[Bug middle-end/50902] New: intVar/dinternal.cc ICEs at -O2 -ftree-vectorize

[Bug middle-end/50902] New: intVar/dinternal.cc ICEs at -O2 -ftree-vectorize

[howarth at nitro dot med.uc.edu]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=50902

             Bug #: 50902
           Summary: intVar/dinternal.cc ICEs at -O2 -ftree-vectorize
    Classification: Unclassified
           Product: gcc
           Version: 4.7.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: middle-end
        AssignedTo: unassigned@gcc.gnu.org
        ReportedBy: howarth@nitro.med.uc.edu


Created attachment 25644
  --> http://gcc.gnu.org/bugzilla/attachment.cgi?id=25644
preprocessed source file for intVar/dinternal.cc

The g++ compiler in current gcc trunk ICEs when compiling intVar/dinternal.cc
from xplor-nih 2.27 with -O2 -ftree-vectorize or -O3.


g++-fsf-4.7 -c dinternal.cc -O2 -ftree-vectorize -fpermissive -DX_MMAP_FLAGS=0
-DFORTRAN_INIT -fno-common -DDARWIN -D_REENTRANT -DNDEBUG
-I/Users/howarth/xplor-nih-2.27/intVar/
-I/Users/howarth/xplor-nih-2.27/arch/Darwin_11_x86_64/include -DCPLUSPLUS
-DUSE_CDS_NAMESPACE -I/Users/howarth/xplor-nih-2.27/intVar/
-I/Users/howarth/xplor-nih-2.27/arch/Darwin_11_x86_64/include
-I/Users/howarth/xplor-nih-2.27/CDSlib -I/Users/howarth/xplor-nih-2.27/common
In file included from dinternal.cc:1251:0:
/Users/howarth/xplor-nih-2.27/CDSlib/matrixTools.cc: In instantiation of
'MATRIX MatrixTools::callInverse(const MATRIX&,
MatrixTools::InverseResults<FullMatrix<typename MATRIX::ElementType> >) [with
MATRIX = InertiaTensor; typename MATRIX::ElementType = double]':
/Users/howarth/xplor-nih-2.27/CDSlib/matrixTools.cc:237:35:   required from
'MATRIX MatrixTools::inverse(const MATRIX&,
MatrixTools::InverseResults<typename MATRIX::MatrixType>) [with MATRIX =
InertiaTensor; typename MATRIX::MatrixType = FullMatrix<double>]'
dinternal.cc:945:29:   required from here
/Users/howarth/xplor-nih-2.27/CDSlib/matrixTools.cc:170:2: warning: 'callTRF'
was not declared in this scope, and no declarations were found by
argument-dependent lookup at the point of instantiation [-fpermissive]
/Users/howarth/xplor-nih-2.27/CDSlib/matrixTools.cc:293:1: note:
'template<class Number> void callTRF(const int&, const int&, Number*, const
int&, int*, int&)' declared here, later in the translation unit
/Users/howarth/xplor-nih-2.27/CDSlib/matrixTools.cc:180:2: warning: 'callTRI'
was not declared in this scope, and no declarations were found by
argument-dependent lookup at the point of instantiation [-fpermissive]
/Users/howarth/xplor-nih-2.27/CDSlib/matrixTools.cc:274:1: note:
'template<class Number> void callTRI(const int&, Number*, const int&, const
int*, Number*, const int&, int&)' declared here, later in the translation unit
In file included from dinternal.cc:1238:0:
/Users/howarth/xplor-nih-2.27/CDSlib/cdsVector.cc: In constructor
'CDSVectorBase<T, ALLOC>::CDSVectorBase(int, const T&, ALLOC) [with T = bool;
ALLOC = CDS::DefaultAlloc]':
/Users/howarth/xplor-nih-2.27/CDSlib/cdsVector.cc:25:1: internal compiler
error: in build_vector_from_val, at tree.c:1382

[Bug middle-end/50902] intVar/dinternal.cc ICEs at -O2 -ftree-vectorize

[howarth at nitro dot med.uc.edu]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=50902

--- Comment #1 from Jack Howarth <howarth at nitro dot med.uc.edu> 2011-10-28 16:01:35 UTC ---
Uploaded testcase which can reproduce the ICE with...

g++-fsf-4.7 -c dinternal.ii -O2 -ftree-vectorize -fpermissive -DX_MMAP_FLAGS=0
-DFORTRAN_INIT -fno-common -DDARWIN -D_REENTRANT -DNDEBUG  -DCPLUSPLUS
-DUSE_CDS_NAMESPACE

for gcc trunk built with...

Using built-in specs.
COLLECT_GCC=g++-fsf-4.7
COLLECT_LTO_WRAPPER=/sw/lib/gcc4.7/libexec/gcc/x86_64-apple-darwin11.2.0/4.7.0/lto-wrapper
Target: x86_64-apple-darwin11.2.0
Configured with: ../gcc-4.7-20111028/configure --prefix=/sw
--prefix=/sw/lib/gcc4.7 --mandir=/sw/share/man --infodir=/sw/lib/gcc4.7/info
--with-build-config=bootstrap-lto --enable-stage1-languages=c,lto
--enable-languages=c,c++,fortran,lto,objc,obj-c++,java --with-gmp=/sw
--with-libiconv-prefix=/sw --with-ppl=/sw --with-cloog=/sw --with-mpc=/sw
--with-system-zlib --x-includes=/usr/X11R6/include --x-libraries=/usr/X11R6/lib
--program-suffix=-fsf-4.7 --enable-checking=yes --enable-cloog-backend=isl
Thread model: posix
gcc version 4.7.0 20111028 (experimental) (GCC)

[Bug middle-end/50902] intVar/dinternal.cc ICEs at -O2 -ftree-vectorize

[howarth at nitro dot med.uc.edu]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=50902

--- Comment #2 from Jack Howarth <howarth at nitro dot med.uc.edu> 2011-10-28 16:11:38 UTC ---
Created attachment 25645
  --> http://gcc.gnu.org/bugzilla/attachment.cgi?id=25645
gdb log for single stepping from last call to tree.c:1382 before crash

[Bug middle-end/50902] intVar/dinternal.cc ICEs at -O2 -ftree-vectorize

[howarth at nitro dot med.uc.edu]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=50902

--- Comment #3 from Jack Howarth <howarth at nitro dot med.uc.edu> 2011-10-28 16:35:21 UTC ---
Could this be an unresolved corner-case of PR48098?

[Bug middle-end/50902] intVar/dinternal.cc ICEs at -O2 -ftree-vectorize

[howarth at nitro dot med.uc.edu]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=50902

--- Comment #4 from Jack Howarth <howarth at nitro dot med.uc.edu> 2011-10-28 17:34:30 UTC ---
Created attachment 25647
  --> http://gcc.gnu.org/bugzilla/attachment.cgi?id=25647
preprocessed source for marvin/MarvinNOEPotential.cc

[Bug middle-end/50902] intVar/dinternal.cc ICEs at -O2 -ftree-vectorize

[howarth at nitro dot med.uc.edu]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=50902

--- Comment #5 from Jack Howarth <howarth at nitro dot med.uc.edu> 2011-10-28 17:35:50 UTC ---
The attached MarvinNOEPotential.ii preprocessed source is another instance of
this bug in the xplor-nih 2.27 build...

g++-fsf-4.7 -c MarvinNOEPotential.ii -O2 -ftree-vectorize -fpermissive
-DX_MMAP_FLAGS=0 -DFORTRAN_INIT -fno-common -DDARWIN -D_REENTRANT -DNDEBUG
-DCPLUSPLUS -DUSE_CDS_NAMESPACE -LANG:std
In file included from MarvinNOEPotential.cc:2787:0:
/Users/howarth/xplor-nih-2.27/CDSlib/cdsMatrix.cc: In constructor
‘CDSMatrixBase<T>::CDSMatrixBase(int, int, const T&) [with T = bool]’:
/Users/howarth/xplor-nih-2.27/CDSlib/cdsMatrix.cc:24:1: internal compiler
error: in build_vector_from_val, at tree.c:1382
Please submit a full bug report,
with preprocessed source if appropriate.
See <http://gcc.gnu.org/bugs.html> for instructions.

[Bug middle-end/50902] intVar/dinternal.cc ICEs at -O2 -ftree-vectorize

[howarth at nitro dot med.uc.edu]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=50902

--- Comment #6 from Jack Howarth <howarth at nitro dot med.uc.edu> 2011-10-29 17:39:43 UTC ---
Confirmed that both the dinternal.ii and MarvinNOEPotential.ii test cases
reproduce this ICE using current gcc trunk under x86_64 Fedora 15.

[Bug tree-optimization/50902] [4.7 Regression] intVar/dinternal.cc ICEs at -O2 -ftree-vectorize

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=50902

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
          Component|middle-end                  |tree-optimization
   Target Milestone|---                         |4.7.0
            Summary|intVar/dinternal.cc ICEs at |[4.7 Regression]
                   |-O2 -ftree-vectorize        |intVar/dinternal.cc ICEs at
                   |                            |-O2 -ftree-vectorize

[Bug tree-optimization/50902] [4.7 Regression] intVar/dinternal.cc ICEs at -O2 -ftree-vectorize

[irar at il dot ibm.com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=50902

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
   Last reconfirmed|                            |2011-11-01
                 CC|                            |irar at il dot ibm.com
     Ever Confirmed|0                           |1

--- Comment #7 from Ira Rosen <irar at il dot ibm.com> 2011-11-01 08:25:08 UTC ---
Reduced testcase:

_Bool data[128];
void foo (_Bool *init)
{
 int i;
 for (i = 0; i < 128; i++)
   data[i] = *init;
}


gcc_checking_assert (types_compatible_p (TYPE_MAIN_VARIANT (TREE_TYPE (sc)),
                                           TREE_TYPE (vectype)));

fails since VECTYPE is vector(16) <unnamed-unsigned:8>
and the scalar type is _Bool

[Bug tree-optimization/50902] [4.7 Regression] intVar/dinternal.cc ICEs at -O2 -ftree-vectorize

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=50902

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> 2011-11-01 13:06:57 UTC ---
Mine.

Index: gcc/tree-vect-stmts.c
===================================================================
--- gcc/tree-vect-stmts.c    (revision 180737)
+++ gcc/tree-vect-stmts.c    (working copy)
@@ -4726,11 +4726,20 @@
           /* 4. Handle invariant-load.  */
           if (inv_p && !bb_vinfo)
         {
-          tree vec_inv;
+          tree tem, vec_inv;
           gimple_stmt_iterator gsi2 = *gsi;
           gcc_assert (!strided_load);
           gsi_next (&gsi2);
-          vec_inv = build_vector_from_val (vectype, scalar_dest);
+          tem = scalar_dest;
+          if (!useless_type_conversion_p (TREE_TYPE (vectype),
+                          TREE_TYPE (tem)))
+            {
+              tem = fold_convert (TREE_TYPE (vectype), tem);
+              tem = force_gimple_operand_gsi (&gsi2, tem, true,
+                              NULL_TREE, true,
+                              GSI_SAME_STMT);
+            }
+          vec_inv = build_vector_from_val (vectype, tem);
           new_temp = vect_init_vector (stmt, vec_inv,
                            vectype, &gsi2);
           new_stmt = SSA_NAME_DEF_STMT (new_temp);

[Bug tree-optimization/50902] [4.7 Regression] intVar/dinternal.cc ICEs at -O2 -ftree-vectorize

[howarth at nitro dot med.uc.edu]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=50902

--- Comment #9 from Jack Howarth <howarth at nitro dot med.uc.edu> 2011-11-01 16:53:09 UTC ---
(In reply to comment #8)

I can confirm that the proposed patch resolves the build failures in xplor-nih
without regressions in the xplor-nih testsuite.

[Bug tree-optimization/50902] [4.7 Regression] intVar/dinternal.cc ICEs at -O2 -ftree-vectorize

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=50902

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

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

--- Comment #11 from Richard Guenther <rguenth at gcc dot gnu.org> 2011-11-02 10:20:51 UTC ---
Fixed.

[Bug tree-optimization/50902] [4.7 Regression] intVar/dinternal.cc ICEs at -O2 -ftree-vectorize

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=50902

--- Comment #10 from Richard Guenther <rguenth at gcc dot gnu.org> 2011-11-02 10:20:36 UTC ---
Author: rguenth
Date: Wed Nov  2 10:20:32 2011
New Revision: 180765

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=180765
Log:
2011-11-02  Richard Guenther  <rguenther@suse.de>

    PR tree-optimization/50902
    * tree-vect-stmts.c (vectorizable_load): Properly convert
    an invariant initializer element.

    * gcc.dg/torture/pr50902.c: New testcase.

Added:
    trunk/gcc/testsuite/gcc.dg/torture/pr50902.c
Modified:
    trunk/gcc/ChangeLog
    trunk/gcc/testsuite/ChangeLog
    trunk/gcc/tree-vect-stmts.c