|
From: | Andreas Kuntzagk |
Subject: | Re: Binaries for SLES10 |
Date: | Thu, 29 Oct 2009 08:34:15 +0100 |
User-agent: | Thunderbird 2.0.0.22 (X11/20090625) |
Hi,
From my experience you will have a lot of problems compiling a static version of octave, plus you will not be able yo add any of the external available modules for octave, since it is required dynamic linking, in order for those to be loaded by octave ...
But I don't want static!!! I ran "configure --enable-shared" (also did a "make clean". Also the shared version of that liblapack is available so I don't know why it compiles against the static liblapack.a My experiences with compiling C and C++ are a little dusted (apart from the usual "configure; make; make install") so maybe it's something stupid.
I've managed to build from source a dynamic version for SLES 10, although I was unable to use an up-to-date version of gnuplot, and compiling that from source seems a nightmare ... I cat tell you I've used SRPMS in one of the cluster nodes (build node) to compile and create the rpms, some of the specs required minor chages as they where only available for fedora or so ... Then used the rmps created at that node to install in all the other nodes of the cluster ...
I don't really need rpm's since I have a /usr/local on a NFS for such shared software.
But would you mind to send me your rpm's ? Maybe they are working for me? Or do they have some special dependencies built in?
regards, Andreas
Regards, Mário On Wed, Oct 28, 2009 at 3:41 PM, Andreas Kuntzagk <address@hidden> wrote:I just now started again to work on this problem and ATM it fails while linking against the static liblapack.a with this -fPIC error. I also have a liblapack.so and it should link against this. I configured with "--enable-shared". How do I force it to use the shared liblapack.so (both static and shared are in same directory) regards, Andreas Jaroslav Hajek wrote:On Tue, Oct 20, 2009 at 2:33 PM, Andreas Kuntzagk <address@hidden> wrote:Hi,At my previous just I had to use SUSE as well, and getting Octave up and running was not a particular joyful experience. SUSE is just not very good at packaging basic scientific libraries, so you'll have to compile a lot of stuff yourself from scratch.Yeah, but for this I need to find out what libraries are missing and where to get them.You can get some (semi-old) Octave RPM's for SUSE by searching at http://software.opensuse.org/searchOk, didn't know this page before. There seems to be some 3.0.3 binaries there. I'll try and come back with results.but (at least the last time I used them) they did not come with full functionality (no QHull and no suite-sparse as far as I remember).Actually I don't know what functionality is required here. Have to investigate.I'm sorry that the answer isn't more satisfactory, but in my experience SUSE just isn't very good for scientific work :-(Unfortunately changing distro is not an option right now. This is a big compute cluster (in production) and I can't start from scratch again. For the next time what distro do you propose for scientific computing? regards, AndreasWe have a cluster running SUSE too, and I think I had to compile all libs from scratch as well, except BLAS and LAPACK. OTOH, the work needs only to be done once, and I'm routinely rebuilding Octave (to be used by all other users) from sources every 2-4 weeks... For a modest reward I'd be willing to do the same on your cluster :D But seriously, compiling Octave is surely much more work, but the result is worth it. You can use heavy compiler optimizations (we use Intel C++ with -O3), you can link in vendor-tuned BLAS if you have it (we use AMD's multi-threaded ACML) and most importantly, you get the most recent sources with the latest features. Also, when a bug is fixed, you just download and apply the patch (using mercurial that's trivial) and rebuild, no need to wait for next release... hth_______________________________________________ Help-octave mailing list address@hidden https://www-old.cae.wisc.edu/mailman/listinfo/help-octave
[Prev in Thread] | Current Thread | [Next in Thread] |