Since x86 is not constrained to 32-bit instructions, do you think it would be better if this 3 input arbitrary logic function instruction had a destination register that is independent of the source registers, that is d = f(a,b,c) ?

For SVE2, would it be OK if the 8-bit immediate is loaded into a control register with a separate instruction? I’m not sure if that is what Jan Wassenberg meant by “something like MOVPRFX”.