8005031: Some cleanup in c2 to prepare for incremental inlining support

Summary: collection of small changes to prepare for incremental inlining.
Reviewed-by: twisti, kvn

src/share/vm/ci/ciField.cpp

@@ -366,10 +366,12 @@ bool ciField::will_link(ciInstanceKlass* accessing_klass,
 // ------------------------------------------------------------------
 // ciField::print
 void ciField::print() {
-  tty->print("<ciField ");
+  tty->print("<ciField name=");
   _holder->print_name();
   tty->print(".");
   _name->print_symbol();
+  tty->print(" signature=");
+  _signature->print_symbol();
   tty->print(" offset=%d type=", _offset);
   if (_type != NULL) _type->print_name();
   else               tty->print("(reference)");

src/share/vm/compiler/compilerOracle.cpp

@@ -538,6 +538,7 @@ void CompilerOracle::parse_from_line(char* line) {
 
   if (match != NULL) {
     if (!_quiet) {
+      ResourceMark rm;
       tty->print("CompilerOracle: %s ", command_names[command]);
       match->print();
     }

src/share/vm/opto/addnode.cpp

@@ -189,6 +189,11 @@ Node *AddNode::Ideal(PhaseGVN *phase, bool can_reshape) {
       set_req(1, addx);
       set_req(2, a22);
       progress = this;
+      PhaseIterGVN *igvn = phase->is_IterGVN();
+      if (add2->outcnt() == 0 && igvn) {
+        // add disconnected.
+        igvn->_worklist.push(add2);
+      }
     }
   }
 
@@ -624,6 +629,11 @@ Node *AddPNode::Ideal(PhaseGVN *phase, bool can_reshape) {
     if( t22->singleton() && (t22 != Type::TOP) ) {  // Right input is an add of a constant?
       set_req(Address, phase->transform(new (phase->C) AddPNode(in(Base),in(Address),add->in(1))));
       set_req(Offset, add->in(2));
+      PhaseIterGVN *igvn = phase->is_IterGVN();
+      if (add->outcnt() == 0 && igvn) {
+        // add disconnected.
+        igvn->_worklist.push((Node*)add);
+      }
       return this;              // Made progress
     }
   }

src/share/vm/opto/bytecodeInfo.cpp

@@ -403,7 +403,7 @@ const char* InlineTree::check_can_parse(ciMethod* callee) {
 //------------------------------print_inlining---------------------------------
 // Really, the failure_msg can be a success message also.
 void InlineTree::print_inlining(ciMethod* callee_method, int caller_bci, const char* failure_msg) const {
-  CompileTask::print_inlining(callee_method, inline_level(), caller_bci, failure_msg ? failure_msg : "inline");
+  C->print_inlining(callee_method, inline_level(), caller_bci, failure_msg ? failure_msg : "inline");
   if (callee_method == NULL)  tty->print(" callee not monotonic or profiled");
   if (Verbose && callee_method) {
     const InlineTree *top = this;

src/share/vm/opto/callGenerator.cpp

@@ -274,6 +274,9 @@ class LateInlineCallGenerator : public DirectCallGenerator {
   virtual void do_late_inline();
 
   virtual JVMState* generate(JVMState* jvms) {
+    Compile *C = Compile::current();
+    C->print_inlining_skip(this);
+
     // Record that this call site should be revisited once the main
     // parse is finished.
     Compile::current()->add_late_inline(this);
@@ -284,7 +287,6 @@ class LateInlineCallGenerator : public DirectCallGenerator {
     // as is done for allocations and macro expansion.
     return DirectCallGenerator::generate(jvms);
   }
-
 };
 
 
@@ -307,7 +309,9 @@ void LateInlineCallGenerator::do_late_inline() {
 
   // Make sure the state is a MergeMem for parsing.
   if (!map->in(TypeFunc::Memory)->is_MergeMem()) {
-    map->set_req(TypeFunc::Memory, MergeMemNode::make(C, map->in(TypeFunc::Memory)));
+    Node* mem = MergeMemNode::make(C, map->in(TypeFunc::Memory));
+    C->initial_gvn()->set_type_bottom(mem);
+    map->set_req(TypeFunc::Memory, mem);
   }
 
   // Make enough space for the expression stack and transfer the incoming arguments
@@ -320,6 +324,8 @@ void LateInlineCallGenerator::do_late_inline() {
     }
   }
 
+  C->print_inlining_insert(this);
+
   CompileLog* log = C->log();
   if (log != NULL) {
     log->head("late_inline method='%d'", log->identify(method()));
@@ -608,7 +614,7 @@ CallGenerator* CallGenerator::for_method_handle_inline(JVMState* jvms, ciMethod*
         if (cg != NULL && cg->is_inline())
           return cg;
       } else {
-        if (PrintInlining)  CompileTask::print_inlining(callee, jvms->depth() - 1, jvms->bci(), "receiver not constant");
+        if (PrintInlining)  C->print_inlining(callee, jvms->depth() - 1, jvms->bci(), "receiver not constant");
       }
     }
     break;

src/share/vm/opto/callGenerator.hpp

@@ -147,9 +147,9 @@ class CallGenerator : public ResourceObj {
                                                 CallGenerator* cg);
   virtual Node* generate_predicate(JVMState* jvms) { return NULL; };
 
-  static void print_inlining(ciMethod* callee, int inline_level, int bci, const char* msg) {
+  static void print_inlining(Compile* C, ciMethod* callee, int inline_level, int bci, const char* msg) {
     if (PrintInlining)
-      CompileTask::print_inlining(callee, inline_level, bci, msg);
+      C->print_inlining(callee, inline_level, bci, msg);
   }
 };
 

src/share/vm/opto/callnode.cpp

@@ -751,7 +751,7 @@ void CallNode::extract_projections(CallProjections* projs, bool separate_io_proj
         projs->fallthrough_ioproj = pn;
       for (DUIterator j = pn->outs(); pn->has_out(j); j++) {
         Node* e = pn->out(j);
-        if (e->Opcode() == Op_CreateEx && e->in(0)->is_CatchProj()) {
+        if (e->Opcode() == Op_CreateEx && e->in(0)->is_CatchProj() && e->outcnt() > 0) {
           assert(projs->exobj == NULL, "only one");
           projs->exobj = e;
         }

src/share/vm/opto/cfgnode.cpp

@@ -1566,6 +1566,10 @@ Node *PhiNode::Ideal(PhaseGVN *phase, bool can_reshape) {
     Node* n = in(j);            // Get the input
     if (rc == NULL || phase->type(rc) == Type::TOP) {
       if (n != top) {           // Not already top?
+        PhaseIterGVN *igvn = phase->is_IterGVN();
+        if (can_reshape && igvn != NULL) {
+          igvn->_worklist.push(r);
+        }
         set_req(j, top);        // Nuke it down
         progress = this;        // Record progress
       }

src/share/vm/opto/compile.cpp

@@ -610,7 +610,9 @@ Compile::Compile( ciEnv* ci_env, C2Compiler* compiler, ciMethod* target, int osr
                   _trace_opto_output(TraceOptoOutput || method()->has_option("TraceOptoOutput")),
                   _printer(IdealGraphPrinter::printer()),
 #endif
-                  _congraph(NULL) {
+                  _congraph(NULL),
+                  _print_inlining_list(NULL),
+                  _print_inlining(0) {
   C = this;
 
   CompileWrapper cw(this);
@@ -666,6 +668,9 @@ Compile::Compile( ciEnv* ci_env, C2Compiler* compiler, ciMethod* target, int osr
   PhaseGVN gvn(node_arena(), estimated_size);
   set_initial_gvn(&gvn);
 
+  if (PrintInlining) {
+    _print_inlining_list = new (comp_arena())GrowableArray<PrintInliningBuffer>(comp_arena(), 1, 1, PrintInliningBuffer());
+  }
   { // Scope for timing the parser
     TracePhase t3("parse", &_t_parser, true);
 
@@ -754,6 +759,7 @@ Compile::Compile( ciEnv* ci_env, C2Compiler* compiler, ciMethod* target, int osr
       }
     }
     assert(_late_inlines.length() == 0, "should have been processed");
+    dump_inlining();
 
     print_method("Before RemoveUseless", 3);
 
@@ -899,7 +905,9 @@ Compile::Compile( ciEnv* ci_env,
 #endif
     _dead_node_list(comp_arena()),
     _dead_node_count(0),
-    _congraph(NULL) {
+    _congraph(NULL),
+    _print_inlining_list(NULL),
+    _print_inlining(0) {
   C = this;
 
 #ifndef PRODUCT
@@ -3351,3 +3359,11 @@ void Compile::ConstantTable::fill_jump_table(CodeBuffer& cb, MachConstantNode* n
     cb.consts()->relocate((address) constant_addr, relocInfo::internal_word_type);
   }
 }
+
+void Compile::dump_inlining() {
+  if (PrintInlining) {
+    for (int i = 0; i < _print_inlining_list->length(); i++) {
+      tty->print(_print_inlining_list->at(i).ss()->as_string());
+    }
+  }
+}

src/share/vm/opto/compile.hpp

@@ -30,6 +30,7 @@
 #include "code/debugInfoRec.hpp"
 #include "code/exceptionHandlerTable.hpp"
 #include "compiler/compilerOracle.hpp"
+#include "compiler/compileBroker.hpp"
 #include "libadt/dict.hpp"
 #include "libadt/port.hpp"
 #include "libadt/vectset.hpp"
@@ -369,6 +370,61 @@ class Compile : public Phase {
   GrowableArray<CallGenerator*> _late_inlines;  // List of CallGenerators to be revisited after
                                                 // main parsing has finished.
 
+  // Inlining may not happen in parse order which would make
+  // PrintInlining output confusing. Keep track of PrintInlining
+  // pieces in order.
+  class PrintInliningBuffer : public ResourceObj {
+   private:
+    CallGenerator* _cg;
+    stringStream* _ss;
+
+   public:
+    PrintInliningBuffer()
+      : _cg(NULL) { _ss = new stringStream(); }
+
+    stringStream* ss() const { return _ss; }
+    CallGenerator* cg() const { return _cg; }
+    void set_cg(CallGenerator* cg) { _cg = cg; }
+  };
+
+  GrowableArray<PrintInliningBuffer>* _print_inlining_list;
+  int _print_inlining;
+
+ public:
+
+  outputStream* print_inlining_stream() const {
+    return _print_inlining_list->at(_print_inlining).ss();
+  }
+
+  void print_inlining_skip(CallGenerator* cg) {
+    if (PrintInlining) {
+      _print_inlining_list->at(_print_inlining).set_cg(cg);
+      _print_inlining++;
+      _print_inlining_list->insert_before(_print_inlining, PrintInliningBuffer());
+    }
+  }
+
+  void print_inlining_insert(CallGenerator* cg) {
+    if (PrintInlining) {
+      for (int i = 0; i < _print_inlining_list->length(); i++) {
+        if (_print_inlining_list->at(i).cg() == cg) {
+          _print_inlining_list->insert_before(i+1, PrintInliningBuffer());
+          _print_inlining = i+1;
+          _print_inlining_list->at(i).set_cg(NULL);
+          return;
+        }
+      }
+      ShouldNotReachHere();
+    }
+  }
+
+  void print_inlining(ciMethod* method, int inline_level, int bci, const char* msg = NULL) {
+    stringStream ss;
+    CompileTask::print_inlining(&ss, method, inline_level, bci, msg);
+    print_inlining_stream()->print(ss.as_string());
+  }
+
+ private:
   // Matching, CFG layout, allocation, code generation
   PhaseCFG*             _cfg;                   // Results of CFG finding
   bool                  _select_24_bit_instr;   // We selected an instruction with a 24-bit result
@@ -591,7 +647,7 @@ class Compile : public Phase {
   void         reset_dead_node_list()      { _dead_node_list.Reset();
                                              _dead_node_count = 0;
                                            }
-  uint          live_nodes()               {
+  uint          live_nodes() const         {
     int  val = _unique - _dead_node_count;
     assert (val >= 0, err_msg_res("number of tracked dead nodes %d more than created nodes %d", _unique, _dead_node_count));
             return (uint) val;
@@ -702,7 +758,7 @@ class Compile : public Phase {
 
   void              identify_useful_nodes(Unique_Node_List &useful);
   void              update_dead_node_list(Unique_Node_List &useful);
-  void              remove_useless_nodes  (Unique_Node_List &useful);
+  void              remove_useless_nodes (Unique_Node_List &useful);
 
   WarmCallInfo*     warm_calls() const          { return _warm_calls; }
   void          set_warm_calls(WarmCallInfo* l) { _warm_calls = l; }
@@ -711,6 +767,8 @@ class Compile : public Phase {
   // Record this CallGenerator for inlining at the end of parsing.
   void              add_late_inline(CallGenerator* cg) { _late_inlines.push(cg); }
 
+  void dump_inlining();
+
   // Matching, CFG layout, allocation, code generation
   PhaseCFG*         cfg()                       { return _cfg; }
   bool              select_24_bit_instr() const { return _select_24_bit_instr; }

src/share/vm/opto/doCall.cpp

@@ -40,19 +40,24 @@
 #include "prims/nativeLookup.hpp"
 #include "runtime/sharedRuntime.hpp"
 
-void trace_type_profile(ciMethod *method, int depth, int bci, ciMethod *prof_method, ciKlass *prof_klass, int site_count, int receiver_count) {
+void trace_type_profile(Compile* C, ciMethod *method, int depth, int bci, ciMethod *prof_method, ciKlass *prof_klass, int site_count, int receiver_count) {
   if (TraceTypeProfile || PrintInlining NOT_PRODUCT(|| PrintOptoInlining)) {
+    outputStream* out = tty;
     if (!PrintInlining) {
       if (NOT_PRODUCT(!PrintOpto &&) !PrintCompilation) {
         method->print_short_name();
         tty->cr();
       }
       CompileTask::print_inlining(prof_method, depth, bci);
+    } else {
+      out = C->print_inlining_stream();
     }
-    CompileTask::print_inline_indent(depth);
-    tty->print(" \\-> TypeProfile (%d/%d counts) = ", receiver_count, site_count);
-    prof_klass->name()->print_symbol();
-    tty->cr();
+    CompileTask::print_inline_indent(depth, out);
+    out->print(" \\-> TypeProfile (%d/%d counts) = ", receiver_count, site_count);
+    stringStream ss;
+    prof_klass->name()->print_symbol_on(&ss);
+    out->print(ss.as_string());
+    out->cr();
   }
 }
 
@@ -233,13 +238,13 @@ CallGenerator* Compile::call_generator(ciMethod* callee, int vtable_index, bool
           }
           if (miss_cg != NULL) {
             if (next_hit_cg != NULL) {
-              trace_type_profile(jvms->method(), jvms->depth() - 1, jvms->bci(), next_receiver_method, profile.receiver(1), site_count, profile.receiver_count(1));
+              trace_type_profile(C, jvms->method(), jvms->depth() - 1, jvms->bci(), next_receiver_method, profile.receiver(1), site_count, profile.receiver_count(1));
               // We don't need to record dependency on a receiver here and below.
               // Whenever we inline, the dependency is added by Parse::Parse().
               miss_cg = CallGenerator::for_predicted_call(profile.receiver(1), miss_cg, next_hit_cg, PROB_MAX);
             }
             if (miss_cg != NULL) {
-              trace_type_profile(jvms->method(), jvms->depth() - 1, jvms->bci(), receiver_method, profile.receiver(0), site_count, receiver_count);
+              trace_type_profile(C, jvms->method(), jvms->depth() - 1, jvms->bci(), receiver_method, profile.receiver(0), site_count, receiver_count);
               CallGenerator* cg = CallGenerator::for_predicted_call(profile.receiver(0), miss_cg, hit_cg, profile.receiver_prob(0));
               if (cg != NULL)  return cg;
             }

src/share/vm/opto/graphKit.cpp

@@ -1771,11 +1771,21 @@ void GraphKit::replace_call(CallNode* call, Node* result) {
   CallProjections callprojs;
   call->extract_projections(&callprojs, true);
 
-  // Replace all the old call edges with the edges from the inlining result
-  C->gvn_replace_by(callprojs.fallthrough_catchproj, final_state->in(TypeFunc::Control));
-  C->gvn_replace_by(callprojs.fallthrough_memproj,   final_state->in(TypeFunc::Memory));
-  C->gvn_replace_by(callprojs.fallthrough_ioproj,    final_state->in(TypeFunc::I_O));
+  Node* init_mem = call->in(TypeFunc::Memory);
   Node* final_mem = final_state->in(TypeFunc::Memory);
+  Node* final_ctl = final_state->in(TypeFunc::Control);
+  Node* final_io = final_state->in(TypeFunc::I_O);
+
+  // Replace all the old call edges with the edges from the inlining result
+  if (callprojs.fallthrough_catchproj != NULL) {
+    C->gvn_replace_by(callprojs.fallthrough_catchproj, final_ctl);
+  }
+  if (callprojs.fallthrough_memproj != NULL) {
+    C->gvn_replace_by(callprojs.fallthrough_memproj,   final_mem);
+  }
+  if (callprojs.fallthrough_ioproj != NULL) {
+    C->gvn_replace_by(callprojs.fallthrough_ioproj,    final_io);
+  }
 
   // Replace the result with the new result if it exists and is used
   if (callprojs.resproj != NULL && result != NULL) {

src/share/vm/opto/library_call.cpp

@@ -412,16 +412,16 @@ CallGenerator* Compile::make_vm_intrinsic(ciMethod* m, bool is_virtual) {
     break;
 
   case vmIntrinsics::_reverseBytes_c:
-    if (!Matcher::match_rule_supported(Op_ReverseBytesUS)) return false;
+    if (!Matcher::match_rule_supported(Op_ReverseBytesUS)) return NULL;
     break;
   case vmIntrinsics::_reverseBytes_s:
-    if (!Matcher::match_rule_supported(Op_ReverseBytesS))  return false;
+    if (!Matcher::match_rule_supported(Op_ReverseBytesS))  return NULL;
     break;
   case vmIntrinsics::_reverseBytes_i:
-    if (!Matcher::match_rule_supported(Op_ReverseBytesI))  return false;
+    if (!Matcher::match_rule_supported(Op_ReverseBytesI))  return NULL;
     break;
   case vmIntrinsics::_reverseBytes_l:
-    if (!Matcher::match_rule_supported(Op_ReverseBytesL))  return false;
+    if (!Matcher::match_rule_supported(Op_ReverseBytesL))  return NULL;
     break;
 
   case vmIntrinsics::_Reference_get:
@@ -536,7 +536,7 @@ JVMState* LibraryIntrinsic::generate(JVMState* jvms) {
   // Try to inline the intrinsic.
   if (kit.try_to_inline()) {
     if (PrintIntrinsics || PrintInlining NOT_PRODUCT( || PrintOptoInlining) ) {
-      CompileTask::print_inlining(callee, jvms->depth() - 1, bci, is_virtual() ? "(intrinsic, virtual)" : "(intrinsic)");
+      C->print_inlining(callee, jvms->depth() - 1, bci, is_virtual() ? "(intrinsic, virtual)" : "(intrinsic)");
     }
     C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_worked);
     if (C->log()) {
@@ -555,7 +555,7 @@ JVMState* LibraryIntrinsic::generate(JVMState* jvms) {
     if (jvms->has_method()) {
       // Not a root compile.
       const char* msg = is_virtual() ? "failed to inline (intrinsic, virtual)" : "failed to inline (intrinsic)";
-      CompileTask::print_inlining(callee, jvms->depth() - 1, bci, msg);
+      C->print_inlining(callee, jvms->depth() - 1, bci, msg);
     } else {
       // Root compile
       tty->print("Did not generate intrinsic %s%s at bci:%d in",
@@ -585,7 +585,7 @@ Node* LibraryIntrinsic::generate_predicate(JVMState* jvms) {
   Node* slow_ctl = kit.try_to_predicate();
   if (!kit.failing()) {
     if (PrintIntrinsics || PrintInlining NOT_PRODUCT( || PrintOptoInlining) ) {
-      CompileTask::print_inlining(callee, jvms->depth() - 1, bci, is_virtual() ? "(intrinsic, virtual)" : "(intrinsic)");
+      C->print_inlining(callee, jvms->depth() - 1, bci, is_virtual() ? "(intrinsic, virtual)" : "(intrinsic)");
     }
     C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_worked);
     if (C->log()) {
@@ -602,12 +602,12 @@ Node* LibraryIntrinsic::generate_predicate(JVMState* jvms) {
     if (jvms->has_method()) {
       // Not a root compile.
       const char* msg = "failed to generate predicate for intrinsic";
-      CompileTask::print_inlining(kit.callee(), jvms->depth() - 1, bci, msg);
+      C->print_inlining(kit.callee(), jvms->depth() - 1, bci, msg);
     } else {
       // Root compile
-      tty->print("Did not generate predicate for intrinsic %s%s at bci:%d in",
-               vmIntrinsics::name_at(intrinsic_id()),
-               (is_virtual() ? " (virtual)" : ""), bci);
+      C->print_inlining_stream()->print("Did not generate predicate for intrinsic %s%s at bci:%d in",
+                                        vmIntrinsics::name_at(intrinsic_id()),
+                                        (is_virtual() ? " (virtual)" : ""), bci);
     }
   }
   C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_failed);
@@ -3319,7 +3319,7 @@ bool LibraryCallKit::inline_native_subtype_check() {
     Node* arg = args[which_arg];
     arg = null_check(arg);
     if (stopped())  break;
-    args[which_arg] = _gvn.transform(arg);
+    args[which_arg] = arg;
 
     Node* p = basic_plus_adr(arg, class_klass_offset);
     Node* kls = LoadKlassNode::make(_gvn, immutable_memory(), p, adr_type, kls_type);

src/share/vm/opto/stringopts.cpp

@@ -744,7 +744,9 @@ bool StringConcat::validate_control_flow() {
       ctrl_path.push(cn);
       ctrl_path.push(cn->proj_out(0));
       ctrl_path.push(cn->proj_out(0)->unique_out());
-      ctrl_path.push(cn->proj_out(0)->unique_out()->as_Catch()->proj_out(0));
+      if (cn->proj_out(0)->unique_out()->as_Catch()->proj_out(0) != NULL) {
+        ctrl_path.push(cn->proj_out(0)->unique_out()->as_Catch()->proj_out(0));
+      }
     } else {
       ShouldNotReachHere();
     }
@@ -762,6 +764,12 @@ bool StringConcat::validate_control_flow() {
     } else if (ptr->is_IfTrue()) {
       IfNode* iff = ptr->in(0)->as_If();
       BoolNode* b = iff->in(1)->isa_Bool();
+
+      if (b == NULL) {
+        fail = true;
+        break;
+      }
+
       Node* cmp = b->in(1);
       Node* v1 = cmp->in(1);
       Node* v2 = cmp->in(2);
@@ -1408,71 +1416,76 @@ void PhaseStringOpts::replace_string_concat(StringConcat* sc) {
                       Deoptimization::Action_make_not_entrant);
   }
 
-  // length now contains the number of characters needed for the
-  // char[] so create a new AllocateArray for the char[]
-  Node* char_array = NULL;
-  {
-    PreserveReexecuteState preexecs(&kit);
-    // The original jvms is for an allocation of either a String or
-    // StringBuffer so no stack adjustment is necessary for proper
-    // reexecution.  If we deoptimize in the slow path the bytecode
-    // will be reexecuted and the char[] allocation will be thrown away.
-    kit.jvms()->set_should_reexecute(true);
-    char_array = kit.new_array(__ makecon(TypeKlassPtr::make(ciTypeArrayKlass::make(T_CHAR))),
-                               length, 1);
-  }
-
-  // Mark the allocation so that zeroing is skipped since the code
-  // below will overwrite the entire array
-  AllocateArrayNode* char_alloc = AllocateArrayNode::Ideal_array_allocation(char_array, _gvn);
-  char_alloc->maybe_set_complete(_gvn);
-
-  // Now copy the string representations into the final char[]
-  Node* start = __ intcon(0);
-  for (int argi = 0; argi < sc->num_arguments(); argi++) {
-    Node* arg = sc->argument(argi);
-    switch (sc->mode(argi)) {
-      case StringConcat::IntMode: {
-        Node* end = __ AddI(start, string_sizes->in(argi));
-        // getChars words backwards so pass the ending point as well as the start
-        int_getChars(kit, arg, char_array, start, end);
-        start = end;
-        break;
-      }
-      case StringConcat::StringNullCheckMode:
-      case StringConcat::StringMode: {
-        start = copy_string(kit, arg, char_array, start);
-        break;
-      }
-      case StringConcat::CharMode: {
-        __ store_to_memory(kit.control(), kit.array_element_address(char_array, start, T_CHAR),
-                           arg, T_CHAR, char_adr_idx);
-        start = __ AddI(start, __ intcon(1));
-        break;
+  Node* result;
+  if (!kit.stopped()) {
+
+    // length now contains the number of characters needed for the
+    // char[] so create a new AllocateArray for the char[]
+    Node* char_array = NULL;
+    {
+      PreserveReexecuteState preexecs(&kit);
+      // The original jvms is for an allocation of either a String or
+      // StringBuffer so no stack adjustment is necessary for proper
+      // reexecution.  If we deoptimize in the slow path the bytecode
+      // will be reexecuted and the char[] allocation will be thrown away.
+      kit.jvms()->set_should_reexecute(true);
+      char_array = kit.new_array(__ makecon(TypeKlassPtr::make(ciTypeArrayKlass::make(T_CHAR))),
+                                 length, 1);
+    }
+
+    // Mark the allocation so that zeroing is skipped since the code
+    // below will overwrite the entire array
+    AllocateArrayNode* char_alloc = AllocateArrayNode::Ideal_array_allocation(char_array, _gvn);
+    char_alloc->maybe_set_complete(_gvn);
+
+    // Now copy the string representations into the final char[]
+    Node* start = __ intcon(0);
+    for (int argi = 0; argi < sc->num_arguments(); argi++) {
+      Node* arg = sc->argument(argi);
+      switch (sc->mode(argi)) {
+        case StringConcat::IntMode: {
+          Node* end = __ AddI(start, string_sizes->in(argi));
+          // getChars words backwards so pass the ending point as well as the start
+          int_getChars(kit, arg, char_array, start, end);
+          start = end;
+          break;
+        }
+        case StringConcat::StringNullCheckMode:
+        case StringConcat::StringMode: {
+          start = copy_string(kit, arg, char_array, start);
+          break;
+        }
+        case StringConcat::CharMode: {
+          __ store_to_memory(kit.control(), kit.array_element_address(char_array, start, T_CHAR),
+                             arg, T_CHAR, char_adr_idx);
+          start = __ AddI(start, __ intcon(1));
+          break;
+        }
+        default:
+          ShouldNotReachHere();
       }
-      default:
-        ShouldNotReachHere();
     }
-  }
 
-  // If we're not reusing an existing String allocation then allocate one here.
-  Node* result = sc->string_alloc();
-  if (result == NULL) {
-    PreserveReexecuteState preexecs(&kit);
-    // The original jvms is for an allocation of either a String or
-    // StringBuffer so no stack adjustment is necessary for proper
-    // reexecution.
-    kit.jvms()->set_should_reexecute(true);
-    result = kit.new_instance(__ makecon(TypeKlassPtr::make(C->env()->String_klass())));
-  }
+    // If we're not reusing an existing String allocation then allocate one here.
+    result = sc->string_alloc();
+    if (result == NULL) {
+      PreserveReexecuteState preexecs(&kit);
+      // The original jvms is for an allocation of either a String or
+      // StringBuffer so no stack adjustment is necessary for proper
+      // reexecution.
+      kit.jvms()->set_should_reexecute(true);
+      result = kit.new_instance(__ makecon(TypeKlassPtr::make(C->env()->String_klass())));
+    }
 
-  // Intialize the string
-  if (java_lang_String::has_offset_field()) {
-    kit.store_String_offset(kit.control(), result, __ intcon(0));
-    kit.store_String_length(kit.control(), result, length);
+    // Intialize the string
+    if (java_lang_String::has_offset_field()) {
+      kit.store_String_offset(kit.control(), result, __ intcon(0));
+      kit.store_String_length(kit.control(), result, length);
+    }
+    kit.store_String_value(kit.control(), result, char_array);
+  } else {
+    result = C->top();
   }
-  kit.store_String_value(kit.control(), result, char_array);
-
   // hook up the outgoing control and result
   kit.replace_call(sc->end(), result);