What about 8 and 16 bit ints? How about signed vs unsigned? Or what about pixel like data that clamps instead of overflows? What about 64-bit IEEE? What id the SIMD unit is 64 bits wide? Or 256? It just seems so not future and varying implementation proof.