[gmx-developers] Profiling

[David van der Spoel]

On Fri, 31 Oct 2008, Roland Schulz wrote:

>Hi,
>
>as a follow up to the other tools thread (finding functions):
>
>Two profiling tools, which may be not so well known, and I think a very
>useful are:
>
>mpiP (mpip.sourceforge.net): It measures MPI time per code line
>(traceback possible). One only needs to compile with -g and link the mpip
>library. It gives a very quick overview why runtime is different on
>different clusters.
>
>hpctookit (hpctoolkit.org/): It uses sampling to pinpoint the number of
>spent cycles, caches misses, FLOPs to the code line (also for assembler).
>By keeping the sampling interval long enough the overhead is neglictable.
>
>In case someone is interested I'm happy to give some more information or
>examples. 
>
>Can you recommend any other tools besides those and MPE/jumpshot?
>
>Roland
>
Did ayone profile the serial code recently on a modern processor? Back in 
the day we used to make very good use of available hardware (Pentium 3). 
Butit would be interesting to first how efficient the code is now, (% of 
theoretical flops) and where the bottlenecks currently are.



>--
>ORNL/UT Center for Molecular Biophysics cmb.ornl.gov
>
>

David.
________________________________________________________________________
David van der Spoel, PhD, Professor of Biology
Dept. of Cell and Molecular Biology, Uppsala University.
Husargatan 3, Box 596,  	75124 Uppsala, Sweden
phone:	46 18 471 4205		fax: 46 18 511 755
spoel at xray.bmc.uu.se	spoel at gromacs.org   http://xray.bmc.uu.se/~spoel
++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++