CFLAGS Central (Part 2)

[CosminG]

Thank you nxsty.

[Gooserider]

Hi, I have an interesting CPU that seems to be unclear about just what it is... I aquired it second hand, and don't have any paperwork on it.

According to this article on Safe Cflags I have two possible CPU's that it could be, depending on which things one looks at. Unfortunately, each has a slightly different set of CFLAGS, and I'm not sure which I should use, or even if it makes much of a difference. (the GNU Online GCC docs weren't clear about what if any difference there was.)
Doing "cat /proc/cpuinfo" gives me

processor : 0
vendor_id : AuthenticAMD
cpu family : 6
model : 10
model name : AMD Athlon(tm) XP 2500+
stepping : 0
cpu MHz : 1830.030
cache size : 512 KB
fdiv_bug : no
hlt_bug : no
f00f_bug : no
coma_bug : no
fpu : yes
fpu_exception : yes
cpuid level : 1
wp : yes
flags : fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 mmx fxsr sse syscall mmxext 3dnowext 3dnow up ts
bogomips : 3664.15

Note the model, model name, and stepping

The article says (in part):

Athlon XP/Geode NX (AMD)
vendor_id : AuthenticAMD
cpu family : 6
model : 6
stepping : 2
CHOST="i686-pc-linux-gnu"
CFLAGS="-march=athlon-xp -O2 -pipe -fomit-frame-pointer"
CXXFLAGS="${CFLAGS}"

Athlon MP (AMD)
vendor_id : AuthenticAMD
cpu family : 6
model : 10
model name : AMD Sempron(tm)
stepping : 0
CHOST="i686-pc-linux-gnu"
CFLAGS="-march=athlon-mp -O2 -pipe -fomit-frame-pointer"
CXXFLAGS="${CFLAGS}"

As you can see, the model name clearly says its an XP, which is what I was expecting, given that it is a single CPU mobo, and that I was told it was an XP when I got it. However, the model and stepping numbers say that it's an MP. Which is it and which -march=flag should I use, athlon-xp or athlon-mp Does it matter that much? My recollection is that the only significant difference between the XP and MP CPU's is that the MP was intended for use in multi-cpu boards, and the XP was intended for single CPU applications. Given that this is a single CPU box, does it make much difference if I possibly have a "wrong" CPU in it?

BTW, I intend to go back and update the article when I get an answer.

Thanks,

Gooserider

[dalek]

I have a known Athlon XP 2500+ bought form newegg. This is what I get:

Code: Select all

root@smoker / # cat /proc/cpuinfo
processor       : 0
vendor_id       : AuthenticAMD
cpu family      : 6
model           : 10
model name      : AMD Athlon(tm) XP 2500+
stepping        : 0
cpu MHz         : 1830.130
cache size      : 512 KB
fdiv_bug        : no
hlt_bug         : no
f00f_bug        : no
coma_bug        : no
fpu             : yes
fpu_exception   : yes
cpuid level     : 1
wp              : yes
flags           : fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 mmx fxsr sse syscall mmxext 3dnowext 3dnow up ts
bogomips        : 3665.02

root@smoker / # 

Maybe that will help you some.

:

[Gooserider]

Thanks Dalek

That is exactly what I needed to know. I will go edit the article (assuming it lets me) to reflect that XP's can also show the info that you and I have.

Gooserider

[dalek]

I was hoping that would help since I new for sure what I had. I bought mine brand new and in pretty wrappings.

Glad that helped though.

[BaYGoN]

Hi all.

In the next few days, i'll be installing Gentoo 2006.1 on my machine (see my signature) and i want some feedback on my CFLAGS of choice.
This time i want a pure 32 bit enviroment (i have a amd64 enviroment running right now), and after much reading, i picked up these flags:

Code: Select all

CFLAGS="-march=athlon-xp -msse2 -msse3 -O2 -pipe -fomit-frame-pointer"
CHOST="i686-pc-linux-gnu"
CXXFLAGS="${CFLAGS}"

As you can see, i'm very conservative about the flags, simply because i want a good balance between a fast and stable system.
My processor has SSE3 support, i'm very sure of that.

I've seen posts of people with Athlons 64 compiling a 32 enviroment, using athlon64 as -march, and i'm confused about one think:

(From GCC 4.1.1 manual)

k8, opteron, athlon64, athlon-fx
AMD K8 core based CPUs with x86-64 instruction set support. (This supersets MMX, SSE, SSE2, 3dNOW!, enhanced 3dNOW! and 64-bit instruction set extensions.)

Using athlon64 as march implies/activates the 64-bit instruction set extensions, as per the GCC Manual.

Wouldnt be wiser to choose athlon-xp as march and activating sse2 and sse3, as i intend to do?
Or i'm getting it wrong and won't make any difference anyway, given the CHOST being set to i686?

I appreciate any comments on that.

[edit]
Another question: I'll be upgrading my processor with an Athlon X2 before the end of this year. If i use the CFLAGS i have chosen, will i have any sort of problems?

Thanks.

[letto]

Hi ,
I've just recently build two gentoo instalation on the same machine. They were meant to be identical so that i could compare the different CFLAGS and how the size increased with different options.

To make a clone of the firs gentoo install i just copied the /var/lib/portage/world to the second instalation and after updating and emerging kde and koffice did this

Code: Select all

emerge --noreplace `cat world.new`

As the systems were synced at the same time and i copied /etc/portage over, this gave me two almost identical systems.
After a little extra uninstalling the difference was minimal.

Code: Select all

letto@localhost ~/devel/genscript $ diff pkg_list.O2 pkg_list.O3
442c442
< sys-apps/dbus-0.62-r1
---
> sys-apps/dbus-0.62
letto@localhost ~/devel/genscript $                                                                                      

The important difference was in make.conf. For the first system:

Code: Select all

letto@localhost ~/devel/genscript $ cat /etc/make.conf|grep -e FLAGS -e ACCEPT
CFLAGS="-march=k8 -O3 -pipe -msse3 -fprefetch-loop-arrays -fmerge-all-constants -funroll-loops -fpeel-loops -ftracer -ftree-loop-ivcanon -funswitch-loops -ftree-vectorize -fivopts -ftree-loop-im -ftree-loop-linear"
CXXFLAGS="${CFLAGS}"
ACCEPT_KEYWORDS="~amd64"
LDFLAGS="-Wl,-O1 -Wl,--sort-common -Wl,-z,now"
letto@localhost ~/devel/genscript $

It's important to note that the majority of packages were build without -fivopts here.
And for the second system:

Code: Select all

(chr) letto@localhost ~/devel/genscript $ cat /etc/make.conf|grep -e FLAGS -e ACCEPT
CFLAGS="-march=k8 -O2 -pipe -msse3 -ftree-vectorize"
CXXFLAGS="${CFLAGS}"
ACCEPT_KEYWORDS="~amd64"
LDFLAGS="-Wl,-O1"
(chr) letto@localhost ~/devel/genscript $

Now to measure the difference in size of the binaries i made a bash script:

Code: Select all

letto@localhost ~/devel/genscript $ cat bin_size.sh
#!/bin/bash

TOTAL=0
LIBS=0

for i in `seq 1 20`;
do
        DIR=`echo $PATH|cut -d: -f$i`
        if [[ -e $DIR ]] ; then
                TOTAL=`expr $TOTAL + $(du $DIR |cut -f1)`
                du -h $DIR
        fi
done

echo binaries $TOTAL

echo
for i in `cat /etc/ld.so.conf |grep -v \#`
do
        #echo $i
        if [[ -e $i ]] ; then
                LIBS=`expr $LIBS + $( du -s $i|cut -f1) `
                du -hs $i
        fi
done

TOTAL=`expr $TOTAL + $LIBS`
echo libs $LIBS
echo $TOTAL

letto@localhost ~/devel/genscript $

and the result of running the script

Code: Select all

letto@localhost ~/devel/genscript $ ./bin_size.sh
0       /usr/local/bin
191M    /usr/bin
11M     /bin
0       /opt/bin
581K    /usr/x86_64-pc-linux-gnu/gcc-bin/4.1.1
2.0M    /opt/blackdown-jdk-1.4.2.03/bin
904K    /opt/blackdown-jdk-1.4.2.03/jre/bin
56M     /usr/kde/3.5/bin
7.6M    /usr/qt/3/bin
13M     /usr/games/bin
binaries 286729

0       /usr/local/lib
8.4M    //usr/lib32/opengl/nvidia/lib
8.3M    //usr/lib64/opengl/nvidia/lib
0       /lib
0       /usr/lib
0       /usr/local/lib
22M     /lib64
817M    /usr/lib64
0       /usr/local/lib64
2.3M    /lib32
44M     /usr/lib32
0       /usr/local/lib32
0       /usr/x86_64-pc-linux-gnu/lib
80M     /usr/lib/gcc/x86_64-pc-linux-gnu/4.1.1
5.0M    /usr/lib/gcc/x86_64-pc-linux-gnu/4.1.1/32
969K    /usr/lib64/nspr
4.7M    /usr/lib64/nss
0       /usr/lib
14M     /opt/blackdown-jdk-1.4.2.03/jre/lib/amd64/
44K     /opt/blackdown-jdk-1.4.2.03/jre/lib/amd64/native_threads/
6.5M    /opt/blackdown-jdk-1.4.2.03/jre/lib/amd64/server/
7.3M    /emul/linux/x86/usr/qt/3/lib
0       /usr/kde/3.5/lib
184M    /usr/kde/3.5//lib64
0       /usr/qt/3/lib
19M     /usr/qt/3/lib64
2.8M    /emul/linux/x86/lib
43M     /emul/linux/x86/usr/lib
2.8M    /emul/linux/x86/lib
43M     /emul/linux/x86/usr/lib
5.8M    /usr/games/lib
809K    /usr/lib64/libstdc++-v3/
libs 1347635
1634364
letto@localhost ~/devel/genscript $

and on the second system

Code: Select all

(chr) letto@localhost ~/devel/genscript $ ./bin_size.sh
0       /usr/local/bin
160M    /usr/bin
9.0M    /bin
0       /opt/bin
581K    /usr/x86_64-pc-linux-gnu/gcc-bin/4.1.1
2.0M    /opt/blackdown-jdk-1.4.2.03/bin
904K    /opt/blackdown-jdk-1.4.2.03/jre/bin
49M     /usr/kde/3.5/bin
7.0M    /usr/qt/3/bin
9.9M    /usr/games/bin
binaries 243084

0       /usr/local/lib
8.4M    /usr/lib32/opengl/nvidia/lib
8.3M    /usr/lib64/opengl/nvidia/lib
0       /lib
0       /usr/lib
0       /usr/local/lib
21M     /lib64
804M    /usr/lib64
0       /usr/local/lib64
2.3M    /lib32
44M     /usr/lib32
0       /usr/local/lib32
0       /usr/x86_64-pc-linux-gnu/lib
80M     /usr/lib/gcc/x86_64-pc-linux-gnu/4.1.1
5.0M    /usr/lib/gcc/x86_64-pc-linux-gnu/4.1.1/32
877K    /usr/lib64/nspr
4.7M    /usr/lib64/nss
0       /usr/lib
14M     /opt/blackdown-jdk-1.4.2.03/jre/lib/amd64/
44K     /opt/blackdown-jdk-1.4.2.03/jre/lib/amd64/native_threads/
6.5M    /opt/blackdown-jdk-1.4.2.03/jre/lib/amd64/server/
7.3M    /emul/linux/x86/usr/qt/3/lib
0       /usr/kde/3.5/lib
171M    /usr/kde/3.5//lib64
0       /usr/qt/3/lib
19M     /usr/qt/3/lib64
2.8M    /emul/linux/x86/lib
43M     /emul/linux/x86/usr/lib
2.8M    /emul/linux/x86/lib
43M     /emul/linux/x86/usr/lib
4.6M    /usr/games/lib
817K    /usr/lib64/libstdc++-v3/
libs 1318182
1561266
(chr) letto@localhost ~/devel/genscript $

Seems that for the executables i got a 17.95% size increase and for the libraries a 2.23% while having an overall 4.68% increase. It's important to note however that this includes some -bin packages and packages that filter CFLAGS out.
Anyway now i wanted to measure the difference between the files in /usr/bin and this is what i did :

Code: Select all

letto@localhost ~/devel/genscript $ cat bin_dir.sh
#!/bin/bash

TOTAL=0

for i in `ls /usr/bin`
do
        B1=$(ls -l /mnt/disk/usr/bin/$i|cut -f3 -dt|cut -f1 -dS)
        B2=$(ls -l /usr/bin/$i|cut -f3 -dt|cut -f1 -dS)
        echo $(./proc $B1 $B2) $i $B1 $B2
done
letto@localhost ~/devel/genscript $

proc is a little c program that divides the numbers and outputs a float number
Here is an interesting section of the output of ./bin_dir.sh |sort :

Code: Select all

100.9 % paperconf 9448 9368
101.2 % giffiltr 7168 7080
101.3 % gnome-cups-manager 108624 107232
101.6 % osx 77720 76472
101.9 % epsilon_thumb_test 7560 7416
102.9 % pat2dat 144296 140248
106.3 % dat2c 27552 25920
107.5 % cupstestdsc 11136 10360
108.0 % autoconf 27 25
108.0 % autoheader 27 25
108.0 % autom4te 27 25
108.0 % autoreconf 27 25
108.0 % autoscan 27 25
108.0 % autoupdate 27 25
108.0 % ifnames 27 25
110.1 % ecore-config 2585 2347
110.8 % textconv 15576 14056
111.0 % ospent 19576 17632
19.3 % pgmtexture 19576 101464
42.8 % pnmconvol 32176 75136
43.0 % pbmupc 12200 28360
47.5 % cabextract 52792 111200
47.5 % dc 40512 85232
49.3 % ppmtoilbm 37000 75032
49.7 % pgmmorphconv 8264 16632
50.9 % cxpm 23728 46592
52.9 % pbmpage 8016 15144
53.0 % ilbmtoppm 33400 63064
53.2 % funzip 21880 41096
53.5 % tcrequant 76920 143776
53.7 % rletopaint 37368 69608
54.0 % pcxtoppm 11640 21560
54.3 % bc 77864 143416
54.6 % cameratopam 93624 171424
55.7 % grotty 82320 147888
56.6 % eyuvtoppm 8776 15512
56.6 % gpgv 325728 574984
57.1 % unrar 175488 307408
57.3 % grolj4 82360 143832
57.6 % unarj 21040 36504
57.8 % fant 44872 77680
57.8 % rleldmap 44880 77688
57.9 % mpeg2encode 78880 136232
58.0 % pbmtoln03 8096 13952
58.2 % grn 60952 104696
58.3 % pre-grohtml 74400 127680
58.4 % grodvi 86456 147928
58.5 % pgmmedian 10296 17592
58.9 % ppmrough 10048 17072
59.3 % pnmtoddif 12408 20928
59.3 % rletorla 41120 69392
59.3 % texindex 23976 40400
59.4 % ppmtopcx 13272 22336
59.6 % gif2epsn 17528 29392
59.6 % pnmtofiasco 145744 244512
59.7 % grops 128808 215752
59.8 % install-info 29136 48744
59.9 % msggrep 97944 163544
60.1 % refer 123304 205256
60.4 % mpeg3cat 162440 268944
60.4 % mpeg3dump 162440 268944
60.7 % bggen 11816 19480
60.9 % pktopbm 11760 19296
61.0 % mpeg3toc 166504 273008
61.0 % soxmix 309416 506992
61.1 % sox 309096 506032
61.2 % ppmtogif 15512 25328
61.3 % gpgsplit 41288 67408
61.5 % gif2x11 19624 31888
61.5 % rlesplice 40744 66256
61.8 % grolbp 95664 154768
61.8 % picttoppm 48592 78624

pgmtexture seems to be a very good example of a program that is affected gcc optimizations. There are also a lot of programs that are at 100% wich probably means that they are not affected by the optimizations. The "program density" tends to be higher as the number aproaches 100%

[ann0]

Here's mine, what do you think?

Code: Select all

CFLAGS="-march=pentium2 -mtune=pentium2 -O2 -pipe -ftracer -fweb -fforce-addr -fomit-frame-pointer 
-fexpensive-optimizations -frerun-cse-after-loop -frerun-loop-opt -falign-functions=4 -Qn -fno-ident"

[golding]

I have bought an AMD Athlon64-3800+ 939 and re-used my existing install, AMD Athlon XP 2500+ Barton, and everything is running fast and really sweet.

What I would like to do, as I don't want to use a chrooted env for the odd 32bit app, is run my AMD64 in a 32bit enhanced environment.

I would like to know, providing any body knows of course, if I left my CHOST ("i686-pc-linux-gnu") as is and ONLY changed my CFLAGS to reflect the CPU, will this run 32bit compile or will the 64 bit CFLAGS automatically produce a 64bit compiled system?

Current config:

Code: Select all

CFLAGS="-march=athlon-xp -mtune=athlon-xp -O2 -mmmx -m3dnow -msse -pipe -fomit-frame-pointer -momit-leaf-frame-pointer -fno-ident"
CHOST="i686-pc-linux-gnu"

Proposed config:

Code: Select all

CFLAGS="-march=athlon64 -mtune=athlon64 -O3 -mmmx -mmmxext -m3dnow -m3dnowext -msse -msse2 -pipe -fomit-frame-pointer -momit-leaf-frame-pointer -fno-ident"
CHOST="i686-pc-linux-gnu"

[cybermonkey]

Hi all,

I'm building a new system this weekend, A celeron D 2.53ghz.

I don't need 64 bitness and have mainly used AMD in the past so a few pointers from the CFLAGS god's
that dwell in this thread would be appreciated So.......

What would be the best cflags to use?

Excuse me if this has already been posted, But this is one hell of a thread to get through

[cybermonkey]

Hi all,

I'm building a new system this weekend, A celeron D 2.53ghz.

I don't need 64 bitness and have mainly used AMD in the past so a few pointers from the CFLAGS god's
that dwell in this thread would be appreciated So.......

What would be the best cflags to use?

Excuse me if this has already been posted, But this is one hell of a thread to get through

[SOLVED]:

Found this on the wiki:-

Code: Select all

CHOST="i686-pc-linux-gnu"
CFLAGS="-march=prescott -O2 -pipe -fomit-frame-pointer"
CXXFLAGS="${CFLAGS}"

[/SOLVED]

[cybermonkey]

I have bought an AMD Athlon64-3800+ 939 and re-used my existing install, AMD Athlon XP 2500+ Barton, and everything is running fast and really sweet.

What I would like to do, as I don't want to use a chrooted env for the odd 32bit app, is run my AMD64 in a 32bit enhanced environment.

I would like to know, providing any body knows of course, if I left my CHOST ("i686-pc-linux-gnu") as is and ONLY changed my CFLAGS to reflect the CPU, will this run 32bit compile or will the 64 bit CFLAGS automatically produce a 64bit compiled system?

Current config:

Code: Select all

CFLAGS="-march=athlon-xp -mtune=athlon-xp -O2 -mmmx -m3dnow -msse -pipe -fomit-frame-pointer -momit-leaf-frame-pointer -fno-ident"
CHOST="i686-pc-linux-gnu"

Proposed config:

Code: Select all

CFLAGS="-march=athlon64 -mtune=athlon64 -O3 -mmmx -mmmxext -m3dnow -m3dnowext -msse -msse2 -pipe -fomit-frame-pointer -momit-leaf-frame-pointer -fno-ident"
CHOST="i686-pc-linux-gnu"

Also from the gentoo-wiki:

Code: Select all

Athlon 64 X2 (AMD) 
vendor_id: AuthenticAMD
cpu family: 15
model: 43
stepping: 1 or 2
model name: AMD Athlon(tm) X2 Dual Core Processor



32 bit 

CHOST="i686-pc-linux-gnu"
CFLAGS="-march=athlon64 -O2 -pipe -fomit-frame-pointer"
CXXFLAGS="${CFLAGS}"
64 bit 

CHOST="x86_64-pc-linux-gnu"
CFLAGS="-march=athlon64 -O2 -pipe"
CXXFLAGS="${CFLAGS}"
If you have a newer Athlon64("Venice" or "San Diego")/Athlon64-X2("Manchester" or "Toledo") (check for pni in cat /proc/cpuinfo) you can also add -msse3 to your CFLAGS to enable SSE3 support. Any chip using the 90nm process except for the "Winchester" class Athlon64 supports SSE3. 

Also bear in mind that all Athlon64 X2 and Opteron 165, 170, 175, 180, and 185 processors are dual-core CPUs so make sure that you use MAKEOPTS="-j3" in your make.conf. This doesn't affect the code but might reduce compile times. 

[golding]

Thankyou cybermonkey, I must be going blind in my dotage, didn't even see that when I checked the wiki

[ConiKost]

Could I use -mfpmath=sse on a AMD Athlon 64 X2 4600+ ?

Or not a good idea?

[golding]

Could I use -mfpmath=sse on a AMD Athlon 64 X2 4600+ ?

Or not a good idea?

My understanding is that it doesn't really improve anything as it is assumed by the compiler from the -march=athlon64 flag.
But it can, sometimes, break compilation on a couple of apps, like OpenOffice.org from source (not the bin) & mozilla for example.

BTW, don't forget to change MAKEOPTS (in make.conf) to "-j3" for the extra, internal, cpu, i.e.

Code: Select all

MAKEOPTS="-j3"

[baigsabeeh]

Is -O3 safe and what are the benefits of it? The wiki makes it a little too vague.

[baigsabeeh]

What is -fPIC for and how is it dangerous? I'm using it right now to complement the hashstyles and I've compiled half of my system without a hitch. Whereas when I didn't have it, it would stop after every 10 or 15 packages.

[-JeaN-]

Is -O3 safe and what are the benefits of it? The wiki makes it a little too vague.

It seems -O3 makes bigger binaries, and implies a very little performance gain ( 2% or 3% ).
Therefore, if you really want to have that little extra, do it, but be aware that the bigger binaries is only suitable if your CPU cache and installed RAM is big enough, and even further more it takes lot more compile time.

I have an Intel Core 2 Duo T7200 on my laptop, with 2GB of RAM, and still I didn't find it important enough over -O2.

Here are my parameters on my make.conf :

Code: Select all

CHOST="x86_64-pc-linux-gnu"
CFLAGS="-O2 -pipe -march=nocona -fomit-frame-pointer -fno-ident"
CXXFLAGS="${CFLAGS}"
MAKEOPTS="-j3"

And I'm very happy with it, I think it's very stable and compiles fast.

Btw, the -f-inline-hidden-visibles ( don't know the exact syntax ) was breaking my compiles, specially for Koffice. Since I removed it, everything went smooth, so I suggest you remove it too.

And the -momit-leaf-frame-pointer shouldn't be used with -fomit-frame-pointer.

Good luck and have fun

[-JeaN-]

What is -fPIC for and how is it dangerous? I'm using it right now to complement the hashstyles and I've compiled half of my system without a hitch. Whereas when I didn't have it, it would stop after every 10 or 15 packages.

From the "man gcc" :

-fpic
If supported for the target machines, generate position-independent code, suitable for use in a shared library.
-fPIC
If supported for the target machine, emit position-independent code, suitable for dynamic linking, even if branches need large displacements.

So basically, it's important in some cases, but overall you can get happily rid of it, since I suppose most ebuilds which will need it will commit this flag when needed.

I suggest you take it out of your CFLAGS and only include it on package base, if needed. I don't think it improves any of the performance.

Cheers

[VeliuX]

Hello, i own an athlon64 3800+ x2, on an a8n32-sli delux mobo, i was wondering if there were any dual core amd64 cflags other than "O2 -march=athlon64 -pipe" That i could possibly use? I mean flags that would increase some speed but not hurt stability in a major way?
I set "Makeopts="-j3"" cause i have a dual core proc. Also when i put -march=athlon64.....now will theses flags that are included in the athlon64 part be beneficial to the athlon64 dual cores or x2s, or is there another -march setting i could set? Thanks

[#2]

This is what I have so far.

Code: Select all

CFLAGS="-O2 -march=athlon-xp -mno-tls-direct-seg-refs -m32 -pipe -fomit-frame-pointer -momit-leaf-frame-pointer" 

USE="... sse 3dnow" 

LDFLAGS="-Wl,-O1 -Wl,--enable-new-dtags -Wl,--sort-common -s" 

LINGUAS="en_GB"

Should I add:
-fforce-addr
-falign-functions=16 -fno-align-jumps -fno-align-labels -falign-loops=4
-fprefetch-loop-arrays

Also I have read. http://gcc.gnu.org/gcc-4.0/changes.html#visibility and http://gcc.gnu.org/wiki/Visibility I'm not sure how to implement this.
I this correct? CXXFLAGS="${CFLAGS} -fvisibility=hidden -fvisibility-inlines-hidden"

My CHOST is i586. :/ I know it's a same about that with Sabayon.

Thanks in advance.

[Arfrever]

Code: Select all

CFLAGS="(...) -fomit-frame-pointer -momit-leaf-frame-pointer"

"-fomit-frame-pointer" includes all actions caused by "-momit-leaf-frame-pointer", so remove "-momit-leaf-frame-pointer".

Code: Select all

LDFLAGS="(...) -Wl,--enable-new-dtags (...)"

"--enable-new-dtags" is default in Gentoo, so remove it.

-fforce-addr

"spidermonkey" will fail to build.

-fprefetch-loop-arrays

Yes.

Also I have read. http://gcc.gnu.org/gcc-4.0/changes.html#visibility and http://gcc.gnu.org/wiki/Visibility I'm not sure how to implement this.
I this correct? CXXFLAGS="${CFLAGS} -fvisibility=hidden -fvisibility-inlines-hidden"

Yes, but use only "-fvisibility-inlines-hidden". Some packages will fail to build even with alone "-fvisibility-inlines-hidden" (Maybe KOffice?).

Sveikinu
Arfrever

[#2]

Ačiū Arfrever.

What do you think of the -falign-functions=16 -fno-align-jumps -fno-align-labels -falign-loops=4 for the Athlon XP?

Any thing else you would add?

[Zvezda]

Here's a semi C-flag related question:

What is a good (common) program that uses lots and lots of floating point instructions to test the benefits/disadvantages of -mfpmath=sse vs 387 on?

[edit]

This is an Athlon-xp system compiled only with -march=athlon-xp -O2 -fomit-frame-pointer -pipe

[mv]

[...] -fPIC [...]
So basically, it's important in some cases, but overall you can get happily rid of it, since I suppose most ebuilds which will need it will commit this flag when needed.

I suggest you take it out of your CFLAGS and only include it on package base, if needed. I don't think it improves any of the performance.

I have very bad experience not using it in many cases: Many packages will then not be able to link with -Wl,-z,now, because some libraries were compiled without this flag (i.e. my experience is that a huge amount of ebuilds is buggy w.r.t. -fPIC). What is worse is that linking without -Wl,-z,now might even be a security issue. So I would suggest to use these flags whenever possible and only omit them if something breaks (despite the annoying warnings of portage). It does probably not improve the performance (on x86 actually quite the opposite, on amd64 it might improve the performance slightly), but it improves the security.