[molpro-user] Installing molpro - GA toolkit needed, or can we use openmpi by itself?

[Andy May]

Benji,

This error is very strange and indicates some of the Molpro source files 
are missing, can you confirm that 'abdata.ascii' is actually present? 
The tarball certainly contains these:

~/test> tar -xzf ~/molpro.2010.1.23.tar.gz
~/test> ls Molpro/lib/*.ascii
Molpro/lib/abdata.ascii  Molpro/lib/rysrw.ascii

It's worth also checking the file permissions, but I don't see how only 
these files could have somehow changed since the rest of the source 
files are obviously readable for you to be able to compile them.

Can you try generating the auxiliary files ABDATA and RYSRW again by:

rm -f lib/{ABDATA,RYSRW,init.log}
make

and once again check the contents of lib/init.log.

Best wishes,

Andy

On 07/09/11 04:55, Benji Wakely wrote:
> Output of lib/init.log:
>
> ----------------
> cat lib/init.log
>
>   ? Error
>   ? The abdata.ascii file does not exist
>   ? The problem occurs in cnvrt
> <error type="Error" location="cnvrt" message="The abdata.ascii file does not exist"/>
> [hercules005:28469] *** Process received signal ***
> [hercules005:28469] Signal: Segmentation fault (11)
> [hercules005:28469] Signal code: Address not mapped (1)
> [hercules005:28469] Failing at address: (nil)
> [hercules005:28469] *** End of error message ***
> --------------------------------------------------------------------------
> mpirun noticed that process rank 0 with PID 28469 on node hercules005.latrobe.edu.au exited on signal 11 (Segmentation fault).
> --------------------------------------------------------------------------
> ----------------
>
> I can invoke mpirun with a simple mpi script,
> (Not running on the head node), and it seems to work correctly.
>
> (For what it's worth, openmpi was compiled with:
>   CC=pgcc CXX=pgcpp F77=pgf77 FC=pgf90 ./configure --prefix=/usr/local/openmpi/1.4.3-pgi --with-openib --with-tm=/usr/local/torque/latest --enable-static --enable-shared
> )
>
> (the 'with-tm' is torque v.2.5.5)
>
>
> Here's a verbose listing of the process I'm using to compile/make molpro,
> for public scrutiny/possible embarrassment:
>
> # set up basic molpro files
> - set up ~/.molpro/token
> - extract molpro source file (molpro.2010.1.23.tar.gz)
>
> # set up PGI compilers
> - ensure PGI version 11.8 binaries are in PATH
> - ensure LD_LIBRARY_PATH includes PGI 'lib' and 'libso' directories
>
> # set up openmpi (openmpi compiled against PGI 11.8)
> - ensure CPPFLAGS references openmpi 'include' directory
> - ensure LDFLAGS references openmpi 'lib' directory
> - ensure MPI_DIR and MPI_HOME reference openmpi base directory
> - ensure LD_LIBRARY_PATH includes openmpi 'lib' directory
>
> # configure molpro
> export TMPDIR=/tmp
> CONFIGURE_OPTIONS="-openmpi -pgf90 -pgcc -blaspath /usr/local/pgi/linux86-64/11.8/lib -mpp -mppbase /usr/local/openmpi/1.4.3-pgi/include -var INSTBIN=/usr/local/molpro/2010.1-openmpi-pgi/bin -var INSTLIB=/usr/local/molpro/2010.1-openmpi-pgi/lib -var INSTHTML=/usr/local/molpro/2010.1-openmpi-pgi/html"
> ./configure -batch $CONFIGURE_OPTIONS
>
> make -j8
>
> ...I can't spot anything I'm doing that's obviously wrong.
>
> --Benji
>
> Benji Wakely - b.wakely at latrobe.edu.au
> Unix Systems Administrator
>
> Work phone: +613 9479 5499
> mobile: +614 34 307 667
>
>
>> -----Original Message-----
>> From: mayaj1 at Cardiff.ac.uk [mailto:mayaj1 at Cardiff.ac.uk]
>> Sent: Tuesday, 6 September 2011 11:18 PM
>> To: Benji Wakely
>> Cc: molpro-user at molpro.net
>> Subject: Re: [molpro-user] Installing molpro - GA toolkit needed, or can
>> we use openmpi by itself?
>>
>> Benji,
>>
>> I see nothing obviously wrong with your CONFIG file, I've just tested
>> building a copy of Molpro with pretty much the same versions of
>> compilers, libraries etc. as you, and it works fine.
>>
>> The problem you saw is that the auxiliary file ABDATA is not present.
>> This file (and others) are created at 'make' time by running an
>> initialization job. Can you check the contents of lib/init.log which
>> should have the output from running that job? It looks as though your
>> openmpi build has been configured with some special options to interact
>> with the queue system, does this prevent the use of mpirun on the
>> (head?) node where you are building?
>>
>> Best wishes,
>>
>> Andy
>>
>> On 06/09/11 00:37, Benji Wakely wrote:
>>> Hi Andy, thanks for the response.
>>>
>>> 1)
>>> When you say 'MPI2', this includes the openmpi implementation of MPI2,
>> yes?
>>> (openmpi doesn't use mpd as such - but mpiexec seems to be working,
>>>    which I believe means it's working.  Corrections welcomed!)
>>>
>>> I've attached the generated CONFIG file from the PGI attempt.
>>>
>>> I can provide the output from the compilation process if the CONFIG
>> file doesn't help.
>>>
>>> Best,
>>> --Benji
>>>
>>> Benji Wakely - b.wakely at latrobe.edu.au
>>> Unix Systems Administrator
>>>
>>> Work phone: +613 9479 5499
>>> mobile: +614 34 307 667