[sbryngelson]

What

In non-MPI builds the scalar allreduce min/max/sum helpers and the gather helper in m_mpi_common never assigned their output argument (the body was only inside #ifdef MFC_MPI), so callers got an undefined result. Add the serial #else branches: var_glb = var_loc for the reductions and gathered_vector = my_vector for the gather.

Verification

Testing. Built with GPU offload (OpenMP, AMD) and ran the full test suite with this fix and seven other independent fixes applied together: 729 tests in 4 shards, 0 failures (two shards on MI300A, two on MI250X).

Prepared with Claude Code (AI-assisted); I reviewed the change and ran the tests.

[github-actions]

Claude Code Review

Head SHA: 054f641

Files changed:

• 1
• src/common/m_mpi_common.fpp

Findings:

• medium src/common/m_mpi_common.fpp: In s_mpi_gather_data, the new #else branch does gathered_vector = my_vector. gathered_vector is allocatable, intent(out) and is never explicitly allocated in the non-MPI path. This relies on F2003 automatic (re)allocation on assignment. gfortran and ifx enable it by default, but nvfortran needs -Mallocatable=03 and some Cray/legacy flag sets disable it. To be portable, write allocate (gathered_vector(counts)); gathered_vector = my_vector. This would also match the MPI branch, which allocates explicitly. Confidence: moderate, because it depends on the compiler flags in use.

The s_mpi_allreduce_sum, s_mpi_allreduce_min and s_mpi_allreduce_max #else copies (var_glb = var_loc) look correct for the serial case.

Note: .claude-review/context was empty, so I checked the hunks against the working-tree src/common/m_mpi_common.fpp.

[github-actions]

Lines of Code

File	Lines	Diff
src/common/m_mpi_common.fpp	1502	+8

Directory	Lines	Diff
common	10552	+8
total	47638	+8