Expensive loops... can they be avoided?

Hi!

I need to render 2D slices cut from 3D spherical grid data. So far, I
am using a brute force method which, using two FOR loops, is
unfortunately rather slow (although working just fine). The problem
and my approach boil down to this:

-- A[i,j] is the input array, with r[i] and chi[j] being curvilinear
coordinates (not necessarily uniformly spaced)
-- B[k,l] is the output array with x[l] and y[l] being uniformly
spaced Cartesian coordinates (to be plotted with TVSCALE)

for k=0,nx-1 do begin
    for l=0,ny-1 do begin

        curr = sqrt(x[k]^2+y[l]^2)
        curchi = atan(x[k],y[l])

        if "curchi or curr are out of bounds" then begin
            B[k,l] = environment_value
            continue
        endif

        foo = min( r - curr ,nearestr,/absolute)
        foo = min( chi - curchi ,nearestchi,/absolute)

        B[k,l] = A[nearestr,nearestchi]

    endfor
endfor


The calculation of "curr" and "curchi" can easily be done outside the
loops of course. But what can I do with the rest?


Thanks,
Rainer