Dear Lammps users,
by looking at the documentation, the "Output" part of the "MPI task
section" of the "LAMMPS screen output" stands only for the time spent on
dump and thermo.
http://lammps.sandia.gov/doc/Section_start.html#lammps-screen-output
However, when these dump or thermo commands induce compute (stress,
voronoi, ...), is the time spent calculating these properties included
in the "Output" part or in the "Modify" part of the breakdown of the CPU
run time?
Regards!
Julien.
Dear Lammps users,
by looking at the documentation, the "Output" part of the "MPI task
section" of the "LAMMPS screen output" stands only for the time spent on
dump and thermo.http://lammps.sandia.gov/doc/Section_start.html#lammps-screen-output
However, when these dump or thermo commands induce compute (stress,
voronoi, ...), is the time spent calculating these properties included
in the "Output" part or in the "Modify" part of the breakdown of the CPU
run time?
that is not easy to say, because these computations may be queried at
different times but most are set up to do the computation only once
per timestep and cache the results.
in some cases, e.g. for compute stress/atom or compute */tally the
majority of the calculation is performed during the Pair section.
if you want more detailed information on where time is spent, you'll
be better off using a profiling tool. for a first look, i usually use
"perf", which provides kernel level profiling without the need for
instrumentation and next to no overhead. just run lammps with "perf
record" on a sufficiently small test example and then analyse the run
with "perf report".
axel.