#941 added subphase support to loadComm

[bradybray]

fixes #941 implementing subphases in loadComm

[codecov]

Codecov Report

Merging #1081 into develop will increase coverage by 0.00%.
The diff coverage is 91.30%.

@@           Coverage Diff            @@
##           develop    #1081   +/-   ##
========================================
  Coverage    75.60%   75.61%           
========================================
  Files          719      719           
  Lines        27239    27255   +16     
========================================
+ Hits         20595    20609   +14     
- Misses        6644     6646    +2     

Impacted Files	Coverage Δ
src/vt/vrt/collection/balance/elm_stats.h	100.00% <ø> (ø)
src/vt/vrt/collection/balance/node_stats.h	100.00% <ø> (ø)
src/vt/vrt/collection/balance/node_stats.cc	74.72% <66.66%> (-0.28%)	⬇️
src/vt/vrt/collection/balance/elm_stats.cc	77.65% <100.00%> (+2.36%)	⬆️
src/vt/vrt/collection/balance/elm_stats.impl.h	100.00% <100.00%> (ø)

[PhilMiller]

So, to carry this the rest of the way through, we'll need a per-subphase structure alongside node_comm_ in node_stats.h, and appropriate recording into it in NodeStats::addNodeStats. That means ElementStats will need to record it in ElementStats::recv* to provide the data later

[bradybray]

@PhilMiller

So, to carry this the rest of the way through, we'll need a per-subphase structure alongside node_comm_ in node_stats.h, and appropriate recording into it in NodeStats::addNodeStats. That means ElementStats will need to record it in ElementStats::recv* to provide the data later

vt/src/vt/vrt/collection/balance/node_stats.h

Lines 257 to 267 in 1a4c8bc

	std::unordered_map<PhaseType, SubphaseLoadMapType> node_subphase_data_;
	/// Local migration type-free lambdas for each object
	std::unordered_map<ElementIDType,MigrateFnType> node_migrate_;
	/// Map of temporary ID to permanent ID
	std::unordered_map<ElementIDType,ElementIDType> node_temp_to_perm_;
	/// Map of permanent ID to temporary ID
	std::unordered_map<ElementIDType,ElementIDType> node_perm_to_temp_;
	/// Map from element temporary ID to the collection's virtual proxy (untyped)
	std::unordered_map<ElementIDType,VirtualProxyType> node_collection_lookup_;
	/// Node communication graph for each local object
	std::unordered_map<PhaseType, CommMapType> node_comm_;

So I see there is a node_subphase_data_ structure on line 257. Is this not a sufficient structure for this implementation?

Ah, as I wrote this I saw something on line 267. Would
std::unordered_map<SubphaseType, CommMapType> node_comm_s; be sufficient for the structure?
I'll start peeking into the other files to provide the additional implementation. I'm sure I'll have some more comments to follow ;)

[lifflander]

Ah, as I wrote this I saw something on line 267. Would
std::unordered_map<SubphaseType, CommMapType> node_comm_s; be sufficient for the structure?
I'll start peeking into the other files to provide the additional implementation. I'm sure I'll have some more comments to follow ;)

I think you will to refactor to be std::unordered_map<PhaseType, std::unordered_map<SubphaseType, CommMapType>> node_comm_;

[bradybray]

@lifflander @PhilMiller /usr/bin/ld: final link failed: No space left on device This directory gets full?

[PhilMiller]

It looks like the last commit, "Checkpoint before rebase", reverts all of the intended changes in the rest of the PR.

[bradybray]

It looks like the last commit, "Checkpoint before rebase", reverts all of the intended changes in the rest of the PR.

Ok. Do I revert back to the previous commit then push?

[PhilMiller]

I think you would want to git reset to the previous commit

[bradybray]

I think you would want to git reset to the previous commit

Yep that is what I ended up doing. I force pushed the commit.

[bradybray]

I think you would want to git reset to the previous commit

As for the conflicts; which side of the conflict do I take? I assume I should be taking the side from develop and not my own branch? The conflicts appear to be from the runtime library.

[PhilMiller]

There shouldn't be conflicts at all - your changes don't touch those files.