| Re: Getting array indices [message #62698] |
Tue, 23 September 2008 14:07  |
Jean H.
Messages: 472
Registered: July 2006
|
Senior Member |
|
|
> That takes about twice as long as the "MOD" method on my machine:
> x86_64 Win32 Windows Microsoft Windows 7.0 Oct 25 2007 64 64
Interesting... indeed, for small arrays, it is about the same speed...
but the mod method is much faster on bigger arrays! (same speed with the
/dim keyword or not, 'verts' is used in both methods, so there is no
penalty for re-creating the index array)
Jean
|
|
|
|
| Re: Getting array indices [message #62699 is a reply to message #62698] |
Tue, 23 September 2008 14:17  |
Brian Larsen
Messages: 270
Registered: June 2006
|
Senior Member |
|
|
> Because of "indgen(ns)" or because of array_indices?
Because of array_indices according to this test:
;; .run test.pro
array = findgen(15, 3)
ind_in = lindgen(n_elements(array))
t1 = systime(/sec)
FOR i = 0UL, 1e5 DO BEGIN
ind = array_indices(array, ind_in
ENDFOR
print, systime(/sec)-t1
t1 = systime(/sec)
FOR i = 0UL, 1e5 DO BEGIN
sz = size(array)
nx = sz[1]
ny = sz[2]
verts = lindgen(nx*ny)
verts = transpose( [ [verts mod nx], [verts/nx] ] )
ENDFOR
print, systime(/sec)-t1
END
IDL> .run test
% Compiled module: $MAIN$.
1.2473059
0.50232387
IDL> print, !version
{ i386 darwin unix Mac OS X 6.4.1 Sep 25 2007 32 64}
For closure I whipped this together into a quick function
all_array_indices.pro so I don't have to remember how to do this
again.
http://people.bu.edu/balarsen/IDLdoc/all_array_indices.html
Brian
------------------------------------------------------------ --------------
Brian Larsen
Boston University
Center for Space Physics
http://people.bu.edu/balarsen/Home/IDL
|
|
|
|
| Re: Getting array indices [message #62700 is a reply to message #62698] |
Tue, 23 September 2008 13:47 |
pgrigis
Messages: 436
Registered: September 2007
|
Senior Member |
|
|
On Sep 23, 4:23 pm, Vince Hradil <hrad...@yahoo.com> wrote:
> On Sep 23, 12:53 pm, Jean H <jghas...@DELTHIS.ucalgary.ANDTHIS.ca>
> wrote:
>
>>> Here's a way to get verts:
>
>>> sz = size(array)
>>> nx = sz[0]
>>> ny = sz[1]
>>> nz = sz[2]
>>> ns = sz[sz[0]+2]
>>> verts = findgen(ns)
>
>> Shouldn't it be indgen() ...
>
>>> verts = transpose([ [verts mod nx], [verts/nx mod ny], [verts/nx/
>>> ny] ])
>
>>> BTW, I'd like to find a faster way, if there is one.
>
>> what about this:
>> print, array_indices(arr, indgen(ns))
>
>> Jean
>
> That takes about twice as long as the "MOD" method on my machine:
> x86_64 Win32 Windows Microsoft Windows 7.0 Oct 25 2007 64 64
Because of "indgen(ns)" or because of array_indices?
Paolo
|
|
|
|
| Re: Getting array indices [message #62701 is a reply to message #62700] |
Tue, 23 September 2008 13:41 |
Brian Larsen
Messages: 270
Registered: June 2006
|
Senior Member |
|
|
Interesting what you learn when you dig into a routine...
array_indices.pro
<<snip>>
for d=ndim-1, 1, -1 do begin
; Remove indices from even higher dimensions.
if (d lt ndim-1) then $
temp mod= dimProduct[d]
; Indices for higher dimensions.
result[d, *] = temp/dimProduct[d-1]
endfor
<</snip>>
of course some people say that all you get when you dig that deep is
dirty :)
Brian
------------------------------------------------------------ --------------
Brian Larsen
Boston University
Center for Space Physics
http://people.bu.edu/balarsen/Home/IDL
|
|
|
|
comp.lang.idl-pvwave archive
Members
Search
Help
Login
Home
