[Bug target/35496] New: [4.4 Regression] test failures between revs. 132950 and 132974

[Bug target/35496] New: [4.4 Regression] test failures between revs. 132950 and 132974

[dominiq at lps dot ens dot fr]

On i686-apple-darwin9 with rev. 132974, I see the following regressions (not in
rev. 132950):

FAIL: gcc.dg/bf-ms-layout-2.c execution test
FAIL: gcc.dg/bf-ms-layout.c execution test

FAIL: gfortran.dg/array_constructor_12.f90  -O3 -fomit-frame-pointer  execution
test
FAIL: gfortran.dg/array_constructor_12.f90  -O3 -fomit-frame-pointer
-funroll-loops  execution test
FAIL: gfortran.dg/array_constructor_12.f90  -O3 -fomit-frame-pointer
-funroll-all-loops -finline-functions  execution test
FAIL: gfortran.dg/array_constructor_12.f90  -O3 -g  execution test

The failures of gfortran.dg/array_constructor_12.f90 result from a segmentation
fault. All failures disappear with -m64. Between the two builds I have replaced
the first patch (comment #3) for PR33642 by the second one (comment #4).


-- 
           Summary: [4.4 Regression] test failures between revs. 132950 and
                    132974
           Product: gcc
           Version: 4.4.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: target
        AssignedTo: unassigned at gcc dot gnu dot org
        ReportedBy: dominiq at lps dot ens dot fr
 GCC build triplet: i686-apple-darwin9
  GCC host triplet: i686-apple-darwin9
GCC target triplet: i686-apple-darwin9


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=35496

[Bug target/35496] [4.4 Regression] test failures between revs. 132950 and 132974

[dominiq at lps dot ens dot fr]

------- Comment #1 from dominiq at lps dot ens dot fr  2008-03-08 17:10 -------
The failures disappear if I revert revision 132966:

2008-03-05  H.J. Lu  <hongjiu.lu@intel.com>

        * config/i386/i386-modes.def: Use 4 byte alignment on DI for
        32bit host.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=35496

[Bug target/35496] [4.4 Regression] test failures between revs. 132950 and 132974

[hjl dot tools at gmail dot com]

------- Comment #2 from hjl dot tools at gmail dot com  2008-03-08 17:39 -------
Created an attachment (id=15282)
 --> (http://gcc.gnu.org/bugzilla/attachment.cgi?id=15282&action=view)
A patch

Can you try this patch?


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=35496

[Bug target/35496] [4.4 Regression] test failures between revs. 132950 and 132974

[hjl dot tools at gmail dot com]

------- Comment #3 from hjl dot tools at gmail dot com  2008-03-08 17:41 -------
Created an attachment (id=15283)
 --> (http://gcc.gnu.org/bugzilla/attachment.cgi?id=15283&action=view)
A real patch

Correct patch.


-- 

hjl dot tools at gmail dot com changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
  Attachment #15282|0                           |1
        is obsolete|                            |


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=35496

[Bug target/35496] [4.4 Regression] test failures between revs. 132950 and 132974

[ubizjak at gmail dot com]

------- Comment #4 from ubizjak at gmail dot com  2008-03-08 18:16 -------
(In reply to comment #0)

> FAIL: gfortran.dg/array_constructor_12.f90  -O3 -fomit-frame-pointer  execution
> test
> FAIL: gfortran.dg/array_constructor_12.f90  -O3 -fomit-frame-pointer
> -funroll-loops  execution test
> FAIL: gfortran.dg/array_constructor_12.f90  -O3 -fomit-frame-pointer
> -funroll-all-loops -finline-functions  execution test
> FAIL: gfortran.dg/array_constructor_12.f90  -O3 -g  execution test
> 
> The failures of gfortran.dg/array_constructor_12.f90 result from a segmentation
> fault. All failures disappear with -m64. Between the two builds I have replaced
> the first patch (comment #3) for PR33642 by the second one (comment #4).

This is vectorizer in action. Note that -msse2 is the default for
i686-apple-darwin, and the testcase indeed fails with -O2 -msse2
-ftree-vectorize on linux too.

It looks like wrong assumption for DImode alignment in the vectorizer to me.

Confirmed on linux using "-O2 -msse2 -ftree-vectorize".


-- 

ubizjak at gmail dot com changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
     Ever Confirmed|0                           |1
   Last reconfirmed|0000-00-00 00:00:00         |2008-03-08 18:16:11
               date|                            |


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=35496

[Bug target/35496] [4.4 Regression] test failures between revs. 132950 and 132974

[ubizjak at gmail dot com]

------- Comment #5 from ubizjak at gmail dot com  2008-03-08 18:18 -------
(In reply to comment #3)
> Created an attachment (id=15283)
 --> (http://gcc.gnu.org/bugzilla/attachment.cgi?id=15283&action=view) [edit]
> A real patch
> 
> Correct patch.

> -ADJUST_ALIGNMENT (DI, (TARGET_64BIT || TARGET_ALIGN_DOUBLE) ? 8 : 4);
> +ADJUST_ALIGNMENT (DI, (TARGET_64BIT
> +		       || TARGET_MACHO
> +		       || TARGET_ALIGN_DOUBLE) ? 8 : 4);

This failure is not darwin specific, so the fix is wrong. Please see Comment
#4.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=35496

[Bug target/35496] [4.4 Regression] test failures between revs. 132950 and 132974

[ubizjak at gmail dot com]

------- Comment #6 from ubizjak at gmail dot com  2008-03-08 18:31 -------
Segfault is here due to unaligned access:

        ...
        movdqa  .LC1, %xmm0
        movdqa  .LC2, %xmm2
        ...


        .align 8
.LC1:
        .long   2147483647
        .long   0
        .long   -2147483647
        .long   0
        .align 8
.LC2:
        .long   4
        .long   0
        .long   4
        .long   0


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=35496

[Bug target/35496] [4.4 Regression] test failures between revs. 132950 and 132974

[dominiq at lps dot ens dot fr]

------- Comment #7 from dominiq at lps dot ens dot fr  2008-03-08 22:31 -------
Subject: Re:  [4.4 Regression] test failures between revs.
 132950 and 132974

> Can you try this patch?

For the record, the right patch works on intel-apple-darwin9


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=35496

[Bug target/35496] [4.4 Regression] test failures between revs. 132950 and 132974

[ubizjak at gmail dot com]

------- Comment #8 from ubizjak at gmail dot com  2008-03-09 11:39 -------
Patch in testing:

Index: i386.c
===================================================================
--- i386.c      (revision 133051)
+++ i386.c      (working copy)
@@ -16789,7 +16789,8 @@
 int
 ix86_constant_alignment (tree exp, int align)
 {
-  if (TREE_CODE (exp) == REAL_CST)
+  if (TREE_CODE (exp) == REAL_CST || TREE_CODE (exp) == VECTOR_CST
+      || TREE_CODE (exp) == INTEGER_CST)
     {
       if (TYPE_MODE (TREE_TYPE (exp)) == DFmode && align < 64)
        return 64;


-- 

ubizjak at gmail dot com changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
         AssignedTo|unassigned at gcc dot gnu   |ubizjak at gmail dot com
                   |dot org                     |
             Status|NEW                         |ASSIGNED
   Last reconfirmed|2008-03-08 18:16:11         |2008-03-09 11:39:09
               date|                            |


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=35496

[Bug target/35496] [4.4 Regression] test failures between revs. 132950 and 132974

[uros at gcc dot gnu dot org]

------- Comment #9 from uros at gcc dot gnu dot org  2008-03-09 13:38 -------
Subject: Bug 35496

Author: uros
Date: Sun Mar  9 13:37:52 2008
New Revision: 133052

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=133052
Log:
        PR target/35496
        * config/i386/i386.c (ix86_constant_algnment): Compute alignment using
        ALIGN_MODE_128 for VECTOR_CST and INTEGER_CST in addition to REAL_CST.


Modified:
    trunk/gcc/ChangeLog
    trunk/gcc/config/i386/i386.c


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=35496

[Bug target/35496] [4.4 Regression] test failures between revs. 132950 and 132974

[hjl dot tools at gmail dot com]

------- Comment #10 from hjl dot tools at gmail dot com  2008-03-09 14:06 -------
Can we add a few testcases for Linux/ia32?


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=35496

[Bug target/35496] [4.4 Regression] test failures between revs. 132950 and 132974

[ubizjak at gmail dot com]

------- Comment #11 from ubizjak at gmail dot com  2008-03-09 17:09 -------
(In reply to comment #10)
> Can we add a few testcases for Linux/ia32?

http://gcc.gnu.org/ml/gcc-patches/2008-03/msg00585.html


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=35496

[Bug target/35496] [4.4 Regression] test failures between revs. 132950 and 132974

[dominiq at lps dot ens dot fr]

------- Comment #12 from dominiq at lps dot ens dot fr  2008-03-09 19:32 -------
On i686-apple-darwin9 the commited patch of comment #8 fixes the failure for
gfortran.dg/array_constructor_12.f90  -O3, but not the failures for
gcc.dg/bf-ms-layout*.c which require the patch from comment #5.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=35496

[Bug target/35496] [4.4 Regression] test failures between revs. 132950 and 132974

[ubizjak at gmail dot com]

------- Comment #13 from ubizjak at gmail dot com  2008-03-09 21:26 -------
(In reply to comment #12)
> On i686-apple-darwin9 the commited patch of comment #8 fixes the failure for
> gfortran.dg/array_constructor_12.f90  -O3, but not the failures for
> gcc.dg/bf-ms-layout*.c which require the patch from comment #5.

The patch from Comment #5 disables alignment optimization for darwin. FWIW,
these tests fail on x86_64-linux* and i686-linux*, and the failure is probably
connected with PR 27948.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=35496

[Bug target/35496] [4.4 Regression] test failures between revs. 132950 and 132974

[hjl dot tools at gmail dot com]

------- Comment #14 from hjl dot tools at gmail dot com  2008-03-09 21:48 -------
Does MS bitfield struct require 8byte alignment for long long field
for 32bit target?


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=35496

[Bug target/35496] [4.4 Regression] test failures between revs. 132950 and 132974

[hjl dot tools at gmail dot com]

------- Comment #15 from hjl dot tools at gmail dot com  2008-03-09 21:54 -------
We may need to update ADJUST_FIELD_ALIGN to support MS-bitfield.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=35496

[Bug target/35496] [4.4 Regression] test failures between revs. 132950 and 132974

[ubizjak at gmail dot com]

------- Comment #16 from ubizjak at gmail dot com  2008-03-10 12:48 -------
(In reply to comment #14)
> Does MS bitfield struct require 8byte alignment for long long field
> for 32bit target?

Can you try following patch (this will aling DImode values to their natural
alignment - if bigger than default alignment - when MS bitfield is used):

Index: stor-layout.c
===================================================================
--- stor-layout.c       (revision 133071)
+++ stor-layout.c       (working copy)
@@ -699,6 +699,9 @@ update_alignment_for_field (record_layou
                 && ! integer_zerop (DECL_SIZE (rli->prev_field)))))
        {
          unsigned int type_align = TYPE_ALIGN (type);
+         unsigned int type_size
+           = tree_low_cst (TYPE_SIZE_UNIT (type), 1) * BITS_PER_UNIT;
+         type_align = MAX (type_align, type_size);
          type_align = MAX (type_align, desired_align);
          if (maximum_field_alignment != 0)
            type_align = MIN (type_align, maximum_field_alignment);


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=35496

[Bug target/35496] [4.4 Regression] test failures between revs. 132950 and 132974

[ubizjak at gmail dot com]

------- Comment #17 from ubizjak at gmail dot com  2008-03-10 13:15 -------
(In reply to comment #16)
> (In reply to comment #14)
> > Does MS bitfield struct require 8byte alignment for long long field
> > for 32bit target?
> 
> Can you try following patch (this will aling DImode values to their natural
> alignment - if bigger than default alignment - when MS bitfield is used):

Ah, we can use:

+         unsigned int type_size
+           = tree_low_cst (TYPE_SIZE (type), 1);


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=35496

[Bug target/35496] [4.4 Regression] test failures between revs. 132950 and 132974

[uros at gcc dot gnu dot org]

------- Comment #18 from uros at gcc dot gnu dot org  2008-03-10 14:14 -------
Subject: Bug 35496

Author: uros
Date: Mon Mar 10 14:13:40 2008
New Revision: 133079

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=133079
Log:
        Backport from mainline:
        2008-03-09  Uros Bizjak  <ubizjak@gmail.com>

        PR target/35496
        * config/i386/i386.c (ix86_constant_alignment): Compute alignment using
        ALIGN_MODE_128 for VECTOR_CST and INTEGER_CST in addition to REAL_CST.

        2008-03-04  Uros Bizjak  <ubizjak@gmail.com>

        PR middle-end/35456
        * fold-const.c (fold_cond_expr_with_comparison): Prevent
        transformations for modes that have signed zeros.
        * ifcvt.c (noce_try_abs): Ditto.

testsuite/ChangeLog:

        Backport from mainline:
        2008-03-04  Uros Bizjak  <ubizjak@gmail.com>

        PR middle-end/35456
        * gcc.c-torture/execute/pr35456.c: New test.


Added:
    branches/gcc-4_3-branch/gcc/testsuite/gcc.c-torture/execute/pr35456.c
Modified:
    branches/gcc-4_3-branch/gcc/ChangeLog
    branches/gcc-4_3-branch/gcc/config/i386/i386.c
    branches/gcc-4_3-branch/gcc/fold-const.c
    branches/gcc-4_3-branch/gcc/testsuite/ChangeLog


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=35496

[Bug target/35496] [4.4 Regression] test failures between revs. 132950 and 132974

[dominiq at lps dot ens dot fr]

------- Comment #19 from dominiq at lps dot ens dot fr  2008-03-10 14:15 -------
Subject: Re:  [4.4 Regression] test failures between revs.
 132950 and 132974

> Can you try following patch ...

Both variants work for bf-ms-layout*.c, starting regtest ASAP.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=35496

[Bug target/35496] [4.4 Regression] test failures between revs. 132950 and 132974

[dominiq at lps dot ens dot fr]

------- Comment #20 from dominiq at lps dot ens dot fr  2008-03-10 20:45 -------
On i686-apple-darwin9 regtested without regression with the patch of comment
#17.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=35496

[Bug target/35496] [4.4 Regression] test failures between revs. 132950 and 132974

[pinskia at gcc dot gnu dot org]

-- 

pinskia at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |pinskia at gcc dot gnu dot
                   |                            |org
           Severity|normal                      |critical
   Target Milestone|---                         |4.4.0


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=35496

[Bug target/35496] [4.4 Regression] test failures between revs. 132950 and 132974

[rguenth at gcc dot gnu dot org]

-- 

rguenth at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Keywords|                            |wrong-code
           Priority|P3                          |P1


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=35496

[Bug target/35496] [4.4 Regression] test failures between revs. 132950 and 132974

[uros at gcc dot gnu dot org]

------- Comment #21 from uros at gcc dot gnu dot org  2008-03-19 14:51 -------
Subject: Bug 35496

Author: uros
Date: Wed Mar 19 14:50:15 2008
New Revision: 133343

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=133343
Log:
        PR target/35496
        * stor-layout.c (update_alignment_for_field): Set minimum alignment
        of the underlying type of a MS bitfield layout to the natural
        alignment of the type.


Modified:
    trunk/gcc/ChangeLog
    trunk/gcc/stor-layout.c


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=35496

[Bug target/35496] [4.4 Regression] test failures between revs. 132950 and 132974

[ubizjak at gmail dot com]

------- Comment #22 from ubizjak at gmail dot com  2008-03-19 14:57 -------
Fixed.


-- 

ubizjak at gmail dot com changed:

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


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=35496

[Bug target/35496] [4.4 Regression] test failures between revs. 132950 and 132974

[kkojima at gcc dot gnu dot org]

------- Comment #23 from kkojima at gcc dot gnu dot org  2008-03-23 00:03 -------
It seems that the patch of #21 causes a segfault of cc1 for
the testcase below with -mms-bitfields on i686-pc-linux-gnu.

struct foo
{
  unsigned int i;
  unsigned char a[];
};


-- 

kkojima at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |kkojima at gcc dot gnu dot
                   |                            |org


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=35496

[Bug target/35496] [4.4 Regression] test failures between revs. 132950 and 132974

[uros at gcc dot gnu dot org]

------- Comment #24 from uros at gcc dot gnu dot org  2008-03-23 12:08 -------
Subject: Bug 35496

Author: uros
Date: Sun Mar 23 12:07:34 2008
New Revision: 133461

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=133461
Log:
        Revert:
        2008-03-05  H.J. Lu  <hongjiu.lu@intel.com>

        * config/i386/i386-modes.def: Use 4 byte alignment on DI for
        32bit host.

        2008-03-19  Uros Bizjak  <ubizjak@gmail.com>

        PR target/35496
        * stor-layout.c (update_alignment_for_field): Set minimum alignment
        of the underlying type of a MS bitfield layout to the natural
        alignment of the type.

        2008-03-22  Uros Bizjak  <ubizjak@gmail.com>

        * config/i386/i386.c (assign_386_stack_local): Align DImode slots
        to their natural alignment to avoid store forwarding stalls.


Modified:
    trunk/gcc/ChangeLog
    trunk/gcc/config/i386/i386-modes.def
    trunk/gcc/config/i386/i386.c
    trunk/gcc/stor-layout.c


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=35496

[Bug target/35496] [4.4 Regression] test failures between revs. 132950 and 132974

[hjl at gcc dot gnu dot org]

------- Comment #25 from hjl at gcc dot gnu dot org  2008-03-23 20:52 -------
Subject: Bug 35496

Author: hjl
Date: Sun Mar 23 20:51:29 2008
New Revision: 133466

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=133466
Log:
2008-03-23  Uros Bizjak  <ubizjak@gmail.com>

        Revert:
        2008-03-05  H.J. Lu  <hongjiu.lu@intel.com>

        * config/i386/i386-modes.def: Use 4 byte alignment on DI for
        32bit host.

        2008-03-19  Uros Bizjak  <ubizjak@gmail.com>

        PR target/35496
        * stor-layout.c (update_alignment_for_field): Set minimum alignment
        of the underlying type of a MS bitfield layout to the natural
        alignment of the type.

Modified:
    branches/stack/gcc/ChangeLog.stackalign
    branches/stack/gcc/config/i386/i386-modes.def
    branches/stack/gcc/stor-layout.c


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=35496