At least in the past, there have been a bunch of places that assume that every data type is the same size (i.e. you can stuff ints/unsigneds/floats in the same bits without losing anything). That would need to change. Plus, we currently don't have any 64-bit data types in our shader compiler and assembly backends. These are not really hard problems to solve, but there's enough that it's non-trivial work and probably a bunch of tedium.
Originally Posted by przemoli