Open MPI logo

Open MPI User's Mailing List Archives

  |   Home   |   Support   |   FAQ   |   all Open MPI User's mailing list

Subject: Re: [OMPI users] recompiling 1.3.1 with intels
From: Francesco Pietra (chiendarret_at_[hidden])
Date: 2009-04-08 12:39:43


On Wed, Apr 8, 2009 at 6:10 PM, Gus Correa <gus_at_[hidden]> wrote:
> Hi Francesco, list
>
> As Jeff suggested, the problem is not likely to be on the
> Intel compilers or on OpenMPI.
> I would guess it is the way information is being passed to the Amber
> configuration script that may be the cause.
> I am not familiar to Amber (we don't do computational Chemistry here),
> so it is hard to tell.
>
> Maybe if you send some information on how you configure Amber
> A) Amber's configure command line,
> B) output of Amber's configure --help , and
> C) the error messages Amber returned, etc.
>
> Also, have you tried the Amber mailing list (if it exists)?
> They should know more about Amber than we do.
>
> Two things that I found weird about your Amber configuration
> command (on an old posting) were:
> 1) that you didn't tell it where
> your openmpi was installed (no path), and
> 2) that you gave it the Intel compiler name (ifort),
> whereas I would guess the OpenMPI wrapper
> (mpif90 or mpif90, maybe with a full path)
> would be perhaps a better choice.
> Ifort doesn't know anything about your OpenMPI
> libraries and include files, whereas the OpenMPI
> mpif90/mpif77 know everything.
>
> PS - As for Intel compilers, I would stick to release 10.
> I tried to install release 11 two weeks ago.
> They changed something in the installation script that
> broke the installation procedure,
> would not allow me to choose the install directory,
> modified their directory structure names, etc.
> The net result was that I couldn't install, lest test it.

Hi Gus:
All above are sound and precious suggestions. I took force from your
mail. Although the flag "-openmpi" and "ifort" for the compiler
worked nicely with me at the time of amd64 etch, I'll try the full
paths and the names you suggest.

As to intel 11, I made a deb package with program "alien". Did not
install. A fixing is required. There is a young smart colleague in
Australia who is modifying his script to fix the deb package (if you
are interested at all in deb packages, I'll keep you informed; his
script for versions 9 and 10 worked nicely).

I am getting very strange results, such as files that are not found
(while they are where expected and the env variables are correct in
the cheat between the message passing interface and amber. One thing:
I have noticed on several occasions that if the situation is not
perfectly correct from the unix point of view, debian refused to do.
It has recently occurred with DOT2, who was used with Mac unix, but
debian found incorrect stuff (the installation code was corrected by
very kind DOT2 people and a paper is now on the web on the ACS
journals.

I'll try to reorganize the situation and when the problems are clear
may be I'll ask again advice here with the information you suggested.

thanks
francesco
>
> I hope this helps.
> Gus Correa
> ---------------------------------------------------------------------
> Gustavo Correa
> Lamont-Doherty Earth Observatory - Columbia University
> Palisades, NY, 10964-8000 - USA
> ---------------------------------------------------------------------
>
>
> Jeff Squyres wrote:
>>
>> On Apr 8, 2009, at 5:08 AM, Francesco Pietra wrote:
>>
>>> As I was unable to compile the parallel code Amber10 with openmpi
>>> 1.3.1 (fully tested) and intel compilers and mkl version 10 on debian
>>> amd64 lenny, I'll try with the intels version 11. If it does not work
>>> in my hands, I'll move to gnu compilers.
>>>
>>> Question: to recompile openmpi 1.3.1. on intels 11 should I first do
>>> anything on openmpi which was compiled with intels 10?
>>>
>>
>>
>> I test Intel v10 compilers frequently; they work fine with Open MPI.
>>
>> As far as we know, Intel v11 compilers work fine with Open MPI as well.
>>
> _______________________________________________
> users mailing list
> users_at_[hidden]
> http://www.open-mpi.org/mailman/listinfo.cgi/users
>