[Wien] WIEN2k 09 poor benchmarks

[Peter Blaha]

WIEN2k09 must not be slower than previous versions.

In fact there is a small improvement in the parallelization of the LOs 
in hamilt.F (probably only noticeable with very many nodes, since this 
small part (1-2 % of hamilt) was not parallelized at all, but becomes 
important only for more than 64 nodes).

hns and full diagonalization was not modified.

fftw is used ONLY in lapw0. While the speedup is modest, memory 
requirement for large cases is drastically reduced.

Scott Beardsley schrieb:
> I've had a test out the new version of WIEN2k (09). I'm seeing a 
> disturbing trend. It looks like 9 is *two times slower* for both hybrid 
> and mpi runs starting at 2cpus[1]. Both of these runs were with binaries 
> compiled in an identical fashion*. Same compiler, same scalapack 
> library, same math library, same mpi library, and same hardware. Can 
> someone else double check my work and run the mpi-benchmark code also 
> for both 8 and 9?
> 
> * Compiled identically with the exception of FFTW. v9 seems to add fftw 
> whereas v8 didn't have that requirement. I've tried the mpi-benchmark 
> with both single and double precision fftw libraries but that didn't 
> affect the results.
> 
> Thanks,
> Scott
> ----------
> [1] http://wiki.cse.ucdavis.edu/support:hpc:software:wien2k#results
> _______________________________________________
> Wien mailing list
> Wien at zeus.theochem.tuwien.ac.at
> http://zeus.theochem.tuwien.ac.at/mailman/listinfo/wien

-- 
-----------------------------------------
Peter Blaha
Inst. Materials Chemistry, TU Vienna
Getreidemarkt 9, A-1060 Vienna, Austria
Tel: +43-1-5880115671
Fax: +43-1-5880115698
email: pblaha at theochem.tuwien.ac.at
-----------------------------------------