>>>> "der" == der Mouse <mouse
at rodents.montreal.qc.ca> writes:
>> A byte is the smallest INDIVIDUALLY
addressable unit of data on a
>> system.
> The PDP-10 is an excellent example of when this
isn't true. The
> smallest addressable unit is a word, which is 36 bits. A byte is,
> as noted, anything between 0 and 36 bits. Bytes are stored in a
> word, as many as can be fitted. To access bytes on a PDP-10, you
> have a byte pointer, which consist of a word address, and a bit
> pointer, and byte size.
der> That sounds a whole lot like a hardware-supported way of
der> addressing an object of an arbitrary size in bits. And that
der> would mean that bytes of any size *are* individually
der> addressible.
"Addressable" is a somewhat slippery concept.
In the PDP-10, the architecture description clearly is in terms of
36-bit words as the addressable entity. Loads and stores act on them,
memory addresses (for example in jumps) are addresses of 36-bit words,
and so forth. But yes, you could claim that the byte operations
"address" smaller units. A PDP-10 person would argue they don't --
they address words, and then extract or insert the smaller bytes.
CDC 6000s have something analogous, a set of character instructions
that interpret 60-bit words as collections of 10 6-bit characters.
I don't think a strict literal reading of bits of spec will give you
clean answers. You have to go by what the intent of the designers was
when they created their models.
For example, older Alphas can't load/store bytes either, you have to
extract or insert them. But the memory addresses, at least as seen by
the program, are considered to be byte addresses, so one would call
an Alpha byte-addressed even though it can't load and store bytes.
Maybe a good test is to look at program addresses. If those look like
byte addresses -- though often they will have alignment rules -- then
I'd call the beast byte addressable. By that test, Alpha and PDP-11
clearly are, and PDP-10 and CDC 6000 clearly are not. (That assumes
von Neumann machines -- it doesn't help with Harvard machines.)
> The fact that people today seem to believe that
byte addressable
> is the only possible thing, along with a byte being 8 bits, is
> plain and simply because they haven't seen any other.
der> There even are relatively modern machines which don't fit it
der> very well. Some DSPs, for example, have 32 bits as their
der> smallest directly addressible unit.
Some don't even use power of 2 word lengths, e.g., the 24 bit Motorola
ones.
paul