make origin and target write size match in the mpi_put call

[clevelam]

Background

• This fixes the MPI_Put call in quick_index to specify the target buffer size equal to the origin buffer size. Originally @clevelam thought that it was supposed to be the off processor buffer size. This might also explain why we had some issues at large mpi_put writes (over 10K). For now we will leave in the chunk writing to save temporary memory space.

Purpose of Pull Request

• Fixes re-git issue #1355

Description of changes

• Remove un-needed off processor buffer size
• Make target buffer size equal to the origin buffer size in the MPI_Put call used by quick_index

Status

• Reference:
  • Pre-Merge Code Review
  • CDash Status
• Checks
  • [x] Travis/Appveyor CI checks pass
  • [x] Code coverage does not decrease
  • [x] Valgrind test passes
  • [x] Toss3 checks pass
  • [x] Trinitite checks pass
  • [x] Code reviewed/approved, sufficient DbC checks, testing, documentation

[KineticTheory]

@alexrlongne Can you review this change?

[KineticTheory]

I'm curios to see if this change will fix the test failure on MSVC. I'll try it now.

[codecov[bot]]

Codecov Report

Merging #1117 (a6a6456) into develop (fc46b6d) will decrease coverage by 0.0%. The diff coverage is 100.0%.

@@            Coverage Diff            @@
##           develop   #1117     +/-   ##
=========================================
- Coverage     89.0%   88.9%   -0.1%     
=========================================
  Files          373     373             
  Lines        19573   19568      -5     
=========================================
- Hits         17420   17415      -5     
  Misses        2153    2153             

[alexrlongne]

LGTM

[alexrlongne]

I don't understand the drop in coverage so I'm good to merge whenever.

[KineticTheory]

I don't understand the drop in coverage so I'm good to merge whenever.

I don't like the way CodeCov does this. The PR has 5 less lines of code and so it has 5 fewer lines 'hit' by the coverage checker. This equates to a loss of coverage.

[KineticTheory]

I'm curios to see if this change will fix the test failure on MSVC. I'll try it now.

It still fails. We'll omit the 3 rank tests when using MSVC.