8023597: Optimize G1 barriers code for unsafe load_store

Avoid loading old values in G1 pre-barriers for inlined unsafe load_store nodes.

Reviewed-by: kvn, tonyp

hotspot/src/share/vm/opto/graphKit.cpp

@@ -1501,6 +1501,25 @@ void GraphKit::pre_barrier(bool do_load,
   }
 }
 
+bool GraphKit::can_move_pre_barrier() const {
+  BarrierSet* bs = Universe::heap()->barrier_set();
+  switch (bs->kind()) {
+    case BarrierSet::G1SATBCT:
+    case BarrierSet::G1SATBCTLogging:
+      return true; // Can move it if no safepoint
+
+    case BarrierSet::CardTableModRef:
+    case BarrierSet::CardTableExtension:
+    case BarrierSet::ModRef:
+      return true; // There is no pre-barrier
+
+    case BarrierSet::Other:
+    default      :
+      ShouldNotReachHere();
+  }
+  return false;
+}
+
 void GraphKit::post_barrier(Node* ctl,
                             Node* store,
                             Node* obj,
@@ -3551,6 +3570,8 @@ void GraphKit::g1_write_barrier_pre(bool do_load,
   } else {
     // In this case both val_type and alias_idx are unused.
     assert(pre_val != NULL, "must be loaded already");
+    // Nothing to be done if pre_val is null.
+    if (pre_val->bottom_type() == TypePtr::NULL_PTR) return;
     assert(pre_val->bottom_type()->basic_type() == T_OBJECT, "or we shouldn't be here");
   }
   assert(bt == T_OBJECT, "or we shouldn't be here");

hotspot/src/share/vm/opto/graphKit.hpp

@@ -695,6 +695,10 @@ class GraphKit : public Phase {
   void write_barrier_post(Node *store, Node* obj,
                           Node* adr,  uint adr_idx, Node* val, bool use_precise);
 
+  // Allow reordering of pre-barrier with oop store and/or post-barrier.
+  // Used for load_store operations which loads old value.
+  bool can_move_pre_barrier() const;
+
   // G1 pre/post barriers
   void g1_write_barrier_pre(bool do_load,
                             Node* obj,

hotspot/src/share/vm/opto/library_call.cpp

@@ -2756,10 +2756,28 @@ bool LibraryCallKit::inline_unsafe_load_store(BasicType type, LoadStoreKind kind
       newval = _gvn.makecon(TypePtr::NULL_PTR);
 
     // Reference stores need a store barrier.
+    if (kind == LS_xchg) {
+      // If pre-barrier must execute before the oop store, old value will require do_load here.
+      if (!can_move_pre_barrier()) {
         pre_barrier(true /* do_load*/,
                     control(), base, adr, alias_idx, newval, value_type->make_oopptr(),
                     NULL /* pre_val*/,
                     T_OBJECT);
+      } // Else move pre_barrier to use load_store value, see below.
+    } else if (kind == LS_cmpxchg) {
+      // Same as for newval above:
+      if (_gvn.type(oldval) == TypePtr::NULL_PTR) {
+        oldval = _gvn.makecon(TypePtr::NULL_PTR);
+      }
+      // The only known value which might get overwritten is oldval.
+      pre_barrier(false /* do_load */,
+                  control(), NULL, NULL, max_juint, NULL, NULL,
+                  oldval /* pre_val */,
+                  T_OBJECT);
+    } else {
+      ShouldNotReachHere();
+    }
+
 #ifdef _LP64
     if (adr->bottom_type()->is_ptr_to_narrowoop()) {
       Node *newval_enc = _gvn.transform(new (C) EncodePNode(newval, newval->bottom_type()->make_narrowoop()));
@@ -2795,15 +2813,26 @@ bool LibraryCallKit::inline_unsafe_load_store(BasicType type, LoadStoreKind kind
   Node* proj = _gvn.transform(new (C) SCMemProjNode(load_store));
   set_memory(proj, alias_idx);
 
-  // Add the trailing membar surrounding the access
-  insert_mem_bar(Op_MemBarCPUOrder);
-  insert_mem_bar(Op_MemBarAcquire);
-
+  if (type == T_OBJECT && kind == LS_xchg) {
 #ifdef _LP64
-  if (type == T_OBJECT && adr->bottom_type()->is_ptr_to_narrowoop() && kind == LS_xchg) {
+    if (adr->bottom_type()->is_ptr_to_narrowoop()) {
       load_store = _gvn.transform(new (C) DecodeNNode(load_store, load_store->get_ptr_type()));
     }
 #endif
+    if (can_move_pre_barrier()) {
+      // Don't need to load pre_val. The old value is returned by load_store.
+      // The pre_barrier can execute after the xchg as long as no safepoint
+      // gets inserted between them.
+      pre_barrier(false /* do_load */,
+                  control(), NULL, NULL, max_juint, NULL, NULL,
+                  load_store /* pre_val */,
+                  T_OBJECT);
+    }
+  }
+
+  // Add the trailing membar surrounding the access
+  insert_mem_bar(Op_MemBarCPUOrder);
+  insert_mem_bar(Op_MemBarAcquire);
 
   assert(type2size[load_store->bottom_type()->basic_type()] == type2size[rtype], "result type should match");
   set_result(load_store);