[Bug c++/70091] Codegen emits dead load on x86-64

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=70091

Richard Biener <rguenth at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Keywords|                            |wrong-code
             Status|UNCONFIRMED                 |NEW
   Last reconfirmed|                            |2016-03-07
     Ever confirmed|0                           |1
      Known to fail|                            |5.3.0

--- Comment #1 from Richard Biener <rguenth at gcc dot gnu.org> ---
The C++ FE emits

;; Function void bitfield::and_assign() volatile (null)
;; enabled by -tree-original


<<cleanup_point <<< Unknown tree: expr_stmt
  (void) (((volatile struct bitfield *) this)->_raw = *(volatile struct
bitfield *) this;, bitfield::_and ((unsigned int) ((volatile struct bitfield *)
this)->_raw, 1);) >>>>>;
<<cleanup_point <<< Unknown tree: expr_stmt
  (void) (((volatile struct bitfield *) this)->_raw = (unsigned int) ((volatile
struct bitfield *) this)->_raw & 1) >>>>>;

there's the odd

((volatile struct bitfield *) this)->_raw = *(volatile struct bitfield *) this;

statement which causes this.  But the store goes away somehow.  GIMPLE:

void bitfield::and_assign() volatile (volatile struct bitfield * const this)
{
  struct bitfield vol.0;
  unsigned int D.2339;
  unsigned int D.2340;
  unsigned int D.2341;
  unsigned int D.2342;

  vol.0 = *this;
  D.2339 = this->_raw;
  D.2340 = bitfield::_and (D.2339, 1);
  this->_raw = D.2340;


Confirmed as FE bug.