6934604: enable parts of EliminateAutoBox by default

Summary: Resurrected autobox elimination code and enabled part of it by default.
Reviewed-by: roland, twisti

src/share/vm/ci/ciInstanceKlass.cpp

@@ -211,12 +211,41 @@ bool ciInstanceKlass::is_java_lang_Object() const {
 
 // ------------------------------------------------------------------
 // ciInstanceKlass::uses_default_loader
-bool ciInstanceKlass::uses_default_loader() {
+bool ciInstanceKlass::uses_default_loader() const {
   // Note:  We do not need to resolve the handle or enter the VM
   // in order to test null-ness.
   return _loader == NULL;
 }
 
+// ------------------------------------------------------------------
+
+/**
+ * Return basic type of boxed value for box klass or T_OBJECT if not.
+ */
+BasicType ciInstanceKlass::box_klass_type() const {
+  if (uses_default_loader() && is_loaded()) {
+    return SystemDictionary::box_klass_type(get_Klass());
+  } else {
+    return T_OBJECT;
+  }
+}
+
+/**
+ * Is this boxing klass?
+ */
+bool ciInstanceKlass::is_box_klass() const {
+  return is_java_primitive(box_klass_type());
+}
+
+/**
+ *  Is this boxed value offset?
+ */
+bool ciInstanceKlass::is_boxed_value_offset(int offset) const {
+  BasicType bt = box_klass_type();
+  return is_java_primitive(bt) &&
+         (offset == java_lang_boxing_object::value_offset_in_bytes(bt));
+}
+
 // ------------------------------------------------------------------
 // ciInstanceKlass::is_in_package
 //

src/share/vm/ci/ciInstanceKlass.hpp

@@ -217,10 +217,14 @@ public:
   ciInstanceKlass* implementor();
 
   // Is the defining class loader of this class the default loader?
-  bool uses_default_loader();
+  bool uses_default_loader() const;
 
   bool is_java_lang_Object() const;
 
+  BasicType box_klass_type() const;
+  bool is_box_klass() const;
+  bool is_boxed_value_offset(int offset) const;
+
   // Is this klass in the given package?
   bool is_in_package(const char* packagename) {
     return is_in_package(packagename, (int) strlen(packagename));

src/share/vm/ci/ciMethod.cpp

@@ -1179,6 +1179,44 @@ bool ciMethod::has_jsrs       () const {         FETCH_FLAG_FROM_VM(has_jsrs);
 bool ciMethod::is_accessor    () const {         FETCH_FLAG_FROM_VM(is_accessor); }
 bool ciMethod::is_initializer () const {         FETCH_FLAG_FROM_VM(is_initializer); }
 
+bool ciMethod::is_boxing_method() const {
+  if (holder()->is_box_klass()) {
+    switch (intrinsic_id()) {
+      case vmIntrinsics::_Boolean_valueOf:
+      case vmIntrinsics::_Byte_valueOf:
+      case vmIntrinsics::_Character_valueOf:
+      case vmIntrinsics::_Short_valueOf:
+      case vmIntrinsics::_Integer_valueOf:
+      case vmIntrinsics::_Long_valueOf:
+      case vmIntrinsics::_Float_valueOf:
+      case vmIntrinsics::_Double_valueOf:
+        return true;
+      default:
+        return false;
+    }
+  }
+  return false;
+}
+
+bool ciMethod::is_unboxing_method() const {
+  if (holder()->is_box_klass()) {
+    switch (intrinsic_id()) {
+      case vmIntrinsics::_booleanValue:
+      case vmIntrinsics::_byteValue:
+      case vmIntrinsics::_charValue:
+      case vmIntrinsics::_shortValue:
+      case vmIntrinsics::_intValue:
+      case vmIntrinsics::_longValue:
+      case vmIntrinsics::_floatValue:
+      case vmIntrinsics::_doubleValue:
+        return true;
+      default:
+        return false;
+    }
+  }
+  return false;
+}
+
 BCEscapeAnalyzer  *ciMethod::get_bcea() {
 #ifdef COMPILER2
   if (_bcea == NULL) {

src/share/vm/ci/ciMethod.hpp

@@ -298,6 +298,8 @@ class ciMethod : public ciMetadata {
   bool is_initializer () const;
   bool can_be_statically_bound() const           { return _can_be_statically_bound; }
   void dump_replay_data(outputStream* st);
+  bool is_boxing_method() const;
+  bool is_unboxing_method() const;
 
   // Print the bytecodes of this method.
   void print_codes_on(outputStream* st);

src/share/vm/classfile/vmSymbols.hpp

@@ -68,7 +68,7 @@
   template(java_lang_Float,                           "java/lang/Float")                          \
   template(java_lang_Double,                          "java/lang/Double")                         \
   template(java_lang_Byte,                            "java/lang/Byte")                           \
-  template(java_lang_Byte_Cache,                      "java/lang/Byte$ByteCache")                 \
+  template(java_lang_Byte_ByteCache,                  "java/lang/Byte$ByteCache")                 \
   template(java_lang_Short,                           "java/lang/Short")                          \
   template(java_lang_Short_ShortCache,                "java/lang/Short$ShortCache")               \
   template(java_lang_Integer,                         "java/lang/Integer")                        \

src/share/vm/compiler/compileBroker.cpp

@@ -1854,8 +1854,10 @@ void CompileBroker::invoke_compiler_on_method(CompileTask* task) {
     tty->print("%7d ", (int) tty->time_stamp().milliseconds());  // print timestamp
     tty->print("%4d ", compile_id);    // print compilation number
     tty->print("%s ", (is_osr ? "%" : " "));
-    int code_size = (task->code() == NULL) ? 0 : task->code()->total_size();
-    tty->print_cr("size: %d time: %d inlined: %d bytes", code_size, (int)time.milliseconds(), task->num_inlined_bytecodes());
+    if (task->code() != NULL) {
+      tty->print("size: %d(%d) ", task->code()->total_size(), task->code()->insts_size());
+    }
+    tty->print_cr("time: %d inlined: %d bytes", (int)time.milliseconds(), task->num_inlined_bytecodes());
   }
 
   if (PrintCodeCacheOnCompilation)

src/share/vm/opto/bytecodeInfo.cpp

@@ -97,6 +97,11 @@ static bool is_init_with_ea(ciMethod* callee_method,
          );
 }
 
+static bool is_unboxing_method(ciMethod* callee_method, Compile* C) {
+  // Force inlining unboxing accessor.
+  return C->eliminate_boxing() && callee_method->is_unboxing_method();
+}
+
 // positive filter: should callee be inlined?
 bool InlineTree::should_inline(ciMethod* callee_method, ciMethod* caller_method,
                                int caller_bci, ciCallProfile& profile,
@@ -144,6 +149,7 @@ bool InlineTree::should_inline(ciMethod* callee_method, ciMethod* caller_method,
   // bump the max size if the call is frequent
   if ((freq >= InlineFrequencyRatio) ||
       (call_site_count >= InlineFrequencyCount) ||
+      is_unboxing_method(callee_method, C) ||
       is_init_with_ea(callee_method, caller_method, C)) {
 
     max_inline_size = C->freq_inline_size();
@@ -237,8 +243,25 @@ bool InlineTree::should_not_inline(ciMethod *callee_method,
     return false;
   }
 
+  if (callee_method->should_not_inline()) {
+    set_msg("disallowed by CompilerOracle");
+    return true;
+  }
+
+#ifndef PRODUCT
+  if (ciReplay::should_not_inline(callee_method)) {
+    set_msg("disallowed by ciReplay");
+    return true;
+  }
+#endif
+
   // Now perform checks which are heuristic
 
+  if (is_unboxing_method(callee_method, C)) {
+    // Inline unboxing methods.
+    return false;
+  }
+
   if (!callee_method->force_inline()) {
     if (callee_method->has_compiled_code() &&
         callee_method->instructions_size() > InlineSmallCode) {
@@ -260,18 +283,6 @@ bool InlineTree::should_not_inline(ciMethod *callee_method,
     }
   }
 
-  if (callee_method->should_not_inline()) {
-    set_msg("disallowed by CompilerOracle");
-    return true;
-  }
-
-#ifndef PRODUCT
-  if (ciReplay::should_not_inline(callee_method)) {
-    set_msg("disallowed by ciReplay");
-    return true;
-  }
-#endif
-
   if (UseStringCache) {
     // Do not inline StringCache::profile() method used only at the beginning.
     if (callee_method->name() == ciSymbol::profile_name() &&
@@ -296,9 +307,8 @@ bool InlineTree::should_not_inline(ciMethod *callee_method,
     }
 
     if (is_init_with_ea(callee_method, caller_method, C)) {
-
       // Escape Analysis: inline all executed constructors
-
+      return false;
     } else if (!callee_method->was_executed_more_than(MIN2(MinInliningThreshold,
                                                            CompileThreshold >> 1))) {
       set_msg("executed < MinInliningThreshold times");

src/share/vm/opto/c2_globals.hpp

@@ -442,12 +442,15 @@
   notproduct(bool, PrintEliminateLocks, false,                              \
           "Print out when locks are eliminated")                            \
                                                                             \
-  diagnostic(bool, EliminateAutoBox, false,                                 \
-          "Private flag to control optimizations for autobox elimination")  \
+  product(bool, EliminateAutoBox, true,                                     \
+          "Control optimizations for autobox elimination")                  \
                                                                             \
   product(intx, AutoBoxCacheMax, 128,                                       \
           "Sets max value cached by the java.lang.Integer autobox cache")   \
                                                                             \
+  experimental(bool, AggressiveUnboxing, false,                             \
+          "Control optimizations for aggressive boxing elimination")        \
+                                                                            \
   product(bool, DoEscapeAnalysis, true,                                     \
           "Perform escape analysis")                                        \
                                                                             \

src/share/vm/opto/c2compiler.cpp

@@ -125,9 +125,10 @@ void C2Compiler::compile_method(ciEnv* env,
   bool subsume_loads = SubsumeLoads;
   bool do_escape_analysis = DoEscapeAnalysis &&
     !env->jvmti_can_access_local_variables();
+  bool eliminate_boxing = EliminateAutoBox;
   while (!env->failing()) {
     // Attempt to compile while subsuming loads into machine instructions.
-    Compile C(env, this, target, entry_bci, subsume_loads, do_escape_analysis);
+    Compile C(env, this, target, entry_bci, subsume_loads, do_escape_analysis, eliminate_boxing);
 
 
     // Check result and retry if appropriate.
@@ -142,6 +143,12 @@ void C2Compiler::compile_method(ciEnv* env,
         do_escape_analysis = false;
         continue;  // retry
       }
+      if (C.has_boxed_value()) {
+        // Recompile without boxing elimination regardless failure reason.
+        assert(eliminate_boxing, "must make progress");
+        eliminate_boxing = false;
+        continue;  // retry
+      }
       // Pass any other failure reason up to the ciEnv.
       // Note that serious, irreversible failures are already logged
       // on the ciEnv via env->record_method_not_compilable().

src/share/vm/opto/callGenerator.cpp

@@ -134,7 +134,7 @@ JVMState* DirectCallGenerator::generate(JVMState* jvms) {
     kit.C->log()->elem("direct_call bci='%d'", jvms->bci());
   }
 
-  CallStaticJavaNode *call = new (kit.C) CallStaticJavaNode(tf(), target, method(), kit.bci());
+  CallStaticJavaNode *call = new (kit.C) CallStaticJavaNode(kit.C, tf(), target, method(), kit.bci());
   _call_node = call;  // Save the call node in case we need it later
   if (!is_static) {
     // Make an explicit receiver null_check as part of this call.
@@ -304,29 +304,34 @@ class LateInlineCallGenerator : public DirectCallGenerator {
 
 void LateInlineCallGenerator::do_late_inline() {
   // Can't inline it
-  if (call_node() == NULL || call_node()->outcnt() == 0 ||
-      call_node()->in(0) == NULL || call_node()->in(0)->is_top()) {
+  CallStaticJavaNode* call = call_node();
+  if (call == NULL || call->outcnt() == 0 ||
+      call->in(0) == NULL || call->in(0)->is_top()) {
     return;
   }
 
-  const TypeTuple *r = call_node()->tf()->domain();
+  const TypeTuple *r = call->tf()->domain();
   for (int i1 = 0; i1 < method()->arg_size(); i1++) {
-    if (call_node()->in(TypeFunc::Parms + i1)->is_top() && r->field_at(TypeFunc::Parms + i1) != Type::HALF) {
+    if (call->in(TypeFunc::Parms + i1)->is_top() && r->field_at(TypeFunc::Parms + i1) != Type::HALF) {
       assert(Compile::current()->inlining_incrementally(), "shouldn't happen during parsing");
       return;
     }
   }
 
-  if (call_node()->in(TypeFunc::Memory)->is_top()) {
+  if (call->in(TypeFunc::Memory)->is_top()) {
     assert(Compile::current()->inlining_incrementally(), "shouldn't happen during parsing");
     return;
   }
 
-  CallStaticJavaNode* call = call_node();
+  Compile* C = Compile::current();
+  // Remove inlined methods from Compiler's lists.
+  if (call->is_macro()) {
+    C->remove_macro_node(call);
+  }
 
   // Make a clone of the JVMState that appropriate to use for driving a parse
-  Compile* C = Compile::current();
-  JVMState* jvms     = call->jvms()->clone_shallow(C);
+  JVMState* old_jvms = call->jvms();
+  JVMState* jvms = old_jvms->clone_shallow(C);
   uint size = call->req();
   SafePointNode* map = new (C) SafePointNode(size, jvms);
   for (uint i1 = 0; i1 < size; i1++) {
@@ -340,16 +345,23 @@ void LateInlineCallGenerator::do_late_inline() {
     map->set_req(TypeFunc::Memory, mem);
   }
 
-  // Make enough space for the expression stack and transfer the incoming arguments
-  int nargs    = method()->arg_size();
+  uint nargs = method()->arg_size();
+  // blow away old call arguments
+  Node* top = C->top();
+  for (uint i1 = 0; i1 < nargs; i1++) {
+    map->set_req(TypeFunc::Parms + i1, top);
+  }
   jvms->set_map(map);
+
+  // Make enough space in the expression stack to transfer
+  // the incoming arguments and return value.
   map->ensure_stack(jvms, jvms->method()->max_stack());
-  if (nargs > 0) {
-    for (int i1 = 0; i1 < nargs; i1++) {
-      map->set_req(i1 + jvms->argoff(), call->in(TypeFunc::Parms + i1));
-    }
+  for (uint i1 = 0; i1 < nargs; i1++) {
+    map->set_argument(jvms, i1, call->in(TypeFunc::Parms + i1));
   }
 
+  // This check is done here because for_method_handle_inline() method
+  // needs jvms for inlined state.
   if (!do_late_inline_check(jvms)) {
     map->disconnect_inputs(NULL, C);
     return;
@@ -480,6 +492,26 @@ CallGenerator* CallGenerator::for_string_late_inline(ciMethod* method, CallGener
   return new LateInlineStringCallGenerator(method, inline_cg);
 }
 
+class LateInlineBoxingCallGenerator : public LateInlineCallGenerator {
+
+ public:
+  LateInlineBoxingCallGenerator(ciMethod* method, CallGenerator* inline_cg) :
+    LateInlineCallGenerator(method, inline_cg) {}
+
+  virtual JVMState* generate(JVMState* jvms) {
+    Compile *C = Compile::current();
+    C->print_inlining_skip(this);
+
+    C->add_boxing_late_inline(this);
+
+    JVMState* new_jvms =  DirectCallGenerator::generate(jvms);
+    return new_jvms;
+  }
+};
+
+CallGenerator* CallGenerator::for_boxing_late_inline(ciMethod* method, CallGenerator* inline_cg) {
+  return new LateInlineBoxingCallGenerator(method, inline_cg);
+}
 
 //---------------------------WarmCallGenerator--------------------------------
 // Internal class which handles initial deferral of inlining decisions.

src/share/vm/opto/callGenerator.hpp

@@ -125,6 +125,7 @@ class CallGenerator : public ResourceObj {
   static CallGenerator* for_late_inline(ciMethod* m, CallGenerator* inline_cg);
   static CallGenerator* for_mh_late_inline(ciMethod* caller, ciMethod* callee, bool input_not_const);
   static CallGenerator* for_string_late_inline(ciMethod* m, CallGenerator* inline_cg);
+  static CallGenerator* for_boxing_late_inline(ciMethod* m, CallGenerator* inline_cg);
 
   // How to make a call but defer the decision whether to inline or not.
   static CallGenerator* for_warm_call(WarmCallInfo* ci,

src/share/vm/opto/callnode.cpp

@@ -523,7 +523,9 @@ void JVMState::dump_spec(outputStream *st) const {
 
 
 void JVMState::dump_on(outputStream* st) const {
-  if (_map && !((uintptr_t)_map & 1)) {
+  bool print_map = _map && !((uintptr_t)_map & 1) &&
+                  ((caller() == NULL) || (caller()->map() != _map));
+  if (print_map) {
     if (_map->len() > _map->req()) {  // _map->has_exceptions()
       Node* ex = _map->in(_map->req());  // _map->next_exception()
       // skip the first one; it's already being printed
@@ -532,7 +534,10 @@ void JVMState::dump_on(outputStream* st) const {
         ex->dump(1);
       }
     }
-    _map->dump(2);
+    _map->dump(Verbose ? 2 : 1);
+  }
+  if (caller() != NULL) {
+    caller()->dump_on(st);
   }
   st->print("JVMS depth=%d loc=%d stk=%d arg=%d mon=%d scalar=%d end=%d mondepth=%d sp=%d bci=%d reexecute=%s method=",
              depth(), locoff(), stkoff(), argoff(), monoff(), scloff(), endoff(), monitor_depth(), sp(), bci(), should_reexecute()?"true":"false");
@@ -546,9 +551,6 @@ void JVMState::dump_on(outputStream* st) const {
       _method->print_codes_on(bci(), bci()+1, st);
     }
   }
-  if (caller() != NULL) {
-    caller()->dump_on(st);
-  }
 }
 
 // Extra way to dump a jvms from the debugger,
@@ -584,6 +586,15 @@ JVMState* JVMState::clone_deep(Compile* C) const {
   return n;
 }
 
+/**
+ * Reset map for all callers
+ */
+void JVMState::set_map_deep(SafePointNode* map) {
+  for (JVMState* p = this; p->_caller != NULL; p = p->_caller) {
+    p->set_map(map);
+  }
+}
+
 //=============================================================================
 uint CallNode::cmp( const Node &n ) const
 { return _tf == ((CallNode&)n)._tf && _jvms == ((CallNode&)n)._jvms; }
@@ -663,17 +674,49 @@ uint CallNode::match_edge(uint idx) const {
 // Determine whether the call could modify the field of the specified
 // instance at the specified offset.
 //
-bool CallNode::may_modify(const TypePtr *addr_t, PhaseTransform *phase) {
-  const TypeOopPtr *adrInst_t  = addr_t->isa_oopptr();
-
-  // If not an OopPtr or not an instance type, assume the worst.
-  // Note: currently this method is called only for instance types.
-  if (adrInst_t == NULL || !adrInst_t->is_known_instance()) {
-    return true;
+bool CallNode::may_modify(const TypeOopPtr *t_oop, PhaseTransform *phase) {
+  assert((t_oop != NULL), "sanity");
+  if (t_oop->is_known_instance()) {
+    // The instance_id is set only for scalar-replaceable allocations which
+    // are not passed as arguments according to Escape Analysis.
+    return false;
   }
-  // The instance_id is set only for scalar-replaceable allocations which
-  // are not passed as arguments according to Escape Analysis.
-  return false;
+  if (t_oop->is_ptr_to_boxed_value()) {
+    ciKlass* boxing_klass = t_oop->klass();
+    if (is_CallStaticJava() && as_CallStaticJava()->is_boxing_method()) {
+      // Skip unrelated boxing methods.
+      Node* proj = proj_out(TypeFunc::Parms);
+      if ((proj == NULL) || (phase->type(proj)->is_instptr()->klass() != boxing_klass)) {
+        return false;
+      }
+    }
+    if (is_CallJava() && as_CallJava()->method() != NULL) {
+      ciMethod* meth = as_CallJava()->method();
+      if (meth->is_accessor()) {
+        return false;
+      }
+      // May modify (by reflection) if an boxing object is passed
+      // as argument or returned.
+      if (returns_pointer() && (proj_out(TypeFunc::Parms) != NULL)) {
+        Node* proj = proj_out(TypeFunc::Parms);
+        const TypeInstPtr* inst_t = phase->type(proj)->isa_instptr();
+        if ((inst_t != NULL) && (!inst_t->klass_is_exact() ||
+                                 (inst_t->klass() == boxing_klass))) {
+          return true;
+        }
+      }
+      const TypeTuple* d = tf()->domain();
+      for (uint i = TypeFunc::Parms; i < d->cnt(); i++) {
+        const TypeInstPtr* inst_t = d->field_at(i)->isa_instptr();
+        if ((inst_t != NULL) && (!inst_t->klass_is_exact() ||
+                                 (inst_t->klass() == boxing_klass))) {
+          return true;
+        }
+      }
+      return false;
+    }
+  }
+  return true;
 }
 
 // Does this call have a direct reference to n other than debug information?
@@ -1020,6 +1063,7 @@ void SafePointNode::grow_stack(JVMState* jvms, uint grow_by) {
   int scloff = jvms->scloff();
   int endoff = jvms->endoff();
   assert(endoff == (int)req(), "no other states or debug info after me");
+  assert(jvms->scl_size() == 0, "parsed code should not have scalar objects");
   Node* top = Compile::current()->top();
   for (uint i = 0; i < grow_by; i++) {
     ins_req(monoff, top);
@@ -1035,6 +1079,7 @@ void SafePointNode::push_monitor(const FastLockNode *lock) {
   const int MonitorEdges = 2;
   assert(JVMState::logMonitorEdges == exact_log2(MonitorEdges), "correct MonitorEdges");
   assert(req() == jvms()->endoff(), "correct sizing");
+  assert((jvms()->scl_size() == 0), "parsed code should not have scalar objects");
   int nextmon = jvms()->scloff();
   if (GenerateSynchronizationCode) {
     add_req(lock->box_node());
@@ -1050,6 +1095,7 @@ void SafePointNode::push_monitor(const FastLockNode *lock) {
 
 void SafePointNode::pop_monitor() {
   // Delete last monitor from debug info
+  assert((jvms()->scl_size() == 0), "parsed code should not have scalar objects");
   debug_only(int num_before_pop = jvms()->nof_monitors());
   const int MonitorEdges = (1<<JVMState::logMonitorEdges);
   int scloff = jvms()->scloff();
@@ -1154,6 +1200,7 @@ AllocateNode::AllocateNode(Compile* C, const TypeFunc *atype,
   init_class_id(Class_Allocate);
   init_flags(Flag_is_macro);
   _is_scalar_replaceable = false;
+  _is_non_escaping = false;
   Node *topnode = C->top();
 
   init_req( TypeFunc::Control  , ctrl );
@@ -1169,8 +1216,6 @@ AllocateNode::AllocateNode(Compile* C, const TypeFunc *atype,
 }
 
 //=============================================================================
-uint AllocateArrayNode::size_of() const { return sizeof(*this); }
-
 Node* AllocateArrayNode::Ideal(PhaseGVN *phase, bool can_reshape) {
   if (remove_dead_region(phase, can_reshape))  return this;
   // Don't bother trying to transform a dead node
@@ -1235,6 +1280,8 @@ Node *AllocateArrayNode::make_ideal_length(const TypeOopPtr* oop_type, PhaseTran
       //   - the narrow_length is 0
       //   - the narrow_length is not wider than length
       assert(narrow_length_type == TypeInt::ZERO ||
+             length_type->is_con() && narrow_length_type->is_con() &&
+                (narrow_length_type->_hi <= length_type->_lo) ||
              (narrow_length_type->_hi <= length_type->_hi &&
               narrow_length_type->_lo >= length_type->_lo),
              "narrow type must be narrower than length type");

src/share/vm/opto/callnode.hpp

@@ -49,6 +49,7 @@ class       CallLeafNode;
 class         CallLeafNoFPNode;
 class     AllocateNode;
 class       AllocateArrayNode;
+class     BoxLockNode;
 class     LockNode;
 class     UnlockNode;
 class JVMState;
@@ -235,7 +236,6 @@ public:
 
   int            loc_size() const { return stkoff() - locoff(); }
   int            stk_size() const { return monoff() - stkoff(); }
-  int            arg_size() const { return monoff() - argoff(); }
   int            mon_size() const { return scloff() - monoff(); }
   int            scl_size() const { return endoff() - scloff(); }
 
@@ -298,6 +298,7 @@ public:
   // Miscellaneous utility functions
   JVMState* clone_deep(Compile* C) const;    // recursively clones caller chain
   JVMState* clone_shallow(Compile* C) const; // retains uncloned caller
+  void      set_map_deep(SafePointNode *map);// reset map for all callers
 
 #ifndef PRODUCT
   void      format(PhaseRegAlloc *regalloc, const Node *n, outputStream* st) const;
@@ -439,7 +440,7 @@ public:
   static  bool           needs_polling_address_input();
 
 #ifndef PRODUCT
-  virtual void              dump_spec(outputStream *st) const;
+  virtual void           dump_spec(outputStream *st) const;
 #endif
 };
 
@@ -554,10 +555,10 @@ public:
   virtual bool        guaranteed_safepoint()  { return true; }
   // For macro nodes, the JVMState gets modified during expansion, so when cloning
   // the node the JVMState must be cloned.
-  virtual void        clone_jvms() { }   // default is not to clone
+  virtual void        clone_jvms(Compile* C) { }   // default is not to clone
 
   // Returns true if the call may modify n
-  virtual bool        may_modify(const TypePtr *addr_t, PhaseTransform *phase);
+  virtual bool        may_modify(const TypeOopPtr *t_oop, PhaseTransform *phase);
   // Does this node have a use of n other than in debug information?
   bool                has_non_debug_use(Node *n);
   // Returns the unique CheckCastPP of a call
@@ -630,9 +631,15 @@ class CallStaticJavaNode : public CallJavaNode {
   virtual uint cmp( const Node &n ) const;
   virtual uint size_of() const; // Size is bigger
 public:
-  CallStaticJavaNode(const TypeFunc* tf, address addr, ciMethod* method, int bci)
+  CallStaticJavaNode(Compile* C, const TypeFunc* tf, address addr, ciMethod* method, int bci)
     : CallJavaNode(tf, addr, method, bci), _name(NULL) {
     init_class_id(Class_CallStaticJava);
+    if (C->eliminate_boxing() && (method != NULL) && method->is_boxing_method()) {
+      init_flags(Flag_is_macro);
+      C->add_macro_node(this);
+    }
+    _is_scalar_replaceable = false;
+    _is_non_escaping = false;
   }
   CallStaticJavaNode(const TypeFunc* tf, address addr, const char* name, int bci,
                      const TypePtr* adr_type)
@@ -640,13 +647,31 @@ public:
     init_class_id(Class_CallStaticJava);
     // This node calls a runtime stub, which often has narrow memory effects.
     _adr_type = adr_type;
+    _is_scalar_replaceable = false;
+    _is_non_escaping = false;
   }
-  const char *_name;            // Runtime wrapper name
+  const char *_name;      // Runtime wrapper name
+
+  // Result of Escape Analysis
+  bool _is_scalar_replaceable;
+  bool _is_non_escaping;
 
   // If this is an uncommon trap, return the request code, else zero.
   int uncommon_trap_request() const;
   static int extract_uncommon_trap_request(const Node* call);
 
+  bool is_boxing_method() const {
+    return is_macro() && (method() != NULL) && method()->is_boxing_method();
+  }
+  // Later inlining modifies the JVMState, so we need to clone it
+  // when the call node is cloned (because it is macro node).
+  virtual void  clone_jvms(Compile* C) {
+    if ((jvms() != NULL) && is_boxing_method()) {
+      set_jvms(jvms()->clone_deep(C));
+      jvms()->set_map_deep(this);
+    }
+  }
+
   virtual int         Opcode() const;
 #ifndef PRODUCT
   virtual void        dump_spec(outputStream *st) const;
@@ -748,12 +773,12 @@ public:
     ParmLimit
   };
 
-  static const TypeFunc* alloc_type() {
+  static const TypeFunc* alloc_type(const Type* t) {
     const Type** fields = TypeTuple::fields(ParmLimit - TypeFunc::Parms);
     fields[AllocSize]   = TypeInt::POS;
     fields[KlassNode]   = TypeInstPtr::NOTNULL;
     fields[InitialTest] = TypeInt::BOOL;
-    fields[ALength]     = TypeInt::INT;  // length (can be a bad length)
+    fields[ALength]     = t;  // length (can be a bad length)
 
     const TypeTuple *domain = TypeTuple::make(ParmLimit, fields);
 
@@ -766,21 +791,26 @@ public:
     return TypeFunc::make(domain, range);
   }
 
-  bool _is_scalar_replaceable;  // Result of Escape Analysis
+  // Result of Escape Analysis
+  bool _is_scalar_replaceable;
+  bool _is_non_escaping;
 
   virtual uint size_of() const; // Size is bigger
   AllocateNode(Compile* C, const TypeFunc *atype, Node *ctrl, Node *mem, Node *abio,
                Node *size, Node *klass_node, Node *initial_test);
   // Expansion modifies the JVMState, so we need to clone it
-  virtual void  clone_jvms() {
-    set_jvms(jvms()->clone_deep(Compile::current()));
+  virtual void  clone_jvms(Compile* C) {
+    if (jvms() != NULL) {
+      set_jvms(jvms()->clone_deep(C));
+      jvms()->set_map_deep(this);
+    }
   }
   virtual int Opcode() const;
   virtual uint ideal_reg() const { return Op_RegP; }
   virtual bool        guaranteed_safepoint()  { return false; }
 
   // allocations do not modify their arguments
-  virtual bool        may_modify(const TypePtr *addr_t, PhaseTransform *phase) { return false;}
+  virtual bool        may_modify(const TypeOopPtr *t_oop, PhaseTransform *phase) { return false;}
 
   // Pattern-match a possible usage of AllocateNode.
   // Return null if no allocation is recognized.
@@ -815,10 +845,6 @@ public:
   // are defined in graphKit.cpp, which sets up the bidirectional relation.)
   InitializeNode* initialization();
 
-  // Return the corresponding storestore barrier (or null if none).
-  // Walks out edges to find it...
-  MemBarStoreStoreNode* storestore();
-
   // Convenience for initialization->maybe_set_complete(phase)
   bool maybe_set_complete(PhaseGVN* phase);
 };
@@ -840,7 +866,6 @@ public:
     set_req(AllocateNode::ALength,        count_val);
   }
   virtual int Opcode() const;
-  virtual uint size_of() const; // Size is bigger
   virtual Node *Ideal(PhaseGVN *phase, bool can_reshape);
 
   // Dig the length operand out of a array allocation site.
@@ -918,7 +943,7 @@ public:
   void set_nested()      { _kind = Nested; set_eliminated_lock_counter(); }
 
   // locking does not modify its arguments
-  virtual bool may_modify(const TypePtr *addr_t, PhaseTransform *phase){ return false;}
+  virtual bool may_modify(const TypeOopPtr *t_oop, PhaseTransform *phase){ return false;}
 
 #ifndef PRODUCT
   void create_lock_counter(JVMState* s);
@@ -965,8 +990,11 @@ public:
 
   virtual Node *Ideal(PhaseGVN *phase, bool can_reshape);
   // Expansion modifies the JVMState, so we need to clone it
-  virtual void  clone_jvms() {
-    set_jvms(jvms()->clone_deep(Compile::current()));
+  virtual void  clone_jvms(Compile* C) {
+    if (jvms() != NULL) {
+      set_jvms(jvms()->clone_deep(C));
+      jvms()->set_map_deep(this);
+    }
   }
 
   bool is_nested_lock_region(); // Is this Lock nested?

src/share/vm/opto/cfgnode.cpp

@@ -806,7 +806,7 @@ PhiNode* PhiNode::split_out_instance(const TypePtr* at, PhaseIterGVN *igvn) cons
       Node *in = ophi->in(i);
       if (in == NULL || igvn->type(in) == Type::TOP)
         continue;
-      Node *opt = MemNode::optimize_simple_memory_chain(in, at, igvn);
+      Node *opt = MemNode::optimize_simple_memory_chain(in, t_oop, NULL, igvn);
       PhiNode *optphi = opt->is_Phi() ? opt->as_Phi() : NULL;
       if (optphi != NULL && optphi->adr_type() == TypePtr::BOTTOM) {
         opt = node_map[optphi->_idx];
@@ -1921,7 +1921,7 @@ Node *PhiNode::Ideal(PhaseGVN *phase, bool can_reshape) {
     const TypePtr* at = adr_type();
     for( uint i=1; i<req(); ++i ) {// For all paths in
       Node *ii = in(i);
-      Node *new_in = MemNode::optimize_memory_chain(ii, at, phase);
+      Node *new_in = MemNode::optimize_memory_chain(ii, at, NULL, phase);
       if (ii != new_in ) {
         set_req(i, new_in);
         progress = this;

src/share/vm/opto/compile.cpp

@@ -418,6 +418,7 @@ void Compile::remove_useless_nodes(Unique_Node_List &useful) {
   }
   // clean up the late inline lists
   remove_useless_late_inlines(&_string_late_inlines, useful);
+  remove_useless_late_inlines(&_boxing_late_inlines, useful);
   remove_useless_late_inlines(&_late_inlines, useful);
   debug_only(verify_graph_edges(true/*check for no_dead_code*/);)
 }
@@ -485,6 +486,12 @@ void Compile::print_compile_messages() {
     tty->print_cr("** Bailout: Recompile without escape analysis          **");
     tty->print_cr("*********************************************************");
   }
+  if (_eliminate_boxing != EliminateAutoBox && PrintOpto) {
+    // Recompiling without boxing elimination
+    tty->print_cr("*********************************************************");
+    tty->print_cr("** Bailout: Recompile without boxing elimination       **");
+    tty->print_cr("*********************************************************");
+  }
   if (env()->break_at_compile()) {
     // Open the debugger when compiling this method.
     tty->print("### Breaking when compiling: ");
@@ -601,7 +608,8 @@ debug_only( int Compile::_debug_idx = 100000; )
 // the continuation bci for on stack replacement.
 
 
-Compile::Compile( ciEnv* ci_env, C2Compiler* compiler, ciMethod* target, int osr_bci, bool subsume_loads, bool do_escape_analysis )
+Compile::Compile( ciEnv* ci_env, C2Compiler* compiler, ciMethod* target, int osr_bci,
+                  bool subsume_loads, bool do_escape_analysis, bool eliminate_boxing )
                 : Phase(Compiler),
                   _env(ci_env),
                   _log(ci_env->log()),
@@ -617,6 +625,7 @@ Compile::Compile( ciEnv* ci_env, C2Compiler* compiler, ciMethod* target, int osr
                   _warm_calls(NULL),
                   _subsume_loads(subsume_loads),
                   _do_escape_analysis(do_escape_analysis),
+                  _eliminate_boxing(eliminate_boxing),
                   _failure_reason(NULL),
                   _code_buffer("Compile::Fill_buffer"),
                   _orig_pc_slot(0),
@@ -638,6 +647,7 @@ Compile::Compile( ciEnv* ci_env, C2Compiler* compiler, ciMethod* target, int osr
                   _congraph(NULL),
                   _late_inlines(comp_arena(), 2, 0, NULL),
                   _string_late_inlines(comp_arena(), 2, 0, NULL),
+                  _boxing_late_inlines(comp_arena(), 2, 0, NULL),
                   _late_inlines_pos(0),
                   _number_of_mh_late_inlines(0),
                   _inlining_progress(false),
@@ -906,6 +916,7 @@ Compile::Compile( ciEnv* ci_env,
     _orig_pc_slot_offset_in_bytes(0),
     _subsume_loads(true),
     _do_escape_analysis(false),
+    _eliminate_boxing(false),
     _failure_reason(NULL),
     _code_buffer("Compile::Fill_buffer"),
     _has_method_handle_invokes(false),
@@ -1016,6 +1027,7 @@ void Compile::Init(int aliaslevel) {
   set_has_split_ifs(false);
   set_has_loops(has_method() && method()->has_loops()); // first approximation
   set_has_stringbuilder(false);
+  set_has_boxed_value(false);
   _trap_can_recompile = false;  // no traps emitted yet
   _major_progress = true; // start out assuming good things will happen
   set_has_unsafe_access(false);
@@ -1807,6 +1819,38 @@ void Compile::inline_string_calls(bool parse_time) {
   _string_late_inlines.trunc_to(0);
 }
 
+// Late inlining of boxing methods
+void Compile::inline_boxing_calls(PhaseIterGVN& igvn) {
+  if (_boxing_late_inlines.length() > 0) {
+    assert(has_boxed_value(), "inconsistent");
+
+    PhaseGVN* gvn = initial_gvn();
+    set_inlining_incrementally(true);
+
+    assert( igvn._worklist.size() == 0, "should be done with igvn" );
+    for_igvn()->clear();
+    gvn->replace_with(&igvn);
+
+    while (_boxing_late_inlines.length() > 0) {
+      CallGenerator* cg = _boxing_late_inlines.pop();
+      cg->do_late_inline();
+      if (failing())  return;
+    }
+    _boxing_late_inlines.trunc_to(0);
+
+    {
+      ResourceMark rm;
+      PhaseRemoveUseless pru(gvn, for_igvn());
+    }
+
+    igvn = PhaseIterGVN(gvn);
+    igvn.optimize();
+
+    set_inlining_progress(false);
+    set_inlining_incrementally(false);
+  }
+}
+
 void Compile::inline_incrementally_one(PhaseIterGVN& igvn) {
   assert(IncrementalInline, "incremental inlining should be on");
   PhaseGVN* gvn = initial_gvn();
@@ -1831,7 +1875,7 @@ void Compile::inline_incrementally_one(PhaseIterGVN& igvn) {
 
   {
     ResourceMark rm;
-    PhaseRemoveUseless pru(C->initial_gvn(), C->for_igvn());
+    PhaseRemoveUseless pru(gvn, for_igvn());
   }
 
   igvn = PhaseIterGVN(gvn);
@@ -1929,12 +1973,25 @@ void Compile::Optimize() {
 
   if (failing())  return;
 
-  inline_incrementally(igvn);
+  {
+    NOT_PRODUCT( TracePhase t2("incrementalInline", &_t_incrInline, TimeCompiler); )
+    inline_incrementally(igvn);
+  }
 
   print_method("Incremental Inline", 2);
 
   if (failing())  return;
 
+  if (eliminate_boxing()) {
+    NOT_PRODUCT( TracePhase t2("incrementalInline", &_t_incrInline, TimeCompiler); )
+    // Inline valueOf() methods now.
+    inline_boxing_calls(igvn);
+
+    print_method("Incremental Boxing Inline", 2);
+
+    if (failing())  return;
+  }
+
   // No more new expensive nodes will be added to the list from here
   // so keep only the actual candidates for optimizations.
   cleanup_expensive_nodes(igvn);
@@ -2896,6 +2953,7 @@ void Compile::final_graph_reshaping_impl( Node *n, Final_Reshape_Counts &frc) {
     }
     break;
   case Op_MemBarStoreStore:
+  case Op_MemBarRelease:
     // Break the link with AllocateNode: it is no longer useful and
     // confuses register allocation.
     if (n->req() > MemBarNode::Precedent) {

src/share/vm/opto/compile.hpp

@@ -262,6 +262,7 @@ class Compile : public Phase {
   const bool            _save_argument_registers; // save/restore arg regs for trampolines
   const bool            _subsume_loads;         // Load can be matched as part of a larger op.
   const bool            _do_escape_analysis;    // Do escape analysis.
+  const bool            _eliminate_boxing;      // Do boxing elimination.
   ciMethod*             _method;                // The method being compiled.
   int                   _entry_bci;             // entry bci for osr methods.
   const TypeFunc*       _tf;                    // My kind of signature
@@ -287,6 +288,7 @@ class Compile : public Phase {
   bool                  _has_split_ifs;         // True if the method _may_ have some split-if
   bool                  _has_unsafe_access;     // True if the method _may_ produce faults in unsafe loads or stores.
   bool                  _has_stringbuilder;     // True StringBuffers or StringBuilders are allocated
+  bool                  _has_boxed_value;       // True if a boxed object is allocated
   int                   _max_vector_size;       // Maximum size of generated vectors
   uint                  _trap_hist[trapHistLength];  // Cumulative traps
   bool                  _trap_can_recompile;    // Have we emitted a recompiling trap?
@@ -375,6 +377,8 @@ class Compile : public Phase {
                                                       // main parsing has finished.
   GrowableArray<CallGenerator*> _string_late_inlines; // same but for string operations
 
+  GrowableArray<CallGenerator*> _boxing_late_inlines; // same but for boxing operations
+
   int                           _late_inlines_pos;    // Where in the queue should the next late inlining candidate go (emulate depth first inlining)
   uint                          _number_of_mh_late_inlines; // number of method handle late inlining still pending
 
@@ -486,8 +490,12 @@ class Compile : public Phase {
   // instructions that subsume a load may result in an unschedulable
   // instruction sequence.
   bool              subsume_loads() const       { return _subsume_loads; }
-  // Do escape analysis.
+  /** Do escape analysis. */
   bool              do_escape_analysis() const  { return _do_escape_analysis; }
+  /** Do boxing elimination. */
+  bool              eliminate_boxing() const    { return _eliminate_boxing; }
+  /** Do aggressive boxing elimination. */
+  bool              aggressive_unboxing() const { return _eliminate_boxing && AggressiveUnboxing; }
   bool              save_argument_registers() const { return _save_argument_registers; }
 
 
@@ -527,6 +535,8 @@ class Compile : public Phase {
   void          set_has_unsafe_access(bool z)   { _has_unsafe_access = z; }
   bool              has_stringbuilder() const   { return _has_stringbuilder; }
   void          set_has_stringbuilder(bool z)   { _has_stringbuilder = z; }
+  bool              has_boxed_value() const     { return _has_boxed_value; }
+  void          set_has_boxed_value(bool z)     { _has_boxed_value = z; }
   int               max_vector_size() const     { return _max_vector_size; }
   void          set_max_vector_size(int s)      { _max_vector_size = s; }
   void          set_trap_count(uint r, uint c)  { assert(r < trapHistLength, "oob");        _trap_hist[r] = c; }
@@ -579,12 +589,12 @@ class Compile : public Phase {
 #endif
   }
 
-  int           macro_count()                   { return _macro_nodes->length(); }
-  int           predicate_count()               { return _predicate_opaqs->length();}
-  int           expensive_count()               { return _expensive_nodes->length(); }
-  Node*         macro_node(int idx)             { return _macro_nodes->at(idx); }
-  Node*         predicate_opaque1_node(int idx) { return _predicate_opaqs->at(idx);}
-  Node*         expensive_node(int idx)         { return _expensive_nodes->at(idx); }
+  int           macro_count()             const { return _macro_nodes->length(); }
+  int           predicate_count()         const { return _predicate_opaqs->length();}
+  int           expensive_count()         const { return _expensive_nodes->length(); }
+  Node*         macro_node(int idx)       const { return _macro_nodes->at(idx); }
+  Node*         predicate_opaque1_node(int idx) const { return _predicate_opaqs->at(idx);}
+  Node*         expensive_node(int idx)   const { return _expensive_nodes->at(idx); }
   ConnectionGraph* congraph()                   { return _congraph;}
   void set_congraph(ConnectionGraph* congraph)  { _congraph = congraph;}
   void add_macro_node(Node * n) {
@@ -766,7 +776,12 @@ class Compile : public Phase {
   // Decide how to build a call.
   // The profile factor is a discount to apply to this site's interp. profile.
   CallGenerator*    call_generator(ciMethod* call_method, int vtable_index, bool call_does_dispatch, JVMState* jvms, bool allow_inline, float profile_factor, bool allow_intrinsics = true, bool delayed_forbidden = false);
-  bool should_delay_inlining(ciMethod* call_method, JVMState* jvms);
+  bool should_delay_inlining(ciMethod* call_method, JVMState* jvms) {
+    return should_delay_string_inlining(call_method, jvms) ||
+           should_delay_boxing_inlining(call_method, jvms);
+  }
+  bool should_delay_string_inlining(ciMethod* call_method, JVMState* jvms);
+  bool should_delay_boxing_inlining(ciMethod* call_method, JVMState* jvms);
 
   // Helper functions to identify inlining potential at call-site
   ciMethod* optimize_virtual_call(ciMethod* caller, int bci, ciInstanceKlass* klass,
@@ -822,6 +837,10 @@ class Compile : public Phase {
     _string_late_inlines.push(cg);
   }
 
+  void              add_boxing_late_inline(CallGenerator* cg) {
+    _boxing_late_inlines.push(cg);
+  }
+
   void remove_useless_late_inlines(GrowableArray<CallGenerator*>* inlines, Unique_Node_List &useful);
 
   void dump_inlining();
@@ -841,6 +860,7 @@ class Compile : public Phase {
   void inline_incrementally_one(PhaseIterGVN& igvn);
   void inline_incrementally(PhaseIterGVN& igvn);
   void inline_string_calls(bool parse_time);
+  void inline_boxing_calls(PhaseIterGVN& igvn);
 
   // Matching, CFG layout, allocation, code generation
   PhaseCFG*         cfg()                       { return _cfg; }
@@ -913,7 +933,8 @@ class Compile : public Phase {
   // replacement, entry_bci indicates the bytecode for which to compile a
   // continuation.
   Compile(ciEnv* ci_env, C2Compiler* compiler, ciMethod* target,
-          int entry_bci, bool subsume_loads, bool do_escape_analysis);
+          int entry_bci, bool subsume_loads, bool do_escape_analysis,
+          bool eliminate_boxing);
 
   // Second major entry point.  From the TypeFunc signature, generate code
   // to pass arguments from the Java calling convention to the C calling

src/share/vm/opto/doCall.cpp

@@ -176,9 +176,12 @@ CallGenerator* Compile::call_generator(ciMethod* callee, int vtable_index, bool
           // Delay the inlining of this method to give us the
           // opportunity to perform some high level optimizations
           // first.
-          if (should_delay_inlining(callee, jvms)) {
+          if (should_delay_string_inlining(callee, jvms)) {
             assert(!delayed_forbidden, "strange");
             return CallGenerator::for_string_late_inline(callee, cg);
+          } else if (should_delay_boxing_inlining(callee, jvms)) {
+            assert(!delayed_forbidden, "strange");
+            return CallGenerator::for_boxing_late_inline(callee, cg);
           } else if ((should_delay || AlwaysIncrementalInline) && !delayed_forbidden) {
             return CallGenerator::for_late_inline(callee, cg);
           }
@@ -276,7 +279,7 @@ CallGenerator* Compile::call_generator(ciMethod* callee, int vtable_index, bool
 
 // Return true for methods that shouldn't be inlined early so that
 // they are easier to analyze and optimize as intrinsics.
-bool Compile::should_delay_inlining(ciMethod* call_method, JVMState* jvms) {
+bool Compile::should_delay_string_inlining(ciMethod* call_method, JVMState* jvms) {
   if (has_stringbuilder()) {
 
     if ((call_method->holder() == C->env()->StringBuilder_klass() ||
@@ -327,6 +330,13 @@ bool Compile::should_delay_inlining(ciMethod* call_method, JVMState* jvms) {
   return false;
 }
 
+bool Compile::should_delay_boxing_inlining(ciMethod* call_method, JVMState* jvms) {
+  if (eliminate_boxing() && call_method->is_boxing_method()) {
+    set_has_boxed_value(true);
+    return true;
+  }
+  return false;
+}
 
 // uncommon-trap call-sites where callee is unloaded, uninitialized or will not link
 bool Parse::can_not_compile_call_site(ciMethod *dest_method, ciInstanceKlass* klass) {

src/share/vm/opto/escape.cpp

@@ -63,15 +63,19 @@ bool ConnectionGraph::has_candidates(Compile *C) {
   // EA brings benefits only when the code has allocations and/or locks which
   // are represented by ideal Macro nodes.
   int cnt = C->macro_count();
-  for( int i=0; i < cnt; i++ ) {
+  for (int i = 0; i < cnt; i++) {
     Node *n = C->macro_node(i);
-    if ( n->is_Allocate() )
+    if (n->is_Allocate())
       return true;
-    if( n->is_Lock() ) {
+    if (n->is_Lock()) {
       Node* obj = n->as_Lock()->obj_node()->uncast();
-      if( !(obj->is_Parm() || obj->is_Con()) )
+      if (!(obj->is_Parm() || obj->is_Con()))
         return true;
     }
+    if (n->is_CallStaticJava() &&
+        n->as_CallStaticJava()->is_boxing_method()) {
+      return true;
+    }
   }
   return false;
 }
@@ -115,7 +119,7 @@ bool ConnectionGraph::compute_escape() {
   { Compile::TracePhase t3("connectionGraph", &Phase::_t_connectionGraph, true);
 
   // 1. Populate Connection Graph (CG) with PointsTo nodes.
-  ideal_nodes.map(C->unique(), NULL);  // preallocate space
+  ideal_nodes.map(C->live_nodes(), NULL);  // preallocate space
   // Initialize worklist
   if (C->root() != NULL) {
     ideal_nodes.push(C->root());
@@ -152,8 +156,11 @@ bool ConnectionGraph::compute_escape() {
       // escape status of the associated Allocate node some of them
       // may be eliminated.
       storestore_worklist.append(n);
+    } else if (n->is_MemBar() && (n->Opcode() == Op_MemBarRelease) &&
+               (n->req() > MemBarNode::Precedent)) {
+      record_for_optimizer(n);
 #ifdef ASSERT
-    } else if(n->is_AddP()) {
+    } else if (n->is_AddP()) {
       // Collect address nodes for graph verification.
       addp_worklist.append(n);
 #endif
@@ -206,8 +213,15 @@ bool ConnectionGraph::compute_escape() {
   int non_escaped_length = non_escaped_worklist.length();
   for (int next = 0; next < non_escaped_length; next++) {
     JavaObjectNode* ptn = non_escaped_worklist.at(next);
-    if (ptn->escape_state() == PointsToNode::NoEscape &&
-        ptn->scalar_replaceable()) {
+    bool noescape = (ptn->escape_state() == PointsToNode::NoEscape);
+    Node* n = ptn->ideal_node();
+    if (n->is_Allocate()) {
+      n->as_Allocate()->_is_non_escaping = noescape;
+    }
+    if (n->is_CallStaticJava()) {
+      n->as_CallStaticJava()->_is_non_escaping = noescape;
+    }
+    if (noescape && ptn->scalar_replaceable()) {
       adjust_scalar_replaceable_state(ptn);
       if (ptn->scalar_replaceable()) {
         alloc_worklist.append(ptn->ideal_node());
@@ -330,8 +344,10 @@ void ConnectionGraph::add_node_to_connection_graph(Node *n, Unique_Node_List *de
       // Don't mark as processed since call's arguments have to be processed.
       delayed_worklist->push(n);
       // Check if a call returns an object.
-      if (n->as_Call()->returns_pointer() &&
-          n->as_Call()->proj_out(TypeFunc::Parms) != NULL) {
+      if ((n->as_Call()->returns_pointer() &&
+           n->as_Call()->proj_out(TypeFunc::Parms) != NULL) ||
+          (n->is_CallStaticJava() &&
+           n->as_CallStaticJava()->is_boxing_method())) {
         add_call_node(n->as_Call());
       }
     }
@@ -387,8 +403,8 @@ void ConnectionGraph::add_node_to_connection_graph(Node *n, Unique_Node_List *de
     case Op_ConNKlass: {
       // assume all oop constants globally escape except for null
       PointsToNode::EscapeState es;
-      if (igvn->type(n) == TypePtr::NULL_PTR ||
-          igvn->type(n) == TypeNarrowOop::NULL_PTR) {
+      const Type* t = igvn->type(n);
+      if (t == TypePtr::NULL_PTR || t == TypeNarrowOop::NULL_PTR) {
         es = PointsToNode::NoEscape;
       } else {
         es = PointsToNode::GlobalEscape;
@@ -797,6 +813,9 @@ void ConnectionGraph::add_call_node(CallNode* call) {
       // Returns a newly allocated unescaped object.
       add_java_object(call, PointsToNode::NoEscape);
       ptnode_adr(call_idx)->set_scalar_replaceable(false);
+    } else if (meth->is_boxing_method()) {
+      // Returns boxing object
+      add_java_object(call, PointsToNode::NoEscape);
     } else {
       BCEscapeAnalyzer* call_analyzer = meth->get_bcea();
       call_analyzer->copy_dependencies(_compile->dependencies());
@@ -943,6 +962,9 @@ void ConnectionGraph::process_call_arguments(CallNode *call) {
       assert((name == NULL || strcmp(name, "uncommon_trap") != 0), "normal calls only");
 #endif
       ciMethod* meth = call->as_CallJava()->method();
+      if ((meth != NULL) && meth->is_boxing_method()) {
+        break; // Boxing methods do not modify any oops.
+      }
       BCEscapeAnalyzer* call_analyzer = (meth !=NULL) ? meth->get_bcea() : NULL;
       // fall-through if not a Java method or no analyzer information
       if (call_analyzer != NULL) {
@@ -2744,6 +2766,11 @@ void ConnectionGraph::split_unique_types(GrowableArray<Node *>  &alloc_worklist)
           // so it could be eliminated if it has no uses.
           alloc->as_Allocate()->_is_scalar_replaceable = true;
         }
+        if (alloc->is_CallStaticJava()) {
+          // Set the scalar_replaceable flag for boxing method
+          // so it could be eliminated if it has no uses.
+          alloc->as_CallStaticJava()->_is_scalar_replaceable = true;
+        }
         continue;
       }
       if (!n->is_CheckCastPP()) { // not unique CheckCastPP.
@@ -2782,6 +2809,11 @@ void ConnectionGraph::split_unique_types(GrowableArray<Node *>  &alloc_worklist)
         // so it could be eliminated.
         alloc->as_Allocate()->_is_scalar_replaceable = true;
       }
+      if (alloc->is_CallStaticJava()) {
+        // Set the scalar_replaceable flag for boxing method
+        // so it could be eliminated.
+        alloc->as_CallStaticJava()->_is_scalar_replaceable = true;
+      }
       set_escape_state(ptnode_adr(n->_idx), es); // CheckCastPP escape state
       // in order for an object to be scalar-replaceable, it must be:
       //   - a direct allocation (not a call returning an object)
@@ -2911,7 +2943,9 @@ void ConnectionGraph::split_unique_types(GrowableArray<Node *>  &alloc_worklist)
         // Load/store to instance's field
         memnode_worklist.append_if_missing(use);
       } else if (use->is_MemBar()) {
-        memnode_worklist.append_if_missing(use);
+        if (use->in(TypeFunc::Memory) == n) { // Ignore precedent edge
+          memnode_worklist.append_if_missing(use);
+        }
       } else if (use->is_AddP() && use->outcnt() > 0) { // No dead nodes
         Node* addp2 = find_second_addp(use, n);
         if (addp2 != NULL) {
@@ -3028,7 +3062,9 @@ void ConnectionGraph::split_unique_types(GrowableArray<Node *>  &alloc_worklist)
           continue;
         memnode_worklist.append_if_missing(use);
       } else if (use->is_MemBar()) {
-        memnode_worklist.append_if_missing(use);
+        if (use->in(TypeFunc::Memory) == n) { // Ignore precedent edge
+          memnode_worklist.append_if_missing(use);
+        }
 #ifdef ASSERT
       } else if(use->is_Mem()) {
         assert(use->in(MemNode::Memory) != n, "EA: missing memory path");
@@ -3264,7 +3300,12 @@ void ConnectionGraph::dump(GrowableArray<PointsToNode*>& ptnodes_worklist) {
     if (ptn == NULL || !ptn->is_JavaObject())
       continue;
     PointsToNode::EscapeState es = ptn->escape_state();
-    if (ptn->ideal_node()->is_Allocate() && (es == PointsToNode::NoEscape || Verbose)) {
+    if ((es != PointsToNode::NoEscape) && !Verbose) {
+      continue;
+    }
+    Node* n = ptn->ideal_node();
+    if (n->is_Allocate() || (n->is_CallStaticJava() &&
+                             n->as_CallStaticJava()->is_boxing_method())) {
       if (first) {
         tty->cr();
         tty->print("======== Connection graph for ");

src/share/vm/opto/graphKit.cpp

@@ -333,6 +333,7 @@ void GraphKit::combine_exception_states(SafePointNode* ex_map, SafePointNode* ph
   assert(ex_jvms->stkoff() == phi_map->_jvms->stkoff(), "matching locals");
   assert(ex_jvms->sp() == phi_map->_jvms->sp(), "matching stack sizes");
   assert(ex_jvms->monoff() == phi_map->_jvms->monoff(), "matching JVMS");
+  assert(ex_jvms->scloff() == phi_map->_jvms->scloff(), "matching scalar replaced objects");
   assert(ex_map->req() == phi_map->req(), "matching maps");
   uint tos = ex_jvms->stkoff() + ex_jvms->sp();
   Node*         hidden_merge_mark = root();
@@ -409,7 +410,7 @@ void GraphKit::combine_exception_states(SafePointNode* ex_map, SafePointNode* ph
         while (dst->req() > orig_width)  dst->del_req(dst->req()-1);
       } else {
         assert(dst->is_Phi(), "nobody else uses a hidden region");
-        phi = (PhiNode*)dst;
+        phi = dst->as_Phi();
       }
       if (add_multiple && src->in(0) == ex_control) {
         // Both are phis.
@@ -1438,7 +1439,12 @@ Node* GraphKit::make_load(Node* ctl, Node* adr, const Type* t, BasicType bt,
   } else {
     ld = LoadNode::make(_gvn, ctl, mem, adr, adr_type, t, bt);
   }
-  return _gvn.transform(ld);
+  ld = _gvn.transform(ld);
+  if ((bt == T_OBJECT) && C->do_escape_analysis() || C->eliminate_boxing()) {
+    // Improve graph before escape analysis and boxing elimination.
+    record_for_igvn(ld);
+  }
+  return ld;
 }
 
 Node* GraphKit::store_to_memory(Node* ctl, Node* adr, Node *val, BasicType bt,
@@ -3144,7 +3150,7 @@ Node* GraphKit::new_instance(Node* klass_node,
   set_all_memory(mem); // Create new memory state
 
   AllocateNode* alloc
-    = new (C) AllocateNode(C, AllocateNode::alloc_type(),
+    = new (C) AllocateNode(C, AllocateNode::alloc_type(Type::TOP),
                            control(), mem, i_o(),
                            size, klass_node,
                            initial_slow_test);
@@ -3285,7 +3291,7 @@ Node* GraphKit::new_array(Node* klass_node,     // array klass (maybe variable)
 
   // Create the AllocateArrayNode and its result projections
   AllocateArrayNode* alloc
-    = new (C) AllocateArrayNode(C, AllocateArrayNode::alloc_type(),
+    = new (C) AllocateArrayNode(C, AllocateArrayNode::alloc_type(TypeInt::INT),
                                 control(), mem, i_o(),
                                 size, klass_node,
                                 initial_slow_test,
@@ -3326,10 +3332,9 @@ AllocateNode* AllocateNode::Ideal_allocation(Node* ptr, PhaseTransform* phase) {
   if (ptr == NULL) {     // reduce dumb test in callers
     return NULL;
   }
-  if (ptr->is_CheckCastPP()) {  // strip a raw-to-oop cast
-    ptr = ptr->in(1);
-    if (ptr == NULL)  return NULL;
-  }
+  ptr = ptr->uncast();  // strip a raw-to-oop cast
+  if (ptr == NULL)  return NULL;
+
   if (ptr->is_Proj()) {
     Node* allo = ptr->in(0);
     if (allo != NULL && allo->is_Allocate()) {
@@ -3374,19 +3379,6 @@ InitializeNode* AllocateNode::initialization() {
   return NULL;
 }
 
-// Trace Allocate -> Proj[Parm] -> MemBarStoreStore
-MemBarStoreStoreNode* AllocateNode::storestore() {
-  ProjNode* rawoop = proj_out(AllocateNode::RawAddress);
-  if (rawoop == NULL)  return NULL;
-  for (DUIterator_Fast imax, i = rawoop->fast_outs(imax); i < imax; i++) {
-    Node* storestore = rawoop->fast_out(i);
-    if (storestore->is_MemBarStoreStore()) {
-      return storestore->as_MemBarStoreStore();
-    }
-  }
-  return NULL;
-}
-
 //----------------------------- loop predicates ---------------------------
 
 //------------------------------add_predicate_impl----------------------------

src/share/vm/opto/ifnode.cpp

@@ -673,7 +673,7 @@ const TypeInt* IfNode::filtered_int_type(PhaseGVN* gvn, Node *val, Node* if_proj
 //           /    Region
 //
 Node* IfNode::fold_compares(PhaseGVN* phase) {
-  if (!EliminateAutoBox || Opcode() != Op_If) return NULL;
+  if (!phase->C->eliminate_boxing() || Opcode() != Op_If) return NULL;
 
   Node* this_cmp = in(1)->in(1);
   if (this_cmp != NULL && this_cmp->Opcode() == Op_CmpI &&

src/share/vm/opto/library_call.cpp

@@ -3703,7 +3703,7 @@ LibraryCallKit::generate_method_call(vmIntrinsics::ID method_id, bool is_virtual
   CallJavaNode* slow_call;
   if (is_static) {
     assert(!is_virtual, "");
-    slow_call = new(C) CallStaticJavaNode(tf,
+    slow_call = new(C) CallStaticJavaNode(C, tf,
                            SharedRuntime::get_resolve_static_call_stub(),
                            method, bci());
   } else if (is_virtual) {
@@ -3722,7 +3722,7 @@ LibraryCallKit::generate_method_call(vmIntrinsics::ID method_id, bool is_virtual
                           method, vtable_index, bci());
   } else {  // neither virtual nor static:  opt_virtual
     null_check_receiver();
-    slow_call = new(C) CallStaticJavaNode(tf,
+    slow_call = new(C) CallStaticJavaNode(C, tf,
                                 SharedRuntime::get_resolve_opt_virtual_call_stub(),
                                 method, bci());
     slow_call->set_optimized_virtual(true);

src/share/vm/opto/loopPredicate.cpp

@@ -821,8 +821,8 @@ bool PhaseIdealLoop::loop_predication_impl(IdealLoopTree *loop) {
         loop->dump_head();
       }
 #endif
-    } else if (cl != NULL && loop->is_range_check_if(iff, this, invar)) {
-      assert(proj->_con == predicate_proj->_con, "must match");
+    } else if ((cl != NULL) && (proj->_con == predicate_proj->_con) &&
+               loop->is_range_check_if(iff, this, invar)) {
 
       // Range check for counted loops
       const Node*    cmp    = bol->in(1)->as_Cmp();

src/share/vm/opto/macro.cpp

@@ -666,7 +666,7 @@ bool PhaseMacroExpand::can_eliminate_allocation(AllocateNode *alloc, GrowableArr
         alloc->dump();
       else
         res->dump();
-    } else {
+    } else if (alloc->_is_scalar_replaceable) {
       tty->print("NotScalar (%s)", fail_eliminate);
       if (res == NULL)
         alloc->dump();
@@ -845,18 +845,14 @@ bool PhaseMacroExpand::scalar_replacement(AllocateNode *alloc, GrowableArray <Sa
     // to the allocated object with "sobj"
     int start = jvms->debug_start();
     int end   = jvms->debug_end();
-    for (int i = start; i < end; i++) {
-      if (sfpt->in(i) == res) {
-        sfpt->set_req(i, sobj);
-      }
-    }
+    sfpt->replace_edges_in_range(res, sobj, start, end);
     safepoints_done.append_if_missing(sfpt); // keep it for rollback
   }
   return true;
 }
 
 // Process users of eliminated allocation.
-void PhaseMacroExpand::process_users_of_allocation(AllocateNode *alloc) {
+void PhaseMacroExpand::process_users_of_allocation(CallNode *alloc) {
   Node* res = alloc->result_cast();
   if (res != NULL) {
     for (DUIterator_Last jmin, j = res->last_outs(jmin); j >= jmin; ) {
@@ -899,6 +895,17 @@ void PhaseMacroExpand::process_users_of_allocation(AllocateNode *alloc) {
   // Process other users of allocation's projections
   //
   if (_resproj != NULL && _resproj->outcnt() != 0) {
+    // First disconnect stores captured by Initialize node.
+    // If Initialize node is eliminated first in the following code,
+    // it will kill such stores and DUIterator_Last will assert.
+    for (DUIterator_Fast jmax, j = _resproj->fast_outs(jmax);  j < jmax; j++) {
+      Node *use = _resproj->fast_out(j);
+      if (use->is_AddP()) {
+        // raw memory addresses used only by the initialization
+        _igvn.replace_node(use, C->top());
+        --j; --jmax;
+      }
+    }
     for (DUIterator_Last jmin, j = _resproj->last_outs(jmin); j >= jmin; ) {
       Node *use = _resproj->last_out(j);
       uint oc1 = _resproj->outcnt();
@@ -923,9 +930,6 @@ void PhaseMacroExpand::process_users_of_allocation(AllocateNode *alloc) {
 #endif
           _igvn.replace_node(mem_proj, mem);
         }
-      } else if (use->is_AddP()) {
-        // raw memory addresses used only by the initialization
-        _igvn.replace_node(use, C->top());
       } else  {
         assert(false, "only Initialize or AddP expected");
       }
@@ -953,8 +957,18 @@ void PhaseMacroExpand::process_users_of_allocation(AllocateNode *alloc) {
 }
 
 bool PhaseMacroExpand::eliminate_allocate_node(AllocateNode *alloc) {
-
-  if (!EliminateAllocations || !alloc->_is_scalar_replaceable) {
+  if (!EliminateAllocations || !alloc->_is_non_escaping) {
+    return false;
+  }
+  Node* klass = alloc->in(AllocateNode::KlassNode);
+  const TypeKlassPtr* tklass = _igvn.type(klass)->is_klassptr();
+  Node* res = alloc->result_cast();
+  // Eliminate boxing allocations which are not used
+  // regardless scalar replacable status.
+  bool boxing_alloc = C->eliminate_boxing() &&
+                      tklass->klass()->is_instance_klass()  &&
+                      tklass->klass()->as_instance_klass()->is_box_klass();
+  if (!alloc->_is_scalar_replaceable && (!boxing_alloc || (res != NULL))) {
     return false;
   }
 
@@ -965,14 +979,22 @@ bool PhaseMacroExpand::eliminate_allocate_node(AllocateNode *alloc) {
     return false;
   }
 
+  if (!alloc->_is_scalar_replaceable) {
+    assert(res == NULL, "sanity");
+    // We can only eliminate allocation if all debug info references
+    // are already replaced with SafePointScalarObject because
+    // we can't search for a fields value without instance_id.
+    if (safepoints.length() > 0) {
+      return false;
+    }
+  }
+
   if (!scalar_replacement(alloc, safepoints)) {
     return false;
   }
 
   CompileLog* log = C->log();
   if (log != NULL) {
-    Node* klass = alloc->in(AllocateNode::KlassNode);
-    const TypeKlassPtr* tklass = _igvn.type(klass)->is_klassptr();
     log->head("eliminate_allocation type='%d'",
               log->identify(tklass->klass()));
     JVMState* p = alloc->jvms();
@@ -997,6 +1019,43 @@ bool PhaseMacroExpand::eliminate_allocate_node(AllocateNode *alloc) {
   return true;
 }
 
+bool PhaseMacroExpand::eliminate_boxing_node(CallStaticJavaNode *boxing) {
+  // EA should remove all uses of non-escaping boxing node.
+  if (!C->eliminate_boxing() || boxing->proj_out(TypeFunc::Parms) != NULL) {
+    return false;
+  }
+
+  extract_call_projections(boxing);
+
+  const TypeTuple* r = boxing->tf()->range();
+  assert(r->cnt() > TypeFunc::Parms, "sanity");
+  const TypeInstPtr* t = r->field_at(TypeFunc::Parms)->isa_instptr();
+  assert(t != NULL, "sanity");
+
+  CompileLog* log = C->log();
+  if (log != NULL) {
+    log->head("eliminate_boxing type='%d'",
+              log->identify(t->klass()));
+    JVMState* p = boxing->jvms();
+    while (p != NULL) {
+      log->elem("jvms bci='%d' method='%d'", p->bci(), log->identify(p->method()));
+      p = p->caller();
+    }
+    log->tail("eliminate_boxing");
+  }
+
+  process_users_of_allocation(boxing);
+
+#ifndef PRODUCT
+  if (PrintEliminateAllocations) {
+    tty->print("++++ Eliminated: %d ", boxing->_idx);
+    boxing->method()->print_short_name(tty);
+    tty->cr();
+  }
+#endif
+
+  return true;
+}
 
 //---------------------------set_eden_pointers-------------------------
 void PhaseMacroExpand::set_eden_pointers(Node* &eden_top_adr, Node* &eden_end_adr) {
@@ -2384,6 +2443,9 @@ void PhaseMacroExpand::eliminate_macro_nodes() {
       case Node::Class_AllocateArray:
         success = eliminate_allocate_node(n->as_Allocate());
         break;
+      case Node::Class_CallStaticJava:
+        success = eliminate_boxing_node(n->as_CallStaticJava());
+        break;
       case Node::Class_Lock:
       case Node::Class_Unlock:
         assert(!n->as_AbstractLock()->is_eliminated(), "sanity");
@@ -2424,6 +2486,11 @@ bool PhaseMacroExpand::expand_macro_nodes() {
         C->remove_macro_node(n);
         _igvn._worklist.push(n);
         success = true;
+      } else if (n->Opcode() == Op_CallStaticJava) {
+        // Remove it from macro list and put on IGVN worklist to optimize.
+        C->remove_macro_node(n);
+        _igvn._worklist.push(n);
+        success = true;
       } else if (n->Opcode() == Op_Opaque1 || n->Opcode() == Op_Opaque2) {
         _igvn.replace_node(n, n->in(1));
         success = true;

src/share/vm/opto/macro.hpp

@@ -86,10 +86,11 @@ private:
   Node *value_from_mem(Node *mem, BasicType ft, const Type *ftype, const TypeOopPtr *adr_t, Node *alloc);
   Node *value_from_mem_phi(Node *mem, BasicType ft, const Type *ftype, const TypeOopPtr *adr_t, Node *alloc, Node_Stack *value_phis, int level);
 
+  bool eliminate_boxing_node(CallStaticJavaNode *boxing);
   bool eliminate_allocate_node(AllocateNode *alloc);
   bool can_eliminate_allocation(AllocateNode *alloc, GrowableArray <SafePointNode *>& safepoints);
   bool scalar_replacement(AllocateNode *alloc, GrowableArray <SafePointNode *>& safepoints_done);
-  void process_users_of_allocation(AllocateNode *alloc);
+  void process_users_of_allocation(CallNode *alloc);
 
   void eliminate_card_mark(Node *cm);
   void mark_eliminated_box(Node* box, Node* obj);

src/share/vm/opto/memnode.hpp

@@ -75,8 +75,8 @@ public:
                                       PhaseTransform* phase);
   static bool adr_phi_is_loop_invariant(Node* adr_phi, Node* cast);
 
-  static Node *optimize_simple_memory_chain(Node *mchain, const TypePtr *t_adr, PhaseGVN *phase);
-  static Node *optimize_memory_chain(Node *mchain, const TypePtr *t_adr, PhaseGVN *phase);
+  static Node *optimize_simple_memory_chain(Node *mchain, const TypeOopPtr *t_oop, Node *load, PhaseGVN *phase);
+  static Node *optimize_memory_chain(Node *mchain, const TypePtr *t_adr, Node *load, PhaseGVN *phase);
   // This one should probably be a phase-specific function:
   static bool all_controls_dominate(Node* dom, Node* sub);
 
@@ -1099,7 +1099,7 @@ public:
 
   Node* make_raw_address(intptr_t offset, PhaseTransform* phase);
 
-  bool detect_init_independence(Node* n, bool st_is_pinned, int& count);
+  bool detect_init_independence(Node* n, int& count);
 
   void coalesce_subword_stores(intptr_t header_size, Node* size_in_bytes,
                                PhaseGVN* phase);

src/share/vm/opto/multnode.cpp

@@ -23,6 +23,7 @@
  */
 
 #include "precompiled.hpp"
+#include "opto/callnode.hpp"
 #include "opto/matcher.hpp"
 #include "opto/multnode.hpp"
 #include "opto/opcodes.hpp"
@@ -73,13 +74,26 @@ bool ProjNode::is_CFG() const {
   return (_con == TypeFunc::Control && def->is_CFG());
 }
 
+const Type* ProjNode::proj_type(const Type* t) const {
+  if (t == Type::TOP) {
+    return Type::TOP;
+  }
+  if (t == Type::BOTTOM) {
+    return Type::BOTTOM;
+  }
+  t = t->is_tuple()->field_at(_con);
+  Node* n = in(0);
+  if ((_con == TypeFunc::Parms) &&
+      n->is_CallStaticJava() && n->as_CallStaticJava()->is_boxing_method()) {
+    // The result of autoboxing is always non-null on normal path.
+    t = t->join(TypePtr::NOTNULL);
+  }
+  return t;
+}
+
 const Type *ProjNode::bottom_type() const {
-  if (in(0) == NULL)  return Type::TOP;
-  const Type *tb = in(0)->bottom_type();
-  if( tb == Type::TOP ) return Type::TOP;
-  if( tb == Type::BOTTOM ) return Type::BOTTOM;
-  const TypeTuple *t = tb->is_tuple();
-  return t->field_at(_con);
+  if (in(0) == NULL) return Type::TOP;
+  return proj_type(in(0)->bottom_type());
 }
 
 const TypePtr *ProjNode::adr_type() const {
@@ -115,11 +129,8 @@ void ProjNode::check_con() const {
 
 //------------------------------Value------------------------------------------
 const Type *ProjNode::Value( PhaseTransform *phase ) const {
-  if( !in(0) ) return Type::TOP;
-  const Type *t = phase->type(in(0));
-  if( t == Type::TOP ) return t;
-  if( t == Type::BOTTOM ) return t;
-  return t->is_tuple()->field_at(_con);
+  if (in(0) == NULL) return Type::TOP;
+  return proj_type(phase->type(in(0)));
 }
 
 //------------------------------out_RegMask------------------------------------

src/share/vm/opto/multnode.hpp

@@ -60,6 +60,7 @@ protected:
   virtual uint cmp( const Node &n ) const;
   virtual uint size_of() const;
   void check_con() const;       // Called from constructor.
+  const Type* proj_type(const Type* t) const;
 
 public:
   ProjNode( Node *src, uint con, bool io_use = false )
@@ -83,6 +84,7 @@ public:
   virtual const Type *Value( PhaseTransform *phase ) const;
   virtual uint ideal_reg() const;
   virtual const RegMask &out_RegMask() const;
+
 #ifndef PRODUCT
   virtual void dump_spec(outputStream *st) const;
 #endif

src/share/vm/opto/node.cpp

@@ -67,7 +67,8 @@ void Node::verify_construction() {
   }
   Compile::set_debug_idx(new_debug_idx);
   set_debug_idx( new_debug_idx );
-  assert(Compile::current()->unique() < (UINT_MAX - 1), "Node limit exceeded UINT_MAX");
+  assert(Compile::current()->unique() < (INT_MAX - 1), "Node limit exceeded INT_MAX");
+  assert(Compile::current()->live_nodes() < (uint)MaxNodeLimit, "Live Node limit exceeded limit");
   if (BreakAtNode != 0 && (_debug_idx == BreakAtNode || (int)_idx == BreakAtNode)) {
     tty->print_cr("BreakAtNode: _idx=%d _debug_idx=%d", _idx, _debug_idx);
     BREAKPOINT;
@@ -471,9 +472,9 @@ Node::Node(Node *n0, Node *n1, Node *n2, Node *n3,
 //------------------------------clone------------------------------------------
 // Clone a Node.
 Node *Node::clone() const {
-  Compile *compile = Compile::current();
+  Compile* C = Compile::current();
   uint s = size_of();           // Size of inherited Node
-  Node *n = (Node*)compile->node_arena()->Amalloc_D(size_of() + _max*sizeof(Node*));
+  Node *n = (Node*)C->node_arena()->Amalloc_D(size_of() + _max*sizeof(Node*));
   Copy::conjoint_words_to_lower((HeapWord*)this, (HeapWord*)n, s);
   // Set the new input pointer array
   n->_in = (Node**)(((char*)n)+s);
@@ -492,18 +493,18 @@ Node *Node::clone() const {
     if (x != NULL) x->add_out(n);
   }
   if (is_macro())
-    compile->add_macro_node(n);
+    C->add_macro_node(n);
   if (is_expensive())
-    compile->add_expensive_node(n);
+    C->add_expensive_node(n);
 
-  n->set_idx(compile->next_unique()); // Get new unique index as well
+  n->set_idx(C->next_unique()); // Get new unique index as well
   debug_only( n->verify_construction() );
   NOT_PRODUCT(nodes_created++);
   // Do not patch over the debug_idx of a clone, because it makes it
   // impossible to break on the clone's moment of creation.
   //debug_only( n->set_debug_idx( debug_idx() ) );
 
-  compile->copy_node_notes_to(n, (Node*) this);
+  C->copy_node_notes_to(n, (Node*) this);
 
   // MachNode clone
   uint nopnds;
@@ -518,13 +519,12 @@ Node *Node::clone() const {
                                   (const void*)(&mthis->_opnds), 1));
     mach->_opnds = to;
     for ( uint i = 0; i < nopnds; ++i ) {
-      to[i] = from[i]->clone(compile);
+      to[i] = from[i]->clone(C);
     }
   }
   // cloning CallNode may need to clone JVMState
   if (n->is_Call()) {
-    CallNode *call = n->as_Call();
-    call->clone_jvms();
+    n->as_Call()->clone_jvms(C);
   }
   return n;                     // Return the clone
 }
@@ -811,6 +811,21 @@ int Node::replace_edge(Node* old, Node* neww) {
   return nrep;
 }
 
+/**
+ * Replace input edges in the range pointing to 'old' node.
+ */
+int Node::replace_edges_in_range(Node* old, Node* neww, int start, int end) {
+  if (old == neww)  return 0;  // nothing to do
+  uint nrep = 0;
+  for (int i = start; i < end; i++) {
+    if (in(i) == old) {
+      set_req(i, neww);
+      nrep++;
+    }
+  }
+  return nrep;
+}
+
 //-------------------------disconnect_inputs-----------------------------------
 // NULL out all inputs to eliminate incoming Def-Use edges.
 // Return the number of edges between 'n' and 'this'

src/share/vm/opto/node.hpp

@@ -410,6 +410,7 @@ protected:
   // Find first occurrence of n among my edges:
   int find_edge(Node* n);
   int replace_edge(Node* old, Node* neww);
+  int replace_edges_in_range(Node* old, Node* neww, int start, int end);
   // NULL out all inputs to eliminate incoming Def-Use edges.
   // Return the number of edges between 'n' and 'this'
   int  disconnect_inputs(Node *n, Compile *c);

src/share/vm/opto/parse.hpp

@@ -330,6 +330,7 @@ class Parse : public GraphKit {
   bool          _wrote_final;   // Did we write a final field?
   bool          _count_invocations; // update and test invocation counter
   bool          _method_data_update; // update method data oop
+  Node*         _alloc_with_final;   // An allocation node with final field
 
   // Variables which track Java semantics during bytecode parsing:
 
@@ -370,6 +371,11 @@ class Parse : public GraphKit {
   void      set_wrote_final(bool z)   { _wrote_final = z; }
   bool          count_invocations() const  { return _count_invocations; }
   bool          method_data_update() const { return _method_data_update; }
+  Node*    alloc_with_final() const   { return _alloc_with_final; }
+  void set_alloc_with_final(Node* n)  {
+    assert((_alloc_with_final == NULL) || (_alloc_with_final == n), "different init objects?");
+    _alloc_with_final = n;
+  }
 
   Block*             block()    const { return _block; }
   ciBytecodeStream&  iter()           { return _iter; }
@@ -512,7 +518,7 @@ class Parse : public GraphKit {
 
   // loading from a constant field or the constant pool
   // returns false if push failed (non-perm field constants only, not ldcs)
-  bool push_constant(ciConstant con, bool require_constant = false);
+  bool push_constant(ciConstant con, bool require_constant = false, bool is_autobox_cache = false);
 
   // implementation of object creation bytecodes
   void emit_guard_for_new(ciInstanceKlass* klass);

src/share/vm/opto/parse1.cpp

@@ -390,6 +390,7 @@ Parse::Parse(JVMState* caller, ciMethod* parse_method, float expected_uses)
   _expected_uses = expected_uses;
   _depth = 1 + (caller->has_method() ? caller->depth() : 0);
   _wrote_final = false;
+  _alloc_with_final = NULL;
   _entry_bci = InvocationEntryBci;
   _tf = NULL;
   _block = NULL;
@@ -723,6 +724,8 @@ void Parse::build_exits() {
   // Note:  iophi and memphi are not transformed until do_exits.
   Node* iophi  = new (C) PhiNode(region, Type::ABIO);
   Node* memphi = new (C) PhiNode(region, Type::MEMORY, TypePtr::BOTTOM);
+  gvn().set_type_bottom(iophi);
+  gvn().set_type_bottom(memphi);
   _exits.set_i_o(iophi);
   _exits.set_all_memory(memphi);
 
@@ -738,6 +741,7 @@ void Parse::build_exits() {
     }
     int         ret_size = type2size[ret_type->basic_type()];
     Node*       ret_phi  = new (C) PhiNode(region, ret_type);
+    gvn().set_type_bottom(ret_phi);
     _exits.ensure_stack(ret_size);
     assert((int)(tf()->range()->cnt() - TypeFunc::Parms) == ret_size, "good tf range");
     assert(method()->return_type()->size() == ret_size, "tf agrees w/ method");
@@ -917,7 +921,7 @@ void Parse::do_exits() {
     // such unusual early publications.  But no barrier is needed on
     // exceptional returns, since they cannot publish normally.
     //
-    _exits.insert_mem_bar(Op_MemBarRelease);
+    _exits.insert_mem_bar(Op_MemBarRelease, alloc_with_final());
 #ifndef PRODUCT
     if (PrintOpto && (Verbose || WizardMode)) {
       method()->print_name();

src/share/vm/opto/parse2.cpp

@@ -987,7 +987,7 @@ void Parse::do_ifnull(BoolTest::mask btest, Node *c) {
     uncommon_trap(Deoptimization::Reason_unreached,
                   Deoptimization::Action_reinterpret,
                   NULL, "cold");
-    if (EliminateAutoBox) {
+    if (C->eliminate_boxing()) {
       // Mark the successor blocks as parsed
       branch_block->next_path_num();
       next_block->next_path_num();
@@ -1012,7 +1012,7 @@ void Parse::do_ifnull(BoolTest::mask btest, Node *c) {
 
     if (stopped()) {            // Path is dead?
       explicit_null_checks_elided++;
-      if (EliminateAutoBox) {
+      if (C->eliminate_boxing()) {
         // Mark the successor block as parsed
         branch_block->next_path_num();
       }
@@ -1032,7 +1032,7 @@ void Parse::do_ifnull(BoolTest::mask btest, Node *c) {
 
   if (stopped()) {              // Path is dead?
     explicit_null_checks_elided++;
-    if (EliminateAutoBox) {
+    if (C->eliminate_boxing()) {
       // Mark the successor block as parsed
       next_block->next_path_num();
     }
@@ -1069,7 +1069,7 @@ void Parse::do_if(BoolTest::mask btest, Node* c) {
     uncommon_trap(Deoptimization::Reason_unreached,
                   Deoptimization::Action_reinterpret,
                   NULL, "cold");
-    if (EliminateAutoBox) {
+    if (C->eliminate_boxing()) {
       // Mark the successor blocks as parsed
       branch_block->next_path_num();
       next_block->next_path_num();
@@ -1135,7 +1135,7 @@ void Parse::do_if(BoolTest::mask btest, Node* c) {
     set_control(taken_branch);
 
     if (stopped()) {
-      if (EliminateAutoBox) {
+      if (C->eliminate_boxing()) {
         // Mark the successor block as parsed
         branch_block->next_path_num();
       }
@@ -1154,7 +1154,7 @@ void Parse::do_if(BoolTest::mask btest, Node* c) {
 
   // Branch not taken.
   if (stopped()) {
-    if (EliminateAutoBox) {
+    if (C->eliminate_boxing()) {
       // Mark the successor block as parsed
       next_block->next_path_num();
     }

src/share/vm/opto/parse3.cpp

@@ -150,6 +150,23 @@ void Parse::do_get_xxx(Node* obj, ciField* field, bool is_field) {
     // final field
     if (field->is_static()) {
       // final static field
+      if (C->eliminate_boxing()) {
+        // The pointers in the autobox arrays are always non-null.
+        ciSymbol* klass_name = field->holder()->name();
+        if (field->name() == ciSymbol::cache_field_name() &&
+            field->holder()->uses_default_loader() &&
+            (klass_name == ciSymbol::java_lang_Character_CharacterCache() ||
+             klass_name == ciSymbol::java_lang_Byte_ByteCache() ||
+             klass_name == ciSymbol::java_lang_Short_ShortCache() ||
+             klass_name == ciSymbol::java_lang_Integer_IntegerCache() ||
+             klass_name == ciSymbol::java_lang_Long_LongCache())) {
+          bool require_const = true;
+          bool autobox_cache = true;
+          if (push_constant(field->constant_value(), require_const, autobox_cache)) {
+            return;
+          }
+        }
+      }
       if (push_constant(field->constant_value()))
         return;
     }
@@ -304,11 +321,18 @@ void Parse::do_put_xxx(Node* obj, ciField* field, bool is_field) {
   // out of the constructor.
   if (is_field && field->is_final()) {
     set_wrote_final(true);
+    // Preserve allocation ptr to create precedent edge to it in membar
+    // generated on exit from constructor.
+    if (C->eliminate_boxing() &&
+        adr_type->isa_oopptr() && adr_type->is_oopptr()->is_ptr_to_boxed_value() &&
+        AllocateNode::Ideal_allocation(obj, &_gvn) != NULL) {
+      set_alloc_with_final(obj);
+    }
   }
 }
 
 
-bool Parse::push_constant(ciConstant constant, bool require_constant) {
+bool Parse::push_constant(ciConstant constant, bool require_constant, bool is_autobox_cache) {
   switch (constant.basic_type()) {
   case T_BOOLEAN:  push( intcon(constant.as_boolean()) ); break;
   case T_INT:      push( intcon(constant.as_int())     ); break;
@@ -329,7 +353,7 @@ bool Parse::push_constant(ciConstant constant, bool require_constant) {
       push( zerocon(T_OBJECT) );
       break;
     } else if (require_constant || oop_constant->should_be_constant()) {
-      push( makecon(TypeOopPtr::make_from_constant(oop_constant, require_constant)) );
+      push( makecon(TypeOopPtr::make_from_constant(oop_constant, require_constant, is_autobox_cache)) );
       break;
     } else {
       // we cannot inline the oop, but we can use it later to narrow a type

src/share/vm/opto/parseHelper.cpp

@@ -284,6 +284,11 @@ void Parse::do_new() {
        klass == C->env()->StringBuffer_klass())) {
     C->set_has_stringbuilder(true);
   }
+
+  // Keep track of boxed values for EliminateAutoBox optimizations.
+  if (C->eliminate_boxing() && klass->is_box_klass()) {
+    C->set_has_boxed_value(true);
+  }
 }
 
 #ifndef PRODUCT

src/share/vm/opto/phase.cpp

@@ -64,6 +64,7 @@ elapsedTimer Phase::_t_idealLoopVerify;
 // Subtimers for _t_optimizer
 elapsedTimer   Phase::_t_iterGVN;
 elapsedTimer   Phase::_t_iterGVN2;
+elapsedTimer   Phase::_t_incrInline;
 
 // Subtimers for _t_registerAllocation
 elapsedTimer   Phase::_t_ctorChaitin;
@@ -110,6 +111,7 @@ void Phase::print_timers() {
       tty->print_cr ("      macroEliminate : %3.3f sec", Phase::_t_macroEliminate.seconds());
     }
     tty->print_cr ("      iterGVN        : %3.3f sec", Phase::_t_iterGVN.seconds());
+    tty->print_cr ("      incrInline     : %3.3f sec", Phase::_t_incrInline.seconds());
     tty->print_cr ("      idealLoop      : %3.3f sec", Phase::_t_idealLoop.seconds());
     tty->print_cr ("      idealLoopVerify: %3.3f sec", Phase::_t_idealLoopVerify.seconds());
     tty->print_cr ("      ccp            : %3.3f sec", Phase::_t_ccp.seconds());

src/share/vm/opto/phase.hpp

@@ -100,6 +100,7 @@ protected:
 // Subtimers for _t_optimizer
   static elapsedTimer   _t_iterGVN;
   static elapsedTimer   _t_iterGVN2;
+  static elapsedTimer   _t_incrInline;
 
 // Subtimers for _t_registerAllocation
   static elapsedTimer   _t_ctorChaitin;

src/share/vm/opto/phaseX.cpp

@@ -882,7 +882,7 @@ void PhaseIterGVN::optimize() {
       return;
     }
     Node *n  = _worklist.pop();
-    if (++loop_count >= K * C->unique()) {
+    if (++loop_count >= K * C->live_nodes()) {
       debug_only(n->dump(4);)
       assert(false, "infinite loop in PhaseIterGVN::optimize");
       C->record_method_not_compilable("infinite loop in PhaseIterGVN::optimize");

src/share/vm/opto/type.cpp

@@ -2372,7 +2372,12 @@ TypeOopPtr::TypeOopPtr( TYPES t, PTR ptr, ciKlass* k, bool xk, ciObject* o, int
     _klass_is_exact(xk),
     _is_ptr_to_narrowoop(false),
     _is_ptr_to_narrowklass(false),
+    _is_ptr_to_boxed_value(false),
     _instance_id(instance_id) {
+  if (Compile::current()->eliminate_boxing() && (t == InstPtr) &&
+      (offset > 0) && xk && (k != 0) && k->is_instance_klass()) {
+    _is_ptr_to_boxed_value = k->as_instance_klass()->is_boxed_value_offset(offset);
+  }
 #ifdef _LP64
   if (_offset != 0) {
     if (_offset == oopDesc::klass_offset_in_bytes()) {
@@ -2613,44 +2618,50 @@ const TypeOopPtr* TypeOopPtr::make_from_klass_common(ciKlass *klass, bool klass_
 
 //------------------------------make_from_constant-----------------------------
 // Make a java pointer from an oop constant
-const TypeOopPtr* TypeOopPtr::make_from_constant(ciObject* o, bool require_constant) {
-    assert(!o->is_null_object(), "null object not yet handled here.");
-    ciKlass* klass = o->klass();
-    if (klass->is_instance_klass()) {
-      // Element is an instance
-      if (require_constant) {
-        if (!o->can_be_constant())  return NULL;
-      } else if (!o->should_be_constant()) {
-        return TypeInstPtr::make(TypePtr::NotNull, klass, true, NULL, 0);
-      }
-      return TypeInstPtr::make(o);
-    } else if (klass->is_obj_array_klass()) {
-      // Element is an object array. Recursively call ourself.
-    const Type *etype =
+const TypeOopPtr* TypeOopPtr::make_from_constant(ciObject* o,
+                                                 bool require_constant,
+                                                 bool is_autobox_cache) {
+  assert(!o->is_null_object(), "null object not yet handled here.");
+  ciKlass* klass = o->klass();
+  if (klass->is_instance_klass()) {
+    // Element is an instance
+    if (require_constant) {
+      if (!o->can_be_constant())  return NULL;
+    } else if (!o->should_be_constant()) {
+      return TypeInstPtr::make(TypePtr::NotNull, klass, true, NULL, 0);
+    }
+    return TypeInstPtr::make(o);
+  } else if (klass->is_obj_array_klass()) {
+    // Element is an object array. Recursively call ourself.
+    const TypeOopPtr *etype =
       TypeOopPtr::make_from_klass_raw(klass->as_obj_array_klass()->element_klass());
-      const TypeAry* arr0 = TypeAry::make(etype, TypeInt::make(o->as_array()->length()));
-      // We used to pass NotNull in here, asserting that the sub-arrays
-      // are all not-null.  This is not true in generally, as code can
-      // slam NULLs down in the subarrays.
-      if (require_constant) {
-        if (!o->can_be_constant())  return NULL;
-      } else if (!o->should_be_constant()) {
-        return TypeAryPtr::make(TypePtr::NotNull, arr0, klass, true, 0);
-      }
-    const TypeAryPtr* arr = TypeAryPtr::make(TypePtr::Constant, o, arr0, klass, true, 0);
+    if (is_autobox_cache) {
+      // The pointers in the autobox arrays are always non-null.
+      etype = etype->cast_to_ptr_type(TypePtr::NotNull)->is_oopptr();
+    }
+    const TypeAry* arr0 = TypeAry::make(etype, TypeInt::make(o->as_array()->length()));
+    // We used to pass NotNull in here, asserting that the sub-arrays
+    // are all not-null.  This is not true in generally, as code can
+    // slam NULLs down in the subarrays.
+    if (require_constant) {
+      if (!o->can_be_constant())  return NULL;
+    } else if (!o->should_be_constant()) {
+      return TypeAryPtr::make(TypePtr::NotNull, arr0, klass, true, 0);
+    }
+    const TypeAryPtr* arr = TypeAryPtr::make(TypePtr::Constant, o, arr0, klass, true, 0, InstanceBot, is_autobox_cache);
     return arr;
-    } else if (klass->is_type_array_klass()) {
-      // Element is an typeArray
+  } else if (klass->is_type_array_klass()) {
+    // Element is an typeArray
     const Type* etype =
       (Type*)get_const_basic_type(klass->as_type_array_klass()->element_type());
-      const TypeAry* arr0 = TypeAry::make(etype, TypeInt::make(o->as_array()->length()));
-      // We used to pass NotNull in here, asserting that the array pointer
-      // is not-null. That was not true in general.
-      if (require_constant) {
-        if (!o->can_be_constant())  return NULL;
-      } else if (!o->should_be_constant()) {
-        return TypeAryPtr::make(TypePtr::NotNull, arr0, klass, true, 0);
-      }
+    const TypeAry* arr0 = TypeAry::make(etype, TypeInt::make(o->as_array()->length()));
+    // We used to pass NotNull in here, asserting that the array pointer
+    // is not-null. That was not true in general.
+    if (require_constant) {
+      if (!o->can_be_constant())  return NULL;
+    } else if (!o->should_be_constant()) {
+      return TypeAryPtr::make(TypePtr::NotNull, arr0, klass, true, 0);
+    }
     const TypeAryPtr* arr = TypeAryPtr::make(TypePtr::Constant, o, arr0, klass, true, 0);
     return arr;
   }
@@ -2856,6 +2867,28 @@ const TypeInstPtr *TypeInstPtr::make(PTR ptr,
   return result;
 }
 
+/**
+ *  Create constant type for a constant boxed value
+ */
+const Type* TypeInstPtr::get_const_boxed_value() const {
+  assert(is_ptr_to_boxed_value(), "should be called only for boxed value");
+  assert((const_oop() != NULL), "should be called only for constant object");
+  ciConstant constant = const_oop()->as_instance()->field_value_by_offset(offset());
+  BasicType bt = constant.basic_type();
+  switch (bt) {
+    case T_BOOLEAN:  return TypeInt::make(constant.as_boolean());
+    case T_INT:      return TypeInt::make(constant.as_int());
+    case T_CHAR:     return TypeInt::make(constant.as_char());
+    case T_BYTE:     return TypeInt::make(constant.as_byte());
+    case T_SHORT:    return TypeInt::make(constant.as_short());
+    case T_FLOAT:    return TypeF::make(constant.as_float());
+    case T_DOUBLE:   return TypeD::make(constant.as_double());
+    case T_LONG:     return TypeLong::make(constant.as_long());
+    default:         break;
+  }
+  fatal(err_msg_res("Invalid boxed value type '%s'", type2name(bt)));
+  return NULL;
+}
 
 //------------------------------cast_to_ptr_type-------------------------------
 const Type *TypeInstPtr::cast_to_ptr_type(PTR ptr) const {
@@ -3330,18 +3363,18 @@ const TypeAryPtr *TypeAryPtr::make( PTR ptr, const TypeAry *ary, ciKlass* k, boo
   if (!xk)  xk = ary->ary_must_be_exact();
   assert(instance_id <= 0 || xk || !UseExactTypes, "instances are always exactly typed");
   if (!UseExactTypes)  xk = (ptr == Constant);
-  return (TypeAryPtr*)(new TypeAryPtr(ptr, NULL, ary, k, xk, offset, instance_id))->hashcons();
+  return (TypeAryPtr*)(new TypeAryPtr(ptr, NULL, ary, k, xk, offset, instance_id, false))->hashcons();
 }
 
 //------------------------------make-------------------------------------------
-const TypeAryPtr *TypeAryPtr::make( PTR ptr, ciObject* o, const TypeAry *ary, ciKlass* k, bool xk, int offset, int instance_id ) {
+const TypeAryPtr *TypeAryPtr::make( PTR ptr, ciObject* o, const TypeAry *ary, ciKlass* k, bool xk, int offset, int instance_id, bool is_autobox_cache) {
   assert(!(k == NULL && ary->_elem->isa_int()),
          "integral arrays must be pre-equipped with a class");
   assert( (ptr==Constant && o) || (ptr!=Constant && !o), "" );
   if (!xk)  xk = (o != NULL) || ary->ary_must_be_exact();
   assert(instance_id <= 0 || xk || !UseExactTypes, "instances are always exactly typed");
   if (!UseExactTypes)  xk = (ptr == Constant);
-  return (TypeAryPtr*)(new TypeAryPtr(ptr, o, ary, k, xk, offset, instance_id))->hashcons();
+  return (TypeAryPtr*)(new TypeAryPtr(ptr, o, ary, k, xk, offset, instance_id, is_autobox_cache))->hashcons();
 }
 
 //------------------------------cast_to_ptr_type-------------------------------
@@ -3397,8 +3430,20 @@ const TypeInt* TypeAryPtr::narrow_size_type(const TypeInt* size) const {
   jint max_hi = max_array_length(elem()->basic_type());
   //if (index_not_size)  --max_hi;     // type of a valid array index, FTR
   bool chg = false;
-  if (lo < min_lo) { lo = min_lo; chg = true; }
-  if (hi > max_hi) { hi = max_hi; chg = true; }
+  if (lo < min_lo) {
+    lo = min_lo;
+    if (size->is_con()) {
+      hi = lo;
+    }
+    chg = true;
+  }
+  if (hi > max_hi) {
+    hi = max_hi;
+    if (size->is_con()) {
+      lo = hi;
+    }
+    chg = true;
+  }
   // Negative length arrays will produce weird intermediate dead fast-path code
   if (lo > hi)
     return TypeInt::ZERO;
@@ -3630,7 +3675,7 @@ const Type *TypeAryPtr::xmeet( const Type *t ) const {
 //------------------------------xdual------------------------------------------
 // Dual: compute field-by-field dual
 const Type *TypeAryPtr::xdual() const {
-  return new TypeAryPtr( dual_ptr(), _const_oop, _ary->dual()->is_ary(),_klass, _klass_is_exact, dual_offset(), dual_instance_id() );
+  return new TypeAryPtr( dual_ptr(), _const_oop, _ary->dual()->is_ary(),_klass, _klass_is_exact, dual_offset(), dual_instance_id(), is_autobox_cache() );
 }
 
 //----------------------interface_vs_oop---------------------------------------

src/share/vm/opto/type.hpp

@@ -234,6 +234,9 @@ public:
   bool is_ptr_to_narrowoop() const;
   bool is_ptr_to_narrowklass() const;
 
+  bool is_ptr_to_boxing_obj() const;
+
+
   // Convenience access
   float getf() const;
   double getd() const;
@@ -794,6 +797,7 @@ protected:
   bool          _klass_is_exact;
   bool          _is_ptr_to_narrowoop;
   bool          _is_ptr_to_narrowklass;
+  bool          _is_ptr_to_boxed_value;
 
   // If not InstanceTop or InstanceBot, indicates that this is
   // a particular instance of this type which is distinct.
@@ -826,7 +830,9 @@ public:
   // If the object cannot be rendered as a constant,
   // may return a non-singleton type.
   // If require_constant, produce a NULL if a singleton is not possible.
-  static const TypeOopPtr* make_from_constant(ciObject* o, bool require_constant = false);
+  static const TypeOopPtr* make_from_constant(ciObject* o,
+                                              bool require_constant = false,
+                                              bool not_null_elements = false);
 
   // Make a generic (unclassed) pointer to an oop.
   static const TypeOopPtr* make(PTR ptr, int offset, int instance_id);
@@ -839,7 +845,7 @@ public:
   // compressed oop references.
   bool is_ptr_to_narrowoop_nv() const { return _is_ptr_to_narrowoop; }
   bool is_ptr_to_narrowklass_nv() const { return _is_ptr_to_narrowklass; }
-
+  bool is_ptr_to_boxed_value()   const { return _is_ptr_to_boxed_value; }
   bool is_known_instance()       const { return _instance_id > 0; }
   int  instance_id()             const { return _instance_id; }
   bool is_known_instance_field() const { return is_known_instance() && _offset >= 0; }
@@ -912,6 +918,9 @@ class TypeInstPtr : public TypeOopPtr {
   // Make a pointer to an oop.
   static const TypeInstPtr *make(PTR ptr, ciKlass* k, bool xk, ciObject* o, int offset, int instance_id = InstanceBot );
 
+  /** Create constant type for a constant boxed value */
+  const Type* get_const_boxed_value() const;
+
   // If this is a java.lang.Class constant, return the type for it or NULL.
   // Pass to Type::get_const_type to turn it to a type, which will usually
   // be a TypeInstPtr, but may also be a TypeInt::INT for int.class, etc.
@@ -943,7 +952,12 @@ class TypeInstPtr : public TypeOopPtr {
 //------------------------------TypeAryPtr-------------------------------------
 // Class of Java array pointers
 class TypeAryPtr : public TypeOopPtr {
-  TypeAryPtr( PTR ptr, ciObject* o, const TypeAry *ary, ciKlass* k, bool xk, int offset, int instance_id ) : TypeOopPtr(AryPtr,ptr,k,xk,o,offset, instance_id), _ary(ary) {
+  TypeAryPtr( PTR ptr, ciObject* o, const TypeAry *ary, ciKlass* k, bool xk,
+              int offset, int instance_id, bool is_autobox_cache )
+  : TypeOopPtr(AryPtr,ptr,k,xk,o,offset, instance_id),
+    _ary(ary),
+    _is_autobox_cache(is_autobox_cache)
+ {
 #ifdef ASSERT
     if (k != NULL) {
       // Verify that specified klass and TypeAryPtr::klass() follow the same rules.
@@ -964,6 +978,7 @@ class TypeAryPtr : public TypeOopPtr {
   virtual bool eq( const Type *t ) const;
   virtual int hash() const;     // Type specific hashing
   const TypeAry *_ary;          // Array we point into
+  const bool     _is_autobox_cache;
 
   ciKlass* compute_klass(DEBUG_ONLY(bool verify = false)) const;
 
@@ -974,9 +989,11 @@ public:
   const Type*    elem() const { return _ary->_elem; }
   const TypeInt* size() const { return _ary->_size; }
 
+  bool is_autobox_cache() const { return _is_autobox_cache; }
+
   static const TypeAryPtr *make( PTR ptr, const TypeAry *ary, ciKlass* k, bool xk, int offset, int instance_id = InstanceBot);
   // Constant pointer to array
-  static const TypeAryPtr *make( PTR ptr, ciObject* o, const TypeAry *ary, ciKlass* k, bool xk, int offset, int instance_id = InstanceBot);
+  static const TypeAryPtr *make( PTR ptr, ciObject* o, const TypeAry *ary, ciKlass* k, bool xk, int offset, int instance_id = InstanceBot, bool is_autobox_cache = false);
 
   // Return a 'ptr' version of this type
   virtual const Type *cast_to_ptr_type(PTR ptr) const;
@@ -1504,6 +1521,13 @@ inline bool Type::is_floatingpoint() const {
   return false;
 }
 
+inline bool Type::is_ptr_to_boxing_obj() const {
+  const TypeInstPtr* tp = isa_instptr();
+  return (tp != NULL) && (tp->offset() == 0) &&
+         tp->klass()->is_instance_klass()  &&
+         tp->klass()->as_instance_klass()->is_box_klass();
+}
+
 
 // ===============================================================
 // Things that need to be 64-bits in the 64-bit build but

src/share/vm/runtime/arguments.cpp

@@ -1089,6 +1089,10 @@ void Arguments::set_tiered_flags() {
   if (FLAG_IS_DEFAULT(ReservedCodeCacheSize)) {
     FLAG_SET_DEFAULT(ReservedCodeCacheSize, ReservedCodeCacheSize * 5);
   }
+  if (!UseInterpreter) { // -Xcomp
+    Tier3InvokeNotifyFreqLog = 0;
+    Tier4InvocationThreshold = 0;
+  }
 }
 
 #if INCLUDE_ALL_GCS
@@ -1661,6 +1665,20 @@ void Arguments::set_bytecode_flags() {
 // Aggressive optimization flags  -XX:+AggressiveOpts
 void Arguments::set_aggressive_opts_flags() {
 #ifdef COMPILER2
+  if (AggressiveUnboxing) {
+    if (FLAG_IS_DEFAULT(EliminateAutoBox)) {
+      FLAG_SET_DEFAULT(EliminateAutoBox, true);
+    } else if (!EliminateAutoBox) {
+      // warning("AggressiveUnboxing is disabled because EliminateAutoBox is disabled");
+      AggressiveUnboxing = false;
+    }
+    if (FLAG_IS_DEFAULT(DoEscapeAnalysis)) {
+      FLAG_SET_DEFAULT(DoEscapeAnalysis, true);
+    } else if (!DoEscapeAnalysis) {
+      // warning("AggressiveUnboxing is disabled because DoEscapeAnalysis is disabled");
+      AggressiveUnboxing = false;
+    }
+  }
   if (AggressiveOpts || !FLAG_IS_DEFAULT(AutoBoxCacheMax)) {
     if (FLAG_IS_DEFAULT(EliminateAutoBox)) {
       FLAG_SET_DEFAULT(EliminateAutoBox, true);

src/share/vm/runtime/vmStructs.cpp

@@ -1057,6 +1057,7 @@ typedef BinaryTreeDictionary<Metablock, FreeList> MetablockTreeDictionary;
   c2_nonstatic_field(Compile,            _save_argument_registers, const bool)                                                       \
   c2_nonstatic_field(Compile,            _subsume_loads,           const bool)                                                       \
   c2_nonstatic_field(Compile,            _do_escape_analysis,      const bool)                                                       \
+  c2_nonstatic_field(Compile,            _eliminate_boxing,        const bool)                                                       \
   c2_nonstatic_field(Compile,            _ilt,                     InlineTree*)                                                      \
                                                                                                                                      \
   c2_nonstatic_field(InlineTree,         _caller_jvms,             JVMState*)                                                        \

test/compiler/6934604/TestByteBoxing.java

+/*
+ * Copyright (c) 2013, Oracle and/or its affiliates. All rights reserved.
+ * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
+ *
+ * This code is free software; you can redistribute it and/or modify it
+ * under the terms of the GNU General Public License version 2 only, as
+ * published by the Free Software Foundation.
+ *
+ * This code is distributed in the hope that it will be useful, but WITHOUT
+ * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
+ * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
+ * version 2 for more details (a copy is included in the LICENSE file that
+ * accompanied this code).
+ *
+ * You should have received a copy of the GNU General Public License version
+ * 2 along with this work; if not, write to the Free Software Foundation,
+ * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
+ *
+ * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
+ * or visit www.oracle.com if you need additional information or have any
+ * questions.
+ */
+
+/*
+ * @test
+ * @bug 6934604
+ * @summary enable parts of EliminateAutoBox by default
+ * @run main/othervm -Xbatch -XX:+IgnoreUnrecognizedVMOptions -XX:+EliminateAutoBox TestByteBoxing
+ * @run main/othervm -Xbatch -XX:+IgnoreUnrecognizedVMOptions -XX:+EliminateAutoBox
+ * -XX:CompileCommand=exclude,TestByteBoxing.dummy -XX:CompileCommand=exclude,TestByteBoxing.foo -XX:CompileCommand=exclude,TestByteBoxing.foob TestByteBoxing
+ * @run main/othervm -Xbatch -XX:+IgnoreUnrecognizedVMOptions -XX:-EliminateAutoBox
+ * -XX:CompileCommand=exclude,TestByteBoxing.dummy -XX:CompileCommand=exclude,TestByteBoxing.foo -XX:CompileCommand=exclude,TestByteBoxing.foob TestByteBoxing
+ *
+ */
+
+public class TestByteBoxing {
+
+  static final Byte ibc = new Byte((byte)1);
+
+  //===============================================
+  // Non-inlined methods to test deoptimization info
+  static void dummy()      { }
+  static byte foo(byte i)  { return i; }
+  static Byte foob(byte i) { return Byte.valueOf(i); }
+
+
+  static byte simple(byte i) {
+    Byte ib = new Byte(i);
+    return ib;
+  }
+
+  static byte simpleb(byte i) {
+    Byte ib = Byte.valueOf(i);
+    return ib;
+  }
+
+  static byte simplec() {
+    Byte ib = ibc;
+    return ib;
+  }
+
+  static byte simplef(byte i) {
+    Byte ib = foob(i);
+    return ib;
+  }
+
+  static byte simplep(Byte ib) {
+    return ib;
+  }
+
+  static byte simple2(byte i) {
+    Byte ib1 = new Byte(i);
+    Byte ib2 = new Byte((byte)(i+1));
+    return (byte)(ib1 + ib2);
+  }
+
+  static byte simpleb2(byte i) {
+    Byte ib1 = Byte.valueOf(i);
+    Byte ib2 = Byte.valueOf((byte)(i+1));
+    return (byte)(ib1 + ib2);
+  }
+
+  static byte simplem2(byte i) {
+    Byte ib1 = new Byte(i);
+    Byte ib2 = Byte.valueOf((byte)(i+1));
+    return (byte)(ib1 + ib2);
+  }
+
+  static byte simplep2(byte i, Byte ib1) {
+    Byte ib2 = Byte.valueOf((byte)(i+1));
+    return (byte)(ib1 + ib2);
+  }
+
+  static byte simplec2(byte i) {
+    Byte ib1 = ibc;
+    Byte ib2 = Byte.valueOf((byte)(i+1));
+    return (byte)(ib1 + ib2);
+  }
+
+  //===============================================
+  static byte test(byte i) {
+    Byte ib = new Byte(i);
+    if ((i&1) == 0)
+      ib = (byte)(i+1);
+    return ib;
+  }
+
+  static byte testb(byte i) {
+    Byte ib = i;
+    if ((i&1) == 0)
+      ib = (byte)(i+1);
+    return ib;
+  }
+
+  static byte testm(byte i) {
+    Byte ib = i;
+    if ((i&1) == 0)
+      ib = new Byte((byte)(i+1));
+    return ib;
+  }
+
+  static byte testp(byte i, Byte ib) {
+    if ((i&1) == 0)
+      ib = new Byte((byte)(i+1));
+    return ib;
+  }
+
+  static byte testc(byte i) {
+    Byte ib = ibc;
+    if ((i&1) == 0)
+      ib = new Byte((byte)(i+1));
+    return ib;
+  }
+
+  static byte test2(byte i) {
+    Byte ib1 = new Byte(i);
+    Byte ib2 = new Byte((byte)(i+1));
+    if ((i&1) == 0) {
+      ib1 = new Byte((byte)(i+1));
+      ib2 = new Byte((byte)(i+2));
+    }
+    return (byte)(ib1+ib2);
+  }
+
+  static byte testb2(byte i) {
+    Byte ib1 = i;
+    Byte ib2 = (byte)(i+1);
+    if ((i&1) == 0) {
+      ib1 = (byte)(i+1);
+      ib2 = (byte)(i+2);
+    }
+    return (byte)(ib1 + ib2);
+  }
+
+  static byte testm2(byte i) {
+    Byte ib1 = new Byte(i);
+    Byte ib2 = (byte)(i+1);
+    if ((i&1) == 0) {
+      ib1 = new Byte((byte)(i+1));
+      ib2 = (byte)(i+2);
+    }
+    return (byte)(ib1 + ib2);
+  }
+
+  static byte testp2(byte i, Byte ib1) {
+    Byte ib2 = (byte)(i+1);
+    if ((i&1) == 0) {
+      ib1 = new Byte((byte)(i+1));
+      ib2 = (byte)(i+2);
+    }
+    return (byte)(ib1 + ib2);
+  }
+
+  static byte testc2(byte i) {
+    Byte ib1 = ibc;
+    Byte ib2 = (byte)(i+1);
+    if ((i&1) == 0) {
+      ib1 = (byte)(ibc+1);
+      ib2 = (byte)(i+2);
+    }
+    return (byte)(ib1 + ib2);
+  }
+
+  //===============================================
+  static byte sum(byte[] a) {
+    byte result = 1;
+    for (Byte i : a)
+        result += i;
+    return result;
+  }
+
+  static byte sumb(byte[] a) {
+    Byte result = 1;
+    for (Byte i : a)
+        result = (byte)(result + i);
+    return result;
+  }
+
+  static byte sumc(byte[] a) {
+    Byte result = ibc;
+    for (Byte i : a)
+        result = (byte)(result + i);
+    return result;
+  }
+
+  static byte sumf(byte[] a) {
+    Byte result = foob((byte)1);
+    for (Byte i : a)
+        result = (byte)(result + i);
+    return result;
+  }
+
+  static byte sump(byte[] a, Byte result) {
+    for (Byte i : a)
+        result = (byte)(result + i);
+    return result;
+  }
+
+  static byte sum2(byte[] a) {
+    byte result1 = 1;
+    byte result2 = 1;
+    for (Byte i : a) {
+        result1 += i;
+        result2 += i + 1;
+    }
+    return (byte)(result1 + result2);
+  }
+
+  static byte sumb2(byte[] a) {
+    Byte result1 = 1;
+    Byte result2 = 1;
+    for (Byte i : a) {
+        result1 = (byte)(result1 + i);
+        result2 = (byte)(result2 + i + 1);
+    }
+    return (byte)(result1 + result2);
+  }
+
+  static byte summ2(byte[] a) {
+    Byte result1 = 1;
+    Byte result2 = new Byte((byte)1);
+    for (Byte i : a) {
+        result1 = (byte)(result1 + i);
+        result2 = (byte)(result2 + new Byte((byte)(i + 1)));
+    }
+    return (byte)(result1 + result2);
+  }
+
+  static byte sump2(byte[] a, Byte result2) {
+    Byte result1 = 1;
+    for (Byte i : a) {
+        result1 = (byte)(result1 + i);
+        result2 = (byte)(result2 + i + 1);
+    }
+    return (byte)(result1 + result2);
+  }
+
+  static byte sumc2(byte[] a) {
+    Byte result1 = 1;
+    Byte result2 = ibc;
+    for (Byte i : a) {
+        result1 = (byte)(result1 + i);
+        result2 = (byte)(result2 + i + ibc);
+    }
+    return (byte)(result1 + result2);
+  }
+
+  //===============================================
+  static byte remi_sum() {
+    Byte j = new Byte((byte)1);
+    for (int i = 0; i< 1000; i++) {
+      j = new Byte((byte)(j + 1));
+    }
+    return j;
+  }
+
+  static byte remi_sumb() {
+    Byte j = Byte.valueOf((byte)1);
+    for (int i = 0; i< 1000; i++) {
+      j = (byte)(j + 1);
+    }
+    return j;
+  }
+
+  static byte remi_sumf() {
+    Byte j = foob((byte)1);
+    for (int i = 0; i< 1000; i++) {
+      j = (byte)(j + 1);
+    }
+    return j;
+  }
+
+  static byte remi_sump(Byte j) {
+    for (int i = 0; i< 1000; i++) {
+      j = new Byte((byte)(j + 1));
+    }
+    return j;
+  }
+
+  static byte remi_sumc() {
+    Byte j = ibc;
+    for (int i = 0; i< 1000; i++) {
+      j = (byte)(j + ibc);
+    }
+    return j;
+  }
+
+  static byte remi_sum2() {
+    Byte j1 = new Byte((byte)1);
+    Byte j2 = new Byte((byte)1);
+    for (int i = 0; i< 1000; i++) {
+      j1 = new Byte((byte)(j1 + 1));
+      j2 = new Byte((byte)(j2 + 2));
+    }
+    return (byte)(j1 + j2);
+  }
+
+  static byte remi_sumb2() {
+    Byte j1 = Byte.valueOf((byte)1);
+    Byte j2 = Byte.valueOf((byte)1);
+    for (int i = 0; i< 1000; i++) {
+      j1 = (byte)(j1 + 1);
+      j2 = (byte)(j2 + 2);
+    }
+    return (byte)(j1 + j2);
+  }
+
+  static byte remi_summ2() {
+    Byte j1 = new Byte((byte)1);
+    Byte j2 = Byte.valueOf((byte)1);
+    for (int i = 0; i< 1000; i++) {
+      j1 = new Byte((byte)(j1 + 1));
+      j2 = (byte)(j2 + 2);
+    }
+    return (byte)(j1 + j2);
+  }
+
+  static byte remi_sump2(Byte j1) {
+    Byte j2 = Byte.valueOf((byte)1);
+    for (int i = 0; i< 1000; i++) {
+      j1 = new Byte((byte)(j1 + 1));
+      j2 = (byte)(j2 + 2);
+    }
+    return (byte)(j1 + j2);
+  }
+
+  static byte remi_sumc2() {
+    Byte j1 = ibc;
+    Byte j2 = Byte.valueOf((byte)1);
+    for (int i = 0; i< 1000; i++) {
+      j1 = (byte)(j1 + ibc);
+      j2 = (byte)(j2 + 2);
+    }
+    return (byte)(j1 + j2);
+  }
+
+
+  //===============================================
+  // Safepointa and debug info for deoptimization
+  static byte simple_deop(byte i) {
+    Byte ib = new Byte(foo(i));
+    dummy();
+    return ib;
+  }
+
+  static byte simpleb_deop(byte i) {
+    Byte ib = Byte.valueOf(foo(i));
+    dummy();
+    return ib;
+  }
+
+  static byte simplef_deop(byte i) {
+    Byte ib = foob(i);
+    dummy();
+    return ib;
+  }
+
+  static byte simplep_deop(Byte ib) {
+    dummy();
+    return ib;
+  }
+
+  static byte simplec_deop(byte i) {
+    Byte ib = ibc;
+    dummy();
+    return ib;
+  }
+
+  static byte test_deop(byte i) {
+    Byte ib = new Byte(foo(i));
+    if ((i&1) == 0)
+      ib = foo((byte)(i+1));
+    dummy();
+    return ib;
+  }
+
+  static byte testb_deop(byte i) {
+    Byte ib = foo(i);
+    if ((i&1) == 0)
+      ib = foo((byte)(i+1));
+    dummy();
+    return ib;
+  }
+
+  static byte testf_deop(byte i) {
+    Byte ib = foob(i);
+    if ((i&1) == 0)
+      ib = foo((byte)(i+1));
+    dummy();
+    return ib;
+  }
+
+  static byte testp_deop(byte i, Byte ib) {
+    if ((i&1) == 0)
+      ib = foo((byte)(i+1));
+    dummy();
+    return ib;
+  }
+
+  static byte testc_deop(byte i) {
+    Byte ib = ibc;
+    if ((i&1) == 0)
+      ib = foo((byte)(i+1));
+    dummy();
+    return ib;
+  }
+
+  static byte sum_deop(byte[] a) {
+    byte result = 1;
+    for (Byte i : a)
+        result += foo(i);
+    dummy();
+    return result;
+  }
+
+  static byte sumb_deop(byte[] a) {
+    Byte result = 1;
+    for (Byte i : a)
+        result = (byte)(result + foo(i));
+    dummy();
+    return result;
+  }
+
+  static byte sumf_deop(byte[] a) {
+    Byte result = 1;
+    for (Byte i : a)
+        result = (byte)(result + foob(i));
+    dummy();
+    return result;
+  }
+
+  static byte sump_deop(byte[] a, Byte result) {
+    for (Byte i : a)
+        result = (byte)(result + foob(i));
+    dummy();
+    return result;
+  }
+
+  static byte sumc_deop(byte[] a) {
+    Byte result = ibc;
+    for (Byte i : a)
+        result = (byte)(result + foo(i));
+    dummy();
+    return result;
+  }
+
+  static byte remi_sum_deop() {
+    Byte j = new Byte(foo((byte)1));
+    for (int i = 0; i< 1000; i++) {
+      j = new Byte(foo((byte)(j + 1)));
+    }
+    dummy();
+    return j;
+  }
+
+  static byte remi_sumb_deop() {
+    Byte j = Byte.valueOf(foo((byte)1));
+    for (int i = 0; i< 1000; i++) {
+      j = foo((byte)(j + 1));
+    }
+    dummy();
+    return j;
+  }
+
+  static byte remi_sumf_deop() {
+    Byte j = foob((byte)1);
+    for (int i = 0; i< 1000; i++) {
+      j = foo((byte)(j + 1));
+    }
+    dummy();
+    return j;
+  }
+
+  static byte remi_sump_deop(Byte j) {
+    for (int i = 0; i< 1000; i++) {
+      j = foo((byte)(j + 1));
+    }
+    dummy();
+    return j;
+  }
+
+  static byte remi_sumc_deop() {
+    Byte j = ibc;
+    for (int i = 0; i< 1000; i++) {
+      j = foo((byte)(j + 1));
+    }
+    dummy();
+    return j;
+  }
+
+  //===============================================
+  // Conditional increment
+  static byte remi_sum_cond() {
+    Byte j = new Byte((byte)1);
+    for (int i = 0; i< 1000; i++) {
+      if ((i&1) == 0) {
+        j = new Byte((byte)(j + 1));
+      }
+    }
+    return j;
+  }
+
+  static byte remi_sumb_cond() {
+    Byte j = Byte.valueOf((byte)1);
+    for (int i = 0; i< 1000; i++) {
+      if ((i&1) == 0) {
+        j = (byte)(j + 1);
+      }
+    }
+    return j;
+  }
+
+  static byte remi_sumf_cond() {
+    Byte j = foob((byte)1);
+    for (int i = 0; i< 1000; i++) {
+      if ((i&1) == 0) {
+        j = (byte)(j + 1);
+      }
+    }
+    return j;
+  }
+
+  static byte remi_sump_cond(Byte j) {
+    for (int i = 0; i< 1000; i++) {
+      if ((i&1) == 0) {
+        j = (byte)(j + 1);
+      }
+    }
+    return j;
+  }
+
+  static byte remi_sumc_cond() {
+    Byte j = ibc;
+    for (int i = 0; i< 1000; i++) {
+      if ((i&1) == 0) {
+        j = (byte)(j + ibc);
+      }
+    }
+    return j;
+  }
+
+  static byte remi_sum2_cond() {
+    Byte j1 = new Byte((byte)1);
+    Byte j2 = new Byte((byte)1);
+    for (int i = 0; i< 1000; i++) {
+      if ((i&1) == 0) {
+        j1 = new Byte((byte)(j1 + 1));
+      } else {
+        j2 = new Byte((byte)(j2 + 2));
+      }
+    }
+    return (byte)(j1 + j2);
+  }
+
+  static byte remi_sumb2_cond() {
+    Byte j1 = Byte.valueOf((byte)1);
+    Byte j2 = Byte.valueOf((byte)1);
+    for (int i = 0; i< 1000; i++) {
+      if ((i&1) == 0) {
+        j1 = (byte)(j1 + 1);
+      } else {
+        j2 = (byte)(j2 + 2);
+      }
+    }
+    return (byte)(j1 + j2);
+  }
+
+  static byte remi_summ2_cond() {
+    Byte j1 = new Byte((byte)1);
+    Byte j2 = Byte.valueOf((byte)1);
+    for (int i = 0; i< 1000; i++) {
+      if ((i&1) == 0) {
+        j1 = new Byte((byte)(j1 + 1));
+      } else {
+        j2 = (byte)(j2 + 2);
+      }
+    }
+    return (byte)(j1 + j2);
+  }
+
+  static byte remi_sump2_cond(Byte j1) {
+    Byte j2 = Byte.valueOf((byte)1);
+    for (int i = 0; i< 1000; i++) {
+      if ((i&1) == 0) {
+        j1 = new Byte((byte)(j1 + 1));
+      } else {
+        j2 = (byte)(j2 + 2);
+      }
+    }
+    return (byte)(j1 + j2);
+  }
+
+  static byte remi_sumc2_cond() {
+    Byte j1 = ibc;
+    Byte j2 = Byte.valueOf((byte)1);
+    for (int i = 0; i< 1000; i++) {
+      if ((i&1) == 0) {
+        j1 = (byte)(j1 + ibc);
+      } else {
+        j2 = (byte)(j2 + 2);
+      }
+    }
+    return (byte)(j1 + j2);
+  }
+
+
+  public static void main(String[] args) {
+    final int ntests = 70;
+
+    String[] test_name = new String[] {
+        "simple",      "simpleb",      "simplec",      "simplef",      "simplep",
+        "simple2",     "simpleb2",     "simplec2",     "simplem2",     "simplep2",
+        "simple_deop", "simpleb_deop", "simplec_deop", "simplef_deop", "simplep_deop",
+        "test",        "testb",        "testc",        "testm",        "testp",
+        "test2",       "testb2",       "testc2",       "testm2",       "testp2",
+        "test_deop",   "testb_deop",   "testc_deop",   "testf_deop",   "testp_deop",
+        "sum",         "sumb",         "sumc",         "sumf",         "sump",
+        "sum2",        "sumb2",        "sumc2",        "summ2",        "sump2",
+        "sum_deop",    "sumb_deop",    "sumc_deop",    "sumf_deop",    "sump_deop",
+        "remi_sum",       "remi_sumb",       "remi_sumc",       "remi_sumf",       "remi_sump",
+        "remi_sum2",      "remi_sumb2",      "remi_sumc2",      "remi_summ2",      "remi_sump2",
+        "remi_sum_deop",  "remi_sumb_deop",  "remi_sumc_deop",  "remi_sumf_deop",  "remi_sump_deop",
+        "remi_sum_cond",  "remi_sumb_cond",  "remi_sumc_cond",  "remi_sumf_cond",  "remi_sump_cond",
+        "remi_sum2_cond", "remi_sumb2_cond", "remi_sumc2_cond", "remi_summ2_cond", "remi_sump2_cond"
+    };
+
+    final int[] val = new int[] {
+      -5488, -5488, 12000, -5488, -5488,
+       1024,  1024, -5552,  1024,  1024,
+      -5488, -5488, 12000, -5488, -5488,
+        512,   512,  6256,   512,   512,
+      13024, 13024, -5584, 13024, 13024,
+        512,   512,  6256,   512,   512,
+         45,    45,    45,    45,    45,
+         66,    66,    66,    66,    66,
+         45,    45,    45,    45,    45,
+        -23,   -23,   -23,   -23,   -23,
+        -70,   -70,   -70,   -70,   -70,
+        -23,   -23,   -23,   -23,   -23,
+        -11,   -11,   -11,   -11,   -11,
+        -34,   -34,   -34,   -34,   -34
+    };
+
+    int[] res = new int[ntests];
+    for (int i = 0; i < ntests; i++) {
+      res[i] = 0;
+    }
+
+
+    for (int i = 0; i < 12000; i++) {
+      res[0] += simple((byte)i);
+      res[1] += simpleb((byte)i);
+      res[2] += simplec();
+      res[3] += simplef((byte)i);
+      res[4] += simplep((byte)i);
+
+      res[5] += simple2((byte)i);
+      res[6] += simpleb2((byte)i);
+      res[7] += simplec2((byte)i);
+      res[8] += simplem2((byte)i);
+      res[9] += simplep2((byte)i, (byte)i);
+
+      res[10] += simple_deop((byte)i);
+      res[11] += simpleb_deop((byte)i);
+      res[12] += simplec_deop((byte)i);
+      res[13] += simplef_deop((byte)i);
+      res[14] += simplep_deop((byte)i);
+
+      res[15] += test((byte)i);
+      res[16] += testb((byte)i);
+      res[17] += testc((byte)i);
+      res[18] += testm((byte)i);
+      res[19] += testp((byte)i, (byte)i);
+
+      res[20] += test2((byte)i);
+      res[21] += testb2((byte)i);
+      res[22] += testc2((byte)i);
+      res[23] += testm2((byte)i);
+      res[24] += testp2((byte)i, (byte)i);
+
+      res[25] += test_deop((byte)i);
+      res[26] += testb_deop((byte)i);
+      res[27] += testc_deop((byte)i);
+      res[28] += testf_deop((byte)i);
+      res[29] += testp_deop((byte)i, (byte)i);
+    }
+
+    byte[] ia = new byte[1000];
+    for (int i = 0; i < 1000; i++) {
+      ia[i] = (byte)i;
+    }
+
+    for (int i = 0; i < 100; i++) {
+      res[30] = sum(ia);
+      res[31] = sumb(ia);
+      res[32] = sumc(ia);
+      res[33] = sumf(ia);
+      res[34] = sump(ia, (byte)1);
+
+      res[35] = sum2(ia);
+      res[36] = sumb2(ia);
+      res[37] = sumc2(ia);
+      res[38] = summ2(ia);
+      res[39] = sump2(ia, (byte)1);
+
+      res[40] = sum_deop(ia);
+      res[41] = sumb_deop(ia);
+      res[42] = sumc_deop(ia);
+      res[43] = sumf_deop(ia);
+      res[44] = sump_deop(ia, (byte)1);
+
+      res[45] = remi_sum();
+      res[46] = remi_sumb();
+      res[47] = remi_sumc();
+      res[48] = remi_sumf();
+      res[49] = remi_sump((byte)1);
+
+      res[50] = remi_sum2();
+      res[51] = remi_sumb2();
+      res[52] = remi_sumc2();
+      res[53] = remi_summ2();
+      res[54] = remi_sump2((byte)1);
+
+      res[55] = remi_sum_deop();
+      res[56] = remi_sumb_deop();
+      res[57] = remi_sumc_deop();
+      res[58] = remi_sumf_deop();
+      res[59] = remi_sump_deop((byte)1);
+
+      res[60] = remi_sum_cond();
+      res[61] = remi_sumb_cond();
+      res[62] = remi_sumc_cond();
+      res[63] = remi_sumf_cond();
+      res[64] = remi_sump_cond((byte)1);
+
+      res[65] = remi_sum2_cond();
+      res[66] = remi_sumb2_cond();
+      res[67] = remi_sumc2_cond();
+      res[68] = remi_summ2_cond();
+      res[69] = remi_sump2_cond((byte)1);
+    }
+
+    int failed = 0;
+    for (int i = 0; i < ntests; i++) {
+      if (res[i] != val[i]) {
+        System.err.println(test_name[i] + ": " + res[i] + " != " + val[i]);
+        failed++;
+      }
+    }
+    if (failed > 0) {
+      System.err.println("Failed " + failed + " tests.");
+      throw new InternalError();
+    } else {
+      System.out.println("Passed.");
+    }
+  }
+}

test/compiler/6934604/TestDoubleBoxing.java

+/*
+ * Copyright (c) 2013, Oracle and/or its affiliates. All rights reserved.
+ * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
+ *
+ * This code is free software; you can redistribute it and/or modify it
+ * under the terms of the GNU General Public License version 2 only, as
+ * published by the Free Software Foundation.
+ *
+ * This code is distributed in the hope that it will be useful, but WITHOUT
+ * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
+ * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
+ * version 2 for more details (a copy is included in the LICENSE file that
+ * accompanied this code).
+ *
+ * You should have received a copy of the GNU General Public License version
+ * 2 along with this work; if not, write to the Free Software Foundation,
+ * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
+ *
+ * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
+ * or visit www.oracle.com if you need additional information or have any
+ * questions.
+ */
+
+/*
+ * @test
+ * @bug 6934604
+ * @summary enable parts of EliminateAutoBox by default
+ * @run main/othervm -Xbatch -XX:+IgnoreUnrecognizedVMOptions -XX:+EliminateAutoBox TestDoubleBoxing
+ * @run main/othervm -Xbatch -XX:+IgnoreUnrecognizedVMOptions -XX:+EliminateAutoBox
+ * -XX:CompileCommand=exclude,TestDoubleBoxing.dummy -XX:CompileCommand=exclude,TestDoubleBoxing.foo -XX:CompileCommand=exclude,TestDoubleBoxing.foob TestDoubleBoxing
+ * @run main/othervm -Xbatch -XX:+IgnoreUnrecognizedVMOptions -XX:-EliminateAutoBox
+ * -XX:CompileCommand=exclude,TestDoubleBoxing.dummy -XX:CompileCommand=exclude,TestDoubleBoxing.foo -XX:CompileCommand=exclude,TestDoubleBoxing.foob TestDoubleBoxing
+ *
+ */
+
+public class TestDoubleBoxing {
+
+  static final Double ibc = new Double(1.);
+
+  //===============================================
+  // Non-inlined methods to test deoptimization info
+  static void   dummy()        { }
+  static double foo(double i)  { return i; }
+  static Double foob(double i) { return Double.valueOf(i); }
+
+
+  static double simple(double i) {
+    Double ib = new Double(i);
+    return ib;
+  }
+
+  static double simpleb(double i) {
+    Double ib = Double.valueOf(i);
+    return ib;
+  }
+
+  static double simplec() {
+    Double ib = ibc;
+    return ib;
+  }
+
+  static double simplef(double i) {
+    Double ib = foob(i);
+    return ib;
+  }
+
+  static double simplep(Double ib) {
+    return ib;
+  }
+
+  static double simple2(double i) {
+    Double ib1 = new Double(i);
+    Double ib2 = new Double(i+1.);
+    return ib1 + ib2;
+  }
+
+  static double simpleb2(double i) {
+    Double ib1 = Double.valueOf(i);
+    Double ib2 = Double.valueOf(i+1.);
+    return ib1 + ib2;
+  }
+
+  static double simplem2(double i) {
+    Double ib1 = new Double(i);
+    Double ib2 = Double.valueOf(i+1.);
+    return ib1 + ib2;
+  }
+
+  static double simplep2(double i, Double ib1) {
+    Double ib2 = Double.valueOf(i+1.);
+    return ib1 + ib2;
+  }
+
+  static double simplec2(double i) {
+    Double ib1 = ibc;
+    Double ib2 = Double.valueOf(i+1.);
+    return ib1 + ib2;
+  }
+
+  //===============================================
+  static double test(double f, int i) {
+    Double ib = new Double(f);
+    if ((i&1) == 0)
+      ib = f+1.;
+    return ib;
+  }
+
+  static double testb(double f, int i) {
+    Double ib = f;
+    if ((i&1) == 0)
+      ib = (f+1.);
+    return ib;
+  }
+
+  static double testm(double f, int i) {
+    Double ib = f;
+    if ((i&1) == 0)
+      ib = new Double(f+1.);
+    return ib;
+  }
+
+  static double testp(double f, int i, Double ib) {
+    if ((i&1) == 0)
+      ib = new Double(f+1.);
+    return ib;
+  }
+
+  static double testc(double f, int i) {
+    Double ib = ibc;
+    if ((i&1) == 0)
+      ib = new Double(f+1.);
+    return ib;
+  }
+
+  static double test2(double f, int i) {
+    Double ib1 = new Double(f);
+    Double ib2 = new Double(f+1.);
+    if ((i&1) == 0) {
+      ib1 = new Double(f+1.);
+      ib2 = new Double(f+2.);
+    }
+    return ib1+ib2;
+  }
+
+  static double testb2(double f, int i) {
+    Double ib1 = f;
+    Double ib2 = f+1.;
+    if ((i&1) == 0) {
+      ib1 = (f+1.);
+      ib2 = (f+2.);
+    }
+    return ib1+ib2;
+  }
+
+  static double testm2(double f, int i) {
+    Double ib1 = new Double(f);
+    Double ib2 = f+1.;
+    if ((i&1) == 0) {
+      ib1 = new Double(f+1.);
+      ib2 = (f+2.);
+    }
+    return ib1+ib2;
+  }
+
+  static double testp2(double f, int i, Double ib1) {
+    Double ib2 = f+1.;
+    if ((i&1) == 0) {
+      ib1 = new Double(f+1.);
+      ib2 = (f+2.);
+    }
+    return ib1+ib2;
+  }
+
+  static double testc2(double f, int i) {
+    Double ib1 = ibc;
+    Double ib2 = f+1.;
+    if ((i&1) == 0) {
+      ib1 = (ibc+1.);
+      ib2 = (f+2.);
+    }
+    return ib1+ib2;
+  }
+
+  //===============================================
+  static double sum(double[] a) {
+    double result = 1.;
+    for (Double i : a)
+        result += i;
+    return result;
+  }
+
+  static double sumb(double[] a) {
+    Double result = 1.;
+    for (Double i : a)
+        result += i;
+    return result;
+  }
+
+  static double sumc(double[] a) {
+    Double result = ibc;
+    for (Double i : a)
+        result += i;
+    return result;
+  }
+
+  static double sumf(double[] a) {
+    Double result = foob(1.);
+    for (Double i : a)
+        result += i;
+    return result;
+  }
+
+  static double sump(double[] a, Double result) {
+    for (Double i : a)
+        result += i;
+    return result;
+  }
+
+  static double sum2(double[] a) {
+    double result1 = 1.;
+    double result2 = 1.;
+    for (Double i : a) {
+        result1 += i;
+        result2 += i + 1.;
+    }
+    return result1 + result2;
+  }
+
+  static double sumb2(double[] a) {
+    Double result1 = 1.;
+    Double result2 = 1.;
+    for (Double i : a) {
+        result1 += i;
+        result2 += i + 1.;
+    }
+    return result1 + result2;
+  }
+
+  static double summ2(double[] a) {
+    Double result1 = 1.;
+    Double result2 = new Double(1.);
+    for (Double i : a) {
+        result1 += i;
+        result2 += new Double(i + 1.);
+    }
+    return result1 + result2;
+  }
+
+  static double sump2(double[] a, Double result2) {
+    Double result1 = 1.;
+    for (Double i : a) {
+        result1 += i;
+        result2 += i + 1.;
+    }
+    return result1 + result2;
+  }
+
+  static double sumc2(double[] a) {
+    Double result1 = 1.;
+    Double result2 = ibc;
+    for (Double i : a) {
+        result1 += i;
+        result2 += i + ibc;
+    }
+    return result1 + result2;
+  }
+
+  //===============================================
+  static double remi_sum() {
+    Double j = new Double(1.);
+    for (int i = 0; i< 1000; i++) {
+      j = new Double(j + 1.);
+    }
+    return j;
+  }
+
+  static double remi_sumb() {
+    Double j = Double.valueOf(1.);
+    for (int i = 0; i< 1000; i++) {
+      j = j + 1.;
+    }
+    return j;
+  }
+
+  static double remi_sumf() {
+    Double j = foob(1.);
+    for (int i = 0; i< 1000; i++) {
+      j = j + 1.;
+    }
+    return j;
+  }
+
+  static double remi_sump(Double j) {
+    for (int i = 0; i< 1000; i++) {
+      j = new Double(j + 1.);
+    }
+    return j;
+  }
+
+  static double remi_sumc() {
+    Double j = ibc;
+    for (int i = 0; i< 1000; i++) {
+      j = j + ibc;
+    }
+    return j;
+  }
+
+  static double remi_sum2() {
+    Double j1 = new Double(1.);
+    Double j2 = new Double(1.);
+    for (int i = 0; i< 1000; i++) {
+      j1 = new Double(j1 + 1.);
+      j2 = new Double(j2 + 2.);
+    }
+    return j1 + j2;
+  }
+
+  static double remi_sumb2() {
+    Double j1 = Double.valueOf(1.);
+    Double j2 = Double.valueOf(1.);
+    for (int i = 0; i< 1000; i++) {
+      j1 = j1 + 1.;
+      j2 = j2 + 2.;
+    }
+    return j1 + j2;
+  }
+
+  static double remi_summ2() {
+    Double j1 = new Double(1.);
+    Double j2 = Double.valueOf(1.);
+    for (int i = 0; i< 1000; i++) {
+      j1 = new Double(j1 + 1.);
+      j2 = j2 + 2.;
+    }
+    return j1 + j2;
+  }
+
+  static double remi_sump2(Double j1) {
+    Double j2 = Double.valueOf(1.);
+    for (int i = 0; i< 1000; i++) {
+      j1 = new Double(j1 + 1.);
+      j2 = j2 + 2.;
+    }
+    return j1 + j2;
+  }
+
+  static double remi_sumc2() {
+    Double j1 = ibc;
+    Double j2 = Double.valueOf(1.);
+    for (int i = 0; i< 1000; i++) {
+      j1 = j1 + ibc;
+      j2 = j2 + 2.;
+    }
+    return j1 + j2;
+  }
+
+
+  //===============================================
+  // Safepointa and debug info for deoptimization
+  static double simple_deop(double i) {
+    Double ib = new Double(foo(i));
+    dummy();
+    return ib;
+  }
+
+  static double simpleb_deop(double i) {
+    Double ib = Double.valueOf(foo(i));
+    dummy();
+    return ib;
+  }
+
+  static double simplef_deop(double i) {
+    Double ib = foob(i);
+    dummy();
+    return ib;
+  }
+
+  static double simplep_deop(Double ib) {
+    dummy();
+    return ib;
+  }
+
+  static double simplec_deop(double i) {
+    Double ib = ibc;
+    dummy();
+    return ib;
+  }
+
+  static double test_deop(double f, int i) {
+    Double ib = new Double(foo(f));
+    if ((i&1) == 0)
+      ib = foo(f+1.);
+    dummy();
+    return ib;
+  }
+
+  static double testb_deop(double f, int i) {
+    Double ib = foo(f);
+    if ((i&1) == 0)
+      ib = foo(f+1.);
+    dummy();
+    return ib;
+  }
+
+  static double testf_deop(double f, int i) {
+    Double ib = foob(f);
+    if ((i&1) == 0)
+      ib = foo(f+1.);
+    dummy();
+    return ib;
+  }
+
+  static double testp_deop(double f, int i, Double ib) {
+    if ((i&1) == 0)
+      ib = foo(f+1.);
+    dummy();
+    return ib;
+  }
+
+  static double testc_deop(double f, int i) {
+    Double ib = ibc;
+    if ((i&1) == 0)
+      ib = foo(f+1.);
+    dummy();
+    return ib;
+  }
+
+  static double sum_deop(double[] a) {
+    double result = 1.;
+    for (Double i : a)
+        result += foo(i);
+    dummy();
+    return result;
+  }
+
+  static double sumb_deop(double[] a) {
+    Double result = 1.;
+    for (Double i : a)
+        result += foo(i);
+    dummy();
+    return result;
+  }
+
+  static double sumf_deop(double[] a) {
+    Double result = 1.;
+    for (Double i : a)
+        result += foob(i);
+    dummy();
+    return result;
+  }
+
+  static double sump_deop(double[] a, Double result) {
+    for (Double i : a)
+        result += foob(i);
+    dummy();
+    return result;
+  }
+
+  static double sumc_deop(double[] a) {
+    Double result = ibc;
+    for (Double i : a)
+        result += foo(i);
+    dummy();
+    return result;
+  }
+
+  static double remi_sum_deop() {
+    Double j = new Double(foo(1.));
+    for (int i = 0; i< 1000; i++) {
+      j = new Double(foo(j + 1.));
+    }
+    dummy();
+    return j;
+  }
+
+  static double remi_sumb_deop() {
+    Double j = Double.valueOf(foo(1.));
+    for (int i = 0; i< 1000; i++) {
+      j = foo(j + 1.);
+    }
+    dummy();
+    return j;
+  }
+
+  static double remi_sumf_deop() {
+    Double j = foob(1.);
+    for (int i = 0; i< 1000; i++) {
+      j = foo(j + 1.);
+    }
+    dummy();
+    return j;
+  }
+
+  static double remi_sump_deop(Double j) {
+    for (int i = 0; i< 1000; i++) {
+      j = foo(j + 1.);
+    }
+    dummy();
+    return j;
+  }
+
+  static double remi_sumc_deop() {
+    Double j = ibc;
+    for (int i = 0; i< 1000; i++) {
+      j = foo(j + 1.);
+    }
+    dummy();
+    return j;
+  }
+
+  //===============================================
+  // Conditional increment
+  static double remi_sum_cond() {
+    Double j = new Double(1.);
+    for (int i = 0; i< 1000; i++) {
+      if ((i&1) == 0) {
+        j = new Double(j + 1.);
+      }
+    }
+    return j;
+  }
+
+  static double remi_sumb_cond() {
+    Double j = Double.valueOf(1.);
+    for (int i = 0; i< 1000; i++) {
+      if ((i&1) == 0) {
+        j = j + 1.;
+      }
+    }
+    return j;
+  }
+
+  static double remi_sumf_cond() {
+    Double j = foob(1.);
+    for (int i = 0; i< 1000; i++) {
+      if ((i&1) == 0) {
+        j = j + 1.;
+      }
+    }
+    return j;
+  }
+
+  static double remi_sump_cond(Double j) {
+    for (int i = 0; i< 1000; i++) {
+      if ((i&1) == 0) {
+        j = j + 1.;
+      }
+    }
+    return j;
+  }
+
+  static double remi_sumc_cond() {
+    Double j = ibc;
+    for (int i = 0; i< 1000; i++) {
+      if ((i&1) == 0) {
+        j = j + ibc;
+      }
+    }
+    return j;
+  }
+
+  static double remi_sum2_cond() {
+    Double j1 = new Double(1.);
+    Double j2 = new Double(1.);
+    for (int i = 0; i< 1000; i++) {
+      if ((i&1) == 0) {
+        j1 = new Double(j1 + 1.);
+      } else {
+        j2 = new Double(j2 + 2.);
+      }
+    }
+    return j1 + j2;
+  }
+
+  static double remi_sumb2_cond() {
+    Double j1 = Double.valueOf(1.);
+    Double j2 = Double.valueOf(1.);
+    for (int i = 0; i< 1000; i++) {
+      if ((i&1) == 0) {
+        j1 = j1 + 1.;
+      } else {
+        j2 = j2 + 2.;
+      }
+    }
+    return j1 + j2;
+  }
+
+  static double remi_summ2_cond() {
+    Double j1 = new Double(1.);
+    Double j2 = Double.valueOf(1.);
+    for (int i = 0; i< 1000; i++) {
+      if ((i&1) == 0) {
+        j1 = new Double(j1 + 1.);
+      } else {
+        j2 = j2 + 2.;
+      }
+    }
+    return j1 + j2;
+  }
+
+  static double remi_sump2_cond(Double j1) {
+    Double j2 = Double.valueOf(1.);
+    for (int i = 0; i< 1000; i++) {
+      if ((i&1) == 0) {
+        j1 = new Double(j1 + 1.);
+      } else {
+        j2 = j2 + 2.;
+      }
+    }
+    return j1 + j2;
+  }
+
+  static double remi_sumc2_cond() {
+    Double j1 = ibc;
+    Double j2 = Double.valueOf(1.);
+    for (int i = 0; i< 1000; i++) {
+      if ((i&1) == 0) {
+        j1 = j1 + ibc;
+      } else {
+        j2 = j2 + 2;
+      }
+    }
+    return j1 + j2;
+  }
+
+
+  public static void main(String[] args) {
+    final int ntests = 70;
+
+    String[] test_name = new String[] {
+        "simple",      "simpleb",      "simplec",      "simplef",      "simplep",
+        "simple2",     "simpleb2",     "simplec2",     "simplem2",     "simplep2",
+        "simple_deop", "simpleb_deop", "simplec_deop", "simplef_deop", "simplep_deop",
+        "test",        "testb",        "testc",        "testm",        "testp",
+        "test2",       "testb2",       "testc2",       "testm2",       "testp2",
+        "test_deop",   "testb_deop",   "testc_deop",   "testf_deop",   "testp_deop",
+        "sum",         "sumb",         "sumc",         "sumf",         "sump",
+        "sum2",        "sumb2",        "sumc2",        "summ2",        "sump2",
+        "sum_deop",    "sumb_deop",    "sumc_deop",    "sumf_deop",    "sump_deop",
+        "remi_sum",       "remi_sumb",       "remi_sumc",       "remi_sumf",       "remi_sump",
+        "remi_sum2",      "remi_sumb2",      "remi_sumc2",      "remi_summ2",      "remi_sump2",
+        "remi_sum_deop",  "remi_sumb_deop",  "remi_sumc_deop",  "remi_sumf_deop",  "remi_sump_deop",
+        "remi_sum_cond",  "remi_sumb_cond",  "remi_sumc_cond",  "remi_sumf_cond",  "remi_sump_cond",
+        "remi_sum2_cond", "remi_sumb2_cond", "remi_sumc2_cond", "remi_summ2_cond", "remi_sump2_cond"
+    };
+
+    final double[] val = new double[] {
+       71994000.,  71994000.,    12000.,  71994000.,  71994000.,
+      144000000., 144000000., 72018000., 144000000., 144000000.,
+       71994000.,  71994000.,    12000.,  71994000.,  71994000.,
+       72000000.,  72000000., 36006000.,  72000000.,  72000000.,
+      144012000., 144012000., 72030000., 144012000., 144012000.,
+       72000000.,  72000000., 36006000.,  72000000.,  72000000.,
+         499501.,    499501.,   499501.,    499501.,    499501.,
+        1000002.,   1000002.,  1000002.,   1000002.,   1000002.,
+         499501.,    499501.,   499501.,    499501.,    499501.,
+           1001.,      1001.,     1001.,      1001.,      1001.,
+           3002.,      3002.,     3002.,      3002.,      3002.,
+           1001.,      1001.,     1001.,      1001.,      1001.,
+            501.,       501.,      501.,       501.,       501.,
+           1502.,      1502.,     1502.,      1502.,      1502.
+    };
+
+    double[] res = new double[ntests];
+    for (int i = 0; i < ntests; i++) {
+      res[i] = 0.;
+    }
+
+
+    for (int i = 0; i < 12000; i++) {
+      res[0] += simple(i);
+      res[1] += simpleb(i);
+      res[2] += simplec();
+      res[3] += simplef(i);
+      res[4] += simplep((double)i);
+
+      res[5] += simple2((double)i);
+      res[6] += simpleb2((double)i);
+      res[7] += simplec2((double)i);
+      res[8] += simplem2((double)i);
+      res[9] += simplep2((double)i, (double)i);
+
+      res[10] += simple_deop((double)i);
+      res[11] += simpleb_deop((double)i);
+      res[12] += simplec_deop((double)i);
+      res[13] += simplef_deop((double)i);
+      res[14] += simplep_deop((double)i);
+
+      res[15] += test((double)i, i);
+      res[16] += testb((double)i, i);
+      res[17] += testc((double)i, i);
+      res[18] += testm((double)i, i);
+      res[19] += testp((double)i, i, (double)i);
+
+      res[20] += test2((double)i, i);
+      res[21] += testb2((double)i, i);
+      res[22] += testc2((double)i, i);
+      res[23] += testm2((double)i, i);
+      res[24] += testp2((double)i, i, (double)i);
+
+      res[25] += test_deop((double)i, i);
+      res[26] += testb_deop((double)i, i);
+      res[27] += testc_deop((double)i, i);
+      res[28] += testf_deop((double)i, i);
+      res[29] += testp_deop((double)i, i, (double)i);
+    }
+
+    double[] ia = new double[1000];
+    for (int i = 0; i < 1000; i++) {
+      ia[i] = i;
+    }
+
+    for (int i = 0; i < 100; i++) {
+      res[30] = sum(ia);
+      res[31] = sumb(ia);
+      res[32] = sumc(ia);
+      res[33] = sumf(ia);
+      res[34] = sump(ia, 1.);
+
+      res[35] = sum2(ia);
+      res[36] = sumb2(ia);
+      res[37] = sumc2(ia);
+      res[38] = summ2(ia);
+      res[39] = sump2(ia, 1.);
+
+      res[40] = sum_deop(ia);
+      res[41] = sumb_deop(ia);
+      res[42] = sumc_deop(ia);
+      res[43] = sumf_deop(ia);
+      res[44] = sump_deop(ia, 1.);
+
+      res[45] = remi_sum();
+      res[46] = remi_sumb();
+      res[47] = remi_sumc();
+      res[48] = remi_sumf();
+      res[49] = remi_sump(1.);
+
+      res[50] = remi_sum2();
+      res[51] = remi_sumb2();
+      res[52] = remi_sumc2();
+      res[53] = remi_summ2();
+      res[54] = remi_sump2(1.);
+
+      res[55] = remi_sum_deop();
+      res[56] = remi_sumb_deop();
+      res[57] = remi_sumc_deop();
+      res[58] = remi_sumf_deop();
+      res[59] = remi_sump_deop(1.);
+
+      res[60] = remi_sum_cond();
+      res[61] = remi_sumb_cond();
+      res[62] = remi_sumc_cond();
+      res[63] = remi_sumf_cond();
+      res[64] = remi_sump_cond(1.);
+
+      res[65] = remi_sum2_cond();
+      res[66] = remi_sumb2_cond();
+      res[67] = remi_sumc2_cond();
+      res[68] = remi_summ2_cond();
+      res[69] = remi_sump2_cond(1.);
+    }
+
+    int failed = 0;
+    for (int i = 0; i < ntests; i++) {
+      if (res[i] != val[i]) {
+        System.err.println(test_name[i] + ": " + res[i] + " != " + val[i]);
+        failed++;
+      }
+    }
+    if (failed > 0) {
+      System.err.println("Failed " + failed + " tests.");
+      throw new InternalError();
+    } else {
+      System.out.println("Passed.");
+    }
+  }
+}