Optimisations

[lshaw8317]

With these changes, one accelerates cumulative_sum for large arrays a little bit. For the benchmark bench/ndarray/cumsum_bench.py.

With old code:

N=511, 0.47549494107564294 s
N=645, 1.0405428409576416 s
N=812, 1.7176300684611006 s
N=1023, 3.627310434977214 s
N=1290, 11.596769571304321 s

With optimisation:

N=511, 0.46673170725504554 s
N=645, 1.0527909596761067 s
N=812, 1.718360980351766 s
N=1023, 3.9926660855611167 s
N=1290, 9.691950400670368 s

[FrancescAlted]

LGTM. You may want to update the plot of your recent blog ;-)