comp.lang.idl-pvwave archive: archive

Home » Public Forums » archive » FOR loops removal

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

Re: FOR loops removal [message #62113 is a reply to message #61975]

Fri, 22 August 2008 03:50

ben.bighair
Messages: 221
Registered: April 2007

Senior Member

On Aug 22, 5:08 am, loebasboy <stijn....@gmail.com> wrote:
> On Aug 21, 8:34 pm, Chris <beaum...@ifa.hawaii.edu> wrote:
>
>
>
>> What about something like this?
>
>> s= 2*(array le 2)-1
>> array = (array ne 0)*array + (array eq 0)*(shift(s,1)+shift(s,-1))
>
>> You'll have to manually fix the endpoints, but the rest should be
>> good. I tested this and the first loop on an array of 400,000
>> elements. The first loop rand in .54 seconds, while the second ran in .
>> 054s.
>
>> Also, you should be a little careful when testing whether array equals
>> zero. If the array isn't an integer type (int, byte, long, etc), then
>> its possible to have numbers you think ought to be zero but, because
>> of finite precision arithmetic, have very small nonzero values. If you
>> suspect that this is happening, you can try a test like abs(array) le
>> eps, where eps is something like 10^-5 or something big enough to
>> cover roundoff error but small enough not to treat nonzero data you
>> care about as zero
>
>> chris
>
> Hello, thank you both for the input. Chris, yours doesn't go faster
> also despite it is really elegant (and pointing me to the shift
> function, which I didnt know exist and can come in handy). The reason
> for the lack of speed profit is that I haven't a large array there
> that needs to be processed, instead I have many small array that need
> to be processed (the whole of my program exist of one BIG for loop)
> and there are relatively a lot of zeros in the processed array (which
> is why the WHERE function solution doesn't work either)
>

Hi,

Given your response to Chris I think the teeeny suggestion I have to
share is a moot point. But I'll put it out there because it can be
handy. If I understand Chris' code snippet correctly, there can be a
spped up by using something like this...

mask1 = array ne 0
mask0 = ~mask1
array = mask1*array + mask0*(shift(s,1)+shift(s,-1))

It probably is a slim gain but can add up to a lot if you are
looping. In my simple minded test below I see better than 2x gain
usually.

a = RANDOMN(seed, 2000, 2000) > 0.0
b = A GT 0

PRINT, "~B test"
t0 = SYSTIME(/sec)
c = ~B
print, systime(/sec) - t0

PRINT, "B EQ 0 test"
t0 = SYSTIME(/sec)
c = b EQ 0
print, systime(/sec) - t0

IDL> @not_test
~B test
0.056261063
B EQ 0 test
0.13446379

Cheers,
ben

Report message to a moderator

[Message index]

		FOR loops removal By: loebasboy on Tue, 19 August 2008 05:38
		Re: FOR loops removal By: loebasboy on Thu, 21 August 2008 07:27
		Re: FOR loops removal By: Jeremy Bailin on Thu, 21 August 2008 08:40
		Re: FOR loops removal By: Jeremy Bailin on Thu, 21 August 2008 06:39
		Re: FOR loops removal By: loebasboy on Thu, 21 August 2008 00:59
		Re: FOR loops removal By: Jeremy Bailin on Wed, 20 August 2008 07:07
		Re: FOR loops removal By: loebasboy on Wed, 20 August 2008 05:36
		Re: FOR loops removal By: David Fanning on Tue, 19 August 2008 13:14
		Re: FOR loops removal By: loebasboy on Mon, 25 August 2008 01:31
		Re: FOR loops removal By: Chris[6] on Fri, 22 August 2008 11:19
		Re: FOR loops removal By: ben.bighair on Fri, 22 August 2008 03:50
		Re: FOR loops removal By: loebasboy on Fri, 22 August 2008 02:08
		Re: FOR loops removal By: Chris[6] on Thu, 21 August 2008 11:34

Previous Topic:	Re: Loop breaks IDL
Next Topic:	Where is my function called

-=] Back to Top [=-

[ Syndicate this forum (XML) ] [

] [

]

Current Time: Sun Nov 30 21:08:46 PST 2025

Total time taken to generate the page: 0.24044 seconds