[Bug tree-optimization/15473] New: [tree-ssa] Sibcall optimization for libcalls.

["kazu at cs dot umass dot edu" <gcc-bugzilla at gcc dot gnu dot org>]

Consider:

long long
foo (long long a, long long b)
{
  return a / b;
}

I get:

foo (a, b)
{
<bb 0>:
  return a_1 / b_2;

}

with the assembly code being

foo:
	subl	$12, %esp
	pushl	28(%esp)
	pushl	28(%esp)
	pushl	28(%esp)
	pushl	28(%esp)
	call	__divdi3
	addl	$28, %esp
	ret

Note that the sequence of 4 pushl instructions simply copy 16 bytes
from one location of the stack to another.

If I create a (non-libcall) function with the same arguments like so

long long div64 (long long a, long long b);

long long
bar (long long a, long long b)
{
  return div64 (a, b);
}

I get:

bar (a, b)
{
  long long int T.0;

<bb 0>:
  T.0_3 = div64 (a_1, b_2) [tail call];
  return T.0_3;

}

with the assembly code being

bar:
	jmp	div64

-- 
           Summary: [tree-ssa] Sibcall optimization for libcalls.
           Product: gcc
           Version: 3.5.0
            Status: UNCONFIRMED
          Keywords: pessimizes-code
          Severity: enhancement
          Priority: P2
         Component: tree-optimization
        AssignedTo: unassigned at gcc dot gnu dot org
        ReportedBy: kazu at cs dot umass dot edu
                CC: gcc-bugs at gcc dot gnu dot org


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=15473