Re: ARM and 6502

[pubpc1@library.ucla.edu]

Regnirps wrote:
> Rubywand <rubywand@swbell.net> wrote:
> >     There are some instructions which would be very nice to have. One
> >would be an address-relative JSR. Another is some way to handle data and
> >code address bank switching. (Another is a way to set up multiple large
> >stacks.)  Anyway, it looks like ARM makes it possible to define a 'super
> >6502'. Is this the case?
> 
> I would consider a set of 'macros' that are called like any other 6502 routine,
> but run pure ARM. A good example is miltiply. Change the JSR to your multiply
> routine to a "trap" that runs the ARM multiply instruction. Here we get a huge
> advantage in speed. I think my best 16 bit multiply was between 400 and 500
> cycles where the ARM uses a Boothe algorithm multiplier so it would be about 16
> ARM cycles. You could build up a library of routines like this and slowly
> transition to pure ARM code kind of like the way Apple went from 68000 to
> PowerPC. Once you are doing ARM instead of emulation, we are talking running
> hundreds of times faster.
> 
> I'm still looking at the best way to get started and I think it is with a
> simple 7 series ARM on a card for a slot and an assembler or something and a
> way to write to it from the Apple side.
> 
> Charlie


Charlie, how about first making a math coprocessor card?  Using a patch
to the SANE Tools, all floating point math can be redirected to the ARM.
There have already been cards based on the 68881 and 68882 that do this
and the ARM would be a vast improvement.  Instantaneous FP on the GS.
After you have sell enough ARM math cards, a full emulation coprocessor
card will be feasable.

-Scott G.