in my usual working environment on always 
the same machine, with BLFS systems build using gcc-2.95.3 and 3.1: 
:> time bzip2 mozilla.tar (mozilla.tar beeing about 250MB)
Thats what i got:

1 .) gcc-2.95.3: -O3 -march=i686 :
     209.47 user 
     1.18 system 
     3:30.77 elapsed 
     99%CPU 
2 .) gcc-3.1: -O3 -march=i686:
     197.82user 
     1.87system 
     3:20.94elapsed 
     99%CPU 
3 .) gcc-3.1: -O3 -march=athlon-mp:
     190.04user 
     0.91system 
     3:10.95elapsed
     99%CPU

About 10% better performance from 1.) to 3.) !

Michael




-- 
Unsubscribe: send email to listar at linuxfromscratch.org
and put 'unsubscribe lfs-dev' in the subject header of the message



More information about the lfs-dev mailing list