Functions
	arg (name, default)

	one ()

	gather (a)

Variables
	CASE_DIR

	CKPT

	N = int(sys.argv[1])

str	WARM = int(sys.argv[sys.argv.index("--warm") + 1]) if "--warm" in sys.argv else 5

	RTOL = float(arg("--rtol", 1e-8))

	DUMP = arg("--dump", None)

	comm = MPI.COMM_WORLD

	rank = comm.Get_rank()

int	SINGLE = 1

	S = case.CELLS_PER_D

	NX

	NY

	NZ

	step

	t

	blocks

	vel

	pres

	series

	org

	siz

	sl = tuple(slice(o, o + n) for o, n in zip(org, siz))

	s = pf.Solver(*siz)

	vol = comm.allreduce(sum(b["volume"] for b in s.diagnostics.vof_block_stats()), op=MPI.SUM)

	alpha = vol / (NX * NY * NZ)

	rho_av = case.RHO_L + (case.RHO_G - case.RHO_L) * alpha

	_closure = getattr(s, "set_superficial_velocity", None) or getattr(s, "set_bulk_velocity", None)

	HOST_FLUX = _closure is None

	w0 = time.perf_counter()

	it

	dts

	wall = time.perf_counter() - w0

	flush

dict	fields

tuple	cols

	iters

Detailed Description

Distributed timing of the bubble column from ckpt_t43 (host MPI x OpenMP; the VoF
step-performance harness, WO-0 -- doc/vof_step_performance_design.md).

    export OMPI_MCA_hwloc_base_binding_policy=none
    PYTHONPATH=<PECLET_FLOW_MPI build> OMP_NUM_THREADS=3 OMP_PROC_BIND=false \
        mpirun -np 8 --bind-to none python tests/study/vof_perf/run_mpi.py N [--warm W] [--cheb]
            [--rtol R] [--dump out.npz] [--ckpt PATH] [--case-dir DIR]

Same case as run_peclet.build(), on this rank's ORB block; set_superficial_velocity closes the
column (a build older than it shifts u on the host: global mean via Allreduce). --ckpt and
--case-dir default as in prof.py. Prints one line on rank 0: ms/step and mean pressure iterations.

  --rtol R    MG-PCG relative tolerance (default 1e-8: the case's registered tolerance, register
              2026-10-03; the solver default stays 1e-10)
  --dump F    save u, v, w, p, C, the timed steps' dt (dts) and pressure iterations (iters) in
              prof.py's format (compare with cmp.py); the fields are gathered onto rank 0 at np > 1,
              the VoF block colours (col<id>) are stored at np = 1 only

np = 1 is SINGLE-RANK: at size 1 neither mpi_block nor init_mpi is called (sec. 14.3 H-0 of
doc/vof_step_performance_design.md). init_mpi at size 1 keeps the distributed path on purpose (the
register's np = 1 bit-exactness gate needs it), so a one-rank benchmark that called it timed the
distributed code -- GridHalo self-copies, the box smoother, no single-rank fusions.

Function Documentation

◆ arg()

run_mpi.arg	(	name,
		default
	)

Definition at line 33 of file run_mpi.py.

◆ one()

run_mpi.one ( )

Definition at line 95 of file run_mpi.py.

References one().

Referenced by one().

◆ gather()

run_mpi.gather ( a )

Definition at line 122 of file run_mpi.py.

Variable Documentation

◆ CASE_DIR

run_mpi.CASE_DIR

Initial value:

1= arg("--case-dir", os.path.expanduser(

2 "~/Codes/peclet-examples/benchmarks/bubble-column/scripts"))

Definition at line 37 of file run_mpi.py.

◆ CKPT

run_mpi.CKPT

Initial value:

1= arg("--ckpt", os.path.expanduser(

2 "~/Codes/bubble_column_perf/ckpt_t43.npz"))

Definition at line 39 of file run_mpi.py.

◆ N

run_mpi.N = int(sys.argv[1])

Definition at line 46 of file run_mpi.py.

◆ WARM

str run_mpi.WARM = int(sys.argv[sys.argv.index("--warm") + 1]) if "--warm" in sys.argv else 5

Definition at line 47 of file run_mpi.py.

◆ RTOL

run_mpi.RTOL = float(arg("--rtol", 1e-8))

Definition at line 48 of file run_mpi.py.

◆ DUMP

run_mpi.DUMP = arg("--dump", None)

Definition at line 49 of file run_mpi.py.

◆ comm

run_mpi.comm = MPI.COMM_WORLD

Definition at line 50 of file run_mpi.py.

◆ rank

run_mpi.rank = comm.Get_rank()

Definition at line 51 of file run_mpi.py.

◆ SINGLE

int run_mpi.SINGLE = 1

Definition at line 52 of file run_mpi.py.

◆ S

run_mpi.S = case.CELLS_PER_D

Definition at line 53 of file run_mpi.py.

◆ NX

run_mpi.NX

Definition at line 54 of file run_mpi.py.

◆ NY

run_mpi.NY

Definition at line 54 of file run_mpi.py.

◆ NZ

run_mpi.NZ

Definition at line 54 of file run_mpi.py.