On Thu, Dec 17, 2009 at 1:59 PM, Gabriel Michael Black
<[email protected]> wrote:
>
> Yeah, picking the microops was a big concern for me when I was
> starting out. If you remember, I closely based what we have now on
> that patent I found for what looks like an older, 32 bit version of
> AMD's integer microop set. The SSE microops I just made up, but I
> spend a decent amount of time (a day or two) surveying the
> instructions, grouping them in various ways, identifying common
> behaviors, etc. In the end it's all a guess/approximation and it can
> never be 100% right like has been said, but the choice was far from
> arbitrary. I (and AMD before me, so to speak) did ask "if I'm going to
> implement x86, what uops should I implement to be able to do that
> efficiently".

I didn't mean to imply that there was any fundamental problem, or that
you hadn't implemented the uops with x86 in mind (since clearly you
did).  I just have the vague impression, that if there's a range from
1 to 10 where 1 is "let's just use the MIPS ISA for our uops" and 10
is "let's define a new uop for each macroinstruction", we're somewhere
around 5 and maybe we should be more like 6 or 7.  That's all.

Steve
_______________________________________________
m5-dev mailing list
[email protected]
http://m5sim.org/mailman/listinfo/m5-dev

Reply via email to