Re: 6502 false references?

[jimn8@netcom.com (Jim Nitchals)]

As pointed out by others, indexed stores that don't cross a page
boundary will access the memory once for reading and once for
writing.

Indexed reads or writes that cross page boundaries will read once
at a non-carry-propogated address, and read/write once at the
correct address.

For purposes of emulating an Apple II speaker at $C030, treat the
read/write as a 1 microsecond push of the speaker in the opposite
direction from its current position.  Remember that the speaker
cone is an analog device, having inertia and a physical state that
can be somewhere inbetween "on" and "off".

That's how Electric Duet, Music Construction Set and the like can
make multi-voice music: they push the speaker cone into states that
approximate 3 or more discrete physical positions.

Trivia: I once wrote a variable pulse-width digitizing program that
got quite pleasant sound quality.  The time between pulses was
variable because it spread the carrier frequency semi-randomly,
reducing artifacts.  It worked great for sampled sound, but didn't
apply to music synthesis *because* the carrier frequency changed
on-the-fly.

By the time I built it, Apple II software was rapidly dying in
the marketplace, so it never shipped with a product.

For fun, try jumping "n" cycles into the following binary stream:
301: C9
302: C9
...
37C: C9
37D: C5
37E: EA
37F: 60

Jumping to 37E takes 8 cycles (2 for the NOP, 6 for the RTS)
Jumping to 37D takes 9 cycles (3 for the CMP zp, 6 for the RTS)
Of course, nobody would do a JSR/RTS into code like this (a waste
of perfectly good cycles!) but it should give you an idea of how
6502 code might try to push a speaker around with higher precision.

Multiple cache layers, execution units, and even (gasp) floating
point units make today's processors seem positively *weird* compared
to the predictable behavior of the 6502.  Counting cycles isn't
the art form it used to be.

- Jim

-- 
                    Jim Nitchals  <jimn8@netcom.com>