[molpro-user] molpro + GA + scalapack?

Stefan Becuwe stefan.becuwe at ua.ac.be
Thu Jul 29 07:45:25 BST 2010


Hello,

I built GA 4.3.1 with scalapack support, using Intel mpiifort, mpiicc and 
MKL:

make TARGET=LINUX64 \
      IB_INCLUDE=/usr/include IB_LIB=/usr/lib64 \
      IB_LIB_NAME="-libumad -libverbs -lpthread" \
      ARMCI_NETWORK=OPENIB  MA_USE_ARMCI_MEM=y \
      MSG_COMMS=MPI \
      MPI_LIB=/easy/impi/4.0.0.027/intel64/lib64 \
      MPI_INCLUDE=/easy/impi/4.0.0.027/intel64/include64 \
      USE_SCALAPACK_I8=y \
      SCALAPACK="-L/easy/imkl/10.2.5.035/lib/em64t -lmkl_scalapack_ilp64 -lmkl_blacs_intelmpi_ilp64 -lmkl_intel_ilp64 -lmkl_intel_thread -lmkl_lapack -lmkl_core -liomp5 -lpthread" \
      CC=mpiicc CXX=mpiicpc FC=mpiifort COPT='-O2 -xHOST' FOPT='-O2 -xHOST'

How can I get scalapack support into molpro (2009.1, patchlevel 34)?

I tried using the BLASLIB variable (using the options to build GA)

export BLASLIB="-L/easy/imkl/10.2.5.035/lib/em64t \
-lmkl_scalapack_ilp64 \
-lmkl_blacs_intelmpi_ilp64 -lmkl_intel_ilp64 -lmkl_intel_thread \
-lmkl_lapack -lmkl_core -liomp5"

... which seems to be ignored by molpro's configure, whether I use 
-blaspath or not.

./configure -batch -ifort -icc -blaspath \
/easy/imkl/10.2.5.035/lib/em64t -mpp -mppbase /easy/ga-4-3-1 \
-var LIBS="-L/usr/lib64 -libumad -libverbs -lm" -instroot 
"$SOFTROOTMOLPRO"

The output of configure contains:

Use BLAS library - Intel 64-bit Math Kernel Library (MKL) for Linux
BLASLIB=-L/easy/imkl/10.2.5.035/lib/em64t -lmkl_intel_ilp64 
-lmkl_sequential -lmkl_core

which is clearly not the scalapack version...

Moreover, it says:

parallel=intel-mpi-gfortran

Any idea where the "gfortran" comes from?  I explicitly mentioned -ifort 
(and also GA was built with mpiifort).

Anyway, it starts to compile, but at some point, it says:

compiling ppidd_c.c
failed
ppidd_c.c(41): catastrophic error: could not open source file "sndrcv.h"
       #include <sndrcv.h>

Does this mean that   MSG_COMMS=MPI  is not a good way to build GA for 
molpro?  Should I use   USE_MPI=y   instead (and use TCGMSG-MPI)?


Any idea how to resolve these problems? Thanks in advance!

Regards


Stefan Becuwe



More information about the Molpro-user mailing list