David Case wrote:
>
> On Thu, Mar 22, 2001, wentaofu wrote:
>
> >
> > Thank you for your pervious answer about compiling AMBER6.0 using MPICH.
> > Since there is a bug on the sander_classic using MPICH, I am compiling
> > AMBER6.0 on our SGI 2100 4 processor machine running IRIX6.5. We are using
> > the shared memory "Machine.sgi" file to make it compiled with parallelism.
> > But, when we test run sander_classic on the machine, we find it runs only on
> > one processor even using "setenv MP_SET_NUMTHREADS 4". Is there anything we
> > need to change on the Machine.sgi file to make it can be compiled with
> > parallelism? Thanks!
> >
>
> I don't know of any reason why sander_classic should not run on multiple
> processors when compiled with shared memory. I'm forwarding your question
> to the amber mail-reflector: that's always the best place to ask questions
> of this sort.
>
> ..regards...dave case
Hi,
There is a source code error in sander_classic that I've reported and
which is documented on the Amber web site:
http://amber.ch.ic.ac.uk/bugfixes/6.0/bugfix.20
Apart from that change to the MACHINE file, I've been able to compile
and run sander_classic on multiple processors successfully. What value
does the preprocessor use for ``-DSGI_MP_MAX_PROCESSORS''?. For example,
mine looks like the following:
cat sander.f | /lib/cpp -P -DDPREC -DIRIS -DISTAR2 -DREGNML
-DHAS_FTN_ERFC -DBLAS1 -DSGI_MP -DSGI_MP_MAX_PROCESSORS=32 -DSGIFFT
-DTFP -DCACHE_BLOCKSIZE=2500 > _sander_.f
So if I had 32 processors, and I didn't set the environment variable
MP_SET_NUMTHREADS to another value, I would run on 32 processors.
In your message you mention compiling Amber 6.0 with MPICH. Is that on
your sgi2100? If so, why are you using MPICH? sgi has an optimized
version of MPI for sgi architecture included in the Message Passing
Toolkit (MPT) which comes installed on your machine. To verify you have
it, type ``versions |grep mpt'' and you should be able to something like
the following:
I modules.sw.mpt 05/18/1999 MPT modules support Software
I mpt 05/02/2000 MPT 1.4.0.1
I mpt.relnotes 05/02/2000 MPT 1.4.0.1 Release Notes
I mpt.relnotes.base 05/02/2000 MPT 1.4.0.1 Release Notes
``man mpi'' will get you the man pages. If you don't have it, please
contact your sgi system engineer (or sales person) or your system
administrator so that it gets installed.
Also, you mention there is a bug with sander_classic with MPICH, is that
on Irix or Linux? Like I said, on Irix I used the MPI included in the
MPT but I've successfully used MPICH on Linux/IA32.
Nathalie
--
------------------------------------------------------------------------
Nathalie Godbout, Ph.D. SGI
godbout_at_hudson.sgi.com Chemistry and Biology Applications Group
(781)839-2147 One Cabot Road, Suite 250
(978)562-7450 fax Hudson, MA
vnet:483-2147 01749, USA
------------------------------------------------------------------------
Received on Fri Mar 23 2001 - 09:27:19 PST
