📄 readme.parallel
字号:
July 2000 INSTRUCTIONS FOR USING PARALLEL XPDC1 /Emi Kawamura/I. DOWNLOAD AND INSTALL MPI The first step in using the parallel version of XPDC1 is to download and install the MPI libraries on your system. 1) Go to the website http://www-unix.mcs.anl.gov/mpi/mpich/ and download the file mpich.tar.gz which contains the MPICH portable implementation of MPI. You can also read the Installation Guide and User's Guide at this site in HTTP format. 2) gunzip the file and unarchive it, you will then create a directory called mpich-1.2.0 (or the name of the current version). 3) For version 1.2.0 compiling on a LINUX machine, I used the following steps to compile and install MPICH: First enter the mpich-1.2.0 directory, then type: a) configure --with-device=ch_p4 -rsh=ssh b) make c) make install PREFIX=/usr/local/mpich-1.2.0 In step (a), I configure MPI with the ch_p4 device which assumes a default network and makes no assumption about memory sharing. If your system is configured to use the secure shell ssh, you should also use the option "-rsh=ssh"; otherwise, omit this option. To see all the configuration options, type "configure -usage". In step (b), I compile the code, and in step (c), I install it to the directory /usr/local/mpich-1.2.0. You may choose another installation directory. Also note that you must be root to install to a system directory. 4) Next you may also wish to install the shared memory MPI (ch_shmem) library. If you do not, skip this section. For version 1.2.0 compiling on a LINUX machine, I used the following steps to install the ch_shmem MPI library: First copy the mpich-1.2.0 directory to mpich-1.2.0sh. Enter the mpich-1.2.0sh directory and type: a) make clean b) configure --with-device=ch_shmem -rsh=ssh c) make d) make install PREFIX=/usr/local/mpich-1.2.0sh In step (a), I clean up any object files left over by a previous installation. In step (b), I configure MPI with the ch_shmem device which assumes a shared memory (SMP) configuration. Also, as mentioned before, if your system is not configured to use the secure shell ssh, omit the second configuration option. In step (c), the code is compiled, and in (d) the library is installed in /usr/local/mpich-1.2.0sh. As mentioned before, you may choose another directory, and you must be root to intall to a system directory.II DOWNLOAD AND INSTALL LATEST XGRAFIX LIBRARY Obtain the latest version of XGRAFIX from our CVS repository. This version has minor revisions to enable the parallel version of XPDC1 to work. 1) First edit the Imakefile by UNcommenting the line: MPI_DEFINE = -DMPI_1D 2) Type "xmkmf" to create the Makefile. 3) Type "make" to create the library libXGC250.a 4) Rename libXGC250.a to libXGC250P.a, and move it to the library where you keep all other xgrafix libraries, e.g., /usr/local/lib/xgrafix. 5) Finally, replace the old "xgrafix.h" in /usr/local/include (or wherever it resides) with the current version of xgrafix.hIII DOWNLOAD AND INSTALL PARALLEL XPDC1 Obtain the latest version of XPDC1 from our CVS repository. This includes the parallel version of XPDC1. 1) First read makefile.par.p4, and edit the file so that all the directory locations are correct. 2) Next type: "make -f makefile.par.p4" This will create the executable "xpdc1.par.p4" which uses the default network (ch_p4) MPI library. If you have installed the shared memory (ch_mem) library, and wish to create an executable which uses this library, then, 3) First read makefile.par.sh, and edit the file so that all the directory locations are correct. 4) Next type: "make -f makefile.par.sh" This will create the executable "xpdc1.par.sh" which uses the ch_mem library.IV USING PARALLEL XPDC1 After completing parts I, II, and III, you are now ready to use parallel xpdc1. 1) CH_P4 version: a) Let $(DEST_P4) = <Location of ch_p4 MPI library>. (For the example in I, $(DEST_P4)=/usr/local/mpich-1.2.0) b) Check the file $(DEST_P4)/share/machines.LINUX to make sure the information is correct. c) Type, $(DEST_P4)/bin/mpirun -np <num proc.'s> xpdc1.par.p4 -i <inputfile> -dp <dump period> -d <dumpfile> -s <num. steps> -nox If you do not wish to use the default machines listed in $(DEST_P4)/share/machines.LINUX, the mpirun command has an option "-machinefile <your machinefile>". 2) CH_SHMEM version: a) Let $(DEST_SH) = <Location of ch_mem MPI library>. (For the example in I, $(DEST_SH)=/usr/local/mpich-1.2.0sh) b) Type, $(DEST_SH)/bin/mpirun -np <num proc.'s> xpdc1.par.sh -i <inputfile> -dp <dump period> -d <dumpfile> -s <num. steps> -nox 3) NOTES a) Current version of Parallel XPDC1 requires you to use the -nox option with number of steps and dump period specified. It also assumes that you are starting a simulation from a dump file. If you do not have a dump file, generate one by running the non-parallel XPDC1 for a few timesteps and saving the result. b) Input files and dump files are fully compatible between the parallel and non-parallel versions. c) To observe diagnostics, start a non-parallel xpdc1 with X turned on with a dumpfile generated from the parallel version.
⌨️ 快捷键说明
复制代码
Ctrl + C
搜索代码
Ctrl + F
全屏模式
F11
切换主题
Ctrl + Shift + D
显示快捷键
?
增大字号
Ctrl + =
减小字号
Ctrl + -