Re: variable length instructions, was Chicken or the Egg?

michael@moria.de schreibt:
>The reason for starting with short instructions is that you usually get a
>smaller executable


I retracted an objection, when I found that it's *really necessary* to
start with the *most compact* representation of some code. It's easy
to determine when an address or offset will occupy more bytes, since
this cannot result in a future size reduction of that instruction
argument. But in contrast it's near impossible to determine, when the
reduction of multiple references to a smaller size will move the
target(s) from an outer range into the (assumed) nearer range.


Assume that you have two (or more) instructions, which refer to a
target at an offset, which would just fit into a single byte. Then two
stable states are possible, with regards to code size, where 2
instructions with byte-size arguments will fit, or where 2 instruction
with word-size arguments will move the target away 2 more bytes, so
that word-size arguments really are required.


DoDi
[As I think we've said three times now, this is certainly possible in
theory. I'd still like to hear whether it occurs in practice often
enough to matter. -John]

Return to the comp.compilers page.
Search the comp.compilers archives again.