Unusual slowdown using sum.

Matlab handles sums of arrays pretty well, but seems to have trouble with my application. I do a bit of work with double precision numbers early in the code, then want to take a 1000x1000 array and produce a 1000x1000x1000 with said array. I've done things like this before with larger dimensions and larger numbers of dimensions with output results in milliseconds. The cataclysmic code is here:

    for nn = 1:1000
        for mm = 1:1000    
            if (nn < mm)
                Temparray = Loss(nn:mm,:);
                Grandarray(nn,mm,:)= sum(Temparray,1);
            end  
        end
    end

Because of the symmetry of the problem I only have to compute the equivalent of the upper-right corner. Is there something I'm inherently doing wrong, or is the computation simply enormous?

[Walter Roberson]

It looks to me as if you could probably make use of cumsum() for efficiency, unless the difference in round-off would be too much for you.

[Matthew Reed]

Thank you very much! You've sped this up by an order of magnitude. Thank you for helping an amateur.

Answer 1

[James Tursa]

Are you pre-allocating Grandarray?

[Matthew Reed]

I am.