31 Oct
2010
31 Oct
'10
9:46 a.m.
Date: Sun, 31 Oct 2010 12:04:17 -0400
From: mcguire at neurotica.com
To:
Subject: Re: I/O models (was RE: Happy Birthday VAX 11/780 (influence of))
On 10/30/10 4:52 PM, Tony Duell wrote:
Well there are tricks like the "no-slot clock" idea, like the DalSemi
DS1216. It sits in memory space, acts like memory, but when a series of
patterns are written to it it switches a set of locations (I don't
recall the details) from being a few bytes of RAM to being a few bytes
of RTC registers.
On my NC4000 I/O board I decoded the 0FFFF address and used that for a
quick byte swap. -1 was a quick address to locate and write/read on that
processor. This was much better than shifts.
Dwight
OK,
it's fair to say that there's nothing that precludes memory-mapped I/O =
in any machine (except perhaps physical memory architecture, although I can=
't think of an example). But port-mapped I/O machines had specific instruc=
The obvious problem would be if the memroy map is already totally full.
If you've got a Z80 systme with 64K of memory, it would be perverse to
try memory mapped I/O (you could have some kind of MMU, but why...).
Similarky trying to emmeroy map any kind of I/O on a 2MByte PERQ would be
an 'interesting' exercise...