Re : Multiprecisions numbers

Re : Multiprecisions numbers

[DELCOURT J�r�me]

Sorry, there was a mistyping in my last e-mail.
The correct site for multiprecisions numbers in ada is
http://www.chez.com/bignumber/index.html
There is a mirror site at
http://www.crosswinds.net/~bignumbers/index.html

Re: Multiprecisions numbers

[Pascal Obry]

[-- Warning: decoded text below may be mangled, UTF-8 assumed --]
[-- Attachment #1: Type: text/plain, Size: 744 bytes --]


DELCOURT J�r�me a �crit dans le message <38E5A2FD.63BD@club-internet.fr>...
>Sorry, there was a mistyping in my last e-mail.
>The correct site for multiprecisions numbers in ada is
>http://www.chez.com/bignumber/index.html
>There is a mirror site at
>http://www.crosswinds.net/~bignumbers/index.html

Do you have some benchmarks about the relative speed of these
implementations ?

Pascal.

--|------------------------------------------------------
--| Pascal Obry                           Team-Ada Member
--| 45, rue Gabriel Peri - 78114 Magny Les Hameaux FRANCE
--|------------------------------------------------------
--|         http://perso.wanadoo.fr/pascal.obry
--|
--| "The best way to travel is by means of imagination"

Re: Multiprecisions numbers

[DELCOURT J�r�me]

Pascal Obry wrote:
> 
> DELCOURT J�r�me a �crit dans le message <38E5A2FD.63BD@club-internet.fr>...
> >Sorry, there was a mistyping in my last e-mail.
> >The correct site for multiprecisions numbers in ada is
> >http://www.chez.com/bignumber/index.html
> >There is a mirror site at
> >http://www.crosswinds.net/~bignumbers/index.html
> 
> Do you have some benchmarks about the relative speed of these
> implementations ?
> 
> Pascal.
> 
> --|------------------------------------------------------
> --| Pascal Obry                           Team-Ada Member
> --| 45, rue Gabriel Peri - 78114 Magny Les Hameaux FRANCE
> --|------------------------------------------------------
> --|         http://perso.wanadoo.fr/pascal.obry
> --|
> --| "The best way to travel is by means of imagination"

Very interesting question. Actually, I have no benchmark available.
One of my goal is to look for a benchmark which could be representative
of a common use of that kind of package. The main problem is to define
what is the common use : each program can have a different use of
functions and procedures provided by that package.

I actually think that each programmer is free to choose the best
package for his (or her) needs in a particular program. Because of
that, I actually propose many packages in my site. Obviously, that
means that each programmer has to do a great work before choosing
the best package for his (or her) particular program, and a benchmark
could help.

Any idea on that subject?

Re: Multiprecisions numbers

[Gautier]

Pascal Obry:

> Do you have some benchmarks about the relative speed of these
> implementations ?

This would be very interesting (not time and equipment myself now)
E.g. GMP and other known packages, being in C, must allocate, reallocate,
deallocate memory for multi-integer. It consume time and the memory
gets fragmented very fast, requiring garbage collection that can take
longer than the computing itself. Reversely, pure Ada methods where functions
can return a whole unconstrained array, can work only on stack with
local variables. Anyway, using the "procedure with out parameter" variants
give very quick results on OpenVMS (Compaq Ada V3.5-2) and PC (GNAT/DOS).
No comparisons though from my side.

_____________________________________________
Gautier  --  http://members.xoom.com/gdemont/

Re: Multiprecisions numbers

[DELCOURT J�r�me]

Gautier wrote:
> 
> Pascal Obry:
> 
> > Do you have some benchmarks about the relative speed of these
> > implementations ?
> 
> This would be very interesting (not time and equipment myself now)
> E.g. GMP and other known packages, being in C, must allocate, reallocate,
> deallocate memory for multi-integer. It consume time and the memory
> gets fragmented very fast, requiring garbage collection that can take
> longer than the computing itself. Reversely, pure Ada methods where functions
> can return a whole unconstrained array, can work only on stack with
> local variables. Anyway, using the "procedure with out parameter" variants
> give very quick results on OpenVMS (Compaq Ada V3.5-2) and PC (GNAT/DOS).
> No comparisons though from my side.
> 
> _____________________________________________
> Gautier  --  http://members.xoom.com/gdemont/

You can find packages in ada about multiprecision numbers at
http://www.chez.com/bignumber/index.html.


Jerome Delcourt

Re: Multiprecisions numbers

[Gisle S�lensminde]

In article <38E5BFDD.AB9ED487@maths.unine.ch>, Gautier wrote:
>Pascal Obry:
>
>> Do you have some benchmarks about the relative speed of these
>> implementations ?
>
>This would be very interesting (not time and equipment myself now)
>E.g. GMP and other known packages, being in C, must allocate, reallocate,
>deallocate memory for multi-integer. It consume time and the memory
>gets fragmented very fast, requiring garbage collection that can take
>longer than the computing itself. Reversely, pure Ada methods where functions
>can return a whole unconstrained array, can work only on stack with
>local variables. Anyway, using the "procedure with out parameter" variants
>give very quick results on OpenVMS (Compaq Ada V3.5-2) and PC (GNAT/DOS).
>No comparisons though from my side.

The packages is not (yet?) comparable. The thick binding to AdaGMP 
use controlled types and operator overloading, which certainly is
inefficient, but is convenient to use. If any of the other bignum 
packages not beats the thick binding to GMP, they should consider 
a rewrite of their packages. I will hopefully do some "smart copying"
in the future, so it can be faster. Adding some proceeures which don't
need finalization like the operators is also an idea.

The GMP contains a wealth of functions, more than 100 integer functions,
about 50 floating point functions and 20 rational number functions.
It will be a considerable work to implement this. My motivation for
creating the binding was to experiment with public-key cryptography.
The GMP contains all the functions you need for RSA or Diffie-Hellman
key exchange, like modular exponentiation and multiplicative inverses.
These functions is typically not called many times, but each call is
quite costly (well at least for key exchange, not encryption). 
Is any of the other bignum libraries having these functions?

It will maybe be an idea to coordinate the effort, and make one
single bignum library? (A discussion that is more suited for gnatlist?)


-- 
--
Gisle S�lensminde ( gisle@ii.uib.no )   

ln -s /dev/null ~/.netscape/cookies

Re: Multiprecisions numbers

[DELCOURT J�r�me]

I totally agree with you when you write that we should coordinate
the effort. The improvement of this new, unique package could be faster.

The problem is that I don't understand at all the C language, and
the GMP library is very difficult to install on my MS-DOS computer
(due to the limitation of the filename length). So I haven't studied
the GMP library.

Any comment on that subject?



Jerome Delcourt