[Bug target/63661] New: -O2 miscompiles on OSX 10.10 Yosemite

[Bug target/63661] New: -O2 miscompiles on OSX 10.10 Yosemite

[vbraun.name at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=63661

            Bug ID: 63661
           Summary: -O2 miscompiles on OSX 10.10 Yosemite
           Product: gcc
           Version: 4.9.1
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: target
          Assignee: unassigned at gcc dot gnu.org
          Reporter: vbraun.name at gmail dot com

Created attachment 33821
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=33821&action=edit
Testcase that yields wrong answer when compiled with -O2

The attached selfcontained test_mathieu.c is based on the GSL testsuite. I
tried to make it shorter but simple changes like adding printf tend to "fix"
it. Compilation fails on OSX 10.10 Yosemite when compiled with -O2:

$ gcc -O0 -o test_mathieu test_mathieu.c -lm && ./test_mathieu
loop -2.35802
loop 2
loop 2
loop 2
loop 2

$ gcc -O1 -o test_mathieu test_mathieu.c -lm && ./test_mathieu
loop -2.35802
loop 2
loop 2
loop 2
loop 2

$ gcc -O2 -o test_mathieu test_mathieu.c -lm && ./test_mathieu
loop -2.35802
loop 2.35802
loop 2.35802
loop 2.35802
loop 2.35802

Notes:
* The non-optimized values are the correct ones.
* Works on linux x86_64 with gcc 4.9.1 (Fedora 21 alpha)
* Works with Apple clang

[Bug target/63661] -O2 miscompiles on OSX 10.10 Yosemite

[vbraun.name at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=63661

--- Comment #1 from Volker Braun <vbraun.name at gmail dot com> ---
Created attachment 33822
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=33822&action=edit
Output of gcc -v

[Bug target/63661] -O2 miscompiles on OSX 10.10 Yosemite

[vbraun.name at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=63661

--- Comment #2 from Volker Braun <vbraun.name at gmail dot com> ---
Created attachment 33823
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=33823&action=edit
Preprocessed source

[Bug target/63661] -O2 miscompiles on OSX 10.10 Yosemite

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=63661

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |WAITING
   Last reconfirmed|                            |2014-10-28
     Ever confirmed|0                           |1

--- Comment #3 from Richard Biener <rguenth at gcc dot gnu.org> ---
What function from the libm library ends up getting used for you?  (none is
used on x86_64-linux for me)

[Bug target/63661] -O2 miscompiles on OSX 10.10 Yosemite

[vbraun.name at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=63661

--- Comment #4 from Volker Braun <vbraun.name at gmail dot com> ---
Good point, -lm is not necessary. It was necessary in the original GSL
testsuite case, but I removed that part.

[Bug target/63661] -O2 miscompiles on OSX 10.10 Yosemite

[howarth at bromo dot med.uc.edu]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=63661

howarth at bromo dot med.uc.edu changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |howarth at bromo dot med.uc.edu

--- Comment #5 from howarth at bromo dot med.uc.edu ---
Interestingly, I find on Yosemite that the system gcc produces...

% gcc -O0 -g -o test_mathieu test_mathieu.c -lm && ./test_mathieu
loop -2.35802
loop 2
loop 2
loop 2
loop 2

for -O0/-O1/-O2 when -g is used.

[Bug target/63661] -O2 miscompiles on OSX 10.10 Yosemite

[howarth at bromo dot med.uc.edu]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=63661

--- Comment #6 from howarth at bromo dot med.uc.edu ---
My mistake, I assumed the first output line was an error.

loop -2.35802

[Bug target/63661] -O2 miscompiles on OSX 10.10 Yosemite

[howarth at bromo dot med.uc.edu]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=63661

--- Comment #7 from howarth at bromo dot med.uc.edu ---
Created attachment 33828
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=33828&action=edit
assembly at -O2 from test_mathieu.c using clang as gcc in Xcode 6

[Bug target/63661] -O2 miscompiles on OSX 10.10 Yosemite

[howarth at bromo dot med.uc.edu]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=63661

--- Comment #8 from howarth at bromo dot med.uc.edu ---
Created attachment 33829
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=33829&action=edit
assembly at -O2 from test_mathieu.c using gcc 4.9.1-RC

[Bug target/63661] -O2 miscompiles on OSX 10.10 Yosemite

[howarth at bromo dot med.uc.edu]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=63661

--- Comment #9 from howarth at bromo dot med.uc.edu ---
Created attachment 33830
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=33830&action=edit
assembly at -O1 from test_mathieu.c using gcc 4.9.1-RC

[Bug target/63661] -O2 miscompiles on OSX 10.10 Yosemite

[howarth at bromo dot med.uc.edu]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=63661

--- Comment #10 from howarth at bromo dot med.uc.edu ---
for the testcase compiled with clang as gcc at -O2 on Yosemite

% setenv DYLD_PRINT_BINDINGS
% ./test_mathieu
dyld: bind: libgcc_s.1.dylib:0x1098E0000 = libdyld.dylib:dyld_stub_binder,
*0x1098E0000 = 0x7FFF9B4612A0
dyld: bind: test_mathieu:0x1098C8000 = libdyld.dylib:dyld_stub_binder,
*0x1098C8000 = 0x7FFF9B4612A0
dyld: resolver at 0x7fff9002924f returned 0x7FFF90029160
dyld: lazy bind: libsystem_c.dylib:0x7FFF7D8BB710 =
libsystem_platform.dylib:__platform_strchr, *0x7FFF7D8BB710 = 0x7FFF90029160
dyld: resolver at 0x7fff9002927a returned 0x7FFF9002D240
dyld: lazy bind: libsystem_c.dylib:0x7FFF7D8BB6E8 =
libsystem_platform.dylib:__platform_memmove, *0x7FFF7D8BB6E8 = 0x7FFF9002D240
dyld: resolver at 0x7fff9002927a returned 0x7FFF9002D240
dyld: lazy bind: libsystem_c.dylib:0x7FFF7D8BB6F0 =
libsystem_platform.dylib:__platform_memmove, *0x7FFF7D8BB6F0 = 0x7FFF9002D240
dyld: lazy bind: test_mathieu:0x1098C8010 = libsystem_c.dylib:_printf,
*0x1098C8010 = 0x7FFF8DA9C930
dyld: resolver at 0x7fff90028cad returned 0x7FFF90028B20
dyld: lazy bind: libsystem_c.dylib:0x7FFF7D8BB6D8 =
libsystem_platform.dylib:__platform_memchr, *0x7FFF7D8BB6D8 = 0x7FFF90028B20
loop -2.35802
loop 2.35802
loop 2.35802
loop 2.35802
loop 2.35802

for the testcase compiled with gcc 4.9.2-RC at -O2 on Yosemite

% setenv DYLD_PRINT_BINDINGS
% ./test_mathieu
dyld: bind: test_mathieu:0x103E32010 = libsystem_c.dylib:___stack_chk_guard,
*0x103E32010 = 0x7FFF7D8BF070
dyld: bind: test_mathieu:0x103E32000 = libdyld.dylib:dyld_stub_binder,
*0x103E32000 = 0x7FFF9B4612A0
dyld: resolver at 0x7fff9002924f returned 0x7FFF90029160
dyld: lazy bind: libsystem_c.dylib:0x7FFF7D8BB710 =
libsystem_platform.dylib:__platform_strchr, *0x7FFF7D8BB710 = 0x7FFF90029160
dyld: resolver at 0x7fff9002927a returned 0x7FFF9002D240
dyld: lazy bind: libsystem_c.dylib:0x7FFF7D8BB6E8 =
libsystem_platform.dylib:__platform_memmove, *0x7FFF7D8BB6E8 = 0x7FFF9002D240
dyld: resolver at 0x7fff9002927a returned 0x7FFF9002D240
dyld: lazy bind: libsystem_c.dylib:0x7FFF7D8BB6F0 =
libsystem_platform.dylib:__platform_memmove, *0x7FFF7D8BB6F0 = 0x7FFF9002D240
dyld: lazy bind: test_mathieu:0x103E32028 = libsystem_c.dylib:_printf,
*0x103E32028 = 0x7FFF8DA9C930
dyld: resolver at 0x7fff90028cad returned 0x7FFF90028B20
dyld: lazy bind: libsystem_c.dylib:0x7FFF7D8BB6D8 =
libsystem_platform.dylib:__platform_memchr, *0x7FFF7D8BB6D8 = 0x7FFF90028B20
loop -2.35802
loop 2
loop 2
loop 2
loop 2
dyld: resolver at 0x7fff90028e1a returned 0x7FFF9002DD10
dyld: lazy bind: test_mathieu:0x103E32018 = libsystem_platform.dylib:___bzero,
*0x103E32018 = 0x7FFF9002DD10

[Bug target/63661] -O2 miscompiles on OSX 10.10 Yosemite

[howarth at bromo dot med.uc.edu]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=63661

--- Comment #11 from howarth at bromo dot med.uc.edu ---
This issue isn't specific to Yosemite. I also see the failure at -O2 with
gcc-4.9.1 on darwin13.4.0 using the cctools (as and ld) from Xcode 6.

[Bug target/63661] -O2 miscompiles on OSX 10.10 Yosemite

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=63661

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Target|                            |x86_64-*-*
             Status|WAITING                     |NEW

--- Comment #12 from Richard Biener <rguenth at gcc dot gnu.org> ---
Reproduces with -O2 -fPIC -mtune=nehalem on x86_64-linux for me.

[Bug target/63661] [4.9/5 Regression] -O2 miscompiles on OSX 10.10 Yosemite

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=63661

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
      Known to work|                            |4.8.3
   Target Milestone|---                         |4.9.2
            Summary|-O2 miscompiles on OSX      |[4.9/5 Regression] -O2
                   |10.10 Yosemite              |miscompiles on OSX 10.10
                   |                            |Yosemite
      Known to fail|                            |4.9.1, 5.0

--- Comment #13 from Richard Biener <rguenth at gcc dot gnu.org> ---
4.8 doesn't know nehalem but works with -mtune=native (where 4.9 and 5 break).

[Bug target/63661] [4.9/5 Regression] -O2 miscompiles on OSX 10.10 Yosemite

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=63661

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

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

--- Comment #14 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
Started with r204698 , you can use -O2 -fpic -mtune=corei7 in compilers that
don't grok -mtune=nehalem.

[Bug target/63661] [4.9/5 Regression] -O2 miscompiles on OSX 10.10 Yosemite

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=63661

--- Comment #15 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
Slightly reduced testcase:
static void
foo (double a, double q, double xx, double *ff, double *gx, int e, int ni)
{
  int ii, nn;
  double g1 = *gx;
  ff[ni] = xx;
  if (e == 0)
    {
      for (ii = 0; ii < ni; ii++)
    {
      nn = 100 - ii - 1;
      ff[ni - ii - 1]
        = -1.0 / ((4 * (nn + 1) * (nn + 1) - a) / q + ff[ni - ii]);
    }
    }
  else
    {
      for (ii = 0; ii < ni; ii++)
    {
      nn = 100 - ii - 1;
      ff[ni - ii - 1]
        = -1.0 / (((2 * nn + 1) * (2 * nn + 1) - a) / q + ff[ni - ii]);
    }
    }
  *gx = ff[0] - g1;
}

int
bar (int order, double q, double a, double c[])
{
  int ni, nn, i, e;
  double eps, g1, g2, x1, x2, e1, e2, de, xh, s, ratio, ff[100];
  eps = 1e-10;
  c[0] = 1.0;
  e = 0;
  if (order % 2 != 0)
    e = 1;
  if (q == 0.0)
    {
      for (i = 0; i < 100; i++)
    c[i] = 0.0;
      c[(order - 1) / 2] = 1.0;
      return 0;
    }
  if (order < 5)
    {
      nn = 0;
      s = 0.0;
      if (e == 0)
    ratio = (a - 4) / q;
      else
    ratio = (a - 1 - q) / q;
    }
  else
    {
      if (e == 0)
    {
      c[1] = (a - 4) / q;
      s = 2 * c[0] + 4 * c[1];
      for (i = 2; i < order / 2; i++)
        {
          c[i] = (a - 4 * i * i) / q * c[i - 1] - c[i - 2];
          s += 2 * (i + 1) * c[i];
        }
    }
      else
    {
      c[1] = (a - 1) / q + 1;
      s = c[0] + 3 * c[1];
      for (i = 2; i < order / 2 + 1; i++)
        {
          c[i] = (a - (2 * i - 1) * (2 * i - 1)) / q * c[i - 1] - c[i - 2];
          s += (2 * (i + 1) - 1) * c[i];
        }
    }
      nn = i - 1;
      ratio = c[nn] / c[nn - 1];
    }
  ni = 100 - nn - 1;
  if (e == 0)
    x1 = -q / (4.0 * (100 + 1.0) * (100 + 1.0));
  else
    x1 = -q / ((2.0 * 100 + 1.0) * (2.0 * 100 + 1.0));
  g1 = ratio;
  foo (a, q, x1, ff, &g1, e, ni);
  x2 = g1;
  g2 = ratio;
  while (1)
    {
      __builtin_printf ("loop %g\n", g2);
      e1 = g1 - x1;
      e2 = g2 - x2;
      de = e1 - e2;
      if (__builtin_fabs (de) < eps)
    break;
      xh = (e1 * x2 - e2 * x1) / de;
      x1 = x2;
      g1 = g2;
      x2 = xh;
      g2 = ratio;
      foo (a, q, x2, ff, &g2, e, ni);
    }
  s += 2 * (nn + 1) * c[nn];
  for (i = nn + 1; i < 100; i++)
    {
      c[i] = ff[i - nn - 1] * c[i - 1];
      s += 2 * (i + 1) * c[i];
      if (__builtin_fabs (c[i]) < 1e-20)
    for (; i < 100;)
      c[i++] = 0.0;
    }
  for (i = 0; i < 100; i++)
    c[i] /= s;
  return 0;
}

int
main ()
{
  double c[100];
  bar (1, 5.0, -5.790080598637771, c);
  return 0;
}

[Bug target/63661] [4.9/5 Regression] -O2 miscompiles on OSX 10.10 Yosemite

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=63661

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|4.9.2                       |4.9.3

--- Comment #16 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
GCC 4.9.2 has been released.

[Bug target/63661] [4.9/5 Regression] -O2 miscompiles on OSX 10.10 Yosemite

[vbraun.name at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=63661

Volker Braun <vbraun.name at gmail dot com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
      Known to fail|                            |4.9.2

--- Comment #17 from Volker Braun <vbraun.name at gmail dot com> ---
Bug is still in gcc 4.9.2

[Bug target/63661] [4.9/5 Regression] -O2 miscompiles on OSX 10.10 Yosemite

[jamborm at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=63661

--- Comment #18 from Martin Jambor <jamborm at gcc dot gnu.org> ---
(In reply to Jakub Jelinek from comment #14)
> Started with r204698 , you can use -O2 -fpic -mtune=corei7 in compilers that
> don't grok -mtune=nehalem.

So it's probably mine.  I will take a look but please be patient, it
may take a week or probably two.

[Bug target/63661] [4.9/5 Regression] -O2 miscompiles on OSX 10.10 Yosemite

[mednafen at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=63661

mednafen at gmail dot com changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |mednafen at gmail dot com

--- Comment #19 from mednafen at gmail dot com ---
Created attachment 33870
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=33870&action=edit
Simplified testcase.

Simplified testcase, tested to be broken on x86_64 Linux.

$ /usr/local/gcc-4.9.2/bin/gcc -O2 -fpic -o testcase-alt testcase-alt.c &&
./testcase-alt 
A: 0x00007fffd8416f60
B: 0x4014000000000000 ---- 5


$ /usr/local/gcc-4.9.2/bin/gcc -O2 -fpic -mtune=nehalem -o testcase-alt
testcase-alt.c && ./testcase-alt 
A: 0x00007fffc9fb1730
B: 0x00007fffc9fb1730 ---- 6.95331e-310
Aborted

[Bug target/63661] [4.9/5 Regression] -O2 miscompiles with -mtune=nehalem or corei7

[jamborm at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=63661

--- Comment #20 from Martin Jambor <jamborm at gcc dot gnu.org> ---
So had a look at what my tiny bit in IRA does with this testcase and
while it certainly triggers the bug, I don't think it causes it.  It
performs exactly the same modifications as when -mtune=nehalem is not
specified on the command line.

The functions I introduced split life-ranges of three registers, an SI
mode one, DF mode one and a DI mode one.  The bug occurs when both the
latter two (DF and DI mode) are split.

One suspicious thing is that in the nehalem case, the DF register is
assigned to r12 (as opposed to a stack spill in the non-nehalem case)
but I think that register is killed before being used, and
(coincidentally?) it is being overwritten by the value from the
DI-life-split register (r15) - in what seems like another effort of
IRA to split their life-range but with one crucial instruction being
lost (as in not present after reload).

I'm afraid I'll need help of someone who knows IRA, Vladimir, could
you please have a look at what happens in function bar during IRA and
reload?

[Bug target/63661] [4.9/5 Regression] -O2 miscompiles with -mtune=nehalem or corei7

[hjl.tools at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=63661

--- Comment #21 from H.J. Lu <hjl.tools at gmail dot com> ---
It was fixed by r217783 on trunk and the fix also works
on 4.9 branch.  Is is the real fix?

[Bug target/63661] [4.9/5 Regression] -O2 miscompiles with -mtune=nehalem or corei7

[jamborm at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=63661

Martin Jambor <jamborm at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |renlin.li at arm dot com

--- Comment #22 from Martin Jambor <jamborm at gcc dot gnu.org> ---
(In reply to H.J. Lu from comment #21)
> It was fixed by r217783 on trunk and the fix also works
> on 4.9 branch.  Is is the real fix?

I am not familiar enough with IRA to be really sure but I would say
that yes.  This bug really looks as if IRA and LRA chose the same
register (r12) for a different thing, which is described as the
problem in the email accompanying the patch.  The weird sequence
clobbering the register certainly goes away with just this patch
applied.

So let's wait for a week or so if any of the RA people confirm this or
have any comments and prepare a backport if they don't.

[Bug target/63661] [4.9/5 Regression] -O2 miscompiles with -mtune=nehalem or corei7

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=63661

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Keywords|                            |wrong-code
           Priority|P3                          |P2

[Bug target/63661] [4.9/5 Regression] -O2 miscompiles with -mtune=nehalem or corei7

[renlin.li at arm dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=63661

--- Comment #23 from Renlin Li <renlin.li at arm dot com> ---
I am using x86_64-unknown-linux-gnu-gcc toolchain.

Reproduce the issue in 4.9 branch with "-O2 -fPIC -mtune=native" options.

With 4.9.2 release, I reproduced the problem with "-O2 -fPIC -mtune=nehalem"
options.

I have verified that, after my patch(r217783), both cases are fixed.

Issue is not there anymore in the latest trunk, it is presumably fixed by my
patch already.

[Bug target/63661] [4.9/5 Regression] -O2 miscompiles with -mtune=nehalem or corei7

[renlin at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=63661

--- Comment #24 from renlin at gcc dot gnu.org ---
Author: renlin
Date: Fri Nov 28 11:01:27 2014
New Revision: 218143

URL: https://gcc.gnu.org/viewcvs?rev=218143&root=gcc&view=rev
Log:
Add testcase for PR63661.

2014-11-28  Renlin Li  <renlin.li@arm.com>

    PR target/63661
    * gcc.target/i386/pr63661.c: New test.

Added:
    trunk/gcc/testsuite/gcc.target/i386/pr63661.c
Modified:
    trunk/gcc/testsuite/ChangeLog

[Bug target/63661] [4.9/5 Regression] -O2 miscompiles with -mtune=nehalem or corei7

[renlin at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=63661

--- Comment #25 from renlin at gcc dot gnu.org ---
Author: renlin
Date: Fri Nov 28 11:18:47 2014
New Revision: 218144

URL: https://gcc.gnu.org/viewcvs?rev=218144&root=gcc&view=rev
Log:
Use native tune. nehalem is not able to triggle the issue in trunk any more.

2014-11-28  Renlin Li  <renlin.li@arm.com>

    PR  target/63661
    * gcc.target/i386/pr63661.c: Use native tune.

Modified:
    trunk/gcc/testsuite/ChangeLog
    trunk/gcc/testsuite/gcc.target/i386/pr63661.c

[Bug target/63661] [4.9/5 Regression] -O2 miscompiles with -mtune=nehalem or corei7

[vbraun.name at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=63661

--- Comment #26 from Volker Braun <vbraun.name at gmail dot com> ---
I can confirm that r217783 fixes the original issue on in the GSL testsuite on
OSX Yosemite, too.

[Bug target/63661] [4.9 Regression] -O2 miscompiles with -mtune=nehalem or corei7

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=63661

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
      Known to work|                            |5.0
            Summary|[4.9/5 Regression] -O2      |[4.9 Regression] -O2
                   |miscompiles with            |miscompiles with
                   |-mtune=nehalem or corei7    |-mtune=nehalem or corei7
      Known to fail|5.0                         |

--- Comment #27 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
Thus fixed on the trunk so far.

[Bug target/63661] [4.9 Regression] -O2 miscompiles with -mtune=nehalem or corei7

[jamborm at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=63661

Martin Jambor <jamborm at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|NEW                         |ASSIGNED
           Assignee|unassigned at gcc dot gnu.org      |renlin.li at arm dot com

--- Comment #28 from Martin Jambor <jamborm at gcc dot gnu.org> ---
In https://gcc.gnu.org/ml/gcc-patches/2014-11/msg02502.html Renlin
said he wanted to prepare the backport himself so I'll take the
liberty of assigning the bug to him.  Please close this bug once it
hits the 4.9 branch.  Thanks a lot.

[Bug target/63661] [4.9 Regression] -O2 miscompiles with -mtune=nehalem or corei7

[renlin at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=63661

--- Comment #29 from renlin at gcc dot gnu.org ---
Author: renlin
Date: Wed Dec  3 11:13:50 2014
New Revision: 218306

URL: https://gcc.gnu.org/viewcvs?rev=218306&root=gcc&view=rev
Log:
Backported from mainline

gcc/

2014-12-03  Renlin Li  <Renlin.Li@arm.com>

    PR middle-end/63762
    PR target/63661
    * ira.c (i386ra): Update preferred class.

gcc/testsuite/

2014-12-03  Renlin Li  <Renlin.Li@arm.com>
        H.J. Lu <hongjiu.lu@intel.com>

    PR middle-end/63762
    PR target/63661
    * gcc.dg/pr63762.c: New test.
    * gcc.target/i386/pr63661.c: New test.

Added:
    branches/gcc-4_9-branch/gcc/testsuite/gcc.dg/pr63762.c
    branches/gcc-4_9-branch/gcc/testsuite/gcc.target/i386/pr63661.c
Modified:
    branches/gcc-4_9-branch/gcc/ChangeLog
    branches/gcc-4_9-branch/gcc/ira.c
    branches/gcc-4_9-branch/gcc/testsuite/ChangeLog

[Bug target/63661] [4.9 Regression] -O2 miscompiles with -mtune=nehalem or corei7

[renlin.li at arm dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=63661

Renlin Li <renlin.li at arm dot com> changed:

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

--- Comment #30 from Renlin Li <renlin.li at arm dot com> ---
4.9 branch is fixed by r218306
trunk is fixed by r217783