[Bug target/50123] New: cmpxchg generated for atomic and with zero/or with -1

[Bug target/50123] New: cmpxchg generated for atomic and with zero/or with -1

[bonzini at gnu dot org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=50123

             Bug #: 50123
           Summary: cmpxchg generated for atomic and with zero/or with -1
    Classification: Unclassified
           Product: gcc
           Version: 4.7.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: target
        AssignedTo: unassigned@gcc.gnu.org
        ReportedBy: bonzini@gnu.org


int x;
int f()
{
        return __sync_fetch_and_and(&x, 0);
}
int g()
{
        return __sync_fetch_and_or(&x, -1);
}

f and g can be optimized to atomic xchg on x86.

[Bug target/50123] cmpxchg generated for atomic and with zero/or with -1

[jakub at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=50123

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |aldyh at gcc dot gnu.org,
                   |                            |amacleod at redhat dot com,
                   |                            |jakub at gcc dot gnu.org

--- Comment #1 from Jakub Jelinek <jakub at gcc dot gnu.org> 2011-08-19 06:53:18 UTC ---
While we could do this on i?86/x86_64 now, it sounds better to do it in the
generic code (folding or expansion), but for that we need to wait for the
extended sync builtins that are currently brewing on cxx-mem-model branch
(because __sync_fetch_and_{and,or} are full barriers, while __sync_test_and_set
is not).

[Bug target/50123] cmpxchg generated for atomic and with zero/or with -1

[amacleod at redhat dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=50123

--- Comment #2 from Andrew Macleod <amacleod at redhat dot com> 2011-11-29 19:42:50 UTC ---
Author: amacleod
Date: Tue Nov 29 19:42:44 2011
New Revision: 181815

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=181815
Log:
2011-11-29  Andrew MacLeod  <amacleod@redhat.com>

    PR target/50123
    * optabs.c (maybe_optimize_fetch_op): New.  Look for more optimal
    instructions for a FECTH_OP or OP_FECTH sequence.
    (expand_atomic_fetch_op): Call maybe_optimize_fetch_op.
    * testsuite/gcc.dg/atomic-op-optimize.c: New.  Test for optimizations.


Added:
    trunk/gcc/testsuite/gcc.dg/atomic-op-optimize.c
Modified:
    trunk/gcc/ChangeLog
    trunk/gcc/optabs.c
    trunk/gcc/testsuite/ChangeLog

[Bug target/50123] cmpxchg generated for atomic and with zero/or with -1

[bonzini at gnu dot org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=50123

Paolo Bonzini <bonzini at gnu dot org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |RESOLVED
      Known to work|                            |4.8.2
         Resolution|---                         |FIXED

--- Comment #4 from Paolo Bonzini <bonzini at gnu dot org> ---
4.8.2 works, will use xchg.