[Bug tree-optimization/40071] New: ICE in vectorizable_store, at tree-vect-stmts.c:3108 on mipsel abi=n32 and 64, works at 32

[Bug tree-optimization/40071] New: ICE in vectorizable_store, at tree-vect-stmts.c:3108 on mipsel abi=n32 and 64, works at 32

[laurent at guerby dot net]

As shown in:

http://gcc.gnu.org/ml/gcc-testresults/2009-05/msg00618.html

                === gnat tests ===


Running target unix/-mabi=32

                === gnat Summary for unix/-mabi=32 ===

# of expected passes            635
# of expected failures          6

Running target unix/-mabi=n32
FAIL: gnat.dg/loop_optimization1.adb (test for excess errors)

                === gnat Summary for unix/-mabi=n32 ===

# of expected passes            634
# of unexpected failures        1
# of expected failures          6

Running target unix/-mabi=64
FAIL: gnat.dg/loop_optimization3.adb execution test

                === gnat Summary for unix/-mabi=64 ===

# of expected passes            634
# of unexpected failures        1
# of expected failures          6

                === gnat Summary ===

# of expected passes            1903
# of unexpected failures        2
# of expected failures          18
                === libgomp tests ===


So works in abi=32 but ICE in abi=n32 and abi=64

Executing on host: /home/guerby/build-147209/gcc/gnatmake
--GCC=/home/guerby/build-147209/gcc/xgcc
--GNATBIND=/home/guerby/build-147209/gcc/gnatbind
--GNATLINK=/home/guerby/build-147209/gcc/gnatlink -cargs
-B/home/guerby/build-147209/gcc -largs --XCC=/home/guerby/build-147209/gcc/xgcc
-B/home/guerby/build-147209/gcc -margs
--RTS=/home/guerby/build-147209/mips64el-unknown-linux-gnu/./libada -q -f
/home/guerby/trunk/gcc/testsuite/gnat.dg/loop_optimization1.adb  -O3 -c -S 
-mabi=n32 -o loop_optimization1.s    (timeout = 300)
+===========================GNAT BUG DETECTED==============================+

| 4.5.0 20090506 (experimental) [trunk revision 147209]
(mips64el-unknown-linux-gnu) GCC error:|

| in vectorizable_store, at tree-vect-stmts.c:3108                         |

| Error detected around
/home/guerby/trunk/gcc/testsuite/gnat.dg/loop_optimization1.ads:33|

| Please submit a bug report; see http://gcc.gnu.org/bugs.html.            |

| Use a subject line meaningful to you and us to track the bug.            |

| Include the entire contents of this bug box in the report.               |

| Include the exact gcc or gnatmake command that you entered.              |

| Also include sources listed below in gnatchop format                     |

| (concatenated together with no headers between files).                   |

+==========================================================================+


-- 
           Summary: ICE in vectorizable_store, at tree-vect-stmts.c:3108 on
                    mipsel abi=n32 and 64, works at 32
           Product: gcc
           Version: 4.5.0
            Status: UNCONFIRMED
          Keywords: ice-on-valid-code
          Severity: normal
          Priority: P3
         Component: tree-optimization
        AssignedTo: unassigned at gcc dot gnu dot org
        ReportedBy: laurent at guerby dot net
 GCC build triplet: mipsel-linux-gnu
  GCC host triplet: mipsel-linux-gnu
GCC target triplet: mipsel-linux-gnu


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=40071

[Bug tree-optimization/40071] ICE assert aliasing in vectorizable_store, at tree-vect-stmts.c:3108 on mipsel abi=n32 and 64, works at 32

[laurent at guerby dot net]

------- Comment #1 from laurent at guerby dot net  2009-05-08 18:08 -------
According to line information the assert that fails is about aliasing:

          /* If accesses through a pointer to vectype do not alias the original
             memory reference we have a problem.  This should never happen.  */
          gcc_assert (alias_sets_conflict_p (get_alias_set (data_ref),
                      get_alias_set (gimple_assign_lhs (stmt))));


-- 

laurent at guerby dot net changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
            Summary|ICE in vectorizable_store,  |ICE assert aliasing in
                   |at tree-vect-stmts.c:3108 on|vectorizable_store, at tree-
                   |mipsel abi=n32 and 64, works|vect-stmts.c:3108 on mipsel
                   |at 32                       |abi=n32 and 64, works at 32


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=40071

[Bug tree-optimization/40071] ICE assert aliasing in vectorizable_store, at tree-vect-stmts.c:3108 on mipsel abi=n32 and 64, works at 32

[laurent at guerby dot net]

------- Comment #2 from laurent at guerby dot net  2009-05-08 18:15 -------
The Ada test was added by Eric Botcazou:

http://gcc.gnu.org/ml/gcc-patches/2008-04/msg01204.html
<<
The predictive commoning pass performs load hoisting on an expression that can 
throw, thus breaking the CFG; of course this is -fnon-call-exceptions at work.

The proposed fix is to do what LIM does, i.e. to punt on these expressions.
>>

Might not be the same issue again though.


-- 

laurent at guerby dot net changed:

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


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=40071

[Bug tree-optimization/40071] ICE assert aliasing in vectorizable_store, at tree-vect-stmts.c:3108 on mipsel abi=n32 and 64, works at 32

[ebotcazou at gcc dot gnu dot org]

------- Comment #3 from ebotcazou at gcc dot gnu dot org  2009-05-14 14:54 -------
Richard, the subtype bounds removal patch will expose this on x86-64 as well.

Vectorization is not a priority in Ada for the time being so I'll go ahead, but
do you have an idea about the nature of the problem?


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=40071

[Bug tree-optimization/40071] ICE assert aliasing in vectorizable_store, at tree-vect-stmts.c:3108 on mipsel abi=n32 and 64, works at 32

[rguenther at suse dot de]

------- Comment #4 from rguenther at suse dot de  2009-05-15 08:39 -------
Subject: Re:  ICE assert aliasing in
 vectorizable_store, at tree-vect-stmts.c:3108 on mipsel abi=n32 and 64,
 works at 32

On Thu, 14 May 2009, ebotcazou at gcc dot gnu dot org wrote:

> ------- Comment #3 from ebotcazou at gcc dot gnu dot org  2009-05-14 14:54 -------
> Richard, the subtype bounds removal patch will expose this on x86-64 as well.
> 
> Vectorization is not a priority in Ada for the time being so I'll go ahead, but
> do you have an idea about the nature of the problem?

The assertion ensures that when the vectorizer replaces a store the
replacement store conflicts with at least the statements the original
store conflicted.

I will have a look once I can reproduce this on x86_64.

Richard.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=40071

[Bug tree-optimization/40071] ICE (aliasing assert) in vectorizable_store at tree-vect-stmts.c:3117

[ebotcazou at gcc dot gnu dot org]

------- Comment #5 from ebotcazou at gcc dot gnu dot org  2009-05-15 09:14 -------
Confirmed on x86-64.


-- 

ebotcazou at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
     Ever Confirmed|0                           |1
  GCC build triplet|mipsel-linux-gnu            |
   GCC host triplet|mipsel-linux-gnu            |
 GCC target triplet|mipsel-linux-gnu            |
   Last reconfirmed|0000-00-00 00:00:00         |2009-05-15 09:14:03
               date|                            |
            Summary|ICE assert aliasing in      |ICE (aliasing assert) in
                   |vectorizable_store, at tree-|vectorizable_store at tree-
                   |vect-stmts.c:3108 on mipsel |vect-stmts.c:3117
                   |abi=n32 and 64, works at 32 |


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=40071

[Bug tree-optimization/40071] ICE (aliasing assert) in vectorizable_store at tree-vect-stmts.c:3117

[rguenth at gcc dot gnu dot org]

------- Comment #6 from rguenth at gcc dot gnu dot org  2009-05-15 12:11 -------
We are vectorizing

<bb 4>:

<bb 5>:
  # J1_43 = PHI <J1_36(6), J1_3(4)>
  J1_36 = J1_43 + 1;
  D.2354_35 = (long int) J1_36;
  VIEW_CONVERT_EXPR<struct loop_optimization1__r
*[D.2335:D.2339]>(*_init$P_ARRAY_1)[D.2354_35]{lb: D.2335_26 sz: 8} = 0B;
  if (D.2336_6 != J1_36)
    goto <bb 6>;
  else
    goto <bb 7>;

<bb 6>:
  goto <bb 5>;

using a vector pointer of type vector struct loop_optimization1__r **.

The issue here is that for

VIEW_CONVERT_EXPR<struct loop_optimization1__r
*[D.2335:D.2339]>(*_init$P_ARRAY_1)[D.2354_35]

we use the alias set of *_init$P_ARRAY_1 because the array elements are
non-aliased.  As the vectorizer now creates an alias the resulting
stores through the pointer to the vector type do not conflict with the
original stores and we have created wrong code.

How I love Ada ;)

Well, the fix is to teach the vectorizer to never vectorize arrays with
non-aliased components (or any aggregate with such elements).  Or, as
they were introduced to allow more optimization, finally drop this
non-sensical non-aliased component stuff from the middle-end as well.

:P


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=40071

[Bug tree-optimization/40071] ICE (aliasing assert) in vectorizable_store at tree-vect-stmts.c:3117

[laurent at guerby dot net]

------- Comment #7 from laurent at guerby dot net  2009-05-15 16:23 -------
Indeed, at revision 147567 on x86_64-linux:

http://gcc.gnu.org/ml/gcc-testresults/2009-05/msg01235.html

Running target unix/-m64
FAIL: gnat.dg/loop_optimization1.adb (test for excess errors)


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=40071

[Bug tree-optimization/40071] ICE (aliasing assert) in vectorizable_store at tree-vect-stmts.c:3117

[ebotcazou at gcc dot gnu dot org]

------- Comment #8 from ebotcazou at gcc dot gnu dot org  2009-05-15 21:48 -------
> The issue here is that for
> 
> VIEW_CONVERT_EXPR<struct loop_optimization1__r
> *[D.2335:D.2339]>(*_init$P_ARRAY_1)[D.2354_35]
> 
> we use the alias set of *_init$P_ARRAY_1 because the array elements are
> non-aliased.  As the vectorizer now creates an alias the resulting
> stores through the pointer to the vector type do not conflict with the
> original stores and we have created wrong code.

Thanks for the analysis.  Looks like we lack non-aliased vectors here. ;-)


-- 

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|2009-05-15 09:14:03         |2009-05-15 21:48:29
               date|                            |


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=40071

[Bug tree-optimization/40071] ICE (aliasing assert) in vectorizable_store at tree-vect-stmts.c:3117

[aoliva at gcc dot gnu dot org]

------- Comment #9 from aoliva at gcc dot gnu dot org  2009-05-16 06:52 -------
FWIW, there are lots of failures like this building gnat in stage2, during a
bootstrap with BUILD_CONFIG=bootstrap-O3


-- 

aoliva at gcc dot gnu dot org changed:

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


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=40071

[Bug tree-optimization/40071] ICE (aliasing assert) in vectorizable_store at tree-vect-stmts.c:3117

[ebotcazou at gcc dot gnu dot org]

------- Comment #10 from ebotcazou at gcc dot gnu dot org  2009-10-04 08:15 -------
*** Bug 41563 has been marked as a duplicate of this bug. ***


-- 

ebotcazou at gcc dot gnu dot org changed:

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


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=40071

[Bug tree-optimization/40071] ICE (aliasing assert) in vectorizable_store at tree-vect-stmts.c:3117

[ebotcazou at gcc dot gnu dot org]

------- Comment #11 from ebotcazou at gcc dot gnu dot org  2009-10-09 12:45 -------
Subject: Bug 40071

Author: ebotcazou
Date: Fri Oct  9 12:44:59 2009
New Revision: 152585

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=152585
Log:
        PR tree-optimization/40071
        * tree-vect-data-refs.c (vect_create_data_ref_ptr): Build a ref-all
        pointer if the original data reference doesn't conflict with the
        created vector data reference.  Fix long line.

Added:
    trunk/gcc/testsuite/gnat.dg/opt3.adb
    trunk/gcc/testsuite/gnat.dg/opt3_pkg.ads
Modified:
    trunk/gcc/ChangeLog
    trunk/gcc/testsuite/ChangeLog
    trunk/gcc/tree-vect-data-refs.c


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=40071

[Bug tree-optimization/40071] ICE (aliasing assert) in vectorizable_store at tree-vect-stmts.c:3117

[ebotcazou at gcc dot gnu dot org]

------- Comment #12 from ebotcazou at gcc dot gnu dot org  2009-10-09 12:53 -------
This particular ICE shouldn't occur anymore.  However, there is a couple of
ACATS failures at -O3 on x86 again related to vectorization:

                === acats tests ===
FAIL:   cxa4016
FAIL:   cxb3007

                === acats Summary ===
# of expected passes            2313
# of unexpected failures        2


It's apparently SLP failing to clean up EH edges.


-- 

ebotcazou at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|ASSIGNED                    |RESOLVED
         Resolution|                            |FIXED
   Target Milestone|---                         |4.5.0


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=40071