comp.lang.idl-pvwave archive: archive » FOR loops and efficiency

Home » Public Forums » archive » FOR loops and efficiency

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

Re: FOR loops and efficiency [message #66635 is a reply to message #66568]

Fri, 29 May 2009 13:39

JDS
Messages: 94
Registered: March 2009

Member

On May 28, 1:38 am, Craig Markwardt <craig.markwa...@gmail.com> wrote:
> On May 26, 5:51 pm, JDS <jdtsmith.nos...@yahoo.com> wrote:
>
>
>
>>> I still stand by my rule of thumb. The problem with FOR loops is the
>>> amount of time spent doing loop overhead stuff. If you run your loop
>>> but *take all the calculations out*, and the total execution time is
>>> not perceptible, then you probably won't gain by optimizing/
>>> vectorizing.
>
>> I find that analysis lacking for a few reasons. Consider this
>> example:
> ...
>> IDL> t=systime(1) & for i=0L,10000000L-2 do a[i+1]+=a[i] &print,systime(1)-t
> ...
>> IDL> t=systime(1) & for i=0L,10000000L-2 do begin & end & print,systime(1)-t
>> 0.12700295
> ...
>> Loop overhead is one reason to avoid FOR loops with high iteration
>> count, but it is by no means not the *only* reason. ...
>
> I agree with everything you said. I still stand by my guideline as
> rule of thumb to know when optimization is important. Note that the
> rule of thumb didn't involve trying to compare the execution time of
> an empty loop and a full loop. :-)
>
> By the way, if you put a simple dummy statement like this,
>
> t=systime(1) & for i=0L,10000000L-2 do begin & dummy = 0 & end &
> print,systime(1)-t
>
> Then the execution time is more like 0.5 seconds. While I agree that
> this is not the same as 2.2 seconds, it is definitely more
> comparable.
>
> Craig

Right. Now that I read your rule of thumb more carefully, I see your
point is really "keep it to a small number of iterations." The
problem is, if this is in a function which itself (perhaps later) gets
called millions of times, it will be of no solace that each function
call performs only 100 iterations. But your rule of thumb is actually
useful for all types of performance optimization. Should I optimize?
Only if it takes too long.

JD

Report message to a moderator

[Message index]

		FOR loops and efficiency By: Rachel on Thu, 21 May 2009 09:04
		Re: FOR loops and efficiency By: JDS on Tue, 26 May 2009 14:51
		Re: FOR loops and efficiency By: Craig Markwardt on Wed, 27 May 2009 22:38
		Re: FOR loops and efficiency By: Craig Markwardt on Sat, 23 May 2009 13:24
		Re: FOR loops and efficiency By: JDS on Fri, 29 May 2009 13:39
		Re: For loop By: Hassan on Tue, 27 October 2009 01:16
		Re: For loop By: Michael Galloy on Mon, 26 October 2009 09:47
		Re: For loop By: Hassan on Mon, 26 October 2009 08:45
		Re: For loop By: penteado on Mon, 26 October 2009 08:01

Previous Topic:	Re: surface_polar plot
Next Topic:	Re: Using FFT convolve() : centering error?

-=] Back to Top [=-

[ Syndicate this forum (XML) ] [

] [

]

Current Time: Wed Oct 08 20:06:05 PDT 2025

Total time taken to generate the page: 0.40736 seconds