MPI Configuration configure (8MPI)

NAME

configure - GridMPI configurer

SYNOPSIS

configure [--enable-debug] [--enable-threads] [--disable-dlload] [--disable-prof-feature] [--disable-unix] [--disable-onesided] [--with-vendormpi=no/ibmmpi/fjmpi/mpisx/mpich/mpich2/intelmpi] [--with-binmode=no/32/64] [--with-libckpt[=path]] [--with-libckpt-include=path] [--with-libckpt-lib=path] [--with-score] [--with-libpsp]

OPTIONS

Following options are supported:

--enable-debug
Specifies to enable debugging in compiling the GridMPI library.

--enable-threads
Specifies to enable multi-threading. THREAD SUPPORT IS ONLY LIGHTLY TESTED.

--disable-dlload
Specifies not to use dynamic library loading. It will be specified when the OS does not have libdl.so.

--disable-prof-feature
Specifies to not use MPI profiling feature. It is automatically disabled, if the compiler does not support #pragma weak (weak symbol).

--disable-unix
Specifies not to use Unix domain sockets for communication on a local host. Without this option, MPI processes on the same host may communicate with the loopback network.

--disable-onesided
Specifies not to use one-sided communication of MPI-2. Currently, one-sided is only supported in Linux and compilation fails on other OS. One-sided is automatically disabled, if an ioctl operation F_SETSIG is not defined.

--with-vendormpi=no/ibmmpi/fjmpi/mpisx/mpich/mpich2/intelmpi
Specifies to use Vendor MPI, in which a vendor-supplied MPI library is used in underlying communication. This option is supported for the following platforms: ibmmpi for IBM-MPI on AIX machines (to use the SP switches), fjmpi for Fujitsu-MPI on Solaris/SPARC64 machines, mpisx for NEC-MPI on NEC SX6 machines, mpich for MPICH-based MPI, mpich2 for MPICH2-based MPI, and intelmpi for Intel MPI (fjmpi and mpisx are under development).

--with-binmode=no/32/64
Specifies to use the default binary mode, or 32bit or 64bit binary mode in compilation. This is only meaningful when the compiler supports binary mode by a compiler option (e.g., -m32/-m64 in GCC/IA32, and -q32/-q64 in IBM XL compilers). If the compiler does not support binary mode option, use no. To use 32bit or 64bit option, configure-make-install procedure shall be performed twice, once for 32bit and once for 64bit. After that, mpicc accepts the options -q32/-q64 to switch the binary mode.

--with-libckpt[=path]
Specifies to use checkpoint/restart. The paths to the library and include files are specified by this option: path/include and path/lib are used. The paths can be specified separately by --with-libckpt-include and --with-libckpt-lib. Checkpointing is enabled by default on Linux Fedora Core 3 and 4 on IA32 (32bit) machines. CHECKPOINTING IS SUPPORTED ONLY LINUX/IA32 PLATFORMS, CURRENTLY.

--with-libckpt-include=path
Specifies libckpt include file path. See above.

--with-libckpt-lib=path
Specifies libckpt library path. See above.

--with-libpsp
Specifies to use the PSPacer module. NOTE: PSPacer is a precise software pacing mechanism for Linux. IT IS JUST EXPERIMENTAL, CURRENTLY.

--with-score
DO NOT SPECIFY THIS. NOT SUPPORTED YET. Specifies to use SCore for cluster local communication. It only passes this option to the YAMPII configurer in the subordinate "yampii" directory. NOTE: SCore is a cluster system software with a very fast communication library.

($Date: 2006/09/12 09:44:38 $)