--- a/src/share/vm/opto/graphKit.cpp	Tue Apr 26 12:14:22 2011 -0700
+++ b/src/share/vm/opto/graphKit.cpp	Wed Apr 27 15:40:36 2011 -0700
@@ -3447,9 +3447,22 @@
 
   // Get the alias_index for raw card-mark memory
   int adr_type = Compile::AliasIdxRaw;
+  Node*   zero = __ ConI(0); // Dirty card value
+  BasicType bt = T_BYTE;
+
+  if (UseCondCardMark) {
+    // The classic GC reference write barrier is typically implemented
+    // as a store into the global card mark table.  Unfortunately
+    // unconditional stores can result in false sharing and excessive
+    // coherence traffic as well as false transactional aborts.
+    // UseCondCardMark enables MP "polite" conditional card mark
+    // stores.  In theory we could relax the load from ctrl() to
+    // no_ctrl, but that doesn't buy much latitude.
+    Node* card_val = __ load( __ ctrl(), card_adr, TypeInt::BYTE, bt, adr_type);
+    __ if_then(card_val, BoolTest::ne, zero);
+  }
+
   // Smash zero into card
-  Node*   zero = __ ConI(0);
-  BasicType bt = T_BYTE;
   if( !UseConcMarkSweepGC ) {
     __ store(__ ctrl(), card_adr, zero, bt, adr_type);
   } else {
@@ -3457,6 +3470,10 @@
     __ storeCM(__ ctrl(), card_adr, zero, oop_store, adr_idx, bt, adr_type);
   }
 
+  if (UseCondCardMark) {
+    __ end_if();
+  }
+
   // Final sync IdealKit and GraphKit.
   final_sync(ideal);
 }

--- a/src/share/vm/opto/macro.cpp	Tue Apr 26 12:14:22 2011 -0700
+++ b/src/share/vm/opto/macro.cpp	Wed Apr 27 15:40:36 2011 -0700
@@ -221,9 +221,16 @@
     Node *shift = p2x->unique_out();
     Node *addp = shift->unique_out();
     for (DUIterator_Last jmin, j = addp->last_outs(jmin); j >= jmin; --j) {
-      Node *st = addp->last_out(j);
-      assert(st->is_Store(), "store required");
-      _igvn.replace_node(st, st->in(MemNode::Memory));
+      Node *mem = addp->last_out(j);
+      if (UseCondCardMark && mem->is_Load()) {
+        assert(mem->Opcode() == Op_LoadB, "unexpected code shape");
+        // The load is checking if the card has been written so
+        // replace it with zero to fold the test.
+        _igvn.replace_node(mem, intcon(0));
+        continue;
+      }
+      assert(mem->is_Store(), "store required");
+      _igvn.replace_node(mem, mem->in(MemNode::Memory));
     }
   } else {
     // G1 pre/post barriers

--- a/src/share/vm/runtime/globals.hpp	Tue Apr 26 12:14:22 2011 -0700
+++ b/src/share/vm/runtime/globals.hpp	Wed Apr 27 15:40:36 2011 -0700
@@ -620,6 +620,9 @@
   product(bool, UseSSE42Intrinsics, false,                                  \
           "SSE4.2 versions of intrinsics")                                  \
                                                                             \
+  product(bool, UseCondCardMark, false,                                     \
+          "Check for already marked card before updating card table")       \
+                                                                            \
   develop(bool, TraceCallFixup, false,                                      \
           "traces all call fixups")                                         \
                                                                             \