[Bug middle-end/46142] New: [4.6 Regression] FMA test failures

[Bug middle-end/46142] New: [4.6 Regression] FMA test failures

[hjl.tools at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46142

           Summary: [4.6 Regression] FMA test failures
           Product: gcc
           Version: 4.6.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: middle-end
        AssignedTo: unassigned@gcc.gnu.org
        ReportedBy: hjl.tools@gmail.com
                CC: rth@gcc.gnu.org


On Linux/ia32,  revision 165859 gave

FAIL: gcc.target/i386/fma3-builtin-2.c (internal compiler error)
FAIL: gcc.target/i386/fma3-builtin-2.c (test for excess errors)
FAIL: gcc.target/i386/fma4-builtin-2.c (internal compiler error)
FAIL: gcc.target/i386/fma4-builtin-2.c (test for excess errors)
FAIL: gfortran.dg/vect/fast-math-vect-8.f90 (internal compiler error)
FAIL: gfortran.dg/vect/fast-math-vect-8.f90 (test for excess errors)
ERROR: gcc.target/i386/fma3-builtin-2.c: error executing dg-final: couldn't
open "fma3-builtin-2.s": no such file or directory
ERROR: gcc.target/i386/fma4-builtin-2.c: error executing dg-final: couldn't
open "fma4-builtin-2.s": no such file or directory

Revision 165851 is OK.

[Bug middle-end/46142] [4.6 Regression] FMA test failures

[hjl.tools at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46142

--- Comment #1 from H.J. Lu <hjl.tools at gmail dot com> 2010-10-23 00:40:08 UTC ---
I saw

FAIL: gcc.target/i386/fma3-builtin-2.c (internal compiler error)
FAIL: gcc.target/i386/fma3-builtin-2.c (test for excess errors)
FAIL: gcc.target/i386/fma4-builtin-2.c (internal compiler error)
FAIL: gcc.target/i386/fma4-builtin-2.c (test for excess errors)

only 32bit gcc, not with 64bit gcc using -m32. It could be the 32bit HWI
problem.

[Bug middle-end/46142] [4.6 Regression] FMA test failures

[hjl.tools at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46142

--- Comment #2 from H.J. Lu <hjl.tools at gmail dot com> 2010-10-23 00:41:49 UTC ---
I got

./xgcc -B./ -S ../../src-trunk/gcc/testsuite/gcc.target/i386/fma3-builtin-2.c
-O3 -mfma -mno-fma4 -S  
../../src-trunk/gcc/testsuite/gcc.target/i386/fma3-builtin-2.c: In function
‘vector_fma’:
../../src-trunk/gcc/testsuite/gcc.target/i386/fma3-builtin-2.c:19:1: internal
compiler error: in vect_is_simple_use_1, at tree-vect-stmts.c:5288
Please submit a full bug report,
with preprocessed source if appropriate.
See <http://gcc.gnu.org/bugs.html> for instructions.

[Bug middle-end/46142] [4.6 Regression] FMA test failures

[hjl.tools at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46142

--- Comment #3 from H.J. Lu <hjl.tools at gmail dot com> 2010-10-23 00:46:01 UTC ---
On Fedora 13/x86-64, you can use

CC="gcc -m32" CXX="g++ -m32" ../src-trunk/configure i686-linux

to bootstrap 32bit gcc.  You need to install 32bit libraries
used by gcc.

[Bug middle-end/46142] [4.6 Regression] FMA test failures

[ubizjak at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46142

Uros Bizjak <ubizjak at gmail dot com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |ASSIGNED
   Last reconfirmed|                            |2010.10.31 06:55:48
   Target Milestone|---                         |4.6.0
     Ever Confirmed|0                           |1

--- Comment #4 from Uros Bizjak <ubizjak at gmail dot com> 2010-10-31 06:55:48 UTC ---
Well, well.

static bool
vectorizable_call (gimple stmt, gimple_stmt_iterator *gsi, gimple *vec_stmt)
{
  ...
  enum vect_def_type dt[2] = {vect_unknown_def_type, vect_unknown_def_type};
  ...

  /* Bail out if the function has more than three arguments, we do not have
     interesting builtin functions to vectorize with more than two arguments
     except for fma.  No arguments is also not good.  */
  if (nargs == 0 || nargs > 3)
    return false;
  ...
}

For some reason, valgrind didn't cathc this problem. Anyway, the patch is
obvious:

Index: tree-vect-stmts.c
===================================================================
--- tree-vect-stmts.c    (revision 166104)
+++ tree-vect-stmts.c    (working copy)
@@ -1308,7 +1308,8 @@ vectorizable_call (gimple stmt, gimple_s
   loop_vec_info loop_vinfo = STMT_VINFO_LOOP_VINFO (stmt_info);
   tree fndecl, new_temp, def, rhs_type;
   gimple def_stmt;
-  enum vect_def_type dt[2] = {vect_unknown_def_type, vect_unknown_def_type};
+  enum vect_def_type dt[3]
+    = {vect_unknown_def_type, vect_unknown_def_type, vect_unknown_def_type};
   gimple new_stmt = NULL;
   int ncopies, j;
   VEC(tree, heap) *vargs = NULL;

[Bug tree-optimization/46142] [4.6 Regression] FMA test failures

[uros at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46142

--- Comment #5 from uros at gcc dot gnu.org 2010-10-31 07:18:13 UTC ---
Author: uros
Date: Sun Oct 31 07:18:07 2010
New Revision: 166105

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=166105
Log:
    PR tree-optimization/46142
    * tree-vect-stmts.c (vectorizable_call): Enlarge dt array to
    accommodate third function argument.


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

[Bug tree-optimization/46142] [4.6 Regression] FMA test failures

[ubizjak at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46142

Uros Bizjak <ubizjak at gmail dot com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                URL|                            |http://gcc.gnu.org/ml/gcc-p
                   |                            |atches/2010-10/msg02658.htm
                   |                            |l

--- Comment #6 from Uros Bizjak <ubizjak at gmail dot com> 2010-10-31 07:19:33 UTC ---
Fixed.

[Bug tree-optimization/46142] [4.6 Regression] FMA test failures

[hjl.tools at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46142

--- Comment #7 from H.J. Lu <hjl.tools at gmail dot com> 2010-10-31 14:27:41 UTC ---
When configured with --with-cpu=atom, I got

FAIL: gcc.target/i386/fma3-builtin-2.c scan-assembler-times vfmadd...pd 1
FAIL: gcc.target/i386/fma3-builtin-2.c scan-assembler-times vfmsub...pd 1
FAIL: gcc.target/i386/fma3-builtin-2.c scan-assembler-times vfnmadd...pd 1
FAIL: gcc.target/i386/fma3-builtin-2.c scan-assembler-times vfnmsub...pd 1
FAIL: gcc.target/i386/fma4-builtin-2.c scan-assembler-times vfmaddpd 1
FAIL: gcc.target/i386/fma4-builtin-2.c scan-assembler-times vfmsubpd 1
FAIL: gcc.target/i386/fma4-builtin-2.c scan-assembler-times vfnmaddpd 1
FAIL: gcc.target/i386/fma4-builtin-2.c scan-assembler-times vfnmsubpd 1

since -mtune=atom disables double precision vectorizer.
We need to pass -mtune=generic to those 2 testcases.

[Bug tree-optimization/46142] [4.6 Regression] FMA test failures

[ubizjak at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46142

--- Comment #8 from Uros Bizjak <ubizjak at gmail dot com> 2010-10-31 18:00:58 UTC ---
(In reply to comment #7)

> since -mtune=atom disables double precision vectorizer.
> We need to pass -mtune=generic to those 2 testcases.

The patch to add -mtune=generic to dg-options in preapproved.

[Bug tree-optimization/46142] [4.6 Regression] FMA test failures

[hjl at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46142

--- Comment #9 from hjl at gcc dot gnu.org <hjl at gcc dot gnu.org> 2010-10-31 19:07:13 UTC ---
Author: hjl
Date: Sun Oct 31 19:07:09 2010
New Revision: 166113

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=166113
Log:
Add -mtune=generic to fma3-builtin-2.c/fma4-builtin-2.c.

2010-10-31  H.J. Lu  <hongjiu.lu@intel.com>

    PR target/46142
    * gcc.target/i386/fma3-builtin-2.c: Add -mtune=generic.
    * gcc.target/i386/fma4-builtin-2.c: Likewise.

Modified:
    trunk/gcc/testsuite/ChangeLog
    trunk/gcc/testsuite/gcc.target/i386/fma3-builtin-2.c
    trunk/gcc/testsuite/gcc.target/i386/fma4-builtin-2.c

[Bug tree-optimization/46142] [4.6 Regression] FMA test failures

[jakub at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46142

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

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

--- Comment #10 from Jakub Jelinek <jakub at gcc dot gnu.org> 2010-11-03 08:48:16 UTC ---
I don't see any *fma* failures on current trunk, neither i686-linux nor
x86_64-linux.  Can this be closed now?

[Bug tree-optimization/46142] [4.6 Regression] FMA test failures

[ubizjak at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46142

Uros Bizjak <ubizjak at gmail dot com> changed:

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

--- Comment #11 from Uros Bizjak <ubizjak at gmail dot com> 2010-11-03 08:52:57 UTC ---
Fixed again.