| Large structures in IDL turn CPU to sweet, sticky molasses [message #48736] |
Fri, 12 May 2006 18:02 |
MarioIncandenza
Messages: 231
Registered: February 2005
|
Senior Member |
|
|
I found this, most recently, when using a longish loop (250k) to assign
a structure tag. I thought I was doing something that was crashing IDL,
but indeed no, just making it very, very slow, like "Fly to Bogota for
a cup of coffee" slow.
Now, when I set the structure up as a single structure, so that each
tag is a vector N elements long, I have no performance problem. It's
when I use a large array of structures, I find that
IDL> for i=0l,n do struct.tag[ri[ri[i]:(ri[i+1]-1)]]=x
takes FOREVER (and, notably, actually seems to decelerate, as if
there's a memory issue?), while
IDL> temp=struct.tag * 0.
IDL> for i=0l,n do temp[ri[ri[i]:(ri[i+1]-1)]]=x
IDL> struct.tag = temp
is very fast.
I found a newsgroup thread discussing this issue in IDL, but it's from
back in 2000, and I thought maybe there might be more information about
ways around this problem.
...and yeah, loops are bad, but in this case I'm already using HIST_ND
on the problem, and while I am aware of JD's ninja tricks using the
histogram of the histogram, I worry about the their effects on code
maintainability and sanity maintainability.
|
|
|
|
comp.lang.idl-pvwave archive
Members
Search
Help
Login
Home
