| 6:04:10 PM PDT - Thu, Mar 14th 2013 |
|
11bolts,
1) USE_MPIF4 has no bearing to your DFT run (it affects only the planewave code)
2) You are right when say that the Intel ILP64 libraries don't need the 64to32 trick.
I have just compiled NWChem with the Intel 2013 composer and I have used settings pretty similar to yours
(the only differences are that I am using MPICH2 instead of Intel MPI and that I have not defined BLAS95 as you did).
The binary I generated had no problem running the input file you provided.
The relevant env. variables are the following (using NWChem 6.1 or later versions).
export BLAS_LIB="-L$MKLROOT/lib/intel64 -lmkl_intel_ilp64 -lmkl_sequential -lmkl_core -lpthread -lm"
export BLASOPT="$BLAS_LIB"
export BLAS_SIZE=8
export SCALAPACK_SIZE=8
export SCALAPACK="-L$MKLROOT/lib/intel64 -lmkl_scalapack_ilp64 -lmkl_intel_ilp64 -lmkl_sequential -lmkl_core\ -lmkl_blacs_intelmpi_ilp64 -lpthread -lm"
export SCALAPACK_LIB="$SCALAPACK"
export USE_SCALAPACK=y
3) Big fat warning for compiling NWChem with the Intel Composer XE 2013:
We have already found a few parts of NWChem that required modifications to get correct behaviors when Intel 2013 was used.
We suggest you to either download the development version of NWChem (with the related caveats) or apply the following patches to 6.1.1
http://nwchemgit.github.io/images/Texasmem2.patch.gz
http://nwchemgit.github.io/images/Giaxyz.patch.gz
As a result of these experiences with Intel 2013, I strongly discourage from redefining the value of FOPTIMIZE as you did,
and use the default value instead.
Cheers, Edo
|