-march=pentium4 issue

Message

mapr · Post by **mapr** » Tue May 20, 2003 8:14 am

Could it be enough to simply compile the glibc without optimization (' -O ') ? Or does that simply include a ' -mno-sse2 ' in addition to some other -mno*some-terrific-instruction-set* ??
Thanks
Mario

bsolar · Post by **bsolar** » Tue May 20, 2003 9:33 am

mapr wrote:Could it be enough to simply compile the glibc without optimization (' -O ') ? Or does that simply include a ' -mno-sse2 ' in addition to some other -mno*some-terrific-instruction-set* ??
Thanks
Mario

The culprit was -march=pentium4.

-mno-sse2 disables sse2, thus solving the problem but you have to pay attention since the flag-o-matic might remove it from the CFLAGS for some packages.

If you don't have the corrected gcc yet I suggest you to use -march=pentium3 -mcpu=pentium4. Then you can use the options you want but -msse2.

jo_vermeulen · Post by **jo_vermeulen** » Mon Sep 01, 2003 11:47 am

subodh wrote:Yes this bug has been resolved in gcc-3.2.3 and gcc-3.3. gcc-3.3 is available in portage (hard masked).

Can you tell me where you found that information?
A reference to the gcc Changelog would be great. I have compiled my entire system using -march=pentium4 and would like to know if I have to recompile some packages.

The only thing I saw in the gcc 3.2.3 Changelog concerning -march=pentium4 was this:

IA-32

* Fixed a number of bugs in SSE and MMX intrinsics.
* Fixed common compiler crashes with SSE instruction set enabled
(implied by -march=pentium3, pentium4, athlon-xp)

By the way, I passed the python test.

Thanks in advance,

Jo Vermeulen[/quote]

m0sk · Post by **m0sk** » Mon Sep 01, 2003 2:19 pm

jo_vermeulen wrote:
subodh wrote:Yes this bug has been resolved in gcc-3.2.3 and gcc-3.3
Can you tell me where you found that information?

It is mentioned in bugzilla, check here and here.

Paule · Post by **Paule** » Mon Sep 01, 2003 2:36 pm

I also had the same problem untill I upgraded to gcc 3.3.1 using these flags....

CFLAGS="-march=pentium4 -O3 -mmmx -msse -msse2 -mfpmath=sse -funroll-loops -fprefetch-loop-arrays -pipe"

I recompiled my whole system with "emerge -e world". Since I have recompiled, My system is noticably faster and EVERYTHING works perfectly. Hope this helps.

Paul

__________________________________________________________________________

- Man is not judged by his problems, but by the way in which he solves them.

wishkah · Post by **wishkah** » Mon Sep 01, 2003 3:31 pm

Compiled my whole system with --ffast-math -march=pentium4 -O3 and the above python-int-cast line works (doesn't give an error). But with 3.2.3 it's expected I guess...

BTW, give --ffast-math a try, it really increases performance (compiled the whole sys with it, no prob, but it MIGHT cause probs when used with apps that rely on very precise calculations)

jo_vermeulen · Post by **jo_vermeulen** » Mon Sep 01, 2003 5:08 pm

Paule wrote:I also had the same problem untill I upgraded to gcc 3.3.1 using these flags....

CFLAGS="-march=pentium4 -O3 -mmmx -msse -msse2 -mfpmath=sse -funroll-loops -fprefetch-loop-arrays -pipe"

I recompiled my whole system with "emerge -e world". Since I have recompiled, My system is noticably faster and EVERYTHING works perfectly. Hope this helps.

But how did you upgrade to gcc 3.3.1?

First set the CFLAGS to pentium3, emerged gcc 3.3.1, re-emerged gcc 3.3.1 and compiled it with the pentium4 settings on the new gcc 3.3.1?

And then recompiled your entire system?

Thanks,

Jo Vermeulen

Paule · Post by **Paule** » Tue Sep 02, 2003 1:36 am

jo_vermeulen

"But how did you upgrade to gcc 3.3.1?

First set the CFLAGS to pentium3, emerged gcc 3.3.1, re-emerged gcc 3.3.1 and compiled it with the pentium4 settings on the new gcc 3.3.1?

And then recompiled your entire system?"

Yes.. That's exactly how I did it. Just let it run overnight.

Beetle B. · Post by **Beetle B.** » Tue Sep 02, 2003 7:39 pm

I'll be getting DSL in a few days, so I'll give this a go (virtually all my software was compiled with -march=pentium4, except some of the basic system ones).

I'm worried that this may create plenty of configuration files that "need updating" - does it? It would be a grand pain if it does.

Since most of my software was compiled with -march=pentium4, would it be the same if I just recompiled those programs that were compiled with -march=pentium3?

TIA,

mexicangeek · Post by **mexicangeek** » Wed Sep 03, 2003 4:34 am

I installed my entire system with the PENTIUM4 setting, because there wasn't a warning in /etc/make.conf anymore. (As opposed to the previous RC_3 or 4 install). It wasn't until today when I upgraded protage that I noticed the addition to the make.conf warning of this problem. I think it would be nice to get a recomended solution and explaination from the gentoo team. Please don't misunderstand, I'm not upset about any mishap about the warning no longer being present in make.conf, but I would just like a recomended solution. I have seen posts in this thread about doing "emerge -e world". I have also seen posts state that there is no problem at all with using pentium4. I would like to know what to do, before I go through the process of re-installing my OS again.

P.S. Hey gentoo team...you guys rock!

jo_vermeulen · Post by **jo_vermeulen** » Wed Sep 03, 2003 7:28 am

mexicangeek wrote:Please don't misunderstand, I'm not upset about any mishap about the warning no longer being present in make.conf, but I would just like a recomended solution. I have seen posts in this thread about doing "emerge -e world". I have also seen posts state that there is no problem at all with using pentium4. I would like to know what to do, before I go through the process of re-installing my OS again.

I completely agree with you!

Kind regards,

Jo Vermeulen

totopo · Post by **totopo** » Wed Sep 29, 2004 8:45 pm

Does someone have expirience with these settings?

http://www.freehackers.org/gentoo/gccfl ... _gcc3.html

Regards

virtual · Post by **virtual** » Wed Sep 29, 2004 11:21 pm

Hi,
totopo

Yes those flags seem fine on my pentium 4 3.2GHz HT

But I switch between -Os and -O3 depending on what I compile. I compiled my system (emerge system) with -O3 settings to get the fastest code, accepting an increase in memory usage.

Then I switched to -Os when compiling the various kde packages and openoffice, here I want the executable to be as small as possible and I still get most of the -O2 optimizations, but not the ones that would increase my memory usage at runtime. I do this because kde apps. and openoffice are usualy just waiting for me to type something in a edit box or waiting for me to push a button (of course if I have long and complex fourmulas to calculate in the spreadsheet of openoffice this may change). My main concern here is startup time, I have also prelinked my whole system.

I am uncertain though if "-fomit-frame-pointer" is nessecary as it is implied by the -O flags anyway, but this may be a way to force it on even if the compiler disagrees when only using a -O flag.

totopo · Post by **totopo** » Tue Apr 19, 2005 7:27 pm

Hello, an AMD Sempron processor uses the same flags as AMD XP processor?

or which one should I use?

CHOST="i686-pc-linux-gnu"
CFLAGS="-march=athlon-xp -O3 -pipe -fomit-frame-pointer"
CXXFLAGS="-march=athlon-xp -O3 -pipe -fomit-frame-pointer"

Regards

totopo

-march=pentium4 issue

why wasn't the warning about the pentum4 in make.conf?

Re: why wasn't the warning about the pentum4 in make.conf?

Experience with this?

Sempron