This page describes how to install, and test version PLASMA 2.2.0, a linear algebra package for high-performance computers, on a Linux, Unix and MacOS X System. Windows installation instructions are not contained in this document.
This work was supported by NSF Grant No. CCF-0811642, NSF Grant No. CCF-0811520 and Microsoft.
Introduction
PLASMA is a linear algebra library for high-performance computers. The library includes Fortran 77 and C subroutines for the analysis and solution of systems of simultaneous linear algebraic equations, linear least-squares problems, and matrix eigenvalue problems.
The Parallel Linear Algebra for Scalable Multi-core Architectures (PLASMA) project aims to address the critical and highly disruptive situation that is facing the Linear Algebra and High Performance Computing community due to the introduction of multi-core architectures.
PLASMA’s ultimate goal is to create software frameworks that enable programmers to simplify the process of developing applications that can achieve both high performance and portability across a range of new architectures.
The development of programming models that enforce asynchronous, out of order scheduling of operations is the concept used as the basis for the definition of a scalable yet highly efficient software framework for Computational Linear Algebra applications.
This guide describes how to install, test, and time this release of PLASMA on a Linux, Unix and Mac OS/X System.
In order to install the PLASMA library, you require a C compiler, a Fortran Compiler, a BLAS library, a CBLAS Library, a LAPACK library, the LAPACK C Wrapper Library, and the availability of the pthread library. If you want to use testings and timings routines, you will also need the TMG library included in Lapack for matrices generation. Before PLASMA can be built, or tested, you must define all machine-specific parameters for the architecture to which you are installing PLASMA. All machine-specific parameters are contained in the file make.inc. We provide you an installer to ease the installation process and to install missing librairies. Users are strongly encouraged to use it. We also strongly recommend to install on your system the HWLoc library for a better control of thread binding. (See http://www.open-mpi.org/projects/hwloc/)
Getting the PLASMA Installer
The PLASMA installer is a set of python scripts developed to ease the installation of the PLASMA library. It can automatically download, configure and compile the PLASMA library including its dependencies. It is available on PLASMA download page: http://icl.cs.utk.edu/plasma/software/
PLASMA Installer flags
Here is a list of the flags that can be used to provide the installer information about the system like, for example, the C and Fortran compilers, the BLAS library that is already present on the system and need not be downloaded.
-h or --help : display this help and exit
--prefix=[DIR] : install files in DIR [./install]
--build=[DIR] : libraries are built in DIR [./build]
Contains log, downloads and builds.
--cc=[CMD] : the C compiler. [cc]
--fc=[CMD] : the Fortran compiler. [gfortran]
--cflags=[FLAGS] : the flags for the C compiler [-02]
--fflags=[FLAGS] : the flags for the Fortran compiler [-O2]
--ldflags_c=[flags] : loader flags when main program is in C. Some
compilers (e.g. PGI) require different
options when linking C main programs to
Fortran subroutines and vice-versa
--ldflags_fc=[flags]: loader flags when main program is in
Fortran. Some compilers (e.g. PGI) require
different options when linking Fortran main
programs to C subroutines and vice-versa.
If not set, ldflags_fc = ldflags_c.
--make=[CMD] : the make command [make]
--blaslib=[LIB] : a BLAS library
(path should be absolute if --prefix is used)
--cblaslib=[LIB] : a CBLAS library
(path should be absolute if --prefix is used)
--lapacklib=[LIB] : a Lapack library
(path should be absolute if --prefix is used)
--lapclib=[DIR] : path to a LAPACK C wrapper.
(path should be absolute if --prefix is used)
--downblas : Download and install reference BLAS.
--downcblas : Download and install reference CBLAS.
--downlapack : Download and install reference LAPACK.
--downlapc : Download and install reference LAPACK C Wrapper.
--downall : Download and install all missing external libraries.
If you don't have access to wget or no network
connection, you can provide the following packages
in the directory builddir/download:
http://netlib.org/blas/blas.tgz
http://www.netlib.org/blas/blast-forum/cblas.tgz
http://www.netlib.org/lapack/lapack.tgz
http://icl.cs.utk.edu/projectsfiles/plasma/pubs/lapack_cwrapper.tgz
http://icl.cs.utk.edu/projectsfiles/plasma/pubs/plasma.tar.gz
--[no]testing : enables/disables the testings. All externals
libraries are required and tested if enabled.
Enable by default.
--nbcores : The number of cores to be used by the testing. [2]
--clean : cleans up the installer directory.
The installer will set
OMP_NUM_THREADS=1 GOTO_NUM_THREADS=1 MKL_NUM_THREADS=1
to disable the multithreading within BLAS. Do not forget to set those environment variables for any further PLASMA testing you want to run.
PLASMA Installer Usage
For an installation with gcc, gfortran and Reference BLAS
./setup.py --cc gcc --fc gfortran --downblas
For an installation with ifort, icc and MKL (em64t architecture)
./setup.py --cc icc --fc ifort --blaslib="-lmkl_em64t -lguide"
For an installation with gcc, gfortran and Veclib (Mac OS/X)
./setup.py --cc gcc --fc gfortran --blaslib="-framework veclib"
For an installation with gcc, gfortran, ATLAS
./setup.py --cc gcc --fc gfortran --blaslib="-lf77blas -lcblas -latlas"
For an installation with gcc, gfortran, goto BLAS and 4 cores
./setup.py --cc gcc --fc gfortran --blaslib="-lgoto" --nbcores=4
For an installation with xlc, xlf, essl and 8 cores
./setup.py --cc xlc --fc xlf --blaslib="-lessl" --nbcores=8
Remarks: By default, the major BLAS librairies (ACML, ESSL, Goto, MKL, …) include CBLAS and LAPACK librairies, so you just have to specify which one you want to use as follow. However, ATLAS library does not include LAPACK library and only MKL includes TMG library from LAPACK for matrices generation in tests. In these case, the installer will automatically download for you the missing libraries and compiles them with your specified BLAS library.
PLASMA Installer Support
Please note that this is an alpha version and, even if it has been tested on a wide set of systems, may not work. In case you encounter a problem, your feedback would be greatly appreciated and would be very useful to improve the quality of this installer. Please submit your complaints and suggestions to the PLASMA forum: http://icl.cs.utk.edu/plasma/forum/
Tips and Tricks
Testings are slow
Please note that, in the case where the installer is asked to automatically download and install the BLAS library, the reference BLAS is installed and, thus, very low performance is to be expected. It is strongly recommended that you use an highly optimized BLAS library (like ATLAS, MKL, GotoBLAS, ESSL etc.) and provide a path to its location through the -blaslib flag.
For Mac
Veclib is accessible if you install the Xcode/developer package provides with your Mac OS installation CD. On MAC OS/C you may be required to add --ccflags=' -I/usr/include/sys/'
Processors with Hyper-threading
The plasma installer cannot detect if you have hyper-threading or not on your machine, if you don’t use HWLoc library. In this case, we advice always to limit the number of core for the first testing of the PLASMA library to half the numbers of cores available if you do not know your exact architecture. Using hyper-threading will cause the PLASMA testings to hangs.
Problem with download
If the required packages cannot be automatically downloaded (for example, because no network connection is available on the system), you can take them any way you like from the following URLs and place them in the build/download directory (if the directory does not exist, create it):
LAPACK C WRAPPER : http://icl.cs.utk.edu/projectsfiles/plasma/pubs/lapack_cwrapper.tgz