| mpicc(1) | LAM TOOLS | mpicc(1) |
mpicc, mpiCC / mpic++ - Compile LAM/MPI C/C++ programs.
mpicc [-showme|-showme:compile|-showme:link] ...
mpiCC [-showme|-showme:compile|-showme:link] ...
mpic++ [-showme|-showme:compile|-showme:link] ...
See cc(1) and CC(1) (or whatever your underlying C/C++ compilers are) for all other options.
mpicc and mpiCC (mpic++ is a synonym for mpiCC provided for filenames that do not support case-sensitive filenames) are convenience wrappers for the local native C and C++ compilers. Translation of a LAM program requires the linkage of the LAM specific libraries which may not reside in one of the standard search directories of ld(1). It also often requires the inclusion of header files what may also not be found in a standard location. mpicc passes its arguments to the local native C compiler along with the -I, -L and -l options required by LAM programs. This includes all necessary options for ROMIO and/or C++ bindings support (if ROMIO/C++ support was included when LAM was compiled).
mpiCC (and therefore mpic++) is similar, but invokes the native C++ compiler instead.
The LAM Team strongly encourages using mpicc and mpiCC instead of attempting to link to the LAM libraries manually. This allows the specific implementation of LAM to change without forcing changes to linker directives in users' Makefiles (the specific set of underlying LAM libraries has already changed multiple times, and will likely change again in future versions).
Indeed, since mpicc/mpiCC are very thin wrappers on top of an underlying compiler, there are very, very few compelling reasons not to use mpicc/mpiCC. When it is not possible to use mpicc/mpiCC, the -showme:compile and -showme:link arguments should be used instead. For example:
shell$ cc -c file1.c `mpicc -showme:compile`
shell$ cc -c file2.c `mpicc -showme:compile`
shell$ cc file1.o file2.o `mpicc -showme:link` -o my_mpi_program
By default, mpicc uses the C compiler that was selected when LAM was configured (with the --with-cc flag to ./configure, or by setting the environment variable CC before ./configure was invoked) as the local native C compiler, but this can be overridden by the LAMMPICC environment variable (an older name for this environment variable is LAMHCC -- this also still works, but its use is deprecated).
Likewise, mpiCC uses the C++ compiler that was selected when LAM was configured (with the --with-cpp flag to ./configure, or by setting the environment variable CXX before invoking ./configure) by default, but this can be overridden by the LAMMPICXX environment variable (an older name for this environment variable is LAMHCP -- this also still works, but its use is deprecated).
If the environment variable LAMHOME is set, mpicc and mpiCC will use its value as the location of the LAM installation directory instead of the value that was compiled into mpicc/mpiCC. This means that mpicc/mpiCC will use the value of LAMHOME as the base to create the -I and -L arguments that are passed to the lower-level compiler, not the installation directory that was supplied when mpicc/mpiCC were created. This is almost always a Bad Idea.
The use of LAMHOME is discouraged except for some rare configuration cases in oddly networked sites (in which case your system administrator should probably set this up), and for advanced users with multiple LAM/MPI installations who really know what they are doing; if the LAMHOME environment variable is unintentionally left set, it can lead to tremendous user confusion. For example, if LAMHOME points to LAM installation A, but the user's PATH points to LAM installation B, then even though B's mpicc will be used, the user program will be compiled and linked against LAM installation A.
The LAMHOME environment variable is mainly only left in place for backwards compatibility; it is not required for normal functioning of LAM/MPI. The LAM Team discourages the use of the LAMHOME environment variable, and instead advocates simply setting the PATH properly to switch between multiple LAM/MPI implementations.
Previous versions of LAM encouraged the use of hcc and hcp to compile LAM and/or MPI C and C++ applications, respectively. In very old versions of LAM, hcc and hcp did not automatically add -lmpi to the command line. hcc and hcp were eventually deprecated and replaced with mpicc and mpiCC (or mpic++ on case-insensitive filesystems). The executables hcc and hcp are now simply symbolic links to mpicc and mpic++, respectively, just in case there's anyone out there that still uses those names. It should be harmless to pass in the additional -lmpi; mpicc and mpic++ should silently do the Right Thing (only link in the MPI library once).
| July, 2007 | LAM 7.1.4 |