| Re: Timing results - matrix multiply vs. indexing [message #2501 is a reply to message #2441] |
Thu, 30 June 1994 12:03  |
chase
Messages: 62
Registered: May 1993
|
Member |
|
|
>>>> > "David" == David Landers <landers@tsunami.dseg.ti.com> writes:
In article <1994Jun30.143333.2221@mksol.dseg.ti.com> landers@tsunami.dseg.ti.com (David Landers) writes:
David> Hi gang, About a week or two ago we were discussing ways to
David> make a 2-D array from a 1-D array (which represented constant
David> rows or columns). To summarize, the two (primary) ways to do
David> this are:
David> m = n_elements( array1 )
David> array2 = replicate( 1.0, n ) # array1
David> or
David> array2 = array1( lindgen( n,m ) / n ) )
I seem to recall someone suggesting another method using:
array2 = rebin(array1,m,n,/sample)
I compared this method using the test_arr.pro program David attached
to his post. It was almost twice as fast as the previous best when
array1 is large. It is slightly faster if array1 is 1xN, i.e.
array1 = transpose(array1) then use:
array2 = rebin(array1,n,m,/sample)
Note that rebin can not be used with complex or string arrays. For
this reason I think that I would prefer the slower indexing method
which can handle all types. In all situations where I might have used
this kind of replication speed is not important.
Regards,
Chris
--
===============================
Bldg 24-E188
The Applied Physics Laboratory
The Johns Hopkins University
(301)953-6000 x8529
chris_chase@jhuapl.edu
|
|
|
|
comp.lang.idl-pvwave archive
Members
Search
Help
Login
Home
