[Bug middle-end/48440] New: [4.7 Regression] FAIL: gcc.c-torture/compile/labels-3.c

[Bug middle-end/48440] New: [4.7 Regression] FAIL: gcc.c-torture/compile/labels-3.c

[hjl.tools at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=48440

           Summary: [4.7 Regression] FAIL:
                    gcc.c-torture/compile/labels-3.c
           Product: gcc
           Version: 4.7.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: middle-end
        AssignedTo: unassigned@gcc.gnu.org
        ReportedBy: hjl.tools@gmail.com
                CC: matz@gcc.gnu.org


Revision 171903:

http://gcc.gnu.org/ml/gcc-cvs/2011-04/msg00095.html

breaks gcc.c-torture/compile/labels-3.c for targets where
Pmode == DImode and ptr_mode == SImode like x32. Revert

    * cgraphbuild.c (record_reference): Canonicalize constructor
    values.

fixes the regression.

[Bug middle-end/48440] [4.7 Regression] FAIL: gcc.c-torture/compile/labels-3.c

[hjl.tools at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=48440

--- Comment #1 from H.J. Lu <hjl.tools at gmail dot com> 2011-04-04 22:05:30 UTC ---
The record_reference change remove the cast, which is needed for
Pmode != ptr_mode:

diff -upr bad/x.i.143r.expand good/x.i.143r.expand
--- bad/x.i.143r.expand    2011-04-04 15:02:05.652458274 -0700
+++ good/x.i.143r.expand    2011-04-04 15:02:55.233678782 -0700
@@ -5,7 +5,7 @@ foo (int a)
 {
   void * gotovar.0;
   void * p;
-  static const short int ar[2] = {0, (short unsigned int) &l2 - (short
unsigned int) &l1};
+  static const short int ar[2] = {0, (short int) ((short unsigned int) (int)
&l2 - (short unsigned int) (int) &l1)};
   int D.2689;
   unsigned int D.2688;
   const short int D.2687;
@@ -105,7 +105,7 @@ foo (int a)
 {
   void * gotovar.0;
   void * p;
-  static const short int ar[2] = {0, (short unsigned int) &l2 - (short
unsigned int) &l1};
+  static const short int ar[2] = {0, (short int) ((short unsigned int) (int)
&l2 - (short unsigned int) (int) &l1)};
   int D.2689;
   unsigned int D.2688;
   const short int D.2687;

[Bug middle-end/48440] [4.7 Regression] FAIL: gcc.c-torture/compile/labels-3.c

[hjl.tools at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=48440

--- Comment #2 from H.J. Lu <hjl.tools at gmail dot com> 2011-04-04 22:15:54 UTC ---
This patch:

diff --git a/gcc/cgraphbuild.c b/gcc/cgraphbuild.c
index 3948cf6..c80d7ab 100644
--- a/gcc/cgraphbuild.c
+++ b/gcc/cgraphbuild.c
@@ -56,7 +56,10 @@ record_reference (tree *tp, int *walk_subtrees, void *data)
   t = canonicalize_constructor_val (t);
   if (!t)
     t = *tp;
-  else if (t != *tp)
+  else if (t != *tp
+       && !(Pmode != ptr_mode
+        && TREE_CODE (t) == ADDR_EXPR
+        && TREE_CODE (*tp) == CONVERT_EXPR))
     *tp = t;

   switch (TREE_CODE (t))

seems to work.

[Bug middle-end/48440] [4.7 Regression] FAIL: gcc.c-torture/compile/labels-3.c

[hjl.tools at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=48440

H.J. Lu <hjl.tools at gmail dot com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                URL|                            |http://gcc.gnu.org/ml/gcc-p
                   |                            |atches/2011-04/msg00281.htm
                   |                            |l
   Target Milestone|---                         |4.7.0

--- Comment #3 from H.J. Lu <hjl.tools at gmail dot com> 2011-04-05 00:51:43 UTC ---
A patch is posted at

http://gcc.gnu.org/ml/gcc-patches/2011-04/msg00281.html

[Bug middle-end/48440] [4.7 Regression] FAIL: gcc.c-torture/compile/labels-3.c

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=48440

--- Comment #4 from Richard Guenther <rguenth at gcc dot gnu.org> 2011-04-05 11:02:50 UTC ---
-  static const short int ar[2] = {0, (short unsigned int) &l2 - (short
unsigned int) &l1};
+  static const short int ar[2] = {0, (short int) ((short unsigned int) (int)
&l2 - (short unsigned int) (int) &l1)};

this transformation is completely reasonable for any pointer size.  The
bug must be elsewhere and latent.

[Bug middle-end/48440] [4.7 Regression] FAIL: gcc.c-torture/compile/labels-3.c

[hjl at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=48440

--- Comment #5 from hjl at gcc dot gnu.org <hjl at gcc dot gnu.org> 2011-04-08 00:23:05 UTC ---
Author: hjl
Date: Fri Apr  8 00:23:02 2011
New Revision: 172156

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=172156
Log:
Properly get inner mode for constant op0.

2011-04-06  H.J. Lu  <hongjiu.lu@intel.com>

    PR middle-end/48440
    * expr.c (expand_expr_real_2): Get inner mode from op0 instead
    of treeop0 for constant op0.

Modified:
    branches/x32/gcc/ChangeLog.x32
    branches/x32/gcc/expr.c

[Bug middle-end/48440] [4.7 Regression] FAIL: gcc.c-torture/compile/labels-3.c

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=48440

Richard Guenther <rguenth at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |RESOLVED
         Resolution|                            |INVALID

--- Comment #6 from Richard Guenther <rguenth at gcc dot gnu.org> 2011-10-10 11:44:14 UTC ---
Invalid thus.