[Bug target/61387] [4.10 Regression] ~900 test failures on on x86_64-apple-darwin13 for g++ with -m64 after r211088

["iains at gcc dot gnu.org" <gcc-bugzilla at gcc dot gnu dot org>]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=61387

Iain Sandoe <iains at gcc dot gnu.org> changed:

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

--- Comment #4 from Iain Sandoe <iains at gcc dot gnu.org> ---
some analysis:

1. The codepath:

     predicate that can distinguish between frame and non-frame memory.
     For our purposes here, we can get away with (ab)using a jump pattern,
     because we're going to do no optimization.  */
  if (MEM_P (fnaddr))
    {
      if (sibcall_insn_operand (fnaddr, word_mode))
    {

does not appear to be triggered for x86_64-linux (i put an assert in which was
not triggered in either a bootstrap or complete c/c++ testsuite run).

----

2.

taking say pr58585.c as an example.

struct A
{
  virtual void foo() {}
  void bar();
};
void A::bar() { foo(); }


Linux and Darwin both mark foo() as weak.

However,
Linux says that targetm.binds_local_p (foo) == true.
Darwin says it's false.

So Darwin takes the code-path above and Linux doesn't.  Right now, I'm not sure
which is the right claim on the binds_local_p.

3.

Anyway, assuming that the intention is to unwrap the indirection from the call
- something like:

  if (MEM_P (fnaddr))
    {
      if (sibcall_insn_operand (fnaddr, word_mode))
    {
      fnaddr = XEXP (DECL_RTL (function), 0);
          tmp = gen_rtx_MEM (QImode, fnaddr);
          tmp = gen_rtx_CALL (VOIDmode, tmp, const0_rtx);
          tmp = emit_call_insn (tmp);
          SIBLING_CALL_P (tmp) = 1;
    }
      else
    emit_jump_insn (gen_indirect_jump (fnaddr));
    }

seems more like what was intended.

However, I really think we need to explain the difference in treatment.

Mike? amy thoughts?