src/share/vm/gc_implementation/g1/g1ParScanThreadState.hpp

rev 7471 : 8060025: Object copy time regressions after JDK-8031323 and JDK-8057536
Summary: Evaluate and improve object copy time by micro-optimizations and splitting out slow and fast paths aggressively.
Reviewed-by:
Contributed-by: Tony Printezis <tprintezis@twitter.com>, Thomas Schatzl <thomas.schatzl@oracle.com>

*** 47,57 ****
    G1RemSet* _g1_rem;
  
    G1ParGCAllocator*   _g1_par_allocator;
  
    ageTable            _age_table;
! 
    G1ParScanClosure    _scanner;
  
    size_t           _alloc_buffer_waste;
    size_t           _undo_waste;
  
--- 47,59 ----
    G1RemSet* _g1_rem;
  
    G1ParGCAllocator* _g1_par_allocator;
  
    ageTable          _age_table;
!   in_cset_state_t   _dest[InCSetState::Num];
!   // Local tenuring threshold.
!   uint              _tenuring_threshold;
    G1ParScanClosure  _scanner;
  
    size_t            _alloc_buffer_waste;
    size_t            _undo_waste;
  
*** 80,89 ****
--- 82,99 ----
    void   add_to_undo_waste(size_t waste)         { _undo_waste += waste; }
  
    DirtyCardQueue& dirty_card_queue()             { return _dcq;  }
    G1SATBCardTableModRefBS* ctbs()                { return _ct_bs; }
  
+   in_cset_state_t dest(uint index) const {
+     assert(index < InCSetState::Num,
+            err_msg("dest index out-of-bounds: %u", index));
+     assert(_dest[index] != InCSetState::NotInCSet,
+            err_msg("dest is invalid: %u", index));
+     return _dest[index];
+   }
+ 
   public:
    G1ParScanThreadState(G1CollectedHeap* g1h, uint queue_num, ReferenceProcessor* rp);
    ~G1ParScanThreadState();
  
    ageTable*         age_table()       { return &_age_table;       }
*** 110,120 ****
        if (ctbs()->mark_card_deferred(card_index)) {
          dirty_card_queue().enqueue((jbyte*)ctbs()->byte_for_index(card_index));
        }
      }
    }
-  public:
  
    void set_evac_failure_closure(OopsInHeapRegionClosure* evac_failure_cl) {
      _evac_failure_cl = evac_failure_cl;
    }
  
--- 120,129 ----
*** 191,203 ****
    template <class T> inline void do_oop_evac(T* p, HeapRegion* from);
  
    template <class T> inline void deal_with_reference(T* ref_to_scan);
  
    inline void dispatch_reference(StarTask ref);
   public:
  
!   oop copy_to_survivor_space(oop const obj, markOop const old_mark);
  
    void trim_queue();
  
    inline void steal_and_trim_queue(RefToScanQueueSet *task_queues);
  };
--- 200,223 ----
    template <class T> inline void do_oop_evac(T* p, HeapRegion* from);
  
    template <class T> inline void deal_with_reference(T* ref_to_scan);
  
    inline void dispatch_reference(StarTask ref);
+ 
+   // Tries to allocate word_sz in the PLAB of the next "generation" after trying to
+   // allocate into dest. State is the original (source) cset state for the object
+   // that is allocated for.
+   // Returns a non-NULL pointer if successful, and updates dest if required.
+   HeapWord* allocate_in_next_plab(in_cset_state_t const state,
+                                   in_cset_state_t* dest,
+                                   size_t word_sz,
+                                   AllocationContext_t const context);
+ 
+   inline in_cset_state_t next_state(in_cset_state_t const state, markOop const m, uint& age);
   public:
  
!   oop copy_to_survivor_space(in_cset_state_t const state, oop const obj, markOop const old_mark);
  
    void trim_queue();
  
    inline void steal_and_trim_queue(RefToScanQueueSet *task_queues);
  };