[CPMD-list] cpmd and gfortran

[Axel Kohlmeyer]

On Wed, 26 Mar 2008, Krzysztof Sakrejda wrote:

KS> Hi,

krzysztof,

KS> I'm considering using cpmd with linux for small simulations, so performance
KS> is not a major issue--yet.  Ideally, I would like to use gfortran as the
KS> fortran compiler since it now supports the Cray pointer extensions which
KS> were previously problematic, and it lists cpmd as one of the sets of code it
KS> can successfully compile.  Does anyone here have experience using a
KS> gfortran-compiled cpmd?  Specifically I have two questions:

yes. i'm using gfortran regularly to test portability issues.

KS> 1) any unexpected compilation issues?

unexpected no. :) 

fft (if external) and BLAS/LAPACK (i.e. ATLAS/MKL/ACML/GOTO)
have to be compatible with gfortran.

KS> 2) any dramatic differences in runtime performance?

a large part of the performance comes from the libraries.
on my (intel) based desktop the difference between
gfortran/atlas (bundled with fedora core 6) and intel/mkl
(both version 10) is between 10 and 15%.

newer versions of gfortran may be even faster...
if you run on a large cluster and in parallel,
the performance of the compiler will be even less
important and the network latency will dominate
(and the c compiler used to compile the MPI libraries).

btw: don't go for the highest optimization level.
moderate optimization (-O2) plus loop unrolling
yields the overall fastest binaries. specifically
flags that turn on SIMD should be avoided. those
can affect the perfomance badly. on intel core
architecture and opteron using generic pentiumIII
optimization is a safe and on average very fast choice.
pentium-IV is a different story...

cheers,
   axel.

KS> I realize the Intel compiler is available, but I will probably have a go
KS> with gfortran first.
KS> 
KS> Best regards,
KS> 
KS> Krzysztof
KS> 

-- 
=======================================================================
Axel Kohlmeyer   akohlmey at cmm.chem.upenn.edu   http://www.cmm.upenn.edu
   Center for Molecular Modeling   --   University of Pennsylvania
Department of Chemistry, 231 S.34th Street, Philadelphia, PA 19104-6323
tel: 1-215-898-1582,  fax: 1-215-573-6233,  office-tel: 1-215-898-5425
=======================================================================
If you make something idiot-proof, the universe creates a better idiot.