Re: [PATCH] Update if-conversion for AV

[Devang Patel <dpatel at apple dot com>]

On Sep 21, 2004, at 2:53 PM, Richard Henderson wrote:

> On Thu, Sep 16, 2004 at 09:50:27AM -0700, Devang Patel wrote:
> >     if (else_clause)
> >       {
> > +       tree c2;
> > +       if (!is_gimple_reg(c) && is_gimple_condexpr (c))
> > + 	{
> > + 	  tree new_stmt;
> > + 	  new_stmt = ifc_temp_var (TREE_TYPE (c), unshare_expr (c));
> > + 	  bsi_insert_before (bsi, new_stmt, BSI_SAME_STMT);
> > + 	  c = TREE_OPERAND (new_stmt, 0);
> > + 	}
> > +
> >         /* if 'c' is false then else_clause is reached.  */
> > !       c2 = build1 (TRUTH_NOT_EXPR,
> > ! 		   boolean_type_node,
> > ! 		   unshare_expr (c));
>
> You want invert_truth_value instaed.

OK

> > +   /* Now if possible, merge loop header and block with exit edge.
> > +      This reduces number of basic blocks to 2. Auto vectorizer 
> > addresses
> > +      loops with two nodes only.  */
> > +   if (exit_bb != loop->latch && empty_block_p (loop->latch))
> > +     {
> > +       if (can_merge_blocks_p (loop->header, exit_bb))
> > + 	{
> > + 	  merge_blocks (loop->header, exit_bb);
> > + 	  loop->num_nodes--;
> > + 	}
>
> Why not just run a cleanup_cfg pass?

Thinking like ... Instead of one entire pass why not just
merge two blocks that we know?

I can run a cleanup_cfg() pass in main_tree_if_conversion() so that it
is not used for each loop.

-
Devang