Okay, so today we're gonna pick up where we left off and continue talking about out
of order processors? And things are gonna start to get a little
complicated, but that's good. We don't want to build, well, we want to
try to build simple things, but we also want to build high performance things, so.
Let's start talking about some more complex, out of order processors than we
talked, talked about last time. So last time we talked about In order, in
order, in order processors, and we introduced a scoreboard, and then we
started to talk some about in order, in order, where the front of the machine is
in order and the issue is in order while the write back and the commit are out of
order. It also uses scoreboard, and today we're
going to start talking about things that even have higher performance.
And this is sort of, roughly put in sort of order of performance and sort of what
people build if they actually want high performance in their computer systems.
And we're gonna start talking about in order on the front end, out of order in
the back end, and we're gonna have to introduce things like re-order buffer, and
a store buffer. Then we're gonna talk about a machine
which has in order front end and all the back stages are out, out of order.
Which may not make a whole lot of sense, because out of order commit starts to look
a little odd. Likewise that's the, the case for this
processor. And, and, so why does out of order commit
look a little odd? Well, when you start to have out of order
commits it's very possible that you could if want to have a precise exceptions
point. It, depending on how you sort of implement
that processor, and depending on where you put the commit points.
If you put the commit points let's say at the end of the pipe it's very possible
you'll commit results to your register file, or your architecture register file
before you know that those results are good.
So you've basically just executed instructions and committed information
that's not correct in program work. And we'll show you some of that today.
Then finally we're gonna talk about sort of a, the highest performance thing here,
which is a, in order, fetch. Out of order issue, out of order execute,
and right back, and then finally an in order commit.
So this, you can have precise exceptions, have a commit point at the end of the
pipe. And for this we need to introduce an issue
queue, which is where the instructions sort of live for a while as they're
getting ready to go issue, and this allows us to do out of order issue.
Two different instruction code sequences not dependent on each other.
You should be able to execute these things in parallel if you had enough parallels on
your machine. And we talked about this already.
So I'm gonna skip past the, in order, in order, in order machine.
David WentzlaffAssociate Professor
