Geoff Worboys wrote:

>The point of this little admission of fallibility is just a
>reminder to think about how you may be using the encoded
>information in the future. Will it only EVER be decoded, or
>may you decide pass it around more generically? The answer
>to this question may change the way you decide to implement.
>
>
>

Goeff, my reference to the PDP-11 instruction set was not gratuitous.
The trade off of variable instruction and operand width was an important
part of most successful instruction sets.

The first three Vaxes, star (780), comet (750), and nebula (730) had
full PDP-11 emulation. The high end Vax, Venus (don't even remember the
model number), pushed more of the instruction decoding into hardware
(ecl, ugh), which made PDP-11 emulation more difficult. Some bright
light had the idea of ECO-ing the VAX instruction set with a 16 kit case
instruction to dispatch the full PDP-11 instruction in one fell swoop --
opcode and addressing modes. It was probably the one and only instance
of the instruction, but it did the trick brilliantly. Venus probably
did soft PDP-11 emulation faster than the fastest PDP-11. Looking back
at it, however, it was clear than the clean orthogonality of the PDP-11
instruction set was, in the final analysis, unnecessary. Any crazy
mish-mash of instructions that echoed the the effective frequency in
practice was just as good or even better. That didn't do DEC any good,
but Intel ran with it.