> First, I'd like to thank kevin for such a performance boost with the new dt
> stuff in cvs, it rocks!
Thanks. BTW, the new CVS code is only improved emulation. I got the
DT code working and booting OSes fine. I'll layer it in later
after more work on faster emulation and native application
layer execution, since those can be done short term. Mostly
I need to add some instrumentation to find out when code
is well behaved and worth DT'ing.
I thought I'd focus on short term issues for now, and work
towards a packagable plex86 version, so people can start using
plex86.
> Using the old arch, if I used the caliberating the delay loop using the PIT at
> 1000Hz, and trying to sleep for n loops in that period, it would totally mess
> up (something like 18 million ticks/milisecond), and the same with the method of
> delaying n loops using RDTSC's.
>
> With the new DT stuff, using the former method, it's down to about 800 (80mhz),
> but with RDTSC, it's still at 18 million. No matter what I set the ips count to,
> I still get roughlt the same values. Any ideas? (800 is still way ott, I'm
> expecting something < 200).
In emulation/emulation.c, there is a fudge factor #define at the
beginning. Play with that. It's the machine cycles per instruction
that you want. There's also a factor in system-*.c, that's how
many cycles per microsecond your real machine does.
These will be better handled.
-Kevin
