Re: [new-ra] "#if DENIS"

[Denis Chertykov <denisc at overta dot ru>]

Michael Matz <matz@suse.de> writes:

> Hi,
> 
> On 16 Jul 2003, Denis Chertykov wrote:
> 
> > I remember that reload (reload on insns modified by allocator) was founded
> > by me. After that I have realized assign_stack_slots and etc.
> > I can't remembered why these reloads happened.
> 
> I had to disable actually doing reloads for insns in later passes, as the
> allocator doesn't expect the insns to be changed under it's foot.

Oops. I forgot about this, but I mean only last pass of allocator and
call to emit_colors and assign_stack_local inside it. I mean true
reload not a pre-reload.

> 
> > > To one of those allowed by the constraints.
> >
> > You don't mark insns with spill-pseudos as ra_modified_insns.
> 
> Hmm, I do, in some situations conditionalized on "if (!flag_ra_test)"
> which is false by default, so ra_modified_insns should be set for all
> modifed instruction.

I was wrong here. :(

> Tracking what insns where changed when is a bit of a mess currently
> (there is last_changed_insns for those insns changed in the current
> pass and ra_modified_insns for insns changed at all, and there is
> some overlap between the two.  This should be cleaned up somewhen).

You right. last_changed_insns and ra_modified_insns may be identical.
I have added ra_modified_insns as alternative/complement of df_insn_modify.
You have added last_changed_insns after ra_modified_insns.
So, today we have a three identical things (if last_changed_insns
equal to ra_modified_insns).

> 
> > Such insns isn't rescanned and will have a wrong web->regclass field.
> 
> They all should be rescanned, see above.  But for later passes no reloads
> are added, just the regclass field is set.

It's a thing about which I have worried.
You have disabled changing of insns in non-first pass of pre-reload.
So, if pre-reload want to change insn then it's a wrong insn and
(probably) it can't be rescanned right. The wrong classes will be choosed
for pseudos (ra_refs) and wrong web->regclass will be calculated too.

Denis.