Support one-sided halo for DNDarrays

[ClaudiaComito]

Due Diligence

• General:
  • title of the PR is suitable to appear in the Release Notes
• Implementation:
  • unit tests: all split configurations tested
  • unit tests: multiple dtypes tested
  • documentation updated where needed

Description

Enabling one-sided halo for #1419.

Usage (run on n>1 MPI processes):

# one-sided halo: from next rank only
a = ht.arange(100, split=0).reshape(10, 10)
a.get_halo(2, prev=False)
print("Test one-sided halo = ", a.array_with_halos)

Issue/s resolved: #

Changes proposed:

• introduced boolean kwargs prev and next for DNDarray.get_halo(), both True by default
• data communication to next MPI rank only if prev == True
• data communication to the previous MPI rank only if next == True

Type of change

• New feature (non-breaking change which adds functionality)

Memory requirements

Performance

Does this change modify the behaviour of other functions? If so, which?

no

[ClaudiaComito]

@FOsterfeld this PR is branched off your unfold branch, feel free to experiment here as needed. If you're ok with it, we can merge this into #1419

[github-actions]

Thank you for the PR!

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 91.73%. Comparing base (5cdd986) to head (63aa686).

Additional details and impacted files

@@                                             Coverage Diff                                             @@
##           features/1400-Implement_unfold-operation_similar_to_torch_Tensor_unfold    #1509      +/-   ##
===========================================================================================================
- Coverage                                                                    91.73%   91.73%   -0.01%     
===========================================================================================================
  Files                                                                           80       80              
  Lines                                                                        11692    11696       +4     
===========================================================================================================
+ Hits                                                                         10726    10729       +3     
- Misses                                                                         966      967       +1     

Flag	Coverage Δ
unit	91.73% <100.00%> (-0.01%)	⬇️

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[FOsterfeld]

@ClaudiaComito Thanks for this functionality, it made the implementation of #1419 much simpler! I have not tested the method specifically but all unit tests for unfold passed, so it should behave as intended when using prev=False.