s_mpi_allreduce_integer_sum integer overflow #1601

[danieljvickers]

Description

Fixes integer overflow when counting the number of ghost points at exascale. Simple modification made to the s_mpi_allreduce_integer_sum subroutines to accept 8-byte integers instead of 4-byte.

Closes #1601

Type of change (delete unused ones)

• Bug fix

Testing

I attempted to recreate the bug locally by using 2 ranks, setting both of their sums to just-below the integer limit, and summing them together, which would normally create an overflow. But instead, the correct value was retrieved.

Checklist

__Check these like this [x] to indicate which of the below applies.

[github-actions]

Claude Code Review

Head SHA: 0b38165

Files changed:

• 2
• src/common/m_mpi_common.fpp
• src/simulation/m_ibm.fpp

Findings

[Correctness / Compiler Portability] Breaking API change in a shared common module

s_mpi_allreduce_integer_sum in src/common/m_mpi_common.fpp changes its dummy argument types from integer to integer(kind=8). CLAUDE.md explicitly calls out src/common/ as having "wide blast radius — shared by ALL three executables." This PR only updates callers in m_ibm.fpp. If any other call sites in src/pre_process/, src/post_process/, or other simulation modules pass default-integer arguments to this routine, they will now get a type-mismatch compile error. The two updated call sites correctly use int(..., 8) casts, but the changed signature is a breaking API change that must be verified across all three executables.

-        integer, intent(in)  :: var_loc
-        integer, intent(out) :: var_glb
+        integer(kind=8), intent(in)  :: var_loc
+        integer(kind=8), intent(out) :: var_glb

[Compiler Portability] MPI_INTEGER8 is an optional MPI named type

The MPI_ALLREDUCE call is changed to use MPI_INTEGER8:

-        call MPI_ALLREDUCE(var_loc, var_glb, 1, MPI_INTEGER, MPI_SUM, MPI_COMM_WORLD, ierr)
+        call MPI_ALLREDUCE(var_loc, var_glb, 1, MPI_INTEGER8, MPI_SUM, MPI_COMM_WORLD, ierr)

Per the MPI-3 standard (§6.9.2), MPI_INTEGER8 is an optional named datatype — conforming implementations are not required to define it. All four CI-gated compilers' MPI libraries (OpenMPI/MPICH for gfortran, NVHPC MPI for nvfortran, Cray MPICH for Cray ftn, Intel MPI for ifx) do provide it in practice, but a portable alternative that is always available is MPI_INT64_T (required since MPI-2.2 when the host C type int64_t is available).

[codecov]

Codecov Report

❌ Patch coverage is 80.00000% with 1 line in your changes missing coverage. Please review.
✅ Project coverage is 60.52%. Comparing base (b4be438) to head (0b38165).
⚠️ Report is 1 commits behind head on master.

Files with missing lines	Patch %	Lines
src/simulation/m_ibm.fpp	75.00%	1 Missing ⚠️

Additional details and impacted files

@@           Coverage Diff           @@
##           master    #1605   +/-   ##
=======================================
  Coverage   60.51%   60.52%           
=======================================
  Files          83       83           
  Lines       19905    19904    -1     
  Branches     2950     2950           
=======================================
  Hits        12046    12046           
+ Misses       5866     5865    -1     
  Partials     1993     1993           

☔ 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.