Create a scoped arena allocator and use that for LVN.

This saves more than 0.5s of boot.oat compilation time
on Nexus 5.

TODO: Move other stuff to the scoped allocator. This CL
alone increases the peak memory allocation. By reusing
the memory for other parts of the compilation we should
reduce this overhead.

Change-Id: Ifbc00aab4f3afd0000da818dfe68b96713824a08
diff --git a/compiler/Android.mk b/compiler/Android.mk
index 499f23f..691cee0 100644
--- a/compiler/Android.mk
+++ b/compiler/Android.mk
@@ -86,6 +86,7 @@
 	utils/mips/managed_register_mips.cc \
 	utils/x86/assembler_x86.cc \
 	utils/x86/managed_register_x86.cc \
+	utils/scoped_arena_allocator.cc \
 	buffered_output_stream.cc \
 	compiler_backend.cc \
 	elf_fixup.cc \
diff --git a/compiler/dex/bit_vector_block_iterator.h b/compiler/dex/bit_vector_block_iterator.h
index 0821e9e..0f1c2b6 100644
--- a/compiler/dex/bit_vector_block_iterator.h
+++ b/compiler/dex/bit_vector_block_iterator.h
@@ -44,7 +44,7 @@
     BasicBlock* Next();
 
     void* operator new(size_t size, ArenaAllocator* arena) {
-      return arena->Alloc(size, ArenaAllocator::kAllocGrowableArray);
+      return arena->Alloc(size, kArenaAllocGrowableArray);
     };
     void operator delete(void* p) {}  // Nop.
 
diff --git a/compiler/dex/compiler_ir.h b/compiler/dex/compiler_ir.h
index ee88041..c71f047 100644
--- a/compiler/dex/compiler_ir.h
+++ b/compiler/dex/compiler_ir.h
@@ -25,6 +25,7 @@
 #include "driver/compiler_driver.h"
 #include "driver/dex_compilation_unit.h"
 #include "safe_map.h"
+#include "utils/scoped_arena_allocator.h"
 #include "base/timing_logger.h"
 #include "utils/arena_allocator.h"
 
@@ -82,6 +83,7 @@
 
   // TODO: move memory management to mir_graph, or just switch to using standard containers.
   ArenaAllocator arena;
+  ArenaStack arena_stack;  // Arenas for ScopedArenaAllocator.
 
   UniquePtr<MIRGraph> mir_graph;   // MIR container.
   UniquePtr<Backend> cg;           // Target-specific codegen.
diff --git a/compiler/dex/frontend.cc b/compiler/dex/frontend.cc
index b55b471..1c2d16f 100644
--- a/compiler/dex/frontend.cc
+++ b/compiler/dex/frontend.cc
@@ -98,6 +98,7 @@
     num_regs(0),
     compiler_flip_match(false),
     arena(pool),
+    arena_stack(pool),
     mir_graph(NULL),
     cg(NULL),
     timings("QuickCompiler", true, false) {
@@ -247,9 +248,12 @@
   }
 
   if (cu.enable_debug & (1 << kDebugShowMemoryUsage)) {
-    if (cu.arena.BytesAllocated() > (5 * 1024 *1024)) {
-      MemStats mem_stats(cu.arena);
-      LOG(INFO) << PrettyMethod(method_idx, dex_file) << " " << Dumpable<MemStats>(mem_stats);
+    if (cu.arena.BytesAllocated() > (1 * 1024 *1024) ||
+        cu.arena_stack.PeakBytesAllocated() > 256 * 1024) {
+      MemStats mem_stats(cu.arena.GetMemStats());
+      MemStats peak_stats(cu.arena_stack.GetPeakStats());
+      LOG(INFO) << PrettyMethod(method_idx, dex_file) << " " << Dumpable<MemStats>(mem_stats)
+          << Dumpable<MemStats>(peak_stats);
     }
   }
 
diff --git a/compiler/dex/local_value_numbering.h b/compiler/dex/local_value_numbering.h
index 348bedc..535b613 100644
--- a/compiler/dex/local_value_numbering.h
+++ b/compiler/dex/local_value_numbering.h
@@ -18,6 +18,8 @@
 #define ART_COMPILER_DEX_LOCAL_VALUE_NUMBERING_H_
 
 #include "compiler_internals.h"
+#include "UniquePtr.h"
+#include "utils/scoped_arena_allocator.h"
 
 #define NO_VALUE 0xffff
 #define ARRAY_REF 0xfffe
@@ -73,28 +75,26 @@
   };
 
   // Key is s_reg, value is value name.
-  typedef SafeMap<uint16_t, uint16_t> SregValueMap;
+  typedef SafeMap<uint16_t, uint16_t, std::less<uint16_t>,
+      ScopedArenaAllocatorAdapter<std::pair<uint16_t, uint16_t> > > SregValueMap;
   // Key is concatenation of opcode, operand1, operand2 and modifier, value is value name.
-  typedef SafeMap<uint64_t, uint16_t> ValueMap;
+  typedef SafeMap<uint64_t, uint16_t, std::less<uint64_t>,
+      ScopedArenaAllocatorAdapter<std::pair<uint64_t, uint16_t> > > ValueMap;
   // Key represents a memory address, value is generation.
-  typedef SafeMap<MemoryVersionKey, uint16_t, MemoryVersionKeyComparator> MemoryVersionMap;
+  typedef SafeMap<MemoryVersionKey, uint16_t, MemoryVersionKeyComparator,
+      ScopedArenaAllocatorAdapter<std::pair<MemoryVersionKey, uint16_t> > > MemoryVersionMap;
   // Maps field key to field id for resolved fields.
-  typedef SafeMap<FieldReference, uint32_t, FieldReferenceComparator> FieldIndexMap;
+  typedef SafeMap<FieldReference, uint32_t, FieldReferenceComparator,
+      ScopedArenaAllocatorAdapter<std::pair<FieldReference, uint16_t> > > FieldIndexMap;
+  // A set of value names.
+  typedef std::set<uint16_t, std::less<uint16_t>,
+      ScopedArenaAllocatorAdapter<uint16_t> > ValueNameSet;
 
  public:
-  explicit LocalValueNumbering(CompilationUnit* cu)
-      : cu_(cu),
-        sreg_value_map_(),
-        sreg_wide_value_map_(),
-        value_map_(),
-        next_memory_version_(1u),
-        global_memory_version_(0u),
-        memory_version_map_(),
-        field_index_map_(),
-        non_aliasing_refs_(),
-        null_checked_() {
-    std::fill_n(unresolved_sfield_version_, kFieldTypeCount, 0u);
-    std::fill_n(unresolved_ifield_version_, kFieldTypeCount, 0u);
+  static LocalValueNumbering* Create(CompilationUnit* cu) {
+    UniquePtr<ScopedArenaAllocator> allocator(ScopedArenaAllocator::Create(&cu->arena_stack));
+    void* addr = allocator->Alloc(sizeof(LocalValueNumbering), kArenaAllocMisc);
+    return new(addr) LocalValueNumbering(cu, allocator.release());
   }
 
   static uint64_t BuildKey(uint16_t op, uint16_t operand1, uint16_t operand2, uint16_t modifier) {
@@ -167,7 +167,26 @@
 
   uint16_t GetValueNumber(MIR* mir);
 
+  // Allow delete-expression to destroy a LocalValueNumbering object without deallocation.
+  static void operator delete(void* ptr) { UNUSED(ptr); }
+
  private:
+  LocalValueNumbering(CompilationUnit* cu, ScopedArenaAllocator* allocator)
+      : cu_(cu),
+        allocator_(allocator),
+        sreg_value_map_(std::less<uint16_t>(), allocator->Adapter()),
+        sreg_wide_value_map_(std::less<uint16_t>(), allocator->Adapter()),
+        value_map_(std::less<uint64_t>(), allocator->Adapter()),
+        next_memory_version_(1u),
+        global_memory_version_(0u),
+        memory_version_map_(MemoryVersionKeyComparator(), allocator->Adapter()),
+        field_index_map_(FieldReferenceComparator(), allocator->Adapter()),
+        non_aliasing_refs_(std::less<uint16_t>(), allocator->Adapter()),
+        null_checked_(std::less<uint16_t>(), allocator->Adapter()) {
+    std::fill_n(unresolved_sfield_version_, kFieldTypeCount, 0u);
+    std::fill_n(unresolved_ifield_version_, kFieldTypeCount, 0u);
+  }
+
   uint16_t GetFieldId(const DexFile* dex_file, uint16_t field_idx);
   void AdvanceGlobalMemory();
   uint16_t GetMemoryVersion(uint16_t base, uint16_t field, uint16_t type);
@@ -179,6 +198,7 @@
   void HandlePutObject(MIR* mir);
 
   CompilationUnit* const cu_;
+  UniquePtr<ScopedArenaAllocator> allocator_;
   SregValueMap sreg_value_map_;
   SregValueMap sreg_wide_value_map_;
   ValueMap value_map_;
@@ -189,8 +209,10 @@
   MemoryVersionMap memory_version_map_;
   FieldIndexMap field_index_map_;
   // Value names of references to objects that cannot be reached through a different value name.
-  std::set<uint16_t> non_aliasing_refs_;
-  std::set<uint16_t> null_checked_;
+  ValueNameSet non_aliasing_refs_;
+  ValueNameSet null_checked_;
+
+  DISALLOW_COPY_AND_ASSIGN(LocalValueNumbering);
 };
 
 }  // namespace art
diff --git a/compiler/dex/local_value_numbering_test.cc b/compiler/dex/local_value_numbering_test.cc
index 4599612..ebac871 100644
--- a/compiler/dex/local_value_numbering_test.cc
+++ b/compiler/dex/local_value_numbering_test.cc
@@ -120,7 +120,7 @@
 
   void DoPrepareMIRs(const MIRDef* defs, size_t count) {
     mir_count_ = count;
-    mirs_ = reinterpret_cast<MIR*>(cu_.arena.Alloc(sizeof(MIR) * count, ArenaAllocator::kAllocMIR));
+    mirs_ = reinterpret_cast<MIR*>(cu_.arena.Alloc(sizeof(MIR) * count, kArenaAllocMIR));
     ssa_reps_.resize(count);
     for (size_t i = 0u; i != count; ++i) {
       const MIRDef* def = &defs[i];
@@ -162,11 +162,16 @@
   void PerformLVN() {
     value_names_.resize(mir_count_);
     for (size_t i = 0; i != mir_count_; ++i) {
-      value_names_[i] =  lvn_.GetValueNumber(&mirs_[i]);
+      value_names_[i] =  lvn_->GetValueNumber(&mirs_[i]);
     }
   }
 
-  LocalValueNumberingTest() : pool_(), cu_(&pool_), mir_count_(0u), mirs_(nullptr), lvn_(&cu_) {
+  LocalValueNumberingTest()
+      : pool_(),
+        cu_(&pool_),
+        mir_count_(0u),
+        mirs_(nullptr),
+        lvn_(LocalValueNumbering::Create(&cu_)) {
     cu_.mir_graph.reset(new MIRGraph(&cu_, &cu_.arena));
   }
 
@@ -176,7 +181,7 @@
   MIR* mirs_;
   std::vector<SSARepresentation> ssa_reps_;
   std::vector<uint16_t> value_names_;
-  LocalValueNumbering lvn_;
+  UniquePtr<LocalValueNumbering> lvn_;
 };
 
 TEST_F(LocalValueNumberingTest, TestIGetIGetInvokeIGet) {
diff --git a/compiler/dex/mir_dataflow.cc b/compiler/dex/mir_dataflow.cc
index 9680450..c3954fe 100644
--- a/compiler/dex/mir_dataflow.cc
+++ b/compiler/dex/mir_dataflow.cc
@@ -955,10 +955,10 @@
 
   mir->ssa_rep->num_uses = num_uses;
   mir->ssa_rep->uses = static_cast<int*>(arena_->Alloc(sizeof(int) * num_uses,
-                                                       ArenaAllocator::kAllocDFInfo));
+                                                       kArenaAllocDFInfo));
   // NOTE: will be filled in during type & size inference pass
   mir->ssa_rep->fp_use = static_cast<bool*>(arena_->Alloc(sizeof(bool) * num_uses,
-                                                          ArenaAllocator::kAllocDFInfo));
+                                                          kArenaAllocDFInfo));
 
   for (i = 0; i < num_uses; i++) {
     HandleSSAUse(mir->ssa_rep->uses, d_insn->arg[i], i);
@@ -973,10 +973,10 @@
 
   mir->ssa_rep->num_uses = num_uses;
   mir->ssa_rep->uses = static_cast<int*>(arena_->Alloc(sizeof(int) * num_uses,
-                                                       ArenaAllocator::kAllocDFInfo));
+                                                       kArenaAllocDFInfo));
   // NOTE: will be filled in during type & size inference pass
   mir->ssa_rep->fp_use = static_cast<bool*>(arena_->Alloc(sizeof(bool) * num_uses,
-                                                          ArenaAllocator::kAllocDFInfo));
+                                                          kArenaAllocDFInfo));
 
   for (i = 0; i < num_uses; i++) {
     HandleSSAUse(mir->ssa_rep->uses, d_insn->vC+i, i);
@@ -992,7 +992,7 @@
   for (mir = bb->first_mir_insn; mir != NULL; mir = mir->next) {
     mir->ssa_rep =
         static_cast<struct SSARepresentation *>(arena_->Alloc(sizeof(SSARepresentation),
-                                                              ArenaAllocator::kAllocDFInfo));
+                                                              kArenaAllocDFInfo));
 
     uint64_t df_attributes = oat_data_flow_attributes_[mir->dalvikInsn.opcode];
 
@@ -1042,9 +1042,9 @@
     if (num_uses) {
       mir->ssa_rep->num_uses = num_uses;
       mir->ssa_rep->uses = static_cast<int*>(arena_->Alloc(sizeof(int) * num_uses,
-                                                           ArenaAllocator::kAllocDFInfo));
+                                                           kArenaAllocDFInfo));
       mir->ssa_rep->fp_use = static_cast<bool*>(arena_->Alloc(sizeof(bool) * num_uses,
-                                                              ArenaAllocator::kAllocDFInfo));
+                                                              kArenaAllocDFInfo));
     }
 
     int num_defs = 0;
@@ -1059,9 +1059,9 @@
     if (num_defs) {
       mir->ssa_rep->num_defs = num_defs;
       mir->ssa_rep->defs = static_cast<int*>(arena_->Alloc(sizeof(int) * num_defs,
-                                                           ArenaAllocator::kAllocDFInfo));
+                                                           kArenaAllocDFInfo));
       mir->ssa_rep->fp_def = static_cast<bool*>(arena_->Alloc(sizeof(bool) * num_defs,
-                                                              ArenaAllocator::kAllocDFInfo));
+                                                              kArenaAllocDFInfo));
     }
 
     DecodedInstruction *d_insn = &mir->dalvikInsn;
@@ -1110,7 +1110,7 @@
    */
   bb->data_flow_info->vreg_to_ssa_map =
       static_cast<int*>(arena_->Alloc(sizeof(int) * cu_->num_dalvik_registers,
-                                      ArenaAllocator::kAllocDFInfo));
+                                      kArenaAllocDFInfo));
 
   memcpy(bb->data_flow_info->vreg_to_ssa_map, vreg_to_ssa_map_,
          sizeof(int) * cu_->num_dalvik_registers);
@@ -1147,11 +1147,11 @@
    */
   vreg_to_ssa_map_ =
       static_cast<int*>(arena_->Alloc(sizeof(int) * num_dalvik_reg,
-                                      ArenaAllocator::kAllocDFInfo));
+                                      kArenaAllocDFInfo));
   /* Keep track of the higest def for each dalvik reg */
   ssa_last_defs_ =
       static_cast<int*>(arena_->Alloc(sizeof(int) * num_dalvik_reg,
-                                      ArenaAllocator::kAllocDFInfo));
+                                      kArenaAllocDFInfo));
 
   for (unsigned int i = 0; i < num_dalvik_reg; i++) {
     vreg_to_ssa_map_[i] = i;
@@ -1175,7 +1175,7 @@
       bb->block_type == kExitBlock) {
       bb->data_flow_info =
           static_cast<BasicBlockDataFlow*>(arena_->Alloc(sizeof(BasicBlockDataFlow),
-                                                         ArenaAllocator::kAllocDFInfo));
+                                                         kArenaAllocDFInfo));
       }
   }
 }
diff --git a/compiler/dex/mir_graph.cc b/compiler/dex/mir_graph.cc
index 46e854f..868730f 100644
--- a/compiler/dex/mir_graph.cc
+++ b/compiler/dex/mir_graph.cc
@@ -411,7 +411,7 @@
                                       /* create */ true, /* immed_pred_block_p */ &cur_block);
     SuccessorBlockInfo *successor_block_info =
         static_cast<SuccessorBlockInfo*>(arena_->Alloc(sizeof(SuccessorBlockInfo),
-                                                       ArenaAllocator::kAllocSuccessor));
+                                                       kArenaAllocSuccessor));
     successor_block_info->block = case_block->id;
     successor_block_info->key =
         (insn->dalvikInsn.opcode == Instruction::PACKED_SWITCH) ?
@@ -459,7 +459,7 @@
         catches_.insert(catch_block->start_offset);
       }
       SuccessorBlockInfo *successor_block_info = reinterpret_cast<SuccessorBlockInfo*>
-          (arena_->Alloc(sizeof(SuccessorBlockInfo), ArenaAllocator::kAllocSuccessor));
+          (arena_->Alloc(sizeof(SuccessorBlockInfo), kArenaAllocSuccessor));
       successor_block_info->block = catch_block->id;
       successor_block_info->key = iterator.GetHandlerTypeIndex();
       cur_block->successor_blocks->Insert(successor_block_info);
@@ -518,7 +518,7 @@
   new_block->start_offset = insn->offset;
   cur_block->fall_through = new_block->id;
   new_block->predecessors->Insert(cur_block->id);
-  MIR* new_insn = static_cast<MIR*>(arena_->Alloc(sizeof(MIR), ArenaAllocator::kAllocMIR));
+  MIR* new_insn = static_cast<MIR*>(arena_->Alloc(sizeof(MIR), kArenaAllocMIR));
   *new_insn = *insn;
   insn->dalvikInsn.opcode =
       static_cast<Instruction::Code>(kMirOpCheck);
@@ -602,7 +602,7 @@
 
   /* Parse all instructions and put them into containing basic blocks */
   while (code_ptr < code_end) {
-    MIR *insn = static_cast<MIR *>(arena_->Alloc(sizeof(MIR), ArenaAllocator::kAllocMIR));
+    MIR *insn = static_cast<MIR *>(arena_->Alloc(sizeof(MIR), kArenaAllocMIR));
     insn->offset = current_offset_;
     insn->m_unit_index = current_method_;
     int width = ParseInsn(code_ptr, &insn->dalvikInsn);
@@ -1042,7 +1042,7 @@
     str.append("]--optimized away");
   }
   int length = str.length() + 1;
-  ret = static_cast<char*>(arena_->Alloc(length, ArenaAllocator::kAllocDFInfo));
+  ret = static_cast<char*>(arena_->Alloc(length, kArenaAllocDFInfo));
   strncpy(ret, str.c_str(), length);
   return ret;
 }
@@ -1157,7 +1157,7 @@
 CallInfo* MIRGraph::NewMemCallInfo(BasicBlock* bb, MIR* mir, InvokeType type,
                                   bool is_range) {
   CallInfo* info = static_cast<CallInfo*>(arena_->Alloc(sizeof(CallInfo),
-                                                        ArenaAllocator::kAllocMisc));
+                                                        kArenaAllocMisc));
   MIR* move_result_mir = FindMoveResult(bb, mir);
   if (move_result_mir == NULL) {
     info->result.location = kLocInvalid;
@@ -1167,7 +1167,7 @@
   }
   info->num_arg_words = mir->ssa_rep->num_uses;
   info->args = (info->num_arg_words == 0) ? NULL : static_cast<RegLocation*>
-      (arena_->Alloc(sizeof(RegLocation) * info->num_arg_words, ArenaAllocator::kAllocMisc));
+      (arena_->Alloc(sizeof(RegLocation) * info->num_arg_words, kArenaAllocMisc));
   for (int i = 0; i < info->num_arg_words; i++) {
     info->args[i] = GetRawSrc(mir, i);
   }
@@ -1182,7 +1182,7 @@
 // Allocate a new basic block.
 BasicBlock* MIRGraph::NewMemBB(BBType block_type, int block_id) {
   BasicBlock* bb = static_cast<BasicBlock*>(arena_->Alloc(sizeof(BasicBlock),
-                                                          ArenaAllocator::kAllocBB));
+                                                          kArenaAllocBB));
   bb->block_type = block_type;
   bb->id = block_id;
   // TUNING: better estimate of the exit block predecessors?
@@ -1196,7 +1196,7 @@
 
 void MIRGraph::InitializeConstantPropagation() {
   is_constant_v_ = new (arena_) ArenaBitVector(arena_, GetNumSSARegs(), false);
-  constant_values_ = static_cast<int*>(arena_->Alloc(sizeof(int) * GetNumSSARegs(), ArenaAllocator::kAllocDFInfo));
+  constant_values_ = static_cast<int*>(arena_->Alloc(sizeof(int) * GetNumSSARegs(), kArenaAllocDFInfo));
 }
 
 void MIRGraph::InitializeMethodUses() {
diff --git a/compiler/dex/mir_graph.h b/compiler/dex/mir_graph.h
index 1eb9ef9..85d6d89 100644
--- a/compiler/dex/mir_graph.h
+++ b/compiler/dex/mir_graph.h
@@ -457,7 +457,7 @@
 
   void EnableOpcodeCounting() {
     opcode_count_ = static_cast<int*>(arena_->Alloc(kNumPackedOpcodes * sizeof(int),
-                                                    ArenaAllocator::kAllocMisc));
+                                                    kArenaAllocMisc));
   }
 
   void ShowOpcodeStats();
diff --git a/compiler/dex/mir_optimization.cc b/compiler/dex/mir_optimization.cc
index 243452e..03fc091 100644
--- a/compiler/dex/mir_optimization.cc
+++ b/compiler/dex/mir_optimization.cc
@@ -245,7 +245,7 @@
   }
 
   CompilerTemp *compiler_temp = static_cast<CompilerTemp *>(arena_->Alloc(sizeof(CompilerTemp),
-                                                            ArenaAllocator::kAllocRegAlloc));
+                                                            kArenaAllocRegAlloc));
 
   // Create the type of temp requested. Special temps need special handling because
   // they have a specific virtual register assignment.
@@ -313,7 +313,7 @@
   bool use_lvn = bb->use_lvn;
   UniquePtr<LocalValueNumbering> local_valnum;
   if (use_lvn) {
-    local_valnum.reset(new LocalValueNumbering(cu_));
+    local_valnum.reset(LocalValueNumbering::Create(cu_));
   }
   while (bb != NULL) {
     for (MIR* mir = bb->first_mir_insn; mir != NULL; mir = mir->next) {
@@ -479,7 +479,7 @@
                 DCHECK_EQ(SelectKind(if_true), kSelectMove);
                 DCHECK_EQ(SelectKind(if_false), kSelectMove);
                 int* src_ssa =
-                    static_cast<int*>(arena_->Alloc(sizeof(int) * 3, ArenaAllocator::kAllocDFInfo));
+                    static_cast<int*>(arena_->Alloc(sizeof(int) * 3, kArenaAllocDFInfo));
                 src_ssa[0] = mir->ssa_rep->uses[0];
                 src_ssa[1] = if_true->ssa_rep->uses[0];
                 src_ssa[2] = if_false->ssa_rep->uses[0];
@@ -488,14 +488,14 @@
               }
               mir->ssa_rep->num_defs = 1;
               mir->ssa_rep->defs =
-                  static_cast<int*>(arena_->Alloc(sizeof(int) * 1, ArenaAllocator::kAllocDFInfo));
+                  static_cast<int*>(arena_->Alloc(sizeof(int) * 1, kArenaAllocDFInfo));
               mir->ssa_rep->fp_def =
-                  static_cast<bool*>(arena_->Alloc(sizeof(bool) * 1, ArenaAllocator::kAllocDFInfo));
+                  static_cast<bool*>(arena_->Alloc(sizeof(bool) * 1, kArenaAllocDFInfo));
               mir->ssa_rep->fp_def[0] = if_true->ssa_rep->fp_def[0];
               // Match type of uses to def.
               mir->ssa_rep->fp_use =
                   static_cast<bool*>(arena_->Alloc(sizeof(bool) * mir->ssa_rep->num_uses,
-                                                   ArenaAllocator::kAllocDFInfo));
+                                                   kArenaAllocDFInfo));
               for (int i = 0; i < mir->ssa_rep->num_uses; i++) {
                 mir->ssa_rep->fp_use[i] = mir->ssa_rep->fp_def[0];
               }
@@ -878,7 +878,7 @@
 
 void MIRGraph::DumpCheckStats() {
   Checkstats* stats =
-      static_cast<Checkstats*>(arena_->Alloc(sizeof(Checkstats), ArenaAllocator::kAllocDFInfo));
+      static_cast<Checkstats*>(arena_->Alloc(sizeof(Checkstats), kArenaAllocDFInfo));
   checkstats_ = stats;
   AllNodesIterator iter(this);
   for (BasicBlock* bb = iter.Next(); bb != NULL; bb = iter.Next()) {
diff --git a/compiler/dex/quick/arm/call_arm.cc b/compiler/dex/quick/arm/call_arm.cc
index f426055..0fce5bb 100644
--- a/compiler/dex/quick/arm/call_arm.cc
+++ b/compiler/dex/quick/arm/call_arm.cc
@@ -50,12 +50,12 @@
   }
   // Add the table to the list - we'll process it later
   SwitchTable *tab_rec =
-      static_cast<SwitchTable*>(arena_->Alloc(sizeof(SwitchTable), ArenaAllocator::kAllocData));
+      static_cast<SwitchTable*>(arena_->Alloc(sizeof(SwitchTable), kArenaAllocData));
   tab_rec->table = table;
   tab_rec->vaddr = current_dalvik_offset_;
   uint32_t size = table[1];
   tab_rec->targets = static_cast<LIR**>(arena_->Alloc(size * sizeof(LIR*),
-                                                     ArenaAllocator::kAllocLIR));
+                                                     kArenaAllocLIR));
   switch_tables_.Insert(tab_rec);
 
   // Get the switch value
@@ -99,12 +99,12 @@
   }
   // Add the table to the list - we'll process it later
   SwitchTable *tab_rec =
-      static_cast<SwitchTable*>(arena_->Alloc(sizeof(SwitchTable),  ArenaAllocator::kAllocData));
+      static_cast<SwitchTable*>(arena_->Alloc(sizeof(SwitchTable),  kArenaAllocData));
   tab_rec->table = table;
   tab_rec->vaddr = current_dalvik_offset_;
   uint32_t size = table[1];
   tab_rec->targets =
-      static_cast<LIR**>(arena_->Alloc(size * sizeof(LIR*), ArenaAllocator::kAllocLIR));
+      static_cast<LIR**>(arena_->Alloc(size * sizeof(LIR*), kArenaAllocLIR));
   switch_tables_.Insert(tab_rec);
 
   // Get the switch value
@@ -152,7 +152,7 @@
   const uint16_t* table = cu_->insns + current_dalvik_offset_ + table_offset;
   // Add the table to the list - we'll process it later
   FillArrayData *tab_rec =
-      static_cast<FillArrayData*>(arena_->Alloc(sizeof(FillArrayData), ArenaAllocator::kAllocData));
+      static_cast<FillArrayData*>(arena_->Alloc(sizeof(FillArrayData), kArenaAllocData));
   tab_rec->table = table;
   tab_rec->vaddr = current_dalvik_offset_;
   uint16_t width = tab_rec->table[1];
diff --git a/compiler/dex/quick/arm/target_arm.cc b/compiler/dex/quick/arm/target_arm.cc
index ab1a053..01d669b 100644
--- a/compiler/dex/quick/arm/target_arm.cc
+++ b/compiler/dex/quick/arm/target_arm.cc
@@ -554,13 +554,13 @@
   int num_fp_regs = sizeof(FpRegs)/sizeof(*FpRegs);
   int num_fp_temps = sizeof(fp_temps)/sizeof(*fp_temps);
   reg_pool_ = static_cast<RegisterPool*>(arena_->Alloc(sizeof(*reg_pool_),
-                                                       ArenaAllocator::kAllocRegAlloc));
+                                                       kArenaAllocRegAlloc));
   reg_pool_->num_core_regs = num_regs;
   reg_pool_->core_regs = reinterpret_cast<RegisterInfo*>
-      (arena_->Alloc(num_regs * sizeof(*reg_pool_->core_regs), ArenaAllocator::kAllocRegAlloc));
+      (arena_->Alloc(num_regs * sizeof(*reg_pool_->core_regs), kArenaAllocRegAlloc));
   reg_pool_->num_fp_regs = num_fp_regs;
   reg_pool_->FPRegs = static_cast<RegisterInfo*>
-      (arena_->Alloc(num_fp_regs * sizeof(*reg_pool_->FPRegs), ArenaAllocator::kAllocRegAlloc));
+      (arena_->Alloc(num_fp_regs * sizeof(*reg_pool_->FPRegs), kArenaAllocRegAlloc));
   CompilerInitPool(reg_pool_->core_regs, core_regs, reg_pool_->num_core_regs);
   CompilerInitPool(reg_pool_->FPRegs, FpRegs, reg_pool_->num_fp_regs);
   // Keep special registers from being allocated
diff --git a/compiler/dex/quick/codegen_util.cc b/compiler/dex/quick/codegen_util.cc
index 14469b6..34d3834 100644
--- a/compiler/dex/quick/codegen_util.cc
+++ b/compiler/dex/quick/codegen_util.cc
@@ -358,7 +358,7 @@
 LIR* Mir2Lir::AddWordData(LIR* *constant_list_p, int value) {
   /* Add the constant to the literal pool */
   if (constant_list_p) {
-    LIR* new_value = static_cast<LIR*>(arena_->Alloc(sizeof(LIR), ArenaAllocator::kAllocData));
+    LIR* new_value = static_cast<LIR*>(arena_->Alloc(sizeof(LIR), kArenaAllocData));
     new_value->operands[0] = value;
     new_value->next = *constant_list_p;
     *constant_list_p = new_value;
@@ -829,7 +829,7 @@
   LIR* res = boundary_lir;
   if (cu_->verbose) {
     // Only pay the expense if we're pretty-printing.
-    LIR* new_label = static_cast<LIR*>(arena_->Alloc(sizeof(LIR), ArenaAllocator::kAllocLIR));
+    LIR* new_label = static_cast<LIR*>(arena_->Alloc(sizeof(LIR), kArenaAllocLIR));
     new_label->dalvik_offset = vaddr;
     new_label->opcode = kPseudoCaseLabel;
     new_label->operands[0] = keyVal;
diff --git a/compiler/dex/quick/local_optimizations.cc b/compiler/dex/quick/local_optimizations.cc
index 7a2dce1..6df91e6 100644
--- a/compiler/dex/quick/local_optimizations.cc
+++ b/compiler/dex/quick/local_optimizations.cc
@@ -248,7 +248,7 @@
         /* Only sink store instructions */
         if (sink_distance && !is_this_lir_load) {
           LIR* new_store_lir =
-              static_cast<LIR*>(arena_->Alloc(sizeof(LIR), ArenaAllocator::kAllocLIR));
+              static_cast<LIR*>(arena_->Alloc(sizeof(LIR), kArenaAllocLIR));
           *new_store_lir = *this_lir;
           /*
            * Stop point found - insert *before* the check_lir
@@ -445,7 +445,7 @@
       if (slot >= 0) {
         LIR* cur_lir = prev_inst_list[slot];
         LIR* new_load_lir =
-          static_cast<LIR*>(arena_->Alloc(sizeof(LIR), ArenaAllocator::kAllocLIR));
+          static_cast<LIR*>(arena_->Alloc(sizeof(LIR), kArenaAllocLIR));
         *new_load_lir = *this_lir;
         /*
          * Insertion is guaranteed to succeed since check_lir
diff --git a/compiler/dex/quick/mips/call_mips.cc b/compiler/dex/quick/mips/call_mips.cc
index 88f46fd..234299e 100644
--- a/compiler/dex/quick/mips/call_mips.cc
+++ b/compiler/dex/quick/mips/call_mips.cc
@@ -68,12 +68,12 @@
   }
   // Add the table to the list - we'll process it later
   SwitchTable* tab_rec =
-      static_cast<SwitchTable*>(arena_->Alloc(sizeof(SwitchTable), ArenaAllocator::kAllocData));
+      static_cast<SwitchTable*>(arena_->Alloc(sizeof(SwitchTable), kArenaAllocData));
   tab_rec->table = table;
   tab_rec->vaddr = current_dalvik_offset_;
   int elements = table[1];
   tab_rec->targets =
-      static_cast<LIR**>(arena_->Alloc(elements * sizeof(LIR*), ArenaAllocator::kAllocLIR));
+      static_cast<LIR**>(arena_->Alloc(elements * sizeof(LIR*), kArenaAllocLIR));
   switch_tables_.Insert(tab_rec);
 
   // The table is composed of 8-byte key/disp pairs
@@ -146,12 +146,12 @@
   }
   // Add the table to the list - we'll process it later
   SwitchTable* tab_rec =
-      static_cast<SwitchTable*>(arena_->Alloc(sizeof(SwitchTable), ArenaAllocator::kAllocData));
+      static_cast<SwitchTable*>(arena_->Alloc(sizeof(SwitchTable), kArenaAllocData));
   tab_rec->table = table;
   tab_rec->vaddr = current_dalvik_offset_;
   int size = table[1];
   tab_rec->targets = static_cast<LIR**>(arena_->Alloc(size * sizeof(LIR*),
-                                                      ArenaAllocator::kAllocLIR));
+                                                      kArenaAllocLIR));
   switch_tables_.Insert(tab_rec);
 
   // Get the switch value
@@ -226,7 +226,7 @@
   // Add the table to the list - we'll process it later
   FillArrayData* tab_rec =
       reinterpret_cast<FillArrayData*>(arena_->Alloc(sizeof(FillArrayData),
-                                                     ArenaAllocator::kAllocData));
+                                                     kArenaAllocData));
   tab_rec->table = table;
   tab_rec->vaddr = current_dalvik_offset_;
   uint16_t width = tab_rec->table[1];
diff --git a/compiler/dex/quick/mips/target_mips.cc b/compiler/dex/quick/mips/target_mips.cc
index 85c250d..4f495ee 100644
--- a/compiler/dex/quick/mips/target_mips.cc
+++ b/compiler/dex/quick/mips/target_mips.cc
@@ -467,13 +467,13 @@
   int num_fp_regs = sizeof(FpRegs)/sizeof(*FpRegs);
   int num_fp_temps = sizeof(fp_temps)/sizeof(*fp_temps);
   reg_pool_ = static_cast<RegisterPool*>(arena_->Alloc(sizeof(*reg_pool_),
-                                                       ArenaAllocator::kAllocRegAlloc));
+                                                       kArenaAllocRegAlloc));
   reg_pool_->num_core_regs = num_regs;
   reg_pool_->core_regs = static_cast<RegisterInfo*>
-     (arena_->Alloc(num_regs * sizeof(*reg_pool_->core_regs), ArenaAllocator::kAllocRegAlloc));
+     (arena_->Alloc(num_regs * sizeof(*reg_pool_->core_regs), kArenaAllocRegAlloc));
   reg_pool_->num_fp_regs = num_fp_regs;
   reg_pool_->FPRegs = static_cast<RegisterInfo*>
-      (arena_->Alloc(num_fp_regs * sizeof(*reg_pool_->FPRegs), ArenaAllocator::kAllocRegAlloc));
+      (arena_->Alloc(num_fp_regs * sizeof(*reg_pool_->FPRegs), kArenaAllocRegAlloc));
   CompilerInitPool(reg_pool_->core_regs, core_regs, reg_pool_->num_core_regs);
   CompilerInitPool(reg_pool_->FPRegs, FpRegs, reg_pool_->num_fp_regs);
   // Keep special registers from being allocated
diff --git a/compiler/dex/quick/mir_to_lir-inl.h b/compiler/dex/quick/mir_to_lir-inl.h
index c2d12f6..8b1f81d 100644
--- a/compiler/dex/quick/mir_to_lir-inl.h
+++ b/compiler/dex/quick/mir_to_lir-inl.h
@@ -45,7 +45,7 @@
 
 inline LIR* Mir2Lir::RawLIR(DexOffset dalvik_offset, int opcode, int op0,
                             int op1, int op2, int op3, int op4, LIR* target) {
-  LIR* insn = static_cast<LIR*>(arena_->Alloc(sizeof(LIR), ArenaAllocator::kAllocLIR));
+  LIR* insn = static_cast<LIR*>(arena_->Alloc(sizeof(LIR), kArenaAllocLIR));
   insn->dalvik_offset = dalvik_offset;
   insn->opcode = opcode;
   insn->operands[0] = op0;
diff --git a/compiler/dex/quick/mir_to_lir.cc b/compiler/dex/quick/mir_to_lir.cc
index d9b241e..40ed5ef 100644
--- a/compiler/dex/quick/mir_to_lir.cc
+++ b/compiler/dex/quick/mir_to_lir.cc
@@ -1066,7 +1066,7 @@
   // Hold the labels of each block.
   block_label_list_ =
       static_cast<LIR*>(arena_->Alloc(sizeof(LIR) * mir_graph_->GetNumBlocks(),
-                                      ArenaAllocator::kAllocLIR));
+                                      kArenaAllocLIR));
 
   PreOrderDfsIterator iter(mir_graph_);
   BasicBlock* curr_bb = iter.Next();
diff --git a/compiler/dex/quick/mir_to_lir.h b/compiler/dex/quick/mir_to_lir.h
index 9e0e299..6955577 100644
--- a/compiler/dex/quick/mir_to_lir.h
+++ b/compiler/dex/quick/mir_to_lir.h
@@ -311,7 +311,7 @@
       virtual void Compile() = 0;
 
       static void* operator new(size_t size, ArenaAllocator* arena) {
-        return arena->Alloc(size, ArenaAllocator::kAllocData);
+        return arena->Alloc(size, kArenaAllocData);
       }
 
      protected:
@@ -363,7 +363,7 @@
     // strdup(), but allocates from the arena.
     char* ArenaStrdup(const char* str) {
       size_t len = strlen(str) + 1;
-      char* res = reinterpret_cast<char*>(arena_->Alloc(len, ArenaAllocator::kAllocMisc));
+      char* res = reinterpret_cast<char*>(arena_->Alloc(len, kArenaAllocMisc));
       if (res != NULL) {
         strncpy(res, str, len);
       }
diff --git a/compiler/dex/quick/ralloc_util.cc b/compiler/dex/quick/ralloc_util.cc
index 3a8942e..3cb6fd0 100644
--- a/compiler/dex/quick/ralloc_util.cc
+++ b/compiler/dex/quick/ralloc_util.cc
@@ -907,7 +907,7 @@
   const int promotion_threshold = 1;
   // Allocate the promotion map - one entry for each Dalvik vReg or compiler temp
   promotion_map_ = static_cast<PromotionMap*>
-      (arena_->Alloc(num_regs * sizeof(promotion_map_[0]), ArenaAllocator::kAllocRegAlloc));
+      (arena_->Alloc(num_regs * sizeof(promotion_map_[0]), kArenaAllocRegAlloc));
 
   // Allow target code to add any special registers
   AdjustSpillMask();
@@ -925,10 +925,10 @@
    */
   RefCounts *core_regs =
       static_cast<RefCounts*>(arena_->Alloc(sizeof(RefCounts) * num_regs,
-                                            ArenaAllocator::kAllocRegAlloc));
+                                            kArenaAllocRegAlloc));
   RefCounts *FpRegs =
       static_cast<RefCounts *>(arena_->Alloc(sizeof(RefCounts) * num_regs * 2,
-                                             ArenaAllocator::kAllocRegAlloc));
+                                             kArenaAllocRegAlloc));
   // Set ssa names for original Dalvik registers
   for (int i = 0; i < dalvik_regs; i++) {
     core_regs[i].s_reg = FpRegs[i].s_reg = i;
diff --git a/compiler/dex/quick/x86/call_x86.cc b/compiler/dex/quick/x86/call_x86.cc
index c92d2bb..577f216 100644
--- a/compiler/dex/quick/x86/call_x86.cc
+++ b/compiler/dex/quick/x86/call_x86.cc
@@ -69,12 +69,12 @@
   }
   // Add the table to the list - we'll process it later
   SwitchTable* tab_rec =
-      static_cast<SwitchTable*>(arena_->Alloc(sizeof(SwitchTable), ArenaAllocator::kAllocData));
+      static_cast<SwitchTable*>(arena_->Alloc(sizeof(SwitchTable), kArenaAllocData));
   tab_rec->table = table;
   tab_rec->vaddr = current_dalvik_offset_;
   int size = table[1];
   tab_rec->targets = static_cast<LIR**>(arena_->Alloc(size * sizeof(LIR*),
-                                                      ArenaAllocator::kAllocLIR));
+                                                      kArenaAllocLIR));
   switch_tables_.Insert(tab_rec);
 
   // Get the switch value
@@ -134,7 +134,7 @@
   const uint16_t* table = cu_->insns + current_dalvik_offset_ + table_offset;
   // Add the table to the list - we'll process it later
   FillArrayData* tab_rec =
-      static_cast<FillArrayData*>(arena_->Alloc(sizeof(FillArrayData), ArenaAllocator::kAllocData));
+      static_cast<FillArrayData*>(arena_->Alloc(sizeof(FillArrayData), kArenaAllocData));
   tab_rec->table = table;
   tab_rec->vaddr = current_dalvik_offset_;
   uint16_t width = tab_rec->table[1];
diff --git a/compiler/dex/quick/x86/target_x86.cc b/compiler/dex/quick/x86/target_x86.cc
index 78a2169..083fccb 100644
--- a/compiler/dex/quick/x86/target_x86.cc
+++ b/compiler/dex/quick/x86/target_x86.cc
@@ -457,15 +457,15 @@
   int num_fp_regs = sizeof(FpRegs)/sizeof(*FpRegs);
   int num_fp_temps = sizeof(fp_temps)/sizeof(*fp_temps);
   reg_pool_ = static_cast<RegisterPool*>(arena_->Alloc(sizeof(*reg_pool_),
-                                                       ArenaAllocator::kAllocRegAlloc));
+                                                       kArenaAllocRegAlloc));
   reg_pool_->num_core_regs = num_regs;
   reg_pool_->core_regs =
       static_cast<RegisterInfo*>(arena_->Alloc(num_regs * sizeof(*reg_pool_->core_regs),
-                                               ArenaAllocator::kAllocRegAlloc));
+                                               kArenaAllocRegAlloc));
   reg_pool_->num_fp_regs = num_fp_regs;
   reg_pool_->FPRegs =
       static_cast<RegisterInfo *>(arena_->Alloc(num_fp_regs * sizeof(*reg_pool_->FPRegs),
-                                                ArenaAllocator::kAllocRegAlloc));
+                                                kArenaAllocRegAlloc));
   CompilerInitPool(reg_pool_->core_regs, core_regs, reg_pool_->num_core_regs);
   CompilerInitPool(reg_pool_->FPRegs, FpRegs, reg_pool_->num_fp_regs);
   // Keep special registers from being allocated
diff --git a/compiler/dex/ssa_transformation.cc b/compiler/dex/ssa_transformation.cc
index 4e258ef..8091528 100644
--- a/compiler/dex/ssa_transformation.cc
+++ b/compiler/dex/ssa_transformation.cc
@@ -144,7 +144,7 @@
   /* Allocate num_dalvik_registers bit vector pointers */
   def_block_matrix_ = static_cast<ArenaBitVector**>
       (arena_->Alloc(sizeof(ArenaBitVector *) * num_registers,
-                     ArenaAllocator::kAllocDFInfo));
+                     kArenaAllocDFInfo));
   int i;
 
   /* Initialize num_register vectors with num_blocks bits each */
@@ -384,7 +384,7 @@
   /* Initalize & Clear i_dom_list */
   if (i_dom_list_ == NULL) {
     i_dom_list_ = static_cast<int*>(arena_->Alloc(sizeof(int) * num_reachable_blocks,
-                                                  ArenaAllocator::kAllocDFInfo));
+                                                  kArenaAllocDFInfo));
   }
   for (int i = 0; i < num_reachable_blocks; i++) {
     i_dom_list_[i] = NOTVISITED;
@@ -565,7 +565,7 @@
         continue;
       }
       MIR *phi =
-          static_cast<MIR*>(arena_->Alloc(sizeof(MIR), ArenaAllocator::kAllocDFInfo));
+          static_cast<MIR*>(arena_->Alloc(sizeof(MIR), kArenaAllocDFInfo));
       phi->dalvikInsn.opcode = static_cast<Instruction::Code>(kMirOpPhi);
       phi->dalvikInsn.vA = dalvik_reg;
       phi->offset = phi_bb->start_offset;
@@ -593,13 +593,13 @@
     size_t num_uses = bb->predecessors->Size();
     mir->ssa_rep->num_uses = num_uses;
     int* uses = static_cast<int*>(arena_->Alloc(sizeof(int) * num_uses,
-                                                ArenaAllocator::kAllocDFInfo));
+                                                kArenaAllocDFInfo));
     mir->ssa_rep->uses = uses;
     mir->ssa_rep->fp_use =
-        static_cast<bool*>(arena_->Alloc(sizeof(bool) * num_uses, ArenaAllocator::kAllocDFInfo));
+        static_cast<bool*>(arena_->Alloc(sizeof(bool) * num_uses, kArenaAllocDFInfo));
     BasicBlockId* incoming =
         static_cast<BasicBlockId*>(arena_->Alloc(sizeof(BasicBlockId) * num_uses,
-                                                 ArenaAllocator::kAllocDFInfo));
+                                                 kArenaAllocDFInfo));
     mir->meta.phi_incoming = incoming;
     int idx = 0;
     while (true) {
@@ -629,7 +629,7 @@
 
   /* Save SSA map snapshot */
   int* saved_ssa_map =
-      static_cast<int*>(arena_->Alloc(map_size, ArenaAllocator::kAllocDalvikToSSAMap));
+      static_cast<int*>(arena_->Alloc(map_size, kArenaAllocDalvikToSSAMap));
   memcpy(saved_ssa_map, vreg_to_ssa_map_, map_size);
 
   if (block->fall_through != NullBasicBlockId) {
diff --git a/compiler/dex/vreg_analysis.cc b/compiler/dex/vreg_analysis.cc
index 4d2c051..8769736 100644
--- a/compiler/dex/vreg_analysis.cc
+++ b/compiler/dex/vreg_analysis.cc
@@ -410,7 +410,7 @@
   /* Allocate the location map */
   int max_regs = GetNumSSARegs() + GetMaxPossibleCompilerTemps();
   RegLocation* loc = static_cast<RegLocation*>(arena_->Alloc(max_regs * sizeof(*loc),
-                                                             ArenaAllocator::kAllocRegAlloc));
+                                                             kArenaAllocRegAlloc));
   for (int i = 0; i < GetNumSSARegs(); i++) {
     loc[i] = fresh_loc;
     loc[i].s_reg_low = i;
diff --git a/compiler/utils/allocation.h b/compiler/utils/allocation.h
index 07cd397..b0947ca 100644
--- a/compiler/utils/allocation.h
+++ b/compiler/utils/allocation.h
@@ -26,7 +26,7 @@
  public:
   // Allocate a new ArenaObject of 'size' bytes in the Arena.
   void* operator new(size_t size, ArenaAllocator* allocator) {
-    return allocator->Alloc(size, ArenaAllocator::kAllocMisc);
+    return allocator->Alloc(size, kArenaAllocMisc);
   }
 
   void operator delete(void*, size_t) {
diff --git a/compiler/utils/arena_allocator.cc b/compiler/utils/arena_allocator.cc
index 00c3c57..365b094 100644
--- a/compiler/utils/arena_allocator.cc
+++ b/compiler/utils/arena_allocator.cc
@@ -14,6 +14,9 @@
  * limitations under the License.
  */
 
+#include <algorithm>
+#include <numeric>
+
 #include "arena_allocator.h"
 #include "base/logging.h"
 #include "base/mutex.h"
@@ -28,7 +31,7 @@
 static constexpr size_t kValgrindRedZoneBytes = 8;
 constexpr size_t Arena::kDefaultSize;
 
-static const char* alloc_names[ArenaAllocator::kNumAllocKinds] = {
+static const char* alloc_names[kNumArenaAllocKinds] = {
   "Misc       ",
   "BasicBlock ",
   "LIR        ",
@@ -42,8 +45,69 @@
   "RegAlloc   ",
   "Data       ",
   "Preds      ",
+  "STL        ",
 };
 
+template <bool kCount>
+ArenaAllocatorStatsImpl<kCount>::ArenaAllocatorStatsImpl()
+    : num_allocations_(0u) {
+  std::fill_n(alloc_stats_, arraysize(alloc_stats_), 0u);
+}
+
+template <bool kCount>
+void ArenaAllocatorStatsImpl<kCount>::Copy(const ArenaAllocatorStatsImpl& other) {
+  num_allocations_ = other.num_allocations_;
+  std::copy(other.alloc_stats_, other.alloc_stats_ + arraysize(alloc_stats_), alloc_stats_);
+}
+
+template <bool kCount>
+void ArenaAllocatorStatsImpl<kCount>::RecordAlloc(size_t bytes, ArenaAllocKind kind) {
+  alloc_stats_[kind] += bytes;
+  ++num_allocations_;
+}
+
+template <bool kCount>
+size_t ArenaAllocatorStatsImpl<kCount>::NumAllocations() const {
+  return num_allocations_;
+}
+
+template <bool kCount>
+size_t ArenaAllocatorStatsImpl<kCount>::BytesAllocated() const {
+  const size_t init = 0u;  // Initial value of the correct type.
+  return std::accumulate(alloc_stats_, alloc_stats_ + arraysize(alloc_stats_), init);
+}
+
+template <bool kCount>
+void ArenaAllocatorStatsImpl<kCount>::Dump(std::ostream& os, const Arena* first,
+                                           ssize_t lost_bytes_adjustment) const {
+  size_t malloc_bytes = 0u;
+  size_t lost_bytes = 0u;
+  size_t num_arenas = 0u;
+  for (const Arena* arena = first; arena != nullptr; arena = arena->next_) {
+    malloc_bytes += arena->Size();
+    lost_bytes += arena->RemainingSpace();
+    ++num_arenas;
+  }
+  // The lost_bytes_adjustment is used to make up for the fact that the current arena
+  // may not have the bytes_allocated_ updated correctly.
+  lost_bytes += lost_bytes_adjustment;
+  const size_t bytes_allocated = BytesAllocated();
+  os << " MEM: used: " << bytes_allocated << ", allocated: " << malloc_bytes
+     << ", lost: " << lost_bytes << "\n";
+  size_t num_allocations = ArenaAllocatorStats::NumAllocations();
+  if (num_allocations != 0) {
+    os << "Number of arenas allocated: " << num_arenas << ", Number of allocations: "
+       << num_allocations << ", avg size: " << bytes_allocated / num_allocations << "\n";
+  }
+  os << "===== Allocation by kind\n";
+  for (int i = 0; i < kNumArenaAllocKinds; i++) {
+      os << alloc_names[i] << std::setw(10) << alloc_stats_[i] << "\n";
+  }
+}
+
+// Explicitly instantiate the used implementation.
+template class ArenaAllocatorStatsImpl<kArenaAllocatorCountAllocations>;
+
 Arena::Arena(size_t size)
     : bytes_allocated_(0),
       map_(nullptr),
@@ -110,24 +174,26 @@
   return ret;
 }
 
-void ArenaPool::FreeArena(Arena* arena) {
-  Thread* self = Thread::Current();
+void ArenaPool::FreeArenaChain(Arena* first) {
   if (UNLIKELY(RUNNING_ON_VALGRIND > 0)) {
-    VALGRIND_MAKE_MEM_UNDEFINED(arena->memory_, arena->bytes_allocated_);
+    for (Arena* arena = first; arena != nullptr; arena = arena->next_) {
+      VALGRIND_MAKE_MEM_UNDEFINED(arena->memory_, arena->bytes_allocated_);
+    }
   }
-  {
+  if (first != nullptr) {
+    Arena* last = first;
+    while (last->next_ != nullptr) {
+      last = last->next_;
+    }
+    Thread* self = Thread::Current();
     MutexLock lock(self, lock_);
-    arena->next_ = free_arenas_;
-    free_arenas_ = arena;
+    last->next_ = free_arenas_;
+    free_arenas_ = first;
   }
 }
 
 size_t ArenaAllocator::BytesAllocated() const {
-  size_t total = 0;
-  for (int i = 0; i < kNumAllocKinds; i++) {
-    total += alloc_stats_[i];
-  }
-  return total;
+  return ArenaAllocatorStats::BytesAllocated();
 }
 
 ArenaAllocator::ArenaAllocator(ArenaPool* pool)
@@ -136,9 +202,7 @@
     end_(nullptr),
     ptr_(nullptr),
     arena_head_(nullptr),
-    num_allocations_(0),
     running_on_valgrind_(RUNNING_ON_VALGRIND > 0) {
-  memset(&alloc_stats_[0], 0, sizeof(alloc_stats_));
 }
 
 void ArenaAllocator::UpdateBytesAllocated() {
@@ -158,10 +222,7 @@
       return nullptr;
     }
   }
-  if (kCountAllocations) {
-    alloc_stats_[kind] += rounded_bytes;
-    ++num_allocations_;
-  }
+  ArenaAllocatorStats::RecordAlloc(rounded_bytes, kind);
   uint8_t* ret = ptr_;
   ptr_ += rounded_bytes;
   // Check that the memory is already zeroed out.
@@ -175,11 +236,7 @@
 ArenaAllocator::~ArenaAllocator() {
   // Reclaim all the arenas by giving them back to the thread pool.
   UpdateBytesAllocated();
-  while (arena_head_ != nullptr) {
-    Arena* arena = arena_head_;
-    arena_head_ = arena_head_->next_;
-    pool_->FreeArena(arena);
-  }
+  pool_->FreeArenaChain(arena_head_);
 }
 
 void ArenaAllocator::ObtainNewArenaForAllocation(size_t allocation_size) {
@@ -192,30 +249,24 @@
   end_ = new_arena->End();
 }
 
+MemStats::MemStats(const char* name, const ArenaAllocatorStats* stats, const Arena* first_arena,
+                   ssize_t lost_bytes_adjustment)
+    : name_(name),
+      stats_(stats),
+      first_arena_(first_arena),
+      lost_bytes_adjustment_(lost_bytes_adjustment) {
+}
+
+void MemStats::Dump(std::ostream& os) const {
+  os << name_ << " stats:\n";
+  stats_->Dump(os, first_arena_, lost_bytes_adjustment_);
+}
+
 // Dump memory usage stats.
-void ArenaAllocator::DumpMemStats(std::ostream& os) const {
-  size_t malloc_bytes = 0;
-  // Start out with how many lost bytes we have in the arena we are currently allocating into.
-  size_t lost_bytes(end_ - ptr_);
-  size_t num_arenas = 0;
-  for (Arena* arena = arena_head_; arena != nullptr; arena = arena->next_) {
-    malloc_bytes += arena->Size();
-    if (arena != arena_head_) {
-      lost_bytes += arena->RemainingSpace();
-    }
-    ++num_arenas;
-  }
-  const size_t bytes_allocated = BytesAllocated();
-  os << " MEM: used: " << bytes_allocated << ", allocated: " << malloc_bytes
-     << ", lost: " << lost_bytes << "\n";
-  if (num_allocations_ != 0) {
-    os << "Number of arenas allocated: " << num_arenas << ", Number of allocations: "
-       << num_allocations_ << ", avg size: " << bytes_allocated / num_allocations_ << "\n";
-  }
-  os << "===== Allocation by kind\n";
-  for (int i = 0; i < kNumAllocKinds; i++) {
-      os << alloc_names[i] << std::setw(10) << alloc_stats_[i] << "\n";
-  }
+MemStats ArenaAllocator::GetMemStats() const {
+  ssize_t lost_bytes_adjustment =
+      (arena_head_ == nullptr) ? 0 : (end_ - ptr_) - arena_head_->RemainingSpace();
+  return MemStats("ArenaAllocator", this, arena_head_, lost_bytes_adjustment);
 }
 
 }  // namespace art
diff --git a/compiler/utils/arena_allocator.h b/compiler/utils/arena_allocator.h
index 56cedfe..a6b74f7 100644
--- a/compiler/utils/arena_allocator.h
+++ b/compiler/utils/arena_allocator.h
@@ -20,6 +20,7 @@
 #include <stdint.h>
 #include <stddef.h>
 
+#include "base/macros.h"
 #include "base/mutex.h"
 #include "mem_map.h"
 
@@ -28,6 +29,70 @@
 class Arena;
 class ArenaPool;
 class ArenaAllocator;
+class ArenaStack;
+class ScopedArenaAllocator;
+class MemStats;
+
+static constexpr bool kArenaAllocatorCountAllocations = false;
+
+// Type of allocation for memory tuning.
+enum ArenaAllocKind {
+  kArenaAllocMisc,
+  kArenaAllocBB,
+  kArenaAllocLIR,
+  kArenaAllocMIR,
+  kArenaAllocDFInfo,
+  kArenaAllocGrowableArray,
+  kArenaAllocGrowableBitMap,
+  kArenaAllocDalvikToSSAMap,
+  kArenaAllocDebugInfo,
+  kArenaAllocSuccessor,
+  kArenaAllocRegAlloc,
+  kArenaAllocData,
+  kArenaAllocPredecessors,
+  kArenaAllocSTL,
+  kNumArenaAllocKinds
+};
+
+template <bool kCount>
+class ArenaAllocatorStatsImpl;
+
+template <>
+class ArenaAllocatorStatsImpl<false> {
+ public:
+  ArenaAllocatorStatsImpl() = default;
+  ArenaAllocatorStatsImpl(const ArenaAllocatorStatsImpl& other) = default;
+  ArenaAllocatorStatsImpl& operator = (const ArenaAllocatorStatsImpl& other) = delete;
+
+  void Copy(const ArenaAllocatorStatsImpl& other) { UNUSED(other); }
+  void RecordAlloc(size_t bytes, ArenaAllocKind kind) { UNUSED(bytes); UNUSED(kind); }
+  size_t NumAllocations() const { return 0u; }
+  size_t BytesAllocated() const { return 0u; }
+  void Dump(std::ostream& os, const Arena* first, ssize_t lost_bytes_adjustment) const {
+    UNUSED(os); UNUSED(first); UNUSED(lost_bytes_adjustment);
+  }
+};
+
+template <bool kCount>
+class ArenaAllocatorStatsImpl {
+ public:
+  ArenaAllocatorStatsImpl();
+  ArenaAllocatorStatsImpl(const ArenaAllocatorStatsImpl& other) = default;
+  ArenaAllocatorStatsImpl& operator = (const ArenaAllocatorStatsImpl& other) = delete;
+
+  void Copy(const ArenaAllocatorStatsImpl& other);
+  void RecordAlloc(size_t bytes, ArenaAllocKind kind);
+  size_t NumAllocations() const;
+  size_t BytesAllocated() const;
+  void Dump(std::ostream& os, const Arena* first, ssize_t lost_bytes_adjustment) const;
+
+ private:
+  size_t num_allocations_;
+  // TODO: Use std::array<size_t, kNumArenaAllocKinds> from C++11 when we upgrade the STL.
+  size_t alloc_stats_[kNumArenaAllocKinds];  // Bytes used by various allocation kinds.
+};
+
+typedef ArenaAllocatorStatsImpl<kArenaAllocatorCountAllocations> ArenaAllocatorStats;
 
 class Arena {
  public:
@@ -59,6 +124,9 @@
   Arena* next_;
   friend class ArenaPool;
   friend class ArenaAllocator;
+  friend class ArenaStack;
+  friend class ScopedArenaAllocator;
+  template <bool kCount> friend class ArenaAllocatorStatsImpl;
   DISALLOW_COPY_AND_ASSIGN(Arena);
 };
 
@@ -67,7 +135,7 @@
   ArenaPool();
   ~ArenaPool();
   Arena* AllocArena(size_t size);
-  void FreeArena(Arena* arena);
+  void FreeArenaChain(Arena* first);
 
  private:
   Mutex lock_ DEFAULT_MUTEX_ACQUIRED_AFTER;
@@ -75,28 +143,8 @@
   DISALLOW_COPY_AND_ASSIGN(ArenaPool);
 };
 
-class ArenaAllocator {
+class ArenaAllocator : private ArenaAllocatorStats {
  public:
-  // Type of allocation for memory tuning.
-  enum ArenaAllocKind {
-    kAllocMisc,
-    kAllocBB,
-    kAllocLIR,
-    kAllocMIR,
-    kAllocDFInfo,
-    kAllocGrowableArray,
-    kAllocGrowableBitMap,
-    kAllocDalvikToSSAMap,
-    kAllocDebugInfo,
-    kAllocSuccessor,
-    kAllocRegAlloc,
-    kAllocData,
-    kAllocPredecessors,
-    kNumAllocKinds
-  };
-
-  static constexpr bool kCountAllocations = false;
-
   explicit ArenaAllocator(ArenaPool* pool);
   ~ArenaAllocator();
 
@@ -113,10 +161,7 @@
         return nullptr;
       }
     }
-    if (kCountAllocations) {
-      alloc_stats_[kind] += bytes;
-      ++num_allocations_;
-    }
+    ArenaAllocatorStats::RecordAlloc(bytes, kind);
     uint8_t* ret = ptr_;
     ptr_ += bytes;
     return ret;
@@ -125,7 +170,7 @@
   void* AllocValgrind(size_t bytes, ArenaAllocKind kind);
   void ObtainNewArenaForAllocation(size_t allocation_size);
   size_t BytesAllocated() const;
-  void DumpMemStats(std::ostream& os) const;
+  MemStats GetMemStats() const;
 
  private:
   void UpdateBytesAllocated();
@@ -135,21 +180,22 @@
   uint8_t* end_;
   uint8_t* ptr_;
   Arena* arena_head_;
-  size_t num_allocations_;
-  size_t alloc_stats_[kNumAllocKinds];  // Bytes used by various allocation kinds.
   bool running_on_valgrind_;
 
   DISALLOW_COPY_AND_ASSIGN(ArenaAllocator);
 };  // ArenaAllocator
 
-struct MemStats {
-   public:
-     void Dump(std::ostream& os) const {
-       arena_.DumpMemStats(os);
-     }
-     explicit MemStats(const ArenaAllocator &arena) : arena_(arena) {}
-  private:
-    const ArenaAllocator &arena_;
+class MemStats {
+ public:
+  MemStats(const char* name, const ArenaAllocatorStats* stats, const Arena* first_arena,
+           ssize_t lost_bytes_adjustment = 0);
+  void Dump(std::ostream& os) const;
+
+ private:
+  const char* const name_;
+  const ArenaAllocatorStats* const stats_;
+  const Arena* const first_arena_;
+  const ssize_t lost_bytes_adjustment_;
 };  // MemStats
 
 }  // namespace art
diff --git a/compiler/utils/arena_bit_vector.cc b/compiler/utils/arena_bit_vector.cc
index 220ff14..eff9778 100644
--- a/compiler/utils/arena_bit_vector.cc
+++ b/compiler/utils/arena_bit_vector.cc
@@ -25,13 +25,13 @@
   ~ArenaBitVectorAllocator() {}
 
   virtual void* Alloc(size_t size) {
-    return arena_->Alloc(size, ArenaAllocator::kAllocGrowableBitMap);
+    return arena_->Alloc(size, kArenaAllocGrowableBitMap);
   }
 
   virtual void Free(void*) {}  // Nop.
 
   static void* operator new(size_t size, ArenaAllocator* arena) {
-    return arena->Alloc(sizeof(ArenaBitVectorAllocator), ArenaAllocator::kAllocGrowableBitMap);
+    return arena->Alloc(sizeof(ArenaBitVectorAllocator), kArenaAllocGrowableBitMap);
   }
   static void operator delete(void* p) {}  // Nop.
 
diff --git a/compiler/utils/arena_bit_vector.h b/compiler/utils/arena_bit_vector.h
index 6c14617..1a3d6a3 100644
--- a/compiler/utils/arena_bit_vector.h
+++ b/compiler/utils/arena_bit_vector.h
@@ -55,7 +55,7 @@
     ~ArenaBitVector() {}
 
   static void* operator new(size_t size, ArenaAllocator* arena) {
-     return arena->Alloc(sizeof(ArenaBitVector), ArenaAllocator::kAllocGrowableBitMap);
+     return arena->Alloc(sizeof(ArenaBitVector), kArenaAllocGrowableBitMap);
   }
   static void operator delete(void* p) {}  // Nop.
 
diff --git a/compiler/utils/debug_stack.h b/compiler/utils/debug_stack.h
new file mode 100644
index 0000000..2e02b43
--- /dev/null
+++ b/compiler/utils/debug_stack.h
@@ -0,0 +1,138 @@
+/*
+ * Copyright (C) 2014 The Android Open Source Project
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#ifndef ART_COMPILER_UTILS_DEBUG_STACK_H_
+#define ART_COMPILER_UTILS_DEBUG_STACK_H_
+
+#include "base/logging.h"
+#include "base/macros.h"
+#include "globals.h"
+
+namespace art {
+
+// Helper classes for reference counting to enforce construction/destruction order and
+// usage of the top element of a stack in debug mode with no overhead in release mode.
+
+// Reference counter. No references allowed in destructor or in explicitly called CheckNoRefs().
+template <bool kIsDebug>
+class DebugStackRefCounterImpl;
+// Reference. Allows an explicit check that it's the top reference.
+template <bool kIsDebug>
+class DebugStackReferenceImpl;
+// Indirect top reference. Checks that the reference is the top reference when used.
+template <bool kIsDebug>
+class DebugStackIndirectTopRefImpl;
+
+typedef DebugStackRefCounterImpl<kIsDebugBuild> DebugStackRefCounter;
+typedef DebugStackReferenceImpl<kIsDebugBuild> DebugStackReference;
+typedef DebugStackIndirectTopRefImpl<kIsDebugBuild> DebugStackIndirectTopRef;
+
+// Non-debug mode specializations. This should be optimized away.
+
+template <>
+class DebugStackRefCounterImpl<false> {
+ public:
+  size_t IncrementRefCount() { return 0u; }
+  void DecrementRefCount() { }
+  size_t GetRefCount() const { return 0u; }
+  void CheckNoRefs() const { }
+};
+
+template <>
+class DebugStackReferenceImpl<false> {
+ public:
+  explicit DebugStackReferenceImpl(DebugStackRefCounterImpl<false>* counter) { UNUSED(counter); }
+  DebugStackReferenceImpl(const DebugStackReferenceImpl& other) = default;
+  DebugStackReferenceImpl& operator=(const DebugStackReferenceImpl& other) = default;
+  void CheckTop() { }
+};
+
+template <>
+class DebugStackIndirectTopRefImpl<false> {
+ public:
+  explicit DebugStackIndirectTopRefImpl(DebugStackReferenceImpl<false>* ref) { UNUSED(ref); }
+  DebugStackIndirectTopRefImpl(const DebugStackIndirectTopRefImpl& other) = default;
+  DebugStackIndirectTopRefImpl& operator=(const DebugStackIndirectTopRefImpl& other) = default;
+  void CheckTop() { }
+};
+
+// Debug mode versions.
+
+template <bool kIsDebug>
+class DebugStackRefCounterImpl {
+ public:
+  DebugStackRefCounterImpl() : ref_count_(0u) { }
+  ~DebugStackRefCounterImpl() { CheckNoRefs(); }
+  size_t IncrementRefCount() { return ++ref_count_; }
+  void DecrementRefCount() { --ref_count_; }
+  size_t GetRefCount() const { return ref_count_; }
+  void CheckNoRefs() const { CHECK_EQ(ref_count_, 0u); }
+
+ private:
+  size_t ref_count_;
+};
+
+template <bool kIsDebug>
+class DebugStackReferenceImpl {
+ public:
+  explicit DebugStackReferenceImpl(DebugStackRefCounterImpl<kIsDebug>* counter)
+    : counter_(counter), ref_count_(counter->IncrementRefCount()) {
+  }
+  DebugStackReferenceImpl(const DebugStackReferenceImpl& other)
+    : counter_(other.counter_), ref_count_(counter_->IncrementRefCount()) {
+  }
+  DebugStackReferenceImpl& operator=(const DebugStackReferenceImpl& other) {
+    CHECK(counter_ == other.counter_);
+    return *this;
+  }
+  ~DebugStackReferenceImpl() { counter_->DecrementRefCount(); }
+  void CheckTop() { CHECK_EQ(counter_->GetRefCount(), ref_count_); }
+
+ private:
+  DebugStackRefCounterImpl<true>* counter_;
+  size_t ref_count_;
+};
+
+template <bool kIsDebug>
+class DebugStackIndirectTopRefImpl {
+ public:
+  explicit DebugStackIndirectTopRefImpl(DebugStackReferenceImpl<kIsDebug>* ref)
+      : ref_(ref) {
+    CheckTop();
+  }
+  DebugStackIndirectTopRefImpl(const DebugStackIndirectTopRefImpl& other)
+      : ref_(other.ref_) {
+    CheckTop();
+  }
+  DebugStackIndirectTopRefImpl& operator=(const DebugStackIndirectTopRefImpl& other) {
+    CHECK(ref_ == other->ref_);
+    CheckTop();
+    return *this;
+  }
+  ~DebugStackIndirectTopRefImpl() {
+    CheckTop();
+  }
+  void CheckTop() {
+    ref_->CheckTop();
+  }
+
+ private:
+  DebugStackReferenceImpl<kIsDebug>* ref_;
+};
+
+}  // namespace art
+
+#endif  // ART_COMPILER_UTILS_DEBUG_STACK_H_
diff --git a/compiler/utils/growable_array.h b/compiler/utils/growable_array.h
index 82b6a60..a7d1f0e 100644
--- a/compiler/utils/growable_array.h
+++ b/compiler/utils/growable_array.h
@@ -75,7 +75,7 @@
         num_used_(0),
         kind_(kind) {
       elem_list_ = static_cast<T*>(arena_->Alloc(sizeof(T) * init_length,
-                                                 ArenaAllocator::kAllocGrowableArray));
+                                                 kArenaAllocGrowableArray));
     };
 
 
@@ -89,7 +89,7 @@
          target_length = new_length;
       }
       T* new_array = static_cast<T*>(arena_->Alloc(sizeof(T) * target_length,
-                                                   ArenaAllocator::kAllocGrowableArray));
+                                                   kArenaAllocGrowableArray));
       memcpy(new_array, elem_list_, sizeof(T) * num_allocated_);
       num_allocated_ = target_length;
       elem_list_ = new_array;
@@ -181,7 +181,7 @@
     T* GetRawStorage() const { return elem_list_; }
 
     static void* operator new(size_t size, ArenaAllocator* arena) {
-      return arena->Alloc(sizeof(GrowableArray<T>), ArenaAllocator::kAllocGrowableArray);
+      return arena->Alloc(sizeof(GrowableArray<T>), kArenaAllocGrowableArray);
     };
     static void operator delete(void* p) {}  // Nop.
 
diff --git a/compiler/utils/scoped_arena_allocator.cc b/compiler/utils/scoped_arena_allocator.cc
new file mode 100644
index 0000000..ee3b07e
--- /dev/null
+++ b/compiler/utils/scoped_arena_allocator.cc
@@ -0,0 +1,126 @@
+/*
+ * Copyright (C) 2014 The Android Open Source Project
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#include "scoped_arena_allocator.h"
+
+#include "utils/arena_allocator.h"
+#include <memcheck/memcheck.h>
+
+namespace art {
+
+static constexpr size_t kValgrindRedZoneBytes = 8;
+
+ArenaStack::ArenaStack(ArenaPool* arena_pool)
+  : DebugStackRefCounter(),
+    stats_and_pool_(arena_pool),
+    bottom_arena_(nullptr),
+    top_arena_(nullptr),
+    top_ptr_(nullptr),
+    top_end_(nullptr),
+    running_on_valgrind_(RUNNING_ON_VALGRIND > 0) {
+}
+
+ArenaStack::~ArenaStack() {
+  stats_and_pool_.pool->FreeArenaChain(bottom_arena_);
+}
+
+MemStats ArenaStack::GetPeakStats() const {
+  DebugStackRefCounter::CheckNoRefs();
+  return MemStats("ArenaStack peak", static_cast<const TaggedStats<Peak>*>(&stats_and_pool_),
+                  bottom_arena_);
+}
+
+uint8_t* ArenaStack::AllocateFromNextArena(size_t rounded_bytes) {
+  UpdateBytesAllocated();
+  size_t allocation_size = std::max(Arena::kDefaultSize, rounded_bytes);
+  if (UNLIKELY(top_arena_ == nullptr)) {
+    top_arena_ = bottom_arena_ = stats_and_pool_.pool->AllocArena(allocation_size);
+    top_arena_->next_ = nullptr;
+  } else if (top_arena_->next_ != nullptr && top_arena_->next_->Size() >= allocation_size) {
+    top_arena_ = top_arena_->next_;
+  } else {
+    Arena* tail = top_arena_->next_;
+    top_arena_->next_ = stats_and_pool_.pool->AllocArena(allocation_size);
+    top_arena_ = top_arena_->next_;
+    top_arena_->next_ = tail;
+  }
+  top_end_ = top_arena_->End();
+  // top_ptr_ shall be updated by ScopedArenaAllocator.
+  return top_arena_->Begin();
+}
+
+void ArenaStack::UpdatePeakStatsAndRestore(const ArenaAllocatorStats& restore_stats) {
+  if (PeakStats()->BytesAllocated() < CurrentStats()->BytesAllocated()) {
+    PeakStats()->Copy(*CurrentStats());
+  }
+  CurrentStats()->Copy(restore_stats);
+}
+
+void ArenaStack::UpdateBytesAllocated() {
+  if (top_arena_ != nullptr) {
+    // Update how many bytes we have allocated into the arena so that the arena pool knows how
+    // much memory to zero out. Though ScopedArenaAllocator doesn't guarantee the memory is
+    // zero-initialized, the Arena may be reused by ArenaAllocator which does guarantee this.
+    size_t allocated = static_cast<size_t>(top_ptr_ - top_arena_->Begin());
+    if (top_arena_->bytes_allocated_ < allocated) {
+      top_arena_->bytes_allocated_ = allocated;
+    }
+  }
+}
+
+void* ArenaStack::AllocValgrind(size_t bytes, ArenaAllocKind kind) {
+  size_t rounded_bytes = (bytes + kValgrindRedZoneBytes + 3) & ~3;
+  uint8_t* ptr = top_ptr_;
+  if (UNLIKELY(static_cast<size_t>(top_end_ - ptr) < rounded_bytes)) {
+    ptr = AllocateFromNextArena(rounded_bytes);
+  }
+  CurrentStats()->RecordAlloc(bytes, kind);
+  top_ptr_ = ptr + rounded_bytes;
+  VALGRIND_MAKE_MEM_NOACCESS(ptr + bytes, rounded_bytes - bytes);
+  return ptr;
+}
+
+ScopedArenaAllocator::ScopedArenaAllocator(ArenaStack* arena_stack)
+  : DebugStackReference(arena_stack),
+    DebugStackRefCounter(),
+    ArenaAllocatorStats(*arena_stack->CurrentStats()),
+    arena_stack_(arena_stack),
+    mark_arena_(arena_stack->top_arena_),
+    mark_ptr_(arena_stack->top_ptr_),
+    mark_end_(arena_stack->top_end_) {
+}
+
+ScopedArenaAllocator::~ScopedArenaAllocator() {
+  Reset();
+}
+
+void ScopedArenaAllocator::Reset() {
+  DebugStackReference::CheckTop();
+  DebugStackRefCounter::CheckNoRefs();
+  arena_stack_->UpdatePeakStatsAndRestore(*this);
+  arena_stack_->UpdateBytesAllocated();
+  if (LIKELY(mark_arena_ != nullptr)) {
+    arena_stack_->top_arena_ = mark_arena_;
+    arena_stack_->top_ptr_ = mark_ptr_;
+    arena_stack_->top_end_ = mark_end_;
+  } else if (arena_stack_->bottom_arena_ != nullptr) {
+    mark_arena_ = arena_stack_->top_arena_ = arena_stack_->bottom_arena_;
+    mark_ptr_ = arena_stack_->top_ptr_ = mark_arena_->Begin();
+    mark_end_ = arena_stack_->top_end_ = mark_arena_->End();
+  }
+}
+
+}  // namespace art
diff --git a/compiler/utils/scoped_arena_allocator.h b/compiler/utils/scoped_arena_allocator.h
new file mode 100644
index 0000000..24a8afe
--- /dev/null
+++ b/compiler/utils/scoped_arena_allocator.h
@@ -0,0 +1,244 @@
+/*
+ * Copyright (C) 2014 The Android Open Source Project
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#ifndef ART_COMPILER_UTILS_SCOPED_ARENA_ALLOCATOR_H_
+#define ART_COMPILER_UTILS_SCOPED_ARENA_ALLOCATOR_H_
+
+#include "base/logging.h"
+#include "base/macros.h"
+#include "utils/arena_allocator.h"
+#include "utils/debug_stack.h"
+#include "globals.h"
+
+namespace art {
+
+class ArenaStack;
+class ScopedArenaAllocator;
+
+template <typename T>
+class ScopedArenaAllocatorAdapter;
+
+// Holds a list of Arenas for use by ScopedArenaAllocator stack.
+class ArenaStack : private DebugStackRefCounter {
+ public:
+  explicit ArenaStack(ArenaPool* arena_pool);
+  ~ArenaStack();
+
+  size_t PeakBytesAllocated() {
+    return PeakStats()->BytesAllocated();
+  }
+
+  MemStats GetPeakStats() const;
+
+ private:
+  struct Peak;
+  struct Current;
+  template <typename Tag> struct TaggedStats : ArenaAllocatorStats { };
+  struct StatsAndPool : TaggedStats<Peak>, TaggedStats<Current> {
+    explicit StatsAndPool(ArenaPool* arena_pool) : pool(arena_pool) { }
+    ArenaPool* const pool;
+  };
+
+  ArenaAllocatorStats* PeakStats() {
+    return static_cast<TaggedStats<Peak>*>(&stats_and_pool_);
+  }
+
+  ArenaAllocatorStats* CurrentStats() {
+    return static_cast<TaggedStats<Current>*>(&stats_and_pool_);
+  }
+
+  // Private - access via ScopedArenaAllocator or ScopedArenaAllocatorAdapter.
+  void* Alloc(size_t bytes, ArenaAllocKind kind) ALWAYS_INLINE {
+    if (UNLIKELY(running_on_valgrind_)) {
+      return AllocValgrind(bytes, kind);
+    }
+    size_t rounded_bytes = (bytes + 3) & ~3;
+    uint8_t* ptr = top_ptr_;
+    if (UNLIKELY(static_cast<size_t>(top_end_ - ptr) < rounded_bytes)) {
+      ptr = AllocateFromNextArena(rounded_bytes);
+    }
+    CurrentStats()->RecordAlloc(bytes, kind);
+    top_ptr_ = ptr + rounded_bytes;
+    return ptr;
+  }
+
+  uint8_t* AllocateFromNextArena(size_t rounded_bytes);
+  void UpdatePeakStatsAndRestore(const ArenaAllocatorStats& restore_stats);
+  void UpdateBytesAllocated();
+  void* AllocValgrind(size_t bytes, ArenaAllocKind kind);
+
+  StatsAndPool stats_and_pool_;
+  Arena* bottom_arena_;
+  Arena* top_arena_;
+  uint8_t* top_ptr_;
+  uint8_t* top_end_;
+
+  const bool running_on_valgrind_;
+
+  friend class ScopedArenaAllocator;
+  template <typename T>
+  friend class ScopedArenaAllocatorAdapter;
+
+  DISALLOW_COPY_AND_ASSIGN(ArenaStack);
+};
+
+class ScopedArenaAllocator
+    : private DebugStackReference, private DebugStackRefCounter, private ArenaAllocatorStats {
+ public:
+  // Create a ScopedArenaAllocator directly on the ArenaStack when the scope of
+  // the allocator is not exactly a C++ block scope. For example, an optimization
+  // pass can create the scoped allocator in Start() and destroy it in End().
+  static ScopedArenaAllocator* Create(ArenaStack* arena_stack) {
+    void* addr = arena_stack->Alloc(sizeof(ScopedArenaAllocator), kArenaAllocMisc);
+    ScopedArenaAllocator* allocator = new(addr) ScopedArenaAllocator(arena_stack);
+    allocator->mark_ptr_ = reinterpret_cast<uint8_t*>(addr);
+    return allocator;
+  }
+
+  explicit ScopedArenaAllocator(ArenaStack* arena_stack);
+  ~ScopedArenaAllocator();
+
+  void Reset();
+
+  void* Alloc(size_t bytes, ArenaAllocKind kind) ALWAYS_INLINE {
+    DebugStackReference::CheckTop();
+    return arena_stack_->Alloc(bytes, kind);
+  }
+
+  // ScopedArenaAllocatorAdapter is incomplete here, we need to define this later.
+  ScopedArenaAllocatorAdapter<void> Adapter();
+
+  // Allow a delete-expression to destroy but not deallocate allocators created by Create().
+  static void operator delete(void* ptr) { UNUSED(ptr); }
+
+ private:
+  ArenaStack* const arena_stack_;
+  Arena* mark_arena_;
+  uint8_t* mark_ptr_;
+  uint8_t* mark_end_;
+
+  template <typename T>
+  friend class ScopedArenaAllocatorAdapter;
+
+  DISALLOW_COPY_AND_ASSIGN(ScopedArenaAllocator);
+};
+
+template <>
+class ScopedArenaAllocatorAdapter<void>
+    : private DebugStackReference, private DebugStackIndirectTopRef {
+ public:
+  typedef void value_type;
+  typedef void* pointer;
+  typedef const void* const_pointer;
+
+  template <typename U>
+  struct rebind {
+    typedef ScopedArenaAllocatorAdapter<U> other;
+  };
+
+  explicit ScopedArenaAllocatorAdapter(ScopedArenaAllocator* arena_allocator)
+      : DebugStackReference(arena_allocator),
+        DebugStackIndirectTopRef(arena_allocator),
+        arena_stack_(arena_allocator->arena_stack_) {
+  }
+  template <typename U>
+  ScopedArenaAllocatorAdapter(const ScopedArenaAllocatorAdapter<U>& other)
+      : DebugStackReference(other),
+        DebugStackIndirectTopRef(other),
+        arena_stack_(other.arena_stack_) {
+  }
+  ScopedArenaAllocatorAdapter(const ScopedArenaAllocatorAdapter& other) = default;
+  ScopedArenaAllocatorAdapter& operator=(const ScopedArenaAllocatorAdapter& other) = default;
+  ~ScopedArenaAllocatorAdapter() = default;
+
+ private:
+  ArenaStack* arena_stack_;
+
+  template <typename U>
+  friend class ScopedArenaAllocatorAdapter;
+};
+
+// Adapter for use of ScopedArenaAllocator in STL containers.
+template <typename T>
+class ScopedArenaAllocatorAdapter : private DebugStackReference, private DebugStackIndirectTopRef {
+ public:
+  typedef T value_type;
+  typedef T* pointer;
+  typedef T& reference;
+  typedef const T* const_pointer;
+  typedef const T& const_reference;
+  typedef size_t size_type;
+  typedef ptrdiff_t difference_type;
+
+  template <typename U>
+  struct rebind {
+    typedef ScopedArenaAllocatorAdapter<U> other;
+  };
+
+  explicit ScopedArenaAllocatorAdapter(ScopedArenaAllocator* arena_allocator)
+      : DebugStackReference(arena_allocator),
+        DebugStackIndirectTopRef(arena_allocator),
+        arena_stack_(arena_allocator->arena_stack_) {
+  }
+  template <typename U>
+  ScopedArenaAllocatorAdapter(const ScopedArenaAllocatorAdapter<U>& other)
+      : DebugStackReference(other),
+        DebugStackIndirectTopRef(other),
+        arena_stack_(other.arena_stack_) {
+  }
+  ScopedArenaAllocatorAdapter(const ScopedArenaAllocatorAdapter& other) = default;
+  ScopedArenaAllocatorAdapter& operator=(const ScopedArenaAllocatorAdapter& other) = default;
+  ~ScopedArenaAllocatorAdapter() = default;
+
+  size_type max_size() const {
+    return static_cast<size_type>(-1) / sizeof(T);
+  }
+
+  pointer address(reference x) const { return &x; }
+  const_pointer address(const_reference x) const { return &x; }
+
+  pointer allocate(size_type n, ScopedArenaAllocatorAdapter<void>::pointer hint = nullptr) {
+    DCHECK_LE(n, max_size());
+    DebugStackIndirectTopRef::CheckTop();
+    return reinterpret_cast<T*>(arena_stack_->Alloc(n * sizeof(T), kArenaAllocSTL));
+  }
+  void deallocate(pointer p, size_type n) {
+    DebugStackIndirectTopRef::CheckTop();
+  }
+
+  void construct(pointer p, const_reference val) {
+    DebugStackIndirectTopRef::CheckTop();
+    new (static_cast<void*>(p)) value_type(val);
+  }
+  void destroy(pointer p) {
+    DebugStackIndirectTopRef::CheckTop();
+    p->~value_type();
+  }
+
+ private:
+  ArenaStack* arena_stack_;
+
+  template <typename U>
+  friend class ScopedArenaAllocatorAdapter;
+};
+
+inline ScopedArenaAllocatorAdapter<void> ScopedArenaAllocator::Adapter() {
+  return ScopedArenaAllocatorAdapter<void>(this);
+}
+
+}  // namespace art
+
+#endif  // ART_COMPILER_UTILS_SCOPED_ARENA_ALLOCATOR_H_