comp.lang.idl-pvwave archive: archive » Re: CUDA version of RANDOMN?

Home » Public Forums » archive » Re: CUDA version of RANDOMN?

Show: Today's Messages :: Show Polls :: Message Navigator
E-mail to friend

Re: CUDA version of RANDOMN? [message #62002 is a reply to message #61999]

Fri, 15 August 2008 08:28

wlandsman
Messages: 743
Registered: June 2000

Senior Member

On Aug 15, 11:16 am, "hotplainr...@gmail.com" <hotplainr...@gmail.com>
wrote:
> On Aug 16, 12:28 am, wlandsman <wlands...@gmail.com> wrote:
>
>
>
>> On Aug 15, 10:11 am, "hotplainr...@gmail.com" <hotplainr...@gmail.com>
>> wrote:
>
>>> Hey guys,
>
>>> I need to write a kernel to replace the IDL RANDOMN POISSON
>
>>> for loop
>>> for loop
>>> for loop
>>> c = data[x,y,b]
>>> if c gt 0.0 then begin
>>> n = RANDOMN( seedP, POISSON=c )
>>> endif else begin
>>> n = 0
>>> endelse
>>> data[x,y,b] = n
>>> endfor
>>> endfor
>>> endfor
>
>>> Could someone point out an example code of how RANDOMN POISSON so that
>>> I can implement it in CUDA?
>
>> Your best bet is to probably look at the Poisson generating algorithm
>> in "Numerical Recipes in C" if you are going to implement it CUDA.
>
>> I have implemented the "Numerical Recipes in C" algorithm into the IDL
>> procedure poidev.pro at http://idlastro.gsfc.nasa.gov/ftp/pro/math/poidev.pro.
>> Although poidev.pro is normally slower than calling randomn(POISSON=),
>> it has advantages for just the problem you describe, which can be
>> written as simply
>
>> data = poidev(data)
>
>> rather than using a triple FOR loop. --Wayne
>
> Thanks for the reply. I was about to use your code until I discovered
> the problem of achieving this.
>
> c = data[x,y,b]
> if c gt 0.0 then begin
> n = RANDOMN( seedP, POISSON=c )
> endif else begin
> n = 0
> endelse
>
> I guess the only way is to code a poisson kernel and then do tiling on
> the data.

Yes, that does mean the code becomes 3 lines instead of 1

g = where( data GT 0, Ng ,complement=g1, Ncomplement=Ng1)
if Ng GT 0 then data[g] = poidev(data[g])
if Ng1 GT 0 then data[g1] = 0

--Wayne

Report message to a moderator

[Message index]

		Re: CUDA version of RANDOMN? By: hotplainrice@gmail.co on Fri, 15 August 2008 17:58
		Re: CUDA version of RANDOMN? By: wlandsman on Fri, 15 August 2008 10:15
		Re: CUDA version of RANDOMN? By: wlandsman on Fri, 15 August 2008 08:28
		Re: CUDA version of RANDOMN? By: hotplainrice@gmail.co on Fri, 15 August 2008 08:16
		Re: CUDA version of RANDOMN? By: hotplainrice@gmail.co on Fri, 15 August 2008 08:14
		Re: CUDA version of RANDOMN? By: wlandsman on Fri, 15 August 2008 07:28

Previous Topic:	Re: Need help with an Iterative solution in IDL (relative newb question)
Next Topic:	Re: A very simple question about map_grid

-=] Back to Top [=-

[ Syndicate this forum (XML) ] [

] [

]

Current Time: Fri Oct 10 07:22:55 PDT 2025

Total time taken to generate the page: 0.48117 seconds