[Bug tree-optimization/48290] New: FAIL: gcc.dg/vect/pr38529.c, ICE in vect_get_vec_def_for_operand, at tree-vect-stmts.c:1072

[Bug tree-optimization/48290] New: FAIL: gcc.dg/vect/pr38529.c, ICE in vect_get_vec_def_for_operand, at tree-vect-stmts.c:1072

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=48290

           Summary: FAIL: gcc.dg/vect/pr38529.c, ICE in
                    vect_get_vec_def_for_operand, at
                    tree-vect-stmts.c:1072
           Product: gcc
           Version: 4.7.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: tree-optimization
        AssignedTo: unassigned@gcc.gnu.org
        ReportedBy: rguenth@gcc.gnu.org
                CC: irar@gcc.gnu.org


The vectorizer seems to mishandle PHIs for

<bb 3>:

<bb 4>:
  # j_17 = PHI <j_5(3), 0(7)>
  # ivtmp.6_14 = PHI <ivtmp.6_15(3), 17(7)>
  j_5 = j_17 + 1;
  ivtmp.6_15 = ivtmp.6_14 - 1;
  if (ivtmp.6_15 != 0)
    goto <bb 3>;
  else
    goto <bb 5>;

<bb 5>:
  # a_I_lsm.4_2 = PHI <0.0(4)>
  a[i_16] = a_I_lsm.4_2;
  i_6 = i_16 + 1;
  ivtmp.5_12 = ivtmp.5_10 - 1;
  if (ivtmp.5_12 != 0)
    goto <bb 6>;
  else
    goto <bb 8>;

<bb 6>:

<bb 7>:
  # i_16 = PHI <0(2), i_6(6)>
  # ivtmp.5_10 = PHI <4(2), ivtmp.5_12(6)>
  goto <bb 4>;


in outer loop vectorization.  Somehow this PHI isn't processed at all.

While I will fix copyprop further to propagate the zero to its single use
I think the vectorizer shouldn't ICE on this case either.

Ira, can you have a look?

Thanks,
Richard.

[Bug tree-optimization/48290] FAIL: gcc.dg/vect/pr38529.c, ICE in vect_get_vec_def_for_operand, at tree-vect-stmts.c:1072

[irar at il dot ibm.com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=48290

Ira Rosen <irar at il dot ibm.com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |irar at il dot ibm.com

--- Comment #1 from Ira Rosen <irar at il dot ibm.com> 2011-03-25 18:04:20 UTC ---
We don't expect a PHI that is neither induction nor reduction as a def stmt of
an operand. I'll try to fix this on Sunday.

Ira

[Bug tree-optimization/48290] FAIL: gcc.dg/vect/pr38529.c, ICE in vect_get_vec_def_for_operand, at tree-vect-stmts.c:1072

[dominiq at lps dot ens.fr]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=48290

Dominique d'Humieres <dominiq at lps dot ens.fr> changed:

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

--- Comment #2 from Dominique d'Humieres <dominiq at lps dot ens.fr> 2011-03-27 15:59:56 UTC ---
*** Bug 48300 has been marked as a duplicate of this bug. ***

[Bug tree-optimization/48290] FAIL: gcc.dg/vect/pr38529.c, ICE in vect_get_vec_def_for_operand, at tree-vect-stmts.c:1072

[irar at il dot ibm.com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=48290

--- Comment #3 from Ira Rosen <irar at il dot ibm.com> 2011-03-28 10:30:55 UTC ---
I am going to test this patch. It checks that if we have a phi in outer loop in
the basic block after the inner loop, then this phi is really inner loop's exit
phi, i.e., its operand is defined in the inner loop. 

Index: tree-vect-loop.c
===================================================================
--- tree-vect-loop.c    (revision 171466)
+++ tree-vect-loop.c    (working copy)
@@ -1184,11 +1184,11 @@ vect_analyze_loop_operations (loop_vec_i
               print_gimple_stmt (vect_dump, phi, 0, TDF_SLIM);
             }

+          /* Inner-loop loop-closed exit phi in outer-loop vectorization
+             (i.e. a phi in the tail of the outer-loop).  */
           if (! is_loop_header_bb_p (bb))
             {
-              /* inner-loop loop-closed exit phi in outer-loop vectorization
-                 (i.e. a phi in the tail of the outer-loop).
-                 FORNOW: we currently don't support the case that these phis
+              /* FORNOW: we currently don't support the case that these phis
                  are not used in the outerloop (unless it is double reduction,
                  i.e., this phi is vect_reduction_def), cause this case
                  requires to actually do something here.  */
@@ -1202,6 +1202,32 @@ vect_analyze_loop_operations (loop_vec_i
                              "Unsupported loop-closed phi in outer-loop.");
                   return false;
                 }
+
+              /* If PHI is used in the outer loop, we check that its operand
+                 is defined in the inner loop.  */
+              if (STMT_VINFO_RELEVANT_P (stmt_info))
+                {
+                  tree phi_op;
+                  gimple op_def_stmt;
+
+                  if (gimple_phi_num_args (phi) != 1)
+                    return false;
+
+                  phi_op = PHI_ARG_DEF (phi, 0);
+                  if (TREE_CODE (phi_op) != SSA_NAME)
+                    return false;
+
+                  op_def_stmt = SSA_NAME_DEF_STMT (phi_op);
+                  if (!op_def_stmt || !vinfo_for_stmt (op_def_stmt))
+                    return false;
+
+                  if (STMT_VINFO_RELEVANT (vinfo_for_stmt (op_def_stmt))
+                        != vect_used_in_outer
+                      && STMT_VINFO_RELEVANT (vinfo_for_stmt (op_def_stmt))
+                           != vect_used_in_outer_by_reduction)
+                    return false;
+                }
+
               continue;
             }

[Bug tree-optimization/48290] FAIL: gcc.dg/vect/pr38529.c, ICE in vect_get_vec_def_for_operand, at tree-vect-stmts.c:1072

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=48290

--- Comment #4 from Richard Guenther <rguenth at gcc dot gnu.org> 2011-03-28 10:38:27 UTC ---
Thanks Ira.

[Bug tree-optimization/48290] FAIL: gcc.dg/vect/pr38529.c, ICE in vect_get_vec_def_for_operand, at tree-vect-stmts.c:1072

[dominiq at lps dot ens.fr]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=48290

--- Comment #5 from Dominique d'Humieres <dominiq at lps dot ens.fr> 2011-03-28 18:26:52 UTC ---
The patch in comment #3 fixes the ICE, but the test still fails:

FAIL: gcc.dg/vect/pr38529.c scan-tree-dump-times vect "OUTER LOOP VECTORIZED" 1

Compiling with -Ofast -ftree-vectorizer-verbose=2 I get

/opt/gcc/_clean/gcc/testsuite/gcc.dg/vect/pr38529.c:11: note: not vectorized:
unsupported data-type
/opt/gcc/_clean/gcc/testsuite/gcc.dg/vect/pr38529.c:6: note: vectorized 0 loops
in function.

With revision 171398 I get

/opt/gcc/_clean/gcc/testsuite/gcc.dg/vect/pr38529.c:10: note: LOOP VECTORIZED.
/opt/gcc/_clean/gcc/testsuite/gcc.dg/vect/pr38529.c:10: note: OUTER LOOP
VECTORIZED.
/opt/gcc/_clean/gcc/testsuite/gcc.dg/vect/pr38529.c:6: note: vectorized 1 loops
in function.

No other test fails in gcc/testsuite/*/vect/* (full test for tonight).

[Bug tree-optimization/48290] FAIL: gcc.dg/vect/pr38529.c, ICE in vect_get_vec_def_for_operand, at tree-vect-stmts.c:1072

[irar at il dot ibm.com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=48290

--- Comment #6 from Ira Rosen <irar at il dot ibm.com> 2011-03-29 07:04:08 UTC ---
(In reply to comment #5)
> The patch in comment #3 fixes the ICE, but the test still fails:
> 
> FAIL: gcc.dg/vect/pr38529.c scan-tree-dump-times vect "OUTER LOOP VECTORIZED" 1
> 

This is expected, because the patch only prevents vectorization of unexpected
phis. As Richard wrote, copyprop will have to be fixed to propagate the zero.
After that the loop will get vectorized again.

[Bug tree-optimization/48290] FAIL: gcc.dg/vect/pr38529.c, ICE in vect_get_vec_def_for_operand, at tree-vect-stmts.c:1072

[rguenther at suse dot de]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=48290

--- Comment #7 from rguenther at suse dot de <rguenther at suse dot de> 2011-03-29 09:22:15 UTC ---
On Mon, 28 Mar 2011, dominiq at lps dot ens.fr wrote:

> http://gcc.gnu.org/bugzilla/show_bug.cgi?id=48290
> 
> --- Comment #5 from Dominique d'Humieres <dominiq at lps dot ens.fr> 2011-03-28 18:26:52 UTC ---
> The patch in comment #3 fixes the ICE, but the test still fails:
> 
> FAIL: gcc.dg/vect/pr38529.c scan-tree-dump-times vect "OUTER LOOP VECTORIZED" 1
> 
> Compiling with -Ofast -ftree-vectorizer-verbose=2 I get
> 
> /opt/gcc/_clean/gcc/testsuite/gcc.dg/vect/pr38529.c:11: note: not vectorized:
> unsupported data-type
> /opt/gcc/_clean/gcc/testsuite/gcc.dg/vect/pr38529.c:6: note: vectorized 0 loops
> in function.

That's expected.  I suppose the testcase is no longer testing what it
was supposed to test (even before, as LIM was applying store-sinking to
the store).

[Bug tree-optimization/48290] FAIL: gcc.dg/vect/pr38529.c, ICE in vect_get_vec_def_for_operand, at tree-vect-stmts.c:1072

[irar at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=48290

--- Comment #8 from irar at gcc dot gnu.org 2011-03-29 10:26:30 UTC ---
Author: irar
Date: Tue Mar 29 10:26:25 2011
New Revision: 171657

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=171657
Log:

        PR tree-optimization/48290
        * tree-vect-loop.c (vect_analyze_loop_operations): In outer loop
        vectorization, check that relevant phis in the basic block after 
        the inner loop are really inner loop's exit phis.


Modified:
    trunk/gcc/ChangeLog
    trunk/gcc/tree-vect-loop.c

[Bug tree-optimization/48290] FAIL: gcc.dg/vect/pr38529.c, ICE in vect_get_vec_def_for_operand, at tree-vect-stmts.c:1072

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=48290

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

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

--- Comment #10 from Richard Guenther <rguenth at gcc dot gnu.org> 2011-04-15 14:03:26 UTC ---
And fixed.

[Bug tree-optimization/48290] FAIL: gcc.dg/vect/pr38529.c, ICE in vect_get_vec_def_for_operand, at tree-vect-stmts.c:1072

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=48290

--- Comment #9 from Richard Guenther <rguenth at gcc dot gnu.org> 2011-04-15 14:02:50 UTC ---
Author: rguenth
Date: Fri Apr 15 14:02:41 2011
New Revision: 172494

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=172494
Log:
2011-04-15  Richard Guenther  <rguenther@suse.de>

    PR tree-optimization/48290
    * tree-ssa-copy.c (copy_prop_visit_phi_node): Propagate constants.
    Properly decide inhibiting propagation based on the valueized
    operand.  Do loop-closed SSA form preserving here ...
    (init_copy_prop): ... not here.

Modified:
    trunk/gcc/ChangeLog
    trunk/gcc/tree-ssa-copy.c