[Bug middle-end/49628] New: [4.7 Regression] 447.dealII in SPEC CPU 2006 failed to build

[Bug middle-end/49628] New: [4.7 Regression] 447.dealII in SPEC CPU 2006 failed to build

[hjl.tools at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=49628

           Summary: [4.7 Regression] 447.dealII in SPEC CPU 2006 failed to
                    build
           Product: gcc
           Version: 4.7.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: middle-end
        AssignedTo: unassigned@gcc.gnu.org
        ReportedBy: hjl.tools@gmail.com
                CC: rguenther@suse.de


On Linux/x86, revision 175704:

http://gcc.gnu.org/ml/gcc-cvs/2011-06/msg01196.html

gave:

g++  -O3 -funroll-loops -ffast-math -fwhole-program -flto=jobserver
-fuse-linker-plugin  -DSPEC_CPU_LP64        auto_derivative_function.o
block_sparse_matrix.o block_sparse_matrix_ez.o block_sparsity_pattern.o
block_vector.o compressed_sparsity_pattern.o data_out.o data_out_base.o
data_out_faces.o data_out_rotation.o data_out_stack.o
derivative_approximation.o dof_accessor.o dof_constraints.o dof_handler.o
dof_levels.o dof_renumbering.o dof_tools.o error_estimator.o exceptions.o fe.o
fe_data.o fe_dgp.o fe_dgp_1d.o fe_dgp_2d.o fe_dgp_3d.o fe_dgp_nonparametric.o
fe_dgq.o fe_dgq_1d.o fe_dgq_2d.o fe_dgq_3d.o fe_nedelec.o fe_nedelec_1d.o
fe_nedelec_2d.o fe_nedelec_3d.o fe_q.o fe_q_1d.o fe_q_2d.o fe_q_3d.o
fe_q_hierarchical.o fe_raviart_thomas.o fe_system.o fe_tools.o fe_values.o
filtered_matrix.o full_matrix.double.o full_matrix.float.o function.o
function_derivative.o function_lib.o function_lib_cutoff.o function_time.o
geometry_info.o grid_generator.o grid_in.o grid_out.all_dimensions.o grid_out.o
grid_refinement.o grid_reordering.o histogram.o intergrid_map.o
job_identifier.o log.o mapping.o mapping_c1.o mapping_cartesian.o mapping_q.o
mapping_q1.o mapping_q1_eulerian.o matrices.all_dimensions.o matrices.o
matrix_lib.o matrix_out.o memory_consumption.o mg_base.o mg_dof_accessor.o
mg_dof_handler.o mg_dof_tools.o mg_smoother.o mg_transfer_block.o
mg_transfer_prebuilt.o mg_transfer_block.all_dimensions.o
multigrid.all_dimensions.o multithread_info.o parameter_handler.o
persistent_tria.o polynomial.o polynomial_space.o programid.o quadrature.o
quadrature_lib.o solution_transfer.o solver_control.o sparse_matrix.double.o
sparse_matrix.float.o sparse_matrix_ez.double.o sparse_matrix_ez.float.o
sparsity_pattern.o step-14.o subscriptor.o swappable_vector.o tensor.o
tensor_product_polynomials.o tria.all_dimensions.o tria.o tria_accessor.o
tria_boundary.o tria_boundary_lib.o vector.o vector.long_double.o
vectors.all_dimensions.o fe_dgp_monomial.o fe_poly.o polynomials_bdm.o
polynomials_p.o vectors.o                     -o dealII
mapping_q1.cc: In member function 'compute_fill':
mapping_q1.cc:442:0: internal compiler error: Floating point exception
Please submit a full bug report,
with preprocessed source if appropriate.
See <http://gcc.gnu.org/bugs.html> for instructions.
make[4]: *** [/tmp/ccLsl9Do.ltrans4.ltrans.o] Error 1
lto-wrapper: make returned 2 exit status
/usr/local/bin/ld: lto-wrapper failed
collect2: error: ld returned 1 exit status

[Bug middle-end/49628] [4.7 Regression] 447.dealII in SPEC CPU 2006 failed to build

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=49628

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |ASSIGNED
   Last reconfirmed|                            |2011.07.04 13:38:59
         AssignedTo|unassigned at gcc dot       |rguenth at gcc dot gnu.org
                   |gnu.org                     |
   Target Milestone|---                         |4.7.0
     Ever Confirmed|0                           |1

--- Comment #1 from Richard Guenther <rguenth at gcc dot gnu.org> 2011-07-04 13:38:59 UTC ---
Mine.  Reducing from mapping_q1.cc which triggers this at -O3 -fipa-pta w/o
LTO.

[Bug middle-end/49628] [4.7 Regression] 447.dealII in SPEC CPU 2006 failed to build

[hjl.tools at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=49628

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|ASSIGNED                    |UNCONFIRMED
   Target Milestone|4.7.0                       |---
     Ever Confirmed|1                           |0

--- Comment #2 from H.J. Lu <hjl.tools at gmail dot com> 2011-07-04 14:01:54 UTC ---
Program received signal SIGFPE, Arithmetic exception.
0x000000000110149c in vector_alignment_reachable_p (dr=0x1960370)
    at ../../src/gcc/tree-vect-data-refs.c:1114
1114          if ((nelements - mis_in_elements) % GROUP_SIZE (stmt_info))
Missing separate debuginfos, use: debuginfo-install glibc-2.14-3.4.f15.x86_64
gmp-4.3.2-3.fc15.x86_64 libgcc-4.6.0-9.fc15.x86_64
libmpc-0.8.3-0.3.svn855.fc15.x86_64 libstdc++-4.6.0-9.fc15.x86_64
mpfr-3.0.0-4.fc15.x86_64 zlib-1.2.5-3.fc15.x86_64
(gdb) bt
#0  0x000000000110149c in vector_alignment_reachable_p (dr=0x1960370)
    at ../../src/gcc/tree-vect-data-refs.c:1114
#1  0x0000000001101e7b in vect_enhance_data_refs_alignment (
    loop_vinfo=0x1a3e750) at ../../src/gcc/tree-vect-data-refs.c:1505
#2  0x0000000000c373e7 in vect_analyze_loop_2 (loop_vinfo=0x1a3e750)
    at ../../src/gcc/tree-vect-loop.c:1554
#3  0x0000000000c375de in vect_analyze_loop (loop=0x7fffefe66550)
    at ../../src/gcc/tree-vect-loop.c:1627
#4  0x0000000000c53fae in vectorize_loops ()
    at ../../src/gcc/tree-vectorizer.c:199
#5  0x0000000000b7dc13 in tree_vectorize ()
    at ../../src/gcc/tree-ssa-loop.c:218
#6  0x00000000008f9f39 in execute_one_pass (pass=0x1737320)
    at ../../src/gcc/passes.c:2061
#7  0x00000000008fa11f in execute_pass_list (pass=0x1737320)
    at ../../src/gcc/passes.c:2116
#8  0x00000000008fa140 in execute_pass_list (pass=0x1737140)
    at ../../src/gcc/passes.c:2117
#9  0x00000000008fa140 in execute_pass_list (pass=0x1736840)
    at ../../src/gcc/passes.c:2117
#10 0x0000000000a92c3c in tree_rest_of_compilation (fndecl=0x7ffff0416100)
    at ../../src/gcc/tree-optimize.c:416
#11 0x000000000057a047 in cgraph_expand_function (node=0x7ffff030b360)
---Type <return> to continue, or q <return> to quit---
    at ../../src/gcc/cgraphunit.c:1792
#12 0x000000000057a206 in cgraph_expand_all_functions ()
    at ../../src/gcc/cgraphunit.c:1851
#13 0x000000000057a902 in cgraph_optimize () at ../../src/gcc/cgraphunit.c:2121
#14 0x00000000004af417 in lto_main () at ../../src/gcc/lto/lto.c:2800
#15 0x00000000009f006e in compile_file () at ../../src/gcc/toplev.c:548
#16 0x00000000009f22b3 in do_compile () at ../../src/gcc/toplev.c:1895
#17 0x00000000009f2419 in toplev_main (argc=20, argv=0x7fffffffe1d8)
    at ../../src/gcc/toplev.c:1967
#18 0x00000000004b21f0 in main (argc=20, argv=0x7fffffffe1d8)
    at ../../src/gcc/main.c:36
(gdb) p stmt_info->size
$3 = 0
(gdb)

[Bug middle-end/49628] [4.7 Regression] 447.dealII in SPEC CPU 2006 failed to build

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=49628

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|rguenther at suse dot de    |
   Target Milestone|---                         |4.7.0

[Bug middle-end/49628] [4.7 Regression] 447.dealII in SPEC CPU 2006 failed to build

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=49628

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |ASSIGNED
   Last reconfirmed|2011-07-04 13:38:59         |2011.07.04 14:33:23
     Ever Confirmed|0                           |1

--- Comment #3 from Richard Guenther <rguenth at gcc dot gnu.org> 2011-07-04 14:33:23 UTC ---
Testcase:

#include <vector>
template <int rank, int dim> class Tensor;
template <int dim> class Tensor<1,dim> {
public:
    explicit Tensor (const bool initialize = true);
    Tensor (const Tensor<1,dim> &);
    Tensor<1,dim> & operator = (const Tensor<1,dim> &);
    double values[(dim!=0) ? (dim) : 1];
};
template <int dim>
inline Tensor<1,dim> & Tensor<1,dim>::operator = (const Tensor<1,dim> &p)
{
    for (unsigned int i=0; i<dim; ++i)
      values[i] = p.values[i];
};
template <int dim> class Quadrature {
public:
    const unsigned int n_quadrature_points;
};
class MappingQ1
{
    class InternalData  {
    public:
        std::vector<Tensor<1,3> > shape_derivatives;
        unsigned int n_shape_functions;
    };
    void compute_data (const Quadrature<3> &quadrature, InternalData &data)
const;
};
void MappingQ1::compute_data (const Quadrature<3> &q, InternalData &data) const
{
    const unsigned int n_q_points = q.n_quadrature_points;
    data.shape_derivatives.resize(data.n_shape_functions * n_q_points);
}

[Bug middle-end/49628] [4.7 Regression] 447.dealII in SPEC CPU 2006 failed to build

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=49628

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |irar at gcc dot gnu.org

--- Comment #4 from Richard Guenther <rguenth at gcc dot gnu.org> 2011-07-04 15:06:51 UTC ---
Vectorizing the loop

<bb 8>:
  # __position_11 = PHI <__position_34(9), __position$_M_current_15(7)>
  D.2964_46 = __x_copy.values[0];
  __position_11->values[0] = D.2964_46;
  D.2964_51 = __x_copy.values[1];
  __position_11->values[1] = D.2964_51;
  D.2964_56 = __x_copy.values[2];
  __position_11->values[2] = D.2964_56;
  __position_34 = __position_11 + 24;
  if (D.2961_29 != __position_34)
    goto <bb 9>;
  else
    goto <bb 10>;

<bb 9>:
  goto <bb 8>;

I wonder where we are supposed to set GROUP_SIZE here and how it possibly
can connect to my change ... ah, the invariant loads from __x_copy.values
no longer will inhibit vectorization early.  The group is now indeed
the __x_copy.values[] loads.

The following fixes it (or rather, avoids looking at this group):

Index: gcc/tree-vect-data-refs.c
===================================================================
--- gcc/tree-vect-data-refs.c   (revision 175802)
+++ gcc/tree-vect-data-refs.c   (working copy)
@@ -1495,12 +1495,19 @@ vect_enhance_data_refs_alignment (loop_v
           && GROUP_FIRST_ELEMENT (stmt_info) != stmt)
         continue;

+      /* For invariant accesses there is nothing to enhance.  */
+      if (integer_zerop (DR_STEP (dr)))
+       continue;
+
       supportable_dr_alignment = vect_supportable_dr_alignment (dr, true);
       do_peeling = vector_alignment_reachable_p (dr);
       if (do_peeling)

[Bug middle-end/49628] [4.7 Regression] 447.dealII in SPEC CPU 2006 failed to build

[irar at il dot ibm.com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=49628

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

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

--- Comment #5 from Ira Rosen <irar at il dot ibm.com> 2011-07-05 06:43:18 UTC ---
(In reply to comment #4)

> I wonder where we are supposed to set GROUP_SIZE here 
In vect_analyze_data_ref_access(),
but your patch does this:

  /* Allow invariant loads in loops.  */
  if (loop_vinfo && dr_step == 0)
    return DR_IS_READ (dr);

preventing detection of interleaving groups for invariant loads. Which is ok,
since you don't support them anyway.

> and how it possibly
> can connect to my change ... ah, the invariant loads from __x_copy.values
> no longer will inhibit vectorization early.  The group is now indeed
> the __x_copy.values[] loads.
> 
> The following fixes it (or rather, avoids looking at this group):

I think this change is correct, there is indeed nothing to enhance.

But this case revealed a problem in STMT_VINFO_STRIDED_ACCESS - it only checks
that the interleaving chain exists, but even if it exists, the access analysis
can later fail, so we should also check the group size:

Index: tree-vectorizer.h
===================================================================
--- tree-vectorizer.h   (revision 175681)
+++ tree-vectorizer.h   (working copy)
@@ -545,7 +547,7 @@ typedef struct _stmt_vec_info {
 #define STMT_VINFO_GROUP_GAP(S)            (S)->gap
 #define STMT_VINFO_GROUP_SAME_DR_STMT(S)   (S)->same_dr_stmt
 #define STMT_VINFO_GROUP_READ_WRITE_DEPENDENCE(S)  (S)->read_write_dep
-#define STMT_VINFO_STRIDED_ACCESS(S)      ((S)->first_element != NULL &&
(S)->data_ref_info)
+#define STMT_VINFO_STRIDED_ACCESS(S)      ((S)->first_element != NULL &&
(S)->data_ref_info && (S)->size > 0)

 #define GROUP_FIRST_ELEMENT(S)          (S)->first_element
 #define GROUP_NEXT_ELEMENT(S)           (S)->next_element

Thanks,
Ira

> 
> Index: gcc/tree-vect-data-refs.c
> ===================================================================
> --- gcc/tree-vect-data-refs.c   (revision 175802)
> +++ gcc/tree-vect-data-refs.c   (working copy)
> @@ -1495,12 +1495,19 @@ vect_enhance_data_refs_alignment (loop_v
>            && GROUP_FIRST_ELEMENT (stmt_info) != stmt)
>          continue;
> 
> +      /* For invariant accesses there is nothing to enhance.  */
> +      if (integer_zerop (DR_STEP (dr)))
> +       continue;
> +
>        supportable_dr_alignment = vect_supportable_dr_alignment (dr, true);
>        do_peeling = vector_alignment_reachable_p (dr);
>        if (do_peeling)

[Bug middle-end/49628] [4.7 Regression] 447.dealII in SPEC CPU 2006 failed to build

[irar at il dot ibm.com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=49628

--- Comment #6 from Ira Rosen <irar at il dot ibm.com> 2011-07-05 08:18:09 UTC ---
Sorry, I was wrong about group size check. It is ok without your patch. So,
this should be enough:

Index: tree-vect-data-refs.c
===================================================================
--- tree-vect-data-refs.c       (revision 175785)
+++ tree-vect-data-refs.c       (working copy)
@@ -2304,7 +2304,10 @@ vect_analyze_data_ref_access (struct dat

   /* Allow invariant loads in loops.  */
   if (loop_vinfo && dr_step == 0)
-    return DR_IS_READ (dr);
+    {
+      GROUP_FIRST_ELEMENT (vinfo_for_stmt (stmt)) = NULL;
+      return DR_IS_READ (dr);
+    }

   if (loop && nested_in_vect_loop_p (loop, stmt))
     {

Thanks,
Ira

[Bug middle-end/49628] [4.7 Regression] 447.dealII in SPEC CPU 2006 failed to build

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=49628

--- Comment #7 from Richard Guenther <rguenth at gcc dot gnu.org> 2011-07-05 08:53:46 UTC ---
(In reply to comment #5)
> (In reply to comment #4)
> 
> > I wonder where we are supposed to set GROUP_SIZE here 
> In vect_analyze_data_ref_access(),
> but your patch does this:
> 
>   /* Allow invariant loads in loops.  */
>   if (loop_vinfo && dr_step == 0)
>     return DR_IS_READ (dr);
> 
> preventing detection of interleaving groups for invariant loads. Which is ok,
> since you don't support them anyway.

Yeah, I noticed we don't vectorize loops similar to

struct X { float values[4]; };
void foo (struct X * dst, struct X * src, int n)
{
  int i;
  for (i = 0; i < n; ++i)
    {
      dst->values[i*4] = src->values[0];
      dst->values[i*4+1] = src->values[1];
      dst->values[i*4+2] = src->values[2];
      dst->values[i*4+3] = src->values[3];
    }
}

with different reasons based on the signedness of i (ugh).  We do vectorize
the above with SLP when LIM/PRE move the loads from src out of the loop
though.

[Bug middle-end/49628] [4.7 Regression] 447.dealII in SPEC CPU 2006 failed to build

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=49628

--- Comment #8 from Richard Guenther <rguenth at gcc dot gnu.org> 2011-07-05 11:24:30 UTC ---
Author: rguenth
Date: Tue Jul  5 11:24:26 2011
New Revision: 175847

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=175847
Log:
2011-07-05  Richard Guenther  <rguenther@suse.de>

    PR tree-optimization/49518
    PR tree-optimization/49628
    * tree-vect-data-refs.c (vect_enhance_data_refs_alignment): Skip
    irrelevant and invariant data-references.
    (vect_analyze_data_ref_access): For invariant loads clear the
    group association.

    * g++.dg/torture/pr49628.C: New testcase.
    * gcc.dg/torture/pr49518.c: Likewise.

Added:
    trunk/gcc/testsuite/g++.dg/torture/pr49628.C
    trunk/gcc/testsuite/gcc.dg/torture/pr49518.c
Modified:
    trunk/gcc/ChangeLog
    trunk/gcc/testsuite/ChangeLog
    trunk/gcc/tree-vect-data-refs.c

[Bug middle-end/49628] [4.7 Regression] 447.dealII in SPEC CPU 2006 failed to build

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=49628

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

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

--- Comment #9 from Richard Guenther <rguenth at gcc dot gnu.org> 2011-07-05 11:25:45 UTC ---
Fixed.

[Bug middle-end/49628] [4.7 Regression] 447.dealII in SPEC CPU 2006 failed to build

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=49628

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |dcb314 at hotmail dot com

--- Comment #10 from Richard Guenther <rguenth at gcc dot gnu.org> 2011-07-06 08:11:05 UTC ---
*** Bug 49652 has been marked as a duplicate of this bug. ***

[Bug middle-end/49628] [4.7 Regression] 447.dealII in SPEC CPU 2006 failed to build

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=49628

--- Comment #11 from Richard Guenther <rguenth at gcc dot gnu.org> 2011-09-08 10:31:43 UTC ---
Author: rguenth
Date: Thu Sep  8 10:31:39 2011
New Revision: 178685

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=178685
Log:
2011-09-08  Richard Guenther  <rguenther@suse.de>

    Backport from mainline
    2011-07-04  Richard Guenther  <rguenther@suse.de>

    PR tree-optimization/49518
    PR tree-optimization/49628
    * tree-vect-data-refs.c (vect_enhance_data_refs_alignment): Skip
    irrelevant data-references.

    * g++.dg/torture/pr49628.C: New testcase.
    * gcc.dg/torture/pr49518.c: Likewise.

Added:
    branches/gcc-4_6-branch/gcc/testsuite/g++.dg/torture/pr49628.C
    branches/gcc-4_6-branch/gcc/testsuite/gcc.dg/torture/pr49518.c
Modified:
    branches/gcc-4_6-branch/gcc/ChangeLog
    branches/gcc-4_6-branch/gcc/testsuite/ChangeLog
    branches/gcc-4_6-branch/gcc/tree-vect-data-refs.c