comp.lang.idl-pvwave archive: archive » Sky Falling, etc. : Array substitution + addition with plus-equal (+=)

Home » Public Forums » archive » Sky Falling, etc. : Array substitution + addition with plus-equal (+=)

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

Re: Sky Falling, etc. : Array substitution + addition with plus-equal (+=) [message #53015 is a reply to message #52941]

Tue, 13 March 2007 13:37

JD Smith
Messages: 850
Registered: December 1999

Senior Member

On Tue, 13 Mar 2007 11:35:53 -0700, Ed Hyer wrote:

> Thanks for the great explanations, that one really had my head
> spinning.
>
>> One additional point is worth mentioning: when setting large arrays, the
>> "offset" method of specifying a single index on the LHS is much faster
>> than the '*' method of building the full index list to match up the
>> dimensions of left and right-hand side arrays:
>
> Well, yes. Which is how I came across this problem. The "offset"
> method does not work for "+=" or any of the "iterative" operations,
> because of the behavior you described.
>
> My application is building a grand [X,Y,T] sum, from contributions of
> a routine which returns an [X,Y,n] array, with n varying from, say, 1
> to 10. Without the increment operation, I have to do
>
> totals=fltarr(2,2,6);
> add=GetAdd(InputArgs,ReturnT0=T0); Get ADD and the initial index in T
> (T0)
> sz=size(add)
> if(sz[0] eq 2) then NT=1 else NT=sz[3]; get size of 3rd dimension
> T1=T0 + NT - 1
> previous_totals=totals[*,*,T0:T1]; extract subarray corresponding to
> ADD
> totals[0,0,t0] = previous_totals + add ; copy new totals back into
> TOTALS
>
> Which comes at considerable computational cost. But, that's life.
> Thanks again for the explanations.

Yes, that's an issue, but then again, without knowing in advance how
large the computed array will be on the RHS, you can't blame IDL for
not knowing to expand totals[0,0,t0] into a large array of the desired
size. The only minor suggestion is that if you are doing
totals[*,*,T0:T1] many times in an inner loop, you could pre-compute
the corresponding index list as a properly dimensioned array, and
subscript with that instead. Won't save memory but will save repeated
recomputation of the same list of indices. Another option, if 'add'
is a large chunk, is to keep a large array of zeroes the same size as
'totals', use the offset trick to stick 'add' into it, and then
increment the entire 'totals' array. Probably won't save much unless
'add' is already a large fraction of the size of 'totals'.

JD

Report message to a moderator

[Message index]

		Sky Falling, etc. : Array substitution + addition with plus-equal (+=) By: MarioIncandenza on Mon, 12 March 2007 14:17
		Re: Sky Falling, etc. : Array substitution + addition with plus-equal (+=) By: JD Smith on Tue, 13 March 2007 13:37
		Re: Sky Falling, etc. : Array substitution + addition with plus-equal (+=) By: MarioIncandenza on Tue, 13 March 2007 11:35
		Re: Sky Falling, etc. : Array substitution + addition with plus-equal (+=) By: JD Smith on Tue, 13 March 2007 10:46
		Re: Sky Falling, etc. : Array substitution + addition with plus-equal (+=) By: Foldy Lajos on Tue, 13 March 2007 05:26
		Re: Sky Falling, etc. : Array substitution + addition with plus-equal (+=) By: Y.T. on Tue, 13 March 2007 00:19

Previous Topic:	randomn problem
Next Topic:	Re: Array resize with arbitrary arithmetic

-=] Back to Top [=-

[ Syndicate this forum (XML) ] [

] [

]

Current Time: Fri Oct 10 15:36:04 PDT 2025

Total time taken to generate the page: 0.48447 seconds