GCC 4.3.0 testing!

Message

Belliash · Post by **Belliash** » Sun Nov 19, 2006 2:47 pm

A new forward propagation pass on RTL was added. The new pass replaces several slower transformations, resulting in compile-time improvements as well as better code generation in some cases.

What do you think about it?
I know, it's in early, stage 1, but maybe someone wanna test it?
It's in portage.

Voltago · Post by **Voltago** » Sun Nov 19, 2006 2:50 pm

Morpheouss wrote:I know, it's in early, stage 1, but maybe someone wanna test it?

Nah. Gonna wait until ecj is merged into the trunk as java compiler.

Belliash · Post by **Belliash** » Sun Nov 19, 2006 2:57 pm

ecj? what's that?

Voltago · Post by **Voltago** » Sun Nov 19, 2006 3:00 pm

Eclipse Compiler for Java. It will replace gcj and bring support for JDK-5.
EDIT: For JDK-5 language features (generics, annotations), that is.

Phenax · Post by **Phenax** » Sun Nov 19, 2006 4:23 pm

I was going to try it on the day of the release, but it didn't even compile here. 4.2.0 compiles fine, and fails on very few packages (Most notably Perl).

Belliash · Post by **Belliash** » Sun Nov 19, 2006 4:34 pm

are there any patches for 4.2 or 4.3?

rhill · Post by **rhill** » Sun Nov 19, 2006 5:46 pm

no, there's not much point in writing patches for an alpha compiler that still has most of the projects yet to be merged. 4.2 is relatively stable but we don't support it and we haven't migrated the patchset and probably won't until release.

f0rk · Post by **f0rk** » Sun Nov 19, 2006 10:16 pm

Gcc 4.3.0-alpha20061111 is c00l !

I'm using gentoo with these unstable flags:
LDFLAGSHASHSTYLE=" -Wl,--hash-style=both"
CFLAGS="-Os -march=pentium3 -frename-registers -fweb -pipe -fomit-frame-pointer -funit-at-a-time -freorder-blocks -fno-ident -freorder-blocks-and-partition -fgcse-sm -fgcse-las -fgcse-after-reload -fmerge-all-constants -combine"
CXXFLAGS="${CFLAGS} -fvisibility-inlines-hidden"
LDFLAGS="-Wl,-O1${LDFLAGSHASHSTYLE}"

Earlier when I used gcc-4.1.1 whole system compiled well but except python.
It doesn't emerge with stable gcc-4.1.1 (segfault)
Now I use gcc-4.3 as a default compiler.
With new experimental gcc python compiled and works fine !!!

And compile-time rather reduced...

Now I migrated from original qt-3.3.6-r4 to nxsty's QT.
And it was compiled faster than with gcc 4.1.1
(I won about 40 min !)

Belliash · Post by **Belliash** » Mon Nov 20, 2006 3:28 pm

f0rk wrote:Gcc 4.3.0-alpha20061111 is c00l !
I'm using gentoo with these unstable flags:
LDFLAGSHASHSTYLE=" -Wl,--hash-style=both"
CFLAGS="-Os -march=pentium3 -frename-registers -fweb -pipe -fomit-frame-pointer -funit-at-a-time -freorder-blocks -fno-ident -freorder-blocks-and-partition -fgcse-sm -fgcse-las -fgcse-after-reload -fmerge-all-constants -combine"
CXXFLAGS="${CFLAGS} -fvisibility-inlines-hidden"
LDFLAGS="-Wl,-O1${LDFLAGSHASHSTYLE}"

Whole system compiled well but except python.
It doesn't emerge with stable gcc-4.1.1 (segfault)
With new experimental gcc it compiled and works fine !!!

And compile-time rather reduced...

Now I migrated from original qt-3.3.6-r4 to nxsty's QT.
And it was compiled faster than with gcc 4.1.1
(I won about 40 min !)

I don't understand something..
1) Have you successfull compiled python using GCC 4.3?
2) You have compiled QT 40minutes faster, haven't you?

Another question: Why don't you upgrade ebuild for gcc 4.3.0?

f0rk · Post by **f0rk** » Mon Nov 20, 2006 4:14 pm

Sorry. Updated. See above!

1) Yes, I compiled successfull python using GCC 4.3
Earlier I couldn't do it with gcc 4.1.1 (using unsafe flags)
2) Yes! You understood right!
genlop -t qt says:
//----------- gcc 4.1.1-r2:
Thu Nov 16 15:46:03 2006 >>> x11-libs/qt-3.3.6-r4
merge time: 2 hours, 1 minute and 29 seconds.
//----------- gcc 4.3.0-alpha:
Mon Nov 20 04:21:53 2006 >>> x11-libs/qt-3.3.6-r4
merge time: 1 hour, 17 minutes and 30 seconds.

I have pentium-3 750 Hz on my laptop.
I'm using glibc-2.5 and binutils-2.17.50.0.5 with hashstyle.

Belliash · Post by **Belliash** » Mon Nov 20, 2006 4:18 pm

questions to f0rk:
1) Have you successful recompiled all packages in your system using GCC 4.3?
2) Have you compiled firefox? Does it work faster? Doesn't it fault?
3) Have you compiled KDE?
4) Do you use beryl? How does it work after compiltion using GCC 4.3?
5) Have you compiled OpenOffice? If QT compiled 40minutes faster, maybe OOo will compile much faster too?
6) Why don't you use tree vectorization? It had to be repaired in GCC 4.2
7) Do you use prelink?

Do you use x86 ~x86, amd64 or maybe ~amd64?
9) Wchi python version have you compiled?

Thanks!

f0rk · Post by **f0rk** » Mon Nov 20, 2006 5:02 pm

Morpheouss wrote:questions to f0rk:
1) Have you successful recompiled all packages in your system using GCC 4.3?
2) Have you compiled firefox? Does it work faster? Doesn't it fault?
3) Have you compiled KDE?
4) Do you use beryl? How does it work after compiltion using GCC 4.3?
5) Have you compiled OpenOffice? If QT compiled 40minutes faster, maybe OOo will compile much faster too?
6) Why don't you use tree vectorization? It had to be repaired in GCC 4.2
7) Do you use prelink?
Do you use x86 ~x86, amd64 or maybe ~amd64?
9) Wchi python version have you compiled?

Thanks!

1) No. I didn't try to do it.
But I think it wouldn't be a problem!

IMHO: it is more useful than 4.1 and rather stable at the same time...
2) I use opera instead firefox. But now I'm emering seamonkey (mozilla like)...
And there is no faults ...
I suggest firefox also will be compiled well.
3) Some part of KDE:
kde-base/kdelibs-3.5.5
kde-base/libkonq-3.5.5
kde-base/ksmserver-3.5.5
kde-base/kcontrol-3.5.5
kde-base/kdesktop-3.5.5
kde-base/kicker-3.5.5
kde-base/konqueror-3.5.5
and some themes...
All packages were compiled fine!
I'm using kde 3.5.5 + KIP.
4) No, unfortunatelly I don't use it, because I have only ATI Rage 128 8 mb on my laptop...

5) No. I'm using openoffice-bin. Good idea about attempt to compile it! May be I'll try it later...
6) What is "tree vectorization" ??? I didn't hear about it.
7) Yes. I do.
By the way: patched glibc + patched qt + patched kde + prelink + no2-sources help me to reduce time of kde startup. It takes about 4-5 sec !!!

(It takes about 10-13 seconds on my friend's desktop with intel conroe core duo

)

I use: x86 + ~x86 for some packages, but packages.keywords becomes rather big and I'm pondering about migrating to ~x86 only.
9) python-2.4.3-r4

Belliash · Post by **Belliash** » Mon Nov 20, 2006 5:16 pm

CFLAGS="-ftree-vectorize"

it speeds up bigsoftware like KDE, OpenOffice, Firefox imho should speedup too.

It would be great if you try to emerge -e world

and it end successful...

Also, you could try to recompile OOo then...

I you had no trubles with it, i would compile new 64-bit Gentoo using GCC 4.3

using this make.conf:

Code: Select all

# MAKE.CONF made by Morpheouss!
# 19/11/2006

# Architecture specific
CHOST="x86_64-pc-linux-gnu"
CFLAGS="-march=athlon64 -mtune=athlon64 -O2 -s -pipe -fomit-frame-pointer -mfpmath=sse,387 -msse -msse2 -msse3 -mmmx -m3dnow -ffast-math -ftracer -finline-limit=1200 -fno-ident -fforce-addr -fpeel-loops -fprefetch-loop-arrays -funroll-loops -funswitch-loops -ftree-vectorize -fprefetch-loop-arrays -frerun-cse-after-loop -momit-leaf-frame-pointer -maccumulate-outgoing-args -mno-align-stringops -minline-all-stringops -DNDEBUG -DG_DISABLE_ASSERT -DNO_DEBUG"
CXXFLAGS="${CFLAGS} -fpermissive -fno-enforce-eh-specs -fvisibility-inlines-hidden"
LDFLAGS="-Wl,-O1 -Wl,--enable-new-dtags -Wl,--as-needed -Wl,--hash-style=both -Wl,--as-needed -Bdirect -s"

PORTAGE_STRIP_FLAGS="--strip-all --discard-all -R .comment -R .note -R .note.ABI-tag"

# Locale Settings
LINGUAS="en pl"
LANG="en_US"
LC_ALL="en_US"

# Mirrors
GENTOO_MIRRORS="ftp://ftp.belnet.be/mirror/rsync.gentoo.org/gentoo/ http://gentoo.zie.pg.gda.pl http://gentoo.po.opole.pl http://gentoo.prz.rzeszow.pl ftp://mirror.icis.pcz.pl/gentoo/ ftp://gentoo.po.opole.pl"

# Portage Specific Settings
ACCEPT_KEYWORDS="amd64 ~amd64"
ALSA_CARDS="emu10k1"
AUTOCLEAN="yes"
CCACHE_DIR="/var/tmp/ccache"
CCACHE_SIZE="2G"
CLEAN_DELAY="0"
CONFIG_PROTECT="/etc"
DISTDIR="/usr/portage/distfiles"
EMERGE_WARNING_DELAY="5"
FEATURES="autoconfig digest distlocks nodoc parallel-fetch sandbox sfperms -metadata-transfer"
INPUT_DEVICES="mouse keyboard"
MAKEOPTS="-j2 -s"
PORT_LOGDIR=/var/log/portage
PORTAGE_TMPDIR=/var/tmp
PORTDIR_OVERLAY="/usr/local/portage-morph /usr/local/portage-xgl"
SYNC="rsync://rsync.europe.gentoo.org/gentoo-portage"
VIDEO_CARDS="nv nvidia"

# USE Flags
USE="3dnow 3dnowext aac alsa avi bitmap-fonts cairo cdr cdrom cups divx4linux dvd dvdr dvdread flash gif glitz gtk2 hal jpeg jpeg2k kde kdeenablefinal mmx mmxext mp3 mpeg nptl nptlonly nvidia ogg oggvorbis opengl pic png qt qt3 speedup sse sse2 sse3 tga tiff truetype type1-fonts udev usb userlocales v4l vorbis wmf X xml2 xv xvid -arts -fortran -gdbm -gnome -gpm -ipv6 -motif -ncurses -nls -oss -perl -python -samba -slang -spell -unicode -xmms"

That's the reason why i ask you so much.
I wanna try GCC 4.3 but befere that i must know if i'll be able to compile base system packages... (eg coreutils

).

So, .... Could you try emerge -e world using GCC 4.3.0?

f0rk · Post by **f0rk** » Mon Nov 20, 2006 5:35 pm

1. About flags:
DON'T USE -ffast-math !!! NEVER !!!

It will break your system.
I'm sure! I tried to use it, but every time I have to abandon it.
What differences are between my:
LDFLAGS="-Wl,-O1 -Wl,--hash-style=both"
and your:
LDFLAGS="-Wl,-O1 -Wl,--enable-new-dtags -Wl,--as-needed -Wl,--hash-style=both -Wl,--as-needed -Bdirect -s"
Are you sure that hashstyle doesn't include -Bdirect ?
2. OK, I will try to do emerge -e world...
But may be not today...

By the way, my email: 4ndrey[at]gmail.com, icq 4064065 .

Belliash · Post by **Belliash** » Mon Nov 20, 2006 5:42 pm

1) --as-needed speeds up big software -Bdirect speeds up dynamic library loading...

2) -ffast-math is filtered out by most packages... From that, which i use only ZSH and HAL have troubles with it

I have tested it long time

By the way, i had to remove it, but i've forgotten

Thanks!

3) i'll be really glad if you try to reemerge world

Big thanks!

I'm waiting for infos

playfool · Post by **playfool** » Mon Nov 20, 2006 5:48 pm

the best thing I've heard about -ffast-math was from rac "If I wanted math done wrong I'd ask my cat"

Belliash · Post by **Belliash** » Mon Nov 20, 2006 5:51 pm

playfool wrote:the best thing I've heard about -ffast-math was from rac "If I wanted math done wrong I'd ask my cat"

OT: Maybe should i ask mine cat?

I've already removed this flag

cruzki · Post by **cruzki** » Mon Nov 20, 2006 6:13 pm

LDFLAGS="-Wl,-O1 -Wl,--enable-new-dtags -Wl,--as-needed -Wl,--hash-style=both -Wl,--as-needed -Bdirect -s"

are you sure that this LDFLAGS will work?

1º what convination of glibc and binutils will you use?¿?¿?¿? (and what overlay will you use?)

2º this line has two --as-needed and miss one Wl, I think this is more correct.

Code: Select all

LDFLAGS="-Wl,-O1 -Wl,--enable-new-dtags -Wl,--as-needed -Wl,--hash-style=both -Wl, -Bdirect -s"

maybe I tree to campile a similar system this weekend... depends of yours experiencies (or how mach I will be boring this weekend ... )

Belliash · Post by **Belliash** » Mon Nov 20, 2006 6:28 pm

Yes, there shouldn't be --as-needed twice

You're right.

But i think this is the best one:

Code: Select all

LDFLAGS="-Wl,-O1,--enable-new-dtags,--sort-common,--hash-style=both,--as-needed -Bdirect -s"

Using this make.conf, i have compiled successful system using GCC 4.0.1, 4.0.2, 4.1.1
Never tried using GCC 4.0.3, 4.2.0, 4.3.0

Also new is --hash-style=both LDFLAG

But i think it should break system in connection with --as-needed and -Bdirect

I'll back to nxsty's binutils 2.17.50.0.5 which have patches for, which allows me to use -Bdirect.
Also wanna back to nxsty's glibc 2.5, becouse it allows me to use -Bdirect and have patches which improves performance on amd64 (mostly patched libmath).

Changes, which i've made to make.conf:
* added --hash-style=both
* removed -ffash-math

On listed above version of GCC everything worked for me good on amd64...
On x86 i have a lot of troubles when i tried to use this flags

Conclusion: AMD64 is easier to better optimise than x86

vipernicus · Post by **vipernicus** » Mon Nov 20, 2006 8:10 pm

Morpheouss wrote:

Code: Select all

# MAKE.CONF made by Morpheouss!
# 19/11/2006

# Architecture specific
CHOST="x86_64-pc-linux-gnu"
CFLAGS="-march=athlon64 -mtune=athlon64 -O2 -s -pipe -fomit-frame-pointer -mfpmath=sse,387 -msse -msse2 -msse3 -mmmx -m3dnow -ffast-math -ftracer -finline-limit=1200 -fno-ident -fforce-addr -fpeel-loops -fprefetch-loop-arrays -funroll-loops -funswitch-loops -ftree-vectorize -fprefetch-loop-arrays -frerun-cse-after-loop -momit-leaf-frame-pointer -maccumulate-outgoing-args -mno-align-stringops -minline-all-stringops -DNDEBUG -DG_DISABLE_ASSERT -DNO_DEBUG"
CXXFLAGS="${CFLAGS} -fpermissive -fno-enforce-eh-specs -fvisibility-inlines-hidden"
LDFLAGS="-Wl,-O1 -Wl,--enable-new-dtags -Wl,--as-needed -Wl,--hash-style=both -Wl,--as-needed -Bdirect -s"

PORTAGE_STRIP_FLAGS="--strip-all --discard-all -R .comment -R .note -R .note.ABI-tag"

# Locale Settings
LINGUAS="en pl"
LANG="en_US"
LC_ALL="en_US"

# Mirrors
GENTOO_MIRRORS="ftp://ftp.belnet.be/mirror/rsync.gentoo.org/gentoo/ http://gentoo.zie.pg.gda.pl http://gentoo.po.opole.pl http://gentoo.prz.rzeszow.pl ftp://mirror.icis.pcz.pl/gentoo/ ftp://gentoo.po.opole.pl"

# Portage Specific Settings
ACCEPT_KEYWORDS="amd64 ~amd64"
ALSA_CARDS="emu10k1"
AUTOCLEAN="yes"
CCACHE_DIR="/var/tmp/ccache"
CCACHE_SIZE="2G"
CLEAN_DELAY="0"
CONFIG_PROTECT="/etc"
DISTDIR="/usr/portage/distfiles"
EMERGE_WARNING_DELAY="5"
FEATURES="autoconfig digest distlocks nodoc parallel-fetch sandbox sfperms -metadata-transfer"
INPUT_DEVICES="mouse keyboard"
MAKEOPTS="-j2 -s"
PORT_LOGDIR=/var/log/portage
PORTAGE_TMPDIR=/var/tmp
PORTDIR_OVERLAY="/usr/local/portage-morph /usr/local/portage-xgl"
SYNC="rsync://rsync.europe.gentoo.org/gentoo-portage"
VIDEO_CARDS="nv nvidia"

# USE Flags
USE="3dnow 3dnowext aac alsa avi bitmap-fonts cairo cdr cdrom cups divx4linux dvd dvdr dvdread flash gif glitz gtk2 hal jpeg jpeg2k kde kdeenablefinal mmx mmxext mp3 mpeg nptl nptlonly nvidia ogg oggvorbis opengl pic png qt qt3 speedup sse sse2 sse3 tga tiff truetype type1-fonts udev usb userlocales v4l vorbis wmf X xml2 xv xvid -arts -fortran -gdbm -gnome -gpm -ipv6 -motif -ncurses -nls -oss -perl -python -samba -slang -spell -unicode -xmms"

There are some big flaws in your make.conf.

Code: Select all

# Architecture specific
CHOST="x86_64-pc-linux-gnu"
CFLAGS="-march=athlon64 -mtune=athlon64 -O2 -s -pipe -fomit-frame-pointer -mfpmath=sse,387 -msse -msse2 -msse3 -mmmx -m3dnow -ffast-math -ftracer -finline-limit=1200 -fno-ident -fforce-addr -fpeel-loops -fprefetch-loop-arrays -funroll-loops -funswitch-loops -ftree-vectorize -fprefetch-loop-arrays -frerun-cse-after-loop -momit-leaf-frame-pointer -maccumulate-outgoing-args -mno-align-stringops -minline-all-stringops -DNDEBUG -DG_DISABLE_ASSERT -DNO_DEBUG"
CXXFLAGS="${CFLAGS} -fpermissive -fno-enforce-eh-specs -fvisibility-inlines-hidden"
LDFLAGS="-Wl,-O1 -Wl,--enable-new-dtags -Wl,--as-needed -Wl,--hash-style=both -Wl,--as-needed -Bdirect -s"

Code: Select all

Redundant flags:
CFLAGS: -mtune=athlon64, -s, -msse -msse2 -msse3 -mmmx -m3dnow, -momit-leaf-frame-pointer
CXXFLAGS: -fvisibility-inlines-hidden, -fpermissive
LDFLAGS: --enable-new-dtags, --hash-style=both, -Bdirect, -s

When you use -march you do not want to use -mtune as well. Gentoo by default strips. You -march defines -msse -msse2 -mmmx -m3dnow. When you use -momit-leaf-frame-pointer in conjunction with -fomit-frame-pointer, -fomit-frame-pointer is ignored and only -momit-leaf-frame-pointer is used, and if you use only -fomit-frame-pointer, then the leaf-frame-pointer is omitted as well (I believe). -fvisibility-lines-hidden is already enabled in the ebuilds for the apps that can benefit from it. -fpermissive is ridiculous to use, because it downgrades diagnostics so that you can compile bad code. --enable-new-dtags is enabled by Gentoo by default, you just won't see it in the compile. --hash-style=both will not work in conjunction with -Bdirect, one of these two flags is being ignored.

Code: Select all

Flags that will slow you down:
CFLAGS: -mfpmath=sse,387, -fpeel-loops, -fprefetch-loop-arrays, -funroll-loops, -funswitch-loops, -frerun-cse-after-loop 
CXXFLAGS: 
LDFLAGS:

-mfpmath=sse,387 will only slow you down. All of the rest will bloat your binaries to the point of sluggishness. Whenever your binaries are larger, they use up more RAM, more CPU cache, and also cause more cache misses. Cache misses increase latency. So imagine if you will, you created a binary that uses more resources and cause cache misses, so you not only have extra overhead, but also increased latencies!

Code: Select all

Flags with stability problems:
CFLAGS: -ffast-math, -ftree-vectorize
CXXFLAGS: -fno-enforce-eh-specs
LDFLAGS:

-ffast-math is known to break just about everything, and normally whenever it can benefit a program, Gentoo Devs use it in the ebuild. -ftree-vectorize is known to cause stability issues with gtk programs and zlib. -fno-enforce-eh-specs decreases size, but also make applications not work the way they are supposed to, for instance when a program should have crashed, it just keeps running.

Code: Select all

My recommendations:
CHOST="x86_64-pc-linux-gnu"
CFLAGS="-march=athlon64 -O2 -pipe -fomit-frame-pointer -fno-ident"
CXXFLAGS="${CFLAGS}"
LDFLAGS="-Wl,-O1 -Wl,--as-needed -Wl,-z,now -Wl,--hash-style=gnu"

And in many situations -Os might be better than -O2.

f0rk · Post by **f0rk** » Mon Nov 20, 2006 10:34 pm

Morpheouss, don't use gcc-4.3-alpha to build your system!
glibc-2.5 doesn't want to compile...

vipernicus, thanks!
I simplify my CFLAGS and changed my LDFLAGS

cruzki · Post by **cruzki** » Tue Nov 21, 2006 9:27 am

@vipernicus

It's safe to use hash-style=gnu?¿?¿?

What are your recomendation to an x86 machine?¿?¿?, what do you think about the CFLAGS that use nesl247 in the evolution-emission guide?¿?¿

Code: Select all

#nesl247's CFLAGS in the evo-emi 4.0
CHOST="i686-pc-linux-gnu"
CFLAGS="-O2 -march=pentium4 -pipe -fomit-frame-pointer"
CFLAGS="${CFLAGS} -fno-ident -frename-registers -combine -fprefetch-loop-arrays -fweb -fgcse-after-reload"
CXXFLAGS="${CFLAGS} -fvisibility-inlines-hidden"
LDFLAGS="-Wl,-O1 -Wl,--as-needed -Wl,-znow"
LDFLAGS="${LDFLAGS} -Wl,--hash-style=both"
MAKEOPTS="-j2"

PS: where I will learnd about the new options in GCC-4.2 or 4.3 for a non experiment programer like kernel-for-newbies??¿?

Belliash · Post by **Belliash** » Tue Nov 21, 2006 12:11 pm

@vipernicus:

1) i've already removed -ffast-math
2) -ftree-vectorize speeds up, but it's broken since 4.0.2 i think, they told, they'll repair it in 4.2.0
3) Some packages filters flags out... So if it filter -momit-frame-pointer, then package can use -fomit-frame-pointer
4) -mfpmath=sse,387 slows down on alder AMD CPUs. AMD's SSE implementation was terrible... I know, that, i had Athlon XP t-bird. But it speeds up on Intel's CPUS and on K8. 387 is old, and slow FPU. This flag forces it to use newer and better SSE... If some package cannot use SSE, then it uses 387.
5) -msse3 => GCC does not support SSE3 and SSE4, you can add -msse3 -msse4, but it does nothing

Look at KInfoCenter/OpenGL: I see there "GeForce 6600/PCI/SSE2/3DNOW!" GCC can't enable SSE3!

So you're wrong in some points

@f0rk:
1) Maybe we have to make new ebuild for newer version of GCC 4.3.0? Eg dated at 20061121 ?

2) Have you tried GCC 4.2.0? Any problems? Recompiled world? How much faster is it?

vipernicus · Post by **vipernicus** » Tue Nov 21, 2006 1:03 pm

cruzki wrote:@vipernicus

It's safe to use hash-style=gnu?¿?¿?

What are your recomendation to an x86 machine?¿?¿?, what do you think about the CFLAGS that use nesl247 in the evolution-emission guide?¿?¿
Code: Select all
#nesl247's CFLAGS in the evo-emi 4.0
CHOST="i686-pc-linux-gnu"
CFLAGS="-O2 -march=pentium4 -pipe -fomit-frame-pointer"
CFLAGS="${CFLAGS} -fno-ident -frename-registers -combine -fprefetch-loop-arrays -fweb -fgcse-after-reload"
CXXFLAGS="${CFLAGS} -fvisibility-inlines-hidden"
LDFLAGS="-Wl,-O1 -Wl,--as-needed -Wl,-znow"
LDFLAGS="${LDFLAGS} -Wl,--hash-style=both"
MAKEOPTS="-j2"
PS: where I will learnd about the new options in GCC-4.2 or 4.3 for a non experiment programer like kernel-for-newbies??¿?

evo-mission flags are great, if you want support, keep those flags.

vipernicus · Post by **vipernicus** » Tue Nov 21, 2006 1:12 pm

Morpheouss wrote:@vipernicus:

1) i've already removed -ffast-math
2) -ftree-vectorize speeds up, but it's broken since 4.0.2 i think, they told, they'll repair it in 4.2.0
3) Some packages filters flags out... So if it filter -momit-frame-pointer, then package can use -fomit-frame-pointer
4) -mfpmath=sse,387 slows down on alder AMD CPUs. AMD's SSE implementation was terrible... I know, that, i had Athlon XP t-bird. But it speeds up on Intel's CPUS and on K8. 387 is old, and slow FPU. This flag forces it to use newer and better SSE... If some package cannot use SSE, then it uses 387.
5) -msse3 => GCC does not support SSE3 and SSE4, you can add -msse3 -msse4, but it does nothing Look at KInfoCenter/OpenGL: I see there "GeForce 6600/PCI/SSE2/3DNOW!" GCC can't enable SSE3!

1) good job
2) hope you removed it too
3) And every package that doesn't filter either ONLY uses -momit-leaf-frame-pointer, but if you really want a worthwhile optimization, you'd want them to only use -fomit-frame-pointer. It's like having the choice between throwing 50 pounds off the truck or 150.
4) -march=athlon64 uses -mfpmath=sse by default
5) redundant - 1 a : exceeding what is necessary or normal : SUPERFLUOUS b : characterized by or containing an excess; specifically : using more words than necessary

GCC 4.3.0 testing!

GCC 4.3.0 testing!

Re: GCC 4.3.0 testing!