Timing StatisticsΒΆ
The amrclaw and geoclaw Fortran codes provide some timing information at the end of a run (and also stored at the end of the file fort.amr in the output directory).
Typical output looks like this:
============================== Timing Data ==============================
Integration Time (stepgrid + BC + overhead)
Level Wall Time (seconds) CPU Time (seconds) Total Cell Updates
1 2.850 2.853 0.288E+07
2 13.214 41.552 0.373E+08
3 92.774 370.259 0.260E+09
total 108.838 414.664 0.301E+09
All levels:
stepgrid 101.440 392.473
BC/ghost cells 5.014 19.801
Regridding 40.508 40.447
Output (valout) 15.413 15.402
Total time: 165.483 472.470
Using 4 thread(s)
Note: The CPU times are summed over all threads.
Total time includes more than the subroutines listed above
=========================================================================
This was generated by running the code in $CLAW/amrclaw/examples/advection_3d_swirl with a third level of AMR added (and additional refinement factor 2) so that more grids are generated.
Note the following:
All times are in seconds.
For this example, more than 85% of the integration time is spent on the finest Level 3 grids.
OpenMP was used in this run, specifying OMP_NUM_THREADS=4. Grid patches are then split up between threads (see Using OpenMP). This test problem has enough grid patches that the time spent in stepgrid (where updating each patch using the finite volume method) uses roughly 1/4 as much wall time as CPU time, and similarly for filling ghost cells.
Regridding and output are done in serial mode, however, so the wall time for these portions agree with the CPU time.