Re: Vectorization question

[Pavel Romashkin <promashkin(at)cmdl.noaa.gov>]

Hi Liam,
I came up with a different approach. For short arrays in a loop, it is 3
times slower than your method, Liam. However, with A being 1,000,000 and
both X and B 100,000 elements long, your method could not allocate
memory on my machine, while mine ran in 0.07 s. Check out the code
below. The loops were put in to get runtime extimates.
I run 5.3 on PowerMac G4, 192 MB total RAM, 64 MB allocated to IDL.
Cheers,
Pavel

pro pavel, a, b, x
out = a
start = systime(1)
;for i =0, 10000 do begin
ind = x[uniq(x, sort(x))]
loc = value_locate(x, ind)
sum_b = total(b, /cumulative)
res = [0, sum_b[loc], 0]
a_values = (res-shift(res, 1))[1:n_elements(res)-2]
out[ind] = a_values
;endfor
print, systime(1) - start
;print, out, format='(10i4)'
end

pro liam, a, b, x
out = a
start = systime(1)
;for i =0, 10000 do begin
tmp = intarr(n_elements(a), n_elements(x))
tmp[x, indgen(n_elements(x))] = b
out = a + total(tmp, 2)
;endfor
print, systime(1) - start
;print,  out, format='(10i4)'
end