Tests a regression in float addition for optimizing. The second argument
could be now be a constant for floating point numbers.