NWchem 6.3 compile problems with Intel compilers: ics-2013.0.028
Just Got Here
12:21:40 PM PDT - Tue, Aug 20th 2013
I am trying to compile NWchem 6.3 using the Intel compilers (2013.0.028) and the Intel mpi. I am getting a lot of “ifort: command line warning #10006: ignoring unknown option” errors.

I am compiling on an SandyBridge machine with 16 processors and 128GB RAM. The OS is SLES11 SP1. The network is a 10GE with a Mellanox ConnectX NIC.

The steps I used to compile are:

cd /lustre01/vol0/crookscj/nwchem-6.3-src.2013-05-28
  1. the following source setups up the PATH and LD_LIBRARY_PATH for the Intel compilers.
source /tstapps/global/bin/devsetup
setenv LARGE_FILES TRUE
setenv USE_NOFSCHECK TRUE
setenv TCGRSH /usr/bin/ssh
setenv NWCHEM_TOP /lustre01/vol0/crookscj/nwchem-6.3-src.2013-05-28
setenv NWCHEM_TARGET LINUX64
setenv USE_MPI y
setenv USE_MPIF y
setenv USE_MPIF4 y
setenv MPI_LOC $I_MPI_ROOT
setenv MPI_LIB $I_MPI_ROOT/lib64
setenv MPI_INCLUDE $I_MPI_ROOT/include64
setenv BLASOPT "-mkl"
setenv ARMCI_NETWORK MPI-SPAWN
setenv FC mpiifort
setenv CC mpiicc
setenv CXX mpiicpc
setenv FXX mpiifort


cd src

make nwchem_config NWCHEM_MODULES=all
  1. I had to add the FC=mpiifort option to the command line. It did not seem to pick it up
  2. from the setenv, although the CC option did.
make FC=mpiifort |& tee make.log01

When I do the compile I get the file following errors:

ifort: command line warning #10006: ignoring unknown option '-dM'
ifort: command line warning #10006: ignoring unknown option '-sTdIn_'
ifort: command line error: no files specified; for help type "ifort -help"
ifort: command line warning #10006: ignoring unknown option '-dM'
ifort: command line warning #10006: ignoring unknown option '-sTdIn_'
ifort: command line error: no files specified; for help type "ifort -help"
ifort: command line warning #10006: ignoring unknown option '-dM'
ifort: command line warning #10006: ignoring unknown option '-sTdIn_'
ifort: command line error: no files specified; for help type "ifort -help"
/bin/sh: line 0: [: too many arguments

I do get a binary, however, when I try to run with an input file that works with 6.1.1 I get the following errors.

MA fatal error: MA_sizeof: invalid datatype: 343597384693
MA fatal error: MA_sizeof: invalid datatype: 343597384693
MA fatal error: MA_sizeof: invalid datatype: 343597384693
MA fatal error: MA_sizeof: invalid datatype: 343597384693

I can upload the make.log01 and input file, if you let me know how/where you want it.

Thanks for your help.

Cindy

Forum Vet
3:47:02 PM PDT - Tue, Aug 20th 2013
Cindy
These are the setting I would you recommend to use "exclusively" (i.e. please unsetenv the one that appears in your list and not in this one below). For the make command, please type make FC=ifort

setenv MPI_HOME location of your mpi installation
setenv MKLROOT location of your MKL installation
setenv MPI_INCLUDE "$MPI_HOME"/include
setenv MPI_LIB "$MPI_HOME"/lib
export LIBMPI="-lmpi -lmpigf -lmpigi -lrt -lpthread"
setenv USE_NOFSCHECK TRUE
setenv NWCHEM_TOP /lustre01/vol0/crookscj/nwchem-6.3-src.2013-05-28
setenv NWCHEM_TARGET LINUX64
setenv USE_MPI y
setenv USE_MPIF y
setenv USE_MPIF4 y
setenv BLASOPT " -L$(MKLROOT)/lib/intel64 -lmkl_intel_ilp64 -lmkl_sequential -lmkl_core -lpthread -lm"
setenv BLAS_LIB "$BLASOPT"
setenv BLAS_SIZE 8
setenv ARMCI_NETWORK MPI-TS

Just Got Here
7:06:40 AM PDT - Sat, Aug 24th 2013
Thank you very much for your reply! I will give these a try on Monday.

Just Got Here
1:47:04 PM PDT - Mon, Aug 26th 2013
I was able to compile 6.3 today and have a test job running. I am curious, however, about two things. 1. If I use FC=ifort, why do I not use CC=icc? 2. Why do I not use the intel mpi wrapper scripts (mpiifort, mpiicc)?

Thanks again for your help.

Cindy

Clicked A Few Times
10:34:08 AM PDT - Thu, May 8th 2014
I am curious about why this is as well. Previous versions of NWChem compiled with mpiifort, so why not this one?


Forum >> NWChem's corner >> Compiling NWChem