Re: [tree-ssa] [PATCH]: Fix DCE not updating bb->end_tree_p

[Andrew MacLeod <amacleod at redhat dot com>]

On Tue, 2003-07-15 at 16:58, Daniel Berlin wrote:
> Okay to apply?
> Bootstrapped with java, c++, f77, objc, c on x86 with PRE on, and works
> fine.
> 
> 
> 2003-07-15  Daniel Berlin  <dberlin@dberlin.org>
> 
> 	* tree-cfg.c (remove_stmt): Update bb->end_tree_p properly when
> 	stmt_p is the end of the bb.
> 
> Index: tree-cfg.c
> ===================================================================
> RCS file: /cvs/gcc/gcc/gcc/Attic/tree-cfg.c,v
> retrieving revision 1.1.4.127
> diff -u -3 -p -r1.1.4.127 tree-cfg.c
> --- tree-cfg.c	14 Jul 2003 21:18:48 -0000	1.1.4.127
> +++ tree-cfg.c	15 Jul 2003 20:50:23 -0000
> @@ -2045,6 +2045,21 @@ remove_stmt (tree *stmt_p)
>  	SSA_NAME_DEF_STMT (vdef) = build_empty_stmt ();
>      }
> 
> +  /* Update end_tree_p when stmt_p is the end of the tree.  We don't
> +     need to handle the COMPOUND_EXPR case, since that's handled
> +     below.  We don't need to handle the one statement bb case, since
> +     we will remove the bb anyway.  */
> +  if (bb && bb->end_tree_p == stmt_p && bb->head_tree_p != stmt_p
> +      && TREE_CODE (*stmt_p) != COMPOUND_EXPR)
> +    {
> +      block_stmt_iterator bsi;
> +      bsi = bsi_start (bb);
> +      while (bsi_stmt_ptr (bsi) != stmt_p)
> +        bsi_next_in_bb (&bsi, bb);
> +      bsi_prev (&bsi);
> +      bb->end_tree_p = bsi_container (bsi);
> +    }
> +
>    stmt->common.ann = NULL;
> 
>    /* The RHS of a MODIFY_EXPR has an annotation for the benefit of

gah. You realize that bsi_prev() currently redoes what you just did all
over again, so you are currently iterating from bsi_start() twice?

why do the bsi_prev() when you can keep track of the previous stmt
youself while you are looping?

Andrew