[Bug c++/46309] New: optimization a==3||a==1

[Bug c++/46309] New: optimization a==3||a==1

[eidletni at mail dot ru]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46309

           Summary: optimization a==3||a==1
           Product: gcc
           Version: 4.6.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: c++
        AssignedTo: unassigned@gcc.gnu.org
        ReportedBy: eidletni@mail.ru


$g++ -v
Using built-in specs.
COLLECT_GCC=g++
COLLECT_LTO_WRAPPER=/home/eid-letni/opt/gcc/libexec/gcc/i686-pc-linux-gnu/4.6.0/lto-wrapper
Target: i686-pc-linux-gnu
Configured with: ../configure --prefix=/home/eid-letni/opt/gcc
--enable-languages=c,c++
Thread model: posix
gcc version 4.6.0 20101105 (experimental) (GCC) 

$uname -a
Linux eidletni 2.6.35-22-generic #35-Ubuntu SMP Sat Oct 16 20:36:48 UTC 2010
i686 GNU/Linux

$g++ -O3 -c main.ii

func good_comp have optimiz without jump:

bool good_comp(int a, int b)
{
    return (a==3)||(a==1);
}

asm with no jumps:

00000020 <_Z9good_compii>:
  20:    8b 54 24 04              mov    0x4(%esp),%edx
  24:    83 fa 01                 cmp    $0x1,%edx
  27:    0f 94 c0                 sete   %al
  2a:    83 fa 03                 cmp    $0x3,%edx
  2d:    0f 94 c2                 sete   %dl
  30:    09 d0                    or     %edx,%eax
  32:    c3                       ret    

using "same" code with temps:

bool bad_comp(int a, int b)
{
    bool tmp1 = (a==3);
    bool tmp2 = (a==1);
    return tmp1||tmp2;
}

asm(with jump):

00000000 <_Z8bad_compii>:
   0:    8b 54 24 04              mov    0x4(%esp),%edx
   4:    b8 01 00 00 00           mov    $0x1,%eax
   9:    83 fa 03                 cmp    $0x3,%edx
   c:    74 06                    je     14 <_Z8bad_compii+0x14>
   e:    83 fa 01                 cmp    $0x1,%edx
  11:    0f 94 c0                 sete   %al
  14:    f3 c3                    repz ret

[Bug c++/46309] optimization a==3||a==1

[eidletni at mail dot ru]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46309

--- Comment #1 from eidletni at mail dot ru 2010-11-05 08:36:58 UTC ---
Created attachment 22290
  --> http://gcc.gnu.org/bugzilla/attachment.cgi?id=22290
c++ code

[Bug c++/46309] optimization a==3||a==1

[svfuerst at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46309

Steven Fuerst <svfuerst at gmail dot com> changed:

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

--- Comment #2 from Steven Fuerst <svfuerst at gmail dot com> 2010-11-06 18:55:48 UTC ---
This is even better, for this particular case.

mov        0x4(%esp),%eax
and        $-3, %eax
cmp        $1, %eax
sete%al
ret

[Bug c++/46309] optimization a==3||a==1

[paolo.carlini at oracle dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46309

Paolo Carlini <paolo.carlini at oracle dot com> changed:

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

--- Comment #3 from Paolo Carlini <paolo.carlini at oracle dot com> 2011-09-23 22:40:32 UTC ---
If anything, doesn't seem a C++ issue... Richard can you have a look?

[Bug tree-optimization/46309] optimization a==3||a==1

[jakub at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46309

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |jakub at gcc dot gnu.org
          Component|c++                         |tree-optimization

--- Comment #4 from Jakub Jelinek <jakub at gcc dot gnu.org> 2011-09-24 08:36:27 UTC ---
With s/bool/int/ we generate identical code for both, without jumps.
Anyway, more important is that we don't seem to optimize the
x == CST1 || x == CST2 where __builtin_popcount (CST1 ^ CST2) == 1
in fold-const.c and don't even optimize range tests if the constants alternate
and, more importantly, don't seem to do anything similar to fold_range_test at
the GIMPLE level.

int
f1 (int a, int b)
{
  int v1 = (a == 3);
  int v2 = (a == 1);
  int v3 = (a == 4);
  int v4 = (a == 2);
  return v1 || v2 || v3 || v4;
}

int
f2 (int a, int b)
{
  int v1 = (a == 1);
  int v2 = (a == 2);
  int v3 = (a == 3);
  int v4 = (a == 4);
  return v1 || v2 || v3 || v4;
}

int
f3 (int a, int b)
{
  int v1 = (a == 3);
  int v2 = (a == 1);
  return v1 || v2;
}

int
f4 (int a, int b)
{
  int v1 = (a == 1);
  int v2 = (a == 2);
  return v1 || v2;
}

int
f5 (int a, int b)
{
  return a == 3 || a == 1 || a == 4 || a == 2;
}

int
f6 (int a, int b)
{
  return a == 1 || a == 2 || a == 3 || a == 4;
}

int
f7 (int a, int b)
{
  return a == 3 || a == 1;
}

int
f8 (int a, int b)
{
  return a == 1 || a == 2;
}

int
f9 (int a, int b)
{
  return a == 3 || a == 1 || a == 2 || a == 4;
}

[Bug tree-optimization/46309] optimization a==3||a==1

[jakub at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46309

--- Comment #5 from Jakub Jelinek <jakub at gcc dot gnu.org> 2011-09-30 15:00:18 UTC ---
Author: jakub
Date: Fri Sep 30 15:00:12 2011
New Revision: 179388

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=179388
Log:
    PR tree-optimization/46309
    * fold-const.c (make_range, merge_ranges): Remove prototypes.
    (make_range_step): New function.
    (make_range): Use it.
    * tree.h (make_range_step): New prototypes.
    * Makefile.in (tree-ssa-reassoc.o): Depend on $(DIAGNOSTIC_CORE_H).
    * tree-ssa-reassoc.c: Include diagnostic-core.h.
    (struct range_entry): New type.
    (init_range_entry, range_entry_cmp, update_range_test,
    optimize_range_tests): New functions.
    (reassociate_bb): Call optimize_range_tests.

    * gcc.dg/pr46309.c: New test.

Added:
    trunk/gcc/testsuite/gcc.dg/pr46309.c
Modified:
    trunk/gcc/ChangeLog
    trunk/gcc/Makefile.in
    trunk/gcc/fold-const.c
    trunk/gcc/testsuite/ChangeLog
    trunk/gcc/tree-ssa-reassoc.c
    trunk/gcc/tree.h

[Bug tree-optimization/46309] optimization a==3||a==1

[paolo.carlini at oracle dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46309

Paolo Carlini <paolo.carlini at oracle dot com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
   Last reconfirmed|                            |2011-09-30
     Ever Confirmed|0                           |1

--- Comment #6 from Paolo Carlini <paolo.carlini at oracle dot com> 2011-09-30 20:10:37 UTC ---
Thank for this, Jakub! I'm confirming the issue ;) Shall we close it as fixed
for 4.7.0 or would it make sense to backport something?

[Bug tree-optimization/46309] optimization a==3||a==1

[jakub at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46309

--- Comment #7 from Jakub Jelinek <jakub at gcc dot gnu.org> 2011-09-30 20:18:59 UTC ---
Well, the original issue isn't fully fixed.  If the gimplifier decides to split
the conditions into multiple basic blocks, i.e. if it isn't
tmp1 = a == 1;
tmp2 = a == 3;
tmp3 = a == 4;
tmp4 = a == 2;
tmp5 = tmp1 | tmp2;
tmp6 = tmp5 | tmp3;
x = tmp6 | tmp4;
but instead
if (a == 1)
  goto <bb X>;
if (a == 3)
  goto <bb X>;
if (a == 4)
  goto <bb X>;
if (a == 2)
  goto <bb X>;
x = 0;
goto <bb Y>;
X:
x = 1;
or similar (e.g. two comparisons BIT_IOR_EXPRed together, then jump), then this
optimization isn't performed.

[Bug tree-optimization/46309] optimization a==3||a==1

[paolo.carlini at oracle dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46309

--- Comment #8 from Paolo Carlini <paolo.carlini at oracle dot com> 2011-09-30 20:25:14 UTC ---
I see... thanks.

[Bug tree-optimization/46309] optimization a==3||a==1

[pinskia at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46309

--- Comment #9 from Andrew Pinski <pinskia at gcc dot gnu.org> 2011-11-23 00:58:23 UTC ---
FAIL: gcc.dg/pr46309.c scan-tree-dump-times reassoc1 "Optimizing range tests
a_[0-9]*.D. -.1, 1. and -.2, 2. and -.3, 3. and -.4, 4.[
]* into" 2
FAIL: gcc.dg/pr46309.c scan-tree-dump-times reassoc1 "Optimizing range tests
a_[0-9]*.D. -.1, 1. and -.3, 3.[
]* into" 1
FAIL: gcc.dg/pr46309.c scan-tree-dump-times reassoc1 "Optimizing range tests
a_[0-9]*.D. -.1, 1. and -.2, 2.[
]* into" 1
FAIL: gcc.dg/pr46309.c scan-tree-dump-times reassoc1 "Optimizing range tests
a_[0-9]*.D. -.0, 31. and -.64, 95.[
]* into" 2
FAIL: gcc.dg/pr46309.c scan-tree-dump-times reassoc1 "Optimizing range tests
a_[0-9]*.D. -.128, 159. and -.192, 223.[
]* into" 1
FAIL: gcc.dg/pr46309.c scan-tree-dump-times reassoc2 "Optimizing range tests
D.[0-9]*_[0-9]* -.0, 31. and -.128, 159.[
]* into" 1


On mips64-linux-gnu

[Bug tree-optimization/46309] optimization a==3||a==1

[jakub at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46309

--- Comment #10 from Jakub Jelinek <jakub at gcc dot gnu.org> 2012-10-31 08:45:36 UTC ---
Author: jakub
Date: Wed Oct 31 08:45:27 2012
New Revision: 193028

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=193028
Log:
    PR tree-optimization/19105
    PR tree-optimization/21643
    PR tree-optimization/46309
    * tree-ssa-reassoc.c (init_range_entry): Add STMT argument
    and use it if EXP is NULL.
    (update_range_test): Handle OPCODE equal to ERROR_MARK
    and oe->op NULL.
    (optimize_range_tests): Likewise.
    (final_range_test_p, suitable_cond_bb, no_side_effect_bb, get_ops,
    maybe_optimize_range_tests): New functions.
    (reassociate_bb): Call maybe_optimize_range_tests if last
    stmt of bb is GIMPLE_COND that hasn't been visited yet.

    * gcc.dg/pr19105.c: New test.
    * gcc.dg/pr21643.c: New test.
    * gcc.dg/pr46309-2.c: New test.
    * gcc.c-torture/execute/pr46309.c: New test.

Added:
    trunk/gcc/testsuite/gcc.c-torture/execute/pr46309.c
    trunk/gcc/testsuite/gcc.dg/pr19105.c
    trunk/gcc/testsuite/gcc.dg/pr21643.c
    trunk/gcc/testsuite/gcc.dg/pr46309-2.c
Modified:
    trunk/gcc/ChangeLog
    trunk/gcc/testsuite/ChangeLog
    trunk/gcc/tree-ssa-reassoc.c

[Bug tree-optimization/46309] optimization a==3||a==1

[jakub at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46309

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|NEW                         |RESOLVED
         Resolution|                            |FIXED
   Target Milestone|---                         |4.8.0

--- Comment #11 from Jakub Jelinek <jakub at gcc dot gnu.org> 2012-10-31 08:54:40 UTC ---
Fixed.

[Bug tree-optimization/46309] optimization a==3||a==1

[eidletni at mail dot ru]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46309

eidletni at mail dot ru changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|RESOLVED                    |CLOSED

--- Comment #12 from eidletni at mail dot ru 2012-10-31 19:09:32 UTC ---
Cool, thank you!

[Bug tree-optimization/46309] optimization a==3||a==1

[rdsandiford at googlemail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46309

--- Comment #13 from rdsandiford at googlemail dot com <rdsandiford at googlemail dot com> 2012-10-31 22:36:48 UTC ---
"eidletni at mail dot ru" <gcc-bugzilla@gcc.gnu.org> writes:
> Cool, thank you!

+1, thanks Jakub