comp.lang.idl-pvwave archive: archive » Re: An optimisation question

Home » Public Forums » archive » Re: An optimisation question

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

Re: An optimisation question [message #79728 is a reply to message #79711]

Tue, 27 March 2012 02:35

Yngvar Larsen
Messages: 134
Registered: January 2010

Senior Member

On Tuesday, 27 March 2012 04:33:08 UTC+2, Bogdanovist wrote:

> Here is a test code snippet isolating this conundrum. In this case the
> REFORM approach is only 5 times slower, so not as bad as my example,
> but why is it slow at all? Surely the FOR loop is not the optimal
> approach here!
>
> pro test_foo_1,basis,lat,lon,ret
> ret = (reform(basis[0,*,*]))[lat,lon]
> end
>
> pro test_foo_2,basis,lat,lon,ret
> for i=0,9999 do $
> ret[i]=basis[0,lat[i],lon[i]]
> end
>
> pro test_foo
> Basis = fltarr(10,1000,1000)
> lat = fltarr(10000)
> lon = fltarr(10000)
>
> ret = fltarr(10000)
>
> test_foo_1,basis,lat,lon,ret
> test_foo_2,basis,lat,lon,ret
>
> end

Hm. Interesting. I've noticed before that one-liner 1D loops are quite fast in IDL. I also cannot explain why your TEST_FOO_1 is so slow since REFORM seems not to be the culprit. However, you can do it much more efficiently like this:

pro test_foo_1,basis,lat,lon,ret,slice
ret[0] = basis[slice+lonarr(n_elements(lat)),lat,lon]
end

Profiler report:

IDL> profiler & profiler, /system
IDL> for n=0,99 do test_foo
IDL> profiler, /report
Module Type Count Only(s) Avg.(s) Time(s) Avg.(s)
FLTARR (S) 400 1.646489 0.004116 1.646489 0.004116
LONARR (S) 100 0.001611 0.000016 0.001611 0.000016
N_ELEMENTS (S) 100 0.000101 0.000001 0.000101 0.000001
PROFILER (S) 2 0.000193 0.000097 0.000193 0.000097
TEST_FOO (U) 100 0.001783 0.000018 1.917648 0.019176
TEST_FOO_1 (U) 100 0.015138 0.000151 0.016837 0.000168
TEST_FOO_2 (U) 100 0.252603 0.002526 0.252603 0.002526

PS: In your test programs, LAT and LON are used as index arrays, but declared as floating point arrays. Not that it matters here though.

--
Yngvar

Report message to a moderator

[Message index]

		Re: An optimisation question By: Matt Francis on Tue, 27 March 2012 15:54
		Re: An optimisation question By: Matt Francis on Tue, 27 March 2012 15:45
		Re: An optimisation question By: David Fanning on Tue, 27 March 2012 15:00
		Re: An optimisation question By: Matt Francis on Tue, 27 March 2012 14:26
		Re: An optimisation question By: Yngvar Larsen on Tue, 27 March 2012 02:35

Previous Topic:	Re: Undefined Variable: DC_Write_Fixed/DC_Write_Free - PV WAVE
Next Topic:	Cellular automata

-=] Back to Top [=-

[ Syndicate this forum (XML) ] [

] [

]

Current Time: Thu Oct 09 23:41:55 PDT 2025

Total time taken to generate the page: 1.20141 seconds