22 Sep
2004
22 Sep
'04
7:11 a.m.
If
you hold the word width constant, yes, you are right. But that is not
what I was talking about. In many early computers, the data buss and
the word width were the same.
... and many did not. The 'byte' as a convention for talking about
memory is just that, a convention, and fails miserably on machines whose
major casual metric is not a multiple of 8 bits. Many, many machines
were built on a multiple of 6 bits because that's how many it took to
define a character.
For machines which have some architectural feature > 8 but modulo 8 ==
0, 32- and 64-bit wide memory and paths could be byte-addressed. I don't
know for sure, but I would imagine there are 6-bit-character-addressable
instruction sets too.
Until more or less when CPUs fit entirely within silicon, there was no
hard and true correlation between the bit-widths of busses, registers
and paths; this was because constructing those things cost actual money
and scaling of silicon didn't exist. Lots of machines have different
width regs/accumulator, memory, index regs, program counters, arithmetic
units, etc.
(My LGP-21 is a good example: 32-bit accumulator, 31-bit memory, 12-bit
program counter, double-32 product reg, 4- or 6-bit I/O.)
(Nothing in a Microchip Inc PICxxxx except the register files is
byte-width!)
For non-multiple-of-8 machines, the 'byte' is not relevant generally.
Boy with hammer: everything looks like a nail.