[MITgcm-devel] Re: AMD Opteron and optim.x

[Patrick Heimbach]

Hi Daniel,

here's what I did:

1.
I went to
http://www.cs.utexas.edu/users/kgoto/
which has a set of high performance BLAS libraries for 
all kinds of platforms.

2. 
Downloaded
http://www.cs.utexas.edu/users/kgoto/libraries/libgoto_opt64p-r0.94-2.so.gz
which seems to be the suitable one for your Opteron 64 bit multi-threaded
and put them in a directory, e.g. ~/mylib/

3.
As per website instruction,
also need the file  xerbla.f
http://www.cs.utexas.edu/users/kgoto/libraries/xerbla.f
I downloaded it into the MITgcm/optim/ directory as 
xerbla.F (note capital F)
(maybe cleaner to generate an object file and keep it in ~/mylib/ )

4.
in .bashrc added line
export LD_LIBRARY_PATH=~/mylib
(consistent with the directory where I keep the library)

5.
in MITgcm/optim/
edit the Makefile as follows
* add xerbla.F to             $SRC
* add -L~/mylib/              $LIBDIRS
* add -lgoto_opt64-r0.94-2    $LIBS
and ignore
-lblas1

That works.
You can find my setup on dolphin under
~heimbach/MITgcm/optim/

Cheers
-Patrick



Quoting Daniel Lea <daniel.lea at jhuapl.edu>:

> 
> Hi Patrick,
> 
> Glad you successfully got an account.
> 
> my directory is /home/dlea/mitgcm3
> I compile the code in /home/dlea/mitgcm3/dan/exp3g/input
> 
> The makefile is there too. I dont think there are any special machine
> specific things. Maybe thats where i'm going wrong! I use
> 
> ../../../tools/genmake -mods=../code
> make depend
> make
> 
> to compile.
> 
> Thanks
> 
> Dan
> 
> On Tue, 4 May 2004, Patrick Heimbach wrote:
> 
> > Hi Daniel,
> >
> > could you point me to your compile and exec directory on dolphin?
> > To get started I'd need your machine-specific Makefile settings.
> > I'd like to run a quick verification experiment.
> >
> > Thanks
> > -Patrick
> >
> >
> >
> > Quoting Daniel Lea <daniel.lea at jhuapl.edu>:
> >
> > >
> > > Hi Patrick,
> > >
> > > Tom Haine has recently bought a cluster of AMD 64bit Opterons. I can
> > > happily run the MITgcm forward model and adjoint. Very fast it is too.
> > > However, I'm running up against problems with the minimisation routine
> > > optim.x. In particular I can't compile it on that machine. In the optim
> > > directory, if I do make I get:
> > >
> > > "
> > > f77 -o optim.x -C optim_main.o optim_sub.o optim_numbmod.o
> optim_initmod.o
> > > optim_readdata.o optim_writedata.o optim_write_control.o simul.o -L.
> > > -L../lsopt/ -L/opt/SUNWspro/SC4.2/lib/v7 -llsopt_ecco -lblas1
> > >
> >
>
/usr/lib64/gcc-lib/x86_64-suse-linux/3.2.2/../../../../x86_64-suse-linux/bin/ld:
> > > skipping incompatible ../lsopt//libblas1.a when searching for -lblas1
> > >
> >
>
/usr/lib64/gcc-lib/x86_64-suse-linux/3.2.2/../../../../x86_64-suse-linux/bin/ld:
> > > cannot find -lblas1
> > > collect2: ld returned 1 exit status
> > > make: *** [optim.x] Error 1
> > > "
> > >
> > > Perhaps I need a different BLAS. I've no idea where to get one from
> > > though. I did try a couple of other BLAS libraries but I got other
> errors
> > > using them.
> > >
> > > I believe the AMD opteron can emulate a 32 bit CPU too. So I tried
> > > copying over optim.x compiled on an Intel machine. It runs but doesn't
> > > work properly giving the following error:
> > >
> > > "
> > > do_us: off end of record
> > > apparent state: unit 20 named ecco_ctrl_MIT_obctrl.opt0000
> > > last format: list io
> > > lately reading sequential unformatted external IO
> > > "
> > >
> > > Thanks
> > >
> > > Dan
> > >
> > >
> >
> >
> > _/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/
> > Patrick Heimbach     Massachusetts Institute of Technology
> > FON: +1/617/253-5259                    EAPS, Room 54-1518
> > FAX: +1/617/253-4464               77 Massachusetts Avenue
> > mailto:heimbach at mit.edu                 Cambridge MA 02139
> > http://www.mit.edu/~heimbach/                          USA
> >
> 
> 


_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/
Patrick Heimbach     Massachusetts Institute of Technology
FON: +1/617/253-5259                    EAPS, Room 54-1518
FAX: +1/617/253-4464               77 Massachusetts Avenue
mailto:heimbach at mit.edu                 Cambridge MA 02139
http://www.mit.edu/~heimbach/                          USA