lib_ex_pour_HZpp/Partie_2/algebre_lineaire/MV++/mv/README

1 line
4.5 KiB
Plaintext
Executable file

/*+++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++*/
/* */
/* */
/* MV++ Numerical Matrix/Vector C++ Library */
/* MV++ Version 1.5 */
/* */
/* R. Pozo */
/* National Institute of Standards and Technology */
/* */
/* NOTICE */
/* */
/* Permission to use, copy, modify, and distribute this software and */
/* its documentation for any purpose and without fee is hereby granted */
/* provided that this permission notice appear in all copies and */
/* supporting documentation. */
/* */
/* Neither the Institution (National Institute of Standards and Technology) */
/* nor the author makes any representations about the suitability of this */
/* software for any purpose. This software is provided ``as is''without */
/* expressed or implied warranty. */
/* */
/*+++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++*/
MV++ : Numerical Matrix/Vector Classes in C++
(http://math.nist.gov/pozo/mv++.html)
MV++ is a small, efficient, set of concrete vector and simple matrix
classes for numerical computing. It is not intended as a general
vector container class, but rather designed specifically for optimized
numerical computations on RISC and pipelined architectures. It is one
step above a C/C++ array, altough it supports subvector/matrix expressions.
It includes several of the computational kernels found in the
Basic Linear Algebra Subprograms (BLAS), such as scalar updates,
vector sums, dot products and so on.
These classes are the building blocks of larger user-level libraries such
as SparseLib++ and Lapack++.
1) More About MV++
------------------
i) User's Guide and man-style pages are in http://math.nist.gov/pozo/mv++.html
or are available via anonymous ftp from math.nist.gov:pub/pozo/docs/mv++.ps.Z.
ii) Code examples are in ./testing/*.cc
2a) Installating MV++ library (float, double, and complex)
------------------------------------------
i) cd to root directory where MV++ is installed, <mv++>
ii) edit makefile.def to specify your specify your particular C++ compiler
iii) type "make mv"; "make" by itself will provide a list of options.
3) Testing MV++
---------------
i) cd to <mv++>
ii) "make test" will run a test suites and leave their output in
<mv++>/testing/mv_test.out.
4) Using MV++
----------
i) all MV++ include files are in <mv++>/include.
ii) MV++ library is in <mv++>/lib/mvlib.a
iii) to extend non-templated MV++ for vector and matrices of user-defined
classes, see mvvt.h, mvmt.h in <mv++>/include.
These can easily modified with an editor, replacing "$TYPE" with your
specific class name.
5) Help!
--------
Questions, comments, suggestions, etc. can be sent to pozo@cam.nist.gov.
Questions and Answers:
---------------------
o) What types does the non-templated MV++ classes support?
Float, double, int, and (optionally) complex.
o) How do I extend the non-templated MV++ classes to my own data types?
Edit mvvt.h in <mv++>/include and change every occurence of
"$TYPE" to the name of your numerical class. (Don't use cpp or m4
for this, as you'll need to change "$TYPE" in word segments as well.)
One easy way is
sed '1,$s/$$TYPE/Large_Real/g' mvvt.h > mvvLR.h
Do the same for mvvt.cc in <mv++>/src.
For example, an arbitrary-precision numerical class, Large_Real, would
have MV++ classes MV_Vector_Large_Real, MV_ColMat_Large_Real, and so on.
o) what is the file "empty" in <mv++>/lib?
A dummy file; it was put there so our version of shar would
include the <mv++>/lib directory. Otherwise, make reports an error
when trying to build mvlib.a , because the destination directory does not
exist.