61 lines
1.7 KiB
Plaintext
61 lines
1.7 KiB
Plaintext
.\" Man page for mpimcexec
|
|
.\"
|
|
|
|
.TH MPIMCEXEC 1 "@MCKERNEL_RELEASE_DATE@" "Version @MCKERNEL_VERSION@" MCKERNEL @MCKERNEL_VERSION@"
|
|
.SH NAME
|
|
mpimcexec \- run an MPI application on McKernel
|
|
.\"
|
|
|
|
.\" ---------------------------- SYNOPSIS ----------------------------
|
|
.SH SYNOPSIS
|
|
.B mpimcexec \fR [\fIoptions\fR] \fI<command>\fR
|
|
|
|
.\" ---------------------------- DESCRIPTION ----------------------------
|
|
.SH DESCRIPTION
|
|
mpimcexec is a wrapper script for running MPI applications on McKernel.
|
|
It internally calls mpiexec to spawn mcexec on compute nodes, which in
|
|
turn runs \fI<command>\fR on McKernel. mpimcexec specifies a number of
|
|
mcexec arguments that enable high performance execution.
|
|
|
|
.\" ---------------------------- OPTIONS ----------------------------
|
|
.SH OPTIONS
|
|
|
|
.TP
|
|
.B -ppn N, --ppn N, --ranks-per-node N
|
|
Specify the number of MPI ranks per node. This argument is required.
|
|
.TP
|
|
.B -n N, --n N, --ranks N
|
|
Specify the number of total MPI ranks.
|
|
e.g.,
|
|
$ mpimcexec -n 32 -ppn 4 ./a.out
|
|
.br
|
|
In the above example, 32 MPI processes are invoked
|
|
on eight compute nodes each of which has four processes.
|
|
.TP
|
|
.B --nodes N
|
|
Specify the number of compute nodes.
|
|
By default, all nodes, specified by "PJM --mpi proc" option, are used.
|
|
.TP
|
|
.B --env, -env
|
|
Pass an additional environment variable
|
|
.TP
|
|
.B -m N, --numa N
|
|
Specify preferred NUMA node.
|
|
.TP
|
|
.B -h <file name>, ---hostfile <file name>
|
|
Specify a host file for MPI.
|
|
.TP
|
|
.B --help
|
|
Show help message.
|
|
|
|
.PP
|
|
.\" ---------------------------- SEE ALSO ----------------------------
|
|
.SH SEE ALSO
|
|
\fBmcexec\fR (1), \fBmpiexec\fR (1)
|
|
|
|
.\" ---------------------------- AUTHORS ----------------------------
|
|
.SH AUTHORS
|
|
Copyright (C) 2018 McKernel Development Team, RIKEN, Japan
|
|
|
|
|