very poor performance using ELK DFT code

[Carbunco_F_]

Dear all,

I've succeeded in compiling the DFT code named ELK ( http://elk.sourceforge.net/ ) for the intel phi coprocessor, in native mode, but the tests show a very poor performance.

I've used the flags -mkl -mmic after sourcing the compilevars.sh as shown below:

-----------------------------begin of make.inc file

MAKE = make
F90 = ifort
F90_OPTS = -O2 -ip -unroll -no-prec-div -openmp -mmic -L  /opt/intel/lib/mic
F77 = ifort
F77_OPTS = -O2 -ip -unroll -no-prec-div -openmp -mmic -L  /opt/intel/lib/mic

AR = ar

LIB_SYS =
# LAPACK and BLAS libraries
LIB_LPK = -mkl -openmp -mmic

LIB_FFT = fftlib.a

--------------------------end of make.inc file

I hope somebody could help me before I turn on the waterworks...

TIA,

  Carb.

[Sunny_G_Intel]

Hello Carbunco,

Can you please provide us some of the benchmark results you have got and details about what are you comparing your results with. It would great if you can provide us some of your initial profiling results which will help us find out which routines(where you expect better performance) to investigate further. 

Also it would be helpful if you can provide us with the list of changes you made(or the modified code) in order to get the code compiling  and running on Intel® Xeon Phi™ coprocessors.

Thanks