[Bug tree-optimization/46172] New: ICE: in expand_widen_pattern_expr, at optabs.c:522 with -ftree-vectorize -fno-tree-dce

[Bug tree-optimization/46172] New: ICE: in expand_widen_pattern_expr, at optabs.c:522 with -ftree-vectorize -fno-tree-dce

[zsojka at seznam dot cz]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46172

           Summary: ICE: in expand_widen_pattern_expr, at optabs.c:522
                    with -ftree-vectorize -fno-tree-dce
           Product: gcc
           Version: 4.6.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: tree-optimization
        AssignedTo: unassigned@gcc.gnu.org
        ReportedBy: zsojka@seznam.cz


Created attachment 22152
  --> http://gcc.gnu.org/bugzilla/attachment.cgi?id=22152
reduced testcase

Output - 4.6 (trunk):
$ gcc -ftree-vectorize -O -fno-tree-dce pr46172.c  
pr46172.c: In function 'foo':
pr46172.c:3:5: internal compiler error: in expand_widen_pattern_expr, at
optabs.c:522
Please submit a full bug report,
with preprocessed source if appropriate.
See <http://gcc.gnu.org/bugs.html> for instructions.

Output - 4.5:
$ gcc -ftree-vectorize -O -fno-tree-dce pr46172.c 
pr46172.c: In function 'foo':
pr46172.c:3:5: internal compiler error: in expand_expr_real_2, at expr.c:8275
Please submit a full bug report,
with preprocessed source if appropriate.
See <http://gcc.gnu.org/bugs.html> for instructions.

Output - 4.4:
$ gcc -ftree-vectorize -O -fno-tree-dce pr46172.c 
pr46172.c: In function 'foo':
pr46172.c:3: internal compiler error: in c_expand_expr, at c-common.c:5015
Please submit a full bug report,
with preprocessed source if appropriate.
See <http://gcc.gnu.org/bugs.html> for instructions.

Tested revisions:
r165915 - crash
4.5 r165781 - crash
4.4 r165781 - crash
4.4 r149995 - crash

[Bug tree-optimization/46172] [4.2/4.3/4.4/4.5/4.6 Regression] ICE: in expand_widen_pattern_expr, at optabs.c:522 with -ftree-vectorize -fno-tree-dce

[hjl.tools at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46172

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
   Last reconfirmed|                            |2010.10.26 07:11:43
            Summary|ICE: in                     |[4.2/4.3/4.4/4.5/4.6
                   |expand_widen_pattern_expr,  |Regression] ICE: in
                   |at optabs.c:522 with        |expand_widen_pattern_expr,
                   |-ftree-vectorize            |at optabs.c:522 with
                   |-fno-tree-dce               |-ftree-vectorize
                   |                            |-fno-tree-dce
     Ever Confirmed|0                           |1

[Bug tree-optimization/46172] [4.2/4.3/4.4/4.5/4.6 Regression] ICE: in expand_widen_pattern_expr, at optabs.c:522 with -ftree-vectorize -fno-tree-dce

[hjl.tools at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46172

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |hjl.tools at gmail dot com

--- Comment #1 from H.J. Lu <hjl.tools at gmail dot com> 2010-10-26 08:21:32 UTC ---
It is caused by revision 109954:

http://gcc.gnu.org/ml/gcc-cvs/2006-01/msg00752.html

[Bug tree-optimization/46172] [4.3/4.4/4.5/4.6 Regression] ICE: in expand_widen_pattern_expr, at optabs.c:522 with -ftree-vectorize -fno-tree-dce

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46172

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|---                         |4.3.6
            Summary|[4.2/4.3/4.4/4.5/4.6        |[4.3/4.4/4.5/4.6
                   |Regression] ICE: in         |Regression] ICE: in
                   |expand_widen_pattern_expr,  |expand_widen_pattern_expr,
                   |at optabs.c:522 with        |at optabs.c:522 with
                   |-ftree-vectorize            |-ftree-vectorize
                   |-fno-tree-dce               |-fno-tree-dce

[Bug tree-optimization/46172] [4.3/4.4/4.5/4.6 Regression] ICE: in expand_widen_pattern_expr, at optabs.c:522 with -ftree-vectorize -fno-tree-dce

[jakub at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46172

Jakub Jelinek <jakub at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|NEW                         |ASSIGNED
                 CC|                            |jakub at gcc dot gnu.org
         AssignedTo|unassigned at gcc dot       |jakub at gcc dot gnu.org
                   |gnu.org                     |

[Bug tree-optimization/46172] [4.3/4.4/4.5/4.6 Regression] ICE: in expand_widen_pattern_expr, at optabs.c:522 with -ftree-vectorize -fno-tree-dce

[jakub at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46172

Jakub Jelinek <jakub at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|ASSIGNED                    |NEW
         AssignedTo|jakub at gcc dot gnu.org    |unassigned at gcc dot
                   |                            |gnu.org

--- Comment #2 from Jakub Jelinek <jakub at gcc dot gnu.org> 2010-11-02 12:39:16 UTC ---
The problem is that we create DOT_PLUS_EXPR in vect_recog_dot_prod_pattern and
it is kept, because for vector mode the pattern is supported, but then it is
kept also for the final loop handling last few entries and for scalar modes
there is no sdot_prod* pattern.

During expansion DOT_PROD_EXPR is expanded using expand_widen_pattern_expr
which ICEs if there is no handler for it.

To fix this, either the vectorizer would need to undo DOT_PROD_EXPR in the
scalar code if there is no DOT_PROD_EXPR scalar handler, or expansion should
check if there is a handler for it and if not, expand it manually using
multiplication and addition, or all the 3 targets that have
sdot_prod*/udot_prod* patterns should always support both vector and scalar
patterns.

[Bug tree-optimization/46172] [4.3/4.4/4.5/4.6 Regression] ICE: in expand_widen_pattern_expr, at optabs.c:522 with -ftree-vectorize -fno-tree-dce

[irar at il dot ibm.com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46172

--- Comment #3 from Ira Rosen <irar at il dot ibm.com> 2010-11-02 13:11:48 UTC ---
The vectorizer assumes that DCE will be applied after it, and not only in this
case. I don't see why someone would want to use this combination of flags
(-ftree-vectorize -fno-tree-dce). I think it is reasonable to automatically
invoke DCE in case -ftree-vectorize is turned on.

Ira

[Bug tree-optimization/46172] [4.3/4.4/4.5/4.6 Regression] ICE: in expand_widen_pattern_expr, at optabs.c:522 with -ftree-vectorize -fno-tree-dce

[jakub at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46172

--- Comment #4 from Jakub Jelinek <jakub at gcc dot gnu.org> 2010-11-02 13:18:57 UTC ---
Ah, if this is in dead code created by vectorizer that really shouldn't be ever
expanded, then either we need to arrange for DCE to be performed after
vectorizer even if flag_tree_vectorize && !flag_tree_dce (e.g. by using a
different pass structure e.g. instead of pass_dce_loop and use a different gate
for it right after vectorizer), or vectorizer shouldn't create garbage.

-fno-tree-dce is sometimes useful for compiler testing to write testcases e.g.
for expansion and RTL passes bugs, so making it unconditional is not a good
idea.

[Bug tree-optimization/46172] [4.3/4.4/4.5/4.6 Regression] ICE: in expand_widen_pattern_expr, at optabs.c:522 with -ftree-vectorize -fno-tree-dce

[irar at il dot ibm.com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46172

--- Comment #5 from Ira Rosen <irar at il dot ibm.com> 2010-11-02 13:28:14 UTC ---
(In reply to comment #4)
> Ah, if this is in dead code created by vectorizer that really shouldn't be ever
> expanded,

Yes, in this case it's dead code created by the vectorizer that shouldn't be
expanded. But there is also always scalar code that we do not remove except for
stores.

> then either we need to arrange for DCE to be performed after
> vectorizer even if flag_tree_vectorize && !flag_tree_dce (e.g. by using a
> different pass structure e.g. instead of pass_dce_loop and use a different gate
> for it right after vectorizer), 

Yes, this is what I meant.


> or vectorizer shouldn't create garbage.
> 
> -fno-tree-dce is sometimes useful for compiler testing to write testcases e.g.
> for expansion and RTL passes bugs, so making it unconditional is not a good
> idea.

[Bug tree-optimization/46172] [4.3/4.4/4.5/4.6 Regression] ICE: in expand_widen_pattern_expr, at optabs.c:522 with -ftree-vectorize -fno-tree-dce

[jakub at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46172

Jakub Jelinek <jakub at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|NEW                         |ASSIGNED
         AssignedTo|unassigned at gcc dot       |jakub at gcc dot gnu.org
                   |gnu.org                     |

--- Comment #6 from Jakub Jelinek <jakub at gcc dot gnu.org> 2010-11-02 15:20:48 UTC ---
Created attachment 22230
  --> http://gcc.gnu.org/bugzilla/attachment.cgi?id=22230
gcc46-pr46172.patch

Untested fix.

[Bug tree-optimization/46172] [4.3/4.4/4.5/4.6 Regression] ICE: in expand_widen_pattern_expr, at optabs.c:522 with -ftree-vectorize -fno-tree-dce

[jakub at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46172

Jakub Jelinek <jakub at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|ASSIGNED                    |NEW
         AssignedTo|jakub at gcc dot gnu.org    |unassigned at gcc dot
                   |                            |gnu.org

--- Comment #7 from Jakub Jelinek <jakub at gcc dot gnu.org> 2010-11-03 11:18:57 UTC ---
Seems richi prefers a solution in the vectorizer, see
http://gcc.gnu.org/ml/gcc-patches/2010-11/msg00263.html

[Bug tree-optimization/46172] [4.3/4.4/4.5/4.6 Regression] ICE: in expand_widen_pattern_expr, at optabs.c:522 with -ftree-vectorize -fno-tree-dce

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46172

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|NEW                         |ASSIGNED
         AssignedTo|unassigned at gcc dot       |rguenth at gcc dot gnu.org
                   |gnu.org                     |

--- Comment #8 from Richard Guenther <rguenth at gcc dot gnu.org> 2010-11-12 14:13:01 UTC ---
Mine.

[Bug tree-optimization/46172] [4.3/4.4/4.5/4.6 Regression] ICE: in expand_widen_pattern_expr, at optabs.c:522 with -ftree-vectorize -fno-tree-dce

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46172

--- Comment #9 from Richard Guenther <rguenth at gcc dot gnu.org> 2010-11-18 15:17:43 UTC ---
Author: rguenth
Date: Thu Nov 18 15:17:37 2010
New Revision: 166908

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=166908
Log:
2010-11-18  Richard Guenther  <rguenther@suse.de>

    PR tree-optimization/46172
    * tree-vect-loop-manip.c (remove_dead_stmts_from_loop): New
    function.
    (slpeel_tree_peel_loop_to_edge): Call it.

    * gcc.dg/torture/pr46172.c: New testcase.

Added:
    trunk/gcc/testsuite/gcc.dg/torture/pr46172.c
Modified:
    trunk/gcc/ChangeLog
    trunk/gcc/testsuite/ChangeLog
    trunk/gcc/tree-vect-loop-manip.c

[Bug tree-optimization/46172] [4.3/4.4/4.5/4.6 Regression] ICE: in expand_widen_pattern_expr, at optabs.c:522 with -ftree-vectorize -fno-tree-dce

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46172

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|ASSIGNED                    |RESOLVED
      Known to work|                            |4.6.0
         Resolution|                            |FIXED
   Target Milestone|4.3.6                       |4.6.0
      Known to fail|4.6.0                       |

--- Comment #10 from Richard Guenther <rguenth at gcc dot gnu.org> 2010-11-18 15:18:15 UTC ---
Fixed on trunk.  Not worth backporting anywhere.