--- a/src/hotspot/cpu/aarch64/gc/g1/g1BarrierSetAssembler_aarch64.cpp	Tue May 15 09:33:02 2018 +0100
+++ b/src/hotspot/cpu/aarch64/gc/g1/g1BarrierSetAssembler_aarch64.cpp	Tue May 15 05:33:26 2018 -0400
@@ -116,7 +116,7 @@
 
   // Do we need to load the previous value?
   if (obj != noreg) {
-    __ load_heap_oop(pre_val, Address(obj, 0));
+    __ load_heap_oop(pre_val, Address(obj, 0), noreg, noreg, AS_RAW);
   }
 
   // Is the previous value null?
@@ -294,7 +294,7 @@
                        false /* expand_call */);
 
   if (val == noreg) {
-    __ store_heap_oop_null(Address(r3, 0));
+    BarrierSetAssembler::store_at(masm, decorators, type, Address(r3, 0), noreg, noreg, noreg);
   } else {
     // G1 barrier needs uncompressed oop for region cross check.
     Register new_val = val;
@@ -302,7 +302,7 @@
       new_val = rscratch2;
       __ mov(new_val, val);
     }
-    __ store_heap_oop(Address(r3, 0), val);
+    BarrierSetAssembler::store_at(masm, decorators, type, Address(r3, 0), val, noreg, noreg);
     g1_write_barrier_post(masm,
                           r3 /* store_adr */,
                           new_val /* new_val */,

--- a/src/hotspot/cpu/aarch64/gc/shared/barrierSetAssembler_aarch64.cpp	Tue May 15 09:33:02 2018 +0100
+++ b/src/hotspot/cpu/aarch64/gc/shared/barrierSetAssembler_aarch64.cpp	Tue May 15 05:33:26 2018 -0400
@@ -35,11 +35,21 @@
 
   bool on_heap = (decorators & IN_HEAP) != 0;
   bool on_root = (decorators & IN_ROOT) != 0;
+  bool oop_not_null = (decorators & OOP_NOT_NULL) != 0;
   switch (type) {
   case T_OBJECT:
   case T_ARRAY: {
     if (on_heap) {
-      __ load_heap_oop(dst, src);
+      if (UseCompressedOops) {
+        __ ldrw(dst, src);
+        if (oop_not_null) {
+          __ decode_heap_oop_not_null(dst);
+        } else {
+          __ decode_heap_oop(dst);
+        }
+      } else {
+        __ ldr(dst, src);
+      }
     } else {
       assert(on_root, "why else?");
       __ ldr(dst, src);
@@ -57,8 +67,17 @@
   switch (type) {
   case T_OBJECT:
   case T_ARRAY: {
+    val = val == noreg ? zr : val;
     if (on_heap) {
-      __ store_heap_oop(dst, val);
+      if (UseCompressedOops) {
+        assert(!dst.uses(val), "not enough registers");
+        if (val != zr) {
+          __ encode_heap_oop(val);
+        }
+        __ strw(val, dst);
+      } else {
+        __ str(val, dst);
+      }
     } else {
       assert(on_root, "why else?");
       __ str(val, dst);

--- a/src/hotspot/cpu/aarch64/gc/shared/cardTableBarrierSetAssembler_aarch64.cpp	Tue May 15 09:33:02 2018 +0100
+++ b/src/hotspot/cpu/aarch64/gc/shared/cardTableBarrierSetAssembler_aarch64.cpp	Tue May 15 05:33:26 2018 -0400
@@ -90,13 +90,14 @@
 
 void CardTableBarrierSetAssembler::oop_store_at(MacroAssembler* masm, DecoratorSet decorators, BasicType type,
                                                 Address dst, Register val, Register tmp1, Register tmp2) {
+  bool in_heap = (decorators & IN_HEAP) != 0;
   bool on_array = (decorators & IN_HEAP_ARRAY) != 0;
   bool on_anonymous = (decorators & ON_UNKNOWN_OOP_REF) != 0;
   bool precise = on_array || on_anonymous;
-  if (val == noreg) {
-    __ store_heap_oop_null(dst);
-  } else {
-    __ store_heap_oop(dst, val);
+
+  bool needs_post_barrier = val != noreg && in_heap;
+  BarrierSetAssembler::store_at(masm, decorators, type, dst, val, noreg, noreg);
+  if (needs_post_barrier) {
     // flatten object address if needed
     if (!precise || (dst.index() == noreg && dst.offset() == 0)) {
       store_check(masm, dst.base(), dst);

--- a/src/hotspot/cpu/aarch64/interp_masm_aarch64.cpp	Tue May 15 09:33:02 2018 +0100
+++ b/src/hotspot/cpu/aarch64/interp_masm_aarch64.cpp	Tue May 15 05:33:26 2018 -0400
@@ -278,8 +278,7 @@
   resolve_oop_handle(result, tmp);
   // Add in the index
   add(result, result, index);
-  BarrierSetAssembler *bs = BarrierSet::barrier_set()->barrier_set_assembler();
-  bs->load_at(this, IN_HEAP, T_OBJECT, result, Address(result, arrayOopDesc::base_offset_in_bytes(T_OBJECT)), tmp, /*tmp_thread*/ noreg);
+  load_heap_oop(result, Address(result, arrayOopDesc::base_offset_in_bytes(T_OBJECT)));
 }
 
 void InterpreterMacroAssembler::load_resolved_klass_at_offset(

--- a/src/hotspot/cpu/aarch64/macroAssembler_aarch64.cpp	Tue May 15 09:33:02 2018 +0100
+++ b/src/hotspot/cpu/aarch64/macroAssembler_aarch64.cpp	Tue May 15 05:33:26 2018 -0400
@@ -3975,41 +3975,48 @@
   movk(dst, nk & 0xffff);
 }
 
-void MacroAssembler::load_heap_oop(Register dst, Address src)
-{
-  if (UseCompressedOops) {
-    ldrw(dst, src);
-    decode_heap_oop(dst);
+void MacroAssembler::access_load_at(BasicType type, DecoratorSet decorators,
+                                    Register dst, Address src,
+                                    Register tmp1, Register thread_tmp) {
+  BarrierSetAssembler *bs = BarrierSet::barrier_set()->barrier_set_assembler();
+  bool as_raw = (decorators & AS_RAW) != 0;
+  if (as_raw) {
+    bs->BarrierSetAssembler::load_at(this, decorators, type, dst, src, tmp1, thread_tmp);
   } else {
-    ldr(dst, src);
+    bs->load_at(this, decorators, type, dst, src, tmp1, thread_tmp);
   }
 }
 
-void MacroAssembler::load_heap_oop_not_null(Register dst, Address src)
-{
-  if (UseCompressedOops) {
-    ldrw(dst, src);
-    decode_heap_oop_not_null(dst);
+void MacroAssembler::access_store_at(BasicType type, DecoratorSet decorators,
+                                     Address dst, Register src,
+                                     Register tmp1, Register thread_tmp) {
+  BarrierSetAssembler *bs = BarrierSet::barrier_set()->barrier_set_assembler();
+  bool as_raw = (decorators & AS_RAW) != 0;
+  if (as_raw) {
+    bs->BarrierSetAssembler::store_at(this, decorators, type, dst, src, tmp1, thread_tmp);
   } else {
-    ldr(dst, src);
+    bs->store_at(this, decorators, type, dst, src, tmp1, thread_tmp);
   }
 }
 
-void MacroAssembler::store_heap_oop(Address dst, Register src) {
-  if (UseCompressedOops) {
-    assert(!dst.uses(src), "not enough registers");
-    encode_heap_oop(src);
-    strw(src, dst);
-  } else
-    str(src, dst);
+void MacroAssembler::load_heap_oop(Register dst, Address src, Register tmp1,
+                                   Register thread_tmp, DecoratorSet decorators) {
+  access_load_at(T_OBJECT, IN_HEAP | decorators, dst, src, tmp1, thread_tmp);
+}
+
+void MacroAssembler::load_heap_oop_not_null(Register dst, Address src, Register tmp1,
+                                            Register thread_tmp, DecoratorSet decorators) {
+  access_load_at(T_OBJECT, IN_HEAP | OOP_NOT_NULL | decorators, dst, src, tmp1, thread_tmp);
+}
+
+void MacroAssembler::store_heap_oop(Address dst, Register src, Register tmp1,
+                                    Register thread_tmp, DecoratorSet decorators) {
+  access_store_at(T_OBJECT, IN_HEAP | decorators, dst, src, tmp1, thread_tmp);
 }
 
 // Used for storing NULLs.
 void MacroAssembler::store_heap_oop_null(Address dst) {
-  if (UseCompressedOops) {
-    strw(zr, dst);
-  } else
-    str(zr, dst);
+  access_store_at(T_OBJECT, IN_HEAP, dst, noreg, noreg, noreg);
 }
 
 Address MacroAssembler::allocate_metadata_address(Metadata* obj) {

--- a/src/hotspot/cpu/aarch64/macroAssembler_aarch64.hpp	Tue May 15 09:33:02 2018 +0100
+++ b/src/hotspot/cpu/aarch64/macroAssembler_aarch64.hpp	Tue May 15 05:33:26 2018 -0400
@@ -789,10 +789,19 @@
   void resolve_oop_handle(Register result, Register tmp = r5);
   void load_mirror(Register dst, Register method, Register tmp = r5);
 
-  void load_heap_oop(Register dst, Address src);
+  void access_load_at(BasicType type, DecoratorSet decorators, Register dst, Address src,
+                      Register tmp1, Register tmp_thread);
+
+  void access_store_at(BasicType type, DecoratorSet decorators, Address dst, Register src,
+                       Register tmp1, Register tmp_thread);
 
-  void load_heap_oop_not_null(Register dst, Address src);
-  void store_heap_oop(Address dst, Register src);
+  void load_heap_oop(Register dst, Address src, Register tmp1 = noreg,
+                     Register thread_tmp = noreg, DecoratorSet decorators = 0);
+
+  void load_heap_oop_not_null(Register dst, Address src, Register tmp1 = noreg,
+                              Register thread_tmp = noreg, DecoratorSet decorators = 0);
+  void store_heap_oop(Address dst, Register src, Register tmp1 = noreg,
+                      Register tmp_thread = noreg, DecoratorSet decorators = 0);
 
   // currently unimplemented
   // Used for storing NULL. All other oop constants should be

--- a/src/hotspot/cpu/aarch64/methodHandles_aarch64.cpp	Tue May 15 09:33:02 2018 +0100
+++ b/src/hotspot/cpu/aarch64/methodHandles_aarch64.cpp	Tue May 15 05:33:26 2018 -0400
@@ -135,11 +135,11 @@
 
   // Load the invoker, as MH -> MH.form -> LF.vmentry
   __ verify_oop(recv);
-  __ load_heap_oop(method_temp, Address(recv, NONZERO(java_lang_invoke_MethodHandle::form_offset_in_bytes())));
+  __ load_heap_oop(method_temp, Address(recv, NONZERO(java_lang_invoke_MethodHandle::form_offset_in_bytes())), temp2);
   __ verify_oop(method_temp);
-  __ load_heap_oop(method_temp, Address(method_temp, NONZERO(java_lang_invoke_LambdaForm::vmentry_offset_in_bytes())));
+  __ load_heap_oop(method_temp, Address(method_temp, NONZERO(java_lang_invoke_LambdaForm::vmentry_offset_in_bytes())), temp2);
   __ verify_oop(method_temp);
-  __ load_heap_oop(method_temp, Address(method_temp, NONZERO(java_lang_invoke_MemberName::method_offset_in_bytes())));
+  __ load_heap_oop(method_temp, Address(method_temp, NONZERO(java_lang_invoke_MemberName::method_offset_in_bytes())), temp2);
   __ verify_oop(method_temp);
   __ ldr(method_temp, Address(method_temp, NONZERO(java_lang_invoke_ResolvedMethodName::vmtarget_offset_in_bytes())));
 
@@ -311,7 +311,7 @@
       if (VerifyMethodHandles && iid != vmIntrinsics::_linkToInterface) {
         Label L_ok;
         Register temp2_defc = temp2;
-        __ load_heap_oop(temp2_defc, member_clazz);
+        __ load_heap_oop(temp2_defc, member_clazz, temp3);
         load_klass_from_Class(_masm, temp2_defc);
         __ verify_klass_ptr(temp2_defc);
         __ check_klass_subtype(temp1_recv_klass, temp2_defc, temp3, L_ok);

--- a/src/hotspot/cpu/aarch64/stubGenerator_aarch64.cpp	Tue May 15 09:33:02 2018 +0100
+++ b/src/hotspot/cpu/aarch64/stubGenerator_aarch64.cpp	Tue May 15 05:33:26 2018 -0400
@@ -1816,13 +1816,13 @@
     __ align(OptoLoopAlignment);
 
     __ BIND(L_store_element);
-    __ store_heap_oop(__ post(to, UseCompressedOops ? 4 : 8), copied_oop);  // store the oop
+    __ store_heap_oop(__ post(to, UseCompressedOops ? 4 : 8), copied_oop, noreg, noreg, AS_RAW);  // store the oop
     __ sub(count, count, 1);
     __ cbz(count, L_do_card_marks);
 
     // ======== loop entry is here ========
     __ BIND(L_load_element);
-    __ load_heap_oop(copied_oop, __ post(from, UseCompressedOops ? 4 : 8)); // load the oop
+    __ load_heap_oop(copied_oop, __ post(from, UseCompressedOops ? 4 : 8), noreg, noreg, AS_RAW); // load the oop
     __ cbz(copied_oop, L_store_element);
 
     __ load_klass(r19_klass, copied_oop);// query the object klass

--- a/src/hotspot/cpu/aarch64/templateTable_aarch64.cpp	Tue May 15 09:33:02 2018 +0100
+++ b/src/hotspot/cpu/aarch64/templateTable_aarch64.cpp	Tue May 15 05:33:26 2018 -0400
@@ -147,16 +147,14 @@
                          Register val,
                          DecoratorSet decorators) {
   assert(val == noreg || val == r0, "parameter is just for looks");
-  BarrierSetAssembler *bs = BarrierSet::barrier_set()->barrier_set_assembler();
-  bs->store_at(_masm, decorators, T_OBJECT, dst, val, /*tmp1*/ r10, /*tmp2*/ r1);
+  __ store_heap_oop(dst, val, r10, r1, decorators);
 }
 
 static void do_oop_load(InterpreterMacroAssembler* _masm,
                         Address src,
                         Register dst,
                         DecoratorSet decorators) {
-  BarrierSetAssembler *bs = BarrierSet::barrier_set()->barrier_set_assembler();
-  bs->load_at(_masm, decorators, T_OBJECT, dst, src, /*tmp1*/ r10, /*tmp_thread*/ r1);
+  __ load_heap_oop(dst, src, r10, r1, decorators);
 }
 
 Address TemplateTable::at_bcp(int offset) {