Article: AMD's Mobile Strategy
By: Ricardo B (ricardo.b.delete@this.xxxxx.xx), December 18, 2011 1:57 pm
Room: Moderated Discussions
Linus Torvalds (torvalds@linux-foundation.org) on 12/18/11 wrote:
---------------------------
>The thing is, the natural non-serial decode approach
>really does not care much at all. Since you have to handle
>multiple prefixes anyway, there's little difference between
>two or five, and there's little difference between "same
>repeated prefix" or "different prefix".
There's a difference between allowing 0 or N prefixes.
I agree that not caring about repeated prefixes is simpler.
>And in many ways, once you have decided that you want to
>do a non-serial decoder (even if it is just a single insns),
>then you are really likely much better off just being
>able to handle all cases in one go, rather than have to
>have some "multi-cycle special cases". Otherwise you just
>set yourself up for even more complexity.
>
>Of course, the traditional Intel approach has been to have
>mode complexity in the first decoder, so that may be the
>only one that handles any of the complex cases. That likely
>helps complexity quite a bit, and avoids the "cascading"
>issue. But it does handle all cases in the one complex
>decoder.
Actually, Intel resorts to microcode the nasty cases.
---------------------------
>The thing is, the natural non-serial decode approach
>really does not care much at all. Since you have to handle
>multiple prefixes anyway, there's little difference between
>two or five, and there's little difference between "same
>repeated prefix" or "different prefix".
There's a difference between allowing 0 or N prefixes.
I agree that not caring about repeated prefixes is simpler.
>And in many ways, once you have decided that you want to
>do a non-serial decoder (even if it is just a single insns),
>then you are really likely much better off just being
>able to handle all cases in one go, rather than have to
>have some "multi-cycle special cases". Otherwise you just
>set yourself up for even more complexity.
>
>Of course, the traditional Intel approach has been to have
>mode complexity in the first decoder, so that may be the
>only one that handles any of the complex cases. That likely
>helps complexity quite a bit, and avoids the "cascading"
>issue. But it does handle all cases in the one complex
>decoder.
Actually, Intel resorts to microcode the nasty cases.