modified cumsum for partial accumulated sums

This is a modified version of cumsum, where the accumulated sum is determined along the first non-singleton direction of an input tensor "arr". The accumulated sums are stored in the output tensor "val" at locations of the zeros of arr.

[val, ind] = cumsum_restart0( arr, dim )
if dim is not defined, use first non-singleton dimension
val is the cumulative sum, where the accumulation is reset for zero
entries
ind indicates the locations where the acummulated sum is stored

example:
arr = [3 1 1 0 1 2]';
[val, ind] = cum_sum_restart0( arr );
val = [0 0 5 0 0 3]'
ind = [0 0 1 0 0 1]'

3/20/13
koehler.s.a@gmail.com