8026054: New type profiling points: type of return values at calls

Summary: x86 interpreter and c1 type profiling for return values at calls
Reviewed-by: kvn, twisti

src/cpu/x86/vm/globals_x86.hpp

@@ -79,7 +79,7 @@ define_pd_global(bool, UseMembar,            false);
 // GC Ergo Flags
 define_pd_global(uintx, CMSYoungGenPerWorker, 64*M);  // default max size of CMS young gen, per GC worker thread
 
-define_pd_global(uintx, TypeProfileLevel, 1);
+define_pd_global(uintx, TypeProfileLevel, 11);
 
 #define ARCH_FLAGS(develop, product, diagnostic, experimental, notproduct) \
                                                                             \

src/cpu/x86/vm/interp_masm_x86_32.cpp

@@ -1095,7 +1095,7 @@ void InterpreterMacroAssembler::profile_arguments_type(Register mdp, Register ca
     return;
   }
 
-  if (MethodData::profile_arguments()) {
+  if (MethodData::profile_arguments() || MethodData::profile_return()) {
     Label profile_continue;
 
     test_method_data_pointer(mdp, profile_continue);
@@ -1105,35 +1105,95 @@ void InterpreterMacroAssembler::profile_arguments_type(Register mdp, Register ca
     cmpb(Address(mdp, in_bytes(DataLayout::tag_offset()) - off_to_start), is_virtual ? DataLayout::virtual_call_type_data_tag : DataLayout::call_type_data_tag);
     jcc(Assembler::notEqual, profile_continue);
 
-    Label done;
-    int off_to_args = in_bytes(TypeStackSlotEntries::args_data_offset());
-    addptr(mdp, off_to_args);
-
-    for (int i = 0; i < TypeProfileArgsLimit; i++) {
-      if (i > 0) {
-        movl(tmp, Address(mdp, in_bytes(TypeStackSlotEntries::cell_count_offset())-off_to_args));
-        subl(tmp, i*TypeStackSlotEntries::per_arg_count());
-        cmpl(tmp, TypeStackSlotEntries::per_arg_count());
-        jcc(Assembler::less, done);
+    if (MethodData::profile_arguments()) {
+      Label done;
+      int off_to_args = in_bytes(TypeEntriesAtCall::args_data_offset());
+      addptr(mdp, off_to_args);
+
+      for (int i = 0; i < TypeProfileArgsLimit; i++) {
+        if (i > 0 || MethodData::profile_return()) {
+          // If return value type is profiled we may have no argument to profile
+          movl(tmp, Address(mdp, in_bytes(TypeEntriesAtCall::cell_count_offset())-off_to_args));
+          subl(tmp, i*TypeStackSlotEntries::per_arg_count());
+          cmpl(tmp, TypeStackSlotEntries::per_arg_count());
+          jcc(Assembler::less, done);
+        }
+        movptr(tmp, Address(callee, Method::const_offset()));
+        load_unsigned_short(tmp, Address(tmp, ConstMethod::size_of_parameters_offset()));
+        // stack offset o (zero based) from the start of the argument
+        // list, for n arguments translates into offset n - o - 1 from
+        // the end of the argument list
+        subl(tmp, Address(mdp, in_bytes(TypeEntriesAtCall::stack_slot_offset(i))-off_to_args));
+        subl(tmp, 1);
+        Address arg_addr = argument_address(tmp);
+        movptr(tmp, arg_addr);
+
+        Address mdo_arg_addr(mdp, in_bytes(TypeEntriesAtCall::argument_type_offset(i))-off_to_args);
+        profile_obj_type(tmp, mdo_arg_addr);
+
+        int to_add = in_bytes(TypeStackSlotEntries::per_arg_size());
+        addptr(mdp, to_add);
+        off_to_args += to_add;
       }
-      movptr(tmp, Address(callee, Method::const_offset()));
-      load_unsigned_short(tmp, Address(tmp, ConstMethod::size_of_parameters_offset()));
-      subl(tmp, Address(mdp, in_bytes(TypeStackSlotEntries::stack_slot_offset(i))-off_to_args));
-      subl(tmp, 1);
-      Address arg_addr = argument_address(tmp);
-      movptr(tmp, arg_addr);
-
-      Address mdo_arg_addr(mdp, in_bytes(TypeStackSlotEntries::type_offset(i))-off_to_args);
-      profile_obj_type(tmp, mdo_arg_addr);
-
-      int to_add = in_bytes(TypeStackSlotEntries::per_arg_size());
-      addptr(mdp, to_add);
-      off_to_args += to_add;
+
+      if (MethodData::profile_return()) {
+        movl(tmp, Address(mdp, in_bytes(TypeEntriesAtCall::cell_count_offset())-off_to_args));
+        subl(tmp, TypeProfileArgsLimit*TypeStackSlotEntries::per_arg_count());
+      }
+
+      bind(done);
+
+      if (MethodData::profile_return()) {
+        // We're right after the type profile for the last
+        // argument. tmp is the number of cell left in the
+        // CallTypeData/VirtualCallTypeData to reach its end. Non null
+        // if there's a return to profile.
+        assert(ReturnTypeEntry::static_cell_count() < TypeStackSlotEntries::per_arg_count(), "can't move past ret type");
+        shll(tmp, exact_log2(DataLayout::cell_size));
+        addptr(mdp, tmp);
+      }
+      movptr(Address(rbp, frame::interpreter_frame_mdx_offset * wordSize), mdp);
+    } else {
+      assert(MethodData::profile_return(), "either profile call args or call ret");
+      update_mdp_by_constant(mdp, in_bytes(ReturnTypeEntry::size()));
     }
 
-    bind(done);
+    // mdp points right after the end of the
+    // CallTypeData/VirtualCallTypeData, right after the cells for the
+    // return value type if there's one
+
+    bind(profile_continue);
+  }
+}
+
+void InterpreterMacroAssembler::profile_return_type(Register mdp, Register ret, Register tmp) {
+  assert_different_registers(mdp, ret, tmp, rsi);
+  if (ProfileInterpreter && MethodData::profile_return()) {
+    Label profile_continue, done;
+
+    test_method_data_pointer(mdp, profile_continue);
+
+    if (MethodData::profile_return_jsr292_only()) {
+      // If we don't profile all invoke bytecodes we must make sure
+      // it's a bytecode we indeed profile. We can't go back to the
+      // begining of the ProfileData we intend to update to check its
+      // type because we're right after it and we don't known its
+      // length
+      Label do_profile;
+      cmpb(Address(rsi, 0), Bytecodes::_invokedynamic);
+      jcc(Assembler::equal, do_profile);
+      cmpb(Address(rsi, 0), Bytecodes::_invokehandle);
+      jcc(Assembler::equal, do_profile);
+      get_method(tmp);
+      cmpb(Address(tmp, Method::intrinsic_id_offset_in_bytes()), vmIntrinsics::_compiledLambdaForm);
+      jcc(Assembler::notEqual, profile_continue);
+
+      bind(do_profile);
+    }
 
-    movptr(Address(rbp, frame::interpreter_frame_mdx_offset * wordSize), mdp);
+    Address mdo_ret_addr(mdp, -in_bytes(ReturnTypeEntry::size()));
+    mov(tmp, ret);
+    profile_obj_type(tmp, mdo_ret_addr);
 
     bind(profile_continue);
   }

src/cpu/x86/vm/interp_masm_x86_32.hpp

@@ -217,6 +217,7 @@ class InterpreterMacroAssembler: public MacroAssembler {
   void profile_not_taken_branch(Register mdp);
   void profile_obj_type(Register obj, const Address& mdo_addr);
   void profile_arguments_type(Register mdp, Register callee, Register tmp, bool is_virtual);
+  void profile_return_type(Register mdp, Register ret, Register tmp);
   void profile_call(Register mdp);
   void profile_final_call(Register mdp);
   void profile_virtual_call(Register receiver, Register mdp, Register scratch2,

src/cpu/x86/vm/interp_masm_x86_64.cpp

@@ -1120,7 +1120,7 @@ void InterpreterMacroAssembler::profile_arguments_type(Register mdp, Register ca
     return;
   }
 
-  if (MethodData::profile_arguments()) {
+  if (MethodData::profile_arguments() || MethodData::profile_return()) {
     Label profile_continue;
 
     test_method_data_pointer(mdp, profile_continue);
@@ -1130,35 +1130,92 @@ void InterpreterMacroAssembler::profile_arguments_type(Register mdp, Register ca
     cmpb(Address(mdp, in_bytes(DataLayout::tag_offset()) - off_to_start), is_virtual ? DataLayout::virtual_call_type_data_tag : DataLayout::call_type_data_tag);
     jcc(Assembler::notEqual, profile_continue);
 
-    Label done;
-    int off_to_args = in_bytes(TypeStackSlotEntries::args_data_offset());
-    addptr(mdp, off_to_args);
-
-    for (int i = 0; i < TypeProfileArgsLimit; i++) {
-      if (i > 0) {
-        movq(tmp, Address(mdp, in_bytes(TypeStackSlotEntries::cell_count_offset())-off_to_args));
-        subl(tmp, i*TypeStackSlotEntries::per_arg_count());
-        cmpl(tmp, TypeStackSlotEntries::per_arg_count());
-        jcc(Assembler::less, done);
+    if (MethodData::profile_arguments()) {
+      Label done;
+      int off_to_args = in_bytes(TypeEntriesAtCall::args_data_offset());
+      addptr(mdp, off_to_args);
+
+      for (int i = 0; i < TypeProfileArgsLimit; i++) {
+        if (i > 0 || MethodData::profile_return()) {
+          // If return value type is profiled we may have no argument to profile
+          movq(tmp, Address(mdp, in_bytes(TypeEntriesAtCall::cell_count_offset())-off_to_args));
+          subl(tmp, i*TypeStackSlotEntries::per_arg_count());
+          cmpl(tmp, TypeStackSlotEntries::per_arg_count());
+          jcc(Assembler::less, done);
+        }
+        movptr(tmp, Address(callee, Method::const_offset()));
+        load_unsigned_short(tmp, Address(tmp, ConstMethod::size_of_parameters_offset()));
+        subq(tmp, Address(mdp, in_bytes(TypeEntriesAtCall::stack_slot_offset(i))-off_to_args));
+        subl(tmp, 1);
+        Address arg_addr = argument_address(tmp);
+        movptr(tmp, arg_addr);
+
+        Address mdo_arg_addr(mdp, in_bytes(TypeEntriesAtCall::argument_type_offset(i))-off_to_args);
+        profile_obj_type(tmp, mdo_arg_addr);
+
+        int to_add = in_bytes(TypeStackSlotEntries::per_arg_size());
+        addptr(mdp, to_add);
+        off_to_args += to_add;
       }
-      movptr(tmp, Address(callee, Method::const_offset()));
-      load_unsigned_short(tmp, Address(tmp, ConstMethod::size_of_parameters_offset()));
-      subq(tmp, Address(mdp, in_bytes(TypeStackSlotEntries::stack_slot_offset(i))-off_to_args));
-      subl(tmp, 1);
-      Address arg_addr = argument_address(tmp);
-      movptr(tmp, arg_addr);
-
-      Address mdo_arg_addr(mdp, in_bytes(TypeStackSlotEntries::type_offset(i))-off_to_args);
-      profile_obj_type(tmp, mdo_arg_addr);
-
-      int to_add = in_bytes(TypeStackSlotEntries::per_arg_size());
-      addptr(mdp, to_add);
-      off_to_args += to_add;
+
+      if (MethodData::profile_return()) {
+        movq(tmp, Address(mdp, in_bytes(TypeEntriesAtCall::cell_count_offset())-off_to_args));
+        subl(tmp, TypeProfileArgsLimit*TypeStackSlotEntries::per_arg_count());
+      }
+
+      bind(done);
+
+      if (MethodData::profile_return()) {
+        // We're right after the type profile for the last
+        // argument. tmp is the number of cell left in the
+        // CallTypeData/VirtualCallTypeData to reach its end. Non null
+        // if there's a return to profile.
+        assert(ReturnTypeEntry::static_cell_count() < TypeStackSlotEntries::per_arg_count(), "can't move past ret type");
+        shll(tmp, exact_log2(DataLayout::cell_size));
+        addptr(mdp, tmp);
+      }
+      movptr(Address(rbp, frame::interpreter_frame_mdx_offset * wordSize), mdp);
+    } else {
+      assert(MethodData::profile_return(), "either profile call args or call ret");
+      update_mdp_by_constant(mdp, in_bytes(ReturnTypeEntry::size()));
     }
 
-    bind(done);
+    // mdp points right after the end of the
+    // CallTypeData/VirtualCallTypeData, right after the cells for the
+    // return value type if there's one
+
+    bind(profile_continue);
+  }
+}
+
+void InterpreterMacroAssembler::profile_return_type(Register mdp, Register ret, Register tmp) {
+  assert_different_registers(mdp, ret, tmp, r13);
+  if (ProfileInterpreter && MethodData::profile_return()) {
+    Label profile_continue, done;
+
+    test_method_data_pointer(mdp, profile_continue);
+
+    if (MethodData::profile_return_jsr292_only()) {
+      // If we don't profile all invoke bytecodes we must make sure
+      // it's a bytecode we indeed profile. We can't go back to the
+      // begining of the ProfileData we intend to update to check its
+      // type because we're right after it and we don't known its
+      // length
+      Label do_profile;
+      cmpb(Address(r13, 0), Bytecodes::_invokedynamic);
+      jcc(Assembler::equal, do_profile);
+      cmpb(Address(r13, 0), Bytecodes::_invokehandle);
+      jcc(Assembler::equal, do_profile);
+      get_method(tmp);
+      cmpb(Address(tmp, Method::intrinsic_id_offset_in_bytes()), vmIntrinsics::_compiledLambdaForm);
+      jcc(Assembler::notEqual, profile_continue);
+
+      bind(do_profile);
+    }
 
-    movptr(Address(rbp, frame::interpreter_frame_mdx_offset * wordSize), mdp);
+    Address mdo_ret_addr(mdp, -in_bytes(ReturnTypeEntry::size()));
+    mov(tmp, ret);
+    profile_obj_type(tmp, mdo_ret_addr);
 
     bind(profile_continue);
   }

src/cpu/x86/vm/interp_masm_x86_64.hpp

@@ -226,6 +226,7 @@ class InterpreterMacroAssembler: public MacroAssembler {
   void profile_not_taken_branch(Register mdp);
   void profile_obj_type(Register obj, const Address& mdo_addr);
   void profile_arguments_type(Register mdp, Register callee, Register tmp, bool is_virtual);
+  void profile_return_type(Register mdp, Register ret, Register tmp);
   void profile_call(Register mdp);
   void profile_final_call(Register mdp);
   void profile_virtual_call(Register receiver, Register mdp,

src/cpu/x86/vm/templateInterpreter_x86_32.cpp

@@ -194,6 +194,12 @@ address TemplateInterpreterGenerator::generate_return_entry_for(TosState state,
   __ restore_bcp();
   __ restore_locals();
 
+  if (incoming_state == atos) {
+    Register mdp = rbx;
+    Register tmp = rcx;
+    __ profile_return_type(mdp, rax, tmp);
+  }
+
   Label L_got_cache, L_giant_index;
   if (EnableInvokeDynamic) {
     __ cmpb(Address(rsi, 0), Bytecodes::_invokedynamic);

src/cpu/x86/vm/templateInterpreter_x86_64.cpp

@@ -177,6 +177,12 @@ address TemplateInterpreterGenerator::generate_return_entry_for(TosState state,
   __ restore_bcp();
   __ restore_locals();
 
+  if (state == atos) {
+    Register mdp = rbx;
+    Register tmp = rcx;
+    __ profile_return_type(mdp, rax, tmp);
+  }
+
   Label L_got_cache, L_giant_index;
   if (EnableInvokeDynamic) {
     __ cmpb(Address(r13, 0), Bytecodes::_invokedynamic);

src/share/vm/c1/c1_Canonicalizer.cpp

@@ -935,6 +935,7 @@ void Canonicalizer::do_UnsafeGetAndSetObject(UnsafeGetAndSetObject* x) {}
 void Canonicalizer::do_UnsafePrefetchRead (UnsafePrefetchRead*  x) {}
 void Canonicalizer::do_UnsafePrefetchWrite(UnsafePrefetchWrite* x) {}
 void Canonicalizer::do_ProfileCall(ProfileCall* x) {}
+void Canonicalizer::do_ProfileReturnType(ProfileReturnType* x) {}
 void Canonicalizer::do_ProfileInvoke(ProfileInvoke* x) {}
 void Canonicalizer::do_RuntimeCall(RuntimeCall* x) {}
 void Canonicalizer::do_RangeCheckPredicate(RangeCheckPredicate* x) {}

src/share/vm/c1/c1_Canonicalizer.hpp

@@ -104,6 +104,7 @@ class Canonicalizer: InstructionVisitor {
   virtual void do_UnsafePrefetchRead (UnsafePrefetchRead*  x);
   virtual void do_UnsafePrefetchWrite(UnsafePrefetchWrite* x);
   virtual void do_ProfileCall    (ProfileCall*     x);
+  virtual void do_ProfileReturnType (ProfileReturnType*  x);
   virtual void do_ProfileInvoke  (ProfileInvoke*   x);
   virtual void do_RuntimeCall    (RuntimeCall*     x);
   virtual void do_MemBar         (MemBar*          x);

src/share/vm/c1/c1_GraphBuilder.cpp

@@ -1466,9 +1466,22 @@ void GraphBuilder::method_return(Value x) {
     // State at end of inlined method is the state of the caller
     // without the method parameters on stack, including the
     // return value, if any, of the inlined method on operand stack.
+    int invoke_bci = state()->caller_state()->bci();
     set_state(state()->caller_state()->copy_for_parsing());
     if (x != NULL) {
       state()->push(x->type(), x);
+      if (profile_calls() && MethodData::profile_return() && x->type()->is_object_kind()) {
+        ciMethod* caller = state()->scope()->method();
+        ciMethodData* md = caller->method_data_or_null();
+        ciProfileData* data = md->bci_to_data(invoke_bci);
+        if (data->is_CallTypeData() || data->is_VirtualCallTypeData()) {
+          bool has_return = data->is_CallTypeData() ? ((ciCallTypeData*)data)->has_return() : ((ciVirtualCallTypeData*)data)->has_return();
+          // May not be true in case of an inlined call through a method handle intrinsic.
+          if (has_return) {
+            profile_return_type(x, method(), caller, invoke_bci);
+          }
+        }
+      }
     }
     Goto* goto_callee = new Goto(continuation(), false);
 
@@ -2008,6 +2021,9 @@ void GraphBuilder::invoke(Bytecodes::Code code) {
       push(result_type, result);
     }
   }
+  if (profile_calls() && MethodData::profile_return() && result_type->is_object_kind()) {
+    profile_return_type(result, target);
+  }
 }
 
 
@@ -3556,6 +3572,10 @@ bool GraphBuilder::try_inline_intrinsics(ciMethod* callee) {
   Value value = append_split(result);
   if (result_type != voidType) push(result_type, value);
 
+  if (callee != method() && profile_calls() && MethodData::profile_return() && result_type->is_object_kind()) {
+    profile_return_type(result, callee);
+  }
+
   // done
   return true;
 }
@@ -4312,6 +4332,21 @@ void GraphBuilder::profile_call(ciMethod* callee, Value recv, ciKlass* known_hol
   append(new ProfileCall(method(), bci(), callee, recv, known_holder, obj_args, inlined));
 }
 
+void GraphBuilder::profile_return_type(Value ret, ciMethod* callee, ciMethod* m, int invoke_bci) {
+  assert((m == NULL) == (invoke_bci < 0), "invalid method and invalid bci together");
+  if (m == NULL) {
+    m = method();
+  }
+  if (invoke_bci < 0) {
+    invoke_bci = bci();
+  }
+  ciMethodData* md = m->method_data_or_null();
+  ciProfileData* data = md->bci_to_data(invoke_bci);
+  if (data->is_CallTypeData() || data->is_VirtualCallTypeData()) {
+    append(new ProfileReturnType(m , invoke_bci, callee, ret));
+  }
+}
+
 void GraphBuilder::profile_invocation(ciMethod* callee, ValueStack* state) {
   append(new ProfileInvoke(callee, state));
 }

src/share/vm/c1/c1_GraphBuilder.hpp

@@ -375,6 +375,7 @@ 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, Values* obj_args, bool inlined);
+  void profile_return_type(Value ret, ciMethod* callee, ciMethod* m = NULL, int bci = -1);
   void profile_invocation(ciMethod* inlinee, ValueStack* state);
 
   // Shortcuts to profiling control.

src/share/vm/c1/c1_Instruction.hpp

@@ -107,6 +107,7 @@ class       UnsafePrefetch;
 class         UnsafePrefetchRead;
 class         UnsafePrefetchWrite;
 class   ProfileCall;
+class   ProfileReturnType;
 class   ProfileInvoke;
 class   RuntimeCall;
 class   MemBar;
@@ -211,6 +212,7 @@ class InstructionVisitor: public StackObj {
   virtual void do_UnsafePrefetchRead (UnsafePrefetchRead*  x) = 0;
   virtual void do_UnsafePrefetchWrite(UnsafePrefetchWrite* x) = 0;
   virtual void do_ProfileCall    (ProfileCall*     x) = 0;
+  virtual void do_ProfileReturnType (ProfileReturnType*  x) = 0;
   virtual void do_ProfileInvoke  (ProfileInvoke*   x) = 0;
   virtual void do_RuntimeCall    (RuntimeCall*     x) = 0;
   virtual void do_MemBar         (MemBar*          x) = 0;
@@ -2518,6 +2520,38 @@ LEAF(ProfileCall, Instruction)
   }
 };
 
+LEAF(ProfileReturnType, Instruction)
+ private:
+  ciMethod*        _method;
+  ciMethod*        _callee;
+  int              _bci_of_invoke;
+  Value            _ret;
+
+ public:
+  ProfileReturnType(ciMethod* method, int bci, ciMethod* callee, Value ret)
+    : Instruction(voidType)
+    , _method(method)
+    , _callee(callee)
+    , _bci_of_invoke(bci)
+    , _ret(ret)
+  {
+    set_needs_null_check(true);
+    // The ProfileType has side-effects and must occur precisely where located
+    pin();
+  }
+
+  ciMethod* method()             const { return _method; }
+  ciMethod* callee()             const { return _callee; }
+  int bci_of_invoke()            const { return _bci_of_invoke; }
+  Value ret()                    const { return _ret; }
+
+  virtual void input_values_do(ValueVisitor* f)   {
+    if (_ret != NULL) {
+      f->visit(&_ret);
+    }
+  }
+};
+
 // Call some C runtime function that doesn't safepoint,
 // optionally passing the current thread as the first argument.
 LEAF(RuntimeCall, Instruction)

src/share/vm/c1/c1_InstructionPrinter.cpp

@@ -904,6 +904,12 @@ void InstructionPrinter::do_ProfileCall(ProfileCall* x) {
   output()->put(')');
 }
 
+void InstructionPrinter::do_ProfileReturnType(ProfileReturnType* x) {
+  output()->print("profile ret type ");
+  print_value(x->ret());
+  output()->print(" %s.%s", x->method()->holder()->name()->as_utf8(), x->method()->name()->as_utf8());
+  output()->put(')');
+}
 void InstructionPrinter::do_ProfileInvoke(ProfileInvoke* x) {
   output()->print("profile_invoke ");
   output()->print(" %s.%s", x->inlinee()->holder()->name()->as_utf8(), x->inlinee()->name()->as_utf8());

src/share/vm/c1/c1_InstructionPrinter.hpp

@@ -132,6 +132,7 @@ class InstructionPrinter: public InstructionVisitor {
   virtual void do_UnsafePrefetchRead (UnsafePrefetchRead*  x);
   virtual void do_UnsafePrefetchWrite(UnsafePrefetchWrite* x);
   virtual void do_ProfileCall    (ProfileCall*     x);
+  virtual void do_ProfileReturnType (ProfileReturnType*  x);
   virtual void do_ProfileInvoke  (ProfileInvoke*   x);
   virtual void do_RuntimeCall    (RuntimeCall*     x);
   virtual void do_MemBar         (MemBar*          x);

src/share/vm/c1/c1_LIRGenerator.cpp

@@ -3089,7 +3089,7 @@ void LIRGenerator::profile_arguments(ProfileCall* x) {
 
       Bytecodes::Code bc = x->method()->java_code_at_bci(bci);
       int start = 0;
-      int stop = args->number_of_arguments();
+      int stop = data->is_CallTypeData() ? ((ciCallTypeData*)data)->number_of_arguments() : ((ciVirtualCallTypeData*)data)->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();
@@ -3099,7 +3099,7 @@ void LIRGenerator::profile_arguments(ProfileCall* x) {
       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());
+        int off = in_bytes(TypeEntriesAtCall::argument_type_offset(i)) - in_bytes(TypeEntriesAtCall::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());
@@ -3131,6 +3131,21 @@ void LIRGenerator::do_ProfileCall(ProfileCall* x) {
   __ profile_call(x->method(), x->bci_of_invoke(), x->callee(), mdo, recv, tmp, x->known_holder());
 }
 
+void LIRGenerator::do_ProfileReturnType(ProfileReturnType* x) {
+  int bci = x->bci_of_invoke();
+  ciMethodData* md = x->method()->method_data_or_null();
+  ciProfileData* data = md->bci_to_data(bci);
+  assert(data->is_CallTypeData() || data->is_VirtualCallTypeData(), "wrong profile data type");
+  ciReturnTypeEntry* ret = data->is_CallTypeData() ? ((ciCallTypeData*)data)->ret() : ((ciVirtualCallTypeData*)data)->ret();
+  LIR_Opr mdp = LIR_OprFact::illegalOpr;
+  ciKlass* exact = profile_arg_type(md, 0, md->byte_offset_of_slot(data, ret->type_offset()),
+                                    ret->type(), x->ret(), mdp,
+                                    !x->needs_null_check(), x->callee()->signature()->return_type()->as_klass());
+  if (exact != NULL) {
+    md->set_return_type(bci, exact);
+  }
+}
+
 void LIRGenerator::do_ProfileInvoke(ProfileInvoke* x) {
   // We can safely ignore accessors here, since c2 will inline them anyway,
   // accessors are also always mature.

src/share/vm/c1/c1_LIRGenerator.hpp

@@ -536,6 +536,7 @@ class LIRGenerator: public InstructionVisitor, public BlockClosure {
   virtual void do_UnsafePrefetchRead (UnsafePrefetchRead*  x);
   virtual void do_UnsafePrefetchWrite(UnsafePrefetchWrite* x);
   virtual void do_ProfileCall    (ProfileCall*     x);
+  virtual void do_ProfileReturnType (ProfileReturnType* x);
   virtual void do_ProfileInvoke  (ProfileInvoke*   x);
   virtual void do_RuntimeCall    (RuntimeCall*     x);
   virtual void do_MemBar         (MemBar*          x);

src/share/vm/c1/c1_Optimizer.cpp

@@ -531,6 +531,7 @@ public:
   void do_UnsafePrefetchRead (UnsafePrefetchRead*  x);
   void do_UnsafePrefetchWrite(UnsafePrefetchWrite* x);
   void do_ProfileCall    (ProfileCall*     x);
+  void do_ProfileReturnType (ProfileReturnType*  x);
   void do_ProfileInvoke  (ProfileInvoke*   x);
   void do_RuntimeCall    (RuntimeCall*     x);
   void do_MemBar         (MemBar*          x);
@@ -658,6 +659,7 @@ class NullCheckEliminator: public ValueVisitor {
   void handle_ExceptionObject (ExceptionObject* x);
   void handle_Phi             (Phi* x);
   void handle_ProfileCall     (ProfileCall* x);
+  void handle_ProfileReturnType (ProfileReturnType* x);
 };
 
 
@@ -718,6 +720,7 @@ void NullCheckVisitor::do_UnsafePrefetchRead (UnsafePrefetchRead*  x) {}
 void NullCheckVisitor::do_UnsafePrefetchWrite(UnsafePrefetchWrite* x) {}
 void NullCheckVisitor::do_ProfileCall    (ProfileCall*     x) { nce()->clear_last_explicit_null_check();
                                                                 nce()->handle_ProfileCall(x); }
+void NullCheckVisitor::do_ProfileReturnType (ProfileReturnType* x) { nce()->handle_ProfileReturnType(x); }
 void NullCheckVisitor::do_ProfileInvoke  (ProfileInvoke*   x) {}
 void NullCheckVisitor::do_RuntimeCall    (RuntimeCall*     x) {}
 void NullCheckVisitor::do_MemBar         (MemBar*          x) {}
@@ -1142,6 +1145,10 @@ void NullCheckEliminator::handle_ProfileCall(ProfileCall* x) {
   }
 }
 
+void NullCheckEliminator::handle_ProfileReturnType(ProfileReturnType* x) {
+  x->set_needs_null_check(!set_contains(x->ret()));
+}
+
 void Optimizer::eliminate_null_checks() {
   ResourceMark rm;
 

src/share/vm/c1/c1_RangeCheckElimination.hpp

@@ -162,6 +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_ProfileReturnType (ProfileReturnType*  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 */ };

src/share/vm/c1/c1_ValueMap.hpp

@@ -203,6 +203,7 @@ class ValueNumberingVisitor: public InstructionVisitor {
   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_ProfileReturnType (ProfileReturnType*  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 */ };

src/share/vm/ci/ciMethodData.cpp

@@ -137,12 +137,17 @@ void ciReceiverTypeData::translate_receiver_data_from(const ProfileData* data) {
 
 
 void ciTypeStackSlotEntries::translate_type_data_from(const TypeStackSlotEntries* entries) {
-  for (int i = 0; i < number_of_arguments(); i++) {
+  for (int i = 0; i < _number_of_entries; i++) {
     intptr_t k = entries->type(i);
     TypeStackSlotEntries::set_type(i, translate_klass(k));
   }
 }
 
+void ciReturnTypeEntry::translate_type_data_from(const ReturnTypeEntry* ret) {
+  intptr_t k = ret->type();
+  set_type(translate_klass(k));
+}
+
 // Get the data at an arbitrary (sort of) data index.
 ciProfileData* ciMethodData::data_at(int data_index) {
   if (out_of_bounds(data_index)) {
@@ -313,6 +318,20 @@ void ciMethodData::set_argument_type(int bci, int i, ciKlass* k) {
   }
 }
 
+void ciMethodData::set_return_type(int bci, ciKlass* k) {
+  VM_ENTRY_MARK;
+  MethodData* mdo = get_MethodData();
+  if (mdo != NULL) {
+    ProfileData* data = mdo->bci_to_data(bci);
+    if (data->is_CallTypeData()) {
+      data->as_CallTypeData()->set_return_type(k->get_Klass());
+    } else {
+      assert(data->is_VirtualCallTypeData(), "no arguments!");
+      data->as_VirtualCallTypeData()->set_return_type(k->get_Klass());
+    }
+  }
+}
+
 bool ciMethodData::has_escape_info() {
   return eflag_set(MethodData::estimated);
 }
@@ -517,9 +536,7 @@ void ciTypeEntries::print_ciklass(outputStream* st, intptr_t k) {
 }
 
 void ciTypeStackSlotEntries::print_data_on(outputStream* st) const {
-  _pd->tab(st, true);
-  st->print("argument types");
-  for (int i = 0; i < number_of_arguments(); i++) {
+  for (int i = 0; i < _number_of_entries; i++) {
     _pd->tab(st);
     st->print("%d: stack (%u) ", i, stack_slot(i));
     print_ciklass(st, type(i));
@@ -527,9 +544,25 @@ void ciTypeStackSlotEntries::print_data_on(outputStream* st) const {
   }
 }
 
+void ciReturnTypeEntry::print_data_on(outputStream* st) const {
+  _pd->tab(st);
+  st->print("ret ");
+  print_ciklass(st, type());
+  st->cr();
+}
+
 void ciCallTypeData::print_data_on(outputStream* st) const {
   print_shared(st, "ciCallTypeData");
-  args()->print_data_on(st);
+  if (has_arguments()) {
+    tab(st, true);
+    st->print("argument types");
+    args()->print_data_on(st);
+  }
+  if (has_return()) {
+    tab(st, true);
+    st->print("return type");
+    ret()->print_data_on(st);
+  }
 }
 
 void ciReceiverTypeData::print_receiver_data_on(outputStream* st) const {
@@ -561,6 +594,15 @@ void ciVirtualCallData::print_data_on(outputStream* st) const {
 void ciVirtualCallTypeData::print_data_on(outputStream* st) const {
   print_shared(st, "ciVirtualCallTypeData");
   rtd_super()->print_receiver_data_on(st);
-  args()->print_data_on(st);
+  if (has_arguments()) {
+    tab(st, true);
+    st->print("argument types");
+    args()->print_data_on(st);
+  }
+  if (has_return()) {
+    tab(st, true);
+    st->print("return type");
+    ret()->print_data_on(st);
+  }
 }
 #endif

src/share/vm/ci/ciMethodData.hpp

@@ -104,20 +104,55 @@ public:
 #endif
 };
 
+class ciReturnTypeEntry : public ReturnTypeEntry, ciTypeEntries {
+public:
+  void translate_type_data_from(const ReturnTypeEntry* ret);
+
+  ciKlass* valid_type() const {
+    return valid_ciklass(type());
+  }
+
+#ifndef PRODUCT
+  void print_data_on(outputStream* st) const;
+#endif
+};
+
 class ciCallTypeData : public CallTypeData {
 public:
   ciCallTypeData(DataLayout* layout) : CallTypeData(layout) {}
 
   ciTypeStackSlotEntries* args() const { return (ciTypeStackSlotEntries*)CallTypeData::args(); }
+  ciReturnTypeEntry* ret() const { return (ciReturnTypeEntry*)CallTypeData::ret(); }
 
-  virtual void translate_from(const ProfileData* data) {
-    args()->translate_type_data_from(data->as_CallTypeData()->args());
+  void translate_type_data_from(const ProfileData* data) {
+    if (has_arguments()) {
+      args()->translate_type_data_from(data->as_CallTypeData()->args());
+    }
+    if (has_return()) {
+      ret()->translate_type_data_from(data->as_CallTypeData()->ret());
+    }
+  }
+
+  intptr_t argument_type(int i) const {
+    assert(has_arguments(), "no arg type profiling data");
+    return args()->type(i);
   }
 
   ciKlass* valid_argument_type(int i) const {
+    assert(has_arguments(), "no arg type profiling data");
     return args()->valid_type(i);
   }
 
+  intptr_t return_type() const {
+    assert(has_return(), "no ret type profiling data");
+    return ret()->type();
+  }
+
+  ciKlass* valid_return_type() const {
+    assert(has_return(), "no ret type profiling data");
+    return ret()->valid_type();
+  }
+
 #ifndef PRODUCT
   void print_data_on(outputStream* st) const;
 #endif
@@ -179,12 +214,9 @@ class ciVirtualCallTypeData : public VirtualCallTypeData {
 private:
   // Fake multiple inheritance...  It's a ciReceiverTypeData also.
   ciReceiverTypeData* rtd_super() const { return (ciReceiverTypeData*) this; }
-
 public:
   ciVirtualCallTypeData(DataLayout* layout) : VirtualCallTypeData(layout) {}
 
-  ciTypeStackSlotEntries* args() const { return (ciTypeStackSlotEntries*)VirtualCallTypeData::args(); }
-
   void set_receiver(uint row, ciKlass* recv) {
     rtd_super()->set_receiver(row, recv);
   }
@@ -193,16 +225,40 @@ public:
     return rtd_super()->receiver(row);
   }
 
+  ciTypeStackSlotEntries* args() const { return (ciTypeStackSlotEntries*)VirtualCallTypeData::args(); }
+  ciReturnTypeEntry* ret() const { return (ciReturnTypeEntry*)VirtualCallTypeData::ret(); }
+
   // Copy & translate from oop based VirtualCallData
   virtual void translate_from(const ProfileData* data) {
     rtd_super()->translate_receiver_data_from(data);
-    args()->translate_type_data_from(data->as_VirtualCallTypeData()->args());
+    if (has_arguments()) {
+      args()->translate_type_data_from(data->as_VirtualCallTypeData()->args());
+    }
+    if (has_return()) {
+      ret()->translate_type_data_from(data->as_VirtualCallTypeData()->ret());
+    }
+  }
+
+  intptr_t argument_type(int i) const {
+    assert(has_arguments(), "no arg type profiling data");
+    return args()->type(i);
   }
 
   ciKlass* valid_argument_type(int i) const {
+    assert(has_arguments(), "no arg type profiling data");
     return args()->valid_type(i);
   }
 
+  intptr_t return_type() const {
+    assert(has_return(), "no ret type profiling data");
+    return ret()->type();
+  }
+
+  ciKlass* valid_return_type() const {
+    assert(has_return(), "no ret type profiling data");
+    return ret()->valid_type();
+  }
+
 #ifndef PRODUCT
   void print_data_on(outputStream* st) const;
 #endif
@@ -347,6 +403,7 @@ public:
   // If the compiler finds a profiled type that is known statically
   // for sure, set it in the MethodData
   void set_argument_type(int bci, int i, ciKlass* k);
+  void set_return_type(int bci, ciKlass* k);
 
   void load_data();
 

src/share/vm/oops/methodData.cpp

@@ -156,16 +156,31 @@ void JumpData::print_data_on(outputStream* st) const {
 }
 #endif // !PRODUCT
 
-int TypeStackSlotEntries::compute_cell_count(BytecodeStream* stream) {
-  int max = TypeProfileArgsLimit;
-  assert(Bytecodes::is_invoke(stream->code()), "should be invoke");
-  Bytecode_invoke inv(stream->method(), stream->bci());
-
+int TypeStackSlotEntries::compute_cell_count(Symbol* signature, int max) {
   ResourceMark rm;
-  SignatureStream ss(inv.signature());
+  SignatureStream ss(signature);
   int args_count = MIN2(ss.reference_parameter_count(), max);
+  return args_count * per_arg_cell_count;
+}
 
-  return args_count * per_arg_cell_count + (args_count > 0 ? header_cell_count() : 0);
+int TypeEntriesAtCall::compute_cell_count(BytecodeStream* stream) {
+  assert(Bytecodes::is_invoke(stream->code()), "should be invoke");
+  assert(TypeStackSlotEntries::per_arg_count() > ReturnTypeEntry::static_cell_count(), "code to test for arguments/results broken");
+  Bytecode_invoke inv(stream->method(), stream->bci());
+  int args_cell = 0;
+  if (arguments_profiling_enabled()) {
+    args_cell = TypeStackSlotEntries::compute_cell_count(inv.signature(), TypeProfileArgsLimit);
+  }
+  int ret_cell = 0;
+  if (return_profiling_enabled() && (inv.result_type() == T_OBJECT || inv.result_type() == T_ARRAY)) {
+    ret_cell = ReturnTypeEntry::static_cell_count();
+  }
+  int header_cell = 0;
+  if (args_cell + ret_cell > 0) {
+    header_cell = header_cell_count();
+  }
+
+  return header_cell + args_cell + ret_cell;
 }
 
 class ArgumentOffsetComputer : public SignatureInfo {
@@ -197,26 +212,55 @@ public:
   int off_at(int i) const { return _offsets.at(i); }
 };
 
-void TypeStackSlotEntries::post_initialize(BytecodeStream* stream) {
+void TypeStackSlotEntries::post_initialize(Symbol* signature, bool has_receiver) {
   ResourceMark rm;
+  ArgumentOffsetComputer aos(signature, _number_of_entries);
+  aos.total();
+  for (int i = 0; i < _number_of_entries; i++) {
+    set_stack_slot(i, aos.off_at(i) + (has_receiver ? 1 : 0));
+    set_type(i, type_none());
+  }
+}
 
+void CallTypeData::post_initialize(BytecodeStream* stream, MethodData* mdo) {
   assert(Bytecodes::is_invoke(stream->code()), "should be invoke");
   Bytecode_invoke inv(stream->method(), stream->bci());
 
-#ifdef ASSERT
   SignatureStream ss(inv.signature());
-  int count = MIN2(ss.reference_parameter_count(), (int)TypeProfileArgsLimit);
-  assert(count > 0, "room for args type but none found?");
-  check_number_of_arguments(count);
+  if (has_arguments()) {
+#ifdef ASSERT
+    ResourceMark rm;
+    int count = MIN2(ss.reference_parameter_count(), (int)TypeProfileArgsLimit);
+    assert(count > 0, "room for args type but none found?");
+    check_number_of_arguments(count);
 #endif
+    _args.post_initialize(inv.signature(), inv.has_receiver());
+  }
 
-  int start = 0;
-  ArgumentOffsetComputer aos(inv.signature(), number_of_arguments()-start);
-  aos.total();
-  bool has_receiver = inv.has_receiver();
-  for (int i = start; i < number_of_arguments(); i++) {
-    set_stack_slot(i, aos.off_at(i-start) + (has_receiver ? 1 : 0));
-    set_type(i, type_none());
+  if (has_return()) {
+    assert(inv.result_type() == T_OBJECT || inv.result_type() == T_ARRAY, "room for a ret type but doesn't return obj?");
+    _ret.post_initialize();
+  }
+}
+
+void VirtualCallTypeData::post_initialize(BytecodeStream* stream, MethodData* mdo) {
+  assert(Bytecodes::is_invoke(stream->code()), "should be invoke");
+  Bytecode_invoke inv(stream->method(), stream->bci());
+
+  if (has_arguments()) {
+#ifdef ASSERT
+    ResourceMark rm;
+    SignatureStream ss(inv.signature());
+    int count = MIN2(ss.reference_parameter_count(), (int)TypeProfileArgsLimit);
+    assert(count > 0, "room for args type but none found?");
+    check_number_of_arguments(count);
+#endif
+    _args.post_initialize(inv.signature(), inv.has_receiver());
+  }
+
+  if (has_return()) {
+    assert(inv.result_type() == T_OBJECT || inv.result_type() == T_ARRAY, "room for a ret type but doesn't return obj?");
+    _ret.post_initialize();
   }
 }
 
@@ -226,7 +270,7 @@ bool TypeEntries::is_loader_alive(BoolObjectClosure* is_alive_cl, intptr_t p) {
 }
 
 void TypeStackSlotEntries::clean_weak_klass_links(BoolObjectClosure* is_alive_cl) {
-  for (int i = 0; i < number_of_arguments(); i++) {
+  for (int i = 0; i < _number_of_entries; i++) {
     intptr_t p = type(i);
     if (is_loader_alive(is_alive_cl, p)) {
       set_type(i, type_none());
@@ -234,7 +278,18 @@ void TypeStackSlotEntries::clean_weak_klass_links(BoolObjectClosure* is_alive_cl
   }
 }
 
-bool TypeStackSlotEntries::arguments_profiling_enabled() {
+void ReturnTypeEntry::clean_weak_klass_links(BoolObjectClosure* is_alive_cl) {
+  intptr_t p = type();
+  if (is_loader_alive(is_alive_cl, p)) {
+    set_type(type_none());
+  }
+}
+
+bool TypeEntriesAtCall::return_profiling_enabled() {
+  return MethodData::profile_return();
+}
+
+bool TypeEntriesAtCall::arguments_profiling_enabled() {
   return MethodData::profile_arguments();
 }
 
@@ -253,9 +308,7 @@ void TypeEntries::print_klass(outputStream* st, intptr_t k) {
 }
 
 void TypeStackSlotEntries::print_data_on(outputStream* st) const {
-  _pd->tab(st, true);
-  st->print("argument types");
-  for (int i = 0; i < number_of_arguments(); i++) {
+  for (int i = 0; i < _number_of_entries; i++) {
     _pd->tab(st);
     st->print("%d: stack(%u) ", i, stack_slot(i));
     print_klass(st, type(i));
@@ -263,14 +316,38 @@ void TypeStackSlotEntries::print_data_on(outputStream* st) const {
   }
 }
 
+void ReturnTypeEntry::print_data_on(outputStream* st) const {
+  _pd->tab(st);
+  print_klass(st, type());
+  st->cr();
+}
+
 void CallTypeData::print_data_on(outputStream* st) const {
   CounterData::print_data_on(st);
-  _args.print_data_on(st);
+  if (has_arguments()) {
+    tab(st, true);
+    st->print("argument types");
+    _args.print_data_on(st);
+  }
+  if (has_return()) {
+    tab(st, true);
+    st->print("return type");
+    _ret.print_data_on(st);
+  }
 }
 
 void VirtualCallTypeData::print_data_on(outputStream* st) const {
   VirtualCallData::print_data_on(st);
-  _args.print_data_on(st);
+  if (has_arguments()) {
+    tab(st, true);
+    st->print("argument types");
+    _args.print_data_on(st);
+  }
+  if (has_return()) {
+    tab(st, true);
+    st->print("return type");
+    _ret.print_data_on(st);
+  }
 }
 #endif
 
@@ -530,7 +607,7 @@ int MethodData::bytecode_cell_count(Bytecodes::Code code) {
     }
   case Bytecodes::_invokespecial:
   case Bytecodes::_invokestatic:
-    if (MethodData::profile_arguments()) {
+    if (MethodData::profile_arguments() || MethodData::profile_return()) {
       return variable_cell_count;
     } else {
       return CounterData::static_cell_count();
@@ -542,13 +619,13 @@ int MethodData::bytecode_cell_count(Bytecodes::Code code) {
     return JumpData::static_cell_count();
   case Bytecodes::_invokevirtual:
   case Bytecodes::_invokeinterface:
-    if (MethodData::profile_arguments()) {
+    if (MethodData::profile_arguments() || MethodData::profile_return()) {
       return variable_cell_count;
     } else {
       return VirtualCallData::static_cell_count();
     }
   case Bytecodes::_invokedynamic:
-    if (MethodData::profile_arguments()) {
+    if (MethodData::profile_arguments() || MethodData::profile_return()) {
       return variable_cell_count;
     } else {
       return CounterData::static_cell_count();
@@ -596,8 +673,9 @@ int MethodData::compute_data_size(BytecodeStream* stream) {
     case Bytecodes::_invokespecial:
     case Bytecodes::_invokestatic:
     case Bytecodes::_invokedynamic:
-      assert(MethodData::profile_arguments(), "should be collecting args profile");
-      if (profile_arguments_for_invoke(stream->method(), stream->bci())) {
+      assert(MethodData::profile_arguments() || MethodData::profile_return(), "should be collecting args profile");
+      if (profile_arguments_for_invoke(stream->method(), stream->bci()) ||
+          profile_return_for_invoke(stream->method(), stream->bci())) {
         cell_count = CallTypeData::compute_cell_count(stream);
       } else {
         cell_count = CounterData::static_cell_count();
@@ -605,8 +683,9 @@ int MethodData::compute_data_size(BytecodeStream* stream) {
       break;
     case Bytecodes::_invokevirtual:
     case Bytecodes::_invokeinterface: {
-      assert(MethodData::profile_arguments(), "should be collecting args profile");
-      if (profile_arguments_for_invoke(stream->method(), stream->bci())) {
+      assert(MethodData::profile_arguments() || MethodData::profile_return(), "should be collecting args profile");
+      if (profile_arguments_for_invoke(stream->method(), stream->bci()) ||
+          profile_return_for_invoke(stream->method(), stream->bci())) {
         cell_count = VirtualCallTypeData::compute_cell_count(stream);
       } else {
         cell_count = VirtualCallData::static_cell_count();
@@ -699,7 +778,8 @@ int MethodData::initialize_data(BytecodeStream* stream,
   case Bytecodes::_invokespecial:
   case Bytecodes::_invokestatic: {
     int counter_data_cell_count = CounterData::static_cell_count();
-    if (profile_arguments_for_invoke(stream->method(), stream->bci())) {
+    if (profile_arguments_for_invoke(stream->method(), stream->bci()) ||
+        profile_return_for_invoke(stream->method(), stream->bci())) {
       cell_count = CallTypeData::compute_cell_count(stream);
     } else {
       cell_count = counter_data_cell_count;
@@ -721,7 +801,8 @@ int MethodData::initialize_data(BytecodeStream* stream,
   case Bytecodes::_invokevirtual:
   case Bytecodes::_invokeinterface: {
     int virtual_call_data_cell_count = VirtualCallData::static_cell_count();
-    if (profile_arguments_for_invoke(stream->method(), stream->bci())) {
+    if (profile_arguments_for_invoke(stream->method(), stream->bci()) ||
+        profile_return_for_invoke(stream->method(), stream->bci())) {
       cell_count = VirtualCallTypeData::compute_cell_count(stream);
     } else {
       cell_count = virtual_call_data_cell_count;
@@ -736,7 +817,8 @@ int MethodData::initialize_data(BytecodeStream* stream,
   case Bytecodes::_invokedynamic: {
     // %%% should make a type profile for any invokedynamic that takes a ref argument
     int counter_data_cell_count = CounterData::static_cell_count();
-    if (profile_arguments_for_invoke(stream->method(), stream->bci())) {
+    if (profile_arguments_for_invoke(stream->method(), stream->bci()) ||
+        profile_return_for_invoke(stream->method(), stream->bci())) {
       cell_count = CallTypeData::compute_cell_count(stream);
     } else {
       cell_count = counter_data_cell_count;
@@ -778,7 +860,7 @@ int MethodData::initialize_data(BytecodeStream* stream,
     break;
   }
   assert(tag == DataLayout::multi_branch_data_tag ||
-         (MethodData::profile_arguments() &&
+         ((MethodData::profile_arguments() || MethodData::profile_return()) &&
           (tag == DataLayout::call_type_data_tag ||
            tag == DataLayout::counter_data_tag ||
            tag == DataLayout::virtual_call_type_data_tag ||
@@ -1111,7 +1193,7 @@ bool MethodData::profile_jsr292(methodHandle m, int bci) {
 }
 
 int MethodData::profile_arguments_flag() {
-  return TypeProfileLevel;
+  return TypeProfileLevel % 10;
 }
 
 bool MethodData::profile_arguments() {
@@ -1139,3 +1221,31 @@ bool MethodData::profile_arguments_for_invoke(methodHandle m, int bci) {
   return profile_jsr292(m, bci);
 }
 
+int MethodData::profile_return_flag() {
+  return TypeProfileLevel / 10;
+}
+
+bool MethodData::profile_return() {
+  return profile_return_flag() > no_type_profile && profile_return_flag() <= type_profile_all;
+}
+
+bool MethodData::profile_return_jsr292_only() {
+  return profile_return_flag() == type_profile_jsr292;
+}
+
+bool MethodData::profile_all_return() {
+  return profile_return_flag() == type_profile_all;
+}
+
+bool MethodData::profile_return_for_invoke(methodHandle m, int bci) {
+  if (!profile_return()) {
+    return false;
+  }
+
+  if (profile_all_return()) {
+    return true;
+  }
+
+  assert(profile_return_jsr292_only(), "inconsistent");
+  return profile_jsr292(m, bci);
+}

src/share/vm/oops/methodData.hpp

@@ -271,6 +271,7 @@ class     ArgInfoData;
 // data in a structured way.
 class ProfileData : public ResourceObj {
   friend class TypeEntries;
+  friend class ReturnTypeEntry;
   friend class TypeStackSlotEntries;
 private:
 #ifndef PRODUCT
@@ -748,119 +749,60 @@ private:
     per_arg_cell_count
   };
 
-  // Start with a header if needed. It stores the number of cells used
-  // for this call type information. Unless we collect only profiling
-  // for a single argument the number of cells is unknown statically.
-  static int header_cell_count() {
-    return (TypeProfileArgsLimit > 1) ? 1 : 0;
-  }
-
-  static int cell_count_local_offset() {
-     assert(arguments_profiling_enabled() && TypeProfileArgsLimit > 1, "no cell count");
-     return 0;
-   }
-
-  int cell_count_global_offset() const {
-    return _base_off + cell_count_local_offset();
-  }
-
   // offset of cell for stack slot for entry i within ProfileData object
-  int stack_slot_global_offset(int i) const {
+  int stack_slot_offset(int i) const {
     return _base_off + stack_slot_local_offset(i);
   }
 
-  void check_number_of_arguments(int total) {
-    assert(number_of_arguments() == total, "should be set in DataLayout::initialize");
-  }
-
-  // number of cells not counting the header
-  int cell_count_no_header() const {
-    return _pd->uint_at(cell_count_global_offset());
-  }
-
-  static bool arguments_profiling_enabled();
-  static void assert_arguments_profiling_enabled() {
-    assert(arguments_profiling_enabled(), "args profiling should be on");
-  }
-
 protected:
+  const int _number_of_entries;
 
   // offset of cell for type for entry i within ProfileData object
-  int type_global_offset(int i) const {
+  int type_offset(int i) const {
     return _base_off + type_local_offset(i);
   }
 
 public:
 
-  TypeStackSlotEntries(int base_off)
-    : TypeEntries(base_off) {}
-
-  static int compute_cell_count(BytecodeStream* stream);
-
-  static void initialize(DataLayout* dl, int base, int cell_count) {
-    if (TypeProfileArgsLimit > 1) {
-      int off = base + cell_count_local_offset();
-      dl->set_cell_at(off, cell_count - base - header_cell_count());
-    }
-  }
-
-  void post_initialize(BytecodeStream* stream);
+  TypeStackSlotEntries(int base_off, int nb_entries)
+    : TypeEntries(base_off), _number_of_entries(nb_entries) {}
 
-  int number_of_arguments() const {
-    assert_arguments_profiling_enabled();
-    if (TypeProfileArgsLimit > 1) {
-      int cell_count = cell_count_no_header();
-      int nb = cell_count / TypeStackSlotEntries::per_arg_count();
-      assert(nb > 0 && nb <= TypeProfileArgsLimit , "only when we profile args");
-      return nb;
-    } else {
-      assert(TypeProfileArgsLimit == 1, "at least one arg");
-      return 1;
-    }
-  }
+  static int compute_cell_count(Symbol* signature, int max);
 
-  int cell_count() const {
-    assert_arguments_profiling_enabled();
-    if (TypeProfileArgsLimit > 1) {
-      return _base_off + header_cell_count() + _pd->int_at_unchecked(cell_count_global_offset());
-    } else {
-      return _base_off + TypeStackSlotEntries::per_arg_count();
-    }
-  }
+  void post_initialize(Symbol* signature, bool has_receiver);
 
   // offset of cell for stack slot for entry i within this block of cells for a TypeStackSlotEntries
   static int stack_slot_local_offset(int i) {
-    assert_arguments_profiling_enabled();
-    return header_cell_count() + i * per_arg_cell_count + stack_slot_entry;
+    return i * per_arg_cell_count + stack_slot_entry;
   }
 
   // offset of cell for type for entry i within this block of cells for a TypeStackSlotEntries
   static int type_local_offset(int i) {
-    return header_cell_count() + i * per_arg_cell_count + type_entry;
+    return i * per_arg_cell_count + type_entry;
   }
 
   // stack slot for entry i
   uint stack_slot(int i) const {
-    assert(i >= 0 && i < number_of_arguments(), "oob");
-    return _pd->uint_at(stack_slot_global_offset(i));
+    assert(i >= 0 && i < _number_of_entries, "oob");
+    return _pd->uint_at(stack_slot_offset(i));
   }
 
   // set stack slot for entry i
   void set_stack_slot(int i, uint num) {
-    assert(i >= 0 && i < number_of_arguments(), "oob");
-    _pd->set_uint_at(stack_slot_global_offset(i), num);
+    assert(i >= 0 && i < _number_of_entries, "oob");
+    _pd->set_uint_at(stack_slot_offset(i), num);
   }
 
   // type for entry i
   intptr_t type(int i) const {
-    assert(i >= 0 && i < number_of_arguments(), "oob");
-    return _pd->intptr_at(type_global_offset(i));
+    assert(i >= 0 && i < _number_of_entries, "oob");
+    return _pd->intptr_at(type_offset(i));
   }
 
   // set type for entry i
   void set_type(int i, intptr_t k) {
-    assert(i >= 0 && i < number_of_arguments(), "oob");
-    _pd->set_intptr_at(type_global_offset(i), k);
+    assert(i >= 0 && i < _number_of_entries, "oob");
+    _pd->set_intptr_at(type_offset(i), k);
   }
 
   static ByteSize per_arg_size() {
@@ -871,22 +813,50 @@ public:
     return per_arg_cell_count ;
   }
 
-  // Code generation support
-   static ByteSize cell_count_offset() {
-     return in_ByteSize(cell_count_local_offset() * DataLayout::cell_size);
-   }
+  // GC support
+  void clean_weak_klass_links(BoolObjectClosure* is_alive_closure);
+
+#ifndef PRODUCT
+  void print_data_on(outputStream* st) const;
+#endif
+};
+
+// Type entry used for return from a call. A single cell to record the
+// type.
+class ReturnTypeEntry : public TypeEntries {
+
+private:
+  enum {
+    cell_count = 1
+  };
+
+public:
+  ReturnTypeEntry(int base_off)
+    : TypeEntries(base_off) {}
+
+  void post_initialize() {
+    set_type(type_none());
+  }
+
+  intptr_t type() const {
+    return _pd->intptr_at(_base_off);
+  }
 
-   static ByteSize args_data_offset() {
-     return in_ByteSize(header_cell_count() * DataLayout::cell_size);
-   }
+  void set_type(intptr_t k) {
+    _pd->set_intptr_at(_base_off, k);
+  }
 
-   static ByteSize stack_slot_offset(int i) {
-     return in_ByteSize(stack_slot_local_offset(i) * DataLayout::cell_size);
-   }
+  static int static_cell_count() {
+    return cell_count;
+  }
 
-   static ByteSize type_offset(int i) {
-     return in_ByteSize(type_local_offset(i) * DataLayout::cell_size);
-   }
+  static ByteSize size() {
+    return in_ByteSize(cell_count * DataLayout::cell_size);
+  }
+
+  ByteSize type_offset() {
+    return DataLayout::cell_offset(_base_off);
+  }
 
   // GC support
   void clean_weak_klass_links(BoolObjectClosure* is_alive_closure);
@@ -896,23 +866,118 @@ public:
 #endif
 };
 
+// Entries to collect type information at a call: contains arguments
+// (TypeStackSlotEntries), a return type (ReturnTypeEntry) and a
+// number of cells. Because the number of cells for the return type is
+// smaller than the number of cells for the type of an arguments, the
+// number of cells is used to tell how many arguments are profiled and
+// whether a return value is profiled. See has_arguments() and
+// has_return().
+class TypeEntriesAtCall {
+private:
+  static int stack_slot_local_offset(int i) {
+    return header_cell_count() + TypeStackSlotEntries::stack_slot_local_offset(i);
+  }
+
+  static int argument_type_local_offset(int i) {
+    return header_cell_count() + TypeStackSlotEntries::type_local_offset(i);;
+  }
+
+public:
+
+  static int header_cell_count() {
+    return 1;
+  }
+
+  static int cell_count_local_offset() {
+    return 0;
+  }
+
+  static int compute_cell_count(BytecodeStream* stream);
+
+  static void initialize(DataLayout* dl, int base, int cell_count) {
+    int off = base + cell_count_local_offset();
+    dl->set_cell_at(off, cell_count - base - header_cell_count());
+  }
+
+  static bool arguments_profiling_enabled();
+  static bool return_profiling_enabled();
+
+  // Code generation support
+  static ByteSize cell_count_offset() {
+    return in_ByteSize(cell_count_local_offset() * DataLayout::cell_size);
+  }
+
+  static ByteSize args_data_offset() {
+    return in_ByteSize(header_cell_count() * DataLayout::cell_size);
+  }
+
+  static ByteSize stack_slot_offset(int i) {
+    return in_ByteSize(stack_slot_local_offset(i) * DataLayout::cell_size);
+  }
+
+  static ByteSize argument_type_offset(int i) {
+    return in_ByteSize(argument_type_local_offset(i) * DataLayout::cell_size);
+  }
+};
+
 // CallTypeData
 //
 // A CallTypeData is used to access profiling information about a non
-// virtual call for which we collect type information about arguments.
+// virtual call for which we collect type information about arguments
+// and return value.
 class CallTypeData : public CounterData {
 private:
+  // entries for arguments if any
   TypeStackSlotEntries _args;
+  // entry for return type if any
+  ReturnTypeEntry _ret;
+
+  int cell_count_global_offset() const {
+    return CounterData::static_cell_count() + TypeEntriesAtCall::cell_count_local_offset();
+  }
+
+  // number of cells not counting the header
+  int cell_count_no_header() const {
+    return uint_at(cell_count_global_offset());
+  }
+
+  void check_number_of_arguments(int total) {
+    assert(number_of_arguments() == total, "should be set in DataLayout::initialize");
+  }
+
+protected:
+  // An entry for a return value takes less space than an entry for an
+  // argument so if the number of cells exceeds the number of cells
+  // needed for an argument, this object contains type information for
+  // at least one argument.
+  bool has_arguments() const {
+    bool res = cell_count_no_header() >= TypeStackSlotEntries::per_arg_count();
+    assert (!res || TypeEntriesAtCall::arguments_profiling_enabled(), "no profiling of arguments");
+    return res;
+  }
 
 public:
   CallTypeData(DataLayout* layout) :
-    CounterData(layout), _args(CounterData::static_cell_count())  {
+    CounterData(layout),
+    _args(CounterData::static_cell_count()+TypeEntriesAtCall::header_cell_count(), number_of_arguments()),
+    _ret(cell_count() - ReturnTypeEntry::static_cell_count())
+  {
     assert(layout->tag() == DataLayout::call_type_data_tag, "wrong type");
     // Some compilers (VC++) don't want this passed in member initialization list
     _args.set_profile_data(this);
+    _ret.set_profile_data(this);
   }
 
-  const TypeStackSlotEntries* args() const { return &_args; }
+  const TypeStackSlotEntries* args() const {
+    assert(has_arguments(), "no profiling of arguments");
+    return &_args;
+  }
+
+  const ReturnTypeEntry* ret() const {
+    assert(has_return(), "no profiling of return value");
+    return &_ret;
+  }
 
   virtual bool is_CallTypeData() const { return true; }
 
@@ -921,38 +986,60 @@ public:
   }
 
   static int compute_cell_count(BytecodeStream* stream) {
-    return CounterData::static_cell_count() + TypeStackSlotEntries::compute_cell_count(stream);
+    return CounterData::static_cell_count() + TypeEntriesAtCall::compute_cell_count(stream);
   }
 
   static void initialize(DataLayout* dl, int cell_count) {
-    TypeStackSlotEntries::initialize(dl, CounterData::static_cell_count(), cell_count);
+    TypeEntriesAtCall::initialize(dl, CounterData::static_cell_count(), cell_count);
   }
 
-  virtual void post_initialize(BytecodeStream* stream, MethodData* mdo) {
-    _args.post_initialize(stream);
-  }
+  virtual void post_initialize(BytecodeStream* stream, MethodData* mdo);
 
   virtual int cell_count() const {
-    return _args.cell_count();
+    return CounterData::static_cell_count() +
+      TypeEntriesAtCall::header_cell_count() +
+      int_at_unchecked(cell_count_global_offset());
   }
 
-  uint number_of_arguments() const {
-    return args()->number_of_arguments();
+  int number_of_arguments() const {
+    return cell_count_no_header() / TypeStackSlotEntries::per_arg_count();
   }
 
   void set_argument_type(int i, Klass* k) {
+    assert(has_arguments(), "no arguments!");
     intptr_t current = _args.type(i);
     _args.set_type(i, TypeEntries::with_status(k, current));
   }
 
+  void set_return_type(Klass* k) {
+    assert(has_return(), "no return!");
+    intptr_t current = _ret.type();
+    _ret.set_type(TypeEntries::with_status(k, current));
+  }
+
+  // An entry for a return value takes less space than an entry for an
+  // argument, so if the remainder of the number of cells divided by
+  // the number of cells for an argument is not null, a return value
+  // is profiled in this object.
+  bool has_return() const {
+    bool res = (cell_count_no_header() % TypeStackSlotEntries::per_arg_count()) != 0;
+    assert (!res || TypeEntriesAtCall::return_profiling_enabled(), "no profiling of return values");
+    return res;
+  }
+
   // Code generation support
   static ByteSize args_data_offset() {
-    return cell_offset(CounterData::static_cell_count()) + TypeStackSlotEntries::args_data_offset();
+    return cell_offset(CounterData::static_cell_count()) + TypeEntriesAtCall::args_data_offset();
   }
 
   // GC support
   virtual void clean_weak_klass_links(BoolObjectClosure* is_alive_closure) {
-    _args.clean_weak_klass_links(is_alive_closure);
+    if (has_arguments()) {
+      _args.clean_weak_klass_links(is_alive_closure);
+    }
+    if (has_return()) {
+      _ret.clean_weak_klass_links(is_alive_closure);
+    }
   }
 
 #ifndef PRODUCT
@@ -1105,20 +1192,59 @@ public:
 //
 // A VirtualCallTypeData is used to access profiling information about
 // a virtual call for which we collect type information about
-// arguments.
+// arguments and return value.
 class VirtualCallTypeData : public VirtualCallData {
 private:
+  // entries for arguments if any
   TypeStackSlotEntries _args;
+  // entry for return type if any
+  ReturnTypeEntry _ret;
+
+  int cell_count_global_offset() const {
+    return VirtualCallData::static_cell_count() + TypeEntriesAtCall::cell_count_local_offset();
+  }
+
+  // number of cells not counting the header
+  int cell_count_no_header() const {
+    return uint_at(cell_count_global_offset());
+  }
+
+  void check_number_of_arguments(int total) {
+    assert(number_of_arguments() == total, "should be set in DataLayout::initialize");
+  }
+
+protected:
+  // An entry for a return value takes less space than an entry for an
+  // argument so if the number of cells exceeds the number of cells
+  // needed for an argument, this object contains type information for
+  // at least one argument.
+  bool has_arguments() const {
+    bool res = cell_count_no_header() >= TypeStackSlotEntries::per_arg_count();
+    assert (!res || TypeEntriesAtCall::arguments_profiling_enabled(), "no profiling of arguments");
+    return res;
+  }
 
 public:
   VirtualCallTypeData(DataLayout* layout) :
-    VirtualCallData(layout), _args(VirtualCallData::static_cell_count())  {
+    VirtualCallData(layout),
+    _args(VirtualCallData::static_cell_count()+TypeEntriesAtCall::header_cell_count(), number_of_arguments()),
+    _ret(cell_count() - ReturnTypeEntry::static_cell_count())
+  {
     assert(layout->tag() == DataLayout::virtual_call_type_data_tag, "wrong type");
     // Some compilers (VC++) don't want this passed in member initialization list
     _args.set_profile_data(this);
+    _ret.set_profile_data(this);
+  }
+
+  const TypeStackSlotEntries* args() const {
+    assert(has_arguments(), "no profiling of arguments");
+    return &_args;
   }
 
-  const TypeStackSlotEntries* args() const { return &_args; }
+  const ReturnTypeEntry* ret() const {
+    assert(has_return(), "no profiling of return value");
+    return &_ret;
+  }
 
   virtual bool is_VirtualCallTypeData() const { return true; }
 
@@ -1127,39 +1253,61 @@ public:
   }
 
   static int compute_cell_count(BytecodeStream* stream) {
-    return VirtualCallData::static_cell_count() + TypeStackSlotEntries::compute_cell_count(stream);
+    return VirtualCallData::static_cell_count() + TypeEntriesAtCall::compute_cell_count(stream);
   }
 
   static void initialize(DataLayout* dl, int cell_count) {
-    TypeStackSlotEntries::initialize(dl, VirtualCallData::static_cell_count(), cell_count);
+    TypeEntriesAtCall::initialize(dl, VirtualCallData::static_cell_count(), cell_count);
   }
 
-  virtual void post_initialize(BytecodeStream* stream, MethodData* mdo) {
-    _args.post_initialize(stream);
-  }
+  virtual void post_initialize(BytecodeStream* stream, MethodData* mdo);
 
   virtual int cell_count() const {
-    return _args.cell_count();
+    return VirtualCallData::static_cell_count() +
+      TypeEntriesAtCall::header_cell_count() +
+      int_at_unchecked(cell_count_global_offset());
   }
 
-  uint number_of_arguments() const {
-    return args()->number_of_arguments();
+  int number_of_arguments() const {
+    return cell_count_no_header() / TypeStackSlotEntries::per_arg_count();
   }
 
   void set_argument_type(int i, Klass* k) {
+    assert(has_arguments(), "no arguments!");
     intptr_t current = _args.type(i);
     _args.set_type(i, TypeEntries::with_status(k, current));
   }
 
+  void set_return_type(Klass* k) {
+    assert(has_return(), "no return!");
+    intptr_t current = _ret.type();
+    _ret.set_type(TypeEntries::with_status(k, current));
+  }
+
+  // An entry for a return value takes less space than an entry for an
+  // argument, so if the remainder of the number of cells divided by
+  // the number of cells for an argument is not null, a return value
+  // is profiled in this object.
+  bool has_return() const {
+    bool res = (cell_count_no_header() % TypeStackSlotEntries::per_arg_count()) != 0;
+    assert (!res || TypeEntriesAtCall::return_profiling_enabled(), "no profiling of return values");
+    return res;
+  }
+
   // Code generation support
   static ByteSize args_data_offset() {
-    return cell_offset(VirtualCallData::static_cell_count()) + TypeStackSlotEntries::args_data_offset();
+    return cell_offset(VirtualCallData::static_cell_count()) + TypeEntriesAtCall::args_data_offset();
   }
 
   // GC support
   virtual void clean_weak_klass_links(BoolObjectClosure* is_alive_closure) {
     ReceiverTypeData::clean_weak_klass_links(is_alive_closure);
-    _args.clean_weak_klass_links(is_alive_closure);
+    if (has_arguments()) {
+      _args.clean_weak_klass_links(is_alive_closure);
+    }
+    if (has_return()) {
+      _ret.clean_weak_klass_links(is_alive_closure);
+    }
   }
 
 #ifndef PRODUCT
@@ -1691,6 +1839,9 @@ private:
   static bool profile_arguments_jsr292_only();
   static bool profile_all_arguments();
   static bool profile_arguments_for_invoke(methodHandle m, int bci);
+  static int profile_return_flag();
+  static bool profile_all_return();
+  static bool profile_return_for_invoke(methodHandle m, int bci);
 
 public:
   static int header_size() {
@@ -1933,6 +2084,8 @@ public:
   void verify_data_on(outputStream* st);
 
   static bool profile_arguments();
+  static bool profile_return();
+  static bool profile_return_jsr292_only();
 };
 
 #endif // SHARE_VM_OOPS_METHODDATAOOP_HPP

src/share/vm/runtime/globals.hpp

@@ -2649,8 +2649,9 @@ class CommandLineFlags {
           "Enable aggressive optimizations - see arguments.cpp")            \
                                                                             \
   product_pd(uintx, TypeProfileLevel,                                       \
-          "Type profiling of arguments at call:"                            \
-          "0->off ; 1->js292 only; 2->all methods")                         \
+          "=XY, with Y, Type profiling of arguments at call"                \
+          "          X, Type profiling of return value at call"             \
+          "X and Y in 0->off ; 1->js292 only; 2->all methods")              \
                                                                             \
   product(intx, TypeProfileArgsLimit,     2,                                \
           "max number of call arguments to consider for type profiling")    \