comp.lang.idl-pvwave archive: archive » Re: Accelerating a one-line program doing matrix multiplication

Home » Public Forums » archive » Re: Accelerating a one-line program doing matrix multiplication

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

Re: Accelerating a one-line program doing matrix multiplication [message #72665 is a reply to message #72663]

Tue, 28 September 2010 23:57

rogass
Messages: 200
Registered: April 2008

Senior Member

On 28 Sep., 15:10, nata <bernat.puigdomen...@gmail.com> wrote:
> Concatenation is a very slow action in IDL and, if you are copying
> memory, the time of computation increases...
> If v0, v1, v2 and v3 are each of them 3-element vectors then you will
> not see the difference. TEMPORARY function is great when you are
> copying large arrays. I think you can not improve your code because
> the problem is the matrix multiplication and you can not change that.
> Try putting timers to see what's the time to compute each instruction.
>
> tt=SYSTIME(/SEC)
> aux=[[v1],[v2],[v3]]
> PRINT, SYSTIME(/SEc)-tt
>
> tt=SYSTIME(/SEC)
> aux=aux # vc
> PRINT, SYSTIME(/SEC)-tt
>
> etc.
>
> Cheers,
> nata
>
> On Sep 28, 2:17 am, Axel M <axe...@gmail.com> wrote:
>
>
>
>> On 27 Sep., 15:31, nata <bernat.puigdomen...@gmail.com> wrote:
>
>>> You can use the TEMPORARY function if you can set the input to
>>> undefined...
>>> When you do [[v1],[v2],[v3]] you are duplicating data. v1, v2 and v3
>>> are copied and you are not conserving memory.
>
>>> You could try:
>>> RETURN, [[TEMPORARY(v1)],[TEMPORARY(v2)],[TEMPORARY(v3)]] # vc +
>>> REBIN(v0, SIZE(vc, /DIMENSIONS))
>
>>> Cheers,
>>> nata
>
>> Thanks nata.
>
>> v0, v1, v2 and v3 are each of them 3-element vectors. I can add that
>> but, as I understand it, it will only save the place of 12 floating
>> values in memory (48 bytes?).
>
>> But I am happy that you did not see any other obvious thing. I started
>> feeling depressed seeing that I am not being able to improve this
>> single line of code... maybe it is ok, and the whole thing is just
>> slow...? ahh.

Yes, and you can substitute some of your calls:

aux #= vc (makes no copy of aux as far as i know)
invert -> la_invert (much much speedier)
sometimes (replicate({temp:input},newsize)).(0) is faster then rebin
exchange ## with matrix_multiply

Cheers

CR

Report message to a moderator

[Message index]

		Re: Accelerating a one-line program doing matrix multiplication By: Axel Martínez on Wed, 29 September 2010 03:48
		Re: Accelerating a one-line program doing matrix multiplication By: Axel Martínez on Wed, 29 September 2010 03:22
		Re: Accelerating a one-line program doing matrix multiplication By: rogass on Tue, 28 September 2010 23:57
		Re: Accelerating a one-line program doing matrix multiplication By: Jeremy Bailin on Tue, 28 September 2010 08:31
		Re: Accelerating a one-line program doing matrix multiplication By: natha on Tue, 28 September 2010 06:10
		Re: Accelerating a one-line program doing matrix multiplication By: Axel Martínez on Mon, 27 September 2010 23:17
		Re: Accelerating a one-line program doing matrix multiplication By: natha on Mon, 27 September 2010 06:31
		Re: Accelerating a one-line program doing matrix multiplication By: rogass on Wed, 29 September 2010 07:49

Previous Topic:	Coyote Facts
Next Topic:	ENVI and IDL ROIs: Do they play nice?

-=] Back to Top [=-

[ Syndicate this forum (XML) ] [

] [

]

Current Time: Sat Oct 11 14:58:55 PDT 2025

Total time taken to generate the page: 1.03887 seconds