ART: Implement literal pool for arm, fix branch fixup.

Change-Id: Iecc91418bb4ee1c957f42fefb737d0ee2ba960e7
diff --git a/compiler/optimizing/code_generator.h b/compiler/optimizing/code_generator.h
index e6b1f7c..a82b08a 100644
--- a/compiler/optimizing/code_generator.h
+++ b/compiler/optimizing/code_generator.h
@@ -358,6 +358,7 @@
         number_of_register_pairs_(number_of_register_pairs),
         core_callee_save_mask_(core_callee_save_mask),
         fpu_callee_save_mask_(fpu_callee_save_mask),
+        stack_map_stream_(graph->GetArena()),
         is_baseline_(false),
         graph_(graph),
         compiler_options_(compiler_options),
@@ -365,8 +366,7 @@
         block_order_(nullptr),
         current_block_index_(0),
         is_leaf_(true),
-        requires_current_method_(false),
-        stack_map_stream_(graph->GetArena()) {}
+        requires_current_method_(false) {}
 
   // Register allocation logic.
   void AllocateRegistersLocally(HInstruction* instruction) const;
@@ -436,6 +436,8 @@
   const uint32_t core_callee_save_mask_;
   const uint32_t fpu_callee_save_mask_;
 
+  StackMapStream stack_map_stream_;
+
   // Whether we are using baseline.
   bool is_baseline_;
 
@@ -464,8 +466,6 @@
   // Whether an instruction in the graph accesses the current method.
   bool requires_current_method_;
 
-  StackMapStream stack_map_stream_;
-
   friend class OptimizingCFITest;
 
   DISALLOW_COPY_AND_ASSIGN(CodeGenerator);
diff --git a/compiler/optimizing/code_generator_arm.cc b/compiler/optimizing/code_generator_arm.cc
index f773106..ec0d56d 100644
--- a/compiler/optimizing/code_generator_arm.cc
+++ b/compiler/optimizing/code_generator_arm.cc
@@ -392,12 +392,26 @@
       location_builder_(graph, this),
       instruction_visitor_(graph, this),
       move_resolver_(graph->GetArena(), this),
-      assembler_(false /* can_relocate_branches */),
+      assembler_(),
       isa_features_(isa_features) {
   // Save the PC register to mimic Quick.
   AddAllocatedRegister(Location::RegisterLocation(PC));
 }
 
+void CodeGeneratorARM::Finalize(CodeAllocator* allocator) {
+  // Ensure that we fix up branches and literal loads and emit the literal pool.
+  __ FinalizeCode();
+
+  // Adjust native pc offsets in stack maps.
+  for (size_t i = 0, num = stack_map_stream_.GetNumberOfStackMaps(); i != num; ++i) {
+    uint32_t old_position = stack_map_stream_.GetStackMap(i).native_pc_offset;
+    uint32_t new_position = __ GetAdjustedPosition(old_position);
+    stack_map_stream_.SetStackMapNativePcOffset(i, new_position);
+  }
+
+  CodeGenerator::Finalize(allocator);
+}
+
 Location CodeGeneratorARM::AllocateFreeRegister(Primitive::Type type) const {
   switch (type) {
     case Primitive::kPrimLong: {
@@ -2842,7 +2856,7 @@
   Location left = locations->InAt(0);
   Location right = locations->InAt(1);
 
-  NearLabel less, greater, done;
+  Label less, greater, done;
   Primitive::Type type = compare->InputAt(0)->GetType();
   switch (type) {
     case Primitive::kPrimLong: {
@@ -2938,7 +2952,7 @@
                                                           Register temp1,
                                                           Register temp2,
                                                           HInstruction* instruction) {
-  NearLabel fail;
+  Label fail;
   if (offset != 0) {
     __ LoadImmediate(temp1, offset);
     __ add(IP, addr, ShifterOperand(temp1));
@@ -3618,7 +3632,7 @@
                                   Register object,
                                   Register value,
                                   bool can_be_null) {
-  NearLabel is_null;
+  Label is_null;
   if (can_be_null) {
     __ CompareAndBranchIfZero(value, &is_null);
   }
@@ -4047,7 +4061,7 @@
   Register cls = locations->InAt(1).AsRegister<Register>();
   Register out = locations->Out().AsRegister<Register>();
   uint32_t class_offset = mirror::Object::ClassOffset().Int32Value();
-  NearLabel done, zero;
+  Label done, zero;
   SlowPathCodeARM* slow_path = nullptr;
 
   // Return 0 if `obj` is null.
@@ -4104,19 +4118,15 @@
       instruction, locations->InAt(1), locations->GetTemp(0), instruction->GetDexPc());
   codegen_->AddSlowPath(slow_path);
 
-  NearLabel done;
   // avoid null check if we know obj is not null.
   if (instruction->MustDoNullCheck()) {
-    __ CompareAndBranchIfZero(obj, &done);
+    __ CompareAndBranchIfZero(obj, slow_path->GetExitLabel());
   }
   // Compare the class of `obj` with `cls`.
   __ LoadFromOffset(kLoadWord, temp, obj, class_offset);
   __ cmp(temp, ShifterOperand(cls));
   __ b(slow_path->GetEntryLabel(), NE);
   __ Bind(slow_path->GetExitLabel());
-  if (instruction->MustDoNullCheck()) {
-    __ Bind(&done);
-  }
 }
 
 void LocationsBuilderARM::VisitMonitorOperation(HMonitorOperation* instruction) {
diff --git a/compiler/optimizing/code_generator_arm.h b/compiler/optimizing/code_generator_arm.h
index d84f2d3..c5a28ba 100644
--- a/compiler/optimizing/code_generator_arm.h
+++ b/compiler/optimizing/code_generator_arm.h
@@ -289,6 +289,8 @@
     block_labels_.SetSize(GetGraph()->GetBlocks().Size());
   }
 
+  void Finalize(CodeAllocator* allocator) OVERRIDE;
+
   const ArmInstructionSetFeatures& GetInstructionSetFeatures() const {
     return isa_features_;
   }
diff --git a/compiler/optimizing/stack_map_stream.h b/compiler/optimizing/stack_map_stream.h
index bc3653d..550ed70 100644
--- a/compiler/optimizing/stack_map_stream.h
+++ b/compiler/optimizing/stack_map_stream.h
@@ -134,6 +134,11 @@
     return stack_maps_.GetRawStorage()[i];
   }
 
+  void SetStackMapNativePcOffset(size_t i, uint32_t native_pc_offset) {
+    DCHECK_LT(i, stack_maps_.Size());
+    stack_maps_.GetRawStorage()[i].native_pc_offset = native_pc_offset;
+  }
+
   uint32_t ComputeMaxNativePcOffset() const;
 
   // Prepares the stream to fill in a memory region. Must be called before FillIn.