藍色情懷

If you want to use your G4/G5 with Mac OS X for some good numerical work .. this could be useful to you. I have here, binaries, source, documentation and instructions to install Fortran, MPI, OpenMP, PVM, Octave, GDL, Cactus, Globus, RNPL, GIMPS, GRAVSIM, FEYNMAN, Xmgr Grace, GNU Java, etc. on Darwin. If you are interested in harnessing the full power of a dual G4/G5, i.e. using AltiVec and both processors, click here.

Most of these packages have been recently recompiled/tested on TIGER OS X 10.4 and the current XCode Tools. Feel free to contact me with questions or problems! Note: Tiger/Panther does not include /usr/local/bin in its default PATH. Therefore, assuming you're using the default bash shell, please type (or include in /etc/profile) PATH=/usr/local/bin:$PATH and export PATH before using any of these tools.

The binary install option provided here, will literally take a few seconds! Many consider this the quickest and easiest install for some of these packages available on the web. If you choose to get the source and compile, be warned that it may take several hours even when everything goes perfectly!

NEW! I'm adding packages for Intel Macs, gradually. Note that most of those are untested. Please consider contributing your Mac crunching ability to the distributed computing project Einstein@Home. The Mac client has recently been optimized for the G4/G5 including AltiVec.

Computation Tools :: Fortran

GCC 4 (auto-vectorizing gcc with openmp) :

Compiled using source code from the GNU CVS servers. This contains current versions of gfortran (GNU F95, this is a version of the much awaited, free, open source, F95 compiler), gcc (GNU C), g++ (GNU C++) and gcj (GNU Java) compilers that can perform auto-vectorization (i.e. modify code to take advantage of AltiVec/SSE, automatically). Use flag -O3 -ftree-vectorize to enable that option. For more information, see this webpage. Download my binaries, and cd to the download folder. Then gunzip gcc-bin.tar.gz (if your browser didn't do so already) and sudo tar -xvf gcc-bin.tar -C /. It installs everything in /usr/local. You invoke F95 by typing gfortran. You will also need to have Apple's Developer Tools installed. They are included on a separate CD in the retail version of OS X and also available as a free download from Apple's Developer Site. Please install the most current version of Developer Tools. Finally, note that there is an alternate g95 project, that is not part of GCC. They seem to have recently released their own official version of g95 for OS X which you may wish to try. Again, note that that g95 is not the same as this version here. Some reports suggest that this alternate g95 is ahead in the implementation of F95 features, but there are mixed reports regarding its optimization capabilities.

Binaries: gcc-bin.tar.gz (PowerPC), gfortran-bin.tar.gz (PowerPC gfortran only), gcc-intel-bin.tar.gz (Intel Mac only), gfortran-intel-bin.tar.gz (Intel Mac gfortran only), updated March 2006 (Tiger only).
Documentation: click here!

g77 3.4 :

Recently compiled using source code from the GNU CVS servers. This is the FINAL release of g77 (version 3.4.4 compiler). Future versions of GCC will have gfortran (see above). Download my binaries, and cd to the download folder. Then gunzip g77v3.4-bin.tar.gz (if your browser didn't do so already) and sudo tar -xvf g77v3.4-bin.tar -C /. It installs everything in /usr/local. Try it! You will need to have Apple's Developer Tools installed. They are included on a separate CD in the retail version of OS X and also available as a free download from Apple's Developer Site. Please install the most current version of Developer Tools.

Note: On Tiger, with gcc 4.0 you may need to add the flags -lSystemStubs -lcc_dynamic for the last linking step to be successful.

Binaries: g77v3.4-bin.tar.gz (Panther/Tiger), updated June 2005.
Documentation: click here!

Commercial Fortran:

ABSOFT: A commercial option for Fortran on OS X. It works very well, I have used it extensively. It optimizes better than g77. Includes f77, f90, f95 and a rather nice IDE. It also includes AltiVec optimized Math and BLAS libraries. Moreover, along with CBS VAST it can vectorize and parallelize your Fortran codes! In other words, it can automatically optimize your code for AltiVec and dual processors. More details about this feature available at this link. For more information, try the Absoft OS X product page.

NAG: A commercial f95 for OS X. I recently played with it. Does excellent optimization .. yielding much faster executables compared with g77. Very affordable commercial solution. For information, try the NAG Fortran page.

IBM: A f90, C, C++ suite for OS X. IBM released its highly optimized PPC compilers (including Fortran!) for OS X. On my tests they seem to provide a speed gain of about 2X on current G4/G5's over other compilers. In my opinion, the best set of compilers for OS X. Unfortunately, with the Intel switch, these compilers will no longer be updated. In fact, the C/C++ compilers are broken on Tiger.

Absoft website: click here!
IBM: click here!
NAG website: click here!

F2C based Fortran:

This is the oldest Fortran compiler available for OS X. Its been around since OS X was in a public beta state. Download this shell script: buildf2c, type chmod +x buildf2c and then sudo ./buildf2c. The script will grab f2c source from Netlib repositories and install a f2c based compiler in /usr/local/. You are done! The compiler can be envoked by the commands fc or f2c.

Install script: buildf2c
Documentation: click here!

Computation Tools :: MPI and OpenMP

MPI (Message Passing Interface):

MPI is the most widely used framework for parallel computing. With this you can parallel compute over almost any kind of cluster .. Mac's, SGI's, SUN's, Linux .. even over a hybrid or heterogeneous cluster. If you want to use MPI with ssh try this link for instructions.

MPICH: Get the source from the MPICH website, configure, make and install. Or get my binaries, gunzip mpich-bin.tar.gz (if your browser didn't do so already) and sudo tar -xvf mpich-bin.tar -C / and you are done! It installs in /usr/local/mpich/.

LAM MPI: OS X is now officially supported! Download their universal binaries for OS X from their site. An initial release of XMPI for OS X has recently appeared. XMPI graphical user interface for running, debugging and visualizing MPI programs.

There is also UCLA's AppleSeed and MacMPI for Mac ONLY clusters that has been a big success. It indeed is a very easy way to build and run a cluster. In that sense it is very mac-like in spirit. I was able to set up my Macs into a parallel cluster in less than an hour! Running distributed simulations is also very easy using their GUI based Pooch Application. They also provide useful tutorials and detailed documentation on writing and running parallel codes. Check those out at this link.

Source: MPICH
Binaries: mpich2-bin.tar.gz, mpich-bin.tar.gz, mpich2-intel-bin.tar.gz, mpich-intel-bin.tar.gz, updated March 2006.
Documentation: LAM MPI, MPICH

OpenMP (Shared Memory Multiprocessing):

These tools compile code optimized for dual processors based on OpenMP directives. SMP is much easier to learn, compared to MPI .. so this could be useful to some. GCC4 now includes OpenMP support. See above for information and binaries. Another option is to get the latest version of Omni MP compiler source, build and install. If you are interested in a tool that auto-parallelizes your code, click here.

Source: click here!
Documentation: click here!

Apple's Xgrid:

This is a new tool that is built into Tiger, developed by Apple's ACG that makes it incredibly easy to run a cluster of Macs by using Apple's technologies like Bonjour, etc. It can automatically seek available resources and execute your simulation appropriately. Learn more about it at their site. The current version even supports MPI. It is my first choice now for all batch and parallel tasks; I use it daily.

Xgrid Site: click here!

Note: If you are interested in other advanced parallel computing environments, try Parallel Virtual Machine PVM and Unified Parallel C UPC. These appear to have Mac OS X support. Also Sun's Grid Engine SGE, which is a batch queuing system also has an OS X port.

Computation Tools :: Octave

Octave:

OCTAVE: Octave is an open source, Matlab-like numerical analysis software package that is very popular among scientists and engineers. Available here is Octave 2.1.73 (with octave-forge extensions) using Apple's GCC and the g77 3.4 above. Download my binaries, gunzip octave-bin.tar.gz (if your browser didn't do so already) and then sudo tar -xvf octave-bin.tar -C / and you are done! It installs in /usr/local/. Note that this Octave installation does NOT install gnuplot. I recommend installing AquaTerm and gnuplot on OS X, using this link. To enable the octave-forge features add the appropriate lines in your .octaverc.

Source: Octave
Binaries: octave-bin.tar.gz (Panther/Tiger PowerPC only), octave-intel-bin.tar.gz (Intel Mac only), updated March 2006.
Documentation: click here!

SciLab:

Scilab is a scientific software package for numerical computation providing a powerful open computing environment for engineering and scientific applications. For OS X binaries, visit Jean-Phillippe's site on installation options and also this link.

Source: click here!
Documentation: click here!

GDL:

Gnu Data Language (GDL) is an open source implementation of Interactive Data Language (IDL) which is an expensive commercial software package developed by Research Systems Inc. To find out more, visit the GDL website. To install, download my GDL version 0.9 binaries, gunzip gdl-bin.tar.gz (if your browser didn't do so already) and then sudo tar -xvf gdl-bin.tar -C / and you are done! It installs in /usr/local.

Source: GDL
Binaries: gdl-bin.tar.gz (PowerPC), gdl-python-bin.tar.gz (PowerPC, Python module), gdl-intel-bin.tar.gz (Intel Mac), updated March 2006.
Documentation: Refer to IDL's documentation.

Computation Tools :: Cactus, Globus, etc.

CactusCode (Numerical Computing Toolkit):

Cactus is an open source problem solving environment designed for scientists and engineers. Its modular structure ("thorns") easily enables parallel computation across different architectures and collaborative code development between different groups. It is mainly used for solving complicated PDE's. It highly portable, and has some really remarkable features like observing and even "controlling" or "steering" a running simulation using an ordinary web browser! Here is a link to a perpetual demo running in Albert-Einstein-Institute, MPG, Germany: Cactus Demo. Check it out! Cactus compiles and runs great on OS X with GCC 3.3. Get the "last stable release" source. Remember to use the make options ARFLAGS='ruc' USE_RANLIB='yes' CC=gcc CXX=g++. The C and F77 "thorns" compile with no trouble. For the Fortran 90 ones, you will need a commercial Fortran compiler, like the one from Absoft. The above MPI distributions are fully compatible with Cactus! For some more info, check out CactusCode's recent OS X compatibility web page.

Source: click here!
Documentation: click here!

Note: If you are interested in other advanced PDE solving environments, try PETSc from ANL. They now officially support Mac OS X!

Globus (Grid Computing Infrastructure):

The Globus project is about developing technology to build grids for high performance computation. Grids are environments that enable software applications to integrate instruments, displays, computational and information resources that are managed by diverse organizations in widespread locations. As an example, look at this web page that has information on a Cactus based grid computation of the collision of two black holes. See the link below for information on how to get Globus and related Grid Tools running on OS X.

Documentation: click here!

Condor (HTC):

Condor is a software system developed at the University of Wisconsin that provides computing capabilities through efficient capture of cycles on idle machines (HTC). The software, operating within an HTC (High Throughput Computing) rather than a traditional HPC (High Performance Computing) paradigm, organizes machines into clusters that can exchange resources. Condor then hunts for idle workstations to run jobs. Condor can migrate jobs to other machines, if needed. The Condor Project now officially supports Mac OS X!

Binaries: click here!
Documentation: click here!

RNPL (Rapid Numerical Prototyping Language):

This is a great tool that takes as input details about the form of an equation (usually a partial differential equation), and some parameters .. and then spits out a C or FORTRAN Code that solves the equation numerically using established iterative numerical techniques. RNPL can also be used for producing skeleton programs and for converting existing programs. Download the source, compile and install. This source has been modified to compile with OS X. Or get my binary distribution. You probably want to install this in /usr/local, to do so gunzip rnpl-bin.tar.gz (if your browser didn't do so already) and then sudo tar -xvf rnpl-bin.tar -C / and you are done! You will need to use the F77 flag: -fno-second-underscore, for RNPL generated F77 codes to compile properly.

Source: rnpl.tar.gz
Binaries: rnpl-bin.tar.gz, updated 2003.
Documentation: click here!

Computation Tools :: Other Miscellaneous

Great Internet Mersenne Prime Search:

These are OS X clients for the internet based, distributed computing project, GIMPS. The clients are based on the GLUCAS and MLUCAS source codes. They benchmark quite well. The sources compile without any special modifications. The GLUCAS binaries for OS X are available from this link (including dual processor versions), while you can get the MLUCAS binary (compiled statically using IBM's XL compilers) from this link. Keep your Mac crunching!

Source and Binaries: MLUCAS, GLUCAS

The GRAVSIM Package:

GRAVSIM is a package that contains an ever evolving and hopefully growing set of gravitational N body solvers. These solvers attempt to track the motion of a set of bodies moving under the influence of gravity. They are used in many areas of astrophysical study including planetary motion, star cluster formation and stability and galactic formation, stability and interactions (i.e. colliding galaxies). This package is highly AltiVec and DP optimized!

Source and Binaries: click here!
Documentation: click here!

The FEYNMAN Java Framework:

The framework provides the infrastructure every scientist or engineer needs to develop simulations using the Java programming language. The framework eliminates numerous code writing tasks by allowing the developer to focus on the physical aspects of the simulation and the type of numerical algorithm being implemented. This is a new package and still under development, but has considerable promise.

Source and Binaries: click here!
Documentation: click here!

PIXELGLOW Software's MACSTL :

This is a high performance implementation of the Standard Template Library and associated numerics classes, optimized for the Altivec execution engine on the Power Macintosh G4, G5 and other PowerPC machines. How fast is it? Take a look here.

Binaries: click here!
Documentation: click here!

Xmgr Grace :

This is a very easy to use, open-source, X11 based, plotting tool. I've been using it since my graduate school days and still love it. Its excellent especially for students and free! In addition to my binaries below, you'll need Motif for OS X; binaries for which you can get by clicking here. You'll also need Apple X11 installed (which is included with OS X, at least with Panther). Same install procedure as always. First, gunzip motif-bin.tar.gz and gunzip grace-bin.tar.gz if your browser didn't do so already. And then, sudo tar -xvf motif-bin.tar -C / and sudo tar -xvf grace-bin.tar -C /. You're all set. Grace is now installed in /usr/local/grace. By the way, if you are looking for a nice, easy to use, 2D and 3D, plotting package for OS X, try pro Fit. It works great. They have a free trial version, that is not time-limited and is full-featured. It just has some restrictions on the volume of data you can make it handle. Even the full version is very inexpensive!

Binaries: grace-bin.tar.gz, updated April 2005.
Documentation: click here!

Vectorizing and Parallelizing:

If you are interesting in harnessing the full power of a dual G4/G5, i.e. using AltiVec and both processors, click here.