I would caution you that most or all of the details regarding
IA64 on the HP3000 are preliminary and thus "subject to change..."
[the usual disclaimer for work in progress].

Don't get too thirsty for detailed answers yet, because lots of the
details are not ready for public consumption.

> Glenn Cole writes:
[snip]
> >Huh?  If PA code cannot call IA-64 code then how will non-recompiled
> >code call the system intrinsics (which, presumably, at some point call
> >"native" routines) ?  Or is that "presumably" part a mistake?
>
> The first part of Ken's presentation was quite technical, and so
> went over my head at times, but it is my understanding that the
> "legacy" PA-RISC code will be executed by an emulator.
This is correct. PA-RISC code will executed by an "emulator".
Thus it is "software only" that knows about PA Risc (the
HW/Architecture does not know about PA Risk Instructions etc).

[NOTE: it is not called an "emulator", but it is one. The IA64 "emulator"
is more like a combination of our emulator and our translator with
the translation (+ optimization) done at run time.]

So from a HW/Architecture point of view, there is no way for PA code
to call IA code directly. This is sort of irrelevant. The same is true
for CM and PA on todays HPPA boxes (from a HW point of view). Today
we support a "switch subsystem" which is software and it handles going from
one mode to the other.

On IA64 we will be supporting some form of software switch. No details
are available at this point. It is not clear how visible/available this
switch will be made to applications (eg: not clear that we will support
mixed mode application development like we do on HPPA today). But
existing PA code, which will be emulated, will "switch" to native OS
code which implements various functions.

At this point, we have no plans to support a "switch" from IA <> CM
directly. Lots of the details regarding CM are still being worked out.

[rest deleted...]