Re: Loop optimizer misses simple optimisation?

[Jeffrey A Law <law@cygnus.com>]

  In message < Pine.LNX.3.96.980502143213.581F-100000@jinks.home >you write:
  > I don't know enough about division instructions available in m68020+, but
  > it looks to me from the above that there is an instruction that doesn't
  > generate the reminder, and that it is actually generated when the reminder
  > is not needed. Wouldn't it be better to move it to a separate, new udivsi3
  > pattern, instead of relying on REG_UNUSED, which is only available when
  > optimising, I suppose? This would allow the LICM code to do a better job,
  > wouldn't it?
In general, GCC does a poor job of certain optimizations for insns
that have multiple outputs (including LICM, contant propagation, etc).

It might be worth an experiment to split up the divmod insns into
div, mod and a unnamed divmod insn (for the combiner).  Though my
experience with other ports tends to make me believe the best
overall code is produced with the divmod insn.

However, a div or mod operation can not generally be moved out of
a loop unless you know it won't trap (divide by zero).

jeff