Re: A2GS Accelerator Thoughts

[regnirps@aol.com (Regnirps)]

doug.cotton@cmdweb.com (Doug Cotton)
Date: Fri, Jul 10, 1998 18:18 EDT
Message-id: <doug.cotton-ya02408000R1007981718310001@news.earthlink.net> wrote:


>Even using some of the fastest processors around (which are VERY
>EXPENSIVE) you'll be lucky to achieve 40 MHz emulation of the 65816.
>A 400 MHz processor might get you close, though I suspect you'd be
>nearer to 30 MHz than 40, and the processor itself will cost hundreds
>of dollars. I doubt such an approach has much of a chance at being
>commercially viable.

The ARM interprets Java at a 2 to 6 instructions per Java byte-code. My own
experiments show about the same for 6502 or 65816. Thats the full overhead of
the interpreter included. Since 60 MHz ARM710 style ASICS are available for
about $20, you should be able to get around 20MHz emulation. 

With more sophisticated dynamic recompilation that recognizes code combinations
like 32 bit add and loops that do shifts, performance goes way up. The ARM has
a barrell shifter - any number of shifts in a single cycle for example. Also
recognizing addressing mode combinations can reduce several instructions to one
and local or comonly used variables can be kept in registers just for starters.

Even better, calls to ROM routines can be recognized and native code versions
run to perform the tasks. In fact, the memory manager could cause an exception
when ROM is addressed and vector to equivalent native ARM code, etc.

Charlie Springer