I can verify that *some* floating point functioncs (basic ones ususally
ie +, -, etc) can be emulated faster than the stock 80387 or 80387sx
co-processor can do them. The majority of functions can not be emulated
faster though.
-Matt Pritchard
Graphics Engine and Optimization Specialist
-----Original Message-----
From: Uncle Roger [SMTP:sinasohn@crl.com]
Sent: Tuesday, October 21, 1997 6:08 PM
To: Discussion re-collecting of classic computers
Subject: Re: 387 emulators (was: Wanted:387 SX)
At 05:07 PM 10/20/97 +0000, you wrote:
> 2) I saw several emulators, but the only one
I found that actually
worked
> well, the file was called
"FRANKE87" and was German in origin. It
actually
> fooled AutoCad 10 into believing there was a
co-processor chip on
my 386SX
and
actually did speed up FP instructions (measured with CheckIt).
Not. Autocad is
processor heavy program and better unload that FP to
that coprocessor result in even powerful system
when using the
suitable s/w like autocad.
IIRC, Autocad *requires* a mathco, so one has a choice of a) buying a
mathco
(used to be $$$), b) running with an emulator, or c) not running
autocad.
As to whether the system will run faster with or without a software
mathco
emulator, I must admit, it would seem obvious that an emulator would
only
slow the system down (by using more Cpu time to handle the emulator
than
going straight to the CPU -- kinda like buying direct from the mfr and
eliminating the middleman) but I cannot say that that's true without
testing it.
The FRANKE87 program may be really good at what it does, enough to
make a
difference when compared to Intel's idea of FP math. I have to say
that if
Merch says he measured the difference and the emulator is faster, I'll
take
his word for it until proven wrong.
---------------------------------------------------------------------
O-
Uncle Roger "There is pleasure pure in being mad
sinasohn(a)crl.com that none but madmen
know."
Roger Louis Sinasohn & Associates
San Francisco, California
http://www.crl.com/~sinasohn/