floating point coverage computing

The current computing of the coverage used for alpha blending is integer based, with 4 to 12 bits per pixel precision. Has anyone yet developed an algorithm that computes the coverage using floating point? I’m trying to design such an algorithm, but it’s harder than I thought.