Re: IDL, arrays, and memory [message #33896] |
Tue, 04 February 2003 09:29  |
James Kuyper
Messages: 425 Registered: March 2000
|
Senior Member |
|
|
David Fanning wrote:
>
> Sean Raffuse (sean@me.wustl.edu) writes:
>
>> I would like to create a jagged array. My array is something like:
>>
>> array = intarr(3600, 1600, 240 or less)
>>
>> Do I save space by creating a jagged array where the 3rd dimension is of
>> variable size? Is that even possible? Should I just stick with
>> 3600x1600x240?
>
> What language are you thinking of doing this in, Sean? :-)
>
> As a general rule, it is better to get all the memory you
> need at once, then trim. This avoids memory fragmentation
> problems, etc.
>
> But "jagged arrays". I'd like to see this when you are
> finished. :-)
In C, you can implement jagged arrays by defining a two-dimensional
array of pointers to allocated arrays, where each of the allocated
arrays can have a different length. You can write custom functions that
know that the length can vary, and they can be very efficient.
In IDL, while you do have pointers, there's such a big speed difference
between the native array processing facilities and user-written loops
that this is probably not worth the trouble. You're probably better off
sticking with 3600x1600x240.
|
|
|
Re: IDL, arrays, and memory [message #33897 is a reply to message #33896] |
Tue, 04 February 2003 09:31  |
Sean Raffuse
Messages: 46 Registered: July 2001
|
Member |
|
|
>
> However, whereas in C the difference in terms of speed and algorithm
> design is negligible between using a ragged array and a "wasteful"
> full 3D array, this is not true in IDL. In particular, you can't use
> most of IDL's fast array-based operators with an array of pointers;
> you're stuck accessing each element in a loop, which will be markedly
> slower for a data structure of this size.
>
> You must balance the memory saved against the speed and flexibility
> with which you can operate on the data. This is a common theme in
> IDL, which, in many instances, trades increased memory usage for
> greater speed of execution. Often you can find other ways to organize
> the data which reduces the memory footprint while preserving much of
> the same flexibility had by putting it all in a single array. Or you
> can use, e.g., NaN values to fill the "wasted" array elements and
> avoid having to treat them specially.
>
> Good luck,
>
> JD
Exactly what I needed to hear. Speed of execution is indeed a major issue
here. I'll stick with the big array.
|
|
|
Re: IDL, arrays, and memory [message #33898 is a reply to message #33896] |
Tue, 04 February 2003 09:23  |
JD Smith
Messages: 850 Registered: December 1999
|
Senior Member |
|
|
On Tue, 04 Feb 2003 09:54:15 -0700, David Fanning wrote:
> Sean Raffuse (sean@me.wustl.edu) writes:
>
>> I would like to create a jagged array. My array is something like:
>>
>> array = intarr(3600, 1600, 240 or less)
>>
>> Do I save space by creating a jagged array where the 3rd dimension is
>> of variable size? Is that even possible? Should I just stick with
>> 3600x1600x240?
>
> What language are you thinking of doing this in, Sean? :-)
>
> As a general rule, it is better to get all the memory you need at once,
> then trim. This avoids memory fragmentation problems, etc.
>
> But "jagged arrays". I'd like to see this when you are finished. :-)
>
I think Sean probably means "ragged" array, which, in C, is
essentially an array of pointers to variable-length arrays, often
allocated dynamically at run time. It's a standard memory-saving
technique. In fact, when you create a simple static array of strings,
ala:
char *msg[3]={"Eat", "My", "Grits"};
you are implicitly creating a ragged array.
You can easily create such a beast in IDL:
array=ptrarr(3600,1600)
for i=0,n_elements(array)-1 do array[i]=ptr_new(vector_less_than_240)
However, whereas in C the difference in terms of speed and algorithm
design is negligible between using a ragged array and a "wasteful"
full 3D array, this is not true in IDL. In particular, you can't use
most of IDL's fast array-based operators with an array of pointers;
you're stuck accessing each element in a loop, which will be markedly
slower for a data structure of this size.
You must balance the memory saved against the speed and flexibility
with which you can operate on the data. This is a common theme in
IDL, which, in many instances, trades increased memory usage for
greater speed of execution. Often you can find other ways to organize
the data which reduces the memory footprint while preserving much of
the same flexibility had by putting it all in a single array. Or you
can use, e.g., NaN values to fill the "wasted" array elements and
avoid having to treat them specially.
Good luck,
JD
|
|
|
Re: IDL, arrays, and memory [message #33901 is a reply to message #33898] |
Tue, 04 February 2003 08:54  |
David Fanning
Messages: 11724 Registered: August 2001
|
Senior Member |
|
|
Sean Raffuse (sean@me.wustl.edu) writes:
> I would like to create a jagged array. My array is something like:
>
> array = intarr(3600, 1600, 240 or less)
>
> Do I save space by creating a jagged array where the 3rd dimension is of
> variable size? Is that even possible? Should I just stick with
> 3600x1600x240?
What language are you thinking of doing this in, Sean? :-)
As a general rule, it is better to get all the memory you
need at once, then trim. This avoids memory fragmentation
problems, etc.
But "jagged arrays". I'd like to see this when you are
finished. :-)
Cheers,
David
--
David W. Fanning, Ph.D.
Fanning Software Consulting, Inc.
Phone: 970-221-0438, E-mail: david@dfanning.com
Coyote's Guide to IDL Programming: http://www.dfanning.com/
Toll-Free IDL Book Orders: 1-888-461-0155
|
|
|