Speccylator-src 1.0

Short:        Source code for Speccylator
Author:       Richard Carlsson
Uploader:     jrj aapt net au (James Jacobs)
Type:         misc/emu
Version:      1.0
Architecture: generic

This is the source code for the Speccylator emulator
(Aminet:misc/emu/Speccylator.lha).

---

I wrote my Z80 emulator in MC68000 assembler, since I wanted the fastest
possible emulation on my Amiga 500. I only got an '030 processor quite
late in the project, so there is little or no optimization that actually
takes into consideration the properties of more modern CPU architectures
such as caching, large memories, or even the fact that a 68030 can move 32
bits from memory in one bus cycle, while the 68000 could only move 16 bits
at a time. Hence, it is very hard to beat the Speccylator on a 68000 CPU
(unless possibly if you do dynamic compilation instead of interpretation,
which I never got around to trying), but eg. the Spanish ZXAM was quite a
lot faster on a 68030. I did also pride myself on having a very exact
emulation, and last I looked, I did not have a single bug in the Z80
emulation that I knew of. I've toyed with the idea of rewriting the whole
thing in C, to see how fast I could make it while still keeping it
portable. It would be fun, but I have absolutely no time for it, what with
all my other projects.

The emulation core uses 64K 16-bit words of threaded code as a
"translation cache", mirroring the 64K bytes of the normal address space.
(I never had time to try to implement banked memory.) Each word is an
offset from a base pointer in code memory to the particular assembler
routine to be executed. There are several special "trap" routines that do
not correspond to any Z80 instruction; in particular, a zero word points
to the actual routine which decodes an instruction and replaces the zero
with the offset to the proper routine. There is relatively little use of
tables, eg. for BCD arithmetic, since memory access is pretty slow on a
68000. But I probably made too many assumptions about speed, rather than
actually measuring, so there may be many cases where I should have kept a
value (eg. a Z80 register) in memory instead of sacrificing a 68000
register. The design could probably have been a bit cleaner without losing
any speed (or even get a slight speedup).