Hi, Paul.

PE> DN>> On some platforms (not intel), the option of accessing data at
PE> DN>> other than alignment boundaries of a size determined by the
PE> DN>> machine is not available.

PE> PE> Could you tell me the name of one such machine?  On IBM S/370
PE> PE> machines it just makes it slower, so I don't have any real
PE> PE> life examples of it being not allowed at all, although I don't
PE> PE> doubt that there are.  BFN.  Paul.

PE> rc>     On any "IBM" (assuming an 80x86 processor),
fetching from a word
PE> rc> aligned boundary will be much faster. On 386+ machines this becomes a
PE> rc> doubleword boundary.

PE> I'm after a machine where it is not possible at all, not just
PE> slower.  BFN.  Paul.

Burroughs (now Unisys, but they were Burroughs when I worked on them)
Large Systems range were a stack-based architecture with a 48 bit word.
Actually it was a 51 bit word, but the other 3 bits were not directly
addressable by software and were used for such things as identifying
descriptors, uninitialised memory, code vs data, etc.

On those machines, you would have to load the word then mask & shift if
you wanted a byte.

That would also almost certainly have been true for the first computer I
ever worked on, the CDC-3200. I don't remember the register size but it
was bigger than a byte, maybe 32 bits.

Regards, FIM.

 * * Bigamist = Italian fog
@EOT:

---