[Bug target/67946] New: Function multiversioning ICE

[Bug target/67946] New: Function multiversioning ICE

[evstupac at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=67946

            Bug ID: 67946
           Summary: Function multiversioning ICE
           Product: gcc
           Version: 6.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: target
          Assignee: unassigned at gcc dot gnu.org
          Reporter: evstupac at gmail dot com
  Target Milestone: ---

The following test:

#include <immintrin.h>

__m256 x, y, z;

__attribute__((target("avx")))
int bar()
{
  x = _mm256_add_ps (y, z);
  return 1;
}

__attribute__((target("default")))
int bar()
{
  return 2;
}

int
foobar()
{
  if (__builtin_cpu_supports ("avx"))
    return bar();
  else
    return 0;
}

__attribute__((target("sse3")))
int foo(int i)
{
  return foobar() + 1;
}

__attribute__((target("default")))
int foo(int i)
{
  return foobar() + 2;
}

ICE by:

gcc test.cpp -O3 -S -msse4.2 
test.cpp: In function 'int bar()':
test.cpp:8:27: internal compiler error: in expand_expr_real_2, at expr.c:9322
   x = _mm256_add_ps (y, z);
                           ^ 
0x9e0445 expand_expr_real_2(separate_ops*, rtx_def*, machine_mode,
expand_modifier)
        /export/users/gnutester/stability/svn/trunk/gcc/expr.c:9322
0x9cd8da expand_expr_real_1(tree_node*, rtx_def*, machine_mode,
expand_modifier, rtx_def**, bool)
        /export/users/gnutester/stability/svn/trunk/gcc/expr.c:9523
0x9d7f62 store_expr_with_bounds(tree_node*, rtx_def*, int, bool, tree_node*)
        /export/users/gnutester/stability/svn/trunk/gcc/expr.c:5399
0x9da55d expand_assignment(tree_node*, tree_node*, bool)
        /export/users/gnutester/stability/svn/trunk/gcc/expr.c:5171
0x8cb1de expand_gimple_stmt_1
        /export/users/gnutester/stability/svn/trunk/gcc/cfgexpand.c:3581
0x8cb1de expand_gimple_stmt
        /export/users/gnutester/stability/svn/trunk/gcc/cfgexpand.c:3677
0x8cd511 expand_gimple_basic_block
        /export/users/gnutester/stability/svn/trunk/gcc/cfgexpand.c:5681
0x8d41f6 execute
        /export/users/gnutester/stability/svn/trunk/gcc/cfgexpand.c:6293
Please submit a full bug report,
with preprocessed source if appropriate.
Please include the complete backtrace with any bug report.
See <http://gcc.gnu.org/bugs.html> for instructions.

[Bug target/67946] Function multiversioning ICE

[ubizjak at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=67946

--- Comment #1 from Uroš Bizjak <ubizjak at gmail dot com> ---
You should compile this with -mavx, otherwise x, y and z global variables get
expanded in BLKmode. So, we get:

(gdb) p code
$1 = PLUS_EXPR
(gdb) p op0
$2 = (rtx) 0x2aaab05a4918
(gdb) p debug_rtx (op0)
(mem/c:BLK (symbol_ref:DI ("z") [flags 0x2] <var_decl 0x2aaab058da20 z>) [1 z+0
S32 A256])
$3 = void
(gdb) p debug_rtx (op1)
(mem/c:BLK (symbol_ref:DI ("y") [flags 0x2] <var_decl 0x2aaab058d990 y>) [1 y+0
S32 A256])
>From gcc-bugs-return-499393-listarch-gcc-bugs=gcc.gnu.org@gcc.gnu.org Tue Oct 13 07:48:47 2015
Return-Path: <gcc-bugs-return-499393-listarch-gcc-bugs=gcc.gnu.org@gcc.gnu.org>
Delivered-To: listarch-gcc-bugs@gcc.gnu.org
Received: (qmail 104266 invoked by alias); 13 Oct 2015 07:48:47 -0000
Mailing-List: contact gcc-bugs-help@gcc.gnu.org; run by ezmlm
Precedence: bulk
List-Id: <gcc-bugs.gcc.gnu.org>
List-Archive: <http://gcc.gnu.org/ml/gcc-bugs/>
List-Post: <mailto:gcc-bugs@gcc.gnu.org>
List-Help: <mailto:gcc-bugs-help@gcc.gnu.org>
Sender: gcc-bugs-owner@gcc.gnu.org
Delivered-To: mailing list gcc-bugs@gcc.gnu.org
Received: (qmail 99993 invoked by uid 48); 13 Oct 2015 07:48:43 -0000
From: "gerhard.steinmetz.fortran@t-online.de" <gcc-bugzilla@gcc.gnu.org>
To: gcc-bugs@gcc.gnu.org
Subject: [Bug fortran/67938] ICE on using assumed rank character with some intrinsics
Date: Tue, 13 Oct 2015 07:48:00 -0000
X-Bugzilla-Reason: CC
X-Bugzilla-Type: changed
X-Bugzilla-Watch-Reason: None
X-Bugzilla-Product: gcc
X-Bugzilla-Component: fortran
X-Bugzilla-Version: 5.2.1
X-Bugzilla-Keywords:
X-Bugzilla-Severity: normal
X-Bugzilla-Who: gerhard.steinmetz.fortran@t-online.de
X-Bugzilla-Status: UNCONFIRMED
X-Bugzilla-Resolution:
X-Bugzilla-Priority: P3
X-Bugzilla-Assigned-To: unassigned at gcc dot gnu.org
X-Bugzilla-Target-Milestone: ---
X-Bugzilla-Flags:
X-Bugzilla-Changed-Fields:
Message-ID: <bug-67938-4-OhaGCT27TW@http.gcc.gnu.org/bugzilla/>
In-Reply-To: <bug-67938-4@http.gcc.gnu.org/bugzilla/>
References: <bug-67938-4@http.gcc.gnu.org/bugzilla/>
Content-Type: text/plain; charset="UTF-8"
Content-Transfer-Encoding: 7bit
X-Bugzilla-URL: http://gcc.gnu.org/bugzilla/
Auto-Submitted: auto-generated
MIME-Version: 1.0
X-SW-Source: 2015-10/txt/msg00948.txt.bz2
Content-length: 287

https://gcc.gnu.org/bugzilla/show_bug.cgi?idg938

--- Comment #2 from Gerhard Steinmetz <gerhard.steinmetz.fortran@t-online.de> ---
As meanwhile noticed, example z1.f90 from comment #1 overlaps with PR66833.
Examples z2.f90 (comment #0) and z1s.f90 are describing additional effects.

[Bug target/67946] Function multiversioning ICE

[evstupac at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=67946

--- Comment #2 from Stupachenko Evgeny <evstupac at gmail dot com> ---
I see, but shouldn't compiler generate en error on this instead of ICE?

[Bug middle-end/67946] Function multiversioning ICE

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=67946

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
   Last reconfirmed|                            |2015-10-13
          Component|target                      |middle-end
     Ever confirmed|0                           |1

--- Comment #3 from Richard Biener <rguenth at gcc dot gnu.org> ---
I don't think compiling with -mavx is desired here.  The "bug" is that
DECL_MODE doesn't have the same special-casing of vectors as TYPE_MODE has
and ...

/* Holds the machine mode corresponding to the declaration of a variable or
   field.  Always equal to TYPE_MODE (TREE_TYPE (decl)) except for a
   FIELD_DECL.  */
#define DECL_MODE(NODE) (DECL_COMMON_CHECK (NODE)->decl_common.mode)

... the comment also is wrong here.

Not sure if we really want to change the above to sth like

#define DECL_MODE(NODE) \
 (VECTOR_TYPE_P (TREE_TYPE (NODE)) \
  ? vector_type_mode (TREE_TYPE (NODE)) \
  : DECL_COMMON_CHECK (NODE)->decl_common.mode)

though.  Maybe changing references to DECL_MODE to use TYPE_MODE in some
selected places would be enough ...

[Bug middle-end/67946] Function multiversioning ICE

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=67946

Andrew Pinski <pinskia at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|---                         |6.0
      Known to work|                            |10.1.0, 12.0, 6.1.0, 6.4.0,
                   |                            |7.1.0, 8.5.0
         Resolution|---                         |FIXED
           Keywords|                            |ice-on-valid-code
      Known to fail|                            |5.5.0
             Status|NEW                         |RESOLVED
           See Also|                            |https://gcc.gnu.org/bugzill
                   |                            |a/show_bug.cgi?id=79124

--- Comment #5 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
Fixed so closing.  Most likely just like PR 79124.