#142: phase: allow starting at non-zero phase #143
Conversation
| if (auto phase_work = phase_history.find(phase); | ||
| phase_work != phase_history.end()) { | ||
|
|
||
| // Get all objects at specified phase | ||
| auto& object_work_at_phase = phase_work->second.getObjectWork(); | ||
| // Get all objects at specified phase | ||
| auto& object_work_at_phase = phase_work->second.getObjectWork(); | ||
|
|
||
| for (auto const& [elm_id, obj_work] : object_work_at_phase) { | ||
| // fmt::print(" Object Id: {}\n", elm_id); | ||
| objects_at_phase.insert(std::make_pair(elm_id, obj_work)); | ||
| for (auto const& [elm_id, obj_work] : object_work_at_phase) { | ||
| // fmt::print(" Object Id: {}\n", elm_id); | ||
| objects_at_phase.insert(std::make_pair(elm_id, obj_work)); | ||
| } | ||
| } |
There was a problem hiding this comment.
We could add an else block with a warning or error message in case the phase is not found
There was a problem hiding this comment.
I don't think we need a warning or error either of these places. You could try to normalize your edges and find that for the phase you are checking there is no phase work.
| } | ||
| if (communications_to_add.empty()) { | ||
| return; | ||
| if (phaseWork.find(phase) != phaseWork.end()) { |
There was a problem hiding this comment.
Same here, more error handling may be helpful
| * | ||
| * \return the minimim phase in the data | ||
| */ | ||
| PhaseType getMinPhase() const { |
There was a problem hiding this comment.
Currently, we iterate through all ranks twice--first to get the number of phases (getNumPhases()) and then again here to get the minimum phase (getMinPhase()).
We could get both of these values (the number and the min) from a single for loop, e.g. by calling rank_info.getNumPhases() in this function and storing it as a member of info.
There was a problem hiding this comment.
getNumPhases actually just extracts the count of phases from rank 0 (it doesn't look at values to determine if there are holes or where it starts). It then loops through the other ranks as an error check to determine if they also have the same number of phases.
The count of phases is just the size of the unordered_map, which tells us nothing about the min. We must actually examine the values because they are unordered in the map.
Getting the number of phases should be a constant time operation of just looking at the size of the |
|
I have made min phase cached so it doesn't compute each time, which should improve performance. |
Fixes #142