From patchwork Thu Mar 31 20:28:58 2011
Content-Type: text/plain; charset="utf-8"
MIME-Version: 1.0
Content-Transfer-Encoding: 7bit
Subject: [fortran] Fix PR 48352 - regression with ICE with front end
 optimization
Date: Thu, 31 Mar 2011 10:28:58 -0000
From: Thomas Koenig <tkoenig@netcologne.de>
X-Patchwork-Id: 89126
Message-Id: <4D94E40A.2060306@netcologne.de>
To: "fortran@gcc.gnu.org" <fortran@gcc.gnu.org>,
 gcc-patches <gcc-patches@gcc.gnu.org>

Hello world,

the attached patch fixes a 4.7 regression, PR 48352, where a function 
elimination in the expressions for a DO loop caused an ICE.  The ICE was 
caused by interaction of the expression walker with insertion of a 
statement for a DO loop.

Many thanks to Joost for finding the bug and reducing the test case.

To fix the regression, I have disabled this particular optimization for 
expressions within the loop control.

I'd like to overhaul the way that statements are inserted during front 
end optimization, later. This is needed for functions returning arrays 
with bounds not known at compile-time anyway.

Regression-tested.  OK for trunk?

	Thomas

2011-03-31  Thomas Koenig  <tkoenig@gcc.gnu.org>

         PR fortran/48352
         * frontend-passes (cfe_register_funcs):  Don't
         register functions if they appear as iterators in DO loops.

2011-03-31  Thomas Koenig  <tkoenig@gcc.gnu.org>

         PR fortran/48352
         * gfortran.dg/function_optimize_3.f90:  New test.
! { dg-do compile }
! { dg-options "-O" }
! PR 48352 - variable elimination in a DO loop caused segfaults.
! Test case contributed by Joost VandeVondele
program main
  INTEGER, DIMENSION(:), POINTER :: a
  DO I=1,MIN(SIZE(a),SIZE(a))
  ENDDO
END program main


Index: frontend-passes.c
===================================================================
--- frontend-passes.c	(Revision 171793)
+++ frontend-passes.c	(Arbeitskopie)
@@ -137,6 +137,13 @@ static int
 cfe_register_funcs (gfc_expr **e, int *walk_subtrees ATTRIBUTE_UNUSED,
 	  void *data ATTRIBUTE_UNUSED)
 {
+
+  /* FIXME - there is a bug in the insertion code for DO loops.  Bail
+     out here.  */
+
+  if ((*current_code)->op == EXEC_DO)
+    return 0;
+
   if ((*e)->expr_type != EXPR_FUNCTION)
     return 0;