[Bug middle-end/44885] New: [4.6 regression] miscompilation of gnat.dg/aliasing3.adb after mem-ref2

[Bug middle-end/44885] New: [4.6 regression] miscompilation of gnat.dg/aliasing3.adb after mem-ref2

[ebotcazou at gcc dot gnu dot org]

There is a regression in gnat.dg visible on various 32-bit platforms:
  http://gcc.gnu.org/ml/gcc-testresults/2010-07/msg00744.html
  http://gcc.gnu.org/ml/gcc-testresults/2010-07/msg00743.html
  http://gcc.gnu.org/ml/gcc-testresults/2010-07/msg00709.html

The problem is that the new code seems to overlook TYPE_NONALIASED_COMPONENT
when building the MEM_REF and thus gives the wrong alias set to the RTL mem,
causing aliasing memory references to be swapped by the scheduler.


-- 
           Summary: [4.6 regression] miscompilation of gnat.dg/aliasing3.adb
                    after mem-ref2
           Product: gcc
           Version: 4.6.0
            Status: UNCONFIRMED
          Keywords: wrong-code
          Severity: normal
          Priority: P3
         Component: middle-end
        AssignedTo: unassigned at gcc dot gnu dot org
        ReportedBy: ebotcazou at gcc dot gnu dot org


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=44885

[Bug middle-end/44885] [4.6 regression] miscompilation of gnat.dg/aliasing3.adb after mem-ref2

[rguenth at gcc dot gnu dot org]

------- Comment #1 from rguenth at gcc dot gnu dot org  2010-07-09 11:42 -------
Can you elaborate?  The relevant type for the MEM_REF is always that of the
base object, which is either a dereference (thus, of sth addressable) or
a decl (which can't be non-aliased as well, no?).


-- 

rguenth at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|---                         |4.6.0


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=44885

[Bug tree-optimization/44885] [4.6 regression] miscompilation of gnat.dg/aliasing3.adb after mem-ref2

[ebotcazou at gcc dot gnu dot org]

------- Comment #2 from ebotcazou at gcc dot gnu dot org  2010-07-09 13:05 -------
> Can you elaborate?  The relevant type for the MEM_REF is always that of the
> base object, which is either a dereference (thus, of sth addressable) or
> a decl (which can't be non-aliased as well, no?).

In 144t.optimized there is:

  aliasing3_pkg__pointer.18_1->a[1]{lb: 1 sz: 4} = 5;
  D.2290_6 = MEM[(integer *)&aliasing3_pkg__block].a[1]{lb: 1 sz: 4};

and the type (integer *) is wrong.  But you're right, it's probably wrong
before since there is in 014t.veclower:

  aliasing3_pkg__pointer.18 = aliasing3_pkg__pointer;
  aliasing3_pkg__pointer.18->a[1]{lb: 1 sz: 4} = 5;
  aliasing3_pkg.test (&aliasing3_pkg__block.a);

and in 022t.cleanup_cfg:

  aliasing3_pkg__pointer.18 = aliasing3_pkg__pointer;
  aliasing3_pkg__pointer.18->a[1]{lb: 1 sz: 4} = 5;
  aliasing3_pkg.test.isra (&aliasing3_pkg__block.a[1]{lb: 1 sz: 4});

So it looks like IPA-SRA breaks TYPE_NONALIASED_COMPONENT here because it takes
the address of a non-aliased component.


-- 

ebotcazou at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
          Component|middle-end                  |tree-optimization


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=44885

[Bug tree-optimization/44885] [4.6 regression] miscompilation of gnat.dg/aliasing3.adb after mem-ref2

[rguenth at gcc dot gnu dot org]

------- Comment #3 from rguenth at gcc dot gnu dot org  2010-07-09 13:23 -------
Ah, that makes sense.  Confirmed btw.


-- 

rguenth at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |jamborm at gcc dot gnu dot
                   |                            |org
             Status|UNCONFIRMED                 |NEW
     Ever Confirmed|0                           |1
   Last reconfirmed|0000-00-00 00:00:00         |2010-07-09 13:23:31
               date|                            |


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=44885

[Bug tree-optimization/44885] [4.6 regression] miscompilation of gnat.dg/aliasing3.adb after mem-ref2

[rguenth at gcc dot gnu dot org]

------- Comment #4 from rguenth at gcc dot gnu dot org  2010-07-09 13:28 -------
Btw, can we amend the stmt verifier to check that the address of such
a thing is not taken?  I'm thinking of verify_expr here:

    case ADDR_EXPR:
      {
        tree tem;

        gcc_assert (is_gimple_address (t));

        /* Skip any references (they will be checked when we recurse down the
           tree) and ensure that any variable used as a prefix is marked
           addressable.  */
        for (x = TREE_OPERAND (t, 0);
             handled_component_p (x);
             x = TREE_OPERAND (x, 0))
          ;

Eric, you probably know best what should be checked, can you prepare a patch
(that hopefully passes testing ... huh)?


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=44885

[Bug tree-optimization/44885] [4.6 regression] miscompilation of gnat.dg/aliasing3.adb after mem-ref2

[ebotcazou at gcc dot gnu dot org]

------- Comment #5 from ebotcazou at gcc dot gnu dot org  2010-07-09 15:59 -------
> Eric, you probably know best what should be checked, can you prepare a patch
> (that hopefully passes testing ... huh)?

Yes, will do, but not before addressing the issues caught by the new size check
on VIEW_CONVERT_EXPR.  We have dozens of regressions in our internal testsuite.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=44885

[Bug tree-optimization/44885] [4.6 regression] miscompilation of gnat.dg/aliasing3.adb after mem-ref2

[rguenther at suse dot de]

------- Comment #6 from rguenther at suse dot de  2010-07-09 16:01 -------
Subject: Re:  [4.6 regression] miscompilation
 of gnat.dg/aliasing3.adb after mem-ref2

On Fri, 9 Jul 2010, ebotcazou at gcc dot gnu dot org wrote:

> ------- Comment #5 from ebotcazou at gcc dot gnu dot org  2010-07-09 15:59 -------
> > Eric, you probably know best what should be checked, can you prepare a patch
> > (that hopefully passes testing ... huh)?
> 
> Yes, will do, but not before addressing the issues caught by the new size check
> on VIEW_CONVERT_EXPR.  We have dozens of regressions in our internal testsuite.

Heh - and I specifically restricted it to registers to avoid that ... ;)

Richard.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=44885

[Bug tree-optimization/44885] [4.6 regression] miscompilation of gnat.dg/aliasing3.adb after mem-ref2

[ebotcazou at gcc dot gnu dot org]

------- Comment #7 from ebotcazou at gcc dot gnu dot org  2010-07-09 16:14 -------
> Heh - and I specifically restricted it to registers to avoid that ... ;)

Yes, that was very kind of you. :-)  But people do appalling things in Ada with
Unchecked_Conversion.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=44885

[Bug tree-optimization/44885] [4.6 regression] miscompilation of gnat.dg/aliasing3.adb after mem-ref2

[ebotcazou at gcc dot gnu dot org]

------- Comment #8 from ebotcazou at gcc dot gnu dot org  2010-07-28 08:16 -------
Fixing.


-- 

ebotcazou at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
         AssignedTo|unassigned at gcc dot gnu   |ebotcazou at gcc dot gnu dot
                   |dot org                     |org
             Status|NEW                         |ASSIGNED
   Last reconfirmed|2010-07-09 13:23:31         |2010-07-28 08:16:48
               date|                            |


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=44885

[Bug tree-optimization/44885] [4.6 regression] miscompilation of gnat.dg/aliasing3.adb after mem-ref2

[ebotcazou at gcc dot gnu dot org]

------- Comment #9 from ebotcazou at gcc dot gnu dot org  2010-07-28 10:03 -------
Subject: Bug 44885

Author: ebotcazou
Date: Wed Jul 28 10:03:00 2010
New Revision: 162621

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=162621
Log:
        PR tree-optimization/44885
        * tree-sra.c (find_param_candidates): Skip pointer types to arrays
        with non-aliased component.

Modified:
    trunk/gcc/ChangeLog
    trunk/gcc/tree-sra.c


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=44885

[Bug tree-optimization/44885] [4.6 regression] miscompilation of gnat.dg/aliasing3.adb after mem-ref2

[ebotcazou at gcc dot gnu dot org]

------- Comment #10 from ebotcazou at gcc dot gnu dot org  2010-07-28 10:04 -------
Schedule of variations:
    unix/-m32

Running target unix/-m32
Using /usr/share/dejagnu/baseboards/unix.exp as board description file for
target.
Using /usr/share/dejagnu/config/unix.exp as generic interface file for target.
Using /home/eric/svn/gcc/gcc/testsuite/config/default.exp as
tool-and-target-specific interface file.
Running /home/eric/svn/gcc/gcc/testsuite/gnat.dg/dg.exp ...
Running /home/eric/svn/gcc/gcc/testsuite/gnat.dg/specs/specs.exp ...
Running /home/eric/svn/gcc/gcc/testsuite/gnat.dg/style/style.exp ...

                === gnat Summary ===

# of expected passes            815
# of expected failures          10
# of unsupported tests          4


-- 

ebotcazou at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|ASSIGNED                    |RESOLVED
         Resolution|                            |FIXED


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=44885