[Bug tree-optimization/37868] New: code that doesn't breaks TBAA is misoptimized even with -fno-strict-aliasing

[Bug tree-optimization/37868] New: code that doesn't breaks TBAA is misoptimized even with -fno-strict-aliasing

[edwintorok at gmail dot com]

[-- Warning: decoded text below may be mangled, UTF-8 assumed --]
[-- Attachment #1: Type: text/plain, Size: 1459 bytes --]

+++ This bug was initially created as a clone of Bug #37867 +++

Cloning bug on Chris's request.

Running gcc-4.3 -O2 on PR1386.c I get different results than with -O1.

$ gcc-4.3 -O1 PR1386.c && ./a.out
PR1386.c: In function ‘main’:
PR1386.c:15: warning: large integer implicitly truncated to unsigned type
PR1386.c:17: warning: large integer implicitly truncated to unsigned type
bad bits: 0

$ gcc-4.3 -O2 PR1386.c && ./a.out
PR1386.c: In function ‘main’:
PR1386.c:15: warning: large integer implicitly truncated to unsigned type
PR1386.c:17: warning: large integer implicitly truncated to unsigned type
bad bits: fffffffffffffff0

$ g++-4.3 -O2 -fno-strict-aliasing x.c
x.c: In function ‘int main()’:
x.c:15: warning: large integer implicitly truncated to unsigned type
x.c:17: warning: large integer implicitly truncated to unsigned type
edwin@debian:~$ ./a.out
bad bits: fffffffffffffff0


-- 
           Summary: code that doesn't breaks TBAA is misoptimized even with
                    -fno-strict-aliasing
           Product: gcc
           Version: 4.3.2
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: tree-optimization
        AssignedTo: unassigned at gcc dot gnu dot org
        ReportedBy: edwintorok at gmail dot com
 GCC build triplet: x86_64-linux-gnu
  GCC host triplet: x86_64-linux-gnu
GCC target triplet: x86_64-linux-gnu


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=37868

[Bug tree-optimization/37868] code that breaks TBAA is misoptimized even with -fno-strict-aliasing

[sabre at nondot dot org]

------- Comment #1 from sabre at nondot dot org  2008-10-18 18:29 -------
Here is the testcase.  I should work with -fno-strict-aliasing:

#include <stdio.h>
#include <stdint.h>

struct X {
  unsigned char pad : 4;
  uint64_t a : 64;
  uint64_t b : 60;
} __attribute__((packed));

int main (void)
{
  struct X x;
  uint64_t bad_bits;

  x.pad = 255;
  x.a = -1ULL;
  x.b = -1ULL;

  bad_bits = ((uint64_t)-1ULL) ^ *(1+(uint64_t *) &x);
  printf("bad bits: %llx\n", bad_bits);
  return bad_bits != 0;
}


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=37868

[Bug tree-optimization/37868] code that breaks TBAA is misoptimized even with -fno-strict-aliasing

[rguenth at gcc dot gnu dot org]

------- Comment #2 from rguenth at gcc dot gnu dot org  2008-10-18 18:57 -------
I don't see why it _should_ work with -fno-strict-aliasing.  It may.  Note that
this testcase also violates pointer arithmetic rules - you have a pointer that
doesn't point to a valid object.

On the 4.3 branch we still optimize it because we use SFTs for the individual
elements and offset disambiguation selects only one.  On the trunk PTA seems
to prune the symbol.


-- 

rguenth at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |rguenth at gcc dot gnu dot
                   |                            |org


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=37868

[Bug tree-optimization/37868] code that breaks TBAA is misoptimized even with -fno-strict-aliasing

[rguenth at gcc dot gnu dot org]

------- Comment #3 from rguenth at gcc dot gnu dot org  2008-10-18 22:34 -------
It works on the trunk for me with -fno-strict-aliasing, with -fstrict-aliasing
TBAA pruning prunes the points-to set.


-- 

rguenth at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
      Known to fail|                            |4.3.2
      Known to work|                            |4.4.0


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=37868

[Bug tree-optimization/37868] code that breaks TBAA is misoptimized even with -fno-strict-aliasing

[rguenth at gcc dot gnu dot org]

------- Comment #4 from rguenth at gcc dot gnu dot org  2008-10-18 22:42 -------
4.3 is broken because PTA figures

D.1187_2 = { x.a }

from the constraints

x.0_1 = &x
D.1187_2 = x.0_1 + 64

because the 4.3 branch doesn't have the PTA fixes, in particular

          /* If the found variable is not exactly at the pointed to
             result, we have to include the next variable in the
             solution as well.  Otherwise two increments by offset / 2
             do not result in the same or a conservative superset
             solution.  */
          if (temp->offset != curr->offset + rhsoffset
              && temp->next != NULL)
            {
              struct constraint_expr c2;
              c2.var = temp->next->id;
              c2.type = ADDRESSOF;
              c2.offset = 0;
              VEC_safe_push (ce_s, heap, *results, &c2);
            }

see rev. 137573:

 2008-07-07  Richard Guenther  <rguenther@suse.de>

+       * tree-ssa-structalias.c (struct variable_info): Add is_full_var flag.
+       (new_var_info): Set it to false.
+       (solution_set_add): Correctly handle pointers outside a var and
+       inside a field.
+       (type_safe): Treat variables with is_full_var properly.
+       (do_sd_constraint): Likewise.
+       (do_ds_constraint): Likewise.
+       (process_constraint): Remove zeroing offset for !use_field_sensitive.
+       (get_constraint_for_ptr_offset): New function.
+       (get_constraint_for_component_ref): For addresses at least include
+       the last field of the variable.  Handle is_full_vars properly.
+       (get_constraint_for_1): Factor common code, handle POINTER_PLUS_EXPR.
+       (handle_ptr_arith): Remove.
+       (find_func_aliases): Simplify assignment handling.
+       (create_function_info_for): For parameter and result varinfos set
+       is_full_var flag.
+       (create_variable_info_for): Set is_full_var flag whenever we
+       just created a single varinfo for a decl.
+       (init_alias_vars): Initialize use_field_sensitive from
+       max-fields-for-field-sensitive parameter.


-- 

rguenth at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
     Ever Confirmed|0                           |1
           Keywords|                            |wrong-code
   Last reconfirmed|0000-00-00 00:00:00         |2008-10-18 22:42:54
               date|                            |


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=37868

[Bug tree-optimization/37868] [4.3 Regression] code that breaks TBAA is misoptimized even with -fno-strict-aliasing

[rguenth at gcc dot gnu dot org]

------- Comment #5 from rguenth at gcc dot gnu dot org  2008-10-22 11:46 -------
4.2 works, so this is a regression.  Mine.


-- 

rguenth at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
         AssignedTo|unassigned at gcc dot gnu   |rguenth at gcc dot gnu dot
                   |dot org                     |org
             Status|NEW                         |ASSIGNED
      Known to work|4.4.0                       |4.2.4 4.4.0
   Last reconfirmed|2008-10-18 22:42:54         |2008-10-22 11:46:45
               date|                            |
            Summary|code that breaks TBAA is    |[4.3 Regression] code that
                   |misoptimized even with -fno-|breaks TBAA is misoptimized
                   |strict-aliasing             |even with -fno-strict-
                   |                            |aliasing
   Target Milestone|---                         |4.3.3


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=37868

[Bug tree-optimization/37868] [4.3 Regression] code that breaks TBAA is misoptimized even with -fno-strict-aliasing

[rguenth at gcc dot gnu dot org]

------- Comment #6 from rguenth at gcc dot gnu dot org  2008-10-22 13:49 -------
Backporting the fix turns out to be tricky...


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=37868

[Bug tree-optimization/37868] [4.3 Regression] code that breaks TBAA is misoptimized even with -fno-strict-aliasing

[jakub at gcc dot gnu dot org]

-- 

jakub at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Priority|P3                          |P2


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=37868

[Bug tree-optimization/37868] [4.3 Regression] code that breaks TBAA is misoptimized even with -fno-strict-aliasing

[rguenth at gcc dot gnu dot org]

------- Comment #7 from rguenth at gcc dot gnu dot org  2008-11-07 12:50 -------
*** Bug 38048 has been marked as a duplicate of this bug. ***


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=37868

[Bug tree-optimization/37868] [4.3 Regression] code that breaks TBAA is misoptimized even with -fno-strict-aliasing

[rguenth at gcc dot gnu dot org]

------- Comment #8 from rguenth at gcc dot gnu dot org  2008-11-20 12:13 -------
Subject: Bug 37868

Author: rguenth
Date: Thu Nov 20 12:12:01 2008
New Revision: 142040

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=142040
Log:
2008-11-20  Richard Guenther  <rguenther@suse.de>

        PR tree-optimization/37868
        * tree-ssa-structalias.c (set_uids_in_ptset): Add SFTs based on
        pointed to variable and access size.

        * gcc.dg/torture/pr37868.c: New testcase.
        * gcc.c-torture/execute/pr38048-1.c: Likewise.
        * gcc.c-torture/execute/pr38048-2.c: Likewise.

        Backport from mainline:
        2008-07-07  Richard Guenther  <rguenther@suse.de>

        * tree-ssa-structalias.c (struct variable_info): Add is_full_var flag.
        (new_var_info): Set it to false.
        (solution_set_add): Correctly handle pointers outside a var and
        inside a field.
        (type_safe): Treat variables with is_full_var properly.
        (do_sd_constraint): Likewise.
        (do_ds_constraint): Likewise.
        (process_constraint): Remove zeroing offset for !use_field_sensitive.
        (get_constraint_for_ptr_offset): New function.
        (get_constraint_for_component_ref): Handle is_full_vars properly.
        (get_constraint_for): Handle POINTER_PLUS_EXPR.
        (handle_ptr_arith): Remove.
        (find_func_aliases): Handle POINTER_PLUS_EXPR through generic
        get_constraint_for code.
        (create_function_info_for): For parameter and result varinfos set
        is_full_var flag.
        (create_variable_info_for): Set is_full_var flag whenever we
        just created a single varinfo for a decl.
        (init_alias_vars): Initialize use_field_sensitive from
        max-fields-for-field-sensitive parameter.

        * gcc.dg/torture/pta-ptrarith-1.c: New testcase.
        * gcc.dg/torture/pta-ptrarith-2.c: Likewise.

Added:
    branches/gcc-4_3-branch/gcc/testsuite/gcc.c-torture/execute/pr38048-1.c
    branches/gcc-4_3-branch/gcc/testsuite/gcc.c-torture/execute/pr38048-2.c
    branches/gcc-4_3-branch/gcc/testsuite/gcc.dg/torture/pr37868.c
    branches/gcc-4_3-branch/gcc/testsuite/gcc.dg/torture/pta-ptrarith-1.c
    branches/gcc-4_3-branch/gcc/testsuite/gcc.dg/torture/pta-ptrarith-2.c
Modified:
    branches/gcc-4_3-branch/gcc/ChangeLog
    branches/gcc-4_3-branch/gcc/testsuite/ChangeLog
    branches/gcc-4_3-branch/gcc/tree-ssa-structalias.c


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=37868

[Bug tree-optimization/37868] [4.3 Regression] code that breaks TBAA is misoptimized even with -fno-strict-aliasing

[rguenth at gcc dot gnu dot org]

------- Comment #9 from rguenth at gcc dot gnu dot org  2008-11-20 12:14 -------
Fixed.


-- 

rguenth at gcc dot gnu dot org changed:

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


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=37868

[Bug tree-optimization/37868] [4.3 Regression] code that breaks TBAA is misoptimized even with -fno-strict-aliasing

[rguenth at gcc dot gnu dot org]

------- Comment #10 from rguenth at gcc dot gnu dot org  2008-11-20 12:26 -------
Subject: Bug 37868

Author: rguenth
Date: Thu Nov 20 12:25:26 2008
New Revision: 142041

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=142041
Log:
2008-11-20  Richard Guenther  <rguenther@suse.de>

        PR tree-optimization/37868
        * gcc.dg/torture/pr37868.c: New testcase.
        * gcc.c-torture/execute/pr38048-1.c: Likewise.
        * gcc.c-torture/execute/pr38048-2.c: Likewise.

Added:
    trunk/gcc/testsuite/gcc.c-torture/execute/pr38048-1.c
    trunk/gcc/testsuite/gcc.c-torture/execute/pr38048-2.c
    trunk/gcc/testsuite/gcc.dg/torture/pr37868.c
Modified:
    trunk/gcc/testsuite/ChangeLog


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=37868