SPEC MPI2007 Flag Description for SGI MPT and Intel(R) Compiler 11.1

Sections

Selecting one of the following will take you directly to that section:


Optimization Flags


Portability Flags


Compiler Flags


Other Flags


System and Other Tuning Information

SGI MPT 1.25 options and environment variables

Job startup command and options

mpiexec_mpt [ global_opts ] local_opts cmd [ : local_opts cmd ] ...

The mpiexec_mpt command launches a Message Passing Toolkit (MPT) MPI program in a batch scheduler-managed cluster environment. mpiexec_mpt uses the list of cluster nodes it receives from the batch scheduler to generate and issue an appropriate mpirun command to launch the multi-node job.

-n <# of processes> or -np <# of processes>

Use this option to set the number of MPI processes to run the current arg-set.

mpiexec [ global_opts ] local_opts cmd [ : local_opts cmd ] ...

The PBS Pro's mpiexec command provides the standard mpiexec interface on the Altix running ProPack 4 or greater. It provides equivalent functionality to mpiexec_mpt.

Environment variables

MPI_REQUEST_MAX

Determines the maximum number of nonblocking sends and receives that can simultaneously exist for any single MPI process. MPI generates an error message if this limit (or the default, if not set) is exceeded. Default: 16384

MPI_TYPE_MAX

Determines the maximum number of data types that can simultaneously exist for any single MPI process. MPI generates an error message if this limit (or the default, if not set) is exceeded. Default: 1024

MPI_BUFS_THRESHOLD

Determines whether MPT uses per-host or per-process message buffers for communicating with other hosts. Per-host buffers are generally faster but for jobs running across many hosts they can consume a prodigious amount of memory. MPT will use per- host buffers for jobs using up to and including this many hosts and will use per-process buffers for larger host counts. Default: 64

MPI_DSM_DISTRIBUTE (toggle)

If set, NUMA job placement mode is activated. This mode ensures that each MPI process gets a unique CPU and physical memory on the node with which that CPU is associated. Currently, the CPUs are chosen by simply starting at relative CPU 0 and incrementing until all MPI processes have been forked. Default: Not enabled

MPI_IB_RAILS

If the MPI library uses the IB driver as the inter-host interconnect it will by default use a single IB fabric. If this is set to 2, the library will try to make use of multiple available separate IB fabrics and split MPI traffic across them. Default: 1

Other Tuning Information

ulimit -s unlimited

Removes limits on the maximum size of the automatically- extended stack region of the current process and each process it creates.