CFLAGS Central (Part 2)

[bmeagle]

I am getting the following warning, I think it started after the emul32bit stuff got emerged when I emerged mozilla-flash.

* Your CFLAGS contains "-fPIC" which can break packages.
* Your CXXFLAGS contains "-fPIC" which can break packages.
* Your CFLAGS contains "-m64" which can break packages.
* Your CXXFLAGS contains "-m64" which can break packages.
*
* Before you file a bug, please remove these flags and
* re-compile the package in question as well as all its dependencies

My make.conf reveals:

CFLAGS="-O2 -march=athlon64 -fomit-frame-pointer -pipe -mmmx -msse -msse2 -msse3 -m3dnow"
CHOST="x86_64-pc-linux-gnu"
CXXFLAGS="${CFLAGS}

emerge --info reveals:

Portage 2.1.4.4 (default-linux/amd64/2007.0, gcc-4.2.2, glibc-2.7-r1, 2.6.24-gentoo-r1 x86_64)
=================================================================
System uname: 2.6.24-gentoo-r1 x86_64 AMD Athlon(tm) 64 X2 Dual Core Processor 4200+
Timestamp of tree: Mon, 18 Feb 2008 01:47:01 +0000
app-shells/bash: 3.2_p33
dev-java/java-config: 1.3.7, 2.1.4
dev-lang/python: 2.4.3-r4, 2.5.1-r5
dev-python/pycrypto: 2.0.1-r6
sys-apps/baselayout: 1.12.9
sys-apps/sandbox: 1.2.18.1-r2
sys-devel/autoconf: 2.13, 2.61
sys-devel/automake: 1.7.9-r1, 1.8.5-r3, 1.9.6-r2, 1.10
sys-devel/binutils: 2.18-r1
sys-devel/gcc-config: 1.4.0-r4
sys-devel/libtool: 1.5.26
virtual/os-headers: 2.6.24
ACCEPT_KEYWORDS="amd64 ~amd64"
CBUILD="x86_64-pc-linux-gnu"
CFLAGS="-m64 -fPIC"
CHOST="x86_64-pc-linux-gnu"
CONFIG_PROTECT="/etc /usr/kde/3.5/env /usr/kde/3.5/share/config /usr/kde/3.5/shutdown /usr/share/config"
CONFIG_PROTECT_MASK="/etc/env.d /etc/env.d/java/ /etc/fonts/fonts.conf /etc/gconf /etc/revdep-rebuild /etc/terminfo /etc/texmf/web2c /etc/udev/rules.d"
CXXFLAGS="-m64 -fPIC"

Where is the -m64 and -fpic coming from?
Thanx

[Ormaaj]

Hi. Has anyone done a comparison of -march=native and explicitly declaring a cflag for > gcc 4.2.x? What I'm particularly wondering about is that in the guide: http://gentoo-wiki.com/Safe_Cflags it recommends using the native march flag. However, I'm rebuilding my whole system under gcc 4.3 (just for fun, and to see how much stuff I can break.) Its an intel core2 quad, and gcc 4.3 also includes the -march=core2 option. Previously you had to pick different ones depending on if you were building for a 32 bit or 64 bit system (32 in this case). So basically... whats best? the old prescott option, native, or core2?

[cts]

My new baby arrived at my doorstep a few days ago and is ready to dive into work

It's a dual processor quadcore (would that be an octocore machine then? Xeon with 4GB RAM and an Areca 1230 hardware RAID controller with 12x1TB HDDs as RAID6. It should do for a while.

Now, can anyone point out a few things to me in regard of

Code: Select all

make.conf

and it's friends so that I can get up to speed? I also need to recompile "what was on my system image" (ie. all the stuff I just untared onto my system-HDD) I guess as that is generic i686 stuff - how would I do that?

Thanks for any help,
Christian.

PS. One rather strange thing I noticed is that when "Machine Check Exception" is enabled in the Kernel the machine likes to crash, actually quite often. Any ideas why that would happen? Disabling that options makes everything good again.

[nixnut]

merged above post here.

[a.b.]

I'm currently running

Code: Select all

-O2 -march=athlon-xp -pipe -fomit-frame-pointer -momit-leaf-frame-pointer -fno-strict-aliasing"

on a Sempron 2200+ with a cache of 256 KB and 786 MB RAM.

-Would -Os be the better choice here? The difference of this CPU to the full Athlon XP is a smaller cache, IIRC.
-Is there anything else you would recommend? (e.g. specific cxxflags)

[unK]

I would recommend you to get rid of -momit-leaf-frame-pointer flag, it actually does nothing as you have also -fomit-frame-pointer in your flags.

-Would -Os be the better choice here?

I don't think so. I was playing around with cflags some time ago and my system compiled with -Os was noticeably slower than with -O2 (Athlon 2600+ & 768 MB RAM here).

[harvester]

Hi guys, My CPU is a T2310

cpuinfo says its family 6 and model 15

according to the safe cflags wiki, march =nocona and it recomends -02 and -pipe

does that sound right ? it says nocona is for core2duo, i dont think I have that, its a laptop and it says Pentium Dual Core... but according to the cflags wiki, my family and model are core2 duo and recomends march=nocona

Also, any other optimizations you think I can benefit from and are safe ?

[a.b.]

I would recommend you to get rid of -momit-leaf-frame-pointer flag, it actually does nothing as you have also -fomit-frame-pointer in your flags.

-Would -Os be the better choice here?

I don't think so. I was playing around with cflags some time ago and my system compiled with -Os was noticeably slower than with -O2 (Athlon 2600+ & 768 MB RAM here).

I have removed the -momit-leaf-frame-pointer and kept O2 as you advised. No desire to recompile mz whole szstem to check if Os is a good choice on Sempron

[kernelOfTruth]

No desire to recompile mz whole szstem

check your keyboard layout

[kernelOfTruth]

is -fvisibility-inlines-hidden in cxxflags safe to use in combination with the linux-kernel ? or will it make your computer go explode ?

(this is a serious question )

is there even c++ code in the kernel ?

[lotusvale]

has anyone tried -march=core2 ? i have T7200 core 2 duo @ 2 ghz, just wondering if i should use this or prescott.

thanks

[kernelOfTruth]

has anyone tried -march=core2 ? i have T7200 core 2 duo @ 2 ghz, just wondering if i should use this or prescott.

thanks

this only applies to >=gcc-4.3.0

if you're using a core 2 (obviously) then try it out, you'll most likely will run into problems with some apps not compiling, though

so it'll be better right now sticking to =gcc-4.2* and -march=native

[lotusvale]

k thanks

[a.b.]

Some people seem to have flags like "falign-functions", falign-loops and falign-jumps that are also present in O2, but they've specified numbers (e.g. -falign-functions=32). What do those flags do and what does the best value depend on? Cache size?

[mv]

-falign-functions=32). What do those flags do and what does the best value depend on?

It means that the code is enlarged (sometimes filled with noops) so that the code can be put at the beginning of an e.g. 32bit boundary.
The best values depend on the code and the processor (i.e. on the way the processor pipelines commands or caches memory).
gcc selects the defaults on the processor type, so unless you know something special about the code (or want to build code "semi-optimized" for several processors) you should leave the defaults.

[avendesora]

is -fvisibility-inlines-hidden in cxxflags safe to use in combination with the linux-kernel ? or will it make your computer go explode ?

(this is a serious question )

is there even c++ code in the kernel ?

There is no C++ code in the kernel. But do not mess with compiler flags in there.
(Unless you really don't care if it eats your disks, scrambles your ram or eats you children, of course.)

[pdw_hu]

I'm in the process of recompiling with gcc 4.3.1 and -march=core2. Will edit my post if anything comes up...

[exif]

has anyone tried -march=core2 ? i have T7200 core 2 duo @ 2 ghz, just wondering if i should use this or prescott.

thanks

On a new install of my laptop with a T8100 Core2 Duo I re-compiled the toolchain and the base minimal install with march=core2. I had to use the ~amd64 versions of net-tools and glibc in order for everything to compile, but everything looks good now.

I'm now setting up the kernel.

[frak10]

Pentium 4 Williamette 1.8Ghz

CFLAGS="-O2 -march=pentium4 -mtune=pentium4 -pipe -frename-registers -fomit-frame-pointer"
CXXFLAGS="${CFLAGS}"
USE="X alsa apm arts artworkextra avi berkdb bitmap-fonts branding c++ cairo
cdr client-only crypt cups dvd encode f77 foomaticdb fortran gdbm gif
gnome gpm gtk imlib jpeg kde libg++ libwww mad mikmod motif mp3 mp4 mpeg
ncurses nls oggvorbis openal opengl oss pam pdflib png python quicktime
readline sdl spell ssl svga symlink tcpd truetype vorbis x86 xml2 xv zlib
-birdstep -bitdefender -bluetooth -br -brltty -brnet -cluster -qt3 -qt4"

Built entire system from the Network CD. I'd rather build my own system that let a package manager put poorly optimized packages on my disk.

I used to have -funroll-loops, but I found it slowed my system down somewhat.

[kernelOfTruth]

is -fvisibility-inlines-hidden in cxxflags safe to use in combination with the linux-kernel ? or will it make your computer go explode ?

(this is a serious question )

is there even c++ code in the kernel ?

There is no C++ code in the kernel. But do not mess with compiler flags in there.
(Unless you really don't care if it eats your disks, scrambles your ram or eats you children, of course.)

I'm a little adventurous

and thanks for explaining !

anyone using -frtl-abstract-sequences noticed that it causes infinite loops / letting the compilation stall at a point ?

(at least with gcc-4.3.1)

[zaccret]

I tried -march=native on my AMD64 (Venice model).
I executed

Code: Select all

echo 'int main(){return 0;}' > test.c && gcc -v -Q -march=native -O2 test.c -o test && rm test.c test

and it gives me

Code: Select all

options passées:  -v test.c -march=k8-sse3 -msahf --param l1-cache-size=64 --param l1-cache-line-size=64 -mtune=k8 -O2

2 things are interesting :
1 - march=native gives mtune AND march. I thought that march was implying mtune...
2 - It gives k8-sse3 for march and k8 for mtune

Do anybody has explanations ?

[AnXa]

Anyone remember in which document amd did state that these are official cflags for athlon64 processors like 3500+. I tried to look x86_64.org but it wheren't there anymore.

Code: Select all

CFLAGS="-O3 -march=athlon64 -ffast-math -funroll-all-loops -funit-at-a-time -fpeel-loops -ftracer -funswitch-loops -fomit-frame-pointer -pipe"

[Shining Arcanine]

I thought it would be nice if people shared their CFLAGS so people can get a better idea of what works and what does not work. I have an ~amd64 virtual machine running on a Core 2 Quad 9550 system with the following CFLAGS:

CFLAGS="-O3 -pipe -march=core2 -msse4.1 -mcx16 -msahf --param l1-cache-size=32 --param l1-cache-line-size=64 --param l2-cache-size=6144 -ftree-loop-linear -fgcse-sm -fgcse-las -ftree-loop-im -ftree-loop-ivcanon -fivopts -frename-registers -fweb"

I originally had graphite optimizations included too, but those seemed to be more thread parallelism oriented than vector parallelism oriented, so I removed them and rebuilt my system. I am using "-march=core2 -msse4.1 -mcx16 -msahf --param l1-cache-size=32 --param l1-cache-line-size=64 --param l2-cache-size=6144" in place of "-march=native" because I do not want GCC's auto-detection logic to be called on each invocation of GCC. The other flags probably do not make much of a difference, but they work and in the very rare cases that they do make a difference, I feel that they make the assembly code GCC produces closer to what I could write by hand, so I am happy with them.

I also have a laptop running ~x86 Gentoo Linux, but it had some hardware failures, so I do not have access to its CFLAGS at the moment. It is going back to Dell as soon as I receive a box from them that I can use to send it to them for repairs. It is using much more conservative CFLAGS (something like "-O2 -pipe -march=prescott" plus the graphite flags that I plan to remove in the near future), in particular because packages failed to compile correclty with -O3 and also because I did not want to experiment with the flags of a production system. Trying -O3 was the most ambitious thing I did in terms of my laptop's CFLAGS.

[eccerr0r]

Code: Select all

subaru:~/c/Program Files/Ventrilo$ grep Hz /proc/cpuinfo |head -2
model name      : Intel(R) Core(TM)2 Quad CPU    Q9550  @ 2.83GHz
cpu MHz         : 2974.639
subaru:~/c/Program Files/Ventrilo$ grep ^CFLAGS /etc/make.conf 
CFLAGS="-O2 -mtune=i686 -fomit-frame-pointer"

Yes, I'm running 32-bit code on a 64-bit capable machine.

I guess I don't like being that much on the edge ... well maybe a little. Also that this hard disk I'm using with this cpu is meant to be moved from machine to machine and still be usable... I guess I've given up on weird CFLAGS as it somewhat makes debugging harder, as gcc bugs can be exposed, and can't test on other machines to see if behavior matches.

In all honesty, it's a toss up what's "best" - this is in regards to a couple of my AVR GCC projects. I've been futzing around with how to get my code smaller, and -Os goes so far. Hacking my C code around seems to make a bigger difference.

[Ahenobarbi]

How about Gentoo in numbers(forums on that)?