[molpro-user] Compiling molpro2010 - parallel version

[Andy May]

Holger,

This does not look like the latest patched version of Molpro, can you 
download and try the latest tarball (molpro.2010.1.22.tar.gz).

I suspect the parallel failures are because you have manually set Molpro 
to use -lga++, when Molpro is set up to should use -lga. Here are the 
options I use for building a pure tcgmsg version of GA 5-0-2:

./configure --prefix=/some/dir/ga-install 
--with-blas8='-L/opt/intel/mkl/lib/intel64 -lmkl_intel_ilp64 
-lmkl_sequential -lmkl_core' --with-scalapack=no --enable-f77 
F77=/opt/intel/compilerpro-12.0.2.137/bin/intel64/ifort 
CC=/opt/intel/compilerpro-12.0.2.137/bin/intel64/icc 
CXX=/opt/intel/compilerpro-12.0.2.137/bin/intel64/icpc --with-tcgmsg

And then point Molpro to the GA build directory, i.e.:

./configure -batch -icc -ifort -mpp -mppbase /GA/BUILD/dir/ga-5-0-2

Alternatively, you can try:

./configure -batch -icc -ifort -mpp -auto-ga-tcgmsg

which should hopefully build GA automatically.

The serial failures I'm not sure, I guess the first thing to try is with 
the 2010.1.22 tarball.

Best wishes,

Andy

On 15/06/11 13:25, Holger Naundorf wrote:
> On Wed, 2011-06-15 at 11:16 +0100, Andy May wrote:
>> Holger,
>>
>> I'm afraid we don't have access to an AIX machine so are unable to test
>> Molpro on AIX. I suggest to contact the Global Arrays mailing list in
>> order to get at least the GA test program building:
>>
>> http://groups.google.com/group/hpctools
>>
>> For Linux, please can you send the CONFIG file from the build?
>>
> The CONFIG file is attached (additionally the config.log from the GA
> build used - both are build using the same Intel compiler, all GA
> testruns reported success).
>
> Building the serial molpro with the same compiler works flawlessly, but
> results in a handful of failing testjobs
>
> cpp.errout        hybrid_dft_ci.errout    tl_so_i.errout
> h2f_merge.errout  tlbr_mp2_dft_so.errout
>
> These do not crash, they just do not end and keep on running forever, so
> I suspect that I either should use a different compiler version anyway
> or need to tweak some optimization levels...but maybe I am doing
> something fundamentally wrong without noticing...
>
> Regards,
> Holger
>
>> Best wishes,
>>
>> Andy
>>
>> On 15/06/11 11:03, Holger Naundorf wrote:
>>> Dear molpro users/developers,
>>> I am in the process of trying to build a parallel version of molpro2010
>>> for Linux and AIX and keep runing into some problems.
>>>
>>> On AIX:
>>> I managae to build GA 5.0.2 with TCGMSG - I have not yet managed to make
>>> it compile using the IBM provided MPI libaries. Unfortunately the test
>>> programm up to now also refuses to build, so I cannot be certain if the
>>> build works cleanly. So if there is anyone using GA on AIX who can give
>>> advise on building GA there I would appreciate any tips.
>>>
>>> I can then build molpro either linking against the GA build or using
>>> only the IBM MPI libraries, not using GA. (Only hitch is that i cannot
>>> use the system lapack routines, i have to build with -noblas -nolapack
>>> and use the molpro provided routines, otherwise I end up with
>>> unsatisfied symbols). The resulting binary then fails nearly all tests,
>>> prducing either one of the four errors below (with different numbers for
>>> LEN=... nrest=... lenout=..., but always the same four numbers for the
>>> decompression error...)
>>>
>>> ---
>>>    ?ERROR IN GLOBAL_CREATE_INT FOR ARRAY   gltskl     LEN=     105
>>>
>>>    GLOBAL ERROR fehler on processor
>>> 0
>>>
>>> ---
>>>    Decompression error, nrest=******  lenout=    -1
>>>
>>>    GLOBAL ERROR fehler on processor   0
>>>
>>> ---
>>>    ?PROGRAMMING ERROR IN DECOMPRESSION    131060     65536    133123
>>> 131059
>>>    Please contact Molpro support
>>>
>>>    GLOBAL ERROR fehler on processor   0
>>>
>>> ---
>>>    Decompression error, lenout=********  lenbf=   32768
>>>
>>>    GLOBAL ERROR fehler on processor   0
>>> ---
>>> a total of two tests actually pass:
>>>
>>> isobutene_mp2.out
>>> c2h4_d2.out
>>>
>>> This is the same both for the binary linked with GA and with the IBM MPI
>>>
>>> System data:
>>>
>>> xlf -qversion
>>> IBM XL Fortran Enterprise Edition for AIX, V11.1
>>> Version: 11.01.0000.0009
>>>
>>> uname -a
>>> AIX a42 3 5 00C418EF4C00
>>>
>>>
>>>
>>> On Linux I am just starting the build, but I wanted to see if it works
>>> with the same configuration usiing GA build with TCGMSG. Building GA and
>>> molpro was quite a bit easier, but i end up with a binary producing
>>> exactly the same errors as the build on AIX...
>>>
>>> uname -a
>>> Linux cat 2.6.18-194.26.1.el5 #1 SMP Tue Nov 9 12:46:16 EST 2010 x86_64
>>> x86_64 x86_64 GNU/Linux
>>>
>>> ifort -V
>>> Intel(R) Fortran Intel(R) 64 Compiler Professional for applications
>>> running on Intel(R) 64, Version 11.1    Build 20090511 Package ID:
>>> l_cprof_p_11.1.038
>>>
>>> So before I start experimenting with lots of different compiler versions
>>> and builds using MPI I would like to ask if there is a 'recommended'
>>> build environment (using GA or not, based on which MPI libraries, using
>>> certain compiler versions etc...) for Linux builds. (we use big SMP
>>> machines here, so we do not normally use MPI, so my normal approach for
>>> older molpro versions was using the GA with TCGMSG build, which worked
>>> for older releases...)
>>>
>>> Thanks for any hints.
>>>
>>> Regards,
>>> Holger Naundorf
>>>
>>>