14 files changed, 373 insertions, 107 deletions

diff --git a/src/share/vm/c1/c1_Compilation.cpp b/src/share/vm/c1/c1_Compilation.cpp
index 1cdcab542..574fda194 100644
--- a/src/share/vm/c1/c1_Compilation.cpp
+++ b/src/share/vm/c1/c1_Compilation.cpp
@@ -601,6 +601,17 @@ void Compilation::bailout(const char* msg) {
   }
 }
 
+ciKlass* Compilation::cha_exact_type(ciType* type) {
+  if (type != NULL && type->is_loaded() && type->is_instance_klass()) {
+    ciInstanceKlass* ik = type->as_instance_klass();
+    assert(ik->exact_klass() == NULL, "no cha for final klass");
+    if (DeoptC1 && UseCHA && !(ik->has_subklass() || ik->is_interface())) {
+      dependency_recorder()->assert_leaf_type(ik);
+      return ik;
+    }
+  }
+  return NULL;
+}
 
 void Compilation::print_timers() {
   // tty->print_cr("    Native methods         : %6.3f s, Average : %2.3f", CompileBroker::_t_native_compilation.seconds(), CompileBroker::_t_native_compilation.seconds() / CompileBroker::_total_native_compile_count);
diff --git a/src/share/vm/c1/c1_Compilation.hpp b/src/share/vm/c1/c1_Compilation.hpp
index f98ae97f3..7386dc414 100644
--- a/src/share/vm/c1/c1_Compilation.hpp
+++ b/src/share/vm/c1/c1_Compilation.hpp
@@ -246,6 +246,8 @@ class Compilation: public StackObj {
       (RangeCheckElimination || UseLoopInvariantCodeMotion) &&
       method()->method_data()->trap_count(Deoptimization::Reason_none) == 0;
   }
+
+  ciKlass* cha_exact_type(ciType* type);
 };
 
 
diff --git a/src/share/vm/c1/c1_GraphBuilder.cpp b/src/share/vm/c1/c1_GraphBuilder.cpp
index 03d0d75fa..f93a6e890 100644
--- a/src/share/vm/c1/c1_GraphBuilder.cpp
+++ b/src/share/vm/c1/c1_GraphBuilder.cpp
@@ -1658,6 +1658,42 @@ Dependencies* GraphBuilder::dependency_recorder() const {
   return compilation()->dependency_recorder();
 }
 
+// How many arguments do we want to profile?
+Values* GraphBuilder::args_list_for_profiling(int& start, bool may_have_receiver) {
+  int n = 0;
+  assert(start == 0, "should be initialized");
+  if (MethodData::profile_arguments()) {
+    ciProfileData* data = method()->method_data()->bci_to_data(bci());
+    if (data->is_CallTypeData() || data->is_VirtualCallTypeData()) {
+      n = data->is_CallTypeData() ? data->as_CallTypeData()->number_of_arguments() : data->as_VirtualCallTypeData()->number_of_arguments();
+      bool has_receiver = may_have_receiver && Bytecodes::has_receiver(method()->java_code_at_bci(bci()));
+      start = has_receiver ? 1 : 0;
+    }
+  }
+  if (n > 0) {
+    return new Values(n);
+  }
+  return NULL;
+}
+
+// Collect arguments that we want to profile in a list
+Values* GraphBuilder::collect_args_for_profiling(Values* args, bool may_have_receiver) {
+  int start = 0;
+  Values* obj_args = args_list_for_profiling(start, may_have_receiver);
+  if (obj_args == NULL) {
+    return NULL;
+  }
+  int s = obj_args->size();
+  for (int i = start, j = 0; j < s; i++) {
+    if (args->at(i)->type()->is_object_kind()) {
+      obj_args->push(args->at(i));
+      j++;
+    }
+  }
+  assert(s == obj_args->length(), "missed on arg?");
+  return obj_args;
+}
+
 
 void GraphBuilder::invoke(Bytecodes::Code code) {
   bool will_link;
@@ -1957,7 +1993,7 @@ void GraphBuilder::invoke(Bytecodes::Code code) {
       } else if (exact_target != NULL) {
         target_klass = exact_target->holder();
       }
-      profile_call(target, recv, target_klass);
+      profile_call(target, recv, target_klass, collect_args_for_profiling(args, false), false);
     }
   }
 
@@ -3509,7 +3545,7 @@ bool GraphBuilder::try_inline_intrinsics(ciMethod* callee) {
           recv = args->at(0);
           null_check(recv);
         }
-        profile_call(callee, recv, NULL);
+        profile_call(callee, recv, NULL, collect_args_for_profiling(args, true), true);
       }
     }
   }
@@ -3763,7 +3799,28 @@ bool GraphBuilder::try_inline_full(ciMethod* callee, bool holder_known, Bytecode
     compilation()->set_would_profile(true);
 
     if (profile_calls()) {
-      profile_call(callee, recv, holder_known ? callee->holder() : NULL);
+      int start = 0;
+      Values* obj_args = args_list_for_profiling(start, has_receiver);
+      if (obj_args != NULL) {
+        int s = obj_args->size();
+        // if called through method handle invoke, some arguments may have been popped
+        for (int i = args_base+start, j = 0; j < obj_args->size() && i < state()->stack_size(); ) {
+          Value v = state()->stack_at_inc(i);
+          if (v->type()->is_object_kind()) {
+            obj_args->push(v);
+            j++;
+          }
+        }
+#ifdef ASSERT
+        {
+          bool ignored_will_link;
+          ciSignature* declared_signature = NULL;
+          ciMethod* real_target = method()->get_method_at_bci(bci(), ignored_will_link, &declared_signature);
+          assert(s == obj_args->length() || real_target->is_method_handle_intrinsic(), "missed on arg?");
+        }
+#endif
+      }
+      profile_call(callee, recv, holder_known ? callee->holder() : NULL, obj_args, true);
     }
   }
 
@@ -4251,8 +4308,8 @@ void GraphBuilder::print_stats() {
 }
 #endif // PRODUCT
 
-void GraphBuilder::profile_call(ciMethod* callee, Value recv, ciKlass* known_holder) {
-  append(new ProfileCall(method(), bci(), callee, recv, known_holder));
+void GraphBuilder::profile_call(ciMethod* callee, Value recv, ciKlass* known_holder, Values* obj_args, bool inlined) {
+  append(new ProfileCall(method(), bci(), callee, recv, known_holder, obj_args, inlined));
 }
 
 void GraphBuilder::profile_invocation(ciMethod* callee, ValueStack* state) {
diff --git a/src/share/vm/c1/c1_GraphBuilder.hpp b/src/share/vm/c1/c1_GraphBuilder.hpp
index ae5afd4e0..217da78bb 100644
--- a/src/share/vm/c1/c1_GraphBuilder.hpp
+++ b/src/share/vm/c1/c1_GraphBuilder.hpp
@@ -374,7 +374,7 @@ class GraphBuilder VALUE_OBJ_CLASS_SPEC {
 
   void print_inlining(ciMethod* callee, const char* msg = NULL, bool success = true);
 
-  void profile_call(ciMethod* callee, Value recv, ciKlass* predicted_holder);
+  void profile_call(ciMethod* callee, Value recv, ciKlass* predicted_holder, Values* obj_args, bool inlined);
   void profile_invocation(ciMethod* inlinee, ValueStack* state);
 
   // Shortcuts to profiling control.
@@ -386,6 +386,9 @@ class GraphBuilder VALUE_OBJ_CLASS_SPEC {
   bool profile_inlined_calls() { return _compilation->profile_inlined_calls(); }
   bool profile_checkcasts()    { return _compilation->profile_checkcasts();    }
 
+  Values* args_list_for_profiling(int& start, bool may_have_receiver);
+  Values* collect_args_for_profiling(Values* args, bool may_have_receiver);
+
  public:
   NOT_PRODUCT(void print_stats();)
 
diff --git a/src/share/vm/c1/c1_Instruction.cpp b/src/share/vm/c1/c1_Instruction.cpp
index a026b4cbe..e5829611e 100644
--- a/src/share/vm/c1/c1_Instruction.cpp
+++ b/src/share/vm/c1/c1_Instruction.cpp
@@ -104,6 +104,14 @@ void Instruction::state_values_do(ValueVisitor* f) {
   }
 }
 
+ciType* Instruction::exact_type() const {
+  ciType* t =  declared_type();
+  if (t != NULL && t->is_klass()) {
+    return t->as_klass()->exact_klass();
+  }
+  return NULL;
+}
+
 
 #ifndef PRODUCT
 void Instruction::check_state(ValueStack* state) {
@@ -135,9 +143,7 @@ void Instruction::print(InstructionPrinter& ip) {
 
 // perform constant and interval tests on index value
 bool AccessIndexed::compute_needs_range_check() {
-
   if (length()) {
-
     Constant* clength = length()->as_Constant();
     Constant* cindex = index()->as_Constant();
     if (clength && cindex) {
@@ -157,34 +163,8 @@ bool AccessIndexed::compute_needs_range_check() {
 }
 
 
-ciType* Local::exact_type() const {
-  ciType* type = declared_type();
-
-  // for primitive arrays, the declared type is the exact type
-  if (type->is_type_array_klass()) {
-    return type;
-  } else if (type->is_instance_klass()) {
-    ciInstanceKlass* ik = (ciInstanceKlass*)type;
-    if (ik->is_loaded() && ik->is_final() && !ik->is_interface()) {
-      return type;
-    }
-  } else if (type->is_obj_array_klass()) {
-    ciObjArrayKlass* oak = (ciObjArrayKlass*)type;
-    ciType* base = oak->base_element_type();
-    if (base->is_instance_klass()) {
-      ciInstanceKlass* ik = base->as_instance_klass();
-      if (ik->is_loaded() && ik->is_final()) {
-        return type;
-      }
-    } else if (base->is_primitive_type()) {
-      return type;
-    }
-  }
-  return NULL;
-}
-
 ciType* Constant::exact_type() const {
-  if (type()->is_object()) {
+  if (type()->is_object() && type()->as_ObjectType()->is_loaded()) {
     return type()->as_ObjectType()->exact_type();
   }
   return NULL;
@@ -192,19 +172,18 @@ ciType* Constant::exact_type() const {
 
 ciType* LoadIndexed::exact_type() const {
   ciType* array_type = array()->exact_type();
-  if (array_type == NULL) {
-    return NULL;
-  }
-  assert(array_type->is_array_klass(), "what else?");
-  ciArrayKlass* ak = (ciArrayKlass*)array_type;
+  if (array_type != NULL) {
+    assert(array_type->is_array_klass(), "what else?");
+    ciArrayKlass* ak = (ciArrayKlass*)array_type;
 
-  if (ak->element_type()->is_instance_klass()) {
-    ciInstanceKlass* ik = (ciInstanceKlass*)ak->element_type();
-    if (ik->is_loaded() && ik->is_final()) {
-      return ik;
+    if (ak->element_type()->is_instance_klass()) {
+      ciInstanceKlass* ik = (ciInstanceKlass*)ak->element_type();
+      if (ik->is_loaded() && ik->is_final()) {
+        return ik;
+      }
     }
   }
-  return NULL;
+  return Instruction::exact_type();
 }
 
 
@@ -224,22 +203,6 @@ ciType* LoadField::declared_type() const {
 }
 
 
-ciType* LoadField::exact_type() const {
-  ciType* type = declared_type();
-  // for primitive arrays, the declared type is the exact type
-  if (type->is_type_array_klass()) {
-    return type;
-  }
-  if (type->is_instance_klass()) {
-    ciInstanceKlass* ik = (ciInstanceKlass*)type;
-    if (ik->is_loaded() && ik->is_final()) {
-      return type;
-    }
-  }
-  return NULL;
-}
-
-
 ciType* NewTypeArray::exact_type() const {
   return ciTypeArrayKlass::make(elt_type());
 }
@@ -264,16 +227,6 @@ ciType* CheckCast::declared_type() const {
   return klass();
 }
 
-ciType* CheckCast::exact_type() const {
-  if (klass()->is_instance_klass()) {
-    ciInstanceKlass* ik = (ciInstanceKlass*)klass();
-    if (ik->is_loaded() && ik->is_final()) {
-      return ik;
-    }
-  }
-  return NULL;
-}
-
 // Implementation of ArithmeticOp
 
 bool ArithmeticOp::is_commutative() const {
diff --git a/src/share/vm/c1/c1_Instruction.hpp b/src/share/vm/c1/c1_Instruction.hpp
index 9563b720a..466d814c6 100644
--- a/src/share/vm/c1/c1_Instruction.hpp
+++ b/src/share/vm/c1/c1_Instruction.hpp
@@ -322,6 +322,36 @@ class Instruction: public CompilationResourceObj {
     _type = type;
   }
 
+  // Helper class to keep track of which arguments need a null check
+  class ArgsNonNullState {
+  private:
+    int _nonnull_state; // mask identifying which args are nonnull
+  public:
+    ArgsNonNullState()
+      : _nonnull_state(AllBits) {}
+
+    // Does argument number i needs a null check?
+    bool arg_needs_null_check(int i) const {
+      // No data is kept for arguments starting at position 33 so
+      // conservatively assume that they need a null check.
+      if (i >= 0 && i < (int)sizeof(_nonnull_state) * BitsPerByte) {
+        return is_set_nth_bit(_nonnull_state, i);
+      }
+      return true;
+    }
+
+    // Set whether argument number i needs a null check or not
+    void set_arg_needs_null_check(int i, bool check) {
+      if (i >= 0 && i < (int)sizeof(_nonnull_state) * BitsPerByte) {
+        if (check) {
+          _nonnull_state |= nth_bit(i);
+        } else {
+          _nonnull_state &= ~(nth_bit(i));
+        }
+      }
+    }
+  };
+
  public:
   void* operator new(size_t size) throw() {
     Compilation* c = Compilation::current();
@@ -566,7 +596,7 @@ class Instruction: public CompilationResourceObj {
   virtual void other_values_do(ValueVisitor* f)   { /* usually no other - override on demand */ }
           void       values_do(ValueVisitor* f)   { input_values_do(f); state_values_do(f); other_values_do(f); }
 
-  virtual ciType* exact_type() const             { return NULL; }
+  virtual ciType* exact_type() const;
   virtual ciType* declared_type() const          { return NULL; }
 
   // hashing
@@ -689,7 +719,6 @@ LEAF(Local, Instruction)
   int java_index() const                         { return _java_index; }
 
   virtual ciType* declared_type() const          { return _declared_type; }
-  virtual ciType* exact_type() const;
 
   // generic
   virtual void input_values_do(ValueVisitor* f)   { /* no values */ }
@@ -806,7 +835,6 @@ LEAF(LoadField, AccessField)
   {}
 
   ciType* declared_type() const;
-  ciType* exact_type() const;
 
   // generic
   HASHING2(LoadField, !needs_patching() && !field()->is_volatile(), obj()->subst(), offset())  // cannot be eliminated if needs patching or if volatile
@@ -1299,6 +1327,7 @@ BASE(NewArray, StateSplit)
 
   virtual bool needs_exception_state() const     { return false; }
 
+  ciType* exact_type() const                     { return NULL; }
   ciType* declared_type() const;
 
   // generic
@@ -1422,7 +1451,6 @@ LEAF(CheckCast, TypeCheck)
   }
 
   ciType* declared_type() const;
-  ciType* exact_type() const;
 };
 
 
@@ -1490,7 +1518,7 @@ LEAF(Intrinsic, StateSplit)
   vmIntrinsics::ID _id;
   Values*          _args;
   Value            _recv;
-  int              _nonnull_state; // mask identifying which args are nonnull
+  ArgsNonNullState _nonnull_state;
 
  public:
   // preserves_state can be set to true for Intrinsics
@@ -1511,7 +1539,6 @@ LEAF(Intrinsic, StateSplit)
   , _id(id)
   , _args(args)
   , _recv(NULL)
-  , _nonnull_state(AllBits)
   {
     assert(args != NULL, "args must exist");
     ASSERT_VALUES
@@ -1537,21 +1564,12 @@ LEAF(Intrinsic, StateSplit)
   Value receiver() const                         { assert(has_receiver(), "must have receiver"); return _recv; }
   bool preserves_state() const                   { return check_flag(PreservesStateFlag); }
 
-  bool arg_needs_null_check(int i) {
-    if (i >= 0 && i < (int)sizeof(_nonnull_state) * BitsPerByte) {
-      return is_set_nth_bit(_nonnull_state, i);
-    }
-    return true;
+  bool arg_needs_null_check(int i) const {
+    return _nonnull_state.arg_needs_null_check(i);
   }
 
   void set_arg_needs_null_check(int i, bool check) {
-    if (i >= 0 && i < (int)sizeof(_nonnull_state) * BitsPerByte) {
-      if (check) {
-        _nonnull_state |= nth_bit(i);
-      } else {
-        _nonnull_state &= ~(nth_bit(i));
-      }
-    }
+    _nonnull_state.set_arg_needs_null_check(i, check);
   }
 
   // generic
@@ -2450,34 +2468,55 @@ LEAF(UnsafePrefetchWrite, UnsafePrefetch)
 
 LEAF(ProfileCall, Instruction)
  private:
-  ciMethod* _method;
-  int       _bci_of_invoke;
-  ciMethod* _callee;         // the method that is called at the given bci
-  Value     _recv;
-  ciKlass*  _known_holder;
+  ciMethod*        _method;
+  int              _bci_of_invoke;
+  ciMethod*        _callee;         // the method that is called at the given bci
+  Value            _recv;
+  ciKlass*         _known_holder;
+  Values*          _obj_args;       // arguments for type profiling
+  ArgsNonNullState _nonnull_state;  // Do we know whether some arguments are never null?
+  bool             _inlined;        // Are we profiling a call that is inlined
 
  public:
-  ProfileCall(ciMethod* method, int bci, ciMethod* callee, Value recv, ciKlass* known_holder)
+  ProfileCall(ciMethod* method, int bci, ciMethod* callee, Value recv, ciKlass* known_holder, Values* obj_args, bool inlined)
     : Instruction(voidType)
     , _method(method)
     , _bci_of_invoke(bci)
     , _callee(callee)
     , _recv(recv)
     , _known_holder(known_holder)
+    , _obj_args(obj_args)
+    , _inlined(inlined)
   {
     // The ProfileCall has side-effects and must occur precisely where located
     pin();
   }
 
-  ciMethod* method()      { return _method; }
-  int bci_of_invoke()     { return _bci_of_invoke; }
-  ciMethod* callee()      { return _callee; }
-  Value recv()            { return _recv; }
-  ciKlass* known_holder() { return _known_holder; }
+  ciMethod* method()             const { return _method; }
+  int bci_of_invoke()            const { return _bci_of_invoke; }
+  ciMethod* callee()             const { return _callee; }
+  Value recv()                   const { return _recv; }
+  ciKlass* known_holder()        const { return _known_holder; }
+  int nb_profiled_args()         const { return _obj_args == NULL ? 0 : _obj_args->length(); }
+  Value profiled_arg_at(int i)   const { return _obj_args->at(i); }
+  bool arg_needs_null_check(int i) const {
+    return _nonnull_state.arg_needs_null_check(i);
+  }
+  bool inlined()                 const { return _inlined; }
 
-  virtual void input_values_do(ValueVisitor* f)   { if (_recv != NULL) f->visit(&_recv); }
-};
+  void set_arg_needs_null_check(int i, bool check) {
+    _nonnull_state.set_arg_needs_null_check(i, check);
+  }
 
+  virtual void input_values_do(ValueVisitor* f)   {
+    if (_recv != NULL) {
+      f->visit(&_recv);
+    }
+    for (int i = 0; i < nb_profiled_args(); i++) {
+      f->visit(_obj_args->adr_at(i));
+    }
+  }
+};
 
 // Call some C runtime function that doesn't safepoint,
 // optionally passing the current thread as the first argument.
diff --git a/src/share/vm/c1/c1_InstructionPrinter.cpp b/src/share/vm/c1/c1_InstructionPrinter.cpp
index cfca00ab2..c15538116 100644
--- a/src/share/vm/c1/c1_InstructionPrinter.cpp
+++ b/src/share/vm/c1/c1_InstructionPrinter.cpp
@@ -892,6 +892,14 @@ void InstructionPrinter::do_ProfileCall(ProfileCall* x) {
   if (x->known_holder() != NULL) {
     output()->print(", ");
     print_klass(x->known_holder());
+    output()->print(" ");
+  }
+  for (int i = 0; i < x->nb_profiled_args(); i++) {
+    if (i > 0) output()->print(", ");
+    print_value(x->profiled_arg_at(i));
+    if (x->arg_needs_null_check(i)) {
+      output()->print(" [NC]");
+    }
   }
   output()->put(')');
 }
diff --git a/src/share/vm/c1/c1_LIR.cpp b/src/share/vm/c1/c1_LIR.cpp
index cb0ceab90..f40d91333 100644
--- a/src/share/vm/c1/c1_LIR.cpp
+++ b/src/share/vm/c1/c1_LIR.cpp
@@ -1001,6 +1001,17 @@ void LIR_OpVisitState::visit(LIR_Op* op) {
       assert(opProfileCall->_tmp1->is_valid(), "used");  do_temp(opProfileCall->_tmp1);
       break;
     }
+
+// LIR_OpProfileType:
+    case lir_profile_type: {
+      assert(op->as_OpProfileType() != NULL, "must be");
+      LIR_OpProfileType* opProfileType = (LIR_OpProfileType*)op;
+
+      do_input(opProfileType->_mdp); do_temp(opProfileType->_mdp);
+      do_input(opProfileType->_obj);
+      do_temp(opProfileType->_tmp);
+      break;
+    }
   default:
     ShouldNotReachHere();
   }
@@ -1151,6 +1162,10 @@ void LIR_OpProfileCall::emit_code(LIR_Assembler* masm) {
   masm->emit_profile_call(this);
 }
 
+void LIR_OpProfileType::emit_code(LIR_Assembler* masm) {
+  masm->emit_profile_type(this);
+}
+
 // LIR_List
 LIR_List::LIR_List(Compilation* compilation, BlockBegin* block)
   : _operations(8)
@@ -1803,6 +1818,8 @@ const char * LIR_Op::name() const {
      case lir_cas_int:               s = "cas_int";      break;
      // LIR_OpProfileCall
      case lir_profile_call:          s = "profile_call";  break;
+     // LIR_OpProfileType
+     case lir_profile_type:          s = "profile_type";  break;
      // LIR_OpAssert
 #ifdef ASSERT
      case lir_assert:                s = "assert";        break;
@@ -2086,6 +2103,15 @@ void LIR_OpProfileCall::print_instr(outputStream* out) const {
   tmp1()->print(out);          out->print(" ");
 }
 
+// LIR_OpProfileType
+void LIR_OpProfileType::print_instr(outputStream* out) const {
+  out->print("exact = "); exact_klass()->print_name_on(out);
+  out->print("current = "); ciTypeEntries::print_ciklass(out, current_klass());
+  mdp()->print(out);          out->print(" ");
+  obj()->print(out);          out->print(" ");
+  tmp()->print(out);          out->print(" ");
+}
+
 #endif // PRODUCT
 
 // Implementation of LIR_InsertionBuffer
diff --git a/src/share/vm/c1/c1_LIR.hpp b/src/share/vm/c1/c1_LIR.hpp
index d0dca72f3..a339af509 100644
--- a/src/share/vm/c1/c1_LIR.hpp
+++ b/src/share/vm/c1/c1_LIR.hpp
@@ -882,6 +882,7 @@ class    LIR_OpLock;
 class    LIR_OpTypeCheck;
 class    LIR_OpCompareAndSwap;
 class    LIR_OpProfileCall;
+class    LIR_OpProfileType;
 #ifdef ASSERT
 class    LIR_OpAssert;
 #endif
@@ -1005,6 +1006,7 @@ enum LIR_Code {
   , end_opCompareAndSwap
   , begin_opMDOProfile
     , lir_profile_call
+    , lir_profile_type
   , end_opMDOProfile
   , begin_opAssert
     , lir_assert
@@ -1145,6 +1147,7 @@ class LIR_Op: public CompilationResourceObj {
   virtual LIR_OpTypeCheck* as_OpTypeCheck() { return NULL; }
   virtual LIR_OpCompareAndSwap* as_OpCompareAndSwap() { return NULL; }
   virtual LIR_OpProfileCall* as_OpProfileCall() { return NULL; }
+  virtual LIR_OpProfileType* as_OpProfileType() { return NULL; }
 #ifdef ASSERT
   virtual LIR_OpAssert* as_OpAssert() { return NULL; }
 #endif
@@ -1925,8 +1928,8 @@ class LIR_OpProfileCall : public LIR_Op {
 
  public:
   // Destroys recv
-  LIR_OpProfileCall(LIR_Code code, ciMethod* profiled_method, int profiled_bci, ciMethod* profiled_callee, LIR_Opr mdo, LIR_Opr recv, LIR_Opr t1, ciKlass* known_holder)
-    : LIR_Op(code, LIR_OprFact::illegalOpr, NULL)  // no result, no info
+  LIR_OpProfileCall(ciMethod* profiled_method, int profiled_bci, ciMethod* profiled_callee, LIR_Opr mdo, LIR_Opr recv, LIR_Opr t1, ciKlass* known_holder)
+    : LIR_Op(lir_profile_call, LIR_OprFact::illegalOpr, NULL)  // no result, no info
     , _profiled_method(profiled_method)
     , _profiled_bci(profiled_bci)
     , _profiled_callee(profiled_callee)
@@ -1948,6 +1951,45 @@ class LIR_OpProfileCall : public LIR_Op {
   virtual void print_instr(outputStream* out) const PRODUCT_RETURN;
 };
 
+// LIR_OpProfileType
+class LIR_OpProfileType : public LIR_Op {
+ friend class LIR_OpVisitState;
+
+ private:
+  LIR_Opr      _mdp;
+  LIR_Opr      _obj;
+  LIR_Opr      _tmp;
+  ciKlass*     _exact_klass;   // non NULL if we know the klass statically (no need to load it from _obj)
+  intptr_t     _current_klass; // what the profiling currently reports
+  bool         _not_null;      // true if we know statically that _obj cannot be null
+  bool         _no_conflict;   // true if we're profling parameters, _exact_klass is not NULL and we know
+                               // _exact_klass it the only possible type for this parameter in any context.
+
+ public:
+  // Destroys recv
+  LIR_OpProfileType(LIR_Opr mdp, LIR_Opr obj, ciKlass* exact_klass, intptr_t current_klass, LIR_Opr tmp, bool not_null, bool no_conflict)
+    : LIR_Op(lir_profile_type, LIR_OprFact::illegalOpr, NULL)  // no result, no info
+    , _mdp(mdp)
+    , _obj(obj)
+    , _exact_klass(exact_klass)
+    , _current_klass(current_klass)
+    , _tmp(tmp)
+    , _not_null(not_null)
+    , _no_conflict(no_conflict) { }
+
+  LIR_Opr      mdp()              const             { return _mdp;              }
+  LIR_Opr      obj()              const             { return _obj;              }
+  LIR_Opr      tmp()              const             { return _tmp;              }
+  ciKlass*     exact_klass()      const             { return _exact_klass;      }
+  intptr_t     current_klass()    const             { return _current_klass;    }
+  bool         not_null()         const             { return _not_null;         }
+  bool         no_conflict()      const             { return _no_conflict;      }
+
+  virtual void emit_code(LIR_Assembler* masm);
+  virtual LIR_OpProfileType* as_OpProfileType() { return this; }
+  virtual void print_instr(outputStream* out) const PRODUCT_RETURN;
+};
+
 class LIR_InsertionBuffer;
 
 //--------------------------------LIR_List---------------------------------------------------
@@ -2247,7 +2289,10 @@ class LIR_List: public CompilationResourceObj {
                   ciMethod* profiled_method, int profiled_bci);
   // MethodData* profiling
   void profile_call(ciMethod* method, int bci, ciMethod* callee, LIR_Opr mdo, LIR_Opr recv, LIR_Opr t1, ciKlass* cha_klass) {
-    append(new LIR_OpProfileCall(lir_profile_call, method, bci, callee, mdo, recv, t1, cha_klass));
+    append(new LIR_OpProfileCall(method, bci, callee, mdo, recv, t1, cha_klass));
+  }
+  void profile_type(LIR_Address* mdp, LIR_Opr obj, ciKlass* exact_klass, intptr_t current_klass, LIR_Opr tmp, bool not_null, bool no_conflict) {
+    append(new LIR_OpProfileType(LIR_OprFact::address(mdp), obj, exact_klass, current_klass, tmp, not_null, no_conflict));
   }
 
   void xadd(LIR_Opr src, LIR_Opr add, LIR_Opr res, LIR_Opr tmp) { append(new LIR_Op2(lir_xadd, src, add, res, tmp)); }
diff --git a/src/share/vm/c1/c1_LIRAssembler.hpp b/src/share/vm/c1/c1_LIRAssembler.hpp
index 57df2725e..68f249a2a 100644
--- a/src/share/vm/c1/c1_LIRAssembler.hpp
+++ b/src/share/vm/c1/c1_LIRAssembler.hpp
@@ -208,6 +208,7 @@ class LIR_Assembler: public CompilationResourceObj {
   void emit_call(LIR_OpJavaCall* op);
   void emit_rtcall(LIR_OpRTCall* op);
   void emit_profile_call(LIR_OpProfileCall* op);
+  void emit_profile_type(LIR_OpProfileType* op);
   void emit_delay(LIR_OpDelay* op);
 
   void arith_op(LIR_Code code, LIR_Opr left, LIR_Opr right, LIR_Opr dest, CodeEmitInfo* info, bool pop_fpu_stack);
diff --git a/src/share/vm/c1/c1_LIRGenerator.cpp b/src/share/vm/c1/c1_LIRGenerator.cpp
index f71470e5e..3eedf1f31 100644
--- a/src/share/vm/c1/c1_LIRGenerator.cpp
+++ b/src/share/vm/c1/c1_LIRGenerator.cpp
@@ -2571,6 +2571,78 @@ void LIRGenerator::do_Goto(Goto* x) {
 }
 
 
+ciKlass* LIRGenerator::profile_arg_type(ciMethodData* md, int md_base_offset, int md_offset, intptr_t profiled_k, Value arg, LIR_Opr& mdp, bool not_null, ciKlass* signature_k) {
+  ciKlass* result = NULL;
+  bool do_null = !not_null && !TypeEntries::was_null_seen(profiled_k);
+  bool do_update = !TypeEntries::is_type_unknown(profiled_k);
+  // known not to be null or null bit already set and already set to
+  // unknown: nothing we can do to improve profiling
+  if (!do_null && !do_update) {
+    return result;
+  }
+
+  ciKlass* exact_klass = NULL;
+  Compilation* comp = Compilation::current();
+  if (do_update) {
+    // try to find exact type, using CHA if possible, so that loading
+    // the klass from the object can be avoided
+    ciType* type = arg->exact_type();
+    if (type == NULL) {
+      type = arg->declared_type();
+      type = comp->cha_exact_type(type);
+    }
+    assert(type == NULL || type->is_klass(), "type should be class");
+    exact_klass = (type != NULL && type->is_loaded()) ? (ciKlass*)type : NULL;
+
+    do_update = exact_klass == NULL || ciTypeEntries::valid_ciklass(profiled_k) != exact_klass;
+  }
+
+  if (!do_null && !do_update) {
+    return result;
+  }
+
+  ciKlass* exact_signature_k = NULL;
+  if (do_update) {
+    // Is the type from the signature exact (the only one possible)?
+    exact_signature_k = signature_k->exact_klass();
+    if (exact_signature_k == NULL) {
+      exact_signature_k = comp->cha_exact_type(signature_k);
+    } else {
+      result = exact_signature_k;
+      do_update = false;
+      // Known statically. No need to emit any code: prevent
+      // LIR_Assembler::emit_profile_type() from emitting useless code
+      profiled_k = ciTypeEntries::with_status(result, profiled_k);
+    }
+    if (exact_signature_k != NULL && exact_klass != exact_signature_k) {
+      assert(exact_klass == NULL, "arg and signature disagree?");
+      // sometimes the type of the signature is better than the best type
+      // the compiler has
+      exact_klass = exact_signature_k;
+      do_update = exact_klass == NULL || ciTypeEntries::valid_ciklass(profiled_k) != exact_klass;
+    }
+  }
+
+  if (!do_null && !do_update) {
+    return result;
+  }
+
+  if (mdp == LIR_OprFact::illegalOpr) {
+    mdp = new_register(T_METADATA);
+    __ metadata2reg(md->constant_encoding(), mdp);
+    if (md_base_offset != 0) {
+      LIR_Address* base_type_address = new LIR_Address(mdp, md_base_offset, T_ADDRESS);
+      mdp = new_pointer_register();
+      __ leal(LIR_OprFact::address(base_type_address), mdp);
+    }
+  }
+  LIRItem value(arg, this);
+  value.load_item();
+  __ profile_type(new LIR_Address(mdp, md_offset, T_METADATA),
+                  value.result(), exact_klass, profiled_k, new_pointer_register(), not_null, exact_signature_k != NULL);
+  return result;
+}
+
 void LIRGenerator::do_Base(Base* x) {
   __ std_entry(LIR_OprFact::illegalOpr);
   // Emit moves from physical registers / stack slots to virtual registers
@@ -3004,12 +3076,52 @@ void LIRGenerator::do_Intrinsic(Intrinsic* x) {
   }
 }
 
+void LIRGenerator::profile_arguments(ProfileCall* x) {
+  if (MethodData::profile_arguments()) {
+    int bci = x->bci_of_invoke();
+    ciMethodData* md = x->method()->method_data_or_null();
+    ciProfileData* data = md->bci_to_data(bci);
+    if (data->is_CallTypeData() || data->is_VirtualCallTypeData()) {
+      ByteSize extra = data->is_CallTypeData() ? CallTypeData::args_data_offset() : VirtualCallTypeData::args_data_offset();
+      int base_offset = md->byte_offset_of_slot(data, extra);
+      LIR_Opr mdp = LIR_OprFact::illegalOpr;
+      ciTypeStackSlotEntries* args = data->is_CallTypeData() ? ((ciCallTypeData*)data)->args() : ((ciVirtualCallTypeData*)data)->args();
+
+      Bytecodes::Code bc = x->method()->java_code_at_bci(bci);
+      int start = 0;
+      int stop = args->number_of_arguments();
+      if (x->nb_profiled_args() < stop) {
+        // if called through method handle invoke, some arguments may have been popped
+        stop = x->nb_profiled_args();
+      }
+      ciSignature* sig = x->callee()->signature();
+      // method handle call to virtual method
+      bool has_receiver = x->inlined() && !x->callee()->is_static() && !Bytecodes::has_receiver(bc);
+      ciSignatureStream sig_stream(sig, has_receiver ? x->callee()->holder() : NULL);
+      for (int i = 0; i < stop; i++) {
+        int off = in_bytes(TypeStackSlotEntries::type_offset(i)) - in_bytes(TypeStackSlotEntries::args_data_offset());
+        ciKlass* exact = profile_arg_type(md, base_offset, off,
+                                          args->type(i), x->profiled_arg_at(i+start), mdp,
+                                          !x->arg_needs_null_check(i+start), sig_stream.next_klass());
+        if (exact != NULL) {
+          md->set_argument_type(bci, i, exact);
+        }
+      }
+    }
+  }
+}
+
 void LIRGenerator::do_ProfileCall(ProfileCall* x) {
   // Need recv in a temporary register so it interferes with the other temporaries
   LIR_Opr recv = LIR_OprFact::illegalOpr;
   LIR_Opr mdo = new_register(T_OBJECT);
   // tmp is used to hold the counters on SPARC
   LIR_Opr tmp = new_pointer_register();
+
+  if (x->nb_profiled_args() > 0) {
+    profile_arguments(x);
+  }
+
   if (x->recv() != NULL) {
     LIRItem value(x->recv(), this);
     value.load_item();
diff --git a/src/share/vm/c1/c1_LIRGenerator.hpp b/src/share/vm/c1/c1_LIRGenerator.hpp
index 0a0292073..fce938a03 100644
--- a/src/share/vm/c1/c1_LIRGenerator.hpp
+++ b/src/share/vm/c1/c1_LIRGenerator.hpp
@@ -434,6 +434,8 @@ class LIRGenerator: public InstructionVisitor, public BlockClosure {
   void do_ThreadIDIntrinsic(Intrinsic* x);
   void do_ClassIDIntrinsic(Intrinsic* x);
 #endif
+  ciKlass* profile_arg_type(ciMethodData* md, int md_first_offset, int md_offset, intptr_t profiled_k, Value arg, LIR_Opr& mdp, bool not_null, ciKlass* signature_k);
+  void profile_arguments(ProfileCall* x);
 
  public:
   Compilation*  compilation() const              { return _compilation; }
diff --git a/src/share/vm/c1/c1_Optimizer.cpp b/src/share/vm/c1/c1_Optimizer.cpp
index 90dc27972..0be6099aa 100644
--- a/src/share/vm/c1/c1_Optimizer.cpp
+++ b/src/share/vm/c1/c1_Optimizer.cpp
@@ -657,6 +657,7 @@ class NullCheckEliminator: public ValueVisitor {
   void handle_Intrinsic       (Intrinsic* x);
   void handle_ExceptionObject (ExceptionObject* x);
   void handle_Phi             (Phi* x);
+  void handle_ProfileCall     (ProfileCall* x);
 };
 
 
@@ -715,7 +716,8 @@ void NullCheckVisitor::do_UnsafePutObject(UnsafePutObject* x) {}
 void NullCheckVisitor::do_UnsafeGetAndSetObject(UnsafeGetAndSetObject* x) {}
 void NullCheckVisitor::do_UnsafePrefetchRead (UnsafePrefetchRead*  x) {}
 void NullCheckVisitor::do_UnsafePrefetchWrite(UnsafePrefetchWrite* x) {}
-void NullCheckVisitor::do_ProfileCall    (ProfileCall*     x) { nce()->clear_last_explicit_null_check(); }
+void NullCheckVisitor::do_ProfileCall    (ProfileCall*     x) { nce()->clear_last_explicit_null_check();
+                                                                nce()->handle_ProfileCall(x); }
 void NullCheckVisitor::do_ProfileInvoke  (ProfileInvoke*   x) {}
 void NullCheckVisitor::do_RuntimeCall    (RuntimeCall*     x) {}
 void NullCheckVisitor::do_MemBar         (MemBar*          x) {}
@@ -1134,6 +1136,11 @@ void NullCheckEliminator::handle_Phi(Phi* x) {
   }
 }
 
+void NullCheckEliminator::handle_ProfileCall(ProfileCall* x) {
+  for (int i = 0; i < x->nb_profiled_args(); i++) {
+    x->set_arg_needs_null_check(i, !set_contains(x->profiled_arg_at(i)));
+  }
+}
 
 void Optimizer::eliminate_null_checks() {
   ResourceMark rm;
diff --git a/src/share/vm/c1/c1_RangeCheckElimination.hpp b/src/share/vm/c1/c1_RangeCheckElimination.hpp
index ae1a25568..1d7897bda 100644
--- a/src/share/vm/c1/c1_RangeCheckElimination.hpp
+++ b/src/share/vm/c1/c1_RangeCheckElimination.hpp
@@ -162,7 +162,7 @@ public:
     void do_UnsafePrefetchRead (UnsafePrefetchRead*  x) { /* nothing to do */ };
     void do_UnsafePrefetchWrite(UnsafePrefetchWrite* x) { /* nothing to do */ };
     void do_ProfileCall    (ProfileCall*     x) { /* nothing to do */ };
-    void do_ProfileInvoke  (ProfileInvoke*  x)  { /* nothing to do */ };
+    void do_ProfileInvoke  (ProfileInvoke*   x) { /* nothing to do */ };
     void do_RuntimeCall    (RuntimeCall*     x) { /* nothing to do */ };
     void do_MemBar         (MemBar*          x) { /* nothing to do */ };
     void do_RangeCheckPredicate(RangeCheckPredicate* x) { /* nothing to do */ };