#41 Problem with MP2 optimisation on a Bluegene P machine

[Jens Thomas]

I've hit a problem trying to run an MP2 optimisation on 1024 processors on an IBM Bluegene P machine, with what appears to be some
sort of overflow happening to the coordinates.

When the coordinates are printed after the first optimisation step, they are a tad on the big size:

molecule<Molecule>: (
symmetry = c1
unit = "angstrom"
{ n atoms geometry }={
1 Cl [-25445994562042207360991595855872.0000000000 80224398962188972120643712581632.0000000000 10806038220264016548205011402752.0000000000]
2 H [-11024065322025672529989165121536.0000000000 64740970660724829967202541109248.0000000000 -46578671303775149406559630524416.0000000000]
3 H [-26391589815870666914245820547072.0000000000 44930545225534618157457034182656.0000000000 -19463851181640218021604806885376.0000000000]

MPQC was configured with the following:

/home/jmht93/mpqc-trunk/configure \ --host=powerpc-bgp-linux \ --enable-cross-compile \ --enable-static \ --enable-always-use-mpi \ --enable-threads \ --with-default-parallel=mtmpi \ --with-mpi-thread=multiple \ --with-cc="mpixlc_r" \ --with-cxx="mpixlcxx_r" \ --with-f77="bgxlf_r" \ --with-libs="/opt/ibmmath/essl/4.4/lib/libesslbg.a /home/jmht93/mpqc-trunk/my_lapack/liblapack.a"

The version of the code is as follows:

svn info:
URL: http://mpqc.svn.sourceforge.net/svnroot/mpqc/trunk/mpqc
Repository Root: http://mpqc.svn.sourceforge.net/svnroot/mpqc
Repository UUID: 0fb12191-db4b-0410-8866-e78deaee71e5
Revision: 9448
Node Kind: directory
Schedule: normal
Last Changed Author: lkong
Last Changed Rev: 9448
Last Changed Date: 2010-06-20 01:12:01 +0100 (Sun, 20 Jun 2010)

I've uploaded the input and output so they should be attached to this report.

If you require any additional information, please let me know.

Thanks,

Jens

[Jens Thomas]

MPQC output of failing job

[Jens Thomas]

Input file