Re: Combine four insns

[Jeff Law <law at redhat dot com>]

 On 08/06/10 13:43, Bernd Schmidt wrote:
> On 08/06/2010 09:37 PM, Jeff Law wrote:
> > It's also worth noting that some ports have hacks to encourage 4->1 or
> > 4->2 combinations.  Basically they have patterns which represent an
> > intermediate step in a 4->1 or 4->2 combination even if there is no
> > machine instruction which implements the insn appearing in the
> > intermediate step.  I've used (and recommended) this trick numerous
> > times through the years, so I suspect these patterns probably exist in
> > many ports.
> Yes.  Such combiner bridges may still be useful to help with five-insn
> combinations :)
Yea, though I expect there to be an ever-decreasing payoff for allowing 
larger bundles of instructions to be combined.
> However, as I found when trying to fix PR42172 for the Thumb, describing
> insns that don't actually exist on the machine can degrade code quality.
Precisely.  Creating those funky bridge instructions was never a 
solution, it was a hack and sometimes the hack created worse code.  Far 
better to actually fix combine.
Jeff