[PATCH 1/2] Auto-generate maybe_fold_and/or_comparisons from match.pd

[Richard Biener]

On Mon, 9 Sep 2019, Martin Liška wrote:

> On 9/9/19 3:10 PM, Richard Biener wrote:
> > On Mon, 9 Sep 2019, Martin Liška wrote:
> > 
> >> Hi.
> >>
> >> I'm sending slightly updated version of the patch where we
> >> need to properly select type in maybe_fold_comparisons_from_match_pd
> >> function for the created SSA_NAMEs. We can be called for a VECTOR_TYPE
> >> and so that we can't return a boolean_type_node.
> >>
> >> Patch can bootstrap on x86_64-linux-gnu and survives regression tests.
> >>
> >> Ready to be installed?
> > 
> > 2019-07-16  Li Jia He  <helijia@linux.ibm.com>
> >             Martin Liska  <mliska@suse.cz>
> > 
> >         * gimple.h (gimple_init): Declare.
> >         (gimple_size): Likewise.
> >         * gimple.c (gimple_init): Remove static and inline restrictions.
> >         (gimple_alloc): Only allocate memory and call gimple_init.
> >         (gimple_size): Likewise.
> > 
> > Likewise?
> 
> Fixed.
> 
> > 
> >         * tree-ssanames.c (init_ssa_name_imm_use): Use make_ssa_name_fn.
> >         (make_ssa_name_fn): New.
> > 
> > You didn't touch make_ssa_name_fn.
> 
> Likewise here.
> 
> > 
> > Since we're needing another iteration:
> > 
> > +  /* Allocate gimple stmt1 on the stack.  */
> > +  gimple *stmt1 = (gimple *) XALLOCAVEC (char, gimple_size 
> > (GIMPLE_ASSIGN, 2));
> > 
> > You can use gassign *stmt1 here so all the gimple_assign_ fns below
> > get cheaper.
> > 
> > +  if (op.resimplify (NULL, follow_all_ssa_edges))
> > +    {
> > +      if (gimple_simplified_result_is_gimple_val (&op))
> > +       {
> > +         tree res = op.ops[0];
> > +         switch (TREE_CODE (res))
> > +           {
> > +           case SSA_NAME:
> > +               {
> > +                 gimple *def = SSA_NAME_DEF_STMT (res);
> > 
> > you shouldn't expand SSA names here unless that SSA name is
> > exactly lhs1 or lhs2 from above.  So
> 
> Ah, got it.
> 
> > 
> >          if (res == lhs1)
> >            return build2 (...);
> >          else if (res == lhs2)
> >            return build2 (..);
> >          else
> >            return res;
> > 
> > plus you miss the case where 'op' became a simplified comparison
> > in itself.  So,
> 
> Yes, that part is included in part 3. I'm going to send the updated patch
> 3 as well soon.
> 
> > 
> >      if (op.code.is_tree_code ()
> >          && TREE_CODE_CLASS ((enum tree_code)op.code) == tcc_comparison)
> >        {
> >           tree op0 = op.ops[0];
> >           tree op1 = op.ops[1];
> >           if (op0 == lhs1 || op0 == lhs2 || op1 == lhs1 || op1 == lhs2)
> >             return NULL_TREE;  /* not simple */
> >           return build2 ((enum tree_code)op.code, op.type,
> >                          op0, op1);
> >        }
> > 
> > note you need not fold_ again.  It's of course ugly that we
> > need to build a GENERIC tree here but that's the current interface
> > and thus OK at the moment.
> 
> I see. But what I need is to insert newly created GIMPLE assignment to
> the provided gimple sequence (gsi), right?

There is no newly created GIMPLE?

Richard.