This is taken from

 https://mathematica.stackexchange.com/a/104696

This is to get f4cd6b63

Using the example of Andrea's monster meg run at
LO (!) we add two histograms each with 4000 bins.
The old code took 9.29s, the new code takes O(5ms)
so this simple update makes it 1000x faster by not
looping in python but letting numpy handle the
numerics..