View previous topic :: View next topic |
Author |
Message |
z-buffer n00b
Joined: 19 May 2005 Posts: 11 Location: Zweibrücken, Germany
|
Posted: Thu May 19, 2005 3:09 pm Post subject: Successful 64-bit installation on a EM64T Xeon |
|
|
Hi all,
I just wanted to let you know: We succeeded in installing a 64-bit-Gentoo (2005.0) on a Dell PowerEdge 1800 machine (Intel Xeon 2.8 GHz EM64T, 2GB RAM, ...). We booted off the AMD64 disc and used the 'gentoo' kernel, as 'gentoo-em64t' is not available by now (Bug). These are our CHOST and [C|CXX]FLAGS settings:
Code: |
CHOST="x86_64-pc-linux-gnu"
CFLAGS="-march=nocona -O3 -pipe -fomit-frame-pointer -mmmx -msse3 -mfpmath=sse"
CXXFLAGS="${CFLAGS}"
|
As the CPU is capable of HyperThreading, we also used the optimization.
The installation went smoothly, no compilation errors at all, the machine is running fast. If there are any questions concerning our install, feel free to post them here.
Cheers,
Alex. |
|
Back to top |
|
|
azlan Guru
Joined: 11 Nov 2002 Posts: 381 Location: Seattle WA USA
|
Posted: Fri May 20, 2005 3:45 am Post subject: |
|
|
do you have any extra, of those laying around??? _________________ there is no crazy, only violet and non violet.. |
|
Back to top |
|
|
z-buffer n00b
Joined: 19 May 2005 Posts: 11 Location: Zweibrücken, Germany
|
Posted: Fri May 20, 2005 6:55 am Post subject: |
|
|
I'm afraid not ... we have to work with these from time to time. |
|
Back to top |
|
|
mr.ed Tux's lil' helper
Joined: 30 Nov 2004 Posts: 75
|
Posted: Fri May 20, 2005 10:27 am Post subject: |
|
|
Hey,
Congrats on your first install on a em64t system, but i've got a question for u:
Quote: | -mmmx -msse3 -mfpmath=sse |
According to the gcc manual and some googling, these options are allready implied with setting -march=nocona. Is this true, or did u use these just to be sure?
I got a similar setup as yours and never used these flags, so i would like to know if setting these flags would do any good. My understanding was that setting these flags is not necessary...
I'll post my setup info later on, maybe this thread will get sticky or something to be of use to others...
Greetz, Mr-Ed
P.s.: Is there a way to check if these implied flags are indeed used? _________________ (Ex-Pornstar) |
|
Back to top |
|
|
z-buffer n00b
Joined: 19 May 2005 Posts: 11 Location: Zweibrücken, Germany
|
Posted: Fri May 20, 2005 11:46 am Post subject: |
|
|
Hi,
I would love to see a listing of all implications and dependencies of all those potential CFLAGS one could use. We actually used ours the mentioned way to make sure the processor extensions are really used for floating point math and the like. We derived the flags from several forum posts and this Intel presentation on EM64T performance.
The march setting usually *should* imply those settings, yes. If it does, I don't know. I have no idea how to check whether the flags are being used in the installed system, too.
mr.ed wrote: | According to the gcc manual and some googling, these options are allready implied with setting -march=nocona. |
Could you please post some of those example URLs for the settings and implications? I desperately tried to find something like that and never did. Thanks.
Greetings,
-Alex. |
|
Back to top |
|
|
mr.ed Tux's lil' helper
Joined: 30 Nov 2004 Posts: 75
|
Posted: Fri May 20, 2005 2:12 pm Post subject: |
|
|
Hey Alex,
As u requested, here's the link to the gcc manual :
The guide mentions that nocona implies these flags and also mention that sse is automatically used when CHOST="x86_64-pc-linux-gnu".
And here are the cflags (including some other usefull info) i used to compile my system:
Quote: | Portage 2.0.51.22-r1 (default-linux/amd64/2005.0, gcc-3.4.3-20050110, glibc-2.3.5-r0, 2.6.11-reiser4-r9 x86_64)
=================================================================
System uname: 2.6.11-reiser4-r9 x86_64 Intel(R) Xeon(TM) CPU 3.20GHz
Gentoo Base System version 1.6.12
ACCEPT_KEYWORDS="amd64 ~amd64"
CFLAGS="-O3 -march=nocona -mtune=nocona -fomit-frame-pointer -maccumulate-outgoing-args -minline-all-stringops -ffast-math -ftracer -fweb -pipe -fforce-addr -frename-registers -funit-at-a-time -fmove-all-movables -fmerge-all-constants -freduce-all-givs -fbranch-target-load-optimize2 -funroll-all-loops"
CHOST="x86_64-pc-linux-gnu"
CXXFLAGS=" CFLAGS -fvisibility-inlines-hidden"
LDFLAGS="-Wl,-O1 -Wl,--enable-new-dtags -Wl,--sort-common -s"
MAKEOPTS="-j5" |
_________________ (Ex-Pornstar) |
|
Back to top |
|
|
z-buffer n00b
Joined: 19 May 2005 Posts: 11 Location: Zweibrücken, Germany
|
Posted: Wed May 25, 2005 6:33 am Post subject: |
|
|
Thanks for the link, Ed. That is indeed a good starting point for optimization flags, although the dependencies are still not perfectly clear ...
As far as I understand the listing, the only compiler flags implicitly set on our machines are:
- -mtune=nocona by setting -march=nocona
- -mfpmath=sse, -msse, & -msse2 by using a x86_64 compiler
GCC Manual (-mtune=cpu-type) wrote: | Tune to cpu-type everything applicable about the generated code, except for the ABI and the set of available instructions. |
In my understanding, the mentioned "set of available instructions" are extensions like MMX and SSE(x), right? Which would imply, these are not automatically used when tuning to the desired architecture and you would have to activate them yourself with the -mmmx and -msse(x) flags. The -march setting as well as the -mtune setting do not set any extension flags automatically, but the x86_64 compiler does so with the flags listed above. So IMHO you still have to set the -mmmx and -msse3 manually if you want to use them ... I will continue to do so.
HTH,
Alex. |
|
Back to top |
|
|
piramid n00b
Joined: 18 Feb 2004 Posts: 37
|
Posted: Wed May 25, 2005 6:40 am Post subject: |
|
|
I have interest in know if Apache2+PHP5 and Qmail+Vpopmail runs fine in EM64T. Somebody has tried it? |
|
Back to top |
|
|
z-buffer n00b
Joined: 19 May 2005 Posts: 11 Location: Zweibrücken, Germany
|
Posted: Wed May 25, 2005 6:49 am Post subject: |
|
|
We are running Apache2 with mod_PHP 4 on our EM64T Xeon. That works without any problems. |
|
Back to top |
|
|
piramid n00b
Joined: 18 Feb 2004 Posts: 37
|
Posted: Wed May 25, 2005 7:04 am Post subject: |
|
|
z-buffer wrote: | We are running Apache2 with mod_PHP 4 on our EM64T Xeon. That works without any problems. |
Oh Good, now I only need to know if Qmail+Vpopmail combination runs fine. |
|
Back to top |
|
|
mr.ed Tux's lil' helper
Joined: 30 Nov 2004 Posts: 75
|
Posted: Wed May 25, 2005 7:23 am Post subject: |
|
|
Hey Alex,
Sorry for the delay in reaction to your post. As for your findings, the way i looked at it (and i do agree the manual is a bit vague on the subject) setting -march=nocona is enough to enable the instructionset specific for this type of processor.
If u look up the nocona flag in the manual it states:
Quote: | nocona: Improved version of Intel Pentium4 CPU with 64-bit extensions, MMX, SSE, SSE2 and SSE3 instruction set support. |
If u combine this with the flag -march:
Quote: | -march=cpu-type: Generate instructions for the machine type cpu-type. The choices for cpu-type are the same as for -mtune. Moreover, specifying -march=cpu-type implies -mtune=cpu-type. |
The way i look at the last statement is that -march enables the instructionset specific for the processor chosen, because it says: " generate instructions for the machine type cpu-type". This means to me that -mmmx and -msse3 are enabled, because they are cpu-specific for the nocona!
The -mtune flag is just a little brother to -march then IMO, but i allways could be wrong... But setting -march should be enough! Look at it this way: why specify your processor, if you still have to fill in the instructionsets for that processor?
But there's no harm in being sure about it and it does no harm at all when specified. So if u want to be sure those instructions are enabled, just use them!
Hope this has clarified it a bit.
Greetz, Mr-Ed _________________ (Ex-Pornstar) |
|
Back to top |
|
|
z-buffer n00b
Joined: 19 May 2005 Posts: 11 Location: Zweibrücken, Germany
|
Posted: Wed May 25, 2005 8:39 am Post subject: |
|
|
According to the GCC manual, the -mtune flag tunes for everything "except for ... the set of available instructions". The -march flag on the other hand "generates instructions" for the specified cpu. So, -march creates code that *requires* the given processor at runtime, whereas -mtune merely optimizes code to the given processor while still being portable to others.
After even more googling I still haven't found any serious link that explicitly documents the redundancy of the -mmmx and -msse3 flags when using -march=nocona. Even the Intel presentation on optimization of EM64T systems I mentioned in an earlier post, uses the -msse3 flag. So unless otherwise explicitly proven, I will continue to use -mmmx and -msse3. |
|
Back to top |
|
|
piramid n00b
Joined: 18 Feb 2004 Posts: 37
|
Posted: Wed May 25, 2005 9:24 am Post subject: |
|
|
I'm trying to boot from CD Universal disk amd 64 on PowerEdge 1850 and when select Gentoo kernel starts load kenel and then black screen and freezes. Is a two cpu 3.0 with 1GB of ram.
Suggestions ? |
|
Back to top |
|
|
mr.ed Tux's lil' helper
Joined: 30 Nov 2004 Posts: 75
|
Posted: Wed May 25, 2005 9:29 am Post subject: |
|
|
Just use the k8-kernel in the cd, the em64t kernel is b0rked!! _________________ (Ex-Pornstar) |
|
Back to top |
|
|
piramid n00b
Joined: 18 Feb 2004 Posts: 37
|
Posted: Wed May 25, 2005 9:41 am Post subject: |
|
|
mr.ed wrote: | Just use the k8-kernel in the cd, the em64t kernel is b0rked!! |
I checked the burned cdrom and I have this error: "/mnt/dvd_rw/livecd.squashfs: Cannot open: Permission denied"
I burned again the image (with slower writing) and I have again the same message.
Booting in gentoo-nofb (on both cd) I get this post "Decompressing Linux ... invalid compressed format (err=2)" |
|
Back to top |
|
|
mr.ed Tux's lil' helper
Joined: 30 Nov 2004 Posts: 75
|
Posted: Wed May 25, 2005 9:57 am Post subject: |
|
|
That was when using the em64t-kernel on the livecd or the k8 one? Sounds like a corrupt iso to me otherwise.. Try re-downloading it otherwise.. _________________ (Ex-Pornstar) |
|
Back to top |
|
|
stillman Apprentice
Joined: 07 Dec 2002 Posts: 223 Location: Vienna, Austria
|
Posted: Wed Jun 15, 2005 12:16 pm Post subject: |
|
|
i'll have to install a em64t-xeon-dual(nocona) server on friday. it's main task will be to serve as a asterisk-media-gateway. since i've got absolutely no experience with the whole 64bit architecture resp. architectures per se i'm pretty unsure if the installation will run smooth. my main concern is the stable and flawless function of the installed packages. can someone proof(or at least suggest) that asterisk or other amd64-ebuilds will compile and work properly on a em64t installed with the nocona-flag resp. with a amd64-based installation? or would you recommend to run a simple Pentium4-x86-installation to be sure there'll be no trouble? |
|
Back to top |
|
|
DangerDan n00b
Joined: 24 Dec 2004 Posts: 11
|
Posted: Wed Jun 15, 2005 2:29 pm Post subject: |
|
|
since you're going to be using it as an astrisk server, you want it to be as stable as possible - i'd steer clear of the 64 bit compile. the amd64 2005.0 install uses gcc 3.4.x by default. i prefer to use 3.3.x for boxes that i need to be solid. just mho. |
|
Back to top |
|
|
stillman Apprentice
Joined: 07 Dec 2002 Posts: 223 Location: Vienna, Austria
|
Posted: Wed Jun 15, 2005 6:44 pm Post subject: |
|
|
thanks for pointing me there - indeed there seem to be issues when compiling asterisk with gcc-3.4, having searched google for info on that matter. guess i'll stick to x86 for the time beeing. i assume to use standard pentium4 cflags then. |
|
Back to top |
|
|
petrjanda Veteran
Joined: 05 Sep 2003 Posts: 1557 Location: Brno, Czech Republic
|
Posted: Thu Jun 30, 2005 10:01 am Post subject: compiling gcc problem |
|
|
Im trying LFS and I think this is a gentoo related problem. Im running in 64-bits.
Code: |
./configure --host=x86_64-pc-linux-gnu --prefix=/tools --libexecdir=/tools/lib --with-local-prefix=/tools --disable-nls --enable-shared --enable-languages=c
|
Code: |
and 2/libgcc_s.so.1 libgcc_s_32.so
/tools/x86_64-pc-linux-gnu/bin/ld: skipping incompatible /usr/lib/../lib/libc.so when searching for -lc
/tools/x86_64-pc-linux-gnu/bin/ld: skipping incompatible /usr/lib/../lib/libc.a when searching for -lc
/tools/x86_64-pc-linux-gnu/bin/ld: skipping incompatible /usr/lib/libc.so when searching for -lc
/tools/x86_64-pc-linux-gnu/bin/ld: skipping incompatible /usr/lib/libc.a when searching for -lc
/tools/x86_64-pc-linux-gnu/bin/ld: cannot find -lc
collect2: ld returned 1 exit status
make[3]: *** [32/libgcc_s_32.so] Error 1
make[3]: Leaving directory `/mnt/lfs/gcc-3.4.4/gcc'
make[2]: *** [stmp-multilib] Error 2
make[2]: Leaving directory `/mnt/lfs/gcc-3.4.4/gcc'
make[1]: *** [stage1_build] Error 2
make[1]: Leaving directory `/mnt/lfs/gcc-3.4.4/gcc'
make: *** [bootstrap] Error 2
then i get this error while running "make bootstrap"
|
Anyone knows what to do? _________________ There is, a not-born, a not-become, a not-made, a not-compounded. If that unborn, not-become, not-made, not-compounded were not, there would be no escape from this here that is born, become, made and compounded. - Gautama Siddharta |
|
Back to top |
|
|
jannis Guru
Joined: 05 Dec 2004 Posts: 340 Location: Germany / Bavaria / Aschaffenburg
|
Posted: Tue Jul 19, 2005 8:13 pm Post subject: |
|
|
I tried to boot a Pentium4 EM64T from the AMD64-CD and it just rebooted after having loaded the kernel and the initrd to RAM. The machine has 2GB RAM. I tried the "gentoo" (k8) kernel and the gentoo-em64t but non of them worked. The thing just reboots.
Anyone got an idea? |
|
Back to top |
|
|
dildassen n00b
Joined: 05 Aug 2005 Posts: 3
|
Posted: Fri Aug 12, 2005 7:44 am Post subject: Successfull install on Poweredge 2850 |
|
|
Just thought I'd let you guys know that the AMD64 stage1 install worked flawlessly on my Poweredge 2850 servers, 2 x dual Xenon 64bit, 6Gmem, SCSI RAID etc.... |
|
Back to top |
|
|
Vampire999 n00b
Joined: 18 Jan 2005 Posts: 2
|
Posted: Thu Aug 25, 2005 3:10 pm Post subject: |
|
|
When you guys were up and running off your own kernel did it show two cpus or four?
When I was running from the cd it showed me there were four. Now that I have compiled the 2.6.12 kernel specifying the "Intel em64t" I only see two cpu's. |
|
Back to top |
|
|
|