[scilab-Users] generators for grand

[Daniele Andreucci]

michael.baudin at scilab.org wrote on 12/05/10 11:29:
> 
>  Hi,
> 
>  Your report is that, on your machine:
>   * mt, urand, kiss : work
>   * fsultra, clcg2, clcg4 : fail
>  It may happen that there is a bad portability of the source code of 
>  these generators on 64 bits systems.
> 

Actually,

* mt, clcg2, clcg4: work
* fsultra, kiss, urand: fail

as confirmed by running your script:
> 
>  for rngen = [ "mt", "kiss", "clcg2", "clcg4", "urand", "fsultra" ]
>    mprintf("rngen:%s\n",rngen);
>    grand ( "setgen" , rngen );
>    r = grand(1,1,"exp",1);
>    m = grand(1,"mul",10,[0.2;0.5]);
>  end
> 

I ran it by removing the three generators "kiss", "urand", "fsultra"
from the list rngen, and everything worked. If I add one of the three to
the list, Scilab gets stuck: I mean, when the cycle arrives at the
failing generator, it outputs the generator name, and then the
computation never ends, the Scilab window menu commands (like Quit) do
not work, and I need stop the process by means of the Linux kill command.

The instruction actually causing the freeze seems to be the one calling
the "mul" distribution. Indeed, the instruction

>    r = grand(1,1,"exp",1);

is carried out anyway, even with "kiss", "urand", "fsultra". But its
results are suspect (I think): the following is the output of a
modification of the script above, cycling 1000 times for each given
generator, and averaging the variable r:

*** Scilab script:
for rngen = [ "mt", "clcg2", "clcg4" , "kiss", "fsultra", "urand" ]
   mprintf("rngen:%s\n",rngen);
   grand ( "setgen" , rngen );
   r=grand(1,1,"exp",1);
   for i=1:999 do
   r = grand(1,1,"exp",1) + r;
   end
   r=r/1000
//  m = grand(1,"mul",10,[0.2;0.5])
end
***
*** Scilab output:
-->exec("test2.sci");
rngen:mt
 r  =

    0.9827961
rngen:clcg2
 r  =

    1.0035369
rngen:clcg4
 r  =

    1.0353063
rngen:kiss
 r  =

    30419447.
rngen:fsultra
 r  =

    31473190.
rngen:urand
 r  =

    63800551.
***
Moreover, about the other hint:
> Does the following work (after a restart of Scilab):
>
>   grand ( "setgen" , "fsultra" );
>   m = grand(1,"mul",10,[0.2;0.5]);
it does not work, Scilab freezes as explained above.

I performed the trials with Scilab 5.3 beta5 Linux 64 bit, but sporadic
attempts confirmed the same behaviour for Scilab 5.2.2 (standard Debian
Squeeze amd64 package).

Thanks a lot for your help, and for the pointers to the code.

Best regards

Daniele Andreucci

-- 
[ Daniele Andreucci
[ Dip. di Scienze di Base e Applicate per l'Ingegneria
[ via A. Scarpa 16     00161 Roma, Italy
[ tel.    +39 0649766785
[ fax     +39 064957647
[ e-mail:  andreucci at dmmm.uniroma1.it
[ http://www.dmmm.uniroma1.it/~andreucci/
[ gpg pub key id: D3ADC732

-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 262 bytes
Desc: OpenPGP digital signature
URL: <https://lists.scilab.org/pipermail/users/attachments/20101205/8919c3ec/attachment.sig>