Skip to content

DEPENDENCIES.md

Latest commit

 

History

History
187 lines (132 loc) · 5.76 KB

DEPENDENCIES.md

File metadata and controls

187 lines (132 loc) · 5.76 KB

Installation Dependencies

NOTE

It is possible that you won't need to read this document. If you obtained the mls software as part of a shrinkwrap package, look there for the files build-xxx (where xxx might be "nag"), env.linux, and f90. After judiciously editing these, just cd to their directory and type build-xxx.

SUMMARY

Before building the mls software you will need the following

  1. f95 compiler (NAG, Lahey, or Sun)
  2. Installed SDP Toolkit libraries:
    1. PGSTK
    2. HDFEOS
    3. HDF
  3. mls software (probably where you're reading this)
  4. PVM--a library to emulate parallel virtual machines
  5. FFTW--a library to accomplish fast Fourier tranforms
  6. BLAS and LAPACK--libraries to speed up linear algebra

The first step is to find out if you already have (1)-(6). If you do, you may safely ignore the rest of this guide. Obtaining (1), (3), and (6) is not within the scope of this guide, although if you are reading it online, you probably already have (3).

The rest of this guide focuses on how to obtain (2), (5), and (6). After reading it you will probably want to read MakeFC.md or else type

make -f MakeFC disthelp

THE SDP TOOLKIT

The distribution to build the libraries and ancillary files making up the SDP Toolkit can be found at:

https://edhs1.gsfc.nasa.gov/

It comes with instructions for building it for a number of supported platforms. Additional information on how to build it for the linux OS can be found at

https://www.met.ed.ac.uk/~hcp/hdfeos_notes.txt

After building this successfully, keep careful note of its location. An easy way to do this is to source pgs-env.csh or pgs-dev-env.csh in your login shell, assuming you use the csh or tcsh shells. That way automatically defines three environment variables that will make configuring the mls software easier:

HDFEOS_LIB

HDFLIB

PGSLIB

If you get the latest version of the toolkit, you will automatically build two added libraries: hdf5 and hdfeos5. These too will be needed for correctly building the mls software if you use the latest version.

PVM

The distribution to build the libraries and ancillary files making up the PVM software system can be found at

https://www.netlib.org/pvm3/index.html

in particular the file: https://www.netlib.org/pvm3/pvm3.4.3.tgz

It comes with instructions for building for a unix-style platforms. In brief, after downloading the file, say to your home directory, issue the following commands

gunzip pvm3.4.3.tgz
tar -xvf pvm3.4.3.tar
cd pvm3
setenv PVM_ROOT `pwd`
make

Note carefully the value of PVM_ROOT --it will be needed when configuring the mls software.

FFTW

It is possible that workable fftw libraries already exist on your platform. For example, JPL's EOS-MLS computers all have it in /usr/local/lib. You can check this via

me:51% ls /usr/local/lib/\*fftw*

/usr/local/lib/libfftw.a

/usr/local/lib/librfftw.a

/usr/local/lib/libfftw.la*

/usr/local/lib/librfftw.la*

and the result, displayed beneath the ls command, indicates a positive match. In such a case, note that the FFTW_ROOT should take the value of /usr/local/lib.

Otherwise, the distribution to build the libraries and ancillary files making up the FFTW libraries can be found at

https://www.fftw.org/download.html

Follow the instructions found in INSTALL and note carefully where it places libfftw.a and librfftw.a. If you lack root privileges, remember to use

configure --prefix=installation_directory_name

before

make
make install

An extra wrinkle, especially if you are installing onto a linux platform, is that you may have to edit the configure.in file. Without this change, the resulting library will contain object files with the wrong number of underscores appended. In particular, look for the line with

AC_CHECK_PROGS(F77, f77 xlf xlf77 cf77 fl32 g77 fort77 f90 xlf90)

and replace it with

AC_CHECK_PROGS(F77, f95 f77 xlf xlf77 cf77 fl32 g77 fort77 f90 xlf90)

Thanks to Robert Thurstans (JPL) for pointing this out.

BLAS

It is possible that highly-optimized blas libraries exist on your platform or are made available through your f95 compiler or hardware vendor. If so, you should note their paths and names. For example, Intel supplies: libmkl32_p3.a, libmkl32_p4.a and libmkl64_itp.a. Or your compiler vendor or OS vendor may have put a copy in an easy-to-access place: e.g.

me/mlspgs:186%ls /usr/lib/\*blas*

/usr/lib/libblas.a

/usr/lib/libblas.so.3@

/usr/lib/libblas.so.3.0.3*

/usr/lib/libblas.so@

/usr/lib/libblas.so.3.0@

In either case, note

  1. Whether you have such a library; and
  2. If so, its path and name

You have to choose one of the three options:

  1. Use the externally-supplied library of BLAS routines
  2. Use f95 intrinsics to satisfy BLAS routines
  3. Let the mls software build its own BLAS routines

Note that (2) and (3) will always be available, and (3) is the built-in default

LAPACK

Highly-optimized LAPACK libraries may also exist on your platform. Again, note their paths and names.

If you are ambitious, you might check the site named ATLAS for "Automatically Tuned Linear Algebra Software"

https://www.netlib.org/atlas/index.html

so that you can download or indeed build your own. In an albeit artificial benchmark its use resulted in an order-of-magnitude speedup on a dual-processor pentium III doing matrix operations.

You have to choose one of the two options:

  1. Use the externally-supplied library of LAPACK routines
  2. Use f95 intrinsics to satisfy LAPACK routines

Note that (2) will always be available, and is the built-in default.