[Bug tree-optimization/58984] New: wrong code at -Os and above on x86_64-linux-gnu in 64-bit mode

[Bug tree-optimization/58984] New: wrong code at -Os and above on x86_64-linux-gnu in 64-bit mode

[su at cs dot ucdavis.edu]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=58984

            Bug ID: 58984
           Summary: wrong code at -Os and above on x86_64-linux-gnu in
                    64-bit mode
           Product: gcc
           Version: 4.9.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: tree-optimization
          Assignee: unassigned at gcc dot gnu.org
          Reporter: su at cs dot ucdavis.edu

The current gcc trunk and 4.8.x miscompile the following code on x86_64-linux
at -Os and above in 64-bit mode. 

This is a regression from 4.7.x. 

$ gcc-trunk -v
Using built-in specs.
COLLECT_GCC=gcc-trunk
COLLECT_LTO_WRAPPER=/usr/local/gcc-trunk/libexec/gcc/x86_64-unknown-linux-gnu/4.9.0/lto-wrapper
Target: x86_64-unknown-linux-gnu
Configured with: ../gcc-trunk/configure
--enable-languages=c,c++,objc,obj-c++,fortran,lto --disable-werror
--enable-checking=release --with-gmp=/usr/local/gcc-trunk
--with-mpfr=/usr/local/gcc-trunk --with-mpc=/usr/local/gcc-trunk
--with-cloog=/usr/local/gcc-trunk --prefix=/usr/local/gcc-trunk
Thread model: posix
gcc version 4.9.0 20131103 (experimental) [trunk revision 204333] (GCC) 
$ 
$ gcc-trunk -O1 small.c; a.out
1
$ gcc-trunk -Os small.c; a.out
0
$ gcc-4.8.2 -Os small.c; a.out
0
$ gcc-4.7.3 -Os small.c; a.out
1
$ 

--------------------------------

int printf (const char *, ...);

struct S
{
  int f0:8;
  int:6;
  int f1:5;
};

int a, *c = &a, e, n, b;

static int
foo (struct S p)
{
  const unsigned short *f[36];
  for (; e < 2; e++)
    {
      const unsigned short **i = &f[0];
      *c ^= 1;
      if (p.f1)
    {
      *i = 0;
      return b;
    }
    }
  return 0;
}

int
main ()
{
  struct S o = { 1, 1 };
  foo (o);
  n || o.f0;
  printf ("%d\n", a);
  return 0;
}

[Bug tree-optimization/58984] [4.8/4.9 Regression] wrong code at -Os and above on x86_64-linux-gnu in 64-bit mode

[mpolacek at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=58984

Marek Polacek <mpolacek at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
   Last reconfirmed|                            |2013-11-04
                 CC|                            |mpolacek at gcc dot gnu.org
            Summary|wrong code at -Os and above |[4.8/4.9 Regression] wrong
                   |on x86_64-linux-gnu in      |code at -Os and above on
                   |64-bit mode                 |x86_64-linux-gnu in 64-bit
                   |                            |mode
     Ever confirmed|0                           |1

--- Comment #1 from Marek Polacek <mpolacek at gcc dot gnu.org> ---
Confirmed.

[Bug tree-optimization/58984] [4.8/4.9 Regression] wrong code at -Os and above on x86_64-linux-gnu in 64-bit mode

[mpolacek at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=58984

Marek Polacek <mpolacek at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|---                         |4.8.4

[Bug tree-optimization/58984] [4.8/4.9 Regression] wrong code at -Os and above on x86_64-linux-gnu in 64-bit mode

[jakub at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=58984

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Last reconfirmed|2013-11-04 00:00:00         |
                 CC|                            |jakub at gcc dot gnu.org,
                   |                            |jamborm at gcc dot gnu.org
   Target Milestone|4.8.4                       |4.8.3

--- Comment #2 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
Started with r193298.

[Bug tree-optimization/58984] [4.8/4.9 Regression] wrong code at -Os and above on x86_64-linux-gnu in 64-bit mode

[jakub at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=58984

--- Comment #3 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
So, before IPA-CP, foo had:
  _10 = BIT_FIELD_REF <p, 32, 0>;
  _11 = _10 & 507904;
Now, IPA-CP does:
Modification phase of node foo.constprop.0/3
     Aggregate replacements: 0[14]=1, 0[8]=0, 0[0]=1
  _4 = 1;
  _5 = _4 & 507904;
which is of course wrong, because 32 bits at offset 0 in the structure should
contain value 16385 ((1 << 0) | (1 << 14)), rather than 1.

[Bug tree-optimization/58984] [4.8/4.9 Regression] wrong code at -Os and above on x86_64-linux-gnu in 64-bit mode

[jakub at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=58984

--- Comment #4 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
Note that likely since r199252 Aggregate replacements: 0[14]=1, 0[8]=0, 0[0]=1
is replaced with just 0[0]=1, still the effect is exactly the same.

[Bug tree-optimization/58984] [4.8/4.9 Regression] wrong code at -Os and above on x86_64-linux-gnu in 64-bit mode

[jakub at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=58984

--- Comment #5 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
Ah, the reason why r199252 doesn't fix this is that esra changes o.f0 = 1;
into a MEM_REF, essentially *(char *)&o = 1, because the f0 has 8 bits.
So, determine_known_aggregate_parts doesn't consider the o.f0 = 1 store as
bitfield access.
If in the testcase int f0:8 is replaced with char f0, then it is miscompiled
even at -O2, and then of course o.f0 = 1 is not a bitfield, but still, it
overlaps with the bit field representative of the other bitfields.
So, either the replacement code really needs to be taught about bitfields
anyway
(at least look not just at offset, but also size of the known value and not
optimize into a constant if the access is to some bits beyond the known value
extent), or the punting in determine_known_aggregate_parts would need to be
much more conservative.

[Bug tree-optimization/58984] [4.8/4.9 Regression] wrong code at -Os and above on x86_64-linux-gnu in 64-bit mode

[jamborm at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=58984

--- Comment #7 from Martin Jambor <jamborm at gcc dot gnu.org> ---
As far as I'm concerned, the patch is fine.  Thanks for taking care of it.

[Bug tree-optimization/58984] [4.8/4.9 Regression] wrong code at -Os and above on x86_64-linux-gnu in 64-bit mode

[jakub at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=58984

--- Comment #8 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
Well, there is if (*size_p) which should have been if (size_p) obviously,
redoing bootstrap now with that fix.

[Bug tree-optimization/58984] [4.8/4.9 Regression] wrong code at -Os and above on x86_64-linux-gnu in 64-bit mode

[jakub at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=58984

--- Comment #9 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
Author: jakub
Date: Tue Nov  5 12:05:37 2013
New Revision: 204385

URL: http://gcc.gnu.org/viewcvs?rev=204385&root=gcc&view=rev
Log:
    PR tree-optimization/58984
    * ipa-prop.c (ipa_load_from_parm_agg_1): Add SIZE_P argument,
    set *SIZE_P if non-NULL on success.
    (ipa_load_from_parm_agg, ipa_analyze_indirect_call_uses): Adjust
    callers.
    (ipcp_transform_function): Likewise.  Punt if size of access
    is different from TYPE_SIZE on v->value's type.

    * gcc.c-torture/execute/pr58984.c: New test.

Added:
    trunk/gcc/testsuite/gcc.c-torture/execute/pr58984.c
Modified:
    trunk/gcc/ChangeLog
    trunk/gcc/ipa-prop.c
    trunk/gcc/testsuite/ChangeLog

[Bug tree-optimization/58984] [4.8/4.9 Regression] wrong code at -Os and above on x86_64-linux-gnu in 64-bit mode

[jakub at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=58984

--- Comment #10 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
Author: jakub
Date: Tue Nov  5 12:11:17 2013
New Revision: 204387

URL: http://gcc.gnu.org/viewcvs?rev=204387&root=gcc&view=rev
Log:
    PR tree-optimization/58984
    * ipa-prop.c (ipa_load_from_parm_agg_1): Add SIZE_P argument,
    set *SIZE_P if non-NULL on success.
    (ipa_load_from_parm_agg, ipa_analyze_indirect_call_uses): Adjust
    callers.
    (ipcp_transform_function): Likewise.  Punt if size of access
    is different from TYPE_SIZE on v->value's type.

    * gcc.c-torture/execute/pr58984.c: New test.

Added:
    branches/gcc-4_8-branch/gcc/testsuite/gcc.c-torture/execute/pr58984.c
Modified:
    branches/gcc-4_8-branch/gcc/ChangeLog
    branches/gcc-4_8-branch/gcc/ipa-prop.c
    branches/gcc-4_8-branch/gcc/testsuite/ChangeLog

[Bug tree-optimization/58984] [4.8/4.9 Regression] wrong code at -Os and above on x86_64-linux-gnu in 64-bit mode

[law at redhat dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=58984

Jeffrey A. Law <law at redhat dot com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|ASSIGNED                    |RESOLVED
                 CC|                            |law at redhat dot com
         Resolution|---                         |FIXED

--- Comment #11 from Jeffrey A. Law <law at redhat dot com> ---
Fixed on trunk.