[Bug bootstrap/53331] New: [4.8 Regression] AIX bootstrap failure in tree-vect-data-ref compiling matmul_i4

[Bug bootstrap/53331] New: [4.8 Regression] AIX bootstrap failure in tree-vect-data-ref compiling matmul_i4

[dje at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53331

             Bug #: 53331
           Summary: [4.8 Regression] AIX bootstrap failure in
                    tree-vect-data-ref compiling matmul_i4
    Classification: Unclassified
           Product: gcc
           Version: 4.8.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: bootstrap
        AssignedTo: unassigned@gcc.gnu.org
        ReportedBy: dje@gcc.gnu.org


The recent patch to tree-vect-data-ref.c produces an ICE when compiling
matmul_i4 in libgfortran on PowerPC:

/farm/dje/src/src/libgfortran/generated/matmul_i4.c: In function 'matmul_i4':
/farm/dje/src/src/libgfortran/generated/matmul_i4.c:79:1: internal compiler
error: in vect_enhance_data_refs_alignment, at tree-vect-data-refs.c:1963
 matmul_i4 (gfc_array_i4 * const restrict retarray,

[Bug bootstrap/53331] [4.8 Regression] AIX bootstrap failure in tree-vect-data-ref compiling matmul_i4

[dje at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53331

David Edelsohn <dje at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Target|                            |powerpc64-*-*
             Status|UNCONFIRMED                 |NEW
           Keywords|                            |ice-on-valid-code
   Last reconfirmed|                            |2012-05-12
                 CC|                            |rguenth at gcc dot gnu.org
               Host|                            |powerpc64-*-*
     Ever Confirmed|0                           |1
   Target Milestone|---                         |4.8.0
              Build|                            |powerpc64-*-*

--- Comment #1 from David Edelsohn <dje at gcc dot gnu.org> 2012-05-12 14:25:31 UTC ---
/tmp/20120511/./gcc/xgcc -B/tmp/20120511/./gcc/
-B/farm/dje/install/powerpc-ibm-aix5.3.0.0-20120511/powerpc-ibm-aix5.3.0.0/bin/
-B/farm/dje/install/powerpc-ibm-aix5.3.0.0-20120511/powerpc-ibm-aix5.3.0.0/lib/
-isystem
/farm/dje/install/powerpc-ibm-aix5.3.0.0-20120511/powerpc-ibm-aix5.3.0.0/include
-isystem
/farm/dje/install/powerpc-ibm-aix5.3.0.0-20120511/powerpc-ibm-aix5.3.0.0/sys-include
-maix64 -DHAVE_CONFIG_H -I. -I/farm/dje/src/src/libgfortran
-iquote/farm/dje/src/src/libgfortran/io -I/farm/dje/src/src/libgfortran/../gcc
-I/farm/dje/src/src/libgfortran/../gcc/config -I../../.././gcc
-I/farm/dje/src/src/libgfortran/../libgcc -I../../libgcc -std=gnu99 -Wall
-Wstrict-prototypes -Wmissing-prototypes -Wold-style-definition -Wextra
-Wwrite-strings -fcx-fortran-rules -ffunction-sections -fdata-sections
-ftree-vectorize -funroll-loops -std=gnu99 -g -O2 -Wunknown-pragmas -MT
matmul_i4.lo -MD -MP -MF .deps/matmul_i4.Tpo -c
/farm/dje/src/src/libgfortran/generated/matmul_i4.c  -DPIC -o .libs/matmul_i4.o

[Bug bootstrap/53331] [4.8 Regression] AIX bootstrap failure in tree-vect-data-ref compiling matmul_i4

[dominiq at lps dot ens.fr]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53331

--- Comment #2 from Dominique d'Humieres <dominiq at lps dot ens.fr> 2012-05-12 14:27:07 UTC ---
On powerpc-apple-darwin9 I get a similar failure:

../../../work/libgfortran/generated/matmul_i8.c: In function 'matmul_i8':
../../../work/libgfortran/generated/matmul_i8.c:79:1: internal compiler error:
in vect_enhance_data_refs_alignment, at tree-vect-data-refs.c:1963
 matmul_i8 (gfc_array_i8 * const restrict retarray,

[Bug bootstrap/53331] [4.8 Regression] AIX bootstrap failure in tree-vect-data-ref compiling matmul_i4

[dominiq at lps dot ens.fr]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53331

--- Comment #3 from Dominique d'Humieres <dominiq at lps dot ens.fr> 2012-05-13 11:38:44 UTC ---
This is due to revision 187402: r187401 bootstrapped without any problem.

[Bug bootstrap/53331] [4.8 Regression] AIX bootstrap failure in tree-vect-data-ref compiling matmul_i4

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53331

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

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

--- Comment #4 from Richard Guenther <rguenth at gcc dot gnu.org> 2012-05-14 08:14:53 UTC ---
Mine.  Can any of you attach preprocessed source please so I can reproduce with
a cross?  Thx.

[Bug bootstrap/53331] [4.8 Regression] AIX bootstrap failure in tree-vect-data-ref compiling matmul_i4

[dominiq at lps dot ens.fr]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53331

--- Comment #5 from Dominique d'Humieres <dominiq at lps dot ens.fr> 2012-05-14 08:51:14 UTC ---
Created attachment 27398
  --> http://gcc.gnu.org/bugzilla/attachment.cgi?id=27398
preprocessed source matmul_i8.i for powerpc-apple-darwin9

[Bug bootstrap/53331] [4.8 Regression] AIX bootstrap failure in tree-vect-data-ref compiling matmul_i4

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53331

--- Comment #6 from Richard Guenther <rguenth at gcc dot gnu.org> 2012-05-14 10:02:12 UTC ---
Can you try

Index: gcc/tree-vect-data-refs.c
===================================================================
--- gcc/tree-vect-data-refs.c   (revision 187449)
+++ gcc/tree-vect-data-refs.c   (working copy)
@@ -1078,6 +1078,11 @@ vect_verify_datarefs_alignment (loop_vec
           || !STMT_VINFO_VECTORIZABLE (stmt_info))
         continue;

+      /* Strided loads perform only component accesses, alignment is
+        irrelevant for them.  */
+      if (STMT_VINFO_STRIDE_LOAD_P (stmt_info))
+       continue;
+
       supportable_dr_alignment = vect_supportable_dr_alignment (dr, false);
       if (!supportable_dr_alignment)
         {
Index: gcc/tree-vect-stmts.c
===================================================================
--- gcc/tree-vect-stmts.c       (revision 187449)
+++ gcc/tree-vect-stmts.c       (working copy)
@@ -1032,10 +1030,14 @@ vect_model_load_cost (stmt_vec_info stmt
     }

   /* The loads themselves.  */
-  vect_get_load_cost (first_dr, ncopies,
-         ((!STMT_VINFO_GROUPED_ACCESS (stmt_info)) || group_size > 1
-          || slp_node),
-         &inside_cost, &outside_cost);
+  if (STMT_VINFO_STRIDE_LOAD_P (stmt_info))
+    inside_cost
+      += ncopies * TYPE_VECTOR_SUBPARTS (STMT_VINFO_VECTYPE (stmt_info));
+  else
+    vect_get_load_cost (first_dr, ncopies,
+                       ((!STMT_VINFO_GROUPED_ACCESS (stmt_info))
+                        || group_size > 1 || slp_node),
+                       &inside_cost, &outside_cost);

   if (vect_print_dump_info (REPORT_COST))
     fprintf (vect_dump, "vect_model_load_cost: inside_cost = %d, "

[Bug bootstrap/53331] [4.8 Regression] AIX bootstrap failure in tree-vect-data-ref compiling matmul_i4

[dominiq at lps dot ens.fr]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53331

--- Comment #7 from Dominique d'Humieres <dominiq at lps dot ens.fr> 2012-05-14 12:32:26 UTC ---
> Can you try

>From a clean bootstrap at revision 187401, updating gcc and rebuilding
libgfortran has succeeded for 187402 with the patch in comment #6 (with
libgfortran moved away),. Thanks for the patch.

[Bug bootstrap/53331] [4.8 Regression] AIX bootstrap failure in tree-vect-data-ref compiling matmul_i4

[ro at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53331

Rainer Orth <ro at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Target|powerpc64-*-*               |powerpc64-*-*,
                   |                            |sparc-sun-solaris2.*
                 CC|                            |ro at gcc dot gnu.org
               Host|powerpc64-*-*               |powerpc64-*-*,
                   |                            |sparc-sun-solaris2.*
              Build|powerpc64-*-*               |powerpc64-*-*,
                   |                            |sparc-sun-solaris2.*

--- Comment #8 from Rainer Orth <ro at gcc dot gnu.org> 2012-05-14 13:32:59 UTC ---
I'm seeing the same failure on Solaris/SPARC for the 64-bit multilib. If I drop
your patch into the source tree and restart the build, it completes
successfully.
make check still running.

Thanks.
  Rainer

[Bug bootstrap/53331] [4.8 Regression] AIX bootstrap failure in tree-vect-data-ref compiling matmul_i4

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53331

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|WAITING                     |ASSIGNED

--- Comment #9 from Richard Guenther <rguenth at gcc dot gnu.org> 2012-05-14 13:44:22 UTC ---
Ok, I'm going to commit a slightly different variant after testing on x86_64
(with a more reasonable cost).

[Bug bootstrap/53331] [4.8 Regression] AIX bootstrap failure in tree-vect-data-ref compiling matmul_i4

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53331

--- Comment #10 from Richard Guenther <rguenth at gcc dot gnu.org> 2012-05-14 15:01:35 UTC ---
Author: rguenth
Date: Mon May 14 15:01:22 2012
New Revision: 187466

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=187466
Log:
2012-05-14  Richard Guenther  <rguenther@suse.de>

    PR tree-optimization/53331
    * tree-vect-data-refs.c (vect_verify_datarefs_alignment): Ignore
    strided loads.
    * tree-vect-stmts.c (vect_model_load_cost): Handle strided loads.

Modified:
    trunk/gcc/ChangeLog
    trunk/gcc/tree-vect-data-refs.c
    trunk/gcc/tree-vect-stmts.c

[Bug bootstrap/53331] [4.8 Regression] AIX bootstrap failure in tree-vect-data-ref compiling matmul_i4

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53331

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> 2012-05-14 15:02:25 UTC ---
Fixed.