| Re: Inner product of multi-dimensional arrays [message #3507] |
Tue, 07 February 1995 07:06 |
thompson
Messages: 584
Registered: August 1991
|
Senior Member |
|
|
Fergus Gallagher <f.Gallagher@nerc.ac.uk> writes:
> I want to form a matrix mulipication of the form
> C(i,j,k) = B(i,j,r) A(r,k) (summed over r)
> With some index cleverness, I could form a 4-D intermediate array and
> sum this over one index, but this intermediate array wouldn't fit
> into memory in my case.
> Does anyone have a suitable (fast) algorithhm for this sum. Even
> better would a generalized procedure for
> C = A(.....r.....) B(..r.....)
Try this:
DB = (SIZE(B))(1:3) ;Extract dims. Assume 3D
DA = (SIZE(A))(1:2) ; " " " 2D
B = REFORM(B, DB(0)*DB(1), DB(2), /OVERWRITE) ;Reformat into 2D array
C = B # A ;Calculate result
C = REFORM(C, DB(0), DB(1), DA(1), /OVERWRITE) ;Put into correct format
B = REFORM(B, DB(0), DB(1), DB(2), /OVERWRITE) ;Restore B to original format
This technique would only work if the dimension to be summed over is the last
one in B and the first one in A.
I do have a routine called REARRANGE which can rearrange the order of
dimensions in an array. It can found at URL
file://umbra.gsfc.nasa.gov/pub/soho/soft/cds/util/array/rear range.pro
It works best when supported by some CALL_EXTERNAL software written in C, which
can be found at URL
file://umbra.gsfc.nasa.gov/pub/soho/soft/cds/external/
It is designed to operate without this C routine, but it is slower then.
Hope this helps,
Bill Thompson
|
|
|
|
comp.lang.idl-pvwave archive
Members
Search
Help
Login
Home
