View previous topic :: View next topic |
Author |
Message |
rhill Retired Dev
Joined: 22 Oct 2004 Posts: 1629 Location: sk.ca
|
|
Back to top |
|
|
neuron Advocate
Joined: 28 May 2002 Posts: 2371
|
Posted: Tue Jul 12, 2005 3:06 pm Post subject: |
|
|
guess we'll start seeing actual performance increases over the 3.4 series now, anyone have any numbers comparing gcc4 and 4.1? |
|
Back to top |
|
|
scoobydu Veteran
Joined: 16 Feb 2003 Posts: 1076 Location: 'Mind the Gap'
|
Posted: Tue Jul 12, 2005 4:31 pm Post subject: |
|
|
Whats the general opinion on support for gcc4.x?
Is it ready for a non techie to try?
I tried early 4 but without much success .. _________________ Tyan Tiger K8W, 2xOpteron 240,Powerbook5,6 15" 1.5g, Macbook Black 2g, Mac Mini 1g, Ipod P60g.
| Linux - From a windows user perspective| |
|
Back to top |
|
|
fuoco Guru
Joined: 23 May 2004 Posts: 386 Location: Israel
|
Posted: Tue Jul 12, 2005 4:48 pm Post subject: |
|
|
ProPolice is now integrated into vanilla gcc ? I suppose it works the same way with the same flags as with the patch ?
What's the difference between ProPolice and FORTIFY_SOURCE - or whatever it's called ? |
|
Back to top |
|
|
aspenhydrowave n00b
Joined: 08 Dec 2004 Posts: 41
|
Posted: Tue Jul 12, 2005 6:53 pm Post subject: |
|
|
neuron wrote: | guess we'll start seeing actual performance increases over the 3.4 series now, anyone have any numbers comparing gcc4 and 4.1? |
Running some numerical simulation, I've seen the best performance with gcc 3.4. 4.0 was 30% slower and 4.1.0-20050709 is 15% slower. So I guess we're waiting for 4.2 to actually see performance gains... sigh
of course this is just one isolated example; it may be faster in many other areas. |
|
Back to top |
|
|
neuron Advocate
Joined: 28 May 2002 Posts: 2371
|
Posted: Tue Jul 12, 2005 6:55 pm Post subject: |
|
|
aspenhydrowave wrote: | neuron wrote: | guess we'll start seeing actual performance increases over the 3.4 series now, anyone have any numbers comparing gcc4 and 4.1? |
Running some numerical simulation, I've seen the best performance with gcc 3.4. 4.0 was 30% slower and 4.1.0-20050709 is 15% slower. So I guess we're waiting for 4.2 to actually see performance gains... sigh
of course this is just one isolated example; it may be faster in many other areas. |
well have you tried current 4.1? There are a few things in the links in the first post that should have performance improvements. |
|
Back to top |
|
|
luisfelipe Guru
Joined: 09 Apr 2005 Posts: 377
|
Posted: Tue Jul 12, 2005 11:05 pm Post subject: |
|
|
Can I install gcc 4.1 on another slot and use it ??
From what I recall this wasn't possible before due to some conflicts with glibc I guess.
If I want to use gcc 4.1, can I just install it and keep changing it with gcc-config ? |
|
Back to top |
|
|
neuron Advocate
Joined: 28 May 2002 Posts: 2371
|
Posted: Tue Jul 12, 2005 11:56 pm Post subject: |
|
|
luisfelipe wrote: | Can I install gcc 4.1 on another slot and use it ??
From what I recall this wasn't possible before due to some conflicts with glibc I guess.
If I want to use gcc 4.1, can I just install it and keep changing it with gcc-config ? |
emerge -p on gcc4 shows NS for me (new, slot), so it seems so.
I sure hope so because that's what I'm planning to do, just to test it a bit performance wise |
|
Back to top |
|
|
luisfelipe Guru
Joined: 09 Apr 2005 Posts: 377
|
Posted: Wed Jul 13, 2005 12:19 am Post subject: |
|
|
I tought I read it somewhere that it wasn't working, but the GCC 4.x threads are so big I can't
find it again
Well, just finishing compiling kdelibs, gonna try to install that later and see how it goes. |
|
Back to top |
|
|
luisfelipe Guru
Joined: 09 Apr 2005 Posts: 377
|
Posted: Wed Jul 13, 2005 7:11 am Post subject: |
|
|
Well, I'm trying to install it, and I'm getting this error :
/var/tmp/portage/gcc-4.1.0_beta20050709/work/gcc-4.1-20050709/gcc/tree-ssa-live.c: In function 'build_tree_conflict_graph':
/var/tmp/portage/gcc-4.1.0_beta20050709/work/gcc-4.1-20050709/gcc/tree-ssa-live.c:1463: error: corrupted profile info: number of executions for edge 97-121 thought to be 6280
/var/tmp/portage/gcc-4.1.0_beta20050709/work/gcc-4.1-20050709/gcc/tree-ssa-live.c:1463: error: corrupted profile info: number of executions for edge 97-98 thought to be -3705
/var/tmp/portage/gcc-4.1.0_beta20050709/work/gcc-4.1-20050709/gcc/tree-ssa-live.c:1463: error: corrupted profile info: number of executions for edge 147-143 thought to be -2308
/var/tmp/portage/gcc-4.1.0_beta20050709/work/gcc-4.1-20050709/gcc/tree-ssa-live.c:1463: error: corrupted profile info: number of executions for edge 147-148 thought to be 3592
/var/tmp/portage/gcc-4.1.0_beta20050709/work/gcc-4.1-20050709/gcc/tree-ssa-live.c:1463: error: corrupted profile info: number of executions for edge 149-147 thought to be -1291
/var/tmp/portage/gcc-4.1.0_beta20050709/work/gcc-4.1-20050709/gcc/tree-ssa-live.c:1463: error: corrupted profile info: number of executions for edge 149-150 thought to be 9503
make[2]: *** [tree-ssa-live.o] Error 1
make[2]: Leaving directory `/var/tmp/portage/gcc-4.1.0_beta20050709/work/build/gcc'
make[1]: *** [stagefeedback_build] Error 2
make[1]: Leaving directory `/var/tmp/portage/gcc-4.1.0_beta20050709/work/build/gcc'
make: *** [profiledbootstrap] Error 2
!!! ERROR: sys-devel/gcc-4.1.0_beta20050709 failed.
!!! Function gcc_do_make, Line 1211, Exitcode 2 |
|
Back to top |
|
|
thersites Tux's lil' helper
Joined: 12 Oct 2004 Posts: 101 Location: South West China
|
Posted: Wed Jul 13, 2005 11:58 am Post subject: |
|
|
I've been running a gcc-4.1 built system since 20050502.
With binutils-2.16.91.0.1 mozilla-firefox, the last package that was broken with gcc-4.1, sucessfully builds and runs.
However, still, if binutils is built with -O1, the gcc build fails.
So my CFLAGS ...
Code: | CFLAGS="-march=i686 -ftree-ccp -ftree-dce -ftree-dominator-opts -ftree-dse -ftree-ter -ftree-lrs -ftree-sra -ftree-copyrename -ftree-fre -ftree-ch -pipe" |
which is about half of -O1.
With these I can emerge -e world - no kde/gnome! - with only libvorbis patched/hacked.
linux-2.6.13-rc* boots when built with gcc-4.1, but the iptable_filter module will not load and intel8x0/snd_ac97_codec give me no sound. I also have this sound problem when 2.6.13-rc*'s are built with gcc-4.0.1. linux-2.6.12 doesn't boot when built with 4.1 |
|
Back to top |
|
|
scoobydu Veteran
Joined: 16 Feb 2003 Posts: 1076 Location: 'Mind the Gap'
|
Posted: Wed Jul 13, 2005 12:32 pm Post subject: |
|
|
thersites wrote: | I've been running a gcc-4.1 built system since 20050502.
With binutils-2.16.91.0.1 mozilla-firefox, the last package that was broken with gcc-4.1, sucessfully builds and runs.
However, still, if binutils is built with -O1, the gcc build fails.
So my CFLAGS ...
Code: | CFLAGS="-march=i686 -ftree-ccp -ftree-dce -ftree-dominator-opts -ftree-dse -ftree-ter -ftree-lrs -ftree-sra -ftree-copyrename -ftree-fre -ftree-ch -pipe" |
which is about half of -O1.
With these I can emerge -e world - no kde/gnome! - with only libvorbis patched/hacked.
linux-2.6.13-rc* boots when built with gcc-4.1, but the iptable_filter module will not load and intel8x0/snd_ac97_codec give me no sound. I also have this sound problem when 2.6.13-rc*'s are built with gcc-4.0.1. linux-2.6.12 doesn't boot when built with 4.1 |
Thx for the info _________________ Tyan Tiger K8W, 2xOpteron 240,Powerbook5,6 15" 1.5g, Macbook Black 2g, Mac Mini 1g, Ipod P60g.
| Linux - From a windows user perspective| |
|
Back to top |
|
|
Gestalt73 n00b
Joined: 02 Jun 2004 Posts: 29
|
Posted: Wed Jul 13, 2005 9:11 pm Post subject: Compile error for 4.1.0_beta20050709 [SOLVED] |
|
|
UPDATE: Relaxed my CFLAGS to "-01 -march=athlon-xp -pipe" and was able to compete the compile.
The compile failed with my "agressive" flags: "-O3 -march=athlon-xp -fomit-frame-pointer -pipe -ffast-math"
Still a bit early for those I suppose.
Tried twice compiling with GCC 4.0.1 and GCC 3.4.3 and the compile fails here:
Code: | : In function 'loop_delete_branch_edge':
/var/tmp/portage/gcc-4.1.0_beta20050709/work/gcc-4.1-20050709/gcc/cfgloopmanip.c:791: error: coverage mismatch for function 'loop_delete_branch_edge' while reading counter 'arcs'
/var/tmp/portage/gcc-4.1.0_beta20050709/work/gcc-4.1-20050709/gcc/cfgloopmanip.c:791: error: number of counters is 14 instead of 15
make[2]: *** [cfgloopmanip.o] Error 1
make[2]: Leaving directory `/var/tmp/portage/gcc-4.1.0_beta20050709/work/build/gcc'
make[1]: *** [stagefeedback_build] Error 2
make[1]: Leaving directory `/var/tmp/portage/gcc-4.1.0_beta20050709/work/build/gcc'
make: *** [profiledbootstrap] Error 2
!!! ERROR: sys-devel/gcc-4.1.0_beta20050709 failed.
!!! Function gcc_do_make, Line 1211, Exitcode 2
!!! emake failed with profiledbootstrap
!!! If you need support, post the topmost build error, NOT this status message.
|
UPDATED: to show more of error message
UPDATED: to show successful compile
Alan |
|
Back to top |
|
|
rhill Retired Dev
Joined: 22 Oct 2004 Posts: 1629 Location: sk.ca
|
|
Back to top |
|
|
m0p Apprentice
Joined: 20 Jun 2005 Posts: 205 Location: en_GB
|
Posted: Sun Jul 17, 2005 3:55 am Post subject: |
|
|
I am going to install gcc 4.1 to test certain apps, so, before I emerged it, I ran it with -p to see what dependencies it would need. It seems it needs a new binutils. Would I need to use this new binutils for executing apps generated by the new gcc, or only for compiling them, or, is it neccesary to use the new one at all? |
|
Back to top |
|
|
voidengineer n00b
Joined: 11 Feb 2005 Posts: 54
|
Posted: Sun Jul 17, 2005 11:13 am Post subject: Re: Compile error for 4.1.0_beta20050709 [SOLVED] |
|
|
Gestalt73 wrote: | UPDATE: Relaxed my CFLAGS to "-01 -march=athlon-xp -pipe" and was able to compete the compile.
The compile failed with my "agressive" flags: "-O3 -march=athlon-xp -fomit-frame-pointer -pipe -ffast-math"
Still a bit early for those I suppose.
Alan |
It is ALWAYS going to be to "early" or unwise to compile with -ffast-math. This compiler flag really should only be explicitly defined on a per file basis like in makefiles for example. |
|
Back to top |
|
|
rhill Retired Dev
Joined: 22 Oct 2004 Posts: 1629 Location: sk.ca
|
Posted: Sun Jul 17, 2005 12:08 pm Post subject: |
|
|
m0p wrote: | I am going to install gcc 4.1 to test certain apps, so, before I emerged it, I ran it with -p to see what dependencies it would need. It seems it needs a new binutils. Would I need to use this new binutils for executing apps generated by the new gcc, or only for compiling them, or, is it neccesary to use the new one at all? |
I don't think it's neccessary but it doesn't hurt. Just turn on the multislot USE flag and install the new binutils in parallel to the old one and you can play with both. Any version above 2.16 will work, and you should probably use the newer binutils to build gcc itself. I haven't tried building gcc with unmasked versions of binutils or glibc for a quite a while so I don't know if it even works. _________________ by design, by neglect
for a fact or just for effect |
|
Back to top |
|
|
pgrdsl Tux's lil' helper
Joined: 29 Aug 2002 Posts: 93 Location: Southampton, UK
|
Posted: Sun Jul 17, 2005 3:27 pm Post subject: Is it just me, or are the installation directories screwed? |
|
|
I'm trying to work out what has changed between the 4.0* family and 4.1* with regards to the ebuild/installation process.
My installation of 4.1-beta20050709 has some things installed in /usr/lib/gcc/i686-pc-linux-gnu/4.1.0-beta20050709 and some in /usr/lib/gcc/i686-pc-linux-gnu/4.1.0.
Unfortunately, the stuff in 4.1.0 are things like: fixincludes'd headers and libstdc++, and gcc-config is helpfully suggesting that binaries and libraries are in the other one.
Yes, I can frig things to get things to work, but this split installation directory doesn't seem right to me.
I raised this on the gcc bugs and got an immediate response of: --enable-version-specific-runtime-libs overrides all the other directories listed in configure.
Which is fair enough, and what the makefiles say - so why is --enable-version-specific-runtime-libs there?
Anyway, I'm intrigued: how come other people's 4.1 installs are working? _________________ pihl |
|
Back to top |
|
|
m0p Apprentice
Joined: 20 Jun 2005 Posts: 205 Location: en_GB
|
Posted: Sun Jul 17, 2005 4:03 pm Post subject: |
|
|
According to "emerge -p =gcc-4.1.0_beta20050709", the new binutils won't install into another slot, it just upgrades, is there any bugs in the newer binutils?
EDIT: And which binutils should I install out of this list:
Code: | sys-devel/binutils-2.16.1
sys-devel/binutils-2.16-r1
sys-devel/binutils-2.16.91.0.1
sys-devel/binutils-2.16.90.0.3 |
|
|
Back to top |
|
|
rhill Retired Dev
Joined: 22 Oct 2004 Posts: 1629 Location: sk.ca
|
Posted: Sun Jul 17, 2005 4:29 pm Post subject: |
|
|
did you add multislot to your USE flags in make.conf? if you did, try just emerge -pv binutils and see if that comes up as a slotted install.
any of those is fine. i have them all installed. ;P 2.16.1 is the latest testing version and fixes some bugs that were in 2.16, so maybe that would be the one to start with. _________________ by design, by neglect
for a fact or just for effect |
|
Back to top |
|
|
m0p Apprentice
Joined: 20 Jun 2005 Posts: 205 Location: en_GB
|
Posted: Sun Jul 17, 2005 4:32 pm Post subject: |
|
|
Yes, I have multislot, but emerge -pv still doesnt say about creating a new slot. |
|
Back to top |
|
|
m0p Apprentice
Joined: 20 Jun 2005 Posts: 205 Location: en_GB
|
Posted: Tue Jul 19, 2005 6:48 am Post subject: |
|
|
Oh, got it installed now. I have run gcc-config as root, and ran env-update && source /etc/profile but gcc --version still says 3.4.4, but if I run as root it says 4.1.0. What is wrong?
Edit: Woops, selected wrong profile! disregard this post. |
|
Back to top |
|
|
irf2003 Veteran
Joined: 10 Sep 2003 Posts: 1078
|
Posted: Tue Jul 19, 2005 3:50 pm Post subject: Re: GCC 4.1 |
|
|
@everyone, don't play with gcc-4.1 on a live system unless u do not mind it borking, play with it on a spare partition or in a chrooted environment. And while you are playing with it, take it easy with your compiler flags, -O1 in gcc-4.1 land, is beaucoup optimizations, which may or may not work, at the time of writing.
happy Gentooing |
|
Back to top |
|
|
m0p Apprentice
Joined: 20 Jun 2005 Posts: 205 Location: en_GB
|
Posted: Tue Jul 19, 2005 4:06 pm Post subject: |
|
|
I am using gcc 4.1, for external use. I am not compiling anything in portage with it. For example, I am using it to build the nightly sources of Firefox and Thunderbird, but with no good results! Here is the output of the firefox startup script after compiling the latest nightly with 4.1, and a stripped down set of CFLAGS:
Code: | joe@mopbox ~/Firefox_Source/mozilla/dist/bin $ ./firefox
./run-mozilla.sh: line 131: 2456 Segmentation fault "$prog" ${1+"$@"} |
Pretty crappy, eh? |
|
Back to top |
|
|
neuron Advocate
Joined: 28 May 2002 Posts: 2371
|
Posted: Tue Jul 19, 2005 4:09 pm Post subject: |
|
|
well nobody said it was stable yet, just that it's getting there |
|
Back to top |
|
|
|