Linus Torvalds (torvalds.delete@this.linux-foundation.org) on July 27, 2022 10:13 am wrote:
>
> I don't know why mulx does that unusual source, but I assume that Intel did some example loops
> and that it ends up working better when you get it right (possibly exactly because other ops
> want to use %rax for its special use - including that regular old-fashioned 'mul').

Now I wonder. If you have to choose a fixed register, %rdx isn't such a bad choice. But why didn't Intel allow four register arguments? Best guess is that they didn't have any previous four-argument instructions and didn't want to deal with all the implicit assumptions in existing code.

But vblendvps was introduced before mulx and already had four arguments. And I have a hard time explaining why vector registers are different from integer registers in this respect. AVX512 then seems to add a mask register parameter to all existing three-parameter instructions, making four parameters the common case. But arguably that happened after mulx. vblendvps clearly happened before.

So why didn't Intel make mulx fully flexible and give it four parameters?