Tony Duell wrote:
I HATE partial decodes --- it's something
else waiting to 'bite you in
the A..' sometime in the future.
Who said anything about partial decodes?
What I was suggesting was to AND the top 100 address lines (if you can
make a CPU clocking at 1GHz, then you can make a 100 input AND gate).
That gives you an 'I/O page selected' signal, which can be system-wide.
Now decode the bottom 28 address lines fully, AND with the I/O page
selected signal, and use the result as the chip select signal for an I/O
chip.
That's fully decoded. The I/O is in its own page at the top of memory
(like on a PDP11, where it works very well), and you can use the same
instructions to access memory and I/O.
Chuck
-tony
Hi Tony:
O.K. You have described _complete_ decoding, but the description
_implies_ that you _may_ just check a few lines, and make a decision on
the basis of that.
The possible problem that I was seeing, is that IF the description of
how things are separated is obvious that _partial_ decoding will work
for most things, then people (some) will be lazy and, save a few clock
cycles, or speed up a routine, 'for now' --- ignoring the problems
later.
Chuck