[RFC] split pseudos during loop unrolling in RTL unroller

[Richard Biener]

On Thu, Apr 16, 2020 at 7:46 PM Segher Boessenkool
<segher@kernel.crashing.org> wrote:
>
> On Thu, Apr 16, 2020 at 10:33:45AM +0200, Richard Biener wrote:
> > On Wed, Apr 15, 2020 at 11:23 PM Segher Boessenkool
> > > On a general note, we shouldn't depend on some pass that may or may not
> > > clean up the mess we make, when we could just avoid making a mess in the
> > > first place.
> >
> > True - but the issue at hand is not trivial given you have to care for
> > partial defs, uses outside of the loop (or across the backedge), etc.
> > So there's plenty of things to go "wrong" here.
>
> Certainly.  But *all* RTL passes before RA should leave things in "web
> form" (compare to SSA form).  The code in web.c is probably just fine;
> but we shouldn't have one web pass, *all* passes should leave things in
> a useful form!

Yeah well, but RTL is not in SSA form and there's no RTL IL verification
in place to track degradation.  And we even work in the opposite way
when expanding to RTL from SSA, coalescing as much as we can ...

> > > The web pass belongs immediately after expand; but ideally, even expand
> > > would not reuse pseudos anyway.
> >
> > But for example when lower-subreg decomposes things in a way turning
> > partial defs into full defs new opportunities to split the web arise.
>
> But that is only for the new registers it creates, or what am I missing?

No idea, just made up the example that maintaing "SSA" RTL is
not automagic.

> > > Maybe it would be better as some utility routines, not a pass?
> >
> > Sure, but then when do we apply it?
>
> All of the time!  Ideally every pass would leave things in a good shape.
> Usually it is cheapest as well as easiest to do things manually, but for
> some harder cases, such helper routines can be used.
>
> > Ideally scheduling would to
> > register renaming itself and thus not rely on the used pseudos
> > (I'm not sure if it tracks false dependences - I guess it must if it
> > isn't able to rename regs).  That would be a much better place
> > for improvements?
>
> sched2 runs after RA, so it has nothing to do with webs?  And sched1
> doesn't do much relevant here (it doesn't move insns much).
>
> I don't see how this is directly related to register renaming either?

If scheduling ignores "false" dependences (anti dependence with
full defs) then when it schedules across such defs needs to perform
renaming.  Maybe I'm using bogus terms here.

Richard.

>
> Segher