[Bug target/63897] New: [5.0 regression] gcc.dg/torture/vector-2.c fails at on x86_64-apple-darwin14

[Bug target/63897] New: [5.0 regression] gcc.dg/torture/vector-2.c fails at on x86_64-apple-darwin14

[fxcoudert at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=63897

            Bug ID: 63897
           Summary: [5.0 regression] gcc.dg/torture/vector-2.c fails at on
                    x86_64-apple-darwin14
           Product: gcc
           Version: 5.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: target
          Assignee: unassigned at gcc dot gnu.org
          Reporter: fxcoudert at gcc dot gnu.org

Created attachment 33987
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=33987&action=edit
Assembler generated by "gcc -msse vector-2.c -O1 -m32 -S"

gcc.dg/torture/vector-2.c currently fails on trunk, when compiled with "-msse
vector-2.c -O1 -m32". Here is a minimal reproducer:

----------------------------
#define vector __attribute__((vector_size(16) ))

vector int f0(vector int t)
{
 ((int*)&t)[0] = 1;
 return t;
}
vector int f1(vector int t)
{
 ((int*)&t)[1] = 1;
 return t;
}

int main(void)
{
  vector int a = {0, 0, 0, 0};
  vector int b = {1, 0, 0, 0};
  vector int c = {0, 1, 0, 0};
  vector int a0 = f0(a);
  if (__builtin_memcmp (&a0, &b, sizeof(a0)))
    __builtin_abort ();
  a0 = f1(a);
  if (__builtin_memcmp (&a0, &c, sizeof(a0)))
    __builtin_abort ();
  return 0;
}
----------------------------


When compiled with "gcc -msse vector-2.c -O1 -m32", this segfaults at runtime.
The backtrace is:

* thread #1: tid = 0xdc2a68, 0x00001ea9 a.out`main + 81, queue =
'com.apple.main-thread', stop reason = EXC_BAD_ACCESS (code=1, address=0xb0)
    frame #0: 0x00001ea9 a.out`main + 81
a.out`main + 81:
-> 0x1ea9:  movd   0xb0(%eax), %xmm0
   0x1eb1:  pshufd $-0x1f, %xmm0, %xmm0
   0x1eb6:  movaps %xmm0, 0x10(%esp)
   0x1ebb:  subl   $0x4, %esp

The assembler generated is attached.

[Bug target/63897] [5.0 regression] gcc.dg/torture/vector-2.c fails at on x86_64-apple-darwin14

[fxcoudert at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=63897

Francois-Xavier Coudert <fxcoudert at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Keywords|                            |wrong-code
             Target|                            |x86_64-apple-darwin14
             Status|UNCONFIRMED                 |NEW
   Last reconfirmed|                            |2014-11-15
               Host|                            |x86_64-apple-darwin14
     Ever confirmed|0                           |1
              Build|                            |x86_64-apple-darwin14

[Bug target/63897] [5.0 regression] gcc.dg/torture/vector-2.c fails at on x86_64-apple-darwin14

[dominiq at lps dot ens.fr]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=63897

--- Comment #1 from Dominique d'Humieres <dominiq at lps dot ens.fr> ---
I don't see it on my builds.

[Book15] f90/bug% gcc410 -msse -O1 -m32
/opt/gcc/work/gcc/testsuite/gcc.dg/torture/vector-2.c
[Book15] f90/bug% a.out 
[Book15] f90/bug% gcc410 -msse -O1 -m32 pr63897.c
[Book15] f90/bug% a.out 
[Book15] f90/bug% gcc410 -v
Using built-in specs.
COLLECT_GCC=gcc410
COLLECT_LTO_WRAPPER=/opt/gcc/gcc4.10w/libexec/gcc/x86_64-apple-darwin14.0.0/5.0.0/lto-wrapper
Target: x86_64-apple-darwin14.0.0
Configured with: ../work/configure --prefix=/opt/gcc/gcc4.10w
--enable-languages=c,c++,fortran,objc,obj-c++,ada,java,lto --with-gmp=/opt/mp
--with-system-zlib --with-isl=/opt/mp --enable-lto --enable-plugin
--with-arch=corei7 --with-cpu=corei7
Thread model: posix
gcc version 5.0.0 20141115 (experimental) [trunk revision 217602p25] (GCC)

[Bug target/63897] [5.0 regression] gcc.dg/torture/vector-2.c fails at on x86_64-apple-darwin14

[fxcoudert at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=63897

--- Comment #2 from Francois-Xavier Coudert <fxcoudert at gcc dot gnu.org> ---
(In reply to Dominique d'Humieres from comment #1)
> Configured with: ../work/configure --prefix=/opt/gcc/gcc4.10w
> --enable-languages=c,c++,fortran,objc,obj-c++,ada,java,lto
> --with-gmp=/opt/mp --with-system-zlib --with-isl=/opt/mp --enable-lto
> --enable-plugin --with-arch=corei7 --with-cpu=corei7

Apparently, it's because you configure --with-arch=corei7. If I compile with
-march=corei7, it generates good code (well, it doesn't segfault, at least).
Try compiling with -march=core2, you should see it.

[Bug target/63897] [5.0 regression] gcc.dg/torture/vector-2.c fails at on x86_64-apple-darwin14

[dominiq at lps dot ens.fr]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=63897

--- Comment #3 from Dominique d'Humieres <dominiq at lps dot ens.fr> ---
> Try compiling with -march=core2, you should see it.

Confirmed.

[Bug target/63897] [5.0 regression] gcc.dg/torture/vector-2.c fails at on x86_64-apple-darwin14

[dominiq at lps dot ens.fr]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=63897

Dominique d'Humieres <dominiq at lps dot ens.fr> changed:

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

--- Comment #4 from Dominique d'Humieres <dominiq at lps dot ens.fr> ---
This due to r216154.

[Bug target/63897] [5.0 regression] gcc.dg/torture/vector-2.c fails at on x86_64-apple-darwin14

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=63897

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Priority|P3                          |P1
   Target Milestone|---                         |5.0

[Bug target/63897] [5.0 regression] gcc.dg/torture/vector-2.c fails at on x86_64-apple-darwin14

[izamyatin at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=63897

--- Comment #6 from Igor Zamyatin <izamyatin at gmail dot com> ---
after RA=after reload

[Bug target/63897] [5.0 regression] gcc.dg/torture/vector-2.c fails at on x86_64-apple-darwin14

[vmakarov at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=63897

Vladimir Makarov <vmakarov at gcc dot gnu.org> changed:

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

--- Comment #7 from Vladimir Makarov <vmakarov at gcc dot gnu.org> ---
I've reproduced this bug too.  The problem is that although global live info
was updated, other important information (like hard registers conflicts, calls
crossed, live ranges) was not.  It needs to rerun live range again after
changing global live info.

I am working on it.  It needs some time, especially when some solutions would
have a big impact on compiler time.  I hope the fix will be ready on Friday (if
I am very lucky) or at the beginning of the next week.

[Bug target/63897] [5.0 regression] gcc.dg/torture/vector-2.c fails at on x86_64-apple-darwin14

[vmakarov at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=63897

--- Comment #8 from Vladimir Makarov <vmakarov at gcc dot gnu.org> ---
Author: vmakarov
Date: Fri Nov 21 21:29:54 2014
New Revision: 217947

URL: https://gcc.gnu.org/viewcvs?rev=217947&root=gcc&view=rev
Log:
2014-11-21  Vladimir Makarov  <vmakarov@redhat.com>

    PR target/63897
    * lra-lives.c (mark_regno_live, mark_regno_dead): Remove last
    argument.
    (process_bb_lives): Rename dead_insn_p on remove_p
    and global_live_info_p on dead_insn_p.  Calculate local live info
    unconditionally.  Remove last argument in calls mark_regno_live and
    mark_regno_dead.  Reorganize body of EXECUTE_IF_SET_IN_BITMAP.
    (lra_create_live_ranges): Rename to lra_create_live_ranges_1.
    Return bool.  Rename global_live_info_p on dead_insn_p.  Return
    flag of live info change.
    (lra_create_live_ranges): New.


Modified:
    trunk/gcc/ChangeLog
    trunk/gcc/lra-lives.c