src/share/vm/utilities/taskqueue.hpp

rev 4231 : Fix memory ordering in taskqueue for platforms with weak memory ordering (PPC).

Add fence as described in "Correct and Efficient Work-Stealing for Weak Memory Models;
Le, Pop, Cohen, Nardelli; PPoPP 13".  With four or more GC threads bottom can be
older than age in get_bottom().

*** 388,398 ****
  }
  
  template<class E, MEMFLAGS F, unsigned int N>
  bool GenericTaskQueue<E, F, N>::pop_global(E& t) {
    Age oldAge = _age.get();
!   uint localBot = _bottom;
    uint n_elems = size(localBot, oldAge.top());
    if (n_elems == 0) {
      return false;
    }
  
--- 388,405 ----
  }
  
  template<class E, MEMFLAGS F, unsigned int N>
  bool GenericTaskQueue<E, F, N>::pop_global(E& t) {
    Age oldAge = _age.get();
!   // Architectures with weak memory model require a fence here. The
!   // fence has a cumulative effect on getting age and getting bottom.
!   // This way it is guaranteed that bottom is not older than age,
!   // what is crucial for the correctness of the algorithm.
! #if (defined ARM || defined IA64 || defined PPC64)
!   OrderAccess::fence();
! #endif
!   uint localBot = OrderAccess::load_acquire((volatile juint*)&_bottom);
    uint n_elems = size(localBot, oldAge.top());
    if (n_elems == 0) {
      return false;
    }
  
*** 632,642 ****
  };
  
  template<class E, MEMFLAGS F, unsigned int N> inline bool
  GenericTaskQueue<E, F, N>::push(E t) {
    uint localBot = _bottom;
!   assert((localBot >= 0) && (localBot < N), "_bottom out of range.");
    idx_t top = _age.top();
    uint dirty_n_elems = dirty_size(localBot, top);
    assert(dirty_n_elems < N, "n_elems out of range.");
    if (dirty_n_elems < max_elems()) {
      // g++ complains if the volatile result of the assignment is unused.
--- 639,649 ----
  };
  
  template<class E, MEMFLAGS F, unsigned int N> inline bool
  GenericTaskQueue<E, F, N>::push(E t) {
    uint localBot = _bottom;
!   assert(localBot < N, "_bottom out of range.");
    idx_t top = _age.top();
    uint dirty_n_elems = dirty_size(localBot, top);
    assert(dirty_n_elems < N, "n_elems out of range.");
    if (dirty_n_elems < max_elems()) {
      // g++ complains if the volatile result of the assignment is unused.