[Bug c/54894] New: internal compiler error: in vect_get_vec_def_for_operand, at tree-vect-stmts.c:1286

[Bug c/54894] New: internal compiler error: in vect_get_vec_def_for_operand, at tree-vect-stmts.c:1286

[kuszmaul at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54894

             Bug #: 54894
           Summary: internal compiler error: in
                    vect_get_vec_def_for_operand, at
                    tree-vect-stmts.c:1286
    Classification: Unclassified
           Product: gcc
           Version: 4.7.2
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: c
        AssignedTo: unassigned@gcc.gnu.org
        ReportedBy: kuszmaul@gmail.com


Created attachment 28415
  --> http://gcc.gnu.org/bugzilla/attachment.cgi?id=28415
preprocessed code for matrix multiply

I'm having trouble getting gcc to remember alignment information the attached
matrix multiplication code.  So I introduced a temporary variable that is a
pointer to a 16-byte aligned double, and ended up with this error:


$ gcc -O3 -std=c99 mm_tile.c -Wcast-align -o mm_tile
mm_tile.c: In function ‘main’:
mm_tile.c:46:5: internal compiler error: in vect_get_vec_def_for_operand, at
tree-vect-stmts.c:1286
Please submit a full bug report,
with preprocessed source if appropriate.
See <http://bugzilla.redhat.com/bugzilla> for instructions.
Preprocessed source stored into /tmp/ccvYpDPi.out file, please attach this to
your bugreport.
$ gcc --version
gcc (GCC) 4.7.2 20120921 (Red Hat 4.7.2-2)
Copyright (C) 2012 Free Software Foundation, Inc.
This is free software; see the source for copying conditions.  There is NO
warranty; not even for MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
$ cat mm_tile.c
/** BEGIN HIDDEN **/
#include <stdlib.h>
#include <stdio.h>
#include <sys/time.h>

typedef unsigned long long uint64_t;

/* Allow us to change n on the compiler command line with for example -Dn=1024
*/
#ifndef n
#define n 4096
#endif
double A[n][n] __attribute__((aligned(16)));
double B[n][n] __attribute__((aligned(16)));
double C[n][n] __attribute__((aligned(16)));

static float tdiff (struct timeval *start,
            struct timeval *end) {
  return (end->tv_sec-start->tv_sec)
    +1e-6*(end->tv_usec-start->tv_usec);
}

#include "verify.h"

#define tilesize 128

/* void mmbase (double *restrict C, double *restrict A, double *restrict B) {
*/
/*   for (int kh = 0; kh < n; kh += tilesize) { */
/*     for (int il = 0; il < tilesize; il++) { */
/*       for (int kl = 0; kl < tilesize; ++kl) { */
/*         for (int jl = 0; jl < tilesize; jl++) { */
/*           C[il*n+jl] += A[il*n+kh+kl]*B[(kh+kl)*n+jl]; */
/*         } */
/*       } */
/*     } */
/*   } */
/* } */

/* void mm_js(double *restrict C, double *restrict A, double *restrict B) { */
/*   for (int jh = 0; jh < n; jh += tilesize) { */
/*     mmbase(&C[jh], &A[0], &B[jh]); */
/*   } */
/* } */

typedef double adouble __attribute__((__aligned__(16)));

int main(int argc, const char *argv[]) {
  parse_args(argc, argv);
  for (int i = 0; i < n; ++i) {
    for (int j = 0; j < n; ++j) {
      A[i][j] = (double)rand() / (double)RAND_MAX;
      B[i][j] = (double)rand() / (double)RAND_MAX;
      C[i][j] = 0;
    } }

  struct timeval start, end;
  gettimeofday(&start, NULL);
  /** END HIDDEN **/
  for (int ih = 0; ih < n; ih += tilesize) {                  
///\lilabel{block_loop_i}
    for (int jh = 0; jh < n; jh += tilesize) {                
///\lilabel{block_loop_k}
      for (int kh = 0; kh < n; kh += tilesize) {                     
///\lilabel{block_loop_j}
        for (int il = 0; il < tilesize; ++il) {                      
///\lilabel{base_loop_i}
          adouble *Ap = (adouble *)&A[ih+il][kh];
          for (int kl = 0; kl < tilesize; ++kl) {              
///\lilabel{base_loop_k}
            for (int jl = 0; jl < tilesize; ++jl) {                    
///\lilabel{base_loop_j}
              C[ih+il][jh+jl] += Ap[kl] * B[kh+kl][jh+jl];
///\lilabel{base_multiply}
        } } }
      }
      /* mmbase(&C[ih][jh],&A[ih][0],&B[0][jh]); */
    }
    /* mm_js(&C[ih][0], &A[ih][0], &B[0][0]); */
  }
  /** BEGIN HIDDEN **/
  gettimeofday(&end, NULL);
  printf("%0.6f\n", tdiff(&start, &end));
  if (need_to_verify) verify();
  return 0;
}
/** END HIDDEN **/

[Bug tree-optimization/54894] [4.6/4.7/4.8 Regression] internal compiler error: in vect_get_vec_def_for_operand, at tree-vect-stmts.c:1286

[jakub at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54894

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
   Last reconfirmed|                            |2012-10-10
                 CC|                            |jakub at gcc dot gnu.org
   Target Milestone|---                         |4.6.4
            Summary|internal compiler error: in |[4.6/4.7/4.8 Regression]
                   |vect_get_vec_def_for_operan |internal compiler error: in
                   |d, at                       |vect_get_vec_def_for_operan
                   |tree-vect-stmts.c:1286      |d, at
                   |                            |tree-vect-stmts.c:1286
     Ever Confirmed|0                           |1

--- Comment #1 from Jakub Jelinek <jakub at gcc dot gnu.org> 2012-10-10 18:47:32 UTC ---
Started with http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=157088

[Bug tree-optimization/54894] [4.6/4.7/4.8 Regression] internal compiler error: in vect_get_vec_def_for_operand, at tree-vect-stmts.c:1286

[jakub at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54894

--- Comment #2 from Jakub Jelinek <jakub at gcc dot gnu.org> 2012-10-11 12:22:47 UTC ---
You should be using __builtin_assume_aligned builtin, i.e.
  double *Ap = __builtin_assume_aligned (&A[ih+il][kh], 16);
instead of the hacks with the overaligned scalar pointer, that doesn't improve
vectorization, but actually prohibits it.  That said, of course gcc shouldn't
ICE on this.

[Bug tree-optimization/54894] [4.6/4.7/4.8 Regression] internal compiler error: in vect_get_vec_def_for_operand, at tree-vect-stmts.c:1286

[jakub at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54894

--- Comment #3 from Jakub Jelinek <jakub at gcc dot gnu.org> 2012-10-11 12:33:41 UTC ---
I'd say the problem is that useless_type_conversion_p considers the overaligned
double type compatible to double, yet get_vectype_for_scalar_type returns
non-NULL for the normally aligned one and NULL for the overaligned one.
During vectorizer analysis phase, we assume it is enough to call that function
just for a single type for the stmt, not check all 3 types, that succeeds, but
during transform phase we call it on all 3 and ICE.

[Bug tree-optimization/54894] [4.6/4.7/4.8 Regression] internal compiler error: in vect_get_vec_def_for_operand, at tree-vect-stmts.c:1286

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54894

Richard Biener <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 #4 from Richard Biener <rguenth at gcc dot gnu.org> 2012-10-11 14:15:52 UTC ---
(In reply to comment #3)
> I'd say the problem is that useless_type_conversion_p considers the overaligned
> double type compatible to double,

That's ok - alignment does not affect the values.  useless_type_conversion_p
is about value-preserving conversions.

> yet get_vectype_for_scalar_type returns
> non-NULL for the normally aligned one and NULL for the overaligned one.

I think that was a bugfix, maybe not the best way.  get_vectype_for_scalar_type
should probably simply drop the alignment and use natural alignment for
the vector type (thus, based on the mode).  The idea of the fix was
to not need to care about extra padding, but data-ref analysis should
be the proper place to verify possible issues with that.

> During vectorizer analysis phase, we assume it is enough to call that function
> just for a single type for the stmt, not check all 3 types, that succeeds, but
> during transform phase we call it on all 3 and ICE.

Thus:

Index: gcc/tree-vect-stmts.c
===================================================================
--- gcc/tree-vect-stmts.c       (revision 192359)
+++ gcc/tree-vect-stmts.c       (working copy)
@@ -6060,11 +6060,6 @@ get_vectype_for_scalar_type_and_size (tr
       && GET_MODE_CLASS (inner_mode) != MODE_FLOAT)
     return NULL_TREE;

-  /* We can't build a vector type of elements with alignment bigger than
-     their size.  */
-  if (nbytes < TYPE_ALIGN_UNIT (scalar_type))
-    return NULL_TREE;
-
   /* For vector types of elements whose mode precision doesn't
      match their types precision we use a element type of mode
      precision.  The vectorization routines will have to make sure
@@ -6086,6 +6081,11 @@ get_vectype_for_scalar_type_and_size (tr
       && !POINTER_TYPE_P (scalar_type))
     scalar_type = lang_hooks.types.type_for_mode (inner_mode, 1);

+  /* We can't build a vector type of elements with alignment bigger than
+     their size.  */
+  if (nbytes < TYPE_ALIGN_UNIT (scalar_type))
+    scalar_type = lang_hooks.types.type_for_mode (inner_mode, 1);
+
   /* If no size was supplied use the mode the target prefers.   Otherwise
      lookup a vector mode of the specified size.  */
   if (size == 0)

[Bug tree-optimization/54894] [4.6/4.7/4.8 Regression] internal compiler error: in vect_get_vec_def_for_operand, at tree-vect-stmts.c:1286

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54894

--- Comment #5 from Richard Biener <rguenth at gcc dot gnu.org> 2012-10-12 08:00:47 UTC ---
Author: rguenth
Date: Fri Oct 12 08:00:29 2012
New Revision: 192390

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=192390
Log:
2012-10-12  Richard Biener  <rguenther@suse.de>

    PR tree-optimization/54894
    * tree-vect-stmts.c (get_vectype_for_scalar_type_and_size):
    Handle over-aligned scalar types properly.

    * gcc.dg/torture/pr54894.c: New testcase.

Added:
    trunk/gcc/testsuite/gcc.dg/torture/pr54894.c
Modified:
    trunk/gcc/ChangeLog
    trunk/gcc/testsuite/ChangeLog
    trunk/gcc/tree-vect-stmts.c

[Bug tree-optimization/54894] [4.6/4.7 Regression] internal compiler error: in vect_get_vec_def_for_operand, at tree-vect-stmts.c:1286

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54894

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
      Known to work|                            |4.8.0
            Summary|[4.6/4.7/4.8 Regression]    |[4.6/4.7 Regression]
                   |internal compiler error: in |internal compiler error: in
                   |vect_get_vec_def_for_operan |vect_get_vec_def_for_operan
                   |d, at                       |d, at
                   |tree-vect-stmts.c:1286      |tree-vect-stmts.c:1286

--- Comment #6 from Richard Biener <rguenth at gcc dot gnu.org> 2012-10-12 08:09:54 UTC ---
Fixed on trunk sofar.

[Bug tree-optimization/54894] [4.6/4.7 Regression] internal compiler error: in vect_get_vec_def_for_operand, at tree-vect-stmts.c:1286

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54894

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
         Depends on|                            |54976

--- Comment #7 from Richard Biener <rguenth at gcc dot gnu.org> 2012-10-19 10:37:46 UTC ---
Requires the fix for PR54976 on backport.

[Bug tree-optimization/54894] [4.6/4.7 Regression] internal compiler error: in vect_get_vec_def_for_operand, at tree-vect-stmts.c:1286

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54894

--- Comment #8 from Richard Biener <rguenth at gcc dot gnu.org> 2012-11-26 14:26:16 UTC ---
Author: rguenth
Date: Mon Nov 26 14:26:07 2012
New Revision: 193816

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=193816
Log:
2012-11-26  Richard Biener  <rguenther@suse.de>

    Backport from mainline
    2012-10-19  Richard Biener  <rguenther@suse.de>

    PR tree-optimization/54976
    * tree-vect-stmts.c (get_vectype_for_scalar_type_and_size):
    Robustify against odd inner_mode inputs.

    2012-10-12  Richard Biener  <rguenther@suse.de>

    PR tree-optimization/54894
    * tree-vect-stmts.c (get_vectype_for_scalar_type_and_size):
    Handle over-aligned scalar types properly.

    2012-10-02  Richard Guenther  <rguenther@suse.de>

    PR middle-end/54735
    * tree-ssa-pre.c (do_pre): Make sure to update virtual SSA form before
    cleaning up the CFG.


    2012-10-12  Richard Biener  <rguenther@suse.de>

    PR tree-optimization/54894
    * gcc.dg/torture/pr54894.c: New testcase.

    2012-10-02  Richard Guenther  <rguenther@suse.de>

    PR middle-end/54735
    * g++.dg/torture/pr54735.C: New testcase.

Added:
    branches/gcc-4_7-branch/gcc/testsuite/g++.dg/torture/pr54735.C
    branches/gcc-4_7-branch/gcc/testsuite/gcc.dg/torture/pr54894.c
Modified:
    branches/gcc-4_7-branch/gcc/ChangeLog
    branches/gcc-4_7-branch/gcc/testsuite/ChangeLog
    branches/gcc-4_7-branch/gcc/tree-ssa-pre.c
    branches/gcc-4_7-branch/gcc/tree-vect-stmts.c

[Bug tree-optimization/54894] [4.6 Regression] internal compiler error: in vect_get_vec_def_for_operand, at tree-vect-stmts.c:1286

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54894

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
      Known to work|                            |4.7.2
            Summary|[4.6/4.7 Regression]        |[4.6 Regression] internal
                   |internal compiler error: in |compiler error: in
                   |vect_get_vec_def_for_operan |vect_get_vec_def_for_operan
                   |d, at                       |d, at
                   |tree-vect-stmts.c:1286      |tree-vect-stmts.c:1286
      Known to fail|                            |4.7.1

--- Comment #9 from Richard Biener <rguenth at gcc dot gnu.org> 2012-11-26 14:27:23 UTC ---
Fixed for 4.7 as well.

[Bug tree-optimization/54894] [4.6 Regression] internal compiler error: in vect_get_vec_def_for_operand, at tree-vect-stmts.c:1286

[jakub at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54894

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|ASSIGNED                    |RESOLVED
         Resolution|                            |FIXED
   Target Milestone|4.6.4                       |4.7.2

--- Comment #10 from Jakub Jelinek <jakub at gcc dot gnu.org> 2013-04-12 16:27:40 UTC ---
The 4.6 branch has been closed, fixed in GCC 4.7.2.