Compiling on RHEL 7 with MKL

Jump to page 1Prev 16

Forum Vet

1:42:24 PM PST - Fri, Dec 2nd 2016
I don't see any difference on the compilers between mpiifort and ifort ... do you?

Clicked A Few Times

6:52:38 AM PST - Mon, Dec 5th 2016

I apologize, I got greedy with my copy and paste. The output of mpiifort is:

$ mpiifort -v
mpiifort for the Intel(R) MPI Library 5.0 Update 2 for Linux*
Copyright(C) 2003-2014, Intel Corporation.  All rights reserved.
ifort version 15.0.1

Just to recap; after

make clean

on the src directory, I ran

make FC=mpiifort

and got still got the invalid data type error. After

make clean

and

make FC=ifort

, I don't get the error but get poor performance. How can I better trace where this error is coming from?
Can't thank you enough for your help!

Forum Vet

10:52:59 AM PST - Mon, Dec 5th 2016
Please post more details about your perceived bad performance. I hope I have shown you that 1) mpiifort and ifort are the same compiler 2) compiling with mpiifort does not work

Clicked A Few Times

5:37:34 AM PST - Tue, Dec 6th 2016

I had a previous build of NWChem on the cluster, that after OS updates stoppped working. The same application ran with these times:

convergence    iter        energy       DeltaE   RMS-Dens  Diis-err    time
---------------- ----- ----------------- --------- --------- ---------  ------
d= 0,ls=0.0,diis     1   -613.2132694772 -1.54D+03  1.59D+02  2.35D+02     5.6
d= 0,ls=0.0,diis     2   -610.9316964505  2.28D+00  4.38D+01  2.85D+01     9.6
Grid integrated density:     107.999848325420
Requested integration accuracy:   0.10E-05
d= 0,ls=0.0,diis     3   -601.3727976994  9.56D+00  2.66D+00  9.57D+01    13.5
Grid integrated density:     107.999889534094
Requested integration accuracy:   0.10E-05
d= 0,ls=0.0,diis     4   -614.8430123370 -1.35D+01  3.64D+00  1.46D+01    17.4
Grid integrated density:     107.999879003760
Requested integration accuracy:   0.10E-05
d= 0,ls=0.0,diis     5   -612.3853104496  2.46D+00  3.09D+00  1.89D+01    21.3

The new compilation runs with these times:

  convergence    iter        energy       DeltaE   RMS-Dens  Diis-err    time
---------------- ----- ----------------- --------- --------- ---------  ------
d= 0,ls=0.0,diis     1   -613.2132694772 -1.54D+03  1.59D+02  2.35D+02    49.1
d= 0,ls=0.0,diis     2   -610.9316964639  2.28D+00  4.38D+01  2.85D+01    59.5
Grid integrated density:     107.999848528712
Requested integration accuracy:   0.10E-05
d= 0,ls=0.0,diis     3   -601.3727976953  9.56D+00  2.66D+00  9.57D+01    69.8
Grid integrated density:     107.999889746041
Requested integration accuracy:   0.10E-05
d= 0,ls=0.0,diis     4   -614.8430121681 -1.35D+01  3.64D+00  1.46D+01    80.2

I still have the compiled code for the previous NWChem I ran. What files should I go through and compare the compilation steps? I looked in src/tools/config.log and the compilation command looks like this:

$ ../ga-5-3/configure --prefix=/pathto/Nwchem-6.5.revision26243-src.2014-09-10/src/tools/install --with-tcgmsg --with-mpi=-I/pathto/impi/4.1.1.036/intel64/include/ -L/pathto/impi/4.1.1.036/intel64/lib/ -lmpigf -lmpi -lmpigi -ldl -lrt -lpthread --enable-peigs --enable-underscoring --disable-mpi-tests --without-scalapack --without-lapack --with-blas8=-L/usr/lib64/ -llapack -lf77blas -latlas --with-openib CC=cc F77=gfortran ARMCI_DEFAULT_SHMMAX_UBOUND=131072

I'm lost in how intel mpi was used with gfortran. I thought it had to match

Forum Vet

11:21:24 AM PST - Tue, Dec 6th 2016
Try this to see if it improves performances cd $NWCHEM_TOP/src/tools make clean make FC=ifort ARMCI_NETWORK=SOCKETS cd .. make FC=ifort link

Clicked A Few Times

9:45:44 AM PST - Wed, Dec 14th 2016
Hi @Edoapra, I was able to dramatically improve performance by compiling the tools directory with make FC=ifort ARMCI_NETWORK=MPI-MT Thank you for your help

Forum Vet

6:11:27 PM PST - Tue, Dec 27th 2016
Thanks for your feedback

Forum >> NWChem's corner >> Compiling NWChem

Jump to page 1Prev 16