Samuel Newcome
Samuel Newcome
Results from comparing the FLOPs and Hit Rate of master (`88b0d7a8` - technically out of date but the only changes since then were to documentation) and this branch (`ea9d52f3`). Also...
To summarise a recent change to this branch: - @thesamriel had an issue with constant log spam due to FLOP counting metrics not being implemented in the case that `AUTOPAS_LOG_FLOPS=ON`....
Just to summarise the even more recent change: FLOPLogger now logs "Not Implemented" in the CSV field if a functor does not implement `getNumFLOPs` or `getHitRate` This is heavy-handed but...
I changed the title to be clearer
> @SamNewcome @FG-TUM @thesamriel I have results from the individual runs (for single configurations of each container) > ### Runtime for 10000 steps of `SpinodalDecomposition_equilibration` case > LC VLC VL...
Further additional metrics may also be added to this. These include MFUPs/sec, GFLOPs, GFLOPs/sec (potentially the scale of the units may have to be changed).
Issue #707 must be address first before this!
Alternative solution is ideal as it allows the user to avoid overhead of counting FLOPs whilst still measuring times/energies per iteration.
Currently working on the alternative solution, as described in #695. The advantage of this, compared to the ideal solution, is that the "ideal" solution is too coarse-grained to be useful.
Currently a work in process. Will be done before GPU support (hopefully). Tunablitity should be done externally to AutoPas.