IB: check every patch is marked in one pass and one reduction

[sbryngelson]

Summary

Follow-up to #1914, addressing its open Copilot review finding.

s_check_every_patch_marked looped over every global patch, and for each one swept every local cell and ran an MPI reduction. That is num_gbl_ibs grid sweeps plus num_gbl_ibs collectives at setup, and num_gbl_ibs counts every particle-cloud particle (up to 54,000 with the configured limit). The check now:

• makes one pass over the local cells, counting markers by patch id;
• reduces the whole count array in one collective;
• aborts if any patch's count is zero.

It also decodes each marker with s_decode_patch_periodicity before counting. The old check compared the raw marker with == gid, so the cells of a body wrapped across a periodic boundary did not count toward their patch, and a fully wrapped image could wrongly abort a valid run.

Counts are integer(kind=8), reduced with a new s_mpi_allreduce_integer_sum_vec in m_mpi_common (shaped like s_mpi_allreduce_min_vec, with the same non-MPI fallback). The abort and its message are unchanged, and the 16-line comment is cut to three.

The check runs once, from s_ibm_setup, not per time step. So the old cost was a setup-time cost, and the check catches a body that marks no cell at startup, not one that later leaves the domain.

Verification

• pre_process, simulation and post_process build on master d87a5fea with this change on Frontier, in all three configurations: CPU (--no-gpu), OpenMP offload (--gpu mp) and OpenACC (--gpu acc).
• ./mfc.sh precheck passes (all 7 gates).
• All 163 IBM and Example tests pass (CPU, run as CI runs them on Frontier). Every one of them runs this check at setup.
• Negative case: a radius-0.002 disc at a cell corner on a 50×50 grid (dx = 0.02), which contains no cell centre, still aborts on 2 ranks with the same message. The same disc centred on a cell centre marks that cell and runs normally.

AI disclosure

Written with Claude Code (Anthropic).

Acknowledgement

• I confirm this PR meets the above expectations and reflects my own understanding and real-world context.

[Copilot]

Warning

Copilot couldn't run its full agentic review because it didn't start before the timeout. Make sure your repository has a runner available, or add a copilot-code-review.yml file specifying one with the runs-on attribute. See the docs for more details.

Copilot review overview

Review effort: Lite
Findings: 4 · 1 · 3

Open (8)

• Counts are being accumulated and reduced in real(wp), then compared via exact equality to… · New
• Counts are being accumulated and reduced in real(wp), then compared via exact equality to… · New
• Counts are being accumulated and reduced in real(wp), then compared via exact equality to… · New
• Counts are being accumulated and reduced in real(wp), then compared via exact equality to… · New
• This introduces an unconditional MPI allreduce call where the previous code was guarded by `#ifdef… · New
• The per-patch counts are modeled as 2D arrays with a hard-coded second dimension of 1, which… · New
• The per-patch counts are modeled as 2D arrays with a hard-coded second dimension of 1, which… · New
• The per-patch counts are modeled as 2D arrays with a hard-coded second dimension of 1, which… · New

What changed in this PR

Optimizes s_check_every_patch_marked to avoid per-patch grid sweeps and collectives by counting marked cells per patch in a single local pass and performing a single global reduction, while correctly handling periodic-encoded markers.

Changes:

• Replace num_gbl_ibs repeated sweeps/reductions with one local sweep that bins counts by patch id.
• Perform a single collective reduction over the full per-patch count array.
• Decode marker periodicity before counting so wrapped images count toward their owning patch.

File	Description
src/​simulation/​m_ibm.fpp	Refactors “every patch marked” validation to single-pass counting + single reduction and adds periodicity-aware marker decoding.

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

[codecov]

Codecov Report

❌ Patch coverage is 84.61538% with 2 lines in your changes missing coverage. Please review.
✅ Project coverage is 62.82%. Comparing base (ed7a238) to head (97d7106).
⚠️ Report is 1 commits behind head on master.

Files with missing lines	Patch %	Lines
src/common/m_mpi_common.fpp	66.66%	1 Missing ⚠️
src/simulation/m_ibm.fpp	90.00%	0 Missing and 1 partial ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##           master    #1933      +/-   ##
==========================================
- Coverage   62.82%   62.82%   -0.01%     
==========================================
  Files          86       86              
  Lines       22394    22398       +4     
  Branches     3305     3305              
==========================================
+ Hits        14070    14072       +2     
- Misses       6071     6072       +1     
- Partials     2253     2254       +1     

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

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[github-actions]

Lines of Code

File	Lines	Diff
src/common/m_mpi_common.fpp	1494	+10
src/simulation/m_ibm.fpp	1442	+3

Directory	Lines	Diff
common	10426	+10
simulation	27979	+3
total	46933	+13