function boxave, array, xsize, ysize
;+
; NAME:
;       BOXAVE
; PURPOSE:
;       Box-average a 1 or 2 dimensional array.
; EXPLANATION:   
;       This procedure differs from the intrinsic REBIN function in the 
;       following 2 ways: 
;
;       (1) the box size parameter is specified rather than the output 
;               array size
;       (2) for INTEGER arrays, BOXAVE computes intermediate steps using REAL*4 
;               arithmetic.   This is considerably slower than REBIN but avoids 
;               integer truncation
;
;       A version of BOXAVE() that supports 64 bit integers is available for
;       V5.4 or later in http://idlastro.gsfc.nasa.gov/ftp/v54/
; CALLING SEQUENCE:
;       result = BOXAVE( Array, Xsize,[ Ysize ] )     
;
; INPUTS:
;       ARRAY - Two dimensional input Array to be box-averaged.  Array may be 
;               one or 2 dimensions and of any type except character.   
;
; OPTIONAL INPUTS:
;       XSIZE - Size of box in the X direction, over which the array is to
;               be averaged.  If omitted, program will prompt for this 
;               parameter.  
;       YSIZE - For 2 dimensional arrays, the box size in the Y direction.
;               If omitted, then the box size in the X and Y directions are 
;               assumed to be equal
;
; OUTPUT:
;       RESULT - Output array after box averaging.  If the input array has 
;               dimensions XDIM by YDIM, then RESULT has dimensions
;               XDIM/NBOX by YDIM/NBOX.  The type of RESULT is the same as
;               the input array.  However, the averaging is always computed
;               using REAL arithmetic, so that the calculation should be exact.
;               If the box size did not exactly divide the input array, then
;               then not all of the input array will be boxaveraged.
;
; PROCEDURE:
;       BOXAVE boxaverages all points simultaneously using vector subscripting
;
; NOTES:
;       If im_int is a 512 x 512 integer array, then the two statements
;
;               IDL> im = fix(round(rebin(float(im_int), 128, 128)))
;               IDL> im  = boxave( im_int,4)
;
;       give equivalent results.   The use of REBIN is faster, but BOXAVE is
;       is less demanding on virtual memory, since one does not need to make
;       a floating point copy of the entire array.      
;
; REVISION HISTORY:
;       Written, W. Landsman, October 1986
;       Call REBIN for REAL*4 and REAL*8 input arrays, W. Landsman Jan, 1992
;       Removed /NOZERO in output array definition     W. Landsman 1995
;       Fixed occasional integer overflow problem      W. Landsman Sep. 1995
;       Allow unsigned data types                      W. Landsman Jan. 2000
;-
 On_error,2

 if N_params() EQ 0 then $
     message,'Syntax -   out =  BOXAVE( array, xsize, [ysize ])',/NoName

 s = size(array)
 if ( s[0] NE 1 ) and ( s[0] NE 2 ) then $
     message,'Input array (first parameter) must be 1 or 2 dimensional'

 if N_elements(xsize) EQ 0 then read,'BOXAVE: Enter box size: ',xsize 
 if N_elements(ysize) EQ 0 then ysize = xsize

 s = size(array)
 ninx = s[1]                                  
 noutx = ninx/xsize     
 type = s[ s[0] + 1]
 integer = (type LT 4) or (type EQ 12) or (type EQ 13)

 if s[0] EQ 1 then begin                ; 1 dimension?

     if integer then begin 

        if xsize LT 2 then return, array
        counter = lindgen(noutx)*xsize
        output = array[counter]
        for i=1,xsize-1 do output = output + array[counter + i]
        nboxsq = float(xsize)

      endif else return, rebin( array, noutx)     ;Use REBIN if not integer

  endif else begin              ; 2 dimensions

        niny = s[2]
        nouty = niny/ysize
        if integer then begin                        ;Byte, Integer, or Long

           
           nboxsq = float( xsize*ysize )
           output = fltarr( noutx, nouty)     ;Create output array 
           counter = lindgen( noutx*nouty )     
           counter = xsize*(counter mod noutx) + $
                    (ysize*ninx)*long((counter/noutx))

           for i = 0L,xsize-1 do $
           for j = 0L,ysize-1 do $
                 output = output + array[counter + (i + j*ninx)]

        endif else $
           return, rebin( array, noutx, nouty)       ;Use REBIN if not integer
 endelse

 case type of 
 12:  return, uint(round( output/nboxsq ))               ;Unsigned Integer
 13:  return, ulong( round(output/nboxsq))               ;Unsigned Long
  2:  return, fix( round( output/ nboxsq ))              ;Integer
  3:  return, round( output / nboxsq )                   ;Long
  1:  return, byte( round( output/nboxsq) )              ;Byte
 endcase

 end