gentoo 2004.1 gcc produces very slow code!

cnot · n00b Joined: 17 Dec 2003 Posts: 16

I installed gemtoo 2004.1 on an opteron machine. Most things are working OK but I noticed that floating point C code compiled with gcc-3.3.3 runs very slowly. I also have access to a Suse-9 machine so I did some simple tests. I compiled the code on the gentoo machine and the suse machine and compared run times. The code compiled on the Suse machine runs 35% quicker than the code compiled on the gentoo machine when both executables are run on the gentoo machine.

Any ideas?

Jazz · Posted: Thu Jun 17, 2004 6:23 am Post subject:

35%, hmm that looks kinda acurate..

And how exactly did u perform these tests ? under what kernel, Gcc version, and exactly what executable are we talking about here ?

I think its a bit of exxaggeration, eh !
_________________
In 2010, M$ Windows will be a quantum processing emulation layer for a 128-bit mod of a 64-bit hack of a 32-bit patch to a 16-bit GUI for an 8-bit operating system written for a 4-bit processor from a 2-bit company that can't stand 1 bit of competition.

the_sphynx · Posted: Thu Jun 17, 2004 6:24 am Post subject:

What CFLAGS and MAKEOPTS do you use in your /etc/make.conf file?
_________________
Folding@Home User 285941

cnot · n00b Joined: 17 Dec 2003 Posts: 16

the_sphynx · Posted: Thu Jun 17, 2004 7:04 am Post subject:

The reason that I believe that it could be your make.conf is because the Opteron is a VERY fast processor. Here is my make.conf:

RAPUL · Posted: Thu Jun 17, 2004 7:50 am Post subject: Hmmm..

Ok, so you are compiling a program by your own, not emerging something.

So you are right then /etc/make.conf does not matter.

Can you post how do you compile that propietary

program? Or just the Makefile if it is not secret also...
_________________
Entropy rulz world.
Redundancy sux.
World is full of redundancy.
World sux.

cnot · n00b Joined: 17 Dec 2003 Posts: 16

THe makefile is very simple, here it is:

----------------------------------
CFLAGS = -O2
CC=gcc
LIBS =

OBJS = main.o \
comp1.o \
comp2.o

all:: test

test:$(OBJS)
\: ;
$(CC) -o $@ $(OBJS) $(LIBS) -lm
\: ;
------------------------------------------------

Thanks for any help with this.

deltaden · n00b Joined: 09 Aug 2003 Posts: 8

well, if I remember correctly, Suse isn't using standard kernel and gcc version for their AMD64 distrib. Their version of GCC 3.3 is a "special" one with some backport from gcc 3.4, wich seems coherent with what you are saying. The reason is that the 3.3 branch was closed before they wer able to include some important optimisations, but choose to inculde it in their distrib...

Can the explanation be that the maintainers from the amd64 kernel and the gcc branch are working for them ???

I'm not so sure, but I remember reading some infos about that.

cnot · n00b Joined: 17 Dec 2003 Posts: 16

OK, could very well explain why the 2.3.3 produces slow code on the AMD64.

I took the plunge and installed gcc 3.4, recompiled glibc and so far so good! The computational speed is now as expected and I have not had any crashes so far.