This PR changes our performance measurement methodology: it measures simulation times only instead of population generation times + simulation times + stride cruft times. (Professor Broeckhove suggested this change during our beta presentation.)
The numbers in the performance comparison table of our user manual's parallelization chapter have changed, though the overall picture painted by that table remains unchanged—parallelization results in a modest speedup at best. On the plus side, the difference in terms of performance between the OpenMP, TBB and STL implementations is clearer now. Turns out that our homebrew STL implementation is actually better than the TBB implementation. :tada:
I've also taken the liberty to change the formatting of the (timing) information printed in run_stride.cpp. Users should have a better idea of what's going on now.
Howdy partners.
This PR changes our performance measurement methodology: it measures simulation times only instead of population generation times + simulation times + stride cruft times. (Professor Broeckhove suggested this change during our beta presentation.)
The numbers in the performance comparison table of our user manual's parallelization chapter have changed, though the overall picture painted by that table remains unchanged—parallelization results in a modest speedup at best. On the plus side, the difference in terms of performance between the OpenMP, TBB and STL implementations is clearer now. Turns out that our homebrew STL implementation is actually better than the TBB implementation. :tada:
I've also taken the liberty to change the formatting of the (timing) information printed in
run_stride.cpp. Users should have a better idea of what's going on now.