🚀 go-pugleaf

>>>>> "Nathanael" == Nathanael J Henderson <nahender@plains.NoDak.edu> writes:

    Jeff> : Now if only we could get them to port it to the DEC
    Jeff> 275Mhz machine, then we'd : have the fastest mac on
    Jeff> earth!

    Nathanael>     Not likely.  I have a hard time believing the
    Nathanael> emulation hit is small enough to let the Alpha come
    Nathanael> close to native performance of the PowerPC 604 Macs.

Nathanael is correct.  Our CPU coefficient going from 68040 -> 486 is
approximately 1/3.  In other words a 75 MHz 486 runs 68040 code at
about the speed that a 25 MHz 68040 does, but an x86 has some features
that make it a better match for emulation purposes than an Alpha
(condition codes, sub-word registers, relatively small performance
penalty for unaligned accesses).

On the other hand, the next generation of our synthetic CPU has some
cool features that will partially compensate for architecture
mismatches.  I'd guess it's *at least* six months from being seen in
public, though.

Additionally, internally Executor uses native code to do all the
OS/Toolbox work, and we'll be making tools available so that
application writers who want to can add native portions of code for
their CPU intensive work.  I don't think you'll see many
people/companies using these tools in the next year, but beyond that,
who knows.

But time does not stand still.  Apple has some very talented people in
their MAE team, and even though they don't currently support any
little-endian work, who knows.  By the time we have something
screaming on the Alpha, they may have beat us to the punch.

For people who are interested in what our synthetic CPU technology
looked like last year, check out ftp://ftp.ardi.com/pub/SynPaper
(there's also a .tex and .ps form available).  That's basically what
will be shipping in Executor 2.0.  Much of the next generation work
has been completed, but Mat has been tied up doing other things at
ARDI and will remain tied up until 2.0 ships.

--Cliff
ctm@ardi.com