| Re: call external speed [message #12679] |
Thu, 03 September 1998 00:00 |
menakkis
Messages: 37
Registered: June 1998
|
Member |
|
|
curtis volin <cev@U.Arizona.EDU> wrote:
> I use the IDL SVD command very frequently, so I was wondering if it would
> be any faster to create a DLL and call it with call external. I suppose
> that the fundamental question is, are IDL routines optimized for speed?
I haven't done SVD comparisons, but my opinion is that it would not be worth
it to replicate this sort of IDL routine in C if you were implementing the
*same* algorithm - you might come out worse off.
I assume that you know that SVD is a "native" IDL routine (meaning that it is
not written in the IDL programming language, but rather in C already), but
that you're wondering how well RSI does this kind of thing. If their array
arithmetic is anything to go by, it looks like they are actually pretty good
at it. I know of a case where an algorithm coded in C (without "source code
level optimisations") compared pretty shabbily against the same thing coded
in IDL (using array arithmetic fairly extensively).
Having said that, I do think that there is *some* room for improvement
because I have been told that IDL for Windows/i386 is compiled for
"pentiums", and if you have a Pentium II or Pro you can compile your stuff
for "pentium pros" and realize a benefit. There might also be other
compilation optimisations (like the "assume no aliasing" ones - big ones)
that you can use, that might not have been used in IDL's build (?)
Funny that you should mention SVD, though. IDL uses the SVD algorithm from
Numerical Recipes. There are those who react in a kind of derogatory way
when in hearing distance of anyone who carelessly remarks: "I'm using the
Numerical Recipes SVD...". Now personally I don't really know the virtues of
one SVD implementation versus another (although I do believe that there are a
couple of bugs in the NR version that are obscured by that peculiar array
indexing they use), but there may be some benefit in coding a *different*
implementation to IDL's, in C.
Peter Mason
-----== Posted via Deja News, The Leader in Internet Discussion ==-----
http://www.dejanews.com/rg_mkgrp.xp Create Your Own Free Member Forum
|
|
|
|
comp.lang.idl-pvwave archive
Members
Search
Help
Login
Home
