Optimal compile flags for a pentium IV with gcc 3.1...

DataShark · n00b Joined: 26 May 2002 Posts: 28 Location: Portugal

Hí all!

since the vast majority of gcc31 problems seems to be related with the usage of athlon optimizations i'd like to know if there is any (hidden) mess if one is using the new pIV optimizations. If not which is the optimal set ? ( to me only speed(&stability) matters, not compile time nor binary sizes

- the system in cause is a toshiba P4 notebook )

Thanxs in advance

klieber

I haven't heard of too many positive reports of people using gcc 3.1 on Gentoo. See this thread for a couple of examples of why.

--kurt
_________________
The problem with political jokes is that they get elected

Forge · Posted: Wed Jul 03, 2002 6:02 pm Post subject:

I have been unable to install either 1.3a or 1.3b on my P4 system.

With decent optimizations, compiles break at several points (can't recall which).

With stock opts/no opts, I kill compilation after a day or two, since I get sick and tired of having the box down.

amigadave · n00b Joined: 29 Apr 2002 Posts: 55 Location: UK

Gentoo with GCC 3.1 works great for me and several friends. However, we all have Athlon Thunderbird's, using the flag:

Scandium · Posted: Thu Jul 04, 2002 9:32 am Post subject:

P4 optimisations are broken in gcc 3.1 (will be fixed in 3.1.1).

I can't give any technical details I just read it (and didn't go any deeper into it as I have an Athlon and no P4)

trapni · Posted: Thu Jul 04, 2002 2:01 pm Post subject:

d-range · n00b Joined: 02 Jul 2002 Posts: 2

Don't agree with klieber either. As long as you know what your doing, 1.3b works flawlessly. OK, it takes some time and recompiling to get everything right, but that's all for the fun of it.

I compiled stage1 with:

"-march=athlon -O3 -mmmx -m3dnow -malign-double -falign-functions=4 -funroll-loops -fschedule-insns2 -pipe"

stage two with:

"-march=athlon -O3 -mmmx -m3dnow -fomit-frame-pointer -funroll-loops -fschedule-insns2 -pipe"

all kde packages with -O2 instead of -O3, and without -malign-double (with it all kde programs would only crash), and everything else (including X, gnome-libs and all gtk stuff) with

"-march=athlon -O3 -mmmx -m3dnow -O3 -mmmx -m3dnow -fomit-frame-pointer -malign-double -falign-functions=4 -falign-loops=4 -falign-jumps=4 -funroll-loops -fschedule-insns2"

I think the flags above are about as far as you can push athlon optimizations (maybe if you include -ffast-math as well it gets even faster, but some programs, notably mozilla, will have behave strangely with it).

I haven't had any problems with these flags on a Duron 800, and it's blazingly fast and rock-solid stable :-)

Hope this helps somebody!

d-range

rac · Posted: Tue Jul 09, 2002 7:24 am Post subject:

There is a GCC 3.1.1 ebuild that claims to expressly address the P4 optimization issues.
_________________
For every higher wall, there is a taller ladder

Scandium · Posted: Tue Jul 09, 2002 11:10 am Post subject:

yes, seems like there is one since yesterday.
After all, gcc 3.1.1 will be released in a week anyway =)