[molpro-user] Molpro 2010.1 PL20 with mpich2-nemesis and GA 5.0.2

Fri Apr 15 00:02:38 BST 2011

Dear Andy
Yes, the auto build works great. Though after compilation I had still  
the ARMCI error, but it disappeared
after I removed PATH to the previous mpich2 mpiexec in  
the .bash_profile and reinitiated the terminal
session. Now all testjobs passed ok with -n6 option.  Thanks!
Jacek

On Apr 14, 2011, at 11:32 AM, Andy May wrote:

> Jacek,
>
> You don't need to do anything related to MPICH2 or GA, the auto-build
> will download MPICH2, GA etc. and build it for you.
>
> For reference, --with-pm=hydra, is the option to MPICH2 configure, I'm
> not sure if it is the default yet.
>
> Best wishes,
>
> Andy
>
> On 14/04/11 16:29, Jacek Klos wrote:
>> Dear Kirk, Andy
>>
>> Thank you for your suggestions, I will certainly try auto build  
>> option
>> with a fresh molpro source.
>> Could you post what configure flags did you use for mpich2 to use
>> hydra,or this device is chosen by default?
>> Thanks and best wishes
>> Jacek
>> On Apr 14, 2011, at 11:22 AM, Kirk Peterson wrote:
>>
>>> Dear Jacek and Andy,
>>>
>>> I noted some problems using the auto build options last night, but  
>>> I'm
>>> happy to report that after deleting
>>> all object files and recompiling, it now magically works. There was
>>> something evidently not quite right
>>> from my earlier attempts. I've been using hydra for a while now on  
>>> my
>>> linux clusters and it really works
>>> nicely.
>>>
>>> best wishes,
>>>
>>> Kirk
>>>
>>> On Apr 14, 2011, at 7:05 AM, Andy May wrote:
>>>
>>>> Jacek,
>>>>
>>>> I don't believe the problem got fixed in GA, but I don't see the  
>>>> problem
>>>> with MPICH2 1.3.3rc1. I would now recommend the use of hydra  
>>>> instead of
>>>> gforker.
>>>>
>>>> Many of the options you passed to Molpro's configure were default  
>>>> (-i8,
>>>> -nocuda), it is best to give options only if you want to change the
>>>> default behaviour, it is then much easier for others to read your
>>>> configure line.
>>>>
>>>> It might be worth trying our auto-build options, which are still
>>>> experimental, but worked when I tried with 10.6.7 using latest  
>>>> ifort:
>>>>
>>>> ./configure -batch -gcc -ifort -mpp -auto-ga-tcgmsg-mpich2
>>>>
>>>> This worked for me and I didn't see the problems you mentioned.
>>>>
>>>> Best wishes,
>>>>
>>>> Andy
>>>>
>>>> On 13/04/11 18:27, Jacek Klos wrote:
>>>>> Hi
>>>>> I found the following discussion at hpctools google groups:
>>>>> http://groups.google.com/group/hpctools/browse_thread/thread/3d66370b71a5b153
>>>>>
>>>>> It seems similar to the problem I have with GA 5.0.2 and mpich2. I
>>>>> will try openmpi.
>>>>> Was this bug resolved later?
>>>>> Best
>>>>> Jacek
>>>>> __________________________________
>>>>> Jacek Klos, Research Assistant Professor
>>>>> Department of Chemistry and Biochemistry
>>>>> University of Maryland
>>>>> College Park, MD 20742 USA
>>>>> tel: 1-301-405-1384
>>>>> fax: 1-301-314-9121
>>>>> http://www.researcherid.com/rid/A-6457-2008
>>>>> ___________________________________
>>>>>
>>>>> On Apr 13, 2011, at 11:12 AM, Jacek Klos wrote:
>>>>>
>>>>>> Dear Molpro Users and Developers,
>>>>>>
>>>>>> I compiled a source code for 2010.1 PL 20 with GA 5.0.2 and  
>>>>>> mpich2-
>>>>>> nemesis. The machine is Mac OS X 10.6.7 with the newest Intel  
>>>>>> Ifort
>>>>>> Composer XE.
>>>>>> When I run tests, I get the following error:
>>>>>> Running job c6h6_freq_restart.test
>>>>>> 0:Child process terminated prematurely, status=: 0
>>>>>> (rank:0 hostname:mha-macpro5.umd.edu pid:77252):ARMCI DASSERT  
>>>>>> fail.
>>>>>> src/signaltrap.c:SigChldHandler():178 cond:0
>>>>>> **** PROBLEMS WITH JOB c6h6_freq_restart.test
>>>>>> **** For further information, look in the output file
>>>>>> **** /opt/molpro_mpp/molpro2010_GA5.0.2/molpro2010.1/testjobs/
>>>>>> c6h6_freq_restart.errout
>>>>>> Running job ch3_cosmo_dft.test
>>>>>> 0:Child process terminated prematurely, status=: 0
>>>>>> (rank:0 hostname:mha-macpro5.umd.edu pid:77352):ARMCI DASSERT  
>>>>>> fail.
>>>>>> src/signaltrap.c:SigChldHandler():178 cond:0
>>>>>> Running job ch4_b2plyp.test
>>>>>> 0:Child process terminated prematurely, status=: 0
>>>>>> (rank:0 hostname:mha-macpro5.umd.edu pid:77487):ARMCI DASSERT  
>>>>>> fail.
>>>>>> src/signaltrap.c:SigChldHandler():178 cond:0
>>>>>>
>>>>>> Molpro first creates errout with the info of child process  
>>>>>> abortion,
>>>>>> but then molpro continues and finished a job and regular output  
>>>>>> is
>>>>>> created and
>>>>>> it looks like molpro finishes a job OK. Here are my options for
>>>>>> configures:
>>>>>>
>>>>>> Mpich2:
>>>>>> sudo ./configure --with-device=ch3:nemesis --with-pm=gforker --
>>>>>> prefix=/
>>>>>> usr/local/mpich2_1.3.2p1_nemesis --enable-fc FC=ifort --enable-cc
>>>>>> CC=gcc --enable-cxx CXX=g++
>>>>>>
>>>>>> GA 5.0.2
>>>>>> sudo ./configure --with-tcgmsg --with-mpi="-I/usr/local/
>>>>>> mpich2_1.3.2p1_nemesis/include -L/usr/local/ 
>>>>>> mpich2_1.3.2p1_nemesis/lib
>>>>>> -lpmpich -lmpich -lopa -lmpl" --enable-i8 F77=ifort CC=gcc CXX=g 
>>>>>> ++ --
>>>>>> prefix=/usr/local/ga-5-0-2-install
>>>>>>
>>>>>> Molpro 2010.1 patchlevel 20
>>>>>> sudo ./configure -ifort -i8 -gcc -lapack -blas -nocuda -nohdf5 - 
>>>>>> mpp -
>>>>>> mppbase /usr/local/ga-5-0-2 -gforker
>>>>>>
>>>>>> Does anybody have an idea what can be the reason? Global Arrays  
>>>>>> test.x
>>>>>> test by itself tests OK.
>>>>>>
>>>>>> Best wishes
>>>>>> Jacek
>>>>>>
>>>>>> __________________________________
>>>>>> Jacek Klos, Research Assistant Professor
>>>>>> Department of Chemistry and Biochemistry
>>>>>> University of Maryland
>>>>>> College Park, MD 20742 USA
>>>>>> tel: 1-301-405-1384
>>>>>> fax: 1-301-314-9121
>>>>>> http://www.researcherid.com/rid/A-6457-2008
>>>>>> ___________________________________
>>>>>>
>>>>>> _______________________________________________
>>>>>> Molpro-user mailing list
>>>>>> Molpro-user at molpro.net
>>>>>> http://www.molpro.net/mailman/listinfo/molpro-user
>>>>>
>>>>> _______________________________________________
>>>>> Molpro-user mailing list
>>>>> Molpro-user at molpro.net
>>>>> http://www.molpro.net/mailman/listinfo/molpro-user
>>>> _______________________________________________
>>>> Molpro-user mailing list
>>>> Molpro-user at molpro.net
>>>> http://www.molpro.net/mailman/listinfo/molpro-user
>>>
>>