ANN: Schelab or numerics for Scheme -- prototype

[Juan Jose Garcia Ripoll <jjgarcia@ind-cr.uclm.es>]

Hi,

  After a weak of hard work I'm proud to announce that the prototype for
Schelab, a numerical analysis library for Scheme is out. Currently only
simple operations on matrices, such as addition, subtraction and
products are implemented, but up from now it is easy to add any new
feature.
  Schelab is being developed on MzScheme, Guile 1.2 (buggy) and Gambit-C
3.0.


WHAT IT IS
----------

  Schelab is an OO library based on Christian Queinnec's Meroon-V3. It
is intended to become the kernel of a matrix algebra system and
includes

+ A library for garbage collectable memory chunks.
+ A library of classes for foreign data types from FORTRAN and C.
+ A FFI for FORTRAN that needs no glue code.
+ A set of overloadable mathematical operations.
+ Interface to BLAS. (incomplete, but usable)
+ Classes for complex and real matrices plus simple ops (+,-,*)
+ A timing facility.


WHERE IT RUNS
-------------

  This library runs both on top of Guile and of MzScheme, and it can be
easily ported to other environments with these requisites:

+ be able to run Meroon-V3
+ allow the introduction of new primitives in C
+ allow the creation of non movable, but garbage collectable objects
(for the `memory blocks' library)

  These are the requirements for a Linux system:

+ MzScheme or Guile 1.2 or Gambit-C 3.0
+ A C and FORTRAN compilers to build the extensions.
+ The BLAS library compiled as a shared library.

  Currently the C code as well as the Makefiles are prepared for Linux,
which is the only environment I know of. If you get it to run on
DOS/WIN**/MAC or on any other system, please tell me so I can introduce
your changes in the package.


SAMPLE SESSIONS
---------------

  Some timings (in milliseconds) for the three supported environments:

A) GAMBIT-C (3.0) -- Code has not been compiled

> (load "init-gsc.scm")
> (schelab-require '(schelab matops))
> (schelab-load '(tests matrix1))
> (define a (make-real-matrix 100 100 0.1))
> (time (do ((i 100 (- i 1)))((zero? i))(plus a a)))
    253 ms real time
    250 ms cpu time (170 user, 80 system)
    3 collections accounting for 60 ms cpu time (60 user, 0 system)
    8416496 bytes allocated
    7920 minor faults
    no major faults
> (time (do ((i 100 (- i 1)))((zero? i))(minus a a)))
    214 ms real time
    210 ms cpu time (170 user, 40 system)
    1 collection accounting for 20 ms cpu time (20 user, 0 system)
    8402936 bytes allocated
    7928 minor faults
    no major faults
> (time (do ((i 100 (- i 1)))((zero? i))(times a a)))
    3544 ms real time
    3540 ms cpu time (3510 user, 30 system)
    1 collection accounting for 20 ms cpu time (20 user, 0 system)
    8798096 bytes allocated
    8168 minor faults
    no major faults
> (time (do ((j 99 (- j 1))) ((zero? j)) (do ((i 99 (- i 1))) ((zero? i)) (n-set! a (list i j) 1.0))))
    703 ms real time
    700 ms cpu time (680 user, 20 system)
    no collections
    2482620 bytes allocated
    1312 minor faults
    no major faults

B) MzSCHEME:

> (load "init-mz.scm")
> (schelab-require '(schelab matops))
> (define a (make-real-matrix 100 100 0.1))
> (time (do ((i 100 (- i 1)))((zero? i))(plus a a)))
cpu time: 600 real time: 600 gc time: 150
> (time (do ((i 100 (- i 1)))((zero? i))(minus a a)))
cpu time: 610 real time: 606 gc time: 180
> (time (do ((i 100 (- i 1)))((zero? i))(times a a)))
cpu time: 3970 real time: 3963 gc time: 150
> (time (do ((j 99 (- j 1))) ((zero? j)) (do ((i 99 (- i 1))) ((zero? i)) (n-set! a (list i j) 1))))
cpu time: 1470 real time: 1477 gc time: 0

C) GUILE 1.2:

guile> (load "init-guile.scm")
guile> (schelab-require '(schelab matops))
guile> (schelab-require '(system timer))
guile> (define a (make-real-matrix 100 100 0.1))
guile> (time (do ((i 100 (- i 1)))((zero? i))(times a a))
Time: 8155.807
guile>  (time (do ((i 100 (- i 1)))((zero? i))(plus a a)))
Time: 1266.251
guile>  (time (do ((i 100 (- i 1)))((zero? i))(minus a a)))
Time: 1288.696
guile> (time (do ((j 99 (- j 1))) ((zero? j)) (do ((i 99 (- i 1)))
((zero? i)) (n-set! a (list i j) 1))))
Time: 8574.112

D) SCILAB (A most efficient MatLab clone)

-->a = rand(100,100);
-->timer();for i=1:100; a+a; end; timer()*1000
 ans  =
   170.  
-->timer();for i=1:100; a-a; end; timer()*1000 
 ans  =
    170.  
-->timer();for i=1:100; a*a; end; timer()*1000
 ans  =
    3940.
-->timer();for j=1:100; for i=1:100; a(1,i) = 1; end; end; timer()
 ans  =
    0.7  

  As you see, Gambit outperforms all of them and only MzScheme gets
close. I believe that the main reason is of Gambit having a
*COMPILED* version of Meroon, which provides it with a fast
dispatching mechanism => a fast way to check whether an object is or
not of a given class => setting/retrieving fields is faster.


DISCLAIMER
----------

  In case you download it, please remember it's just a prototype: it
lacks all of the functions that you would expect from a matrix algebra
environment, and only provides the kernel plus some useful Scheme
libraries.

0) This is free software. You can hack it, wrap it, jump on top of it or
do whatever you want, but I deserve usual acknowledgement for my work

1) The whole package is provided as is. No guarantee, neither explicit
nor implicit is made about the suitability of the software for your
purpose or needs.

2) You use this software at your own risk: I shall not be resposible for
any damage, loss of data or undesired consequence of using the program.


TO-DO
-----

For the moment these are my main objectives:

+ Add a decent set of solvers for linear systems.

+ Delayed operations.

+ Add hypermatrices and sparse matrices.

+ Add an interface to the FFTPACK library.


WHERE TO GET IT!!!
------------------

        http://www.arrakis.es/~worm/schelab.html
-- 
Juan Jose Garcia Ripoll		job: jjgarcia@ind-cr.uclm.es
Dpto. de Matematicas		home: worm@arrakis.es
E.T.S.I. Industriales		www: http://www.arrakis.es/~worm
Univ. de Castilla-La Mancha
Ciudad Real (Spain)