diff --git a/compiler/Android.mk b/compiler/Android.mk
index 67536f0..60668ed 100644
--- a/compiler/Android.mk
+++ b/compiler/Android.mk
@@ -103,6 +103,7 @@
 	optimizing/code_generator.cc \
 	optimizing/code_generator_arm.cc \
 	optimizing/code_generator_arm64.cc \
+	optimizing/code_generator_mips64.cc \
 	optimizing/code_generator_x86.cc \
 	optimizing/code_generator_x86_64.cc \
 	optimizing/code_generator_utils.cc \
diff --git a/compiler/optimizing/builder.cc b/compiler/optimizing/builder.cc
index 946c060..b568f31 100644
--- a/compiler/optimizing/builder.cc
+++ b/compiler/optimizing/builder.cc
@@ -483,10 +483,11 @@
 
 void HGraphBuilder::Binop_23x_cmp(const Instruction& instruction,
                                   Primitive::Type type,
-                                  HCompare::Bias bias) {
+                                  HCompare::Bias bias,
+                                  uint32_t dex_pc) {
   HInstruction* first = LoadLocal(instruction.VRegB(), type);
   HInstruction* second = LoadLocal(instruction.VRegC(), type);
-  current_block_->AddInstruction(new (arena_) HCompare(type, first, second, bias));
+  current_block_->AddInstruction(new (arena_) HCompare(type, first, second, bias, dex_pc));
   UpdateLocal(instruction.VRegA(), current_block_->GetLastInstruction());
 }
 
@@ -878,17 +879,25 @@
   return true;
 }
 
-mirror::Class* HGraphBuilder::GetOutermostCompilingClass() const {
+static mirror::Class* GetClassFrom(CompilerDriver* driver,
+                                   const DexCompilationUnit& compilation_unit) {
   ScopedObjectAccess soa(Thread::Current());
   StackHandleScope<2> hs(soa.Self());
-  const DexFile& outer_dex_file = *outer_compilation_unit_->GetDexFile();
+  const DexFile& dex_file = *compilation_unit.GetDexFile();
   Handle<mirror::ClassLoader> class_loader(hs.NewHandle(
-      soa.Decode<mirror::ClassLoader*>(dex_compilation_unit_->GetClassLoader())));
-  Handle<mirror::DexCache> outer_dex_cache(hs.NewHandle(
-      outer_compilation_unit_->GetClassLinker()->FindDexCache(outer_dex_file)));
+      soa.Decode<mirror::ClassLoader*>(compilation_unit.GetClassLoader())));
+  Handle<mirror::DexCache> dex_cache(hs.NewHandle(
+      compilation_unit.GetClassLinker()->FindDexCache(dex_file)));
 
-  return compiler_driver_->ResolveCompilingMethodsClass(
-      soa, outer_dex_cache, class_loader, outer_compilation_unit_);
+  return driver->ResolveCompilingMethodsClass(soa, dex_cache, class_loader, &compilation_unit);
+}
+
+mirror::Class* HGraphBuilder::GetOutermostCompilingClass() const {
+  return GetClassFrom(compiler_driver_, *outer_compilation_unit_);
+}
+
+mirror::Class* HGraphBuilder::GetCompilingClass() const {
+  return GetClassFrom(compiler_driver_, *dex_compilation_unit_);
 }
 
 bool HGraphBuilder::IsOutermostCompilingClass(uint16_t type_index) const {
@@ -928,20 +937,20 @@
   const DexFile& outer_dex_file = *outer_compilation_unit_->GetDexFile();
   Handle<mirror::DexCache> outer_dex_cache(hs.NewHandle(
       outer_compilation_unit_->GetClassLinker()->FindDexCache(outer_dex_file)));
-  Handle<mirror::Class> referrer_class(hs.NewHandle(GetOutermostCompilingClass()));
+  Handle<mirror::Class> outer_class(hs.NewHandle(GetOutermostCompilingClass()));
 
   // The index at which the field's class is stored in the DexCache's type array.
   uint32_t storage_index;
-  bool is_referrer_class = (referrer_class.Get() == resolved_field->GetDeclaringClass());
-  if (is_referrer_class) {
-    storage_index = referrer_class->GetDexTypeIndex();
+  bool is_outer_class = (outer_class.Get() == resolved_field->GetDeclaringClass());
+  if (is_outer_class) {
+    storage_index = outer_class->GetDexTypeIndex();
   } else if (outer_dex_cache.Get() != dex_cache.Get()) {
     // The compiler driver cannot currently understand multiple dex caches involved. Just bailout.
     return false;
   } else {
     std::pair<bool, bool> pair = compiler_driver_->IsFastStaticField(
         outer_dex_cache.Get(),
-        referrer_class.Get(),
+        GetCompilingClass(),
         resolved_field,
         field_index,
         &storage_index);
@@ -959,12 +968,12 @@
   HLoadClass* constant = new (arena_) HLoadClass(graph_->GetCurrentMethod(),
                                                  storage_index,
                                                  *dex_compilation_unit_->GetDexFile(),
-                                                 is_referrer_class,
+                                                 is_outer_class,
                                                  dex_pc);
   current_block_->AddInstruction(constant);
 
   HInstruction* cls = constant;
-  if (!is_initialized && !is_referrer_class) {
+  if (!is_initialized && !is_outer_class) {
     cls = new (arena_) HClinitCheck(constant, dex_pc);
     current_block_->AddInstruction(cls);
   }
@@ -2100,27 +2109,27 @@
       break;
 
     case Instruction::CMP_LONG: {
-      Binop_23x_cmp(instruction, Primitive::kPrimLong, HCompare::kNoBias);
+      Binop_23x_cmp(instruction, Primitive::kPrimLong, HCompare::kNoBias, dex_pc);
       break;
     }
 
     case Instruction::CMPG_FLOAT: {
-      Binop_23x_cmp(instruction, Primitive::kPrimFloat, HCompare::kGtBias);
+      Binop_23x_cmp(instruction, Primitive::kPrimFloat, HCompare::kGtBias, dex_pc);
       break;
     }
 
     case Instruction::CMPG_DOUBLE: {
-      Binop_23x_cmp(instruction, Primitive::kPrimDouble, HCompare::kGtBias);
+      Binop_23x_cmp(instruction, Primitive::kPrimDouble, HCompare::kGtBias, dex_pc);
       break;
     }
 
     case Instruction::CMPL_FLOAT: {
-      Binop_23x_cmp(instruction, Primitive::kPrimFloat, HCompare::kLtBias);
+      Binop_23x_cmp(instruction, Primitive::kPrimFloat, HCompare::kLtBias, dex_pc);
       break;
     }
 
     case Instruction::CMPL_DOUBLE: {
-      Binop_23x_cmp(instruction, Primitive::kPrimDouble, HCompare::kLtBias);
+      Binop_23x_cmp(instruction, Primitive::kPrimDouble, HCompare::kLtBias, dex_pc);
       break;
     }
 
diff --git a/compiler/optimizing/builder.h b/compiler/optimizing/builder.h
index 36503ce..052aaf8 100644
--- a/compiler/optimizing/builder.h
+++ b/compiler/optimizing/builder.h
@@ -117,7 +117,10 @@
   template<typename T>
   void Binop_23x_shift(const Instruction& instruction, Primitive::Type type);
 
-  void Binop_23x_cmp(const Instruction& instruction, Primitive::Type type, HCompare::Bias bias);
+  void Binop_23x_cmp(const Instruction& instruction,
+                     Primitive::Type type,
+                     HCompare::Bias bias,
+                     uint32_t dex_pc);
 
   template<typename T>
   void Binop_12x(const Instruction& instruction, Primitive::Type type);
@@ -222,8 +225,12 @@
 
   void MaybeRecordStat(MethodCompilationStat compilation_stat);
 
+  // Returns the outer-most compiling method's class.
   mirror::Class* GetOutermostCompilingClass() const;
 
+  // Returns the class whose method is being compiled.
+  mirror::Class* GetCompilingClass() const;
+
   // Returns whether `type_index` points to the outer-most compiling method's class.
   bool IsOutermostCompilingClass(uint16_t type_index) const;
 
diff --git a/compiler/optimizing/code_generator.cc b/compiler/optimizing/code_generator.cc
index 09f7d86..64f2c9a 100644
--- a/compiler/optimizing/code_generator.cc
+++ b/compiler/optimizing/code_generator.cc
@@ -20,6 +20,7 @@
 #include "code_generator_arm64.h"
 #include "code_generator_x86.h"
 #include "code_generator_x86_64.h"
+#include "code_generator_mips64.h"
 #include "compiled_method.h"
 #include "dex/verified_method.h"
 #include "driver/dex_compilation_unit.h"
@@ -486,6 +487,11 @@
     }
     case kMips:
       return nullptr;
+    case kMips64: {
+      return new mips64::CodeGeneratorMIPS64(graph,
+          *isa_features.AsMips64InstructionSetFeatures(),
+          compiler_options);
+    }
     case kX86: {
       return new x86::CodeGeneratorX86(graph,
            *isa_features.AsX86InstructionSetFeatures(),
@@ -651,18 +657,18 @@
                                  uint32_t dex_pc,
                                  SlowPathCode* slow_path) {
   if (instruction != nullptr) {
-    // The code generated for some type conversions may call the
-    // runtime, thus normally requiring a subsequent call to this
-    // method.  However, the method verifier does not produce PC
-    // information for certain instructions, which are considered "atomic"
-    // (they cannot join a GC).
+    // The code generated for some type conversions and comparisons
+    // may call the runtime, thus normally requiring a subsequent
+    // call to this method. However, the method verifier does not
+    // produce PC information for certain instructions, which are
+    // considered "atomic" (they cannot join a GC).
     // Therefore we do not currently record PC information for such
     // instructions.  As this may change later, we added this special
     // case so that code generators may nevertheless call
     // CodeGenerator::RecordPcInfo without triggering an error in
     // CodeGenerator::BuildNativeGCMap ("Missing ref for dex pc 0x")
     // thereafter.
-    if (instruction->IsTypeConversion()) {
+    if (instruction->IsTypeConversion() || instruction->IsCompare()) {
       return;
     }
     if (instruction->IsRem()) {
diff --git a/compiler/optimizing/code_generator.h b/compiler/optimizing/code_generator.h
index 5b0abd7..b1f1674 100644
--- a/compiler/optimizing/code_generator.h
+++ b/compiler/optimizing/code_generator.h
@@ -97,6 +97,8 @@
     return saved_fpu_stack_offsets_[reg];
   }
 
+  virtual const char* GetDescription() const = 0;
+
  protected:
   static constexpr size_t kMaximumNumberOfExpectedRegisters = 32;
   static constexpr uint32_t kRegisterNotSaved = -1;
diff --git a/compiler/optimizing/code_generator_arm.cc b/compiler/optimizing/code_generator_arm.cc
index f6ae452..7169679 100644
--- a/compiler/optimizing/code_generator_arm.cc
+++ b/compiler/optimizing/code_generator_arm.cc
@@ -69,6 +69,8 @@
         QUICK_ENTRY_POINT(pThrowNullPointer), instruction_, instruction_->GetDexPc(), this);
   }
 
+  const char* GetDescription() const OVERRIDE { return "NullCheckSlowPathARM"; }
+
  private:
   HNullCheck* const instruction_;
   DISALLOW_COPY_AND_ASSIGN(NullCheckSlowPathARM);
@@ -85,6 +87,8 @@
         QUICK_ENTRY_POINT(pThrowDivZero), instruction_, instruction_->GetDexPc(), this);
   }
 
+  const char* GetDescription() const OVERRIDE { return "DivZeroCheckSlowPathARM"; }
+
  private:
   HDivZeroCheck* const instruction_;
   DISALLOW_COPY_AND_ASSIGN(DivZeroCheckSlowPathARM);
@@ -118,6 +122,8 @@
     return successor_;
   }
 
+  const char* GetDescription() const OVERRIDE { return "SuspendCheckSlowPathARM"; }
+
  private:
   HSuspendCheck* const instruction_;
   // If not null, the block to branch to after the suspend check.
@@ -155,6 +161,8 @@
         QUICK_ENTRY_POINT(pThrowArrayBounds), instruction_, instruction_->GetDexPc(), this);
   }
 
+  const char* GetDescription() const OVERRIDE { return "BoundsCheckSlowPathARM"; }
+
  private:
   HBoundsCheck* const instruction_;
   const Location index_location_;
@@ -197,6 +205,8 @@
     __ b(GetExitLabel());
   }
 
+  const char* GetDescription() const OVERRIDE { return "LoadClassSlowPathARM"; }
+
  private:
   // The class this slow path will load.
   HLoadClass* const cls_;
@@ -236,6 +246,8 @@
     __ b(GetExitLabel());
   }
 
+  const char* GetDescription() const OVERRIDE { return "LoadStringSlowPathARM"; }
+
  private:
   HLoadString* const instruction_;
 
@@ -286,6 +298,8 @@
     __ b(GetExitLabel());
   }
 
+  const char* GetDescription() const OVERRIDE { return "TypeCheckSlowPathARM"; }
+
  private:
   HInstruction* const instruction_;
   const Location class_to_check_;
@@ -310,6 +324,8 @@
     arm_codegen->InvokeRuntime(QUICK_ENTRY_POINT(pDeoptimize), instruction_, dex_pc, this);
   }
 
+  const char* GetDescription() const OVERRIDE { return "DeoptimizationSlowPathARM"; }
+
  private:
   HInstruction* const instruction_;
   DISALLOW_COPY_AND_ASSIGN(DeoptimizationSlowPathARM);
diff --git a/compiler/optimizing/code_generator_arm64.cc b/compiler/optimizing/code_generator_arm64.cc
index 3c8f117..7ec6b54 100644
--- a/compiler/optimizing/code_generator_arm64.cc
+++ b/compiler/optimizing/code_generator_arm64.cc
@@ -213,6 +213,8 @@
     CheckEntrypointTypes<kQuickThrowArrayBounds, void, int32_t, int32_t>();
   }
 
+  const char* GetDescription() const OVERRIDE { return "BoundsCheckSlowPathARM64"; }
+
  private:
   HBoundsCheck* const instruction_;
   const Location index_location_;
@@ -233,6 +235,8 @@
     CheckEntrypointTypes<kQuickThrowDivZero, void, void>();
   }
 
+  const char* GetDescription() const OVERRIDE { return "DivZeroCheckSlowPathARM64"; }
+
  private:
   HDivZeroCheck* const instruction_;
   DISALLOW_COPY_AND_ASSIGN(DivZeroCheckSlowPathARM64);
@@ -278,6 +282,8 @@
     __ B(GetExitLabel());
   }
 
+  const char* GetDescription() const OVERRIDE { return "LoadClassSlowPathARM64"; }
+
  private:
   // The class this slow path will load.
   HLoadClass* const cls_;
@@ -319,6 +325,8 @@
     __ B(GetExitLabel());
   }
 
+  const char* GetDescription() const OVERRIDE { return "LoadStringSlowPathARM64"; }
+
  private:
   HLoadString* const instruction_;
 
@@ -337,6 +345,8 @@
     CheckEntrypointTypes<kQuickThrowNullPointer, void, void>();
   }
 
+  const char* GetDescription() const OVERRIDE { return "NullCheckSlowPathARM64"; }
+
  private:
   HNullCheck* const instruction_;
 
@@ -373,6 +383,8 @@
     return successor_;
   }
 
+  const char* GetDescription() const OVERRIDE { return "SuspendCheckSlowPathARM64"; }
+
  private:
   HSuspendCheck* const instruction_;
   // If not null, the block to branch to after the suspend check.
@@ -429,6 +441,8 @@
     __ B(GetExitLabel());
   }
 
+  const char* GetDescription() const OVERRIDE { return "TypeCheckSlowPathARM64"; }
+
  private:
   HInstruction* const instruction_;
   const Location class_to_check_;
@@ -453,6 +467,8 @@
     arm64_codegen->InvokeRuntime(QUICK_ENTRY_POINT(pDeoptimize), instruction_, dex_pc, this);
   }
 
+  const char* GetDescription() const OVERRIDE { return "DeoptimizationSlowPathARM64"; }
+
  private:
   HInstruction* const instruction_;
   DISALLOW_COPY_AND_ASSIGN(DeoptimizationSlowPathARM64);
diff --git a/compiler/optimizing/code_generator_mips64.cc b/compiler/optimizing/code_generator_mips64.cc
new file mode 100644
index 0000000..ab684d4
--- /dev/null
+++ b/compiler/optimizing/code_generator_mips64.cc
@@ -0,0 +1,3282 @@
+/*
+ * Copyright (C) 2015 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 "code_generator_mips64.h"
+
+#include "entrypoints/quick/quick_entrypoints.h"
+#include "entrypoints/quick/quick_entrypoints_enum.h"
+#include "gc/accounting/card_table.h"
+#include "intrinsics.h"
+#include "art_method.h"
+#include "mirror/array-inl.h"
+#include "mirror/class-inl.h"
+#include "offsets.h"
+#include "thread.h"
+#include "utils/mips64/assembler_mips64.h"
+#include "utils/assembler.h"
+#include "utils/stack_checks.h"
+
+namespace art {
+namespace mips64 {
+
+static constexpr int kCurrentMethodStackOffset = 0;
+static constexpr GpuRegister kMethodRegisterArgument = A0;
+
+// We need extra temporary/scratch registers (in addition to AT) in some cases.
+static constexpr GpuRegister TMP = T8;
+static constexpr FpuRegister FTMP = F8;
+
+// ART Thread Register.
+static constexpr GpuRegister TR = S1;
+
+Location Mips64ReturnLocation(Primitive::Type return_type) {
+  switch (return_type) {
+    case Primitive::kPrimBoolean:
+    case Primitive::kPrimByte:
+    case Primitive::kPrimChar:
+    case Primitive::kPrimShort:
+    case Primitive::kPrimInt:
+    case Primitive::kPrimNot:
+    case Primitive::kPrimLong:
+      return Location::RegisterLocation(V0);
+
+    case Primitive::kPrimFloat:
+    case Primitive::kPrimDouble:
+      return Location::FpuRegisterLocation(F0);
+
+    case Primitive::kPrimVoid:
+      return Location();
+  }
+  UNREACHABLE();
+}
+
+Location InvokeDexCallingConventionVisitorMIPS64::GetReturnLocation(Primitive::Type type) const {
+  return Mips64ReturnLocation(type);
+}
+
+Location InvokeDexCallingConventionVisitorMIPS64::GetMethodLocation() const {
+  return Location::RegisterLocation(kMethodRegisterArgument);
+}
+
+Location InvokeDexCallingConventionVisitorMIPS64::GetNextLocation(Primitive::Type type) {
+  Location next_location;
+  if (type == Primitive::kPrimVoid) {
+    LOG(FATAL) << "Unexpected parameter type " << type;
+  }
+
+  if (Primitive::IsFloatingPointType(type) &&
+      (float_index_ < calling_convention.GetNumberOfFpuRegisters())) {
+    next_location = Location::FpuRegisterLocation(
+        calling_convention.GetFpuRegisterAt(float_index_++));
+    gp_index_++;
+  } else if (!Primitive::IsFloatingPointType(type) &&
+             (gp_index_ < calling_convention.GetNumberOfRegisters())) {
+    next_location = Location::RegisterLocation(calling_convention.GetRegisterAt(gp_index_++));
+    float_index_++;
+  } else {
+    size_t stack_offset = calling_convention.GetStackOffsetOf(stack_index_);
+    next_location = Primitive::Is64BitType(type) ? Location::DoubleStackSlot(stack_offset)
+                                                 : Location::StackSlot(stack_offset);
+  }
+
+  // Space on the stack is reserved for all arguments.
+  stack_index_ += Primitive::Is64BitType(type) ? 2 : 1;
+
+  // TODO: review
+
+  // TODO: shouldn't we use a whole machine word per argument on the stack?
+  // Implicit 4-byte method pointer (and such) will cause misalignment.
+
+  return next_location;
+}
+
+Location InvokeRuntimeCallingConvention::GetReturnLocation(Primitive::Type type) {
+  return Mips64ReturnLocation(type);
+}
+
+#define __ down_cast<CodeGeneratorMIPS64*>(codegen)->GetAssembler()->
+#define QUICK_ENTRY_POINT(x) QUICK_ENTRYPOINT_OFFSET(kMips64WordSize, x).Int32Value()
+
+class BoundsCheckSlowPathMIPS64 : public SlowPathCodeMIPS64 {
+ public:
+  BoundsCheckSlowPathMIPS64(HBoundsCheck* instruction,
+                            Location index_location,
+                            Location length_location)
+      : instruction_(instruction),
+        index_location_(index_location),
+        length_location_(length_location) {}
+
+  void EmitNativeCode(CodeGenerator* codegen) OVERRIDE {
+    CodeGeneratorMIPS64* mips64_codegen = down_cast<CodeGeneratorMIPS64*>(codegen);
+    __ Bind(GetEntryLabel());
+    // We're moving two locations to locations that could overlap, so we need a parallel
+    // move resolver.
+    InvokeRuntimeCallingConvention calling_convention;
+    codegen->EmitParallelMoves(index_location_,
+                               Location::RegisterLocation(calling_convention.GetRegisterAt(0)),
+                               Primitive::kPrimInt,
+                               length_location_,
+                               Location::RegisterLocation(calling_convention.GetRegisterAt(1)),
+                               Primitive::kPrimInt);
+    mips64_codegen->InvokeRuntime(QUICK_ENTRY_POINT(pThrowArrayBounds),
+                                  instruction_,
+                                  instruction_->GetDexPc(),
+                                  this);
+    CheckEntrypointTypes<kQuickThrowArrayBounds, void, int32_t, int32_t>();
+  }
+
+  const char* GetDescription() const OVERRIDE { return "BoundsCheckSlowPathMIPS64"; }
+
+ private:
+  HBoundsCheck* const instruction_;
+  const Location index_location_;
+  const Location length_location_;
+
+  DISALLOW_COPY_AND_ASSIGN(BoundsCheckSlowPathMIPS64);
+};
+
+class DivZeroCheckSlowPathMIPS64 : public SlowPathCodeMIPS64 {
+ public:
+  explicit DivZeroCheckSlowPathMIPS64(HDivZeroCheck* instruction) : instruction_(instruction) {}
+
+  void EmitNativeCode(CodeGenerator* codegen) OVERRIDE {
+    CodeGeneratorMIPS64* mips64_codegen = down_cast<CodeGeneratorMIPS64*>(codegen);
+    __ Bind(GetEntryLabel());
+    mips64_codegen->InvokeRuntime(QUICK_ENTRY_POINT(pThrowDivZero),
+                                  instruction_,
+                                  instruction_->GetDexPc(),
+                                  this);
+    CheckEntrypointTypes<kQuickThrowDivZero, void, void>();
+  }
+
+  const char* GetDescription() const OVERRIDE { return "DivZeroCheckSlowPathMIPS64"; }
+
+ private:
+  HDivZeroCheck* const instruction_;
+  DISALLOW_COPY_AND_ASSIGN(DivZeroCheckSlowPathMIPS64);
+};
+
+class LoadClassSlowPathMIPS64 : public SlowPathCodeMIPS64 {
+ public:
+  LoadClassSlowPathMIPS64(HLoadClass* cls,
+                          HInstruction* at,
+                          uint32_t dex_pc,
+                          bool do_clinit)
+      : cls_(cls), at_(at), dex_pc_(dex_pc), do_clinit_(do_clinit) {
+    DCHECK(at->IsLoadClass() || at->IsClinitCheck());
+  }
+
+  void EmitNativeCode(CodeGenerator* codegen) OVERRIDE {
+    LocationSummary* locations = at_->GetLocations();
+    CodeGeneratorMIPS64* mips64_codegen = down_cast<CodeGeneratorMIPS64*>(codegen);
+
+    __ Bind(GetEntryLabel());
+    SaveLiveRegisters(codegen, locations);
+
+    InvokeRuntimeCallingConvention calling_convention;
+    __ LoadConst32(calling_convention.GetRegisterAt(0), cls_->GetTypeIndex());
+    int32_t entry_point_offset = do_clinit_ ? QUICK_ENTRY_POINT(pInitializeStaticStorage)
+                                            : QUICK_ENTRY_POINT(pInitializeType);
+    mips64_codegen->InvokeRuntime(entry_point_offset, at_, dex_pc_, this);
+    if (do_clinit_) {
+      CheckEntrypointTypes<kQuickInitializeStaticStorage, void*, uint32_t>();
+    } else {
+      CheckEntrypointTypes<kQuickInitializeType, void*, uint32_t>();
+    }
+
+    // Move the class to the desired location.
+    Location out = locations->Out();
+    if (out.IsValid()) {
+      DCHECK(out.IsRegister() && !locations->GetLiveRegisters()->ContainsCoreRegister(out.reg()));
+      Primitive::Type type = at_->GetType();
+      mips64_codegen->MoveLocation(out, calling_convention.GetReturnLocation(type), type);
+    }
+
+    RestoreLiveRegisters(codegen, locations);
+    __ B(GetExitLabel());
+  }
+
+  const char* GetDescription() const OVERRIDE { return "LoadClassSlowPathMIPS64"; }
+
+ private:
+  // The class this slow path will load.
+  HLoadClass* const cls_;
+
+  // The instruction where this slow path is happening.
+  // (Might be the load class or an initialization check).
+  HInstruction* const at_;
+
+  // The dex PC of `at_`.
+  const uint32_t dex_pc_;
+
+  // Whether to initialize the class.
+  const bool do_clinit_;
+
+  DISALLOW_COPY_AND_ASSIGN(LoadClassSlowPathMIPS64);
+};
+
+class LoadStringSlowPathMIPS64 : public SlowPathCodeMIPS64 {
+ public:
+  explicit LoadStringSlowPathMIPS64(HLoadString* instruction) : instruction_(instruction) {}
+
+  void EmitNativeCode(CodeGenerator* codegen) OVERRIDE {
+    LocationSummary* locations = instruction_->GetLocations();
+    DCHECK(!locations->GetLiveRegisters()->ContainsCoreRegister(locations->Out().reg()));
+    CodeGeneratorMIPS64* mips64_codegen = down_cast<CodeGeneratorMIPS64*>(codegen);
+
+    __ Bind(GetEntryLabel());
+    SaveLiveRegisters(codegen, locations);
+
+    InvokeRuntimeCallingConvention calling_convention;
+    __ LoadConst32(calling_convention.GetRegisterAt(0), instruction_->GetStringIndex());
+    mips64_codegen->InvokeRuntime(QUICK_ENTRY_POINT(pResolveString),
+                                  instruction_,
+                                  instruction_->GetDexPc(),
+                                  this);
+    CheckEntrypointTypes<kQuickResolveString, void*, uint32_t>();
+    Primitive::Type type = instruction_->GetType();
+    mips64_codegen->MoveLocation(locations->Out(),
+                                 calling_convention.GetReturnLocation(type),
+                                 type);
+
+    RestoreLiveRegisters(codegen, locations);
+    __ B(GetExitLabel());
+  }
+
+  const char* GetDescription() const OVERRIDE { return "LoadStringSlowPathMIPS64"; }
+
+ private:
+  HLoadString* const instruction_;
+
+  DISALLOW_COPY_AND_ASSIGN(LoadStringSlowPathMIPS64);
+};
+
+class NullCheckSlowPathMIPS64 : public SlowPathCodeMIPS64 {
+ public:
+  explicit NullCheckSlowPathMIPS64(HNullCheck* instr) : instruction_(instr) {}
+
+  void EmitNativeCode(CodeGenerator* codegen) OVERRIDE {
+    CodeGeneratorMIPS64* mips64_codegen = down_cast<CodeGeneratorMIPS64*>(codegen);
+    __ Bind(GetEntryLabel());
+    mips64_codegen->InvokeRuntime(QUICK_ENTRY_POINT(pThrowNullPointer),
+                                  instruction_,
+                                  instruction_->GetDexPc(),
+                                  this);
+    CheckEntrypointTypes<kQuickThrowNullPointer, void, void>();
+  }
+
+  const char* GetDescription() const OVERRIDE { return "NullCheckSlowPathMIPS64"; }
+
+ private:
+  HNullCheck* const instruction_;
+
+  DISALLOW_COPY_AND_ASSIGN(NullCheckSlowPathMIPS64);
+};
+
+class SuspendCheckSlowPathMIPS64 : public SlowPathCodeMIPS64 {
+ public:
+  explicit SuspendCheckSlowPathMIPS64(HSuspendCheck* instruction,
+                                      HBasicBlock* successor)
+      : instruction_(instruction), successor_(successor) {}
+
+  void EmitNativeCode(CodeGenerator* codegen) OVERRIDE {
+    CodeGeneratorMIPS64* mips64_codegen = down_cast<CodeGeneratorMIPS64*>(codegen);
+    __ Bind(GetEntryLabel());
+    SaveLiveRegisters(codegen, instruction_->GetLocations());
+    mips64_codegen->InvokeRuntime(QUICK_ENTRY_POINT(pTestSuspend),
+                                  instruction_,
+                                  instruction_->GetDexPc(),
+                                  this);
+    CheckEntrypointTypes<kQuickTestSuspend, void, void>();
+    RestoreLiveRegisters(codegen, instruction_->GetLocations());
+    if (successor_ == nullptr) {
+      __ B(GetReturnLabel());
+    } else {
+      __ B(mips64_codegen->GetLabelOf(successor_));
+    }
+  }
+
+  Label* GetReturnLabel() {
+    DCHECK(successor_ == nullptr);
+    return &return_label_;
+  }
+
+  const char* GetDescription() const OVERRIDE { return "SuspendCheckSlowPathMIPS64"; }
+
+ private:
+  HSuspendCheck* const instruction_;
+  // If not null, the block to branch to after the suspend check.
+  HBasicBlock* const successor_;
+
+  // If `successor_` is null, the label to branch to after the suspend check.
+  Label return_label_;
+
+  DISALLOW_COPY_AND_ASSIGN(SuspendCheckSlowPathMIPS64);
+};
+
+class TypeCheckSlowPathMIPS64 : public SlowPathCodeMIPS64 {
+ public:
+  TypeCheckSlowPathMIPS64(HInstruction* instruction,
+                          Location class_to_check,
+                          Location object_class,
+                          uint32_t dex_pc)
+      : instruction_(instruction),
+        class_to_check_(class_to_check),
+        object_class_(object_class),
+        dex_pc_(dex_pc) {}
+
+  void EmitNativeCode(CodeGenerator* codegen) OVERRIDE {
+    LocationSummary* locations = instruction_->GetLocations();
+    DCHECK(instruction_->IsCheckCast()
+           || !locations->GetLiveRegisters()->ContainsCoreRegister(locations->Out().reg()));
+    CodeGeneratorMIPS64* mips64_codegen = down_cast<CodeGeneratorMIPS64*>(codegen);
+
+    __ Bind(GetEntryLabel());
+    SaveLiveRegisters(codegen, locations);
+
+    // We're moving two locations to locations that could overlap, so we need a parallel
+    // move resolver.
+    InvokeRuntimeCallingConvention calling_convention;
+    codegen->EmitParallelMoves(class_to_check_,
+                               Location::RegisterLocation(calling_convention.GetRegisterAt(0)),
+                               Primitive::kPrimNot,
+                               object_class_,
+                               Location::RegisterLocation(calling_convention.GetRegisterAt(1)),
+                               Primitive::kPrimNot);
+
+    if (instruction_->IsInstanceOf()) {
+      mips64_codegen->InvokeRuntime(QUICK_ENTRY_POINT(pInstanceofNonTrivial),
+                                    instruction_,
+                                    dex_pc_,
+                                    this);
+      Primitive::Type ret_type = instruction_->GetType();
+      Location ret_loc = calling_convention.GetReturnLocation(ret_type);
+      mips64_codegen->MoveLocation(locations->Out(), ret_loc, ret_type);
+      CheckEntrypointTypes<kQuickInstanceofNonTrivial,
+                           uint32_t,
+                           const mirror::Class*,
+                           const mirror::Class*>();
+    } else {
+      DCHECK(instruction_->IsCheckCast());
+      mips64_codegen->InvokeRuntime(QUICK_ENTRY_POINT(pCheckCast), instruction_, dex_pc_, this);
+      CheckEntrypointTypes<kQuickCheckCast, void, const mirror::Class*, const mirror::Class*>();
+    }
+
+    RestoreLiveRegisters(codegen, locations);
+    __ B(GetExitLabel());
+  }
+
+  const char* GetDescription() const OVERRIDE { return "TypeCheckSlowPathMIPS64"; }
+
+ private:
+  HInstruction* const instruction_;
+  const Location class_to_check_;
+  const Location object_class_;
+  uint32_t dex_pc_;
+
+  DISALLOW_COPY_AND_ASSIGN(TypeCheckSlowPathMIPS64);
+};
+
+class DeoptimizationSlowPathMIPS64 : public SlowPathCodeMIPS64 {
+ public:
+  explicit DeoptimizationSlowPathMIPS64(HInstruction* instruction)
+    : instruction_(instruction) {}
+
+  void EmitNativeCode(CodeGenerator* codegen) OVERRIDE {
+    __ Bind(GetEntryLabel());
+    SaveLiveRegisters(codegen, instruction_->GetLocations());
+    DCHECK(instruction_->IsDeoptimize());
+    HDeoptimize* deoptimize = instruction_->AsDeoptimize();
+    uint32_t dex_pc = deoptimize->GetDexPc();
+    CodeGeneratorMIPS64* mips64_codegen = down_cast<CodeGeneratorMIPS64*>(codegen);
+    mips64_codegen->InvokeRuntime(QUICK_ENTRY_POINT(pDeoptimize), instruction_, dex_pc, this);
+  }
+
+  const char* GetDescription() const OVERRIDE { return "DeoptimizationSlowPathMIPS64"; }
+
+ private:
+  HInstruction* const instruction_;
+  DISALLOW_COPY_AND_ASSIGN(DeoptimizationSlowPathMIPS64);
+};
+
+CodeGeneratorMIPS64::CodeGeneratorMIPS64(HGraph* graph,
+                                         const Mips64InstructionSetFeatures& isa_features,
+                                         const CompilerOptions& compiler_options)
+    : CodeGenerator(graph,
+                    kNumberOfGpuRegisters,
+                    kNumberOfFpuRegisters,
+                    0,  // kNumberOfRegisterPairs
+                    ComputeRegisterMask(reinterpret_cast<const int*>(kCoreCalleeSaves),
+                                        arraysize(kCoreCalleeSaves)),
+                    ComputeRegisterMask(reinterpret_cast<const int*>(kFpuCalleeSaves),
+                                        arraysize(kFpuCalleeSaves)),
+                    compiler_options),
+      block_labels_(graph->GetArena(), 0),
+      location_builder_(graph, this),
+      instruction_visitor_(graph, this),
+      move_resolver_(graph->GetArena(), this),
+      isa_features_(isa_features) {
+  // Save RA (containing the return address) to mimic Quick.
+  AddAllocatedRegister(Location::RegisterLocation(RA));
+}
+
+#undef __
+#define __ down_cast<Mips64Assembler*>(GetAssembler())->
+#define QUICK_ENTRY_POINT(x) QUICK_ENTRYPOINT_OFFSET(kMips64WordSize, x).Int32Value()
+
+void CodeGeneratorMIPS64::Finalize(CodeAllocator* allocator) {
+  CodeGenerator::Finalize(allocator);
+}
+
+Mips64Assembler* ParallelMoveResolverMIPS64::GetAssembler() const {
+  return codegen_->GetAssembler();
+}
+
+void ParallelMoveResolverMIPS64::EmitMove(size_t index) {
+  MoveOperands* move = moves_.Get(index);
+  codegen_->MoveLocation(move->GetDestination(), move->GetSource(), move->GetType());
+}
+
+void ParallelMoveResolverMIPS64::EmitSwap(size_t index) {
+  MoveOperands* move = moves_.Get(index);
+  codegen_->SwapLocations(move->GetDestination(), move->GetSource(), move->GetType());
+}
+
+void ParallelMoveResolverMIPS64::RestoreScratch(int reg) {
+  // Pop reg
+  __ Ld(GpuRegister(reg), SP, 0);
+  __ DecreaseFrameSize(kMips64WordSize);
+}
+
+void ParallelMoveResolverMIPS64::SpillScratch(int reg) {
+  // Push reg
+  __ IncreaseFrameSize(kMips64WordSize);
+  __ Sd(GpuRegister(reg), SP, 0);
+}
+
+void ParallelMoveResolverMIPS64::Exchange(int index1, int index2, bool double_slot) {
+  LoadOperandType load_type = double_slot ? kLoadDoubleword : kLoadWord;
+  StoreOperandType store_type = double_slot ? kStoreDoubleword : kStoreWord;
+  // Allocate a scratch register other than TMP, if available.
+  // Else, spill V0 (arbitrary choice) and use it as a scratch register (it will be
+  // automatically unspilled when the scratch scope object is destroyed).
+  ScratchRegisterScope ensure_scratch(this, TMP, V0, codegen_->GetNumberOfCoreRegisters());
+  // If V0 spills onto the stack, SP-relative offsets need to be adjusted.
+  int stack_offset = ensure_scratch.IsSpilled() ? kMips64WordSize : 0;
+  __ LoadFromOffset(load_type,
+                    GpuRegister(ensure_scratch.GetRegister()),
+                    SP,
+                    index1 + stack_offset);
+  __ LoadFromOffset(load_type,
+                    TMP,
+                    SP,
+                    index2 + stack_offset);
+  __ StoreToOffset(store_type,
+                   GpuRegister(ensure_scratch.GetRegister()),
+                   SP,
+                   index2 + stack_offset);
+  __ StoreToOffset(store_type, TMP, SP, index1 + stack_offset);
+}
+
+static dwarf::Reg DWARFReg(GpuRegister reg) {
+  return dwarf::Reg::Mips64Core(static_cast<int>(reg));
+}
+
+// TODO: mapping of floating-point registers to DWARF
+
+void CodeGeneratorMIPS64::GenerateFrameEntry() {
+  __ Bind(&frame_entry_label_);
+
+  bool do_overflow_check = FrameNeedsStackCheck(GetFrameSize(), kMips64) || !IsLeafMethod();
+
+  if (do_overflow_check) {
+    __ LoadFromOffset(kLoadWord,
+                      ZERO,
+                      SP,
+                      -static_cast<int32_t>(GetStackOverflowReservedBytes(kMips64)));
+    RecordPcInfo(nullptr, 0);
+  }
+
+  // TODO: anything related to T9/GP/GOT/PIC/.so's?
+
+  if (HasEmptyFrame()) {
+    return;
+  }
+
+  // Make sure the frame size isn't unreasonably large. Per the various APIs
+  // it looks like it should always be less than 2GB in size, which allows
+  // us using 32-bit signed offsets from the stack pointer.
+  if (GetFrameSize() > 0x7FFFFFFF)
+    LOG(FATAL) << "Stack frame larger than 2GB";
+
+  // Spill callee-saved registers.
+  // Note that their cumulative size is small and they can be indexed using
+  // 16-bit offsets.
+
+  // TODO: increment/decrement SP in one step instead of two or remove this comment.
+
+  uint32_t ofs = FrameEntrySpillSize();
+  __ IncreaseFrameSize(ofs);
+
+  for (int i = arraysize(kCoreCalleeSaves) - 1; i >= 0; --i) {
+    GpuRegister reg = kCoreCalleeSaves[i];
+    if (allocated_registers_.ContainsCoreRegister(reg)) {
+      ofs -= kMips64WordSize;
+      __ Sd(reg, SP, ofs);
+      __ cfi().RelOffset(DWARFReg(reg), ofs);
+    }
+  }
+
+  for (int i = arraysize(kFpuCalleeSaves) - 1; i >= 0; --i) {
+    FpuRegister reg = kFpuCalleeSaves[i];
+    if (allocated_registers_.ContainsFloatingPointRegister(reg)) {
+      ofs -= kMips64WordSize;
+      __ Sdc1(reg, SP, ofs);
+      // TODO: __ cfi().RelOffset(DWARFReg(reg), ofs);
+    }
+  }
+
+  // Allocate the rest of the frame and store the current method pointer
+  // at its end.
+
+  __ IncreaseFrameSize(GetFrameSize() - FrameEntrySpillSize());
+
+  static_assert(IsInt<16>(kCurrentMethodStackOffset),
+                "kCurrentMethodStackOffset must fit into int16_t");
+  __ Sd(kMethodRegisterArgument, SP, kCurrentMethodStackOffset);
+}
+
+void CodeGeneratorMIPS64::GenerateFrameExit() {
+  __ cfi().RememberState();
+
+  // TODO: anything related to T9/GP/GOT/PIC/.so's?
+
+  if (!HasEmptyFrame()) {
+    // Deallocate the rest of the frame.
+
+    __ DecreaseFrameSize(GetFrameSize() - FrameEntrySpillSize());
+
+    // Restore callee-saved registers.
+    // Note that their cumulative size is small and they can be indexed using
+    // 16-bit offsets.
+
+    // TODO: increment/decrement SP in one step instead of two or remove this comment.
+
+    uint32_t ofs = 0;
+
+    for (size_t i = 0; i < arraysize(kFpuCalleeSaves); ++i) {
+      FpuRegister reg = kFpuCalleeSaves[i];
+      if (allocated_registers_.ContainsFloatingPointRegister(reg)) {
+        __ Ldc1(reg, SP, ofs);
+        ofs += kMips64WordSize;
+        // TODO: __ cfi().Restore(DWARFReg(reg));
+      }
+    }
+
+    for (size_t i = 0; i < arraysize(kCoreCalleeSaves); ++i) {
+      GpuRegister reg = kCoreCalleeSaves[i];
+      if (allocated_registers_.ContainsCoreRegister(reg)) {
+        __ Ld(reg, SP, ofs);
+        ofs += kMips64WordSize;
+        __ cfi().Restore(DWARFReg(reg));
+      }
+    }
+
+    DCHECK_EQ(ofs, FrameEntrySpillSize());
+    __ DecreaseFrameSize(ofs);
+  }
+
+  __ Jr(RA);
+
+  __ cfi().RestoreState();
+  __ cfi().DefCFAOffset(GetFrameSize());
+}
+
+void CodeGeneratorMIPS64::Bind(HBasicBlock* block) {
+  __ Bind(GetLabelOf(block));
+}
+
+void CodeGeneratorMIPS64::MoveLocation(Location destination,
+                                       Location source,
+                                       Primitive::Type type) {
+  if (source.Equals(destination)) {
+    return;
+  }
+
+  // A valid move can always be inferred from the destination and source
+  // locations. When moving from and to a register, the argument type can be
+  // used to generate 32bit instead of 64bit moves.
+  bool unspecified_type = (type == Primitive::kPrimVoid);
+  DCHECK_EQ(unspecified_type, false);
+
+  if (destination.IsRegister() || destination.IsFpuRegister()) {
+    if (unspecified_type) {
+      HConstant* src_cst = source.IsConstant() ? source.GetConstant() : nullptr;
+      if (source.IsStackSlot() ||
+          (src_cst != nullptr && (src_cst->IsIntConstant()
+                                  || src_cst->IsFloatConstant()
+                                  || src_cst->IsNullConstant()))) {
+        // For stack slots and 32bit constants, a 64bit type is appropriate.
+        type = destination.IsRegister() ? Primitive::kPrimInt : Primitive::kPrimFloat;
+      } else {
+        // If the source is a double stack slot or a 64bit constant, a 64bit
+        // type is appropriate. Else the source is a register, and since the
+        // type has not been specified, we chose a 64bit type to force a 64bit
+        // move.
+        type = destination.IsRegister() ? Primitive::kPrimLong : Primitive::kPrimDouble;
+      }
+    }
+    DCHECK((destination.IsFpuRegister() && Primitive::IsFloatingPointType(type)) ||
+           (destination.IsRegister() && !Primitive::IsFloatingPointType(type)));
+    if (source.IsStackSlot() || source.IsDoubleStackSlot()) {
+      // Move to GPR/FPR from stack
+      LoadOperandType load_type = source.IsStackSlot() ? kLoadWord : kLoadDoubleword;
+      if (Primitive::IsFloatingPointType(type)) {
+        __ LoadFpuFromOffset(load_type,
+                             destination.AsFpuRegister<FpuRegister>(),
+                             SP,
+                             source.GetStackIndex());
+      } else {
+        // TODO: use load_type = kLoadUnsignedWord when type == Primitive::kPrimNot.
+        __ LoadFromOffset(load_type,
+                          destination.AsRegister<GpuRegister>(),
+                          SP,
+                          source.GetStackIndex());
+      }
+    } else if (source.IsConstant()) {
+      // Move to GPR/FPR from constant
+      GpuRegister gpr = AT;
+      if (!Primitive::IsFloatingPointType(type)) {
+        gpr = destination.AsRegister<GpuRegister>();
+      }
+      if (type == Primitive::kPrimInt || type == Primitive::kPrimFloat) {
+        __ LoadConst32(gpr, GetInt32ValueOf(source.GetConstant()->AsConstant()));
+      } else {
+        __ LoadConst64(gpr, GetInt64ValueOf(source.GetConstant()->AsConstant()));
+      }
+      if (type == Primitive::kPrimFloat) {
+        __ Mtc1(gpr, destination.AsFpuRegister<FpuRegister>());
+      } else if (type == Primitive::kPrimDouble) {
+        __ Dmtc1(gpr, destination.AsFpuRegister<FpuRegister>());
+      }
+    } else {
+      if (destination.IsRegister()) {
+        // Move to GPR from GPR
+        __ Move(destination.AsRegister<GpuRegister>(), source.AsRegister<GpuRegister>());
+      } else {
+        // Move to FPR from FPR
+        if (type == Primitive::kPrimFloat) {
+          __ MovS(destination.AsFpuRegister<FpuRegister>(), source.AsFpuRegister<FpuRegister>());
+        } else {
+          DCHECK_EQ(type, Primitive::kPrimDouble);
+          __ MovD(destination.AsFpuRegister<FpuRegister>(), source.AsFpuRegister<FpuRegister>());
+        }
+      }
+    }
+  } else {  // The destination is not a register. It must be a stack slot.
+    DCHECK(destination.IsStackSlot() || destination.IsDoubleStackSlot());
+    if (source.IsRegister() || source.IsFpuRegister()) {
+      if (unspecified_type) {
+        if (source.IsRegister()) {
+          type = destination.IsStackSlot() ? Primitive::kPrimInt : Primitive::kPrimLong;
+        } else {
+          type = destination.IsStackSlot() ? Primitive::kPrimFloat : Primitive::kPrimDouble;
+        }
+      }
+      DCHECK((destination.IsDoubleStackSlot() == Primitive::Is64BitType(type)) &&
+             (source.IsFpuRegister() == Primitive::IsFloatingPointType(type)));
+      // Move to stack from GPR/FPR
+      StoreOperandType store_type = destination.IsStackSlot() ? kStoreWord : kStoreDoubleword;
+      if (source.IsRegister()) {
+        __ StoreToOffset(store_type,
+                         source.AsRegister<GpuRegister>(),
+                         SP,
+                         destination.GetStackIndex());
+      } else {
+        __ StoreFpuToOffset(store_type,
+                            source.AsFpuRegister<FpuRegister>(),
+                            SP,
+                            destination.GetStackIndex());
+      }
+    } else if (source.IsConstant()) {
+      // Move to stack from constant
+      HConstant* src_cst = source.GetConstant();
+      StoreOperandType store_type = destination.IsStackSlot() ? kStoreWord : kStoreDoubleword;
+      if (destination.IsStackSlot()) {
+        __ LoadConst32(TMP, GetInt32ValueOf(src_cst->AsConstant()));
+      } else {
+        __ LoadConst64(TMP, GetInt64ValueOf(src_cst->AsConstant()));
+      }
+      __ StoreToOffset(store_type, TMP, SP, destination.GetStackIndex());
+    } else {
+      DCHECK(source.IsStackSlot() || source.IsDoubleStackSlot());
+      DCHECK_EQ(source.IsDoubleStackSlot(), destination.IsDoubleStackSlot());
+      // Move to stack from stack
+      if (destination.IsStackSlot()) {
+        __ LoadFromOffset(kLoadWord, TMP, SP, source.GetStackIndex());
+        __ StoreToOffset(kStoreWord, TMP, SP, destination.GetStackIndex());
+      } else {
+        __ LoadFromOffset(kLoadDoubleword, TMP, SP, source.GetStackIndex());
+        __ StoreToOffset(kStoreDoubleword, TMP, SP, destination.GetStackIndex());
+      }
+    }
+  }
+}
+
+void CodeGeneratorMIPS64::SwapLocations(Location loc1,
+                                        Location loc2,
+                                        Primitive::Type type ATTRIBUTE_UNUSED) {
+  DCHECK(!loc1.IsConstant());
+  DCHECK(!loc2.IsConstant());
+
+  if (loc1.Equals(loc2)) {
+    return;
+  }
+
+  bool is_slot1 = loc1.IsStackSlot() || loc1.IsDoubleStackSlot();
+  bool is_slot2 = loc2.IsStackSlot() || loc2.IsDoubleStackSlot();
+  bool is_fp_reg1 = loc1.IsFpuRegister();
+  bool is_fp_reg2 = loc2.IsFpuRegister();
+
+  if (loc2.IsRegister() && loc1.IsRegister()) {
+    // Swap 2 GPRs
+    GpuRegister r1 = loc1.AsRegister<GpuRegister>();
+    GpuRegister r2 = loc2.AsRegister<GpuRegister>();
+    __ Move(TMP, r2);
+    __ Move(r2, r1);
+    __ Move(r1, TMP);
+  } else if (is_fp_reg2 && is_fp_reg1) {
+    // Swap 2 FPRs
+    FpuRegister r1 = loc1.AsFpuRegister<FpuRegister>();
+    FpuRegister r2 = loc2.AsFpuRegister<FpuRegister>();
+    // TODO: Can MOV.S/MOV.D be used here to save one instruction?
+    // Need to distinguish float from double, right?
+    __ Dmfc1(TMP, r2);
+    __ Dmfc1(AT, r1);
+    __ Dmtc1(TMP, r1);
+    __ Dmtc1(AT, r2);
+  } else if (is_slot1 != is_slot2) {
+    // Swap GPR/FPR and stack slot
+    Location reg_loc = is_slot1 ? loc2 : loc1;
+    Location mem_loc = is_slot1 ? loc1 : loc2;
+    LoadOperandType load_type = mem_loc.IsStackSlot() ? kLoadWord : kLoadDoubleword;
+    StoreOperandType store_type = mem_loc.IsStackSlot() ? kStoreWord : kStoreDoubleword;
+    // TODO: use load_type = kLoadUnsignedWord when type == Primitive::kPrimNot.
+    __ LoadFromOffset(load_type, TMP, SP, mem_loc.GetStackIndex());
+    if (reg_loc.IsFpuRegister()) {
+      __ StoreFpuToOffset(store_type,
+                          reg_loc.AsFpuRegister<FpuRegister>(),
+                          SP,
+                          mem_loc.GetStackIndex());
+      // TODO: review this MTC1/DMTC1 move
+      if (mem_loc.IsStackSlot()) {
+        __ Mtc1(TMP, reg_loc.AsFpuRegister<FpuRegister>());
+      } else {
+        DCHECK(mem_loc.IsDoubleStackSlot());
+        __ Dmtc1(TMP, reg_loc.AsFpuRegister<FpuRegister>());
+      }
+    } else {
+      __ StoreToOffset(store_type, reg_loc.AsRegister<GpuRegister>(), SP, mem_loc.GetStackIndex());
+      __ Move(reg_loc.AsRegister<GpuRegister>(), TMP);
+    }
+  } else if (is_slot1 && is_slot2) {
+    move_resolver_.Exchange(loc1.GetStackIndex(),
+                            loc2.GetStackIndex(),
+                            loc1.IsDoubleStackSlot());
+  } else {
+    LOG(FATAL) << "Unimplemented swap between locations " << loc1 << " and " << loc2;
+  }
+}
+
+void CodeGeneratorMIPS64::Move(HInstruction* instruction,
+                               Location location,
+                               HInstruction* move_for) {
+  LocationSummary* locations = instruction->GetLocations();
+  Primitive::Type type = instruction->GetType();
+  DCHECK_NE(type, Primitive::kPrimVoid);
+
+  if (instruction->IsCurrentMethod()) {
+    MoveLocation(location, Location::DoubleStackSlot(kCurrentMethodStackOffset), type);
+  } else if (locations != nullptr && locations->Out().Equals(location)) {
+    return;
+  } else if (instruction->IsIntConstant()
+             || instruction->IsLongConstant()
+             || instruction->IsNullConstant()) {
+    if (location.IsRegister()) {
+      // Move to GPR from constant
+      GpuRegister dst = location.AsRegister<GpuRegister>();
+      if (instruction->IsNullConstant() || instruction->IsIntConstant()) {
+        __ LoadConst32(dst, GetInt32ValueOf(instruction->AsConstant()));
+      } else {
+        __ LoadConst64(dst, instruction->AsLongConstant()->GetValue());
+      }
+    } else {
+      DCHECK(location.IsStackSlot() || location.IsDoubleStackSlot());
+      // Move to stack from constant
+      if (location.IsStackSlot()) {
+        __ LoadConst32(TMP, GetInt32ValueOf(instruction->AsConstant()));
+        __ StoreToOffset(kStoreWord, TMP, SP, location.GetStackIndex());
+      } else {
+        __ LoadConst64(TMP, instruction->AsLongConstant()->GetValue());
+        __ StoreToOffset(kStoreDoubleword, TMP, SP, location.GetStackIndex());
+      }
+    }
+  } else if (instruction->IsTemporary()) {
+    Location temp_location = GetTemporaryLocation(instruction->AsTemporary());
+    MoveLocation(location, temp_location, type);
+  } else if (instruction->IsLoadLocal()) {
+    uint32_t stack_slot = GetStackSlot(instruction->AsLoadLocal()->GetLocal());
+    if (Primitive::Is64BitType(type)) {
+      MoveLocation(location, Location::DoubleStackSlot(stack_slot), type);
+    } else {
+      MoveLocation(location, Location::StackSlot(stack_slot), type);
+    }
+  } else {
+    DCHECK((instruction->GetNext() == move_for) || instruction->GetNext()->IsTemporary());
+    MoveLocation(location, locations->Out(), type);
+  }
+}
+
+Location CodeGeneratorMIPS64::GetStackLocation(HLoadLocal* load) const {
+  Primitive::Type type = load->GetType();
+
+  switch (type) {
+    case Primitive::kPrimNot:
+    case Primitive::kPrimInt:
+    case Primitive::kPrimFloat:
+      return Location::StackSlot(GetStackSlot(load->GetLocal()));
+
+    case Primitive::kPrimLong:
+    case Primitive::kPrimDouble:
+      return Location::DoubleStackSlot(GetStackSlot(load->GetLocal()));
+
+    case Primitive::kPrimBoolean:
+    case Primitive::kPrimByte:
+    case Primitive::kPrimChar:
+    case Primitive::kPrimShort:
+    case Primitive::kPrimVoid:
+      LOG(FATAL) << "Unexpected type " << type;
+  }
+
+  LOG(FATAL) << "Unreachable";
+  return Location::NoLocation();
+}
+
+void CodeGeneratorMIPS64::MarkGCCard(GpuRegister object, GpuRegister value) {
+  Label done;
+  GpuRegister card = AT;
+  GpuRegister temp = TMP;
+  __ Beqzc(value, &done);
+  __ LoadFromOffset(kLoadDoubleword,
+                    card,
+                    TR,
+                    Thread::CardTableOffset<kMips64WordSize>().Int32Value());
+  __ Dsrl(temp, object, gc::accounting::CardTable::kCardShift);
+  __ Daddu(temp, card, temp);
+  __ Sb(card, temp, 0);
+  __ Bind(&done);
+}
+
+void CodeGeneratorMIPS64::SetupBlockedRegisters(bool is_baseline ATTRIBUTE_UNUSED) const {
+  // ZERO, K0, K1, GP, SP, RA are always reserved and can't be allocated.
+  blocked_core_registers_[ZERO] = true;
+  blocked_core_registers_[K0] = true;
+  blocked_core_registers_[K1] = true;
+  blocked_core_registers_[GP] = true;
+  blocked_core_registers_[SP] = true;
+  blocked_core_registers_[RA] = true;
+
+  // AT and TMP(T8) are used as temporary/scratch registers
+  // (similar to how AT is used by MIPS assemblers).
+  blocked_core_registers_[AT] = true;
+  blocked_core_registers_[TMP] = true;
+  blocked_fpu_registers_[FTMP] = true;
+
+  // Reserve suspend and thread registers.
+  blocked_core_registers_[S0] = true;
+  blocked_core_registers_[TR] = true;
+
+  // Reserve T9 for function calls
+  blocked_core_registers_[T9] = true;
+
+  // TODO: review; anything else?
+
+  // TODO: make these two for's conditional on is_baseline once
+  // all the issues with register saving/restoring are sorted out.
+  for (size_t i = 0; i < arraysize(kCoreCalleeSaves); ++i) {
+    blocked_core_registers_[kCoreCalleeSaves[i]] = true;
+  }
+
+  for (size_t i = 0; i < arraysize(kFpuCalleeSaves); ++i) {
+    blocked_fpu_registers_[kFpuCalleeSaves[i]] = true;
+  }
+}
+
+Location CodeGeneratorMIPS64::AllocateFreeRegister(Primitive::Type type) const {
+  if (type == Primitive::kPrimVoid) {
+    LOG(FATAL) << "Unreachable type " << type;
+  }
+
+  if (Primitive::IsFloatingPointType(type)) {
+    size_t reg = FindFreeEntry(blocked_fpu_registers_, kNumberOfFpuRegisters);
+    return Location::FpuRegisterLocation(reg);
+  } else {
+    size_t reg = FindFreeEntry(blocked_core_registers_, kNumberOfGpuRegisters);
+    return Location::RegisterLocation(reg);
+  }
+}
+
+size_t CodeGeneratorMIPS64::SaveCoreRegister(size_t stack_index, uint32_t reg_id) {
+  __ StoreToOffset(kStoreDoubleword, GpuRegister(reg_id), SP, stack_index);
+  return kMips64WordSize;
+}
+
+size_t CodeGeneratorMIPS64::RestoreCoreRegister(size_t stack_index, uint32_t reg_id) {
+  __ LoadFromOffset(kLoadDoubleword, GpuRegister(reg_id), SP, stack_index);
+  return kMips64WordSize;
+}
+
+size_t CodeGeneratorMIPS64::SaveFloatingPointRegister(size_t stack_index, uint32_t reg_id) {
+  __ StoreFpuToOffset(kStoreDoubleword, FpuRegister(reg_id), SP, stack_index);
+  return kMips64WordSize;
+}
+
+size_t CodeGeneratorMIPS64::RestoreFloatingPointRegister(size_t stack_index, uint32_t reg_id) {
+  __ LoadFpuFromOffset(kLoadDoubleword, FpuRegister(reg_id), SP, stack_index);
+  return kMips64WordSize;
+}
+
+void CodeGeneratorMIPS64::DumpCoreRegister(std::ostream& stream, int reg) const {
+  stream << Mips64ManagedRegister::FromGpuRegister(GpuRegister(reg));
+}
+
+void CodeGeneratorMIPS64::DumpFloatingPointRegister(std::ostream& stream, int reg) const {
+  stream << Mips64ManagedRegister::FromFpuRegister(FpuRegister(reg));
+}
+
+void CodeGeneratorMIPS64::InvokeRuntime(int32_t entry_point_offset,
+                                        HInstruction* instruction,
+                                        uint32_t dex_pc,
+                                        SlowPathCode* slow_path) {
+  // TODO: anything related to T9/GP/GOT/PIC/.so's?
+  __ LoadFromOffset(kLoadDoubleword, T9, TR, entry_point_offset);
+  __ Jalr(T9);
+  RecordPcInfo(instruction, dex_pc, slow_path);
+  DCHECK(instruction->IsSuspendCheck()
+      || instruction->IsBoundsCheck()
+      || instruction->IsNullCheck()
+      || instruction->IsDivZeroCheck()
+      || !IsLeafMethod());
+}
+
+void InstructionCodeGeneratorMIPS64::GenerateClassInitializationCheck(SlowPathCodeMIPS64* slow_path,
+                                                                      GpuRegister class_reg) {
+  __ LoadFromOffset(kLoadWord, TMP, class_reg, mirror::Class::StatusOffset().Int32Value());
+  __ LoadConst32(AT, mirror::Class::kStatusInitialized);
+  __ Bltc(TMP, AT, slow_path->GetEntryLabel());
+  // TODO: barrier needed?
+  __ Bind(slow_path->GetExitLabel());
+}
+
+void InstructionCodeGeneratorMIPS64::GenerateMemoryBarrier(MemBarrierKind kind ATTRIBUTE_UNUSED) {
+  __ Sync(0);  // only stype 0 is supported
+}
+
+void InstructionCodeGeneratorMIPS64::GenerateSuspendCheck(HSuspendCheck* instruction,
+                                                          HBasicBlock* successor) {
+  SuspendCheckSlowPathMIPS64* slow_path =
+    new (GetGraph()->GetArena()) SuspendCheckSlowPathMIPS64(instruction, successor);
+  codegen_->AddSlowPath(slow_path);
+
+  __ LoadFromOffset(kLoadUnsignedHalfword,
+                    TMP,
+                    TR,
+                    Thread::ThreadFlagsOffset<kMips64WordSize>().Int32Value());
+  if (successor == nullptr) {
+    __ Bnezc(TMP, slow_path->GetEntryLabel());
+    __ Bind(slow_path->GetReturnLabel());
+  } else {
+    __ Beqzc(TMP, codegen_->GetLabelOf(successor));
+    __ B(slow_path->GetEntryLabel());
+    // slow_path will return to GetLabelOf(successor).
+  }
+}
+
+InstructionCodeGeneratorMIPS64::InstructionCodeGeneratorMIPS64(HGraph* graph,
+                                                               CodeGeneratorMIPS64* codegen)
+      : HGraphVisitor(graph),
+        assembler_(codegen->GetAssembler()),
+        codegen_(codegen) {}
+
+void LocationsBuilderMIPS64::HandleBinaryOp(HBinaryOperation* instruction) {
+  DCHECK_EQ(instruction->InputCount(), 2U);
+  LocationSummary* locations = new (GetGraph()->GetArena()) LocationSummary(instruction);
+  Primitive::Type type = instruction->GetResultType();
+  switch (type) {
+    case Primitive::kPrimInt:
+    case Primitive::kPrimLong: {
+      locations->SetInAt(0, Location::RequiresRegister());
+      HInstruction* right = instruction->InputAt(1);
+      bool can_use_imm = false;
+      if (right->IsConstant()) {
+        int64_t imm = CodeGenerator::GetInt64ValueOf(right->AsConstant());
+        if (instruction->IsAnd() || instruction->IsOr() || instruction->IsXor()) {
+          can_use_imm = IsUint<16>(imm);
+        } else if (instruction->IsAdd()) {
+          can_use_imm = IsInt<16>(imm);
+        } else {
+          DCHECK(instruction->IsSub());
+          can_use_imm = IsInt<16>(-imm);
+        }
+      }
+      if (can_use_imm)
+        locations->SetInAt(1, Location::ConstantLocation(right->AsConstant()));
+      else
+        locations->SetInAt(1, Location::RequiresRegister());
+      locations->SetOut(Location::RequiresRegister(), Location::kNoOutputOverlap);
+      }
+      break;
+
+    case Primitive::kPrimFloat:
+    case Primitive::kPrimDouble:
+      locations->SetInAt(0, Location::RequiresFpuRegister());
+      locations->SetInAt(1, Location::RequiresFpuRegister());
+      locations->SetOut(Location::RequiresFpuRegister(), Location::kNoOutputOverlap);
+      break;
+
+    default:
+      LOG(FATAL) << "Unexpected " << instruction->DebugName() << " type " << type;
+  }
+}
+
+void InstructionCodeGeneratorMIPS64::HandleBinaryOp(HBinaryOperation* instruction) {
+  Primitive::Type type = instruction->GetType();
+  LocationSummary* locations = instruction->GetLocations();
+
+  switch (type) {
+    case Primitive::kPrimInt:
+    case Primitive::kPrimLong: {
+      GpuRegister dst = locations->Out().AsRegister<GpuRegister>();
+      GpuRegister lhs = locations->InAt(0).AsRegister<GpuRegister>();
+      Location rhs_location = locations->InAt(1);
+
+      GpuRegister rhs_reg = ZERO;
+      int64_t rhs_imm = 0;
+      bool use_imm = rhs_location.IsConstant();
+      if (use_imm) {
+        rhs_imm = CodeGenerator::GetInt64ValueOf(rhs_location.GetConstant());
+      } else {
+        rhs_reg = rhs_location.AsRegister<GpuRegister>();
+      }
+
+      if (instruction->IsAnd()) {
+        if (use_imm)
+          __ Andi(dst, lhs, rhs_imm);
+        else
+          __ And(dst, lhs, rhs_reg);
+      } else if (instruction->IsOr()) {
+        if (use_imm)
+          __ Ori(dst, lhs, rhs_imm);
+        else
+          __ Or(dst, lhs, rhs_reg);
+      } else if (instruction->IsXor()) {
+        if (use_imm)
+          __ Xori(dst, lhs, rhs_imm);
+        else
+          __ Xor(dst, lhs, rhs_reg);
+      } else if (instruction->IsAdd()) {
+        if (type == Primitive::kPrimInt) {
+          if (use_imm)
+            __ Addiu(dst, lhs, rhs_imm);
+          else
+            __ Addu(dst, lhs, rhs_reg);
+        } else {
+          if (use_imm)
+            __ Daddiu(dst, lhs, rhs_imm);
+          else
+            __ Daddu(dst, lhs, rhs_reg);
+        }
+      } else {
+        DCHECK(instruction->IsSub());
+        if (type == Primitive::kPrimInt) {
+          if (use_imm)
+            __ Addiu(dst, lhs, -rhs_imm);
+          else
+            __ Subu(dst, lhs, rhs_reg);
+        } else {
+          if (use_imm)
+            __ Daddiu(dst, lhs, -rhs_imm);
+          else
+            __ Dsubu(dst, lhs, rhs_reg);
+        }
+      }
+      break;
+    }
+    case Primitive::kPrimFloat:
+    case Primitive::kPrimDouble: {
+      FpuRegister dst = locations->Out().AsFpuRegister<FpuRegister>();
+      FpuRegister lhs = locations->InAt(0).AsFpuRegister<FpuRegister>();
+      FpuRegister rhs = locations->InAt(1).AsFpuRegister<FpuRegister>();
+      if (instruction->IsAdd()) {
+        if (type == Primitive::kPrimFloat)
+          __ AddS(dst, lhs, rhs);
+        else
+          __ AddD(dst, lhs, rhs);
+      } else if (instruction->IsSub()) {
+        if (type == Primitive::kPrimFloat)
+          __ SubS(dst, lhs, rhs);
+        else
+          __ SubD(dst, lhs, rhs);
+      } else {
+        LOG(FATAL) << "Unexpected floating-point binary operation";
+      }
+      break;
+    }
+    default:
+      LOG(FATAL) << "Unexpected binary operation type " << type;
+  }
+}
+
+void LocationsBuilderMIPS64::HandleShift(HBinaryOperation* instr) {
+  DCHECK(instr->IsShl() || instr->IsShr() || instr->IsUShr());
+
+  LocationSummary* locations = new (GetGraph()->GetArena()) LocationSummary(instr);
+  Primitive::Type type = instr->GetResultType();
+  switch (type) {
+    case Primitive::kPrimInt:
+    case Primitive::kPrimLong: {
+      locations->SetInAt(0, Location::RequiresRegister());
+      locations->SetInAt(1, Location::RegisterOrConstant(instr->InputAt(1)));
+      locations->SetOut(Location::RequiresRegister());
+      break;
+    }
+    default:
+      LOG(FATAL) << "Unexpected shift type " << type;
+  }
+}
+
+void InstructionCodeGeneratorMIPS64::HandleShift(HBinaryOperation* instr) {
+  DCHECK(instr->IsShl() || instr->IsShr() || instr->IsUShr());
+  LocationSummary* locations = instr->GetLocations();
+  Primitive::Type type = instr->GetType();
+
+  switch (type) {
+    case Primitive::kPrimInt:
+    case Primitive::kPrimLong: {
+      GpuRegister dst = locations->Out().AsRegister<GpuRegister>();
+      GpuRegister lhs = locations->InAt(0).AsRegister<GpuRegister>();
+      Location rhs_location = locations->InAt(1);
+
+      GpuRegister rhs_reg = ZERO;
+      int64_t rhs_imm = 0;
+      bool use_imm = rhs_location.IsConstant();
+      if (use_imm) {
+        rhs_imm = CodeGenerator::GetInt64ValueOf(rhs_location.GetConstant());
+      } else {
+        rhs_reg = rhs_location.AsRegister<GpuRegister>();
+      }
+
+      if (use_imm) {
+        uint32_t shift_value = (type == Primitive::kPrimInt)
+          ? static_cast<uint32_t>(rhs_imm & kMaxIntShiftValue)
+          : static_cast<uint32_t>(rhs_imm & kMaxLongShiftValue);
+
+        if (type == Primitive::kPrimInt) {
+          if (instr->IsShl()) {
+            __ Sll(dst, lhs, shift_value);
+          } else if (instr->IsShr()) {
+            __ Sra(dst, lhs, shift_value);
+          } else {
+            __ Srl(dst, lhs, shift_value);
+          }
+        } else {
+          if (shift_value < 32) {
+            if (instr->IsShl()) {
+              __ Dsll(dst, lhs, shift_value);
+            } else if (instr->IsShr()) {
+              __ Dsra(dst, lhs, shift_value);
+            } else {
+              __ Dsrl(dst, lhs, shift_value);
+            }
+          } else {
+            shift_value -= 32;
+            if (instr->IsShl()) {
+              __ Dsll32(dst, lhs, shift_value);
+            } else if (instr->IsShr()) {
+              __ Dsra32(dst, lhs, shift_value);
+            } else {
+              __ Dsrl32(dst, lhs, shift_value);
+            }
+          }
+        }
+      } else {
+        if (type == Primitive::kPrimInt) {
+          if (instr->IsShl()) {
+            __ Sllv(dst, lhs, rhs_reg);
+          } else if (instr->IsShr()) {
+            __ Srav(dst, lhs, rhs_reg);
+          } else {
+            __ Srlv(dst, lhs, rhs_reg);
+          }
+        } else {
+          if (instr->IsShl()) {
+            __ Dsllv(dst, lhs, rhs_reg);
+          } else if (instr->IsShr()) {
+            __ Dsrav(dst, lhs, rhs_reg);
+          } else {
+            __ Dsrlv(dst, lhs, rhs_reg);
+          }
+        }
+      }
+      break;
+    }
+    default:
+      LOG(FATAL) << "Unexpected shift operation type " << type;
+  }
+}
+
+void LocationsBuilderMIPS64::VisitAdd(HAdd* instruction) {
+  HandleBinaryOp(instruction);
+}
+
+void InstructionCodeGeneratorMIPS64::VisitAdd(HAdd* instruction) {
+  HandleBinaryOp(instruction);
+}
+
+void LocationsBuilderMIPS64::VisitAnd(HAnd* instruction) {
+  HandleBinaryOp(instruction);
+}
+
+void InstructionCodeGeneratorMIPS64::VisitAnd(HAnd* instruction) {
+  HandleBinaryOp(instruction);
+}
+
+void LocationsBuilderMIPS64::VisitArrayGet(HArrayGet* instruction) {
+  LocationSummary* locations =
+      new (GetGraph()->GetArena()) LocationSummary(instruction, LocationSummary::kNoCall);
+  locations->SetInAt(0, Location::RequiresRegister());
+  locations->SetInAt(1, Location::RegisterOrConstant(instruction->InputAt(1)));
+  if (Primitive::IsFloatingPointType(instruction->GetType())) {
+    locations->SetOut(Location::RequiresFpuRegister(), Location::kNoOutputOverlap);
+  } else {
+    locations->SetOut(Location::RequiresRegister(), Location::kNoOutputOverlap);
+  }
+}
+
+void InstructionCodeGeneratorMIPS64::VisitArrayGet(HArrayGet* instruction) {
+  LocationSummary* locations = instruction->GetLocations();
+  GpuRegister obj = locations->InAt(0).AsRegister<GpuRegister>();
+  Location index = locations->InAt(1);
+  Primitive::Type type = instruction->GetType();
+
+  switch (type) {
+    case Primitive::kPrimBoolean: {
+      uint32_t data_offset = mirror::Array::DataOffset(sizeof(uint8_t)).Uint32Value();
+      GpuRegister out = locations->Out().AsRegister<GpuRegister>();
+      if (index.IsConstant()) {
+        size_t offset =
+            (index.GetConstant()->AsIntConstant()->GetValue() << TIMES_1) + data_offset;
+        __ LoadFromOffset(kLoadUnsignedByte, out, obj, offset);
+      } else {
+        __ Daddu(TMP, obj, index.AsRegister<GpuRegister>());
+        __ LoadFromOffset(kLoadUnsignedByte, out, TMP, data_offset);
+      }
+      break;
+    }
+
+    case Primitive::kPrimByte: {
+      uint32_t data_offset = mirror::Array::DataOffset(sizeof(int8_t)).Uint32Value();
+      GpuRegister out = locations->Out().AsRegister<GpuRegister>();
+      if (index.IsConstant()) {
+        size_t offset =
+            (index.GetConstant()->AsIntConstant()->GetValue() << TIMES_1) + data_offset;
+        __ LoadFromOffset(kLoadSignedByte, out, obj, offset);
+      } else {
+        __ Daddu(TMP, obj, index.AsRegister<GpuRegister>());
+        __ LoadFromOffset(kLoadSignedByte, out, TMP, data_offset);
+      }
+      break;
+    }
+
+    case Primitive::kPrimShort: {
+      uint32_t data_offset = mirror::Array::DataOffset(sizeof(int16_t)).Uint32Value();
+      GpuRegister out = locations->Out().AsRegister<GpuRegister>();
+      if (index.IsConstant()) {
+        size_t offset =
+            (index.GetConstant()->AsIntConstant()->GetValue() << TIMES_2) + data_offset;
+        __ LoadFromOffset(kLoadSignedHalfword, out, obj, offset);
+      } else {
+        __ Dsll(TMP, index.AsRegister<GpuRegister>(), TIMES_2);
+        __ Daddu(TMP, obj, TMP);
+        __ LoadFromOffset(kLoadSignedHalfword, out, TMP, data_offset);
+      }
+      break;
+    }
+
+    case Primitive::kPrimChar: {
+      uint32_t data_offset = mirror::Array::DataOffset(sizeof(uint16_t)).Uint32Value();
+      GpuRegister out = locations->Out().AsRegister<GpuRegister>();
+      if (index.IsConstant()) {
+        size_t offset =
+            (index.GetConstant()->AsIntConstant()->GetValue() << TIMES_2) + data_offset;
+        __ LoadFromOffset(kLoadUnsignedHalfword, out, obj, offset);
+      } else {
+        __ Dsll(TMP, index.AsRegister<GpuRegister>(), TIMES_2);
+        __ Daddu(TMP, obj, TMP);
+        __ LoadFromOffset(kLoadUnsignedHalfword, out, TMP, data_offset);
+      }
+      break;
+    }
+
+    case Primitive::kPrimInt:
+    case Primitive::kPrimNot: {
+      DCHECK_EQ(sizeof(mirror::HeapReference<mirror::Object>), sizeof(int32_t));
+      uint32_t data_offset = mirror::Array::DataOffset(sizeof(int32_t)).Uint32Value();
+      GpuRegister out = locations->Out().AsRegister<GpuRegister>();
+      LoadOperandType load_type = (type == Primitive::kPrimNot) ? kLoadUnsignedWord : kLoadWord;
+      if (index.IsConstant()) {
+        size_t offset =
+            (index.GetConstant()->AsIntConstant()->GetValue() << TIMES_4) + data_offset;
+        __ LoadFromOffset(load_type, out, obj, offset);
+      } else {
+        __ Dsll(TMP, index.AsRegister<GpuRegister>(), TIMES_4);
+        __ Daddu(TMP, obj, TMP);
+        __ LoadFromOffset(load_type, out, TMP, data_offset);
+      }
+      break;
+    }
+
+    case Primitive::kPrimLong: {
+      uint32_t data_offset = mirror::Array::DataOffset(sizeof(int64_t)).Uint32Value();
+      GpuRegister out = locations->Out().AsRegister<GpuRegister>();
+      if (index.IsConstant()) {
+        size_t offset =
+            (index.GetConstant()->AsIntConstant()->GetValue() << TIMES_8) + data_offset;
+        __ LoadFromOffset(kLoadDoubleword, out, obj, offset);
+      } else {
+        __ Dsll(TMP, index.AsRegister<GpuRegister>(), TIMES_8);
+        __ Daddu(TMP, obj, TMP);
+        __ LoadFromOffset(kLoadDoubleword, out, TMP, data_offset);
+      }
+      break;
+    }
+
+    case Primitive::kPrimFloat: {
+      uint32_t data_offset = mirror::Array::DataOffset(sizeof(float)).Uint32Value();
+      FpuRegister out = locations->Out().AsFpuRegister<FpuRegister>();
+      if (index.IsConstant()) {
+        size_t offset =
+            (index.GetConstant()->AsIntConstant()->GetValue() << TIMES_4) + data_offset;
+        __ LoadFpuFromOffset(kLoadWord, out, obj, offset);
+      } else {
+        __ Dsll(TMP, index.AsRegister<GpuRegister>(), TIMES_4);
+        __ Daddu(TMP, obj, TMP);
+        __ LoadFpuFromOffset(kLoadWord, out, TMP, data_offset);
+      }
+      break;
+    }
+
+    case Primitive::kPrimDouble: {
+      uint32_t data_offset = mirror::Array::DataOffset(sizeof(double)).Uint32Value();
+      FpuRegister out = locations->Out().AsFpuRegister<FpuRegister>();
+      if (index.IsConstant()) {
+        size_t offset =
+            (index.GetConstant()->AsIntConstant()->GetValue() << TIMES_8) + data_offset;
+        __ LoadFpuFromOffset(kLoadDoubleword, out, obj, offset);
+      } else {
+        __ Dsll(TMP, index.AsRegister<GpuRegister>(), TIMES_8);
+        __ Daddu(TMP, obj, TMP);
+        __ LoadFpuFromOffset(kLoadDoubleword, out, TMP, data_offset);
+      }
+      break;
+    }
+
+    case Primitive::kPrimVoid:
+      LOG(FATAL) << "Unreachable type " << instruction->GetType();
+      UNREACHABLE();
+  }
+  codegen_->MaybeRecordImplicitNullCheck(instruction);
+}
+
+void LocationsBuilderMIPS64::VisitArrayLength(HArrayLength* instruction) {
+  LocationSummary* locations = new (GetGraph()->GetArena()) LocationSummary(instruction);
+  locations->SetInAt(0, Location::RequiresRegister());
+  locations->SetOut(Location::RequiresRegister(), Location::kNoOutputOverlap);
+}
+
+void InstructionCodeGeneratorMIPS64::VisitArrayLength(HArrayLength* instruction) {
+  LocationSummary* locations = instruction->GetLocations();
+  uint32_t offset = mirror::Array::LengthOffset().Uint32Value();
+  GpuRegister obj = locations->InAt(0).AsRegister<GpuRegister>();
+  GpuRegister out = locations->Out().AsRegister<GpuRegister>();
+  __ LoadFromOffset(kLoadWord, out, obj, offset);
+  codegen_->MaybeRecordImplicitNullCheck(instruction);
+}
+
+void LocationsBuilderMIPS64::VisitArraySet(HArraySet* instruction) {
+  Primitive::Type value_type = instruction->GetComponentType();
+  bool is_object = value_type == Primitive::kPrimNot;
+  LocationSummary* locations = new (GetGraph()->GetArena()) LocationSummary(
+      instruction,
+      is_object ? LocationSummary::kCall : LocationSummary::kNoCall);
+  if (is_object) {
+    InvokeRuntimeCallingConvention calling_convention;
+    locations->SetInAt(0, Location::RegisterLocation(calling_convention.GetRegisterAt(0)));
+    locations->SetInAt(1, Location::RegisterLocation(calling_convention.GetRegisterAt(1)));
+    locations->SetInAt(2, Location::RegisterLocation(calling_convention.GetRegisterAt(2)));
+  } else {
+    locations->SetInAt(0, Location::RequiresRegister());
+    locations->SetInAt(1, Location::RegisterOrConstant(instruction->InputAt(1)));
+    if (Primitive::IsFloatingPointType(instruction->InputAt(2)->GetType())) {
+      locations->SetInAt(2, Location::RequiresFpuRegister());
+    } else {
+      locations->SetInAt(2, Location::RequiresRegister());
+    }
+  }
+}
+
+void InstructionCodeGeneratorMIPS64::VisitArraySet(HArraySet* instruction) {
+  LocationSummary* locations = instruction->GetLocations();
+  GpuRegister obj = locations->InAt(0).AsRegister<GpuRegister>();
+  Location index = locations->InAt(1);
+  Primitive::Type value_type = instruction->GetComponentType();
+  bool needs_runtime_call = locations->WillCall();
+  bool needs_write_barrier =
+      CodeGenerator::StoreNeedsWriteBarrier(value_type, instruction->GetValue());
+
+  switch (value_type) {
+    case Primitive::kPrimBoolean:
+    case Primitive::kPrimByte: {
+      uint32_t data_offset = mirror::Array::DataOffset(sizeof(uint8_t)).Uint32Value();
+      GpuRegister value = locations->InAt(2).AsRegister<GpuRegister>();
+      if (index.IsConstant()) {
+        size_t offset =
+            (index.GetConstant()->AsIntConstant()->GetValue() << TIMES_1) + data_offset;
+        __ StoreToOffset(kStoreByte, value, obj, offset);
+      } else {
+        __ Daddu(TMP, obj, index.AsRegister<GpuRegister>());
+        __ StoreToOffset(kStoreByte, value, TMP, data_offset);
+      }
+      break;
+    }
+
+    case Primitive::kPrimShort:
+    case Primitive::kPrimChar: {
+      uint32_t data_offset = mirror::Array::DataOffset(sizeof(uint16_t)).Uint32Value();
+      GpuRegister value = locations->InAt(2).AsRegister<GpuRegister>();
+      if (index.IsConstant()) {
+        size_t offset =
+            (index.GetConstant()->AsIntConstant()->GetValue() << TIMES_2) + data_offset;
+        __ StoreToOffset(kStoreHalfword, value, obj, offset);
+      } else {
+        __ Dsll(TMP, index.AsRegister<GpuRegister>(), TIMES_2);
+        __ Daddu(TMP, obj, TMP);
+        __ StoreToOffset(kStoreHalfword, value, TMP, data_offset);
+      }
+      break;
+    }
+
+    case Primitive::kPrimInt:
+    case Primitive::kPrimNot: {
+      if (!needs_runtime_call) {
+        uint32_t data_offset = mirror::Array::DataOffset(sizeof(int32_t)).Uint32Value();
+        GpuRegister value = locations->InAt(2).AsRegister<GpuRegister>();
+        if (index.IsConstant()) {
+          size_t offset =
+              (index.GetConstant()->AsIntConstant()->GetValue() << TIMES_4) + data_offset;
+          __ StoreToOffset(kStoreWord, value, obj, offset);
+        } else {
+          DCHECK(index.IsRegister()) << index;
+          __ Dsll(TMP, index.AsRegister<GpuRegister>(), TIMES_4);
+          __ Daddu(TMP, obj, TMP);
+          __ StoreToOffset(kStoreWord, value, TMP, data_offset);
+        }
+        codegen_->MaybeRecordImplicitNullCheck(instruction);
+        if (needs_write_barrier) {
+          DCHECK_EQ(value_type, Primitive::kPrimNot);
+          codegen_->MarkGCCard(obj, value);
+        }
+      } else {
+        DCHECK_EQ(value_type, Primitive::kPrimNot);
+        codegen_->InvokeRuntime(QUICK_ENTRY_POINT(pAputObject),
+                                instruction,
+                                instruction->GetDexPc(),
+                                nullptr);
+      }
+      break;
+    }
+
+    case Primitive::kPrimLong: {
+      uint32_t data_offset = mirror::Array::DataOffset(sizeof(int64_t)).Uint32Value();
+      GpuRegister value = locations->InAt(2).AsRegister<GpuRegister>();
+      if (index.IsConstant()) {
+        size_t offset =
+            (index.GetConstant()->AsIntConstant()->GetValue() << TIMES_8) + data_offset;
+        __ StoreToOffset(kStoreDoubleword, value, obj, offset);
+      } else {
+        __ Dsll(TMP, index.AsRegister<GpuRegister>(), TIMES_8);
+        __ Daddu(TMP, obj, TMP);
+        __ StoreToOffset(kStoreDoubleword, value, TMP, data_offset);
+      }
+      break;
+    }
+
+    case Primitive::kPrimFloat: {
+      uint32_t data_offset = mirror::Array::DataOffset(sizeof(float)).Uint32Value();
+      FpuRegister value = locations->InAt(2).AsFpuRegister<FpuRegister>();
+      DCHECK(locations->InAt(2).IsFpuRegister());
+      if (index.IsConstant()) {
+        size_t offset =
+            (index.GetConstant()->AsIntConstant()->GetValue() << TIMES_4) + data_offset;
+        __ StoreFpuToOffset(kStoreWord, value, obj, offset);
+      } else {
+        __ Dsll(TMP, index.AsRegister<GpuRegister>(), TIMES_4);
+        __ Daddu(TMP, obj, TMP);
+        __ StoreFpuToOffset(kStoreWord, value, TMP, data_offset);
+      }
+      break;
+    }
+
+    case Primitive::kPrimDouble: {
+      uint32_t data_offset = mirror::Array::DataOffset(sizeof(double)).Uint32Value();
+      FpuRegister value = locations->InAt(2).AsFpuRegister<FpuRegister>();
+      DCHECK(locations->InAt(2).IsFpuRegister());
+      if (index.IsConstant()) {
+        size_t offset =
+            (index.GetConstant()->AsIntConstant()->GetValue() << TIMES_8) + data_offset;
+        __ StoreFpuToOffset(kStoreDoubleword, value, obj, offset);
+      } else {
+        __ Dsll(TMP, index.AsRegister<GpuRegister>(), TIMES_8);
+        __ Daddu(TMP, obj, TMP);
+        __ StoreFpuToOffset(kStoreDoubleword, value, TMP, data_offset);
+      }
+      break;
+    }
+
+    case Primitive::kPrimVoid:
+      LOG(FATAL) << "Unreachable type " << instruction->GetType();
+      UNREACHABLE();
+  }
+
+  // Ints and objects are handled in the switch.
+  if (value_type != Primitive::kPrimInt && value_type != Primitive::kPrimNot) {
+    codegen_->MaybeRecordImplicitNullCheck(instruction);
+  }
+}
+
+void LocationsBuilderMIPS64::VisitBoundsCheck(HBoundsCheck* instruction) {
+  LocationSummary* locations =
+      new (GetGraph()->GetArena()) LocationSummary(instruction, LocationSummary::kNoCall);
+  locations->SetInAt(0, Location::RequiresRegister());
+  locations->SetInAt(1, Location::RequiresRegister());
+  if (instruction->HasUses()) {
+    locations->SetOut(Location::SameAsFirstInput());
+  }
+}
+
+void InstructionCodeGeneratorMIPS64::VisitBoundsCheck(HBoundsCheck* instruction) {
+  LocationSummary* locations = instruction->GetLocations();
+  BoundsCheckSlowPathMIPS64* slow_path = new (GetGraph()->GetArena()) BoundsCheckSlowPathMIPS64(
+      instruction,
+      locations->InAt(0),
+      locations->InAt(1));
+  codegen_->AddSlowPath(slow_path);
+
+  GpuRegister index = locations->InAt(0).AsRegister<GpuRegister>();
+  GpuRegister length = locations->InAt(1).AsRegister<GpuRegister>();
+
+  // length is limited by the maximum positive signed 32-bit integer.
+  // Unsigned comparison of length and index checks for index < 0
+  // and for length <= index simultaneously.
+  // Mips R6 requires lhs != rhs for compact branches.
+  if (index == length) {
+    __ B(slow_path->GetEntryLabel());
+  } else {
+    __ Bgeuc(index, length, slow_path->GetEntryLabel());
+  }
+}
+
+void LocationsBuilderMIPS64::VisitCheckCast(HCheckCast* instruction) {
+  LocationSummary* locations = new (GetGraph()->GetArena()) LocationSummary(
+      instruction,
+      LocationSummary::kCallOnSlowPath);
+  locations->SetInAt(0, Location::RequiresRegister());
+  locations->SetInAt(1, Location::RequiresRegister());
+  locations->AddTemp(Location::RequiresRegister());
+}
+
+void InstructionCodeGeneratorMIPS64::VisitCheckCast(HCheckCast* instruction) {
+  LocationSummary* locations = instruction->GetLocations();
+  GpuRegister obj = locations->InAt(0).AsRegister<GpuRegister>();
+  GpuRegister cls = locations->InAt(1).AsRegister<GpuRegister>();
+  GpuRegister obj_cls = locations->GetTemp(0).AsRegister<GpuRegister>();
+
+  SlowPathCodeMIPS64* slow_path = new (GetGraph()->GetArena()) TypeCheckSlowPathMIPS64(
+      instruction,
+      locations->InAt(1),
+      Location::RegisterLocation(obj_cls),
+      instruction->GetDexPc());
+  codegen_->AddSlowPath(slow_path);
+
+  // TODO: avoid this check if we know obj is not null.
+  __ Beqzc(obj, slow_path->GetExitLabel());
+  // Compare the class of `obj` with `cls`.
+  __ LoadFromOffset(kLoadUnsignedWord, obj_cls, obj, mirror::Object::ClassOffset().Int32Value());
+  __ Bnec(obj_cls, cls, slow_path->GetEntryLabel());
+  __ Bind(slow_path->GetExitLabel());
+}
+
+void LocationsBuilderMIPS64::VisitClinitCheck(HClinitCheck* check) {
+  LocationSummary* locations =
+      new (GetGraph()->GetArena()) LocationSummary(check, LocationSummary::kCallOnSlowPath);
+  locations->SetInAt(0, Location::RequiresRegister());
+  if (check->HasUses()) {
+    locations->SetOut(Location::SameAsFirstInput());
+  }
+}
+
+void InstructionCodeGeneratorMIPS64::VisitClinitCheck(HClinitCheck* check) {
+  // We assume the class is not null.
+  SlowPathCodeMIPS64* slow_path = new (GetGraph()->GetArena()) LoadClassSlowPathMIPS64(
+      check->GetLoadClass(),
+      check,
+      check->GetDexPc(),
+      true);
+  codegen_->AddSlowPath(slow_path);
+  GenerateClassInitializationCheck(slow_path,
+                                   check->GetLocations()->InAt(0).AsRegister<GpuRegister>());
+}
+
+void LocationsBuilderMIPS64::VisitCompare(HCompare* compare) {
+  Primitive::Type in_type = compare->InputAt(0)->GetType();
+
+  LocationSummary::CallKind call_kind = Primitive::IsFloatingPointType(in_type)
+      ? LocationSummary::kCall
+      : LocationSummary::kNoCall;
+
+  LocationSummary* locations = new (GetGraph()->GetArena()) LocationSummary(compare, call_kind);
+
+  switch (in_type) {
+    case Primitive::kPrimLong:
+      locations->SetInAt(0, Location::RequiresRegister());
+      locations->SetInAt(1, Location::RequiresRegister());
+      locations->SetOut(Location::RequiresRegister(), Location::kNoOutputOverlap);
+      break;
+
+    case Primitive::kPrimFloat:
+    case Primitive::kPrimDouble: {
+      InvokeRuntimeCallingConvention calling_convention;
+      locations->SetInAt(0, Location::FpuRegisterLocation(calling_convention.GetFpuRegisterAt(0)));
+      locations->SetInAt(1, Location::FpuRegisterLocation(calling_convention.GetFpuRegisterAt(1)));
+      locations->SetOut(calling_convention.GetReturnLocation(Primitive::kPrimInt));
+      break;
+    }
+
+    default:
+      LOG(FATAL) << "Unexpected type for compare operation " << in_type;
+  }
+}
+
+void InstructionCodeGeneratorMIPS64::VisitCompare(HCompare* instruction) {
+  LocationSummary* locations = instruction->GetLocations();
+  Primitive::Type in_type = instruction->InputAt(0)->GetType();
+
+  //  0 if: left == right
+  //  1 if: left  > right
+  // -1 if: left  < right
+  switch (in_type) {
+    case Primitive::kPrimLong: {
+      GpuRegister dst = locations->Out().AsRegister<GpuRegister>();
+      GpuRegister lhs = locations->InAt(0).AsRegister<GpuRegister>();
+      GpuRegister rhs = locations->InAt(1).AsRegister<GpuRegister>();
+      // TODO: more efficient (direct) comparison with a constant
+      __ Slt(TMP, lhs, rhs);
+      __ Slt(dst, rhs, lhs);
+      __ Subu(dst, dst, TMP);
+      break;
+    }
+
+    case Primitive::kPrimFloat:
+    case Primitive::kPrimDouble: {
+      int32_t entry_point_offset;
+      if (in_type == Primitive::kPrimFloat) {
+        entry_point_offset = instruction->IsGtBias() ? QUICK_ENTRY_POINT(pCmpgFloat)
+                                                     : QUICK_ENTRY_POINT(pCmplFloat);
+      } else {
+        entry_point_offset = instruction->IsGtBias() ? QUICK_ENTRY_POINT(pCmpgDouble)
+                                                     : QUICK_ENTRY_POINT(pCmplDouble);
+      }
+      codegen_->InvokeRuntime(entry_point_offset, instruction, instruction->GetDexPc(), nullptr);
+      break;
+    }
+
+    default:
+      LOG(FATAL) << "Unimplemented compare type " << in_type;
+  }
+}
+
+void LocationsBuilderMIPS64::VisitCondition(HCondition* instruction) {
+  LocationSummary* locations = new (GetGraph()->GetArena()) LocationSummary(instruction);
+  locations->SetInAt(0, Location::RequiresRegister());
+  locations->SetInAt(1, Location::RegisterOrConstant(instruction->InputAt(1)));
+  if (instruction->NeedsMaterialization()) {
+    locations->SetOut(Location::RequiresRegister(), Location::kNoOutputOverlap);
+  }
+}
+
+void InstructionCodeGeneratorMIPS64::VisitCondition(HCondition* instruction) {
+  if (!instruction->NeedsMaterialization()) {
+    return;
+  }
+
+  LocationSummary* locations = instruction->GetLocations();
+
+  GpuRegister dst = locations->Out().AsRegister<GpuRegister>();
+  GpuRegister lhs = locations->InAt(0).AsRegister<GpuRegister>();
+  Location rhs_location = locations->InAt(1);
+
+  GpuRegister rhs_reg = ZERO;
+  int64_t rhs_imm = 0;
+  bool use_imm = rhs_location.IsConstant();
+  if (use_imm) {
+    rhs_imm = CodeGenerator::GetInt32ValueOf(rhs_location.GetConstant());
+  } else {
+    rhs_reg = rhs_location.AsRegister<GpuRegister>();
+  }
+
+  IfCondition if_cond = instruction->GetCondition();
+
+  switch (if_cond) {
+    case kCondEQ:
+    case kCondNE:
+      if (use_imm && IsUint<16>(rhs_imm)) {
+        __ Xori(dst, lhs, rhs_imm);
+      } else {
+        if (use_imm) {
+          rhs_reg = TMP;
+          __ LoadConst32(rhs_reg, rhs_imm);
+        }
+        __ Xor(dst, lhs, rhs_reg);
+      }
+      if (if_cond == kCondEQ) {
+        __ Sltiu(dst, dst, 1);
+      } else {
+        __ Sltu(dst, ZERO, dst);
+      }
+      break;
+
+    case kCondLT:
+    case kCondGE:
+      if (use_imm && IsInt<16>(rhs_imm)) {
+        __ Slti(dst, lhs, rhs_imm);
+      } else {
+        if (use_imm) {
+          rhs_reg = TMP;
+          __ LoadConst32(rhs_reg, rhs_imm);
+        }
+        __ Slt(dst, lhs, rhs_reg);
+      }
+      if (if_cond == kCondGE) {
+        // Simulate lhs >= rhs via !(lhs < rhs) since there's
+        // only the slt instruction but no sge.
+        __ Xori(dst, dst, 1);
+      }
+      break;
+
+    case kCondLE:
+    case kCondGT:
+      if (use_imm && IsInt<16>(rhs_imm + 1)) {
+        // Simulate lhs <= rhs via lhs < rhs + 1.
+        __ Slti(dst, lhs, rhs_imm + 1);
+        if (if_cond == kCondGT) {
+          // Simulate lhs > rhs via !(lhs <= rhs) since there's
+          // only the slti instruction but no sgti.
+          __ Xori(dst, dst, 1);
+        }
+      } else {
+        if (use_imm) {
+          rhs_reg = TMP;
+          __ LoadConst32(rhs_reg, rhs_imm);
+        }
+        __ Slt(dst, rhs_reg, lhs);
+        if (if_cond == kCondLE) {
+          // Simulate lhs <= rhs via !(rhs < lhs) since there's
+          // only the slt instruction but no sle.
+          __ Xori(dst, dst, 1);
+        }
+      }
+      break;
+  }
+}
+
+void LocationsBuilderMIPS64::VisitDiv(HDiv* div) {
+  LocationSummary* locations =
+      new (GetGraph()->GetArena()) LocationSummary(div, LocationSummary::kNoCall);
+  switch (div->GetResultType()) {
+    case Primitive::kPrimInt:
+    case Primitive::kPrimLong:
+      locations->SetInAt(0, Location::RequiresRegister());
+      locations->SetInAt(1, Location::RequiresRegister());
+      locations->SetOut(Location::RequiresRegister(), Location::kNoOutputOverlap);
+      break;
+
+    case Primitive::kPrimFloat:
+    case Primitive::kPrimDouble:
+      locations->SetInAt(0, Location::RequiresFpuRegister());
+      locations->SetInAt(1, Location::RequiresFpuRegister());
+      locations->SetOut(Location::RequiresFpuRegister(), Location::kNoOutputOverlap);
+      break;
+
+    default:
+      LOG(FATAL) << "Unexpected div type " << div->GetResultType();
+  }
+}
+
+void InstructionCodeGeneratorMIPS64::VisitDiv(HDiv* instruction) {
+  Primitive::Type type = instruction->GetType();
+  LocationSummary* locations = instruction->GetLocations();
+
+  switch (type) {
+    case Primitive::kPrimInt:
+    case Primitive::kPrimLong: {
+      GpuRegister dst = locations->Out().AsRegister<GpuRegister>();
+      GpuRegister lhs = locations->InAt(0).AsRegister<GpuRegister>();
+      GpuRegister rhs = locations->InAt(1).AsRegister<GpuRegister>();
+      if (type == Primitive::kPrimInt)
+        __ DivR6(dst, lhs, rhs);
+      else
+        __ Ddiv(dst, lhs, rhs);
+      break;
+    }
+    case Primitive::kPrimFloat:
+    case Primitive::kPrimDouble: {
+      FpuRegister dst = locations->Out().AsFpuRegister<FpuRegister>();
+      FpuRegister lhs = locations->InAt(0).AsFpuRegister<FpuRegister>();
+      FpuRegister rhs = locations->InAt(1).AsFpuRegister<FpuRegister>();
+      if (type == Primitive::kPrimFloat)
+        __ DivS(dst, lhs, rhs);
+      else
+        __ DivD(dst, lhs, rhs);
+      break;
+    }
+    default:
+      LOG(FATAL) << "Unexpected div type " << type;
+  }
+}
+
+void LocationsBuilderMIPS64::VisitDivZeroCheck(HDivZeroCheck* instruction) {
+  LocationSummary* locations =
+      new (GetGraph()->GetArena()) LocationSummary(instruction, LocationSummary::kNoCall);
+  locations->SetInAt(0, Location::RegisterOrConstant(instruction->InputAt(0)));
+  if (instruction->HasUses()) {
+    locations->SetOut(Location::SameAsFirstInput());
+  }
+}
+
+void InstructionCodeGeneratorMIPS64::VisitDivZeroCheck(HDivZeroCheck* instruction) {
+  SlowPathCodeMIPS64* slow_path =
+      new (GetGraph()->GetArena()) DivZeroCheckSlowPathMIPS64(instruction);
+  codegen_->AddSlowPath(slow_path);
+  Location value = instruction->GetLocations()->InAt(0);
+
+  Primitive::Type type = instruction->GetType();
+
+  if ((type != Primitive::kPrimInt) && (type != Primitive::kPrimLong)) {
+      LOG(FATAL) << "Unexpected type " << type << " for DivZeroCheck.";
+  }
+
+  if (value.IsConstant()) {
+    int64_t divisor = codegen_->GetInt64ValueOf(value.GetConstant()->AsConstant());
+    if (divisor == 0) {
+      __ B(slow_path->GetEntryLabel());
+    } else {
+      // A division by a non-null constant is valid. We don't need to perform
+      // any check, so simply fall through.
+    }
+  } else {
+    __ Beqzc(value.AsRegister<GpuRegister>(), slow_path->GetEntryLabel());
+  }
+}
+
+void LocationsBuilderMIPS64::VisitDoubleConstant(HDoubleConstant* constant) {
+  LocationSummary* locations =
+      new (GetGraph()->GetArena()) LocationSummary(constant, LocationSummary::kNoCall);
+  locations->SetOut(Location::ConstantLocation(constant));
+}
+
+void InstructionCodeGeneratorMIPS64::VisitDoubleConstant(HDoubleConstant* cst ATTRIBUTE_UNUSED) {
+  // Will be generated at use site.
+}
+
+void LocationsBuilderMIPS64::VisitExit(HExit* exit) {
+  exit->SetLocations(nullptr);
+}
+
+void InstructionCodeGeneratorMIPS64::VisitExit(HExit* exit ATTRIBUTE_UNUSED) {
+}
+
+void LocationsBuilderMIPS64::VisitFloatConstant(HFloatConstant* constant) {
+  LocationSummary* locations =
+      new (GetGraph()->GetArena()) LocationSummary(constant, LocationSummary::kNoCall);
+  locations->SetOut(Location::ConstantLocation(constant));
+}
+
+void InstructionCodeGeneratorMIPS64::VisitFloatConstant(HFloatConstant* constant ATTRIBUTE_UNUSED) {
+  // Will be generated at use site.
+}
+
+void LocationsBuilderMIPS64::VisitGoto(HGoto* got) {
+  got->SetLocations(nullptr);
+}
+
+void InstructionCodeGeneratorMIPS64::VisitGoto(HGoto* got) {
+  HBasicBlock* successor = got->GetSuccessor();
+  DCHECK(!successor->IsExitBlock());
+  HBasicBlock* block = got->GetBlock();
+  HInstruction* previous = got->GetPrevious();
+  HLoopInformation* info = block->GetLoopInformation();
+
+  if (info != nullptr && info->IsBackEdge(*block) && info->HasSuspendCheck()) {
+    codegen_->ClearSpillSlotsFromLoopPhisInStackMap(info->GetSuspendCheck());
+    GenerateSuspendCheck(info->GetSuspendCheck(), successor);
+    return;
+  }
+  if (block->IsEntryBlock() && (previous != nullptr) && previous->IsSuspendCheck()) {
+    GenerateSuspendCheck(previous->AsSuspendCheck(), nullptr);
+  }
+  if (!codegen_->GoesToNextBlock(block, successor)) {
+    __ B(codegen_->GetLabelOf(successor));
+  }
+}
+
+void InstructionCodeGeneratorMIPS64::GenerateTestAndBranch(HInstruction* instruction,
+                                                           Label* true_target,
+                                                           Label* false_target,
+                                                           Label* always_true_target) {
+  HInstruction* cond = instruction->InputAt(0);
+  HCondition* condition = cond->AsCondition();
+
+  if (cond->IsIntConstant()) {
+    int32_t cond_value = cond->AsIntConstant()->GetValue();
+    if (cond_value == 1) {
+      if (always_true_target != nullptr) {
+        __ B(always_true_target);
+      }
+      return;
+    } else {
+      DCHECK_EQ(cond_value, 0);
+    }
+  } else if (!cond->IsCondition() || condition->NeedsMaterialization()) {
+    // The condition instruction has been materialized, compare the output to 0.
+    Location cond_val = instruction->GetLocations()->InAt(0);
+    DCHECK(cond_val.IsRegister());
+    __ Bnezc(cond_val.AsRegister<GpuRegister>(), true_target);
+  } else {
+    // The condition instruction has not been materialized, use its inputs as
+    // the comparison and its condition as the branch condition.
+    GpuRegister lhs = condition->GetLocations()->InAt(0).AsRegister<GpuRegister>();
+    Location rhs_location = condition->GetLocations()->InAt(1);
+    GpuRegister rhs_reg = ZERO;
+    int32_t rhs_imm = 0;
+    bool use_imm = rhs_location.IsConstant();
+    if (use_imm) {
+      rhs_imm = CodeGenerator::GetInt32ValueOf(rhs_location.GetConstant());
+    } else {
+      rhs_reg = rhs_location.AsRegister<GpuRegister>();
+    }
+
+    IfCondition if_cond = condition->GetCondition();
+    if (use_imm && rhs_imm == 0) {
+      switch (if_cond) {
+        case kCondEQ:
+          __ Beqzc(lhs, true_target);
+          break;
+        case kCondNE:
+          __ Bnezc(lhs, true_target);
+          break;
+        case kCondLT:
+          __ Bltzc(lhs, true_target);
+          break;
+        case kCondGE:
+          __ Bgezc(lhs, true_target);
+          break;
+        case kCondLE:
+          __ Blezc(lhs, true_target);
+          break;
+        case kCondGT:
+          __ Bgtzc(lhs, true_target);
+          break;
+      }
+    } else {
+      if (use_imm) {
+        rhs_reg = TMP;
+        __ LoadConst32(rhs_reg, rhs_imm);
+      }
+      // It looks like we can get here with lhs == rhs. Should that be possible at all?
+      // Mips R6 requires lhs != rhs for compact branches.
+      if (lhs == rhs_reg) {
+        DCHECK(!use_imm);
+        switch (if_cond) {
+          case kCondEQ:
+          case kCondGE:
+          case kCondLE:
+            // if lhs == rhs for a positive condition, then it is a branch
+            __ B(true_target);
+            break;
+          case kCondNE:
+          case kCondLT:
+          case kCondGT:
+            // if lhs == rhs for a negative condition, then it is a NOP
+            break;
+        }
+      } else {
+        switch (if_cond) {
+          case kCondEQ:
+            __ Beqc(lhs, rhs_reg, true_target);
+            break;
+          case kCondNE:
+            __ Bnec(lhs, rhs_reg, true_target);
+            break;
+          case kCondLT:
+            __ Bltc(lhs, rhs_reg, true_target);
+            break;
+          case kCondGE:
+            __ Bgec(lhs, rhs_reg, true_target);
+            break;
+          case kCondLE:
+            __ Bgec(rhs_reg, lhs, true_target);
+            break;
+          case kCondGT:
+            __ Bltc(rhs_reg, lhs, true_target);
+            break;
+        }
+      }
+    }
+  }
+  if (false_target != nullptr) {
+    __ B(false_target);
+  }
+}
+
+void LocationsBuilderMIPS64::VisitIf(HIf* if_instr) {
+  LocationSummary* locations = new (GetGraph()->GetArena()) LocationSummary(if_instr);
+  HInstruction* cond = if_instr->InputAt(0);
+  if (!cond->IsCondition() || cond->AsCondition()->NeedsMaterialization()) {
+    locations->SetInAt(0, Location::RequiresRegister());
+  }
+}
+
+void InstructionCodeGeneratorMIPS64::VisitIf(HIf* if_instr) {
+  Label* true_target = codegen_->GetLabelOf(if_instr->IfTrueSuccessor());
+  Label* false_target = codegen_->GetLabelOf(if_instr->IfFalseSuccessor());
+  Label* always_true_target = true_target;
+  if (codegen_->GoesToNextBlock(if_instr->GetBlock(),
+                                if_instr->IfTrueSuccessor())) {
+    always_true_target = nullptr;
+  }
+  if (codegen_->GoesToNextBlock(if_instr->GetBlock(),
+                                if_instr->IfFalseSuccessor())) {
+    false_target = nullptr;
+  }
+  GenerateTestAndBranch(if_instr, true_target, false_target, always_true_target);
+}
+
+void LocationsBuilderMIPS64::VisitDeoptimize(HDeoptimize* deoptimize) {
+  LocationSummary* locations = new (GetGraph()->GetArena())
+      LocationSummary(deoptimize, LocationSummary::kCallOnSlowPath);
+  HInstruction* cond = deoptimize->InputAt(0);
+  DCHECK(cond->IsCondition());
+  if (cond->AsCondition()->NeedsMaterialization()) {
+    locations->SetInAt(0, Location::RequiresRegister());
+  }
+}
+
+void InstructionCodeGeneratorMIPS64::VisitDeoptimize(HDeoptimize* deoptimize) {
+  SlowPathCodeMIPS64* slow_path = new (GetGraph()->GetArena())
+      DeoptimizationSlowPathMIPS64(deoptimize);
+  codegen_->AddSlowPath(slow_path);
+  Label* slow_path_entry = slow_path->GetEntryLabel();
+  GenerateTestAndBranch(deoptimize, slow_path_entry, nullptr, slow_path_entry);
+}
+
+void LocationsBuilderMIPS64::HandleFieldGet(HInstruction* instruction,
+                                            const FieldInfo& field_info ATTRIBUTE_UNUSED) {
+  LocationSummary* locations =
+      new (GetGraph()->GetArena()) LocationSummary(instruction, LocationSummary::kNoCall);
+  locations->SetInAt(0, Location::RequiresRegister());
+  if (Primitive::IsFloatingPointType(instruction->GetType())) {
+    locations->SetOut(Location::RequiresFpuRegister());
+  } else {
+    locations->SetOut(Location::RequiresRegister(), Location::kNoOutputOverlap);
+  }
+}
+
+void InstructionCodeGeneratorMIPS64::HandleFieldGet(HInstruction* instruction,
+                                                    const FieldInfo& field_info) {
+  Primitive::Type type = field_info.GetFieldType();
+  LocationSummary* locations = instruction->GetLocations();
+  GpuRegister obj = locations->InAt(0).AsRegister<GpuRegister>();
+  LoadOperandType load_type = kLoadUnsignedByte;
+  switch (type) {
+    case Primitive::kPrimBoolean:
+      load_type = kLoadUnsignedByte;
+      break;
+    case Primitive::kPrimByte:
+      load_type = kLoadSignedByte;
+      break;
+    case Primitive::kPrimShort:
+      load_type = kLoadSignedHalfword;
+      break;
+    case Primitive::kPrimChar:
+      load_type = kLoadUnsignedHalfword;
+      break;
+    case Primitive::kPrimInt:
+    case Primitive::kPrimFloat:
+      load_type = kLoadWord;
+      break;
+    case Primitive::kPrimLong:
+    case Primitive::kPrimDouble:
+      load_type = kLoadDoubleword;
+      break;
+    case Primitive::kPrimNot:
+      load_type = kLoadUnsignedWord;
+      break;
+    case Primitive::kPrimVoid:
+      LOG(FATAL) << "Unreachable type " << type;
+      UNREACHABLE();
+  }
+  if (!Primitive::IsFloatingPointType(type)) {
+    DCHECK(locations->Out().IsRegister());
+    GpuRegister dst = locations->Out().AsRegister<GpuRegister>();
+    __ LoadFromOffset(load_type, dst, obj, field_info.GetFieldOffset().Uint32Value());
+  } else {
+    DCHECK(locations->Out().IsFpuRegister());
+    FpuRegister dst = locations->Out().AsFpuRegister<FpuRegister>();
+    __ LoadFpuFromOffset(load_type, dst, obj, field_info.GetFieldOffset().Uint32Value());
+  }
+
+  codegen_->MaybeRecordImplicitNullCheck(instruction);
+  // TODO: memory barrier?
+}
+
+void LocationsBuilderMIPS64::HandleFieldSet(HInstruction* instruction,
+                                            const FieldInfo& field_info ATTRIBUTE_UNUSED) {
+  LocationSummary* locations =
+      new (GetGraph()->GetArena()) LocationSummary(instruction, LocationSummary::kNoCall);
+  locations->SetInAt(0, Location::RequiresRegister());
+  if (Primitive::IsFloatingPointType(instruction->InputAt(1)->GetType())) {
+    locations->SetInAt(1, Location::RequiresFpuRegister());
+  } else {
+    locations->SetInAt(1, Location::RequiresRegister());
+  }
+}
+
+void InstructionCodeGeneratorMIPS64::HandleFieldSet(HInstruction* instruction,
+                                                    const FieldInfo& field_info) {
+  Primitive::Type type = field_info.GetFieldType();
+  LocationSummary* locations = instruction->GetLocations();
+  GpuRegister obj = locations->InAt(0).AsRegister<GpuRegister>();
+  StoreOperandType store_type = kStoreByte;
+  switch (type) {
+    case Primitive::kPrimBoolean:
+    case Primitive::kPrimByte:
+      store_type = kStoreByte;
+      break;
+    case Primitive::kPrimShort:
+    case Primitive::kPrimChar:
+      store_type = kStoreHalfword;
+      break;
+    case Primitive::kPrimInt:
+    case Primitive::kPrimFloat:
+    case Primitive::kPrimNot:
+      store_type = kStoreWord;
+      break;
+    case Primitive::kPrimLong:
+    case Primitive::kPrimDouble:
+      store_type = kStoreDoubleword;
+      break;
+    case Primitive::kPrimVoid:
+      LOG(FATAL) << "Unreachable type " << type;
+      UNREACHABLE();
+  }
+  if (!Primitive::IsFloatingPointType(type)) {
+    DCHECK(locations->InAt(1).IsRegister());
+    GpuRegister src = locations->InAt(1).AsRegister<GpuRegister>();
+    __ StoreToOffset(store_type, src, obj, field_info.GetFieldOffset().Uint32Value());
+  } else {
+    DCHECK(locations->InAt(1).IsFpuRegister());
+    FpuRegister src = locations->InAt(1).AsFpuRegister<FpuRegister>();
+    __ StoreFpuToOffset(store_type, src, obj, field_info.GetFieldOffset().Uint32Value());
+  }
+
+  codegen_->MaybeRecordImplicitNullCheck(instruction);
+  // TODO: memory barriers?
+  if (CodeGenerator::StoreNeedsWriteBarrier(type, instruction->InputAt(1))) {
+    DCHECK(locations->InAt(1).IsRegister());
+    GpuRegister src = locations->InAt(1).AsRegister<GpuRegister>();
+    codegen_->MarkGCCard(obj, src);
+  }
+}
+
+void LocationsBuilderMIPS64::VisitInstanceFieldGet(HInstanceFieldGet* instruction) {
+  HandleFieldGet(instruction, instruction->GetFieldInfo());
+}
+
+void InstructionCodeGeneratorMIPS64::VisitInstanceFieldGet(HInstanceFieldGet* instruction) {
+  HandleFieldGet(instruction, instruction->GetFieldInfo());
+}
+
+void LocationsBuilderMIPS64::VisitInstanceFieldSet(HInstanceFieldSet* instruction) {
+  HandleFieldSet(instruction, instruction->GetFieldInfo());
+}
+
+void InstructionCodeGeneratorMIPS64::VisitInstanceFieldSet(HInstanceFieldSet* instruction) {
+  HandleFieldSet(instruction, instruction->GetFieldInfo());
+}
+
+void LocationsBuilderMIPS64::VisitInstanceOf(HInstanceOf* instruction) {
+  LocationSummary::CallKind call_kind =
+      instruction->IsClassFinal() ? LocationSummary::kNoCall : LocationSummary::kCallOnSlowPath;
+  LocationSummary* locations = new (GetGraph()->GetArena()) LocationSummary(instruction, call_kind);
+  locations->SetInAt(0, Location::RequiresRegister());
+  locations->SetInAt(1, Location::RequiresRegister());
+  // The output does overlap inputs.
+  locations->SetOut(Location::RequiresRegister(), Location::kOutputOverlap);
+}
+
+void InstructionCodeGeneratorMIPS64::VisitInstanceOf(HInstanceOf* instruction) {
+  LocationSummary* locations = instruction->GetLocations();
+  GpuRegister obj = locations->InAt(0).AsRegister<GpuRegister>();
+  GpuRegister cls = locations->InAt(1).AsRegister<GpuRegister>();
+  GpuRegister out = locations->Out().AsRegister<GpuRegister>();
+
+  Label done;
+
+  // Return 0 if `obj` is null.
+  // TODO: Avoid this check if we know `obj` is not null.
+  __ Move(out, ZERO);
+  __ Beqzc(obj, &done);
+
+  // Compare the class of `obj` with `cls`.
+  __ LoadFromOffset(kLoadUnsignedWord, out, obj, mirror::Object::ClassOffset().Int32Value());
+  if (instruction->IsClassFinal()) {
+    // Classes must be equal for the instanceof to succeed.
+    __ Xor(out, out, cls);
+    __ Sltiu(out, out, 1);
+  } else {
+    // If the classes are not equal, we go into a slow path.
+    DCHECK(locations->OnlyCallsOnSlowPath());
+    SlowPathCodeMIPS64* slow_path =
+        new (GetGraph()->GetArena()) TypeCheckSlowPathMIPS64(instruction,
+                                                             locations->InAt(1),
+                                                             locations->Out(),
+                                                             instruction->GetDexPc());
+    codegen_->AddSlowPath(slow_path);
+    __ Bnec(out, cls, slow_path->GetEntryLabel());
+    __ LoadConst32(out, 1);
+    __ Bind(slow_path->GetExitLabel());
+  }
+
+  __ Bind(&done);
+}
+
+void LocationsBuilderMIPS64::VisitIntConstant(HIntConstant* constant) {
+  LocationSummary* locations = new (GetGraph()->GetArena()) LocationSummary(constant);
+  locations->SetOut(Location::ConstantLocation(constant));
+}
+
+void InstructionCodeGeneratorMIPS64::VisitIntConstant(HIntConstant* constant ATTRIBUTE_UNUSED) {
+  // Will be generated at use site.
+}
+
+void LocationsBuilderMIPS64::VisitNullConstant(HNullConstant* constant) {
+  LocationSummary* locations = new (GetGraph()->GetArena()) LocationSummary(constant);
+  locations->SetOut(Location::ConstantLocation(constant));
+}
+
+void InstructionCodeGeneratorMIPS64::VisitNullConstant(HNullConstant* constant ATTRIBUTE_UNUSED) {
+  // Will be generated at use site.
+}
+
+void LocationsBuilderMIPS64::HandleInvoke(HInvoke* invoke) {
+  InvokeDexCallingConventionVisitorMIPS64 calling_convention_visitor;
+  CodeGenerator::CreateCommonInvokeLocationSummary(invoke, &calling_convention_visitor);
+}
+
+void LocationsBuilderMIPS64::VisitInvokeInterface(HInvokeInterface* invoke) {
+  HandleInvoke(invoke);
+  // The register T0 is required to be used for the hidden argument in
+  // art_quick_imt_conflict_trampoline, so add the hidden argument.
+  invoke->GetLocations()->AddTemp(Location::RegisterLocation(T0));
+}
+
+void InstructionCodeGeneratorMIPS64::VisitInvokeInterface(HInvokeInterface* invoke) {
+  // TODO: b/18116999, our IMTs can miss an IncompatibleClassChangeError.
+  GpuRegister temp = invoke->GetLocations()->GetTemp(0).AsRegister<GpuRegister>();
+  uint32_t method_offset = mirror::Class::EmbeddedImTableEntryOffset(
+      invoke->GetImtIndex() % mirror::Class::kImtSize, kMips64PointerSize).Uint32Value();
+  Location receiver = invoke->GetLocations()->InAt(0);
+  uint32_t class_offset = mirror::Object::ClassOffset().Int32Value();
+  Offset entry_point = ArtMethod::EntryPointFromQuickCompiledCodeOffset(kMips64WordSize);
+
+  // Set the hidden argument.
+  __ LoadConst32(invoke->GetLocations()->GetTemp(1).AsRegister<GpuRegister>(),
+                 invoke->GetDexMethodIndex());
+
+  // temp = object->GetClass();
+  if (receiver.IsStackSlot()) {
+    __ LoadFromOffset(kLoadUnsignedWord, temp, SP, receiver.GetStackIndex());
+    __ LoadFromOffset(kLoadUnsignedWord, temp, temp, class_offset);
+  } else {
+    __ LoadFromOffset(kLoadUnsignedWord, temp, receiver.AsRegister<GpuRegister>(), class_offset);
+  }
+  codegen_->MaybeRecordImplicitNullCheck(invoke);
+  // temp = temp->GetImtEntryAt(method_offset);
+  __ LoadFromOffset(kLoadDoubleword, temp, temp, method_offset);
+  // T9 = temp->GetEntryPoint();
+  __ LoadFromOffset(kLoadDoubleword, T9, temp, entry_point.Int32Value());
+  // T9();
+  __ Jalr(T9);
+  DCHECK(!codegen_->IsLeafMethod());
+  codegen_->RecordPcInfo(invoke, invoke->GetDexPc());
+}
+
+void LocationsBuilderMIPS64::VisitInvokeVirtual(HInvokeVirtual* invoke) {
+  // TODO intrinsic function
+  HandleInvoke(invoke);
+}
+
+void LocationsBuilderMIPS64::VisitInvokeStaticOrDirect(HInvokeStaticOrDirect* invoke) {
+  // When we do not run baseline, explicit clinit checks triggered by static
+  // invokes must have been pruned by art::PrepareForRegisterAllocation.
+  DCHECK(codegen_->IsBaseline() || !invoke->IsStaticWithExplicitClinitCheck());
+
+  // TODO - intrinsic function
+  HandleInvoke(invoke);
+
+  // While SetupBlockedRegisters() blocks registers S2-S8 due to their
+  // clobbering somewhere else, reduce further register pressure by avoiding
+  // allocation of a register for the current method pointer like on x86 baseline.
+  // TODO: remove this once all the issues with register saving/restoring are
+  // sorted out.
+  LocationSummary* locations = invoke->GetLocations();
+  Location location = locations->InAt(invoke->GetCurrentMethodInputIndex());
+  if (location.IsUnallocated() && location.GetPolicy() == Location::kRequiresRegister) {
+    locations->SetInAt(invoke->GetCurrentMethodInputIndex(), Location::NoLocation());
+  }
+}
+
+static bool TryGenerateIntrinsicCode(HInvoke* invoke,
+                                     CodeGeneratorMIPS64* codegen ATTRIBUTE_UNUSED) {
+  if (invoke->GetLocations()->Intrinsified()) {
+    // TODO - intrinsic function
+    return true;
+  }
+  return false;
+}
+
+void CodeGeneratorMIPS64::GenerateStaticOrDirectCall(HInvokeStaticOrDirect* invoke, Location temp) {
+  // All registers are assumed to be correctly set up per the calling convention.
+
+  // TODO: Implement all kinds of calls:
+  // 1) boot -> boot
+  // 2) app -> boot
+  // 3) app -> app
+  //
+  // Currently we implement the app -> app logic, which looks up in the resolve cache.
+
+  if (invoke->IsStringInit()) {
+    GpuRegister reg = temp.AsRegister<GpuRegister>();
+    // temp = thread->string_init_entrypoint
+    __ LoadFromOffset(kLoadDoubleword,
+                      reg,
+                      TR,
+                      invoke->GetStringInitOffset());
+    // T9 = temp->entry_point_from_quick_compiled_code_;
+    __ LoadFromOffset(kLoadDoubleword,
+                      T9,
+                      reg,
+                      ArtMethod::EntryPointFromQuickCompiledCodeOffset(
+                          kMips64WordSize).Int32Value());
+    // T9()
+    __ Jalr(T9);
+  } else if (invoke->IsRecursive()) {
+    __ Jalr(&frame_entry_label_, T9);
+  } else {
+    Location current_method = invoke->GetLocations()->InAt(invoke->GetCurrentMethodInputIndex());
+    GpuRegister reg = temp.AsRegister<GpuRegister>();
+    GpuRegister method_reg;
+    if (current_method.IsRegister()) {
+      method_reg = current_method.AsRegister<GpuRegister>();
+    } else {
+      // TODO: use the appropriate DCHECK() here if possible.
+      // DCHECK(invoke->GetLocations()->Intrinsified());
+      DCHECK(!current_method.IsValid());
+      method_reg = reg;
+      __ Ld(reg, SP, kCurrentMethodStackOffset);
+    }
+
+    // temp = temp->dex_cache_resolved_methods_;
+    __ LoadFromOffset(kLoadUnsignedWord,
+                      reg,
+                      method_reg,
+                      ArtMethod::DexCacheResolvedMethodsOffset().Int32Value());
+    // temp = temp[index_in_cache]
+    __ LoadFromOffset(kLoadDoubleword,
+                      reg,
+                      reg,
+                      CodeGenerator::GetCachePointerOffset(invoke->GetDexMethodIndex()));
+    // T9 = temp[offset_of_quick_compiled_code]
+    __ LoadFromOffset(kLoadDoubleword,
+                      T9,
+                      reg,
+                      ArtMethod::EntryPointFromQuickCompiledCodeOffset(
+                          kMips64WordSize).Int32Value());
+    // T9()
+    __ Jalr(T9);
+  }
+
+  DCHECK(!IsLeafMethod());
+}
+
+void InstructionCodeGeneratorMIPS64::VisitInvokeStaticOrDirect(HInvokeStaticOrDirect* invoke) {
+  // When we do not run baseline, explicit clinit checks triggered by static
+  // invokes must have been pruned by art::PrepareForRegisterAllocation.
+  DCHECK(codegen_->IsBaseline() || !invoke->IsStaticWithExplicitClinitCheck());
+
+  if (TryGenerateIntrinsicCode(invoke, codegen_)) {
+    return;
+  }
+
+  LocationSummary* locations = invoke->GetLocations();
+  codegen_->GenerateStaticOrDirectCall(invoke,
+                                       locations->HasTemps()
+                                           ? locations->GetTemp(0)
+                                           : Location::NoLocation());
+  codegen_->RecordPcInfo(invoke, invoke->GetDexPc());
+}
+
+void InstructionCodeGeneratorMIPS64::VisitInvokeVirtual(HInvokeVirtual* invoke) {
+  // TODO: Try to generate intrinsics code.
+  LocationSummary* locations = invoke->GetLocations();
+  Location receiver = locations->InAt(0);
+  GpuRegister temp = invoke->GetLocations()->GetTemp(0).AsRegister<GpuRegister>();
+  size_t method_offset = mirror::Class::EmbeddedVTableEntryOffset(
+      invoke->GetVTableIndex(), kMips64PointerSize).SizeValue();
+  uint32_t class_offset = mirror::Object::ClassOffset().Int32Value();
+  Offset entry_point = ArtMethod::EntryPointFromQuickCompiledCodeOffset(kMips64WordSize);
+
+  // temp = object->GetClass();
+  DCHECK(receiver.IsRegister());
+  __ LoadFromOffset(kLoadUnsignedWord, temp, receiver.AsRegister<GpuRegister>(), class_offset);
+  codegen_->MaybeRecordImplicitNullCheck(invoke);
+  // temp = temp->GetMethodAt(method_offset);
+  __ LoadFromOffset(kLoadDoubleword, temp, temp, method_offset);
+  // T9 = temp->GetEntryPoint();
+  __ LoadFromOffset(kLoadDoubleword, T9, temp, entry_point.Int32Value());
+  // T9();
+  __ Jalr(T9);
+  DCHECK(!codegen_->IsLeafMethod());
+  codegen_->RecordPcInfo(invoke, invoke->GetDexPc());
+}
+
+void LocationsBuilderMIPS64::VisitLoadClass(HLoadClass* cls) {
+  LocationSummary::CallKind call_kind = cls->CanCallRuntime() ? LocationSummary::kCallOnSlowPath
+                                                              : LocationSummary::kNoCall;
+  LocationSummary* locations = new (GetGraph()->GetArena()) LocationSummary(cls, call_kind);
+  locations->SetInAt(0, Location::RequiresRegister());
+  locations->SetOut(Location::RequiresRegister());
+}
+
+void InstructionCodeGeneratorMIPS64::VisitLoadClass(HLoadClass* cls) {
+  LocationSummary* locations = cls->GetLocations();
+  GpuRegister out = locations->Out().AsRegister<GpuRegister>();
+  GpuRegister current_method = locations->InAt(0).AsRegister<GpuRegister>();
+  if (cls->IsReferrersClass()) {
+    DCHECK(!cls->CanCallRuntime());
+    DCHECK(!cls->MustGenerateClinitCheck());
+    __ LoadFromOffset(kLoadUnsignedWord, out, current_method,
+                      ArtMethod::DeclaringClassOffset().Int32Value());
+  } else {
+    DCHECK(cls->CanCallRuntime());
+    __ LoadFromOffset(kLoadUnsignedWord, out, current_method,
+                      ArtMethod::DexCacheResolvedTypesOffset().Int32Value());
+    __ LoadFromOffset(kLoadUnsignedWord, out, out, CodeGenerator::GetCacheOffset(cls->GetTypeIndex()));
+    SlowPathCodeMIPS64* slow_path = new (GetGraph()->GetArena()) LoadClassSlowPathMIPS64(
+        cls,
+        cls,
+        cls->GetDexPc(),
+        cls->MustGenerateClinitCheck());
+    codegen_->AddSlowPath(slow_path);
+    __ Beqzc(out, slow_path->GetEntryLabel());
+    if (cls->MustGenerateClinitCheck()) {
+      GenerateClassInitializationCheck(slow_path, out);
+    } else {
+      __ Bind(slow_path->GetExitLabel());
+    }
+  }
+}
+
+void LocationsBuilderMIPS64::VisitLoadException(HLoadException* load) {
+  LocationSummary* locations =
+      new (GetGraph()->GetArena()) LocationSummary(load, LocationSummary::kNoCall);
+  locations->SetOut(Location::RequiresRegister());
+}
+
+void InstructionCodeGeneratorMIPS64::VisitLoadException(HLoadException* load) {
+  GpuRegister out = load->GetLocations()->Out().AsRegister<GpuRegister>();
+  __ LoadFromOffset(kLoadUnsignedWord, out, TR, Thread::ExceptionOffset<kMips64WordSize>().Int32Value());
+  __ StoreToOffset(kStoreWord, ZERO, TR, Thread::ExceptionOffset<kMips64WordSize>().Int32Value());
+}
+
+void LocationsBuilderMIPS64::VisitLoadLocal(HLoadLocal* load) {
+  load->SetLocations(nullptr);
+}
+
+void InstructionCodeGeneratorMIPS64::VisitLoadLocal(HLoadLocal* load ATTRIBUTE_UNUSED) {
+  // Nothing to do, this is driven by the code generator.
+}
+
+void LocationsBuilderMIPS64::VisitLoadString(HLoadString* load) {
+  LocationSummary* locations =
+      new (GetGraph()->GetArena()) LocationSummary(load, LocationSummary::kCallOnSlowPath);
+  locations->SetInAt(0, Location::RequiresRegister());
+  locations->SetOut(Location::RequiresRegister());
+}
+
+void InstructionCodeGeneratorMIPS64::VisitLoadString(HLoadString* load) {
+  SlowPathCodeMIPS64* slow_path = new (GetGraph()->GetArena()) LoadStringSlowPathMIPS64(load);
+  codegen_->AddSlowPath(slow_path);
+
+  LocationSummary* locations = load->GetLocations();
+  GpuRegister out = locations->Out().AsRegister<GpuRegister>();
+  GpuRegister current_method = locations->InAt(0).AsRegister<GpuRegister>();
+  __ LoadFromOffset(kLoadUnsignedWord, out, current_method,
+                    ArtMethod::DeclaringClassOffset().Int32Value());
+  __ LoadFromOffset(kLoadUnsignedWord, out, out, mirror::Class::DexCacheStringsOffset().Int32Value());
+  __ LoadFromOffset(kLoadUnsignedWord, out, out, CodeGenerator::GetCacheOffset(load->GetStringIndex()));
+  __ Beqzc(out, slow_path->GetEntryLabel());
+  __ Bind(slow_path->GetExitLabel());
+}
+
+void LocationsBuilderMIPS64::VisitLocal(HLocal* local) {
+  local->SetLocations(nullptr);
+}
+
+void InstructionCodeGeneratorMIPS64::VisitLocal(HLocal* local) {
+  DCHECK_EQ(local->GetBlock(), GetGraph()->GetEntryBlock());
+}
+
+void LocationsBuilderMIPS64::VisitLongConstant(HLongConstant* constant) {
+  LocationSummary* locations = new (GetGraph()->GetArena()) LocationSummary(constant);
+  locations->SetOut(Location::ConstantLocation(constant));
+}
+
+void InstructionCodeGeneratorMIPS64::VisitLongConstant(HLongConstant* constant ATTRIBUTE_UNUSED) {
+  // Will be generated at use site.
+}
+
+void LocationsBuilderMIPS64::VisitMonitorOperation(HMonitorOperation* instruction) {
+  LocationSummary* locations =
+      new (GetGraph()->GetArena()) LocationSummary(instruction, LocationSummary::kCall);
+  InvokeRuntimeCallingConvention calling_convention;
+  locations->SetInAt(0, Location::RegisterLocation(calling_convention.GetRegisterAt(0)));
+}
+
+void InstructionCodeGeneratorMIPS64::VisitMonitorOperation(HMonitorOperation* instruction) {
+  codegen_->InvokeRuntime(instruction->IsEnter()
+                              ? QUICK_ENTRY_POINT(pLockObject)
+                              : QUICK_ENTRY_POINT(pUnlockObject),
+                          instruction,
+                          instruction->GetDexPc(),
+                          nullptr);
+  CheckEntrypointTypes<kQuickLockObject, void, mirror::Object*>();
+}
+
+void LocationsBuilderMIPS64::VisitMul(HMul* mul) {
+  LocationSummary* locations =
+      new (GetGraph()->GetArena()) LocationSummary(mul, LocationSummary::kNoCall);
+  switch (mul->GetResultType()) {
+    case Primitive::kPrimInt:
+    case Primitive::kPrimLong:
+      locations->SetInAt(0, Location::RequiresRegister());
+      locations->SetInAt(1, Location::RequiresRegister());
+      locations->SetOut(Location::RequiresRegister(), Location::kNoOutputOverlap);
+      break;
+
+    case Primitive::kPrimFloat:
+    case Primitive::kPrimDouble:
+      locations->SetInAt(0, Location::RequiresFpuRegister());
+      locations->SetInAt(1, Location::RequiresFpuRegister());
+      locations->SetOut(Location::RequiresFpuRegister(), Location::kNoOutputOverlap);
+      break;
+
+    default:
+      LOG(FATAL) << "Unexpected mul type " << mul->GetResultType();
+  }
+}
+
+void InstructionCodeGeneratorMIPS64::VisitMul(HMul* instruction) {
+  Primitive::Type type = instruction->GetType();
+  LocationSummary* locations = instruction->GetLocations();
+
+  switch (type) {
+    case Primitive::kPrimInt:
+    case Primitive::kPrimLong: {
+      GpuRegister dst = locations->Out().AsRegister<GpuRegister>();
+      GpuRegister lhs = locations->InAt(0).AsRegister<GpuRegister>();
+      GpuRegister rhs = locations->InAt(1).AsRegister<GpuRegister>();
+      if (type == Primitive::kPrimInt)
+        __ MulR6(dst, lhs, rhs);
+      else
+        __ Dmul(dst, lhs, rhs);
+      break;
+    }
+    case Primitive::kPrimFloat:
+    case Primitive::kPrimDouble: {
+      FpuRegister dst = locations->Out().AsFpuRegister<FpuRegister>();
+      FpuRegister lhs = locations->InAt(0).AsFpuRegister<FpuRegister>();
+      FpuRegister rhs = locations->InAt(1).AsFpuRegister<FpuRegister>();
+      if (type == Primitive::kPrimFloat)
+        __ MulS(dst, lhs, rhs);
+      else
+        __ MulD(dst, lhs, rhs);
+      break;
+    }
+    default:
+      LOG(FATAL) << "Unexpected mul type " << type;
+  }
+}
+
+void LocationsBuilderMIPS64::VisitNeg(HNeg* neg) {
+  LocationSummary* locations =
+      new (GetGraph()->GetArena()) LocationSummary(neg, LocationSummary::kNoCall);
+  switch (neg->GetResultType()) {
+    case Primitive::kPrimInt:
+    case Primitive::kPrimLong:
+      locations->SetInAt(0, Location::RequiresRegister());
+      locations->SetOut(Location::RequiresRegister(), Location::kNoOutputOverlap);
+      break;
+
+    case Primitive::kPrimFloat:
+    case Primitive::kPrimDouble:
+      locations->SetInAt(0, Location::RequiresFpuRegister());
+      locations->SetOut(Location::RequiresFpuRegister(), Location::kNoOutputOverlap);
+      break;
+
+    default:
+      LOG(FATAL) << "Unexpected neg type " << neg->GetResultType();
+  }
+}
+
+void InstructionCodeGeneratorMIPS64::VisitNeg(HNeg* instruction) {
+  Primitive::Type type = instruction->GetType();
+  LocationSummary* locations = instruction->GetLocations();
+
+  switch (type) {
+    case Primitive::kPrimInt:
+    case Primitive::kPrimLong: {
+      GpuRegister dst = locations->Out().AsRegister<GpuRegister>();
+      GpuRegister src = locations->InAt(0).AsRegister<GpuRegister>();
+      if (type == Primitive::kPrimInt)
+        __ Subu(dst, ZERO, src);
+      else
+        __ Dsubu(dst, ZERO, src);
+      break;
+    }
+    case Primitive::kPrimFloat:
+    case Primitive::kPrimDouble: {
+      FpuRegister dst = locations->Out().AsFpuRegister<FpuRegister>();
+      FpuRegister src = locations->InAt(0).AsFpuRegister<FpuRegister>();
+      if (type == Primitive::kPrimFloat)
+        __ NegS(dst, src);
+      else
+        __ NegD(dst, src);
+      break;
+    }
+    default:
+      LOG(FATAL) << "Unexpected neg type " << type;
+  }
+}
+
+void LocationsBuilderMIPS64::VisitNewArray(HNewArray* instruction) {
+  LocationSummary* locations =
+      new (GetGraph()->GetArena()) LocationSummary(instruction, LocationSummary::kCall);
+  InvokeRuntimeCallingConvention calling_convention;
+  locations->AddTemp(Location::RegisterLocation(calling_convention.GetRegisterAt(0)));
+  locations->SetOut(calling_convention.GetReturnLocation(Primitive::kPrimNot));
+  locations->SetInAt(0, Location::RegisterLocation(calling_convention.GetRegisterAt(1)));
+  locations->SetInAt(1, Location::RegisterLocation(calling_convention.GetRegisterAt(2)));
+}
+
+void InstructionCodeGeneratorMIPS64::VisitNewArray(HNewArray* instruction) {
+  LocationSummary* locations = instruction->GetLocations();
+  // Move an uint16_t value to a register.
+  __ LoadConst32(locations->GetTemp(0).AsRegister<GpuRegister>(), instruction->GetTypeIndex());
+  codegen_->InvokeRuntime(
+      GetThreadOffset<kMips64WordSize>(instruction->GetEntrypoint()).Int32Value(),
+      instruction,
+      instruction->GetDexPc(),
+      nullptr);
+  CheckEntrypointTypes<kQuickAllocArrayWithAccessCheck, void*, uint32_t, int32_t, ArtMethod*>();
+}
+
+void LocationsBuilderMIPS64::VisitNewInstance(HNewInstance* instruction) {
+  LocationSummary* locations =
+      new (GetGraph()->GetArena()) LocationSummary(instruction, LocationSummary::kCall);
+  InvokeRuntimeCallingConvention calling_convention;
+  locations->AddTemp(Location::RegisterLocation(calling_convention.GetRegisterAt(0)));
+  locations->SetInAt(0, Location::RegisterLocation(calling_convention.GetRegisterAt(1)));
+  locations->SetOut(calling_convention.GetReturnLocation(Primitive::kPrimNot));
+}
+
+void InstructionCodeGeneratorMIPS64::VisitNewInstance(HNewInstance* instruction) {
+  LocationSummary* locations = instruction->GetLocations();
+  // Move an uint16_t value to a register.
+  __ LoadConst32(locations->GetTemp(0).AsRegister<GpuRegister>(), instruction->GetTypeIndex());
+  codegen_->InvokeRuntime(
+      GetThreadOffset<kMips64WordSize>(instruction->GetEntrypoint()).Int32Value(),
+      instruction,
+      instruction->GetDexPc(),
+      nullptr);
+  CheckEntrypointTypes<kQuickAllocObjectWithAccessCheck, void*, uint32_t, ArtMethod*>();
+}
+
+void LocationsBuilderMIPS64::VisitNot(HNot* instruction) {
+  LocationSummary* locations = new (GetGraph()->GetArena()) LocationSummary(instruction);
+  locations->SetInAt(0, Location::RequiresRegister());
+  locations->SetOut(Location::RequiresRegister(), Location::kNoOutputOverlap);
+}
+
+void InstructionCodeGeneratorMIPS64::VisitNot(HNot* instruction) {
+  Primitive::Type type = instruction->GetType();
+  LocationSummary* locations = instruction->GetLocations();
+
+  switch (type) {
+    case Primitive::kPrimInt:
+    case Primitive::kPrimLong: {
+      GpuRegister dst = locations->Out().AsRegister<GpuRegister>();
+      GpuRegister src = locations->InAt(0).AsRegister<GpuRegister>();
+      __ Nor(dst, src, ZERO);
+      break;
+    }
+
+    default:
+      LOG(FATAL) << "Unexpected type for not operation " << instruction->GetResultType();
+  }
+}
+
+void LocationsBuilderMIPS64::VisitBooleanNot(HBooleanNot* instruction) {
+  LocationSummary* locations = new (GetGraph()->GetArena()) LocationSummary(instruction);
+  locations->SetInAt(0, Location::RequiresRegister());
+  locations->SetOut(Location::RequiresRegister(), Location::kNoOutputOverlap);
+}
+
+void InstructionCodeGeneratorMIPS64::VisitBooleanNot(HBooleanNot* instruction) {
+  LocationSummary* locations = instruction->GetLocations();
+  __ Xori(locations->Out().AsRegister<GpuRegister>(),
+          locations->InAt(0).AsRegister<GpuRegister>(),
+          1);
+}
+
+void LocationsBuilderMIPS64::VisitNullCheck(HNullCheck* instruction) {
+  LocationSummary* locations =
+      new (GetGraph()->GetArena()) LocationSummary(instruction, LocationSummary::kNoCall);
+  locations->SetInAt(0, Location::RequiresRegister());
+  if (instruction->HasUses()) {
+    locations->SetOut(Location::SameAsFirstInput());
+  }
+}
+
+void InstructionCodeGeneratorMIPS64::GenerateImplicitNullCheck(HNullCheck* instruction) {
+  if (codegen_->CanMoveNullCheckToUser(instruction)) {
+    return;
+  }
+  Location obj = instruction->GetLocations()->InAt(0);
+
+  __ Lw(ZERO, obj.AsRegister<GpuRegister>(), 0);
+  codegen_->RecordPcInfo(instruction, instruction->GetDexPc());
+}
+
+void InstructionCodeGeneratorMIPS64::GenerateExplicitNullCheck(HNullCheck* instruction) {
+  SlowPathCodeMIPS64* slow_path = new (GetGraph()->GetArena()) NullCheckSlowPathMIPS64(instruction);
+  codegen_->AddSlowPath(slow_path);
+
+  Location obj = instruction->GetLocations()->InAt(0);
+
+  __ Beqzc(obj.AsRegister<GpuRegister>(), slow_path->GetEntryLabel());
+}
+
+void InstructionCodeGeneratorMIPS64::VisitNullCheck(HNullCheck* instruction) {
+  if (codegen_->GetCompilerOptions().GetImplicitNullChecks()) {
+    GenerateImplicitNullCheck(instruction);
+  } else {
+    GenerateExplicitNullCheck(instruction);
+  }
+}
+
+void LocationsBuilderMIPS64::VisitOr(HOr* instruction) {
+  HandleBinaryOp(instruction);
+}
+
+void InstructionCodeGeneratorMIPS64::VisitOr(HOr* instruction) {
+  HandleBinaryOp(instruction);
+}
+
+void LocationsBuilderMIPS64::VisitParallelMove(HParallelMove* instruction ATTRIBUTE_UNUSED) {
+  LOG(FATAL) << "Unreachable";
+}
+
+void InstructionCodeGeneratorMIPS64::VisitParallelMove(HParallelMove* instruction) {
+  codegen_->GetMoveResolver()->EmitNativeCode(instruction);
+}
+
+void LocationsBuilderMIPS64::VisitParameterValue(HParameterValue* instruction) {
+  LocationSummary* locations = new (GetGraph()->GetArena()) LocationSummary(instruction);
+  Location location = parameter_visitor_.GetNextLocation(instruction->GetType());
+  if (location.IsStackSlot()) {
+    location = Location::StackSlot(location.GetStackIndex() + codegen_->GetFrameSize());
+  } else if (location.IsDoubleStackSlot()) {
+    location = Location::DoubleStackSlot(location.GetStackIndex() + codegen_->GetFrameSize());
+  }
+  locations->SetOut(location);
+}
+
+void InstructionCodeGeneratorMIPS64::VisitParameterValue(HParameterValue* instruction
+                                                         ATTRIBUTE_UNUSED) {
+  // Nothing to do, the parameter is already at its location.
+}
+
+void LocationsBuilderMIPS64::VisitCurrentMethod(HCurrentMethod* instruction) {
+  LocationSummary* locations =
+      new (GetGraph()->GetArena()) LocationSummary(instruction, LocationSummary::kNoCall);
+  locations->SetOut(Location::RegisterLocation(kMethodRegisterArgument));
+}
+
+void InstructionCodeGeneratorMIPS64::VisitCurrentMethod(HCurrentMethod* instruction
+                                                        ATTRIBUTE_UNUSED) {
+  // Nothing to do, the method is already at its location.
+}
+
+void LocationsBuilderMIPS64::VisitPhi(HPhi* instruction) {
+  LocationSummary* locations = new (GetGraph()->GetArena()) LocationSummary(instruction);
+  for (size_t i = 0, e = instruction->InputCount(); i < e; ++i) {
+    locations->SetInAt(i, Location::Any());
+  }
+  locations->SetOut(Location::Any());
+}
+
+void InstructionCodeGeneratorMIPS64::VisitPhi(HPhi* instruction ATTRIBUTE_UNUSED) {
+  LOG(FATAL) << "Unreachable";
+}
+
+void LocationsBuilderMIPS64::VisitRem(HRem* rem) {
+  Primitive::Type type = rem->GetResultType();
+  LocationSummary::CallKind call_kind =
+      Primitive::IsFloatingPointType(type) ? LocationSummary::kCall : LocationSummary::kNoCall;
+  LocationSummary* locations = new (GetGraph()->GetArena()) LocationSummary(rem, call_kind);
+
+  switch (type) {
+    case Primitive::kPrimInt:
+    case Primitive::kPrimLong:
+      locations->SetInAt(0, Location::RequiresRegister());
+      locations->SetInAt(1, Location::RequiresRegister());
+      locations->SetOut(Location::RequiresRegister(), Location::kNoOutputOverlap);
+      break;
+
+    case Primitive::kPrimFloat:
+    case Primitive::kPrimDouble: {
+      InvokeRuntimeCallingConvention calling_convention;
+      locations->SetInAt(0, Location::FpuRegisterLocation(calling_convention.GetFpuRegisterAt(0)));
+      locations->SetInAt(1, Location::FpuRegisterLocation(calling_convention.GetFpuRegisterAt(1)));
+      locations->SetOut(calling_convention.GetReturnLocation(type));
+      break;
+    }
+
+    default:
+      LOG(FATAL) << "Unexpected rem type " << type;
+  }
+}
+
+void InstructionCodeGeneratorMIPS64::VisitRem(HRem* instruction) {
+  Primitive::Type type = instruction->GetType();
+  LocationSummary* locations = instruction->GetLocations();
+
+  switch (type) {
+    case Primitive::kPrimInt:
+    case Primitive::kPrimLong: {
+      GpuRegister dst = locations->Out().AsRegister<GpuRegister>();
+      GpuRegister lhs = locations->InAt(0).AsRegister<GpuRegister>();
+      GpuRegister rhs = locations->InAt(1).AsRegister<GpuRegister>();
+      if (type == Primitive::kPrimInt)
+        __ ModR6(dst, lhs, rhs);
+      else
+        __ Dmod(dst, lhs, rhs);
+      break;
+    }
+
+    case Primitive::kPrimFloat:
+    case Primitive::kPrimDouble: {
+      int32_t entry_offset = (type == Primitive::kPrimFloat) ? QUICK_ENTRY_POINT(pFmodf)
+                                                             : QUICK_ENTRY_POINT(pFmod);
+      codegen_->InvokeRuntime(entry_offset, instruction, instruction->GetDexPc(), nullptr);
+      break;
+    }
+    default:
+      LOG(FATAL) << "Unexpected rem type " << type;
+  }
+}
+
+void LocationsBuilderMIPS64::VisitMemoryBarrier(HMemoryBarrier* memory_barrier) {
+  memory_barrier->SetLocations(nullptr);
+}
+
+void InstructionCodeGeneratorMIPS64::VisitMemoryBarrier(HMemoryBarrier* memory_barrier) {
+  GenerateMemoryBarrier(memory_barrier->GetBarrierKind());
+}
+
+void LocationsBuilderMIPS64::VisitReturn(HReturn* ret) {
+  LocationSummary* locations = new (GetGraph()->GetArena()) LocationSummary(ret);
+  Primitive::Type return_type = ret->InputAt(0)->GetType();
+  locations->SetInAt(0, Mips64ReturnLocation(return_type));
+}
+
+void InstructionCodeGeneratorMIPS64::VisitReturn(HReturn* ret ATTRIBUTE_UNUSED) {
+  codegen_->GenerateFrameExit();
+}
+
+void LocationsBuilderMIPS64::VisitReturnVoid(HReturnVoid* ret) {
+  ret->SetLocations(nullptr);
+}
+
+void InstructionCodeGeneratorMIPS64::VisitReturnVoid(HReturnVoid* ret ATTRIBUTE_UNUSED) {
+  codegen_->GenerateFrameExit();
+}
+
+void LocationsBuilderMIPS64::VisitShl(HShl* shl) {
+  HandleShift(shl);
+}
+
+void InstructionCodeGeneratorMIPS64::VisitShl(HShl* shl) {
+  HandleShift(shl);
+}
+
+void LocationsBuilderMIPS64::VisitShr(HShr* shr) {
+  HandleShift(shr);
+}
+
+void InstructionCodeGeneratorMIPS64::VisitShr(HShr* shr) {
+  HandleShift(shr);
+}
+
+void LocationsBuilderMIPS64::VisitStoreLocal(HStoreLocal* store) {
+  LocationSummary* locations = new (GetGraph()->GetArena()) LocationSummary(store);
+  Primitive::Type field_type = store->InputAt(1)->GetType();
+  switch (field_type) {
+    case Primitive::kPrimNot:
+    case Primitive::kPrimBoolean:
+    case Primitive::kPrimByte:
+    case Primitive::kPrimChar:
+    case Primitive::kPrimShort:
+    case Primitive::kPrimInt:
+    case Primitive::kPrimFloat:
+      locations->SetInAt(1, Location::StackSlot(codegen_->GetStackSlot(store->GetLocal())));
+      break;
+
+    case Primitive::kPrimLong:
+    case Primitive::kPrimDouble:
+      locations->SetInAt(1, Location::DoubleStackSlot(codegen_->GetStackSlot(store->GetLocal())));
+      break;
+
+    default:
+      LOG(FATAL) << "Unimplemented local type " << field_type;
+  }
+}
+
+void InstructionCodeGeneratorMIPS64::VisitStoreLocal(HStoreLocal* store ATTRIBUTE_UNUSED) {
+}
+
+void LocationsBuilderMIPS64::VisitSub(HSub* instruction) {
+  HandleBinaryOp(instruction);
+}
+
+void InstructionCodeGeneratorMIPS64::VisitSub(HSub* instruction) {
+  HandleBinaryOp(instruction);
+}
+
+void LocationsBuilderMIPS64::VisitStaticFieldGet(HStaticFieldGet* instruction) {
+  HandleFieldGet(instruction, instruction->GetFieldInfo());
+}
+
+void InstructionCodeGeneratorMIPS64::VisitStaticFieldGet(HStaticFieldGet* instruction) {
+  HandleFieldGet(instruction, instruction->GetFieldInfo());
+}
+
+void LocationsBuilderMIPS64::VisitStaticFieldSet(HStaticFieldSet* instruction) {
+  HandleFieldSet(instruction, instruction->GetFieldInfo());
+}
+
+void InstructionCodeGeneratorMIPS64::VisitStaticFieldSet(HStaticFieldSet* instruction) {
+  HandleFieldSet(instruction, instruction->GetFieldInfo());
+}
+
+void LocationsBuilderMIPS64::VisitSuspendCheck(HSuspendCheck* instruction) {
+  new (GetGraph()->GetArena()) LocationSummary(instruction, LocationSummary::kCallOnSlowPath);
+}
+
+void InstructionCodeGeneratorMIPS64::VisitSuspendCheck(HSuspendCheck* instruction) {
+  HBasicBlock* block = instruction->GetBlock();
+  if (block->GetLoopInformation() != nullptr) {
+    DCHECK(block->GetLoopInformation()->GetSuspendCheck() == instruction);
+    // The back edge will generate the suspend check.
+    return;
+  }
+  if (block->IsEntryBlock() && instruction->GetNext()->IsGoto()) {
+    // The goto will generate the suspend check.
+    return;
+  }
+  GenerateSuspendCheck(instruction, nullptr);
+}
+
+void LocationsBuilderMIPS64::VisitTemporary(HTemporary* temp) {
+  temp->SetLocations(nullptr);
+}
+
+void InstructionCodeGeneratorMIPS64::VisitTemporary(HTemporary* temp ATTRIBUTE_UNUSED) {
+  // Nothing to do, this is driven by the code generator.
+}
+
+void LocationsBuilderMIPS64::VisitThrow(HThrow* instruction) {
+  LocationSummary* locations =
+      new (GetGraph()->GetArena()) LocationSummary(instruction, LocationSummary::kCall);
+  InvokeRuntimeCallingConvention calling_convention;
+  locations->SetInAt(0, Location::RegisterLocation(calling_convention.GetRegisterAt(0)));
+}
+
+void InstructionCodeGeneratorMIPS64::VisitThrow(HThrow* instruction) {
+  codegen_->InvokeRuntime(QUICK_ENTRY_POINT(pDeliverException),
+                          instruction,
+                          instruction->GetDexPc(),
+                          nullptr);
+  CheckEntrypointTypes<kQuickDeliverException, void, mirror::Object*>();
+}
+
+void LocationsBuilderMIPS64::VisitTypeConversion(HTypeConversion* conversion) {
+  Primitive::Type input_type = conversion->GetInputType();
+  Primitive::Type result_type = conversion->GetResultType();
+  DCHECK_NE(input_type, result_type);
+
+  if ((input_type == Primitive::kPrimNot) || (input_type == Primitive::kPrimVoid) ||
+      (result_type == Primitive::kPrimNot) || (result_type == Primitive::kPrimVoid)) {
+    LOG(FATAL) << "Unexpected type conversion from " << input_type << " to " << result_type;
+  }
+
+  LocationSummary::CallKind call_kind = LocationSummary::kNoCall;
+  if ((Primitive::IsFloatingPointType(result_type) && input_type == Primitive::kPrimLong) ||
+      (Primitive::IsIntegralType(result_type) && Primitive::IsFloatingPointType(input_type))) {
+    call_kind = LocationSummary::kCall;
+  }
+
+  LocationSummary* locations = new (GetGraph()->GetArena()) LocationSummary(conversion, call_kind);
+
+  if (call_kind == LocationSummary::kNoCall) {
+    if (Primitive::IsFloatingPointType(input_type)) {
+      locations->SetInAt(0, Location::RequiresFpuRegister());
+    } else {
+      locations->SetInAt(0, Location::RequiresRegister());
+    }
+
+    if (Primitive::IsFloatingPointType(result_type)) {
+      locations->SetOut(Location::RequiresFpuRegister(), Location::kNoOutputOverlap);
+    } else {
+      locations->SetOut(Location::RequiresRegister(), Location::kNoOutputOverlap);
+    }
+  } else {
+    InvokeRuntimeCallingConvention calling_convention;
+
+    if (Primitive::IsFloatingPointType(input_type)) {
+      locations->SetInAt(0, Location::FpuRegisterLocation(calling_convention.GetFpuRegisterAt(0)));
+    } else {
+      locations->SetInAt(0, Location::RegisterLocation(calling_convention.GetRegisterAt(0)));
+    }
+
+    locations->SetOut(calling_convention.GetReturnLocation(result_type));
+  }
+}
+
+void InstructionCodeGeneratorMIPS64::VisitTypeConversion(HTypeConversion* conversion) {
+  LocationSummary* locations = conversion->GetLocations();
+  Primitive::Type result_type = conversion->GetResultType();
+  Primitive::Type input_type = conversion->GetInputType();
+
+  DCHECK_NE(input_type, result_type);
+
+  if (Primitive::IsIntegralType(result_type) && Primitive::IsIntegralType(input_type)) {
+    GpuRegister dst = locations->Out().AsRegister<GpuRegister>();
+    GpuRegister src = locations->InAt(0).AsRegister<GpuRegister>();
+
+    switch (result_type) {
+      case Primitive::kPrimChar:
+        __ Andi(dst, src, 0xFFFF);
+        break;
+      case Primitive::kPrimByte:
+        // long is never converted into types narrower than int directly,
+        // so SEB and SEH can be used without ever causing unpredictable results
+        // on 64-bit inputs
+        DCHECK(input_type != Primitive::kPrimLong);
+        __ Seb(dst, src);
+        break;
+      case Primitive::kPrimShort:
+        // long is never converted into types narrower than int directly,
+        // so SEB and SEH can be used without ever causing unpredictable results
+        // on 64-bit inputs
+        DCHECK(input_type != Primitive::kPrimLong);
+        __ Seh(dst, src);
+        break;
+      case Primitive::kPrimInt:
+      case Primitive::kPrimLong:
+        // Sign-extend 32-bit int into bits 32 through 63 for
+        // int-to-long and long-to-int conversions
+        __ Sll(dst, src, 0);
+        break;
+
+      default:
+        LOG(FATAL) << "Unexpected type conversion from " << input_type
+                   << " to " << result_type;
+    }
+  } else if (Primitive::IsFloatingPointType(result_type) && Primitive::IsIntegralType(input_type)) {
+    if (input_type != Primitive::kPrimLong) {
+      FpuRegister dst = locations->Out().AsFpuRegister<FpuRegister>();
+      GpuRegister src = locations->InAt(0).AsRegister<GpuRegister>();
+      __ Mtc1(src, FTMP);
+      if (result_type == Primitive::kPrimFloat) {
+        __ Cvtsw(dst, FTMP);
+      } else {
+        __ Cvtdw(dst, FTMP);
+      }
+    } else {
+      int32_t entry_offset = (result_type == Primitive::kPrimFloat) ? QUICK_ENTRY_POINT(pL2f)
+                                                                    : QUICK_ENTRY_POINT(pL2d);
+      codegen_->InvokeRuntime(entry_offset,
+                              conversion,
+                              conversion->GetDexPc(),
+                              nullptr);
+    }
+  } else if (Primitive::IsIntegralType(result_type) && Primitive::IsFloatingPointType(input_type)) {
+    CHECK(result_type == Primitive::kPrimInt || result_type == Primitive::kPrimLong);
+    int32_t entry_offset;
+    if (result_type != Primitive::kPrimLong) {
+      entry_offset = (input_type == Primitive::kPrimFloat) ? QUICK_ENTRY_POINT(pF2iz)
+                                                           : QUICK_ENTRY_POINT(pD2iz);
+    } else {
+      entry_offset = (input_type == Primitive::kPrimFloat) ? QUICK_ENTRY_POINT(pF2l)
+                                                           : QUICK_ENTRY_POINT(pD2l);
+    }
+    codegen_->InvokeRuntime(entry_offset,
+                            conversion,
+                            conversion->GetDexPc(),
+                            nullptr);
+  } else if (Primitive::IsFloatingPointType(result_type) &&
+             Primitive::IsFloatingPointType(input_type)) {
+    FpuRegister dst = locations->Out().AsFpuRegister<FpuRegister>();
+    FpuRegister src = locations->InAt(0).AsFpuRegister<FpuRegister>();
+    if (result_type == Primitive::kPrimFloat) {
+      __ Cvtsd(dst, src);
+    } else {
+      __ Cvtds(dst, src);
+    }
+  } else {
+    LOG(FATAL) << "Unexpected or unimplemented type conversion from " << input_type
+                << " to " << result_type;
+  }
+}
+
+void LocationsBuilderMIPS64::VisitUShr(HUShr* ushr) {
+  HandleShift(ushr);
+}
+
+void InstructionCodeGeneratorMIPS64::VisitUShr(HUShr* ushr) {
+  HandleShift(ushr);
+}
+
+void LocationsBuilderMIPS64::VisitXor(HXor* instruction) {
+  HandleBinaryOp(instruction);
+}
+
+void InstructionCodeGeneratorMIPS64::VisitXor(HXor* instruction) {
+  HandleBinaryOp(instruction);
+}
+
+void LocationsBuilderMIPS64::VisitBoundType(HBoundType* instruction ATTRIBUTE_UNUSED) {
+  // Nothing to do, this should be removed during prepare for register allocator.
+  LOG(FATAL) << "Unreachable";
+}
+
+void InstructionCodeGeneratorMIPS64::VisitBoundType(HBoundType* instruction ATTRIBUTE_UNUSED) {
+  // Nothing to do, this should be removed during prepare for register allocator.
+  LOG(FATAL) << "Unreachable";
+}
+
+void LocationsBuilderMIPS64::VisitEqual(HEqual* comp) {
+  VisitCondition(comp);
+}
+
+void InstructionCodeGeneratorMIPS64::VisitEqual(HEqual* comp) {
+  VisitCondition(comp);
+}
+
+void LocationsBuilderMIPS64::VisitNotEqual(HNotEqual* comp) {
+  VisitCondition(comp);
+}
+
+void InstructionCodeGeneratorMIPS64::VisitNotEqual(HNotEqual* comp) {
+  VisitCondition(comp);
+}
+
+void LocationsBuilderMIPS64::VisitLessThan(HLessThan* comp) {
+  VisitCondition(comp);
+}
+
+void InstructionCodeGeneratorMIPS64::VisitLessThan(HLessThan* comp) {
+  VisitCondition(comp);
+}
+
+void LocationsBuilderMIPS64::VisitLessThanOrEqual(HLessThanOrEqual* comp) {
+  VisitCondition(comp);
+}
+
+void InstructionCodeGeneratorMIPS64::VisitLessThanOrEqual(HLessThanOrEqual* comp) {
+  VisitCondition(comp);
+}
+
+void LocationsBuilderMIPS64::VisitGreaterThan(HGreaterThan* comp) {
+  VisitCondition(comp);
+}
+
+void InstructionCodeGeneratorMIPS64::VisitGreaterThan(HGreaterThan* comp) {
+  VisitCondition(comp);
+}
+
+void LocationsBuilderMIPS64::VisitGreaterThanOrEqual(HGreaterThanOrEqual* comp) {
+  VisitCondition(comp);
+}
+
+void InstructionCodeGeneratorMIPS64::VisitGreaterThanOrEqual(HGreaterThanOrEqual* comp) {
+  VisitCondition(comp);
+}
+
+}  // namespace mips64
+}  // namespace art
diff --git a/compiler/optimizing/code_generator_mips64.h b/compiler/optimizing/code_generator_mips64.h
new file mode 100644
index 0000000..534154f
--- /dev/null
+++ b/compiler/optimizing/code_generator_mips64.h
@@ -0,0 +1,301 @@
+/*
+ * Copyright (C) 2015 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_OPTIMIZING_CODE_GENERATOR_MIPS64_H_
+#define ART_COMPILER_OPTIMIZING_CODE_GENERATOR_MIPS64_H_
+
+#include "code_generator.h"
+#include "dex/compiler_enums.h"
+#include "driver/compiler_options.h"
+#include "nodes.h"
+#include "parallel_move_resolver.h"
+#include "utils/mips64/assembler_mips64.h"
+
+namespace art {
+namespace mips64 {
+
+// Use a local definition to prevent copying mistakes.
+static constexpr size_t kMips64WordSize = kMips64PointerSize;
+
+
+// InvokeDexCallingConvention registers
+
+static constexpr GpuRegister kParameterCoreRegisters[] =
+    { A1, A2, A3, A4, A5, A6, A7 };
+static constexpr size_t kParameterCoreRegistersLength = arraysize(kParameterCoreRegisters);
+
+static constexpr FpuRegister kParameterFpuRegisters[] =
+    { F13, F14, F15, F16, F17, F18, F19 };
+static constexpr size_t kParameterFpuRegistersLength = arraysize(kParameterFpuRegisters);
+
+
+// InvokeRuntimeCallingConvention registers
+
+static constexpr GpuRegister kRuntimeParameterCoreRegisters[] =
+    { A0, A1, A2, A3, A4, A5, A6, A7 };
+static constexpr size_t kRuntimeParameterCoreRegistersLength =
+    arraysize(kRuntimeParameterCoreRegisters);
+
+static constexpr FpuRegister kRuntimeParameterFpuRegisters[] =
+    { F12, F13, F14, F15, F16, F17, F18, F19 };
+static constexpr size_t kRuntimeParameterFpuRegistersLength =
+    arraysize(kRuntimeParameterFpuRegisters);
+
+
+static constexpr GpuRegister kCoreCalleeSaves[] =
+    { S0, S1, S2, S3, S4, S5, S6, S7, GP, S8, RA };  // TODO: review
+static constexpr FpuRegister kFpuCalleeSaves[] =
+    { F24, F25, F26, F27, F28, F29, F30, F31 };
+
+
+class CodeGeneratorMIPS64;
+
+class InvokeDexCallingConvention : public CallingConvention<GpuRegister, FpuRegister> {
+ public:
+  InvokeDexCallingConvention()
+      : CallingConvention(kParameterCoreRegisters,
+                          kParameterCoreRegistersLength,
+                          kParameterFpuRegisters,
+                          kParameterFpuRegistersLength,
+                          kMips64PointerSize) {}
+
+ private:
+  DISALLOW_COPY_AND_ASSIGN(InvokeDexCallingConvention);
+};
+
+class InvokeDexCallingConventionVisitorMIPS64 : public InvokeDexCallingConventionVisitor {
+ public:
+  InvokeDexCallingConventionVisitorMIPS64() {}
+  virtual ~InvokeDexCallingConventionVisitorMIPS64() {}
+
+  Location GetNextLocation(Primitive::Type type) OVERRIDE;
+  Location GetReturnLocation(Primitive::Type type) const OVERRIDE;
+  Location GetMethodLocation() const OVERRIDE;
+
+ private:
+  InvokeDexCallingConvention calling_convention;
+
+  DISALLOW_COPY_AND_ASSIGN(InvokeDexCallingConventionVisitorMIPS64);
+};
+
+class InvokeRuntimeCallingConvention : public CallingConvention<GpuRegister, FpuRegister> {
+ public:
+  InvokeRuntimeCallingConvention()
+      : CallingConvention(kRuntimeParameterCoreRegisters,
+                          kRuntimeParameterCoreRegistersLength,
+                          kRuntimeParameterFpuRegisters,
+                          kRuntimeParameterFpuRegistersLength,
+                          kMips64PointerSize) {}
+
+  Location GetReturnLocation(Primitive::Type return_type);
+
+ private:
+  DISALLOW_COPY_AND_ASSIGN(InvokeRuntimeCallingConvention);
+};
+
+class ParallelMoveResolverMIPS64 : public ParallelMoveResolverWithSwap {
+ public:
+  ParallelMoveResolverMIPS64(ArenaAllocator* allocator, CodeGeneratorMIPS64* codegen)
+      : ParallelMoveResolverWithSwap(allocator), codegen_(codegen) {}
+
+  void EmitMove(size_t index) OVERRIDE;
+  void EmitSwap(size_t index) OVERRIDE;
+  void SpillScratch(int reg) OVERRIDE;
+  void RestoreScratch(int reg) OVERRIDE;
+
+  void Exchange(int index1, int index2, bool double_slot);
+
+  Mips64Assembler* GetAssembler() const;
+
+ private:
+  CodeGeneratorMIPS64* const codegen_;
+
+  DISALLOW_COPY_AND_ASSIGN(ParallelMoveResolverMIPS64);
+};
+
+class SlowPathCodeMIPS64 : public SlowPathCode {
+ public:
+  SlowPathCodeMIPS64() : entry_label_(), exit_label_() {}
+
+  Label* GetEntryLabel() { return &entry_label_; }
+  Label* GetExitLabel() { return &exit_label_; }
+
+ private:
+  Label entry_label_;
+  Label exit_label_;
+
+  DISALLOW_COPY_AND_ASSIGN(SlowPathCodeMIPS64);
+};
+
+class LocationsBuilderMIPS64 : public HGraphVisitor {
+ public:
+  LocationsBuilderMIPS64(HGraph* graph, CodeGeneratorMIPS64* codegen)
+      : HGraphVisitor(graph), codegen_(codegen) {}
+
+#define DECLARE_VISIT_INSTRUCTION(name, super)     \
+  void Visit##name(H##name* instr);
+
+  FOR_EACH_CONCRETE_INSTRUCTION(DECLARE_VISIT_INSTRUCTION)
+
+#undef DECLARE_VISIT_INSTRUCTION
+
+ private:
+  void HandleInvoke(HInvoke* invoke);
+  void HandleBinaryOp(HBinaryOperation* operation);
+  void HandleShift(HBinaryOperation* operation);
+  void HandleFieldSet(HInstruction* instruction, const FieldInfo& field_info);
+  void HandleFieldGet(HInstruction* instruction, const FieldInfo& field_info);
+
+  InvokeDexCallingConventionVisitorMIPS64 parameter_visitor_;
+
+  CodeGeneratorMIPS64* const codegen_;
+
+  DISALLOW_COPY_AND_ASSIGN(LocationsBuilderMIPS64);
+};
+
+class InstructionCodeGeneratorMIPS64 : public HGraphVisitor {
+ public:
+  InstructionCodeGeneratorMIPS64(HGraph* graph, CodeGeneratorMIPS64* codegen);
+
+#define DECLARE_VISIT_INSTRUCTION(name, super)     \
+  void Visit##name(H##name* instr);
+
+  FOR_EACH_CONCRETE_INSTRUCTION(DECLARE_VISIT_INSTRUCTION)
+
+#undef DECLARE_VISIT_INSTRUCTION
+
+  Mips64Assembler* GetAssembler() const { return assembler_; }
+
+ private:
+  // Generate code for the given suspend check. If not null, `successor`
+  // is the block to branch to if the suspend check is not needed, and after
+  // the suspend call.
+  void GenerateClassInitializationCheck(SlowPathCodeMIPS64* slow_path, GpuRegister class_reg);
+  void GenerateMemoryBarrier(MemBarrierKind kind);
+  void GenerateSuspendCheck(HSuspendCheck* check, HBasicBlock* successor);
+  void HandleBinaryOp(HBinaryOperation* operation);
+  void HandleShift(HBinaryOperation* operation);
+  void HandleFieldSet(HInstruction* instruction, const FieldInfo& field_info);
+  void HandleFieldGet(HInstruction* instruction, const FieldInfo& field_info);
+  void GenerateImplicitNullCheck(HNullCheck* instruction);
+  void GenerateExplicitNullCheck(HNullCheck* instruction);
+  void GenerateTestAndBranch(HInstruction* instruction,
+                             Label* true_target,
+                             Label* false_target,
+                             Label* always_true_target);
+
+  Mips64Assembler* const assembler_;
+  CodeGeneratorMIPS64* const codegen_;
+
+  DISALLOW_COPY_AND_ASSIGN(InstructionCodeGeneratorMIPS64);
+};
+
+class CodeGeneratorMIPS64 : public CodeGenerator {
+ public:
+  CodeGeneratorMIPS64(HGraph* graph,
+                      const Mips64InstructionSetFeatures& isa_features,
+                      const CompilerOptions& compiler_options);
+  virtual ~CodeGeneratorMIPS64() {}
+
+  void GenerateFrameEntry() OVERRIDE;
+  void GenerateFrameExit() OVERRIDE;
+
+  void Bind(HBasicBlock* block) OVERRIDE;
+
+  void Move(HInstruction* instruction, Location location, HInstruction* move_for) OVERRIDE;
+
+  size_t GetWordSize() const OVERRIDE { return kMips64WordSize; }
+
+  size_t GetFloatingPointSpillSlotSize() const OVERRIDE { return kMips64WordSize; }
+
+  uintptr_t GetAddressOf(HBasicBlock* block) const OVERRIDE {
+    return GetLabelOf(block)->Position();
+  }
+
+  HGraphVisitor* GetLocationBuilder() OVERRIDE { return &location_builder_; }
+  HGraphVisitor* GetInstructionVisitor() OVERRIDE { return &instruction_visitor_; }
+  Mips64Assembler* GetAssembler() OVERRIDE { return &assembler_; }
+
+  void MarkGCCard(GpuRegister object, GpuRegister value);
+
+  // Register allocation.
+
+  void SetupBlockedRegisters(bool is_baseline) const OVERRIDE;
+  // AllocateFreeRegister() is only used when allocating registers locally
+  // during CompileBaseline().
+  Location AllocateFreeRegister(Primitive::Type type) const OVERRIDE;
+
+  Location GetStackLocation(HLoadLocal* load) const OVERRIDE;
+
+  size_t SaveCoreRegister(size_t stack_index, uint32_t reg_id);
+  size_t RestoreCoreRegister(size_t stack_index, uint32_t reg_id);
+  size_t SaveFloatingPointRegister(size_t stack_index, uint32_t reg_id);
+  size_t RestoreFloatingPointRegister(size_t stack_index, uint32_t reg_id);
+
+  void DumpCoreRegister(std::ostream& stream, int reg) const OVERRIDE;
+  void DumpFloatingPointRegister(std::ostream& stream, int reg) const OVERRIDE;
+
+  InstructionSet GetInstructionSet() const OVERRIDE { return InstructionSet::kMips64; }
+
+  const Mips64InstructionSetFeatures& GetInstructionSetFeatures() const {
+    return isa_features_;
+  }
+
+  Label* GetLabelOf(HBasicBlock* block) const {
+    return CommonGetLabelOf<Label>(block_labels_.GetRawStorage(), block);
+  }
+
+  void Initialize() OVERRIDE {
+    block_labels_.SetSize(GetGraph()->GetBlocks().Size());
+  }
+
+  void Finalize(CodeAllocator* allocator) OVERRIDE;
+
+  // Code generation helpers.
+
+  void MoveLocation(Location destination, Location source, Primitive::Type type);
+
+  void SwapLocations(Location loc1, Location loc2, Primitive::Type type);
+
+  // Generate code to invoke a runtime entry point.
+  void InvokeRuntime(int32_t offset,
+                     HInstruction* instruction,
+                     uint32_t dex_pc,
+                     SlowPathCode* slow_path);
+
+  ParallelMoveResolver* GetMoveResolver() OVERRIDE { return &move_resolver_; }
+
+  bool NeedsTwoRegisters(Primitive::Type type ATTRIBUTE_UNUSED) const { return false; }
+
+  void GenerateStaticOrDirectCall(HInvokeStaticOrDirect* invoke, Location temp);
+
+ private:
+  // Labels for each block that will be compiled.
+  GrowableArray<Label> block_labels_;
+  Label frame_entry_label_;
+  LocationsBuilderMIPS64 location_builder_;
+  InstructionCodeGeneratorMIPS64 instruction_visitor_;
+  ParallelMoveResolverMIPS64 move_resolver_;
+  Mips64Assembler assembler_;
+  const Mips64InstructionSetFeatures& isa_features_;
+
+  DISALLOW_COPY_AND_ASSIGN(CodeGeneratorMIPS64);
+};
+
+}  // namespace mips64
+}  // namespace art
+
+#endif  // ART_COMPILER_OPTIMIZING_CODE_GENERATOR_MIPS64_H_
diff --git a/compiler/optimizing/code_generator_x86.cc b/compiler/optimizing/code_generator_x86.cc
index e39a1c2..4d106c4 100644
--- a/compiler/optimizing/code_generator_x86.cc
+++ b/compiler/optimizing/code_generator_x86.cc
@@ -56,6 +56,8 @@
     RecordPcInfo(codegen, instruction_, instruction_->GetDexPc());
   }
 
+  const char* GetDescription() const OVERRIDE { return "NullCheckSlowPathX86"; }
+
  private:
   HNullCheck* const instruction_;
   DISALLOW_COPY_AND_ASSIGN(NullCheckSlowPathX86);
@@ -71,6 +73,8 @@
     RecordPcInfo(codegen, instruction_, instruction_->GetDexPc());
   }
 
+  const char* GetDescription() const OVERRIDE { return "DivZeroCheckSlowPathX86"; }
+
  private:
   HDivZeroCheck* const instruction_;
   DISALLOW_COPY_AND_ASSIGN(DivZeroCheckSlowPathX86);
@@ -90,6 +94,8 @@
     __ jmp(GetExitLabel());
   }
 
+  const char* GetDescription() const OVERRIDE { return "DivRemMinusOneSlowPathX86"; }
+
  private:
   Register reg_;
   bool is_div_;
@@ -122,6 +128,8 @@
     RecordPcInfo(codegen, instruction_, instruction_->GetDexPc());
   }
 
+  const char* GetDescription() const OVERRIDE { return "BoundsCheckSlowPathX86"; }
+
  private:
   HBoundsCheck* const instruction_;
   const Location index_location_;
@@ -158,6 +166,8 @@
     return successor_;
   }
 
+  const char* GetDescription() const OVERRIDE { return "SuspendCheckSlowPathX86"; }
+
  private:
   HSuspendCheck* const instruction_;
   HBasicBlock* const successor_;
@@ -188,6 +198,8 @@
     __ jmp(GetExitLabel());
   }
 
+  const char* GetDescription() const OVERRIDE { return "LoadStringSlowPathX86"; }
+
  private:
   HLoadString* const instruction_;
 
@@ -228,6 +240,8 @@
     __ jmp(GetExitLabel());
   }
 
+  const char* GetDescription() const OVERRIDE { return "LoadClassSlowPathX86"; }
+
  private:
   // The class this slow path will load.
   HLoadClass* const cls_;
@@ -293,6 +307,8 @@
     __ jmp(GetExitLabel());
   }
 
+  const char* GetDescription() const OVERRIDE { return "TypeCheckSlowPathX86"; }
+
  private:
   HInstruction* const instruction_;
   const Location class_to_check_;
@@ -318,6 +334,8 @@
     codegen->RecordPcInfo(instruction_, dex_pc, this);
   }
 
+  const char* GetDescription() const OVERRIDE { return "DeoptimizationSlowPathX86"; }
+
  private:
   HInstruction* const instruction_;
   DISALLOW_COPY_AND_ASSIGN(DeoptimizationSlowPathX86);
diff --git a/compiler/optimizing/code_generator_x86_64.cc b/compiler/optimizing/code_generator_x86_64.cc
index bfc827d..e64540b 100644
--- a/compiler/optimizing/code_generator_x86_64.cc
+++ b/compiler/optimizing/code_generator_x86_64.cc
@@ -59,6 +59,8 @@
     RecordPcInfo(codegen, instruction_, instruction_->GetDexPc());
   }
 
+  const char* GetDescription() const OVERRIDE { return "NullCheckSlowPathX86_64"; }
+
  private:
   HNullCheck* const instruction_;
   DISALLOW_COPY_AND_ASSIGN(NullCheckSlowPathX86_64);
@@ -75,6 +77,8 @@
     RecordPcInfo(codegen, instruction_, instruction_->GetDexPc());
   }
 
+  const char* GetDescription() const OVERRIDE { return "DivZeroCheckSlowPathX86_64"; }
+
  private:
   HDivZeroCheck* const instruction_;
   DISALLOW_COPY_AND_ASSIGN(DivZeroCheckSlowPathX86_64);
@@ -105,6 +109,8 @@
     __ jmp(GetExitLabel());
   }
 
+  const char* GetDescription() const OVERRIDE { return "DivRemMinusOneSlowPathX86_64"; }
+
  private:
   const CpuRegister cpu_reg_;
   const Primitive::Type type_;
@@ -140,6 +146,8 @@
     return successor_;
   }
 
+  const char* GetDescription() const OVERRIDE { return "SuspendCheckSlowPathX86_64"; }
+
  private:
   HSuspendCheck* const instruction_;
   HBasicBlock* const successor_;
@@ -174,6 +182,8 @@
     RecordPcInfo(codegen, instruction_, instruction_->GetDexPc());
   }
 
+  const char* GetDescription() const OVERRIDE { return "BoundsCheckSlowPathX86_64"; }
+
  private:
   HBoundsCheck* const instruction_;
   const Location index_location_;
@@ -217,6 +227,8 @@
     __ jmp(GetExitLabel());
   }
 
+  const char* GetDescription() const OVERRIDE { return "LoadClassSlowPathX86_64"; }
+
  private:
   // The class this slow path will load.
   HLoadClass* const cls_;
@@ -257,6 +269,8 @@
     __ jmp(GetExitLabel());
   }
 
+  const char* GetDescription() const OVERRIDE { return "LoadStringSlowPathX86_64"; }
+
  private:
   HLoadString* const instruction_;
 
@@ -312,6 +326,8 @@
     __ jmp(GetExitLabel());
   }
 
+  const char* GetDescription() const OVERRIDE { return "TypeCheckSlowPathX86_64"; }
+
  private:
   HInstruction* const instruction_;
   const Location class_to_check_;
@@ -337,6 +353,8 @@
     codegen->RecordPcInfo(instruction_, dex_pc, this);
   }
 
+  const char* GetDescription() const OVERRIDE { return "DeoptimizationSlowPathX86_64"; }
+
  private:
   HInstruction* const instruction_;
   DISALLOW_COPY_AND_ASSIGN(DeoptimizationSlowPathX86_64);
diff --git a/compiler/optimizing/codegen_test.cc b/compiler/optimizing/codegen_test.cc
index bfed1a8..ca85cf5 100644
--- a/compiler/optimizing/codegen_test.cc
+++ b/compiler/optimizing/codegen_test.cc
@@ -20,6 +20,8 @@
 #include "arch/arm/instruction_set_features_arm.h"
 #include "arch/arm/registers_arm.h"
 #include "arch/arm64/instruction_set_features_arm64.h"
+#include "arch/mips64/instruction_set_features_mips64.h"
+#include "arch/mips64/registers_mips64.h"
 #include "arch/x86/instruction_set_features_x86.h"
 #include "arch/x86/registers_x86.h"
 #include "arch/x86_64/instruction_set_features_x86_64.h"
@@ -27,6 +29,7 @@
 #include "builder.h"
 #include "code_generator_arm.h"
 #include "code_generator_arm64.h"
+#include "code_generator_mips64.h"
 #include "code_generator_x86.h"
 #include "code_generator_x86_64.h"
 #include "common_compiler_test.h"
@@ -40,6 +43,7 @@
 #include "ssa_liveness_analysis.h"
 #include "utils.h"
 #include "utils/arm/managed_register_arm.h"
+#include "utils/mips64/managed_register_mips64.h"
 #include "utils/x86/managed_register_x86.h"
 
 #include "gtest/gtest.h"
@@ -172,6 +176,14 @@
   if (kRuntimeISA == kArm64) {
     Run(allocator, codegenARM64, has_result, expected);
   }
+
+  std::unique_ptr<const Mips64InstructionSetFeatures> features_mips64(
+      Mips64InstructionSetFeatures::FromCppDefines());
+  mips64::CodeGeneratorMIPS64 codegenMIPS64(graph, *features_mips64.get(), compiler_options);
+  codegenMIPS64.CompileBaseline(&allocator, true);
+  if (kRuntimeISA == kMips64) {
+    Run(allocator, codegenMIPS64, has_result, expected);
+  }
 }
 
 template <typename Expected>
@@ -219,6 +231,11 @@
         X86_64InstructionSetFeatures::FromCppDefines());
     x86_64::CodeGeneratorX86_64 codegenX86_64(graph, *features_x86_64.get(), compiler_options);
     RunCodeOptimized(&codegenX86_64, graph, hook_before_codegen, has_result, expected);
+  } else if (kRuntimeISA == kMips64) {
+    std::unique_ptr<const Mips64InstructionSetFeatures> features_mips64(
+        Mips64InstructionSetFeatures::FromCppDefines());
+    mips64::CodeGeneratorMIPS64 codegenMIPS64(graph, *features_mips64.get(), compiler_options);
+    RunCodeOptimized(&codegenMIPS64, graph, hook_before_codegen, has_result, expected);
   }
 }
 
diff --git a/compiler/optimizing/intrinsics_arm.cc b/compiler/optimizing/intrinsics_arm.cc
index 749bedf..71fadfb 100644
--- a/compiler/optimizing/intrinsics_arm.cc
+++ b/compiler/optimizing/intrinsics_arm.cc
@@ -121,6 +121,8 @@
     __ b(GetExitLabel());
   }
 
+  const char* GetDescription() const OVERRIDE { return "IntrinsicSlowPathARM"; }
+
  private:
   // The instruction where this slow path is happening.
   HInvoke* const invoke_;
diff --git a/compiler/optimizing/intrinsics_arm64.cc b/compiler/optimizing/intrinsics_arm64.cc
index c108ad5..8bcb88b 100644
--- a/compiler/optimizing/intrinsics_arm64.cc
+++ b/compiler/optimizing/intrinsics_arm64.cc
@@ -130,6 +130,8 @@
     __ B(GetExitLabel());
   }
 
+  const char* GetDescription() const OVERRIDE { return "IntrinsicSlowPathARM64"; }
+
  private:
   // The instruction where this slow path is happening.
   HInvoke* const invoke_;
diff --git a/compiler/optimizing/intrinsics_x86.cc b/compiler/optimizing/intrinsics_x86.cc
index 424ac7c..b04cc5c 100644
--- a/compiler/optimizing/intrinsics_x86.cc
+++ b/compiler/optimizing/intrinsics_x86.cc
@@ -158,6 +158,8 @@
     __ jmp(GetExitLabel());
   }
 
+  const char* GetDescription() const OVERRIDE { return "IntrinsicSlowPathX86"; }
+
  private:
   // The instruction where this slow path is happening.
   HInvoke* const invoke_;
diff --git a/compiler/optimizing/intrinsics_x86_64.cc b/compiler/optimizing/intrinsics_x86_64.cc
index 8915314..888c7b8 100644
--- a/compiler/optimizing/intrinsics_x86_64.cc
+++ b/compiler/optimizing/intrinsics_x86_64.cc
@@ -149,6 +149,8 @@
     __ jmp(GetExitLabel());
   }
 
+  const char* GetDescription() const OVERRIDE { return "IntrinsicSlowPathX86_64"; }
+
  private:
   // The instruction where this slow path is happening.
   HInvoke* const invoke_;
diff --git a/compiler/optimizing/nodes.h b/compiler/optimizing/nodes.h
index 9443653..78ef13e 100644
--- a/compiler/optimizing/nodes.h
+++ b/compiler/optimizing/nodes.h
@@ -2190,8 +2190,12 @@
     kLtBias,  // return -1 for NaN comparisons
   };
 
-  HCompare(Primitive::Type type, HInstruction* first, HInstruction* second, Bias bias)
-      : HBinaryOperation(Primitive::kPrimInt, first, second), bias_(bias) {
+  HCompare(Primitive::Type type,
+           HInstruction* first,
+           HInstruction* second,
+           Bias bias,
+           uint32_t dex_pc)
+      : HBinaryOperation(Primitive::kPrimInt, first, second), bias_(bias), dex_pc_(dex_pc) {
     DCHECK_EQ(type, first->GetType());
     DCHECK_EQ(type, second->GetType());
   }
@@ -2216,10 +2220,13 @@
 
   bool IsGtBias() { return bias_ == kGtBias; }
 
+  uint32_t GetDexPc() const { return dex_pc_; }
+
   DECLARE_INSTRUCTION(Compare);
 
  private:
   const Bias bias_;
+  const uint32_t dex_pc_;
 
   DISALLOW_COPY_AND_ASSIGN(HCompare);
 };
@@ -4027,6 +4034,8 @@
     return source_.IsInvalid();
   }
 
+  Primitive::Type GetType() const { return type_; }
+
   bool Is64BitMove() const {
     return Primitive::Is64BitType(type_);
   }
diff --git a/compiler/optimizing/optimizing_compiler.cc b/compiler/optimizing/optimizing_compiler.cc
index 8d43ada..ad67813 100644
--- a/compiler/optimizing/optimizing_compiler.cc
+++ b/compiler/optimizing/optimizing_compiler.cc
@@ -290,6 +290,7 @@
 static bool IsInstructionSetSupported(InstructionSet instruction_set) {
   return instruction_set == kArm64
       || (instruction_set == kThumb2 && !kArm32QuickCodeUseSoftFloat)
+      || instruction_set == kMips64
       || instruction_set == kX86
       || instruction_set == kX86_64;
 }
diff --git a/compiler/optimizing/register_allocator.cc b/compiler/optimizing/register_allocator.cc
index e38e49c..7b23d02 100644
--- a/compiler/optimizing/register_allocator.cc
+++ b/compiler/optimizing/register_allocator.cc
@@ -80,6 +80,7 @@
                                                 InstructionSet instruction_set) {
   return instruction_set == kArm64
       || instruction_set == kX86_64
+      || instruction_set == kMips64
       || instruction_set == kArm
       || instruction_set == kX86
       || instruction_set == kThumb2;
diff --git a/compiler/utils/mips64/assembler_mips64.cc b/compiler/utils/mips64/assembler_mips64.cc
index a8b55d1..a192d2f 100644
--- a/compiler/utils/mips64/assembler_mips64.cc
+++ b/compiler/utils/mips64/assembler_mips64.cc
@@ -25,9 +25,9 @@
 namespace art {
 namespace mips64 {
 
-void Mips64Assembler::Emit(int32_t value) {
+void Mips64Assembler::Emit(uint32_t value) {
   AssemblerBuffer::EnsureCapacity ensured(&buffer_);
-  buffer_.Emit<int32_t>(value);
+  buffer_.Emit<uint32_t>(value);
 }
 
 void Mips64Assembler::EmitR(int opcode, GpuRegister rs, GpuRegister rt, GpuRegister rd,
@@ -35,124 +35,62 @@
   CHECK_NE(rs, kNoGpuRegister);
   CHECK_NE(rt, kNoGpuRegister);
   CHECK_NE(rd, kNoGpuRegister);
-  int32_t encoding = opcode << kOpcodeShift |
-                     static_cast<int32_t>(rs) << kRsShift |
-                     static_cast<int32_t>(rt) << kRtShift |
-                     static_cast<int32_t>(rd) << kRdShift |
-                     shamt << kShamtShift |
-                     funct;
+  uint32_t encoding = static_cast<uint32_t>(opcode) << kOpcodeShift |
+                      static_cast<uint32_t>(rs) << kRsShift |
+                      static_cast<uint32_t>(rt) << kRtShift |
+                      static_cast<uint32_t>(rd) << kRdShift |
+                      shamt << kShamtShift |
+                      funct;
   Emit(encoding);
 }
 
 void Mips64Assembler::EmitI(int opcode, GpuRegister rs, GpuRegister rt, uint16_t imm) {
   CHECK_NE(rs, kNoGpuRegister);
   CHECK_NE(rt, kNoGpuRegister);
-  int32_t encoding = opcode << kOpcodeShift |
-                     static_cast<int32_t>(rs) << kRsShift |
-                     static_cast<int32_t>(rt) << kRtShift |
-                     imm;
+  uint32_t encoding = static_cast<uint32_t>(opcode) << kOpcodeShift |
+                      static_cast<uint32_t>(rs) << kRsShift |
+                      static_cast<uint32_t>(rt) << kRtShift |
+                      imm;
   Emit(encoding);
 }
 
-void Mips64Assembler::EmitJ(int opcode, int address) {
-  int32_t encoding = opcode << kOpcodeShift |
-                     address;
+void Mips64Assembler::EmitI21(int opcode, GpuRegister rs, uint32_t imm21) {
+  CHECK_NE(rs, kNoGpuRegister);
+  uint32_t encoding = static_cast<uint32_t>(opcode) << kOpcodeShift |
+                      static_cast<uint32_t>(rs) << kRsShift |
+                      (imm21 & 0x1FFFFF);
+  Emit(encoding);
+}
+
+void Mips64Assembler::EmitJ(int opcode, uint32_t addr26) {
+  uint32_t encoding = static_cast<uint32_t>(opcode) << kOpcodeShift |
+                      (addr26 & 0x3FFFFFF);
   Emit(encoding);
 }
 
 void Mips64Assembler::EmitFR(int opcode, int fmt, FpuRegister ft, FpuRegister fs, FpuRegister fd,
-int funct) {
+                             int funct) {
   CHECK_NE(ft, kNoFpuRegister);
   CHECK_NE(fs, kNoFpuRegister);
   CHECK_NE(fd, kNoFpuRegister);
-  int32_t encoding = opcode << kOpcodeShift |
-                     fmt << kFmtShift |
-                     static_cast<int32_t>(ft) << kFtShift |
-                     static_cast<int32_t>(fs) << kFsShift |
-                     static_cast<int32_t>(fd) << kFdShift |
-                     funct;
+  uint32_t encoding = static_cast<uint32_t>(opcode) << kOpcodeShift |
+                      fmt << kFmtShift |
+                      static_cast<uint32_t>(ft) << kFtShift |
+                      static_cast<uint32_t>(fs) << kFsShift |
+                      static_cast<uint32_t>(fd) << kFdShift |
+                      funct;
   Emit(encoding);
 }
 
-void Mips64Assembler::EmitFI(int opcode, int fmt, FpuRegister rt, uint16_t imm) {
-  CHECK_NE(rt, kNoFpuRegister);
-  int32_t encoding = opcode << kOpcodeShift |
-                     fmt << kFmtShift |
-                     static_cast<int32_t>(rt) << kRtShift |
-                     imm;
+void Mips64Assembler::EmitFI(int opcode, int fmt, FpuRegister ft, uint16_t imm) {
+  CHECK_NE(ft, kNoFpuRegister);
+  uint32_t encoding = static_cast<uint32_t>(opcode) << kOpcodeShift |
+                      fmt << kFmtShift |
+                      static_cast<uint32_t>(ft) << kFtShift |
+                      imm;
   Emit(encoding);
 }
 
-void Mips64Assembler::EmitBranch(GpuRegister rt, GpuRegister rs, Label* label, bool equal) {
-  int offset;
-  if (label->IsBound()) {
-    offset = label->Position() - buffer_.Size();
-  } else {
-    // Use the offset field of the branch instruction for linking the sites.
-    offset = label->position_;
-    label->LinkTo(buffer_.Size());
-  }
-  if (equal) {
-    Beq(rt, rs, (offset >> 2) & kBranchOffsetMask);
-  } else {
-    Bne(rt, rs, (offset >> 2) & kBranchOffsetMask);
-  }
-}
-
-void Mips64Assembler::EmitJump(Label* label, bool link) {
-  int offset;
-  if (label->IsBound()) {
-    offset = label->Position() - buffer_.Size();
-  } else {
-    // Use the offset field of the jump instruction for linking the sites.
-    offset = label->position_;
-    label->LinkTo(buffer_.Size());
-  }
-  if (link) {
-    Jal((offset >> 2) & kJumpOffsetMask);
-  } else {
-    J((offset >> 2) & kJumpOffsetMask);
-  }
-}
-
-int32_t Mips64Assembler::EncodeBranchOffset(int offset, int32_t inst, bool is_jump) {
-  CHECK_ALIGNED(offset, 4);
-  CHECK(IsInt<POPCOUNT(kBranchOffsetMask)>(offset)) << offset;
-
-  // Properly preserve only the bits supported in the instruction.
-  offset >>= 2;
-  if (is_jump) {
-    offset &= kJumpOffsetMask;
-    return (inst & ~kJumpOffsetMask) | offset;
-  } else {
-    offset &= kBranchOffsetMask;
-    return (inst & ~kBranchOffsetMask) | offset;
-  }
-}
-
-int Mips64Assembler::DecodeBranchOffset(int32_t inst, bool is_jump) {
-  // Sign-extend, then left-shift by 2.
-  if (is_jump) {
-    return (((inst & kJumpOffsetMask) << 6) >> 4);
-  } else {
-    return (((inst & kBranchOffsetMask) << 16) >> 14);
-  }
-}
-
-void Mips64Assembler::Bind(Label* label, bool is_jump) {
-  CHECK(!label->IsBound());
-  int bound_pc = buffer_.Size();
-  while (label->IsLinked()) {
-    int32_t position = label->Position();
-    int32_t next = buffer_.Load<int32_t>(position);
-    int32_t offset = is_jump ? bound_pc - position : bound_pc - position - 4;
-    int32_t encoded = Mips64Assembler::EncodeBranchOffset(offset, next, is_jump);
-    buffer_.Store<int32_t>(position, encoded);
-    label->position_ = Mips64Assembler::DecodeBranchOffset(next, is_jump);
-  }
-  label->BindTo(bound_pc);
-}
-
 void Mips64Assembler::Add(GpuRegister rd, GpuRegister rs, GpuRegister rt) {
   EmitR(0, rs, rt, rd, 0, 0x20);
 }
@@ -169,6 +107,10 @@
   EmitI(0x9, rs, rt, imm16);
 }
 
+void Mips64Assembler::Daddu(GpuRegister rd, GpuRegister rs, GpuRegister rt) {
+  EmitR(0, rs, rt, rd, 0, 0x2d);
+}
+
 void Mips64Assembler::Daddiu(GpuRegister rt, GpuRegister rs, uint16_t imm16) {
   EmitI(0x19, rs, rt, imm16);
 }
@@ -181,22 +123,90 @@
   EmitR(0, rs, rt, rd, 0, 0x23);
 }
 
-void Mips64Assembler::Mult(GpuRegister rs, GpuRegister rt) {
+void Mips64Assembler::Dsubu(GpuRegister rd, GpuRegister rs, GpuRegister rt) {
+  EmitR(0, rs, rt, rd, 0, 0x2f);
+}
+
+void Mips64Assembler::MultR2(GpuRegister rs, GpuRegister rt) {
   EmitR(0, rs, rt, static_cast<GpuRegister>(0), 0, 0x18);
 }
 
-void Mips64Assembler::Multu(GpuRegister rs, GpuRegister rt) {
+void Mips64Assembler::MultuR2(GpuRegister rs, GpuRegister rt) {
   EmitR(0, rs, rt, static_cast<GpuRegister>(0), 0, 0x19);
 }
 
-void Mips64Assembler::Div(GpuRegister rs, GpuRegister rt) {
+void Mips64Assembler::DivR2(GpuRegister rs, GpuRegister rt) {
   EmitR(0, rs, rt, static_cast<GpuRegister>(0), 0, 0x1a);
 }
 
-void Mips64Assembler::Divu(GpuRegister rs, GpuRegister rt) {
+void Mips64Assembler::DivuR2(GpuRegister rs, GpuRegister rt) {
   EmitR(0, rs, rt, static_cast<GpuRegister>(0), 0, 0x1b);
 }
 
+void Mips64Assembler::MulR2(GpuRegister rd, GpuRegister rs, GpuRegister rt) {
+  EmitR(0x1c, rs, rt, rd, 0, 2);
+}
+
+void Mips64Assembler::DivR2(GpuRegister rd, GpuRegister rs, GpuRegister rt) {
+  DivR2(rs, rt);
+  Mflo(rd);
+}
+
+void Mips64Assembler::ModR2(GpuRegister rd, GpuRegister rs, GpuRegister rt) {
+  DivR2(rs, rt);
+  Mfhi(rd);
+}
+
+void Mips64Assembler::DivuR2(GpuRegister rd, GpuRegister rs, GpuRegister rt) {
+  DivuR2(rs, rt);
+  Mflo(rd);
+}
+
+void Mips64Assembler::ModuR2(GpuRegister rd, GpuRegister rs, GpuRegister rt) {
+  DivuR2(rs, rt);
+  Mfhi(rd);
+}
+
+void Mips64Assembler::MulR6(GpuRegister rd, GpuRegister rs, GpuRegister rt) {
+  EmitR(0, rs, rt, rd, 2, 0x18);
+}
+
+void Mips64Assembler::DivR6(GpuRegister rd, GpuRegister rs, GpuRegister rt) {
+  EmitR(0, rs, rt, rd, 2, 0x1a);
+}
+
+void Mips64Assembler::ModR6(GpuRegister rd, GpuRegister rs, GpuRegister rt) {
+  EmitR(0, rs, rt, rd, 3, 0x1a);
+}
+
+void Mips64Assembler::DivuR6(GpuRegister rd, GpuRegister rs, GpuRegister rt) {
+  EmitR(0, rs, rt, rd, 2, 0x1b);
+}
+
+void Mips64Assembler::ModuR6(GpuRegister rd, GpuRegister rs, GpuRegister rt) {
+  EmitR(0, rs, rt, rd, 3, 0x1b);
+}
+
+void Mips64Assembler::Dmul(GpuRegister rd, GpuRegister rs, GpuRegister rt) {
+  EmitR(0, rs, rt, rd, 2, 0x1c);
+}
+
+void Mips64Assembler::Ddiv(GpuRegister rd, GpuRegister rs, GpuRegister rt) {
+  EmitR(0, rs, rt, rd, 2, 0x1e);
+}
+
+void Mips64Assembler::Dmod(GpuRegister rd, GpuRegister rs, GpuRegister rt) {
+  EmitR(0, rs, rt, rd, 3, 0x1e);
+}
+
+void Mips64Assembler::Ddivu(GpuRegister rd, GpuRegister rs, GpuRegister rt) {
+  EmitR(0, rs, rt, rd, 2, 0x1f);
+}
+
+void Mips64Assembler::Dmodu(GpuRegister rd, GpuRegister rs, GpuRegister rt) {
+  EmitR(0, rs, rt, rd, 3, 0x1f);
+}
+
 void Mips64Assembler::And(GpuRegister rd, GpuRegister rs, GpuRegister rt) {
   EmitR(0, rs, rt, rd, 0, 0x24);
 }
@@ -225,30 +235,80 @@
   EmitR(0, rs, rt, rd, 0, 0x27);
 }
 
-void Mips64Assembler::Sll(GpuRegister rd, GpuRegister rs, int shamt) {
-  EmitR(0, rs, static_cast<GpuRegister>(0), rd, shamt, 0x00);
+void Mips64Assembler::Seb(GpuRegister rd, GpuRegister rt) {
+  EmitR(0x1f, static_cast<GpuRegister>(0), rt, rd, 0x10, 0x20);
 }
 
-void Mips64Assembler::Srl(GpuRegister rd, GpuRegister rs, int shamt) {
-  EmitR(0, rs, static_cast<GpuRegister>(0), rd, shamt, 0x02);
+void Mips64Assembler::Seh(GpuRegister rd, GpuRegister rt) {
+  EmitR(0x1f, static_cast<GpuRegister>(0), rt, rd, 0x18, 0x20);
 }
 
-void Mips64Assembler::Sra(GpuRegister rd, GpuRegister rs, int shamt) {
-  EmitR(0, rs, static_cast<GpuRegister>(0), rd, shamt, 0x03);
+void Mips64Assembler::Dext(GpuRegister rt, GpuRegister rs, int pos, int size_less_one) {
+  DCHECK(0 <= pos && pos < 32) << pos;
+  DCHECK(0 <= size_less_one && size_less_one < 32) << size_less_one;
+  EmitR(0x1f, rs, rt, static_cast<GpuRegister>(size_less_one), pos, 3);
 }
 
-void Mips64Assembler::Sllv(GpuRegister rd, GpuRegister rs, GpuRegister rt) {
+void Mips64Assembler::Sll(GpuRegister rd, GpuRegister rt, int shamt) {
+  EmitR(0, static_cast<GpuRegister>(0), rt, rd, shamt, 0x00);
+}
+
+void Mips64Assembler::Srl(GpuRegister rd, GpuRegister rt, int shamt) {
+  EmitR(0, static_cast<GpuRegister>(0), rt, rd, shamt, 0x02);
+}
+
+void Mips64Assembler::Sra(GpuRegister rd, GpuRegister rt, int shamt) {
+  EmitR(0, static_cast<GpuRegister>(0), rt, rd, shamt, 0x03);
+}
+
+void Mips64Assembler::Sllv(GpuRegister rd, GpuRegister rt, GpuRegister rs) {
   EmitR(0, rs, rt, rd, 0, 0x04);
 }
 
-void Mips64Assembler::Srlv(GpuRegister rd, GpuRegister rs, GpuRegister rt) {
+void Mips64Assembler::Srlv(GpuRegister rd, GpuRegister rt, GpuRegister rs) {
   EmitR(0, rs, rt, rd, 0, 0x06);
 }
 
-void Mips64Assembler::Srav(GpuRegister rd, GpuRegister rs, GpuRegister rt) {
+void Mips64Assembler::Srav(GpuRegister rd, GpuRegister rt, GpuRegister rs) {
   EmitR(0, rs, rt, rd, 0, 0x07);
 }
 
+void Mips64Assembler::Dsll(GpuRegister rd, GpuRegister rt, int shamt) {
+  EmitR(0, static_cast<GpuRegister>(0), rt, rd, shamt, 0x38);
+}
+
+void Mips64Assembler::Dsrl(GpuRegister rd, GpuRegister rt, int shamt) {
+  EmitR(0, static_cast<GpuRegister>(0), rt, rd, shamt, 0x3a);
+}
+
+void Mips64Assembler::Dsra(GpuRegister rd, GpuRegister rt, int shamt) {
+  EmitR(0, static_cast<GpuRegister>(0), rt, rd, shamt, 0x3b);
+}
+
+void Mips64Assembler::Dsll32(GpuRegister rd, GpuRegister rt, int shamt) {
+  EmitR(0, static_cast<GpuRegister>(0), rt, rd, shamt, 0x3c);
+}
+
+void Mips64Assembler::Dsrl32(GpuRegister rd, GpuRegister rt, int shamt) {
+  EmitR(0, static_cast<GpuRegister>(0), rt, rd, shamt, 0x3e);
+}
+
+void Mips64Assembler::Dsra32(GpuRegister rd, GpuRegister rt, int shamt) {
+  EmitR(0, static_cast<GpuRegister>(0), rt, rd, shamt, 0x3f);
+}
+
+void Mips64Assembler::Dsllv(GpuRegister rd, GpuRegister rt, GpuRegister rs) {
+  EmitR(0, rs, rt, rd, 0, 0x14);
+}
+
+void Mips64Assembler::Dsrlv(GpuRegister rd, GpuRegister rt, GpuRegister rs) {
+  EmitR(0, rs, rt, rd, 0, 0x16);
+}
+
+void Mips64Assembler::Dsrav(GpuRegister rd, GpuRegister rt, GpuRegister rs) {
+  EmitR(0, rs, rt, rd, 0, 0x17);
+}
+
 void Mips64Assembler::Lb(GpuRegister rt, GpuRegister rs, uint16_t imm16) {
   EmitI(0x20, rs, rt, imm16);
 }
@@ -281,6 +341,19 @@
   EmitI(0xf, static_cast<GpuRegister>(0), rt, imm16);
 }
 
+void Mips64Assembler::Dahi(GpuRegister rs, uint16_t imm16) {
+  EmitI(1, rs, static_cast<GpuRegister>(6), imm16);
+}
+
+void Mips64Assembler::Dati(GpuRegister rs, uint16_t imm16) {
+  EmitI(1, rs, static_cast<GpuRegister>(0x1e), imm16);
+}
+
+void Mips64Assembler::Sync(uint32_t stype) {
+  EmitR(0, static_cast<GpuRegister>(0), static_cast<GpuRegister>(0),
+           static_cast<GpuRegister>(0), stype & 0x1f, 0xf);
+}
+
 void Mips64Assembler::Mfhi(GpuRegister rd) {
   EmitR(0, static_cast<GpuRegister>(0), static_cast<GpuRegister>(0), rd, 0, 0x10);
 }
@@ -321,34 +394,121 @@
   EmitI(0xb, rs, rt, imm16);
 }
 
-void Mips64Assembler::Beq(GpuRegister rt, GpuRegister rs, uint16_t imm16) {
+void Mips64Assembler::Beq(GpuRegister rs, GpuRegister rt, uint16_t imm16) {
   EmitI(0x4, rs, rt, imm16);
   Nop();
 }
 
-void Mips64Assembler::Bne(GpuRegister rt, GpuRegister rs, uint16_t imm16) {
+void Mips64Assembler::Bne(GpuRegister rs, GpuRegister rt, uint16_t imm16) {
   EmitI(0x5, rs, rt, imm16);
   Nop();
 }
 
-void Mips64Assembler::J(uint32_t address) {
-  EmitJ(0x2, address);
+void Mips64Assembler::J(uint32_t addr26) {
+  EmitJ(0x2, addr26);
   Nop();
 }
 
-void Mips64Assembler::Jal(uint32_t address) {
-  EmitJ(0x2, address);
+void Mips64Assembler::Jal(uint32_t addr26) {
+  EmitJ(0x3, addr26);
   Nop();
 }
 
-void Mips64Assembler::Jr(GpuRegister rs) {
-  EmitR(0, rs, static_cast<GpuRegister>(0), static_cast<GpuRegister>(0), 0, 0x09);  // Jalr zero, rs
+void Mips64Assembler::Jalr(GpuRegister rd, GpuRegister rs) {
+  EmitR(0, rs, static_cast<GpuRegister>(0), rd, 0, 0x09);
   Nop();
 }
 
 void Mips64Assembler::Jalr(GpuRegister rs) {
-  EmitR(0, rs, static_cast<GpuRegister>(0), RA, 0, 0x09);
-  Nop();
+  Jalr(RA, rs);
+}
+
+void Mips64Assembler::Jr(GpuRegister rs) {
+  Jalr(ZERO, rs);
+}
+
+void Mips64Assembler::Auipc(GpuRegister rs, uint16_t imm16) {
+  EmitI(0x3B, rs, static_cast<GpuRegister>(0x1E), imm16);
+}
+
+void Mips64Assembler::Jic(GpuRegister rt, uint16_t imm16) {
+  EmitI(0x36, static_cast<GpuRegister>(0), rt, imm16);
+}
+
+void Mips64Assembler::Jialc(GpuRegister rt, uint16_t imm16) {
+  EmitI(0x3E, static_cast<GpuRegister>(0), rt, imm16);
+}
+
+void Mips64Assembler::Bltc(GpuRegister rs, GpuRegister rt, uint16_t imm16) {
+  CHECK_NE(rs, ZERO);
+  CHECK_NE(rt, ZERO);
+  CHECK_NE(rs, rt);
+  EmitI(0x17, rs, rt, imm16);
+}
+
+void Mips64Assembler::Bltzc(GpuRegister rt, uint16_t imm16) {
+  CHECK_NE(rt, ZERO);
+  EmitI(0x17, rt, rt, imm16);
+}
+
+void Mips64Assembler::Bgtzc(GpuRegister rt, uint16_t imm16) {
+  CHECK_NE(rt, ZERO);
+  EmitI(0x17, static_cast<GpuRegister>(0), rt, imm16);
+}
+
+void Mips64Assembler::Bgec(GpuRegister rs, GpuRegister rt, uint16_t imm16) {
+  CHECK_NE(rs, ZERO);
+  CHECK_NE(rt, ZERO);
+  CHECK_NE(rs, rt);
+  EmitI(0x16, rs, rt, imm16);
+}
+
+void Mips64Assembler::Bgezc(GpuRegister rt, uint16_t imm16) {
+  CHECK_NE(rt, ZERO);
+  EmitI(0x16, rt, rt, imm16);
+}
+
+void Mips64Assembler::Blezc(GpuRegister rt, uint16_t imm16) {
+  CHECK_NE(rt, ZERO);
+  EmitI(0x16, static_cast<GpuRegister>(0), rt, imm16);
+}
+
+void Mips64Assembler::Bltuc(GpuRegister rs, GpuRegister rt, uint16_t imm16) {
+  CHECK_NE(rs, ZERO);
+  CHECK_NE(rt, ZERO);
+  CHECK_NE(rs, rt);
+  EmitI(0x7, rs, rt, imm16);
+}
+
+void Mips64Assembler::Bgeuc(GpuRegister rs, GpuRegister rt, uint16_t imm16) {
+  CHECK_NE(rs, ZERO);
+  CHECK_NE(rt, ZERO);
+  CHECK_NE(rs, rt);
+  EmitI(0x6, rs, rt, imm16);
+}
+
+void Mips64Assembler::Beqc(GpuRegister rs, GpuRegister rt, uint16_t imm16) {
+  CHECK_NE(rs, ZERO);
+  CHECK_NE(rt, ZERO);
+  CHECK_NE(rs, rt);
+  EmitI(0x8, (rs < rt) ? rs : rt, (rs < rt) ? rt : rs, imm16);
+}
+
+void Mips64Assembler::Bnec(GpuRegister rs, GpuRegister rt, uint16_t imm16) {
+  CHECK_NE(rs, ZERO);
+  CHECK_NE(rt, ZERO);
+  CHECK_NE(rs, rt);
+  EmitI(0x18, (rs < rt) ? rs : rt, (rs < rt) ? rt : rs, imm16);
+}
+
+void Mips64Assembler::Beqzc(GpuRegister rs, uint32_t imm21) {
+  CHECK_NE(rs, ZERO);
+  EmitI21(0x36, rs, imm21);
+}
+
+void Mips64Assembler::Bnezc(GpuRegister rs, uint32_t imm21) {
+  CHECK_NE(rs, ZERO);
+  EmitI21(0x3E, rs, imm21);
 }
 
 void Mips64Assembler::AddS(FpuRegister fd, FpuRegister fs, FpuRegister ft) {
@@ -368,23 +528,19 @@
 }
 
 void Mips64Assembler::AddD(FpuRegister fd, FpuRegister fs, FpuRegister ft) {
-  EmitFR(0x11, 0x11, static_cast<FpuRegister>(ft), static_cast<FpuRegister>(fs),
-         static_cast<FpuRegister>(fd), 0x0);
+  EmitFR(0x11, 0x11, ft, fs, fd, 0x0);
 }
 
 void Mips64Assembler::SubD(FpuRegister fd, FpuRegister fs, FpuRegister ft) {
-  EmitFR(0x11, 0x11, static_cast<FpuRegister>(ft), static_cast<FpuRegister>(fs),
-         static_cast<FpuRegister>(fd), 0x1);
+  EmitFR(0x11, 0x11, ft, fs, fd, 0x1);
 }
 
 void Mips64Assembler::MulD(FpuRegister fd, FpuRegister fs, FpuRegister ft) {
-  EmitFR(0x11, 0x11, static_cast<FpuRegister>(ft), static_cast<FpuRegister>(fs),
-         static_cast<FpuRegister>(fd), 0x2);
+  EmitFR(0x11, 0x11, ft, fs, fd, 0x2);
 }
 
 void Mips64Assembler::DivD(FpuRegister fd, FpuRegister fs, FpuRegister ft) {
-  EmitFR(0x11, 0x11, static_cast<FpuRegister>(ft), static_cast<FpuRegister>(fs),
-         static_cast<FpuRegister>(fd), 0x3);
+  EmitFR(0x11, 0x11, ft, fs, fd, 0x3);
 }
 
 void Mips64Assembler::MovS(FpuRegister fd, FpuRegister fs) {
@@ -392,16 +548,47 @@
 }
 
 void Mips64Assembler::MovD(FpuRegister fd, FpuRegister fs) {
-  EmitFR(0x11, 0x11, static_cast<FpuRegister>(0), static_cast<FpuRegister>(fs),
-         static_cast<FpuRegister>(fd), 0x6);
+  EmitFR(0x11, 0x11, static_cast<FpuRegister>(0), fs, fd, 0x6);
+}
+
+void Mips64Assembler::NegS(FpuRegister fd, FpuRegister fs) {
+  EmitFR(0x11, 0x10, static_cast<FpuRegister>(0), fs, fd, 0x7);
+}
+
+void Mips64Assembler::NegD(FpuRegister fd, FpuRegister fs) {
+  EmitFR(0x11, 0x11, static_cast<FpuRegister>(0), fs, fd, 0x7);
+}
+
+void Mips64Assembler::Cvtsw(FpuRegister fd, FpuRegister fs) {
+  EmitFR(0x11, 0x14, static_cast<FpuRegister>(0), fs, fd, 0x20);
+}
+
+void Mips64Assembler::Cvtdw(FpuRegister fd, FpuRegister fs) {
+  EmitFR(0x11, 0x14, static_cast<FpuRegister>(0), fs, fd, 0x21);
+}
+
+void Mips64Assembler::Cvtsd(FpuRegister fd, FpuRegister fs) {
+  EmitFR(0x11, 0x11, static_cast<FpuRegister>(0), fs, fd, 0x20);
+}
+
+void Mips64Assembler::Cvtds(FpuRegister fd, FpuRegister fs) {
+  EmitFR(0x11, 0x10, static_cast<FpuRegister>(0), fs, fd, 0x21);
 }
 
 void Mips64Assembler::Mfc1(GpuRegister rt, FpuRegister fs) {
   EmitFR(0x11, 0x00, static_cast<FpuRegister>(rt), fs, static_cast<FpuRegister>(0), 0x0);
 }
 
-void Mips64Assembler::Mtc1(FpuRegister ft, GpuRegister rs) {
-  EmitFR(0x11, 0x04, ft, static_cast<FpuRegister>(rs), static_cast<FpuRegister>(0), 0x0);
+void Mips64Assembler::Mtc1(GpuRegister rt, FpuRegister fs) {
+  EmitFR(0x11, 0x04, static_cast<FpuRegister>(rt), fs, static_cast<FpuRegister>(0), 0x0);
+}
+
+void Mips64Assembler::Dmfc1(GpuRegister rt, FpuRegister fs) {
+  EmitFR(0x11, 0x01, static_cast<FpuRegister>(rt), fs, static_cast<FpuRegister>(0), 0x0);
+}
+
+void Mips64Assembler::Dmtc1(GpuRegister rt, FpuRegister fs) {
+  EmitFR(0x11, 0x05, static_cast<FpuRegister>(rt), fs, static_cast<FpuRegister>(0), 0x0);
 }
 
 void Mips64Assembler::Lwc1(FpuRegister ft, GpuRegister rs, uint16_t imm16) {
@@ -430,45 +617,330 @@
         static_cast<GpuRegister>(0), 0, 0x0);
 }
 
-void Mips64Assembler::Move(GpuRegister rt, GpuRegister rs) {
-  EmitI(0x19, rs, rt, 0);   // Daddiu
+void Mips64Assembler::Move(GpuRegister rd, GpuRegister rs) {
+  Or(rd, rs, ZERO);
 }
 
-void Mips64Assembler::Clear(GpuRegister rt) {
-  EmitR(0, static_cast<GpuRegister>(0), static_cast<GpuRegister>(0), rt, 0, 0x20);
+void Mips64Assembler::Clear(GpuRegister rd) {
+  Move(rd, ZERO);
 }
 
-void Mips64Assembler::Not(GpuRegister rt, GpuRegister rs) {
-  EmitR(0, static_cast<GpuRegister>(0), rs, rt, 0, 0x27);
+void Mips64Assembler::Not(GpuRegister rd, GpuRegister rs) {
+  Nor(rd, rs, ZERO);
 }
 
-void Mips64Assembler::Mul(GpuRegister rd, GpuRegister rs, GpuRegister rt) {
-  Mult(rs, rt);
-  Mflo(rd);
+void Mips64Assembler::LoadConst32(GpuRegister rd, int32_t value) {
+  if (IsUint<16>(value)) {
+    // Use OR with (unsigned) immediate to encode 16b unsigned int.
+    Ori(rd, ZERO, value);
+  } else if (IsInt<16>(value)) {
+    // Use ADD with (signed) immediate to encode 16b signed int.
+    Addiu(rd, ZERO, value);
+  } else {
+    Lui(rd, value >> 16);
+    if (value & 0xFFFF)
+      Ori(rd, rd, value);
+  }
 }
 
-void Mips64Assembler::Div(GpuRegister rd, GpuRegister rs, GpuRegister rt) {
-  Div(rs, rt);
-  Mflo(rd);
+void Mips64Assembler::LoadConst64(GpuRegister rd, int64_t value) {
+  int bit31 = (value & UINT64_C(0x80000000)) != 0;
+
+  // Loads with 1 instruction.
+  if (IsUint<16>(value)) {
+    Ori(rd, ZERO, value);
+  } else if (IsInt<16>(value)) {
+    Daddiu(rd, ZERO, value);
+  } else if ((value & 0xFFFF) == 0 && IsInt<16>(value >> 16)) {
+    Lui(rd, value >> 16);
+  } else if (IsInt<32>(value)) {
+    // Loads with 2 instructions.
+    Lui(rd, value >> 16);
+    Ori(rd, rd, value);
+  } else if ((value & 0xFFFF0000) == 0 && IsInt<16>(value >> 32)) {
+    Ori(rd, ZERO, value);
+    Dahi(rd, value >> 32);
+  } else if ((value & UINT64_C(0xFFFFFFFF0000)) == 0) {
+    Ori(rd, ZERO, value);
+    Dati(rd, value >> 48);
+  } else if ((value & 0xFFFF) == 0 &&
+             (-32768 - bit31) <= (value >> 32) && (value >> 32) <= (32767 - bit31)) {
+    Lui(rd, value >> 16);
+    Dahi(rd, (value >> 32) + bit31);
+  } else if ((value & 0xFFFF) == 0 && ((value >> 31) & 0x1FFFF) == ((0x20000 - bit31) & 0x1FFFF)) {
+    Lui(rd, value >> 16);
+    Dati(rd, (value >> 48) + bit31);
+  } else {
+    int shift_cnt = CTZ(value);
+    int64_t tmp = value >> shift_cnt;
+    if (IsUint<16>(tmp)) {
+      Ori(rd, ZERO, tmp);
+      if (shift_cnt < 32)
+        Dsll(rd, rd, shift_cnt);
+      else
+        Dsll32(rd, rd, shift_cnt & 31);
+    } else if (IsInt<16>(tmp)) {
+      Daddiu(rd, ZERO, tmp);
+      if (shift_cnt < 32)
+        Dsll(rd, rd, shift_cnt);
+      else
+        Dsll32(rd, rd, shift_cnt & 31);
+    } else if (IsInt<32>(tmp)) {
+      // Loads with 3 instructions.
+      Lui(rd, tmp >> 16);
+      Ori(rd, rd, tmp);
+      if (shift_cnt < 32)
+        Dsll(rd, rd, shift_cnt);
+      else
+        Dsll32(rd, rd, shift_cnt & 31);
+    } else {
+      shift_cnt = 16 + CTZ(value >> 16);
+      tmp = value >> shift_cnt;
+      if (IsUint<16>(tmp)) {
+        Ori(rd, ZERO, tmp);
+        if (shift_cnt < 32)
+          Dsll(rd, rd, shift_cnt);
+        else
+          Dsll32(rd, rd, shift_cnt & 31);
+        Ori(rd, rd, value);
+      } else if (IsInt<16>(tmp)) {
+        Daddiu(rd, ZERO, tmp);
+        if (shift_cnt < 32)
+          Dsll(rd, rd, shift_cnt);
+        else
+          Dsll32(rd, rd, shift_cnt & 31);
+        Ori(rd, rd, value);
+      } else {
+        // Loads with 3-4 instructions.
+        uint64_t tmp2 = value;
+        bool used_lui = false;
+        if (((tmp2 >> 16) & 0xFFFF) != 0 || (tmp2 & 0xFFFFFFFF) == 0) {
+          Lui(rd, tmp2 >> 16);
+          used_lui = true;
+        }
+        if ((tmp2 & 0xFFFF) != 0) {
+          if (used_lui)
+            Ori(rd, rd, tmp2);
+          else
+            Ori(rd, ZERO, tmp2);
+        }
+        if (bit31) {
+          tmp2 += UINT64_C(0x100000000);
+        }
+        if (((tmp2 >> 32) & 0xFFFF) != 0) {
+          Dahi(rd, tmp2 >> 32);
+        }
+        if (tmp2 & UINT64_C(0x800000000000)) {
+          tmp2 += UINT64_C(0x1000000000000);
+        }
+        if ((tmp2 >> 48) != 0) {
+          Dati(rd, tmp2 >> 48);
+        }
+      }
+    }
+  }
 }
 
-void Mips64Assembler::Rem(GpuRegister rd, GpuRegister rs, GpuRegister rt) {
-  Div(rs, rt);
-  Mfhi(rd);
+void Mips64Assembler::Addiu32(GpuRegister rt, GpuRegister rs, int32_t value, GpuRegister rtmp) {
+  if (IsInt<16>(value)) {
+    Addiu(rt, rs, value);
+  } else {
+    LoadConst32(rtmp, value);
+    Addu(rt, rs, rtmp);
+  }
 }
 
-void Mips64Assembler::AddConstant64(GpuRegister rt, GpuRegister rs, int32_t value) {
-  CHECK((value >= -32768) && (value <= 32766));
-  Daddiu(rt, rs, value);
+void Mips64Assembler::Daddiu64(GpuRegister rt, GpuRegister rs, int64_t value, GpuRegister rtmp) {
+  if (IsInt<16>(value)) {
+    Daddiu(rt, rs, value);
+  } else {
+    LoadConst64(rtmp, value);
+    Daddu(rt, rs, rtmp);
+  }
 }
 
-void Mips64Assembler::LoadImmediate64(GpuRegister rt, int32_t value) {
-  CHECK((value >= -32768) && (value <= 32766));
-  Daddiu(rt, ZERO, value);
+//
+// MIPS64R6 branches
+//
+//
+// Unconditional (pc + 32-bit signed offset):
+//
+//   auipc    at, ofs_high
+//   jic      at, ofs_low
+//   // no delay/forbidden slot
+//
+//
+// Conditional (pc + 32-bit signed offset):
+//
+//   b<cond>c   reg, +2      // skip next 2 instructions
+//   auipc      at, ofs_high
+//   jic        at, ofs_low
+//   // no delay/forbidden slot
+//
+//
+// Unconditional (pc + 32-bit signed offset) and link:
+//
+//   auipc    reg, ofs_high
+//   daddiu   reg, ofs_low
+//   jialc    reg, 0
+//   // no delay/forbidden slot
+//
+//
+// TODO: use shorter instruction sequences whenever possible.
+//
+
+void Mips64Assembler::Bind(Label* label) {
+  CHECK(!label->IsBound());
+  int32_t bound_pc = buffer_.Size();
+
+  // Walk the list of the branches (auipc + jic pairs) referring to and preceding this label.
+  // Embed the previously unknown pc-relative addresses in them.
+  while (label->IsLinked()) {
+    int32_t position = label->Position();
+    // Extract the branch (instruction pair)
+    uint32_t auipc = buffer_.Load<uint32_t>(position);
+    uint32_t jic = buffer_.Load<uint32_t>(position + 4);  // actually, jic or daddiu
+
+    // Extract the location of the previous pair in the list (walking the list backwards;
+    // the previous pair location was stored in the immediate operands of the instructions)
+    int32_t prev = (auipc << 16) | (jic & 0xFFFF);
+
+    // Get the pc-relative address
+    uint32_t offset = bound_pc - position;
+    offset += (offset & 0x8000) << 1;  // account for sign extension in jic/daddiu
+
+    // Embed it in the two instructions
+    auipc = (auipc & 0xFFFF0000) | (offset >> 16);
+    jic = (jic & 0xFFFF0000) | (offset & 0xFFFF);
+
+    // Save the adjusted instructions
+    buffer_.Store<uint32_t>(position, auipc);
+    buffer_.Store<uint32_t>(position + 4, jic);
+
+    // On to the previous branch in the list...
+    label->position_ = prev;
+  }
+
+  // Now make the label object contain its own location
+  // (it will be used by the branches referring to and following this label)
+  label->BindTo(bound_pc);
+}
+
+void Mips64Assembler::B(Label* label) {
+  if (label->IsBound()) {
+    // Branch backwards (to a preceding label), distance is known
+    uint32_t offset = label->Position() - buffer_.Size();
+    CHECK_LE(static_cast<int32_t>(offset), 0);
+    offset += (offset & 0x8000) << 1;  // account for sign extension in jic
+    Auipc(AT, offset >> 16);
+    Jic(AT, offset);
+  } else {
+    // Branch forward (to a following label), distance is unknown
+    int32_t position = buffer_.Size();
+    // The first branch forward will have 0 in its pc-relative address (copied from label's
+    // position). It will be the terminator of the list of forward-reaching branches.
+    uint32_t prev = label->position_;
+    Auipc(AT, prev >> 16);
+    Jic(AT, prev);
+    // Now make the link object point to the location of this branch
+    // (this forms a linked list of branches preceding this label)
+    label->LinkTo(position);
+  }
+}
+
+void Mips64Assembler::Jalr(Label* label, GpuRegister indirect_reg) {
+  if (label->IsBound()) {
+    // Branch backwards (to a preceding label), distance is known
+    uint32_t offset = label->Position() - buffer_.Size();
+    CHECK_LE(static_cast<int32_t>(offset), 0);
+    offset += (offset & 0x8000) << 1;  // account for sign extension in daddiu
+    Auipc(indirect_reg, offset >> 16);
+    Daddiu(indirect_reg, indirect_reg, offset);
+    Jialc(indirect_reg, 0);
+  } else {
+    // Branch forward (to a following label), distance is unknown
+    int32_t position = buffer_.Size();
+    // The first branch forward will have 0 in its pc-relative address (copied from label's
+    // position). It will be the terminator of the list of forward-reaching branches.
+    uint32_t prev = label->position_;
+    Auipc(indirect_reg, prev >> 16);
+    Daddiu(indirect_reg, indirect_reg, prev);
+    Jialc(indirect_reg, 0);
+    // Now make the link object point to the location of this branch
+    // (this forms a linked list of branches preceding this label)
+    label->LinkTo(position);
+  }
+}
+
+void Mips64Assembler::Bltc(GpuRegister rs, GpuRegister rt, Label* label) {
+  Bgec(rs, rt, 2);
+  B(label);
+}
+
+void Mips64Assembler::Bltzc(GpuRegister rt, Label* label) {
+  Bgezc(rt, 2);
+  B(label);
+}
+
+void Mips64Assembler::Bgtzc(GpuRegister rt, Label* label) {
+  Blezc(rt, 2);
+  B(label);
+}
+
+void Mips64Assembler::Bgec(GpuRegister rs, GpuRegister rt, Label* label) {
+  Bltc(rs, rt, 2);
+  B(label);
+}
+
+void Mips64Assembler::Bgezc(GpuRegister rt, Label* label) {
+  Bltzc(rt, 2);
+  B(label);
+}
+
+void Mips64Assembler::Blezc(GpuRegister rt, Label* label) {
+  Bgtzc(rt, 2);
+  B(label);
+}
+
+void Mips64Assembler::Bltuc(GpuRegister rs, GpuRegister rt, Label* label) {
+  Bgeuc(rs, rt, 2);
+  B(label);
+}
+
+void Mips64Assembler::Bgeuc(GpuRegister rs, GpuRegister rt, Label* label) {
+  Bltuc(rs, rt, 2);
+  B(label);
+}
+
+void Mips64Assembler::Beqc(GpuRegister rs, GpuRegister rt, Label* label) {
+  Bnec(rs, rt, 2);
+  B(label);
+}
+
+void Mips64Assembler::Bnec(GpuRegister rs, GpuRegister rt, Label* label) {
+  Beqc(rs, rt, 2);
+  B(label);
+}
+
+void Mips64Assembler::Beqzc(GpuRegister rs, Label* label) {
+  Bnezc(rs, 2);
+  B(label);
+}
+
+void Mips64Assembler::Bnezc(GpuRegister rs, Label* label) {
+  Beqzc(rs, 2);
+  B(label);
 }
 
 void Mips64Assembler::LoadFromOffset(LoadOperandType type, GpuRegister reg, GpuRegister base,
                                      int32_t offset) {
+  if (!IsInt<16>(offset)) {
+    LoadConst32(AT, offset);
+    Daddu(AT, AT, base);
+    base = AT;
+    offset = 0;
+  }
+
   switch (type) {
     case kLoadSignedByte:
       Lb(reg, base, offset);
@@ -489,23 +961,25 @@
       Lwu(reg, base, offset);
       break;
     case kLoadDoubleword:
-      // TODO: alignment issues ???
       Ld(reg, base, offset);
       break;
-    default:
-      LOG(FATAL) << "UNREACHABLE";
   }
 }
 
 void Mips64Assembler::LoadFpuFromOffset(LoadOperandType type, FpuRegister reg, GpuRegister base,
                                         int32_t offset) {
-  CHECK((offset >= -32768) && (offset <= 32766));
+  if (!IsInt<16>(offset)) {
+    LoadConst32(AT, offset);
+    Daddu(AT, AT, base);
+    base = AT;
+    offset = 0;
+  }
+
   switch (type) {
     case kLoadWord:
       Lwc1(reg, base, offset);
       break;
     case kLoadDoubleword:
-      // TODO: alignment issues ???
       Ldc1(reg, base, offset);
       break;
     default:
@@ -542,6 +1016,13 @@
 
 void Mips64Assembler::StoreToOffset(StoreOperandType type, GpuRegister reg, GpuRegister base,
                                     int32_t offset) {
+  if (!IsInt<16>(offset)) {
+    LoadConst32(AT, offset);
+    Daddu(AT, AT, base);
+    base = AT;
+    offset = 0;
+  }
+
   switch (type) {
     case kStoreByte:
       Sb(reg, base, offset);
@@ -553,7 +1034,6 @@
       Sw(reg, base, offset);
       break;
     case kStoreDoubleword:
-      // TODO: alignment issues ???
       Sd(reg, base, offset);
       break;
     default:
@@ -563,6 +1043,13 @@
 
 void Mips64Assembler::StoreFpuToOffset(StoreOperandType type, FpuRegister reg, GpuRegister base,
                                        int32_t offset) {
+  if (!IsInt<16>(offset)) {
+    LoadConst32(AT, offset);
+    Daddu(AT, AT, base);
+    base = AT;
+    offset = 0;
+  }
+
   switch (type) {
     case kStoreWord:
       Swc1(reg, base, offset);
@@ -613,10 +1100,12 @@
       // only increment stack offset.
       offset += size;
     } else if (reg.IsFpuRegister()) {
-      StoreFpuToOffset((size == 4) ? kStoreWord : kStoreDoubleword, reg.AsFpuRegister(), SP, offset);
+      StoreFpuToOffset((size == 4) ? kStoreWord : kStoreDoubleword,
+          reg.AsFpuRegister(), SP, offset);
       offset += size;
     } else if (reg.IsGpuRegister()) {
-      StoreToOffset((size == 4) ? kStoreWord : kStoreDoubleword, reg.AsGpuRegister(), SP, offset);
+      StoreToOffset((size == 4) ? kStoreWord : kStoreDoubleword,
+          reg.AsGpuRegister(), SP, offset);
       offset += size;
     }
   }
@@ -650,14 +1139,14 @@
 }
 
 void Mips64Assembler::IncreaseFrameSize(size_t adjust) {
-  CHECK_ALIGNED(adjust, kStackAlignment);
-  AddConstant64(SP, SP, -adjust);
+  CHECK_ALIGNED(adjust, kFramePointerSize);
+  Daddiu64(SP, SP, static_cast<int32_t>(-adjust));
   cfi_.AdjustCFAOffset(adjust);
 }
 
 void Mips64Assembler::DecreaseFrameSize(size_t adjust) {
-  CHECK_ALIGNED(adjust, kStackAlignment);
-  AddConstant64(SP, SP, adjust);
+  CHECK_ALIGNED(adjust, kFramePointerSize);
+  Daddiu64(SP, SP, static_cast<int32_t>(adjust));
   cfi_.AdjustCFAOffset(-adjust);
 }
 
@@ -702,7 +1191,7 @@
                                             ManagedRegister mscratch) {
   Mips64ManagedRegister scratch = mscratch.AsMips64();
   CHECK(scratch.IsGpuRegister()) << scratch;
-  LoadImmediate64(scratch.AsGpuRegister(), imm);
+  LoadConst32(scratch.AsGpuRegister(), imm);
   StoreToOffset(kStoreWord, scratch.AsGpuRegister(), SP, dest.Int32Value());
 }
 
@@ -710,7 +1199,9 @@
                                                ManagedRegister mscratch) {
   Mips64ManagedRegister scratch = mscratch.AsMips64();
   CHECK(scratch.IsGpuRegister()) << scratch;
-  LoadImmediate64(scratch.AsGpuRegister(), imm);
+  // TODO: it's unclear wether 32 or 64 bits need to be stored (Arm64 and x86/x64 disagree?).
+  // Is this function even referenced anywhere else in the code?
+  LoadConst32(scratch.AsGpuRegister(), imm);
   StoreToOffset(kStoreDoubleword, scratch.AsGpuRegister(), S1, dest.Int32Value());
 }
 
@@ -719,7 +1210,7 @@
                                                  ManagedRegister mscratch) {
   Mips64ManagedRegister scratch = mscratch.AsMips64();
   CHECK(scratch.IsGpuRegister()) << scratch;
-  AddConstant64(scratch.AsGpuRegister(), SP, fr_offs.Int32Value());
+  Daddiu64(scratch.AsGpuRegister(), SP, fr_offs.Int32Value());
   StoreToOffset(kStoreDoubleword, scratch.AsGpuRegister(), S1, thr_offs.Int32Value());
 }
 
@@ -757,20 +1248,24 @@
   LoadFromOffset(kLoadUnsignedWord, dest.AsGpuRegister(),
                  base.AsMips64().AsGpuRegister(), offs.Int32Value());
   if (kPoisonHeapReferences && poison_reference) {
-    Subu(dest.AsGpuRegister(), ZERO, dest.AsGpuRegister());
+    // TODO: review
+    // Negate the 32-bit ref
+    Dsubu(dest.AsGpuRegister(), ZERO, dest.AsGpuRegister());
+    // And constrain it to 32 bits (zero-extend into bits 32 through 63) as on Arm64 and x86/64
+    Dext(dest.AsGpuRegister(), dest.AsGpuRegister(), 0, 31);
   }
 }
 
 void Mips64Assembler::LoadRawPtr(ManagedRegister mdest, ManagedRegister base,
-                               Offset offs) {
+                                 Offset offs) {
   Mips64ManagedRegister dest = mdest.AsMips64();
-  CHECK(dest.IsGpuRegister() && dest.IsGpuRegister()) << dest;
+  CHECK(dest.IsGpuRegister() && base.AsMips64().IsGpuRegister());
   LoadFromOffset(kLoadDoubleword, dest.AsGpuRegister(),
                  base.AsMips64().AsGpuRegister(), offs.Int32Value());
 }
 
 void Mips64Assembler::LoadRawPtrFromThread64(ManagedRegister mdest,
-                                         ThreadOffset<8> offs) {
+                                             ThreadOffset<8> offs) {
   Mips64ManagedRegister dest = mdest.AsMips64();
   CHECK(dest.IsGpuRegister());
   LoadFromOffset(kLoadDoubleword, dest.AsGpuRegister(), S1, offs.Int32Value());
@@ -849,7 +1344,7 @@
 }
 
 void Mips64Assembler::Copy(FrameOffset dest, ManagedRegister src_base, Offset src_offset,
-                         ManagedRegister mscratch, size_t size) {
+                           ManagedRegister mscratch, size_t size) {
   GpuRegister scratch = mscratch.AsMips64().AsGpuRegister();
   CHECK(size == 4 || size == 8) << size;
   if (size == 4) {
@@ -866,7 +1361,7 @@
 }
 
 void Mips64Assembler::Copy(ManagedRegister dest_base, Offset dest_offset, FrameOffset src,
-                         ManagedRegister mscratch, size_t size) {
+                           ManagedRegister mscratch, size_t size) {
   GpuRegister scratch = mscratch.AsMips64().AsGpuRegister();
   CHECK(size == 4 || size == 8) << size;
   if (size == 4) {
@@ -888,8 +1383,8 @@
 }
 
 void Mips64Assembler::Copy(ManagedRegister dest, Offset dest_offset,
-                         ManagedRegister src, Offset src_offset,
-                         ManagedRegister mscratch, size_t size) {
+                           ManagedRegister src, Offset src_offset,
+                           ManagedRegister mscratch, size_t size) {
   GpuRegister scratch = mscratch.AsMips64().AsGpuRegister();
   CHECK(size == 4 || size == 8) << size;
   if (size == 4) {
@@ -912,12 +1407,14 @@
 }
 
 void Mips64Assembler::MemoryBarrier(ManagedRegister) {
+  // TODO: sync?
   UNIMPLEMENTED(FATAL) << "no mips64 implementation";
 }
 
 void Mips64Assembler::CreateHandleScopeEntry(ManagedRegister mout_reg,
-                                    FrameOffset handle_scope_offset,
-                                    ManagedRegister min_reg, bool null_allowed) {
+                                             FrameOffset handle_scope_offset,
+                                             ManagedRegister min_reg,
+                                             bool null_allowed) {
   Mips64ManagedRegister out_reg = mout_reg.AsMips64();
   Mips64ManagedRegister in_reg = min_reg.AsMips64();
   CHECK(in_reg.IsNoRegister() || in_reg.IsGpuRegister()) << in_reg;
@@ -933,20 +1430,20 @@
       in_reg = out_reg;
     }
     if (!out_reg.Equals(in_reg)) {
-      LoadImmediate64(out_reg.AsGpuRegister(), 0);
+      LoadConst32(out_reg.AsGpuRegister(), 0);
     }
-    EmitBranch(in_reg.AsGpuRegister(), ZERO, &null_arg, true);
-    AddConstant64(out_reg.AsGpuRegister(), SP, handle_scope_offset.Int32Value());
-    Bind(&null_arg, false);
+    Beqzc(in_reg.AsGpuRegister(), &null_arg);
+    Daddiu64(out_reg.AsGpuRegister(), SP, handle_scope_offset.Int32Value());
+    Bind(&null_arg);
   } else {
-    AddConstant64(out_reg.AsGpuRegister(), SP, handle_scope_offset.Int32Value());
+    Daddiu64(out_reg.AsGpuRegister(), SP, handle_scope_offset.Int32Value());
   }
 }
 
 void Mips64Assembler::CreateHandleScopeEntry(FrameOffset out_off,
-                                    FrameOffset handle_scope_offset,
-                                    ManagedRegister mscratch,
-                                    bool null_allowed) {
+                                             FrameOffset handle_scope_offset,
+                                             ManagedRegister mscratch,
+                                             bool null_allowed) {
   Mips64ManagedRegister scratch = mscratch.AsMips64();
   CHECK(scratch.IsGpuRegister()) << scratch;
   if (null_allowed) {
@@ -956,30 +1453,30 @@
     // Null values get a handle scope entry value of 0.  Otherwise, the handle scope entry is
     // the address in the handle scope holding the reference.
     // e.g. scratch = (scratch == 0) ? 0 : (SP+handle_scope_offset)
-    EmitBranch(scratch.AsGpuRegister(), ZERO, &null_arg, true);
-    AddConstant64(scratch.AsGpuRegister(), SP, handle_scope_offset.Int32Value());
-    Bind(&null_arg, false);
+    Beqzc(scratch.AsGpuRegister(), &null_arg);
+    Daddiu64(scratch.AsGpuRegister(), SP, handle_scope_offset.Int32Value());
+    Bind(&null_arg);
   } else {
-    AddConstant64(scratch.AsGpuRegister(), SP, handle_scope_offset.Int32Value());
+    Daddiu64(scratch.AsGpuRegister(), SP, handle_scope_offset.Int32Value());
   }
   StoreToOffset(kStoreDoubleword, scratch.AsGpuRegister(), SP, out_off.Int32Value());
 }
 
 // Given a handle scope entry, load the associated reference.
 void Mips64Assembler::LoadReferenceFromHandleScope(ManagedRegister mout_reg,
-                                          ManagedRegister min_reg) {
+                                                   ManagedRegister min_reg) {
   Mips64ManagedRegister out_reg = mout_reg.AsMips64();
   Mips64ManagedRegister in_reg = min_reg.AsMips64();
   CHECK(out_reg.IsGpuRegister()) << out_reg;
   CHECK(in_reg.IsGpuRegister()) << in_reg;
   Label null_arg;
   if (!out_reg.Equals(in_reg)) {
-    LoadImmediate64(out_reg.AsGpuRegister(), 0);
+    LoadConst32(out_reg.AsGpuRegister(), 0);
   }
-  EmitBranch(in_reg.AsGpuRegister(), ZERO, &null_arg, true);
+  Beqzc(in_reg.AsGpuRegister(), &null_arg);
   LoadFromOffset(kLoadDoubleword, out_reg.AsGpuRegister(),
                  in_reg.AsGpuRegister(), 0);
-  Bind(&null_arg, false);
+  Bind(&null_arg);
 }
 
 void Mips64Assembler::VerifyObject(ManagedRegister /*src*/, bool /*could_be_null*/) {
@@ -1022,7 +1519,7 @@
 }
 
 void Mips64Assembler::GetCurrentThread(FrameOffset offset,
-                                     ManagedRegister /*mscratch*/) {
+                                       ManagedRegister /*mscratch*/) {
   StoreToOffset(kStoreDoubleword, S1, SP, offset.Int32Value());
 }
 
@@ -1032,13 +1529,13 @@
   buffer_.EnqueueSlowPath(slow);
   LoadFromOffset(kLoadDoubleword, scratch.AsGpuRegister(),
                  S1, Thread::ExceptionOffset<8>().Int32Value());
-  EmitBranch(scratch.AsGpuRegister(), ZERO, slow->Entry(), false);
+  Bnezc(scratch.AsGpuRegister(), slow->Entry());
 }
 
 void Mips64ExceptionSlowPath::Emit(Assembler* sasm) {
   Mips64Assembler* sp_asm = down_cast<Mips64Assembler*>(sasm);
 #define __ sp_asm->
-  __ Bind(&entry_, false);
+  __ Bind(&entry_);
   if (stack_adjust_ != 0) {  // Fix up the frame.
     __ DecreaseFrameSize(stack_adjust_);
   }
@@ -1048,6 +1545,7 @@
   // Set up call to Thread::Current()->pDeliverException
   __ LoadFromOffset(kLoadDoubleword, T9, S1,
                     QUICK_ENTRYPOINT_OFFSET(8, pDeliverException).Int32Value());
+  // TODO: check T9 usage
   __ Jr(T9);
   // Call never returns
   __ Break();
diff --git a/compiler/utils/mips64/assembler_mips64.h b/compiler/utils/mips64/assembler_mips64.h
index 38419ab..88cc4bc 100644
--- a/compiler/utils/mips64/assembler_mips64.h
+++ b/compiler/utils/mips64/assembler_mips64.h
@@ -56,13 +56,31 @@
   void Addi(GpuRegister rt, GpuRegister rs, uint16_t imm16);
   void Addu(GpuRegister rd, GpuRegister rs, GpuRegister rt);
   void Addiu(GpuRegister rt, GpuRegister rs, uint16_t imm16);
-  void Daddiu(GpuRegister rt, GpuRegister rs, uint16_t imm16);
+  void Daddu(GpuRegister rd, GpuRegister rs, GpuRegister rt);  // MIPS64
+  void Daddiu(GpuRegister rt, GpuRegister rs, uint16_t imm16);  // MIPS64
   void Sub(GpuRegister rd, GpuRegister rs, GpuRegister rt);
   void Subu(GpuRegister rd, GpuRegister rs, GpuRegister rt);
-  void Mult(GpuRegister rs, GpuRegister rt);
-  void Multu(GpuRegister rs, GpuRegister rt);
-  void Div(GpuRegister rs, GpuRegister rt);
-  void Divu(GpuRegister rs, GpuRegister rt);
+  void Dsubu(GpuRegister rd, GpuRegister rs, GpuRegister rt);  // MIPS64
+
+  void MultR2(GpuRegister rs, GpuRegister rt);  // R2
+  void MultuR2(GpuRegister rs, GpuRegister rt);  // R2
+  void DivR2(GpuRegister rs, GpuRegister rt);  // R2
+  void DivuR2(GpuRegister rs, GpuRegister rt);  // R2
+  void MulR2(GpuRegister rd, GpuRegister rs, GpuRegister rt);  // R2
+  void DivR2(GpuRegister rd, GpuRegister rs, GpuRegister rt);  // R2
+  void ModR2(GpuRegister rd, GpuRegister rs, GpuRegister rt);  // R2
+  void DivuR2(GpuRegister rd, GpuRegister rs, GpuRegister rt);  // R2
+  void ModuR2(GpuRegister rd, GpuRegister rs, GpuRegister rt);  // R2
+  void MulR6(GpuRegister rd, GpuRegister rs, GpuRegister rt);  // R6
+  void DivR6(GpuRegister rd, GpuRegister rs, GpuRegister rt);  // R6
+  void ModR6(GpuRegister rd, GpuRegister rs, GpuRegister rt);  // R6
+  void DivuR6(GpuRegister rd, GpuRegister rs, GpuRegister rt);  // R6
+  void ModuR6(GpuRegister rd, GpuRegister rs, GpuRegister rt);  // R6
+  void Dmul(GpuRegister rd, GpuRegister rs, GpuRegister rt);  // MIPS64 R6
+  void Ddiv(GpuRegister rd, GpuRegister rs, GpuRegister rt);  // MIPS64 R6
+  void Dmod(GpuRegister rd, GpuRegister rs, GpuRegister rt);  // MIPS64 R6
+  void Ddivu(GpuRegister rd, GpuRegister rs, GpuRegister rt);  // MIPS64 R6
+  void Dmodu(GpuRegister rd, GpuRegister rs, GpuRegister rt);  // MIPS64 R6
 
   void And(GpuRegister rd, GpuRegister rs, GpuRegister rt);
   void Andi(GpuRegister rt, GpuRegister rs, uint16_t imm16);
@@ -72,40 +90,72 @@
   void Xori(GpuRegister rt, GpuRegister rs, uint16_t imm16);
   void Nor(GpuRegister rd, GpuRegister rs, GpuRegister rt);
 
-  void Sll(GpuRegister rd, GpuRegister rs, int shamt);
-  void Srl(GpuRegister rd, GpuRegister rs, int shamt);
-  void Sra(GpuRegister rd, GpuRegister rs, int shamt);
-  void Sllv(GpuRegister rd, GpuRegister rs, GpuRegister rt);
-  void Srlv(GpuRegister rd, GpuRegister rs, GpuRegister rt);
-  void Srav(GpuRegister rd, GpuRegister rs, GpuRegister rt);
+  void Seb(GpuRegister rd, GpuRegister rt);  // R2+
+  void Seh(GpuRegister rd, GpuRegister rt);  // R2+
+  void Dext(GpuRegister rs, GpuRegister rt, int pos, int size_less_one);  // MIPS64
+
+  void Sll(GpuRegister rd, GpuRegister rt, int shamt);
+  void Srl(GpuRegister rd, GpuRegister rt, int shamt);
+  void Sra(GpuRegister rd, GpuRegister rt, int shamt);
+  void Sllv(GpuRegister rd, GpuRegister rt, GpuRegister rs);
+  void Srlv(GpuRegister rd, GpuRegister rt, GpuRegister rs);
+  void Srav(GpuRegister rd, GpuRegister rt, GpuRegister rs);
+  void Dsll(GpuRegister rd, GpuRegister rt, int shamt);  // MIPS64
+  void Dsrl(GpuRegister rd, GpuRegister rt, int shamt);  // MIPS64
+  void Dsra(GpuRegister rd, GpuRegister rt, int shamt);  // MIPS64
+  void Dsll32(GpuRegister rd, GpuRegister rt, int shamt);  // MIPS64
+  void Dsrl32(GpuRegister rd, GpuRegister rt, int shamt);  // MIPS64
+  void Dsra32(GpuRegister rd, GpuRegister rt, int shamt);  // MIPS64
+  void Dsllv(GpuRegister rd, GpuRegister rt, GpuRegister rs);  // MIPS64
+  void Dsrlv(GpuRegister rd, GpuRegister rt, GpuRegister rs);  // MIPS64
+  void Dsrav(GpuRegister rd, GpuRegister rt, GpuRegister rs);  // MIPS64
 
   void Lb(GpuRegister rt, GpuRegister rs, uint16_t imm16);
   void Lh(GpuRegister rt, GpuRegister rs, uint16_t imm16);
   void Lw(GpuRegister rt, GpuRegister rs, uint16_t imm16);
-  void Ld(GpuRegister rt, GpuRegister rs, uint16_t imm16);
+  void Ld(GpuRegister rt, GpuRegister rs, uint16_t imm16);  // MIPS64
   void Lbu(GpuRegister rt, GpuRegister rs, uint16_t imm16);
   void Lhu(GpuRegister rt, GpuRegister rs, uint16_t imm16);
-  void Lwu(GpuRegister rt, GpuRegister rs, uint16_t imm16);
+  void Lwu(GpuRegister rt, GpuRegister rs, uint16_t imm16);  // MIPS64
   void Lui(GpuRegister rt, uint16_t imm16);
-  void Mfhi(GpuRegister rd);
-  void Mflo(GpuRegister rd);
+  void Dahi(GpuRegister rs, uint16_t imm16);  // MIPS64 R6
+  void Dati(GpuRegister rs, uint16_t imm16);  // MIPS64 R6
+  void Sync(uint32_t stype);
+  void Mfhi(GpuRegister rd);  // R2
+  void Mflo(GpuRegister rd);  // R2
 
   void Sb(GpuRegister rt, GpuRegister rs, uint16_t imm16);
   void Sh(GpuRegister rt, GpuRegister rs, uint16_t imm16);
   void Sw(GpuRegister rt, GpuRegister rs, uint16_t imm16);
-  void Sd(GpuRegister rt, GpuRegister rs, uint16_t imm16);
+  void Sd(GpuRegister rt, GpuRegister rs, uint16_t imm16);  // MIPS64
 
   void Slt(GpuRegister rd, GpuRegister rs, GpuRegister rt);
   void Sltu(GpuRegister rd, GpuRegister rs, GpuRegister rt);
   void Slti(GpuRegister rt, GpuRegister rs, uint16_t imm16);
   void Sltiu(GpuRegister rt, GpuRegister rs, uint16_t imm16);
 
-  void Beq(GpuRegister rt, GpuRegister rs, uint16_t imm16);
-  void Bne(GpuRegister rt, GpuRegister rs, uint16_t imm16);
-  void J(uint32_t address);
-  void Jal(uint32_t address);
-  void Jr(GpuRegister rs);
+  void Beq(GpuRegister rs, GpuRegister rt, uint16_t imm16);
+  void Bne(GpuRegister rs, GpuRegister rt, uint16_t imm16);
+  void J(uint32_t addr26);
+  void Jal(uint32_t addr26);
+  void Jalr(GpuRegister rd, GpuRegister rs);
   void Jalr(GpuRegister rs);
+  void Jr(GpuRegister rs);
+  void Auipc(GpuRegister rs, uint16_t imm16);  // R6
+  void Jic(GpuRegister rt, uint16_t imm16);  // R6
+  void Jialc(GpuRegister rt, uint16_t imm16);  // R6
+  void Bltc(GpuRegister rs, GpuRegister rt, uint16_t imm16);  // R6
+  void Bltzc(GpuRegister rt, uint16_t imm16);  // R6
+  void Bgtzc(GpuRegister rt, uint16_t imm16);  // R6
+  void Bgec(GpuRegister rs, GpuRegister rt, uint16_t imm16);  // R6
+  void Bgezc(GpuRegister rt, uint16_t imm16);  // R6
+  void Blezc(GpuRegister rt, uint16_t imm16);  // R6
+  void Bltuc(GpuRegister rs, GpuRegister rt, uint16_t imm16);  // R6
+  void Bgeuc(GpuRegister rs, GpuRegister rt, uint16_t imm16);  // R6
+  void Beqc(GpuRegister rs, GpuRegister rt, uint16_t imm16);  // R6
+  void Bnec(GpuRegister rs, GpuRegister rt, uint16_t imm16);  // R6
+  void Beqzc(GpuRegister rs, uint32_t imm21);  // R6
+  void Bnezc(GpuRegister rs, uint32_t imm21);  // R6
 
   void AddS(FpuRegister fd, FpuRegister fs, FpuRegister ft);
   void SubS(FpuRegister fd, FpuRegister fs, FpuRegister ft);
@@ -117,9 +167,18 @@
   void DivD(FpuRegister fd, FpuRegister fs, FpuRegister ft);
   void MovS(FpuRegister fd, FpuRegister fs);
   void MovD(FpuRegister fd, FpuRegister fs);
+  void NegS(FpuRegister fd, FpuRegister fs);
+  void NegD(FpuRegister fd, FpuRegister fs);
+
+  void Cvtsw(FpuRegister fd, FpuRegister fs);
+  void Cvtdw(FpuRegister fd, FpuRegister fs);
+  void Cvtsd(FpuRegister fd, FpuRegister fs);
+  void Cvtds(FpuRegister fd, FpuRegister fs);
 
   void Mfc1(GpuRegister rt, FpuRegister fs);
-  void Mtc1(FpuRegister ft, GpuRegister rs);
+  void Mtc1(GpuRegister rt, FpuRegister fs);
+  void Dmfc1(GpuRegister rt, FpuRegister fs);  // MIPS64
+  void Dmtc1(GpuRegister rt, FpuRegister fs);  // MIPS64
   void Lwc1(FpuRegister ft, GpuRegister rs, uint16_t imm16);
   void Ldc1(FpuRegister ft, GpuRegister rs, uint16_t imm16);
   void Swc1(FpuRegister ft, GpuRegister rs, uint16_t imm16);
@@ -127,15 +186,33 @@
 
   void Break();
   void Nop();
-  void Move(GpuRegister rt, GpuRegister rs);
-  void Clear(GpuRegister rt);
-  void Not(GpuRegister rt, GpuRegister rs);
-  void Mul(GpuRegister rd, GpuRegister rs, GpuRegister rt);
-  void Div(GpuRegister rd, GpuRegister rs, GpuRegister rt);
-  void Rem(GpuRegister rd, GpuRegister rs, GpuRegister rt);
+  void Move(GpuRegister rd, GpuRegister rs);
+  void Clear(GpuRegister rd);
+  void Not(GpuRegister rd, GpuRegister rs);
 
-  void AddConstant64(GpuRegister rt, GpuRegister rs, int32_t value);
-  void LoadImmediate64(GpuRegister rt, int32_t value);
+  // Higher level composite instructions
+  void LoadConst32(GpuRegister rd, int32_t value);
+  void LoadConst64(GpuRegister rd, int64_t value);  // MIPS64
+
+  void Addiu32(GpuRegister rt, GpuRegister rs, int32_t value, GpuRegister rtmp = AT);
+  void Daddiu64(GpuRegister rt, GpuRegister rs, int64_t value, GpuRegister rtmp = AT);  // MIPS64
+
+  void Bind(Label* label);  // R6
+  void B(Label* label);  // R6
+  void Jalr(Label* label, GpuRegister indirect_reg = RA);  // R6
+  // TODO: implement common for R6 and non-R6 interface for conditional branches?
+  void Bltc(GpuRegister rs, GpuRegister rt, Label* label);  // R6
+  void Bltzc(GpuRegister rt, Label* label);  // R6
+  void Bgtzc(GpuRegister rt, Label* label);  // R6
+  void Bgec(GpuRegister rs, GpuRegister rt, Label* label);  // R6
+  void Bgezc(GpuRegister rt, Label* label);  // R6
+  void Blezc(GpuRegister rt, Label* label);  // R6
+  void Bltuc(GpuRegister rs, GpuRegister rt, Label* label);  // R6
+  void Bgeuc(GpuRegister rs, GpuRegister rt, Label* label);  // R6
+  void Beqc(GpuRegister rs, GpuRegister rt, Label* label);  // R6
+  void Bnec(GpuRegister rs, GpuRegister rt, Label* label);  // R6
+  void Beqzc(GpuRegister rs, Label* label);  // R6
+  void Bnezc(GpuRegister rs, Label* label);  // R6
 
   void EmitLoad(ManagedRegister m_dst, GpuRegister src_register, int32_t src_offset, size_t size);
   void LoadFromOffset(LoadOperandType type, GpuRegister reg, GpuRegister base, int32_t offset);
@@ -144,10 +221,7 @@
   void StoreFpuToOffset(StoreOperandType type, FpuRegister reg, GpuRegister base, int32_t offset);
 
   // Emit data (e.g. encoded instruction or immediate) to the instruction stream.
-  void Emit(int32_t value);
-  void EmitBranch(GpuRegister rt, GpuRegister rs, Label* label, bool equal);
-  void EmitJump(Label* label, bool link);
-  void Bind(Label* label, bool is_jump);
+  void Emit(uint32_t value);
 
   //
   // Overridden common assembler high-level functionality
@@ -269,13 +343,11 @@
  private:
   void EmitR(int opcode, GpuRegister rs, GpuRegister rt, GpuRegister rd, int shamt, int funct);
   void EmitI(int opcode, GpuRegister rs, GpuRegister rt, uint16_t imm);
-  void EmitJ(int opcode, int address);
+  void EmitI21(int opcode, GpuRegister rs, uint32_t imm21);
+  void EmitJ(int opcode, uint32_t addr26);
   void EmitFR(int opcode, int fmt, FpuRegister ft, FpuRegister fs, FpuRegister fd, int funct);
   void EmitFI(int opcode, int fmt, FpuRegister rt, uint16_t imm);
 
-  int32_t EncodeBranchOffset(int offset, int32_t inst, bool is_jump);
-  int DecodeBranchOffset(int32_t inst, bool is_jump);
-
   DISALLOW_COPY_AND_ASSIGN(Mips64Assembler);
 };
 
diff --git a/compiler/utils/mips64/constants_mips64.h b/compiler/utils/mips64/constants_mips64.h
index 8b7697c..f57498d 100644
--- a/compiler/utils/mips64/constants_mips64.h
+++ b/compiler/utils/mips64/constants_mips64.h
@@ -67,7 +67,7 @@
   static const uint32_t kBreakPointInstruction = 0x0000000D;
 
   bool IsBreakPoint() {
-    return ((*reinterpret_cast<const uint32_t*>(this)) & 0xFC0000CF) == kBreakPointInstruction;
+    return ((*reinterpret_cast<const uint32_t*>(this)) & 0xFC00003F) == kBreakPointInstruction;
   }
 
   // Instructions are read out of a code stream. The only way to get a
diff --git a/disassembler/disassembler_mips.cc b/disassembler/disassembler_mips.cc
index ac81737..70ca88d 100644
--- a/disassembler/disassembler_mips.cc
+++ b/disassembler/disassembler_mips.cc
@@ -106,6 +106,9 @@
   { kRTypeMask | (0x1f << 21), 62 | (1 << 21), "drotr32", "DTA", },
   { kRTypeMask, 62, "dsrl32", "DTA", },
   { kRTypeMask, 63, "dsra32", "DTA", },
+  { kRTypeMask, (31u << kOpcodeShift) | 3, "dext", "TSAZ", },
+  { kRTypeMask | (0x1f << 21) | (0x1f << 6), (31u << 26) | (16 << 6) | 32, "seb", "DT", },
+  { kRTypeMask | (0x1f << 21) | (0x1f << 6), (31u << 26) | (24 << 6) | 32, "seh", "DT", },
 
   // SPECIAL0
   { kSpecial0Mask | 0x7ff, (2 << 6) | 24, "mul", "DST" },
@@ -150,13 +153,17 @@
   { kITypeMask | (0x1f << 16), 1 << kOpcodeShift | (16 << 16), "bltzal", "SB" },
   { kITypeMask | (0x1f << 16), 1 << kOpcodeShift | (18 << 16), "bltzall", "SB" },
   { kITypeMask | (0x1f << 16), 6 << kOpcodeShift | (0 << 16), "blez", "SB" },
+  { kITypeMask, 6 << kOpcodeShift, "bgeuc", "STB" },
   { kITypeMask | (0x1f << 16), 7 << kOpcodeShift | (0 << 16), "bgtz", "SB" },
+  { kITypeMask, 7 << kOpcodeShift, "bltuc", "STB" },
   { kITypeMask | (0x1f << 16), 1 << kOpcodeShift | (6 << 16), "dahi", "Si", },
   { kITypeMask | (0x1f << 16), 1 << kOpcodeShift | (30 << 16), "dati", "Si", },
 
   { 0xffff0000, (4 << kOpcodeShift), "b", "B" },
   { 0xffff0000, (1 << kOpcodeShift) | (17 << 16), "bal", "B" },
 
+  { kITypeMask, 8 << kOpcodeShift, "beqc", "STB" },
+
   { kITypeMask, 8 << kOpcodeShift, "addi", "TSi", },
   { kITypeMask, 9 << kOpcodeShift, "addiu", "TSi", },
   { kITypeMask, 10 << kOpcodeShift, "slti", "TSi", },
@@ -166,6 +173,83 @@
   { kITypeMask, 14 << kOpcodeShift, "xori", "TSi", },
   { kITypeMask | (0x1f << 21), 15 << kOpcodeShift, "lui", "TI", },
   { kITypeMask, 15 << kOpcodeShift, "aui", "TSI", },
+
+  { kITypeMask | (0x1f << 21), 22 << kOpcodeShift, "blezc", "TB" },
+
+  // TODO: de-dup
+  { kITypeMask | (0x3ff << 16), (22 << kOpcodeShift) | (1  << 21) | (1  << 16), "bgezc", "TB" },
+  { kITypeMask | (0x3ff << 16), (22 << kOpcodeShift) | (2  << 21) | (2  << 16), "bgezc", "TB" },
+  { kITypeMask | (0x3ff << 16), (22 << kOpcodeShift) | (3  << 21) | (3  << 16), "bgezc", "TB" },
+  { kITypeMask | (0x3ff << 16), (22 << kOpcodeShift) | (4  << 21) | (4  << 16), "bgezc", "TB" },
+  { kITypeMask | (0x3ff << 16), (22 << kOpcodeShift) | (5  << 21) | (5  << 16), "bgezc", "TB" },
+  { kITypeMask | (0x3ff << 16), (22 << kOpcodeShift) | (6  << 21) | (6  << 16), "bgezc", "TB" },
+  { kITypeMask | (0x3ff << 16), (22 << kOpcodeShift) | (7  << 21) | (7  << 16), "bgezc", "TB" },
+  { kITypeMask | (0x3ff << 16), (22 << kOpcodeShift) | (8  << 21) | (8  << 16), "bgezc", "TB" },
+  { kITypeMask | (0x3ff << 16), (22 << kOpcodeShift) | (9  << 21) | (9  << 16), "bgezc", "TB" },
+  { kITypeMask | (0x3ff << 16), (22 << kOpcodeShift) | (10 << 21) | (10 << 16), "bgezc", "TB" },
+  { kITypeMask | (0x3ff << 16), (22 << kOpcodeShift) | (11 << 21) | (11 << 16), "bgezc", "TB" },
+  { kITypeMask | (0x3ff << 16), (22 << kOpcodeShift) | (12 << 21) | (12 << 16), "bgezc", "TB" },
+  { kITypeMask | (0x3ff << 16), (22 << kOpcodeShift) | (13 << 21) | (13 << 16), "bgezc", "TB" },
+  { kITypeMask | (0x3ff << 16), (22 << kOpcodeShift) | (14 << 21) | (14 << 16), "bgezc", "TB" },
+  { kITypeMask | (0x3ff << 16), (22 << kOpcodeShift) | (15 << 21) | (15 << 16), "bgezc", "TB" },
+  { kITypeMask | (0x3ff << 16), (22 << kOpcodeShift) | (16 << 21) | (16 << 16), "bgezc", "TB" },
+  { kITypeMask | (0x3ff << 16), (22 << kOpcodeShift) | (17 << 21) | (17 << 16), "bgezc", "TB" },
+  { kITypeMask | (0x3ff << 16), (22 << kOpcodeShift) | (18 << 21) | (18 << 16), "bgezc", "TB" },
+  { kITypeMask | (0x3ff << 16), (22 << kOpcodeShift) | (19 << 21) | (19 << 16), "bgezc", "TB" },
+  { kITypeMask | (0x3ff << 16), (22 << kOpcodeShift) | (20 << 21) | (20 << 16), "bgezc", "TB" },
+  { kITypeMask | (0x3ff << 16), (22 << kOpcodeShift) | (21 << 21) | (21 << 16), "bgezc", "TB" },
+  { kITypeMask | (0x3ff << 16), (22 << kOpcodeShift) | (22 << 21) | (22 << 16), "bgezc", "TB" },
+  { kITypeMask | (0x3ff << 16), (22 << kOpcodeShift) | (23 << 21) | (23 << 16), "bgezc", "TB" },
+  { kITypeMask | (0x3ff << 16), (22 << kOpcodeShift) | (24 << 21) | (24 << 16), "bgezc", "TB" },
+  { kITypeMask | (0x3ff << 16), (22 << kOpcodeShift) | (25 << 21) | (25 << 16), "bgezc", "TB" },
+  { kITypeMask | (0x3ff << 16), (22 << kOpcodeShift) | (26 << 21) | (26 << 16), "bgezc", "TB" },
+  { kITypeMask | (0x3ff << 16), (22 << kOpcodeShift) | (27 << 21) | (27 << 16), "bgezc", "TB" },
+  { kITypeMask | (0x3ff << 16), (22 << kOpcodeShift) | (28 << 21) | (28 << 16), "bgezc", "TB" },
+  { kITypeMask | (0x3ff << 16), (22 << kOpcodeShift) | (29 << 21) | (29 << 16), "bgezc", "TB" },
+  { kITypeMask | (0x3ff << 16), (22 << kOpcodeShift) | (30 << 21) | (30 << 16), "bgezc", "TB" },
+  { kITypeMask | (0x3ff << 16), (22 << kOpcodeShift) | (31 << 21) | (31 << 16), "bgezc", "TB" },
+
+  { kITypeMask, 22 << kOpcodeShift, "bgec", "STB" },
+
+  { kITypeMask | (0x1f << 21), 23 << kOpcodeShift, "bgtzc", "TB" },
+
+  // TODO: de-dup
+  { kITypeMask | (0x3ff << 16), (23 << kOpcodeShift) | (1  << 21) | (1  << 16), "bltzc", "TB" },
+  { kITypeMask | (0x3ff << 16), (23 << kOpcodeShift) | (2  << 21) | (2  << 16), "bltzc", "TB" },
+  { kITypeMask | (0x3ff << 16), (23 << kOpcodeShift) | (3  << 21) | (3  << 16), "bltzc", "TB" },
+  { kITypeMask | (0x3ff << 16), (23 << kOpcodeShift) | (4  << 21) | (4  << 16), "bltzc", "TB" },
+  { kITypeMask | (0x3ff << 16), (23 << kOpcodeShift) | (5  << 21) | (5  << 16), "bltzc", "TB" },
+  { kITypeMask | (0x3ff << 16), (23 << kOpcodeShift) | (6  << 21) | (6  << 16), "bltzc", "TB" },
+  { kITypeMask | (0x3ff << 16), (23 << kOpcodeShift) | (7  << 21) | (7  << 16), "bltzc", "TB" },
+  { kITypeMask | (0x3ff << 16), (23 << kOpcodeShift) | (8  << 21) | (8  << 16), "bltzc", "TB" },
+  { kITypeMask | (0x3ff << 16), (23 << kOpcodeShift) | (9  << 21) | (9  << 16), "bltzc", "TB" },
+  { kITypeMask | (0x3ff << 16), (23 << kOpcodeShift) | (10 << 21) | (10 << 16), "bltzc", "TB" },
+  { kITypeMask | (0x3ff << 16), (23 << kOpcodeShift) | (11 << 21) | (11 << 16), "bltzc", "TB" },
+  { kITypeMask | (0x3ff << 16), (23 << kOpcodeShift) | (12 << 21) | (12 << 16), "bltzc", "TB" },
+  { kITypeMask | (0x3ff << 16), (23 << kOpcodeShift) | (13 << 21) | (13 << 16), "bltzc", "TB" },
+  { kITypeMask | (0x3ff << 16), (23 << kOpcodeShift) | (14 << 21) | (14 << 16), "bltzc", "TB" },
+  { kITypeMask | (0x3ff << 16), (23 << kOpcodeShift) | (15 << 21) | (15 << 16), "bltzc", "TB" },
+  { kITypeMask | (0x3ff << 16), (23 << kOpcodeShift) | (16 << 21) | (16 << 16), "bltzc", "TB" },
+  { kITypeMask | (0x3ff << 16), (23 << kOpcodeShift) | (17 << 21) | (17 << 16), "bltzc", "TB" },
+  { kITypeMask | (0x3ff << 16), (23 << kOpcodeShift) | (18 << 21) | (18 << 16), "bltzc", "TB" },
+  { kITypeMask | (0x3ff << 16), (23 << kOpcodeShift) | (19 << 21) | (19 << 16), "bltzc", "TB" },
+  { kITypeMask | (0x3ff << 16), (23 << kOpcodeShift) | (20 << 21) | (20 << 16), "bltzc", "TB" },
+  { kITypeMask | (0x3ff << 16), (23 << kOpcodeShift) | (21 << 21) | (21 << 16), "bltzc", "TB" },
+  { kITypeMask | (0x3ff << 16), (23 << kOpcodeShift) | (22 << 21) | (22 << 16), "bltzc", "TB" },
+  { kITypeMask | (0x3ff << 16), (23 << kOpcodeShift) | (23 << 21) | (23 << 16), "bltzc", "TB" },
+  { kITypeMask | (0x3ff << 16), (23 << kOpcodeShift) | (24 << 21) | (24 << 16), "bltzc", "TB" },
+  { kITypeMask | (0x3ff << 16), (23 << kOpcodeShift) | (25 << 21) | (25 << 16), "bltzc", "TB" },
+  { kITypeMask | (0x3ff << 16), (23 << kOpcodeShift) | (26 << 21) | (26 << 16), "bltzc", "TB" },
+  { kITypeMask | (0x3ff << 16), (23 << kOpcodeShift) | (27 << 21) | (27 << 16), "bltzc", "TB" },
+  { kITypeMask | (0x3ff << 16), (23 << kOpcodeShift) | (28 << 21) | (28 << 16), "bltzc", "TB" },
+  { kITypeMask | (0x3ff << 16), (23 << kOpcodeShift) | (29 << 21) | (29 << 16), "bltzc", "TB" },
+  { kITypeMask | (0x3ff << 16), (23 << kOpcodeShift) | (30 << 21) | (30 << 16), "bltzc", "TB" },
+  { kITypeMask | (0x3ff << 16), (23 << kOpcodeShift) | (31 << 21) | (31 << 16), "bltzc", "TB" },
+
+  { kITypeMask, 23 << kOpcodeShift, "bltc", "STB" },
+
+  { kITypeMask, 24 << kOpcodeShift, "bnec", "STB" },
+
   { kITypeMask, 25 << kOpcodeShift, "daddiu", "TSi", },
   { kITypeMask, 29 << kOpcodeShift, "daui", "TSi", },
 
@@ -180,9 +264,22 @@
   { kITypeMask, 43u << kOpcodeShift, "sw", "TO", },
   { kITypeMask, 49u << kOpcodeShift, "lwc1", "tO", },
   { kITypeMask, 53u << kOpcodeShift, "ldc1", "tO", },
+  { kITypeMask | (0x1f << 21), 54u << kOpcodeShift, "jic", "Ti" },
+  { kITypeMask | (1 << 21), (54u << kOpcodeShift) | (1 << 21), "beqzc", "Sb" },  // TODO: de-dup?
+  { kITypeMask | (1 << 22), (54u << kOpcodeShift) | (1 << 22), "beqzc", "Sb" },
+  { kITypeMask | (1 << 23), (54u << kOpcodeShift) | (1 << 23), "beqzc", "Sb" },
+  { kITypeMask | (1 << 24), (54u << kOpcodeShift) | (1 << 24), "beqzc", "Sb" },
+  { kITypeMask | (1 << 25), (54u << kOpcodeShift) | (1 << 25), "beqzc", "Sb" },
   { kITypeMask, 55u << kOpcodeShift, "ld", "TO", },
   { kITypeMask, 57u << kOpcodeShift, "swc1", "tO", },
+  { kITypeMask | (0x1f << 16), (59u << kOpcodeShift) | (30 << 16), "auipc", "Si" },
   { kITypeMask, 61u << kOpcodeShift, "sdc1", "tO", },
+  { kITypeMask | (0x1f << 21), 62u << kOpcodeShift, "jialc", "Ti" },
+  { kITypeMask | (1 << 21), (62u << kOpcodeShift) | (1 << 21), "bnezc", "Sb" },  // TODO: de-dup?
+  { kITypeMask | (1 << 22), (62u << kOpcodeShift) | (1 << 22), "bnezc", "Sb" },
+  { kITypeMask | (1 << 23), (62u << kOpcodeShift) | (1 << 23), "bnezc", "Sb" },
+  { kITypeMask | (1 << 24), (62u << kOpcodeShift) | (1 << 24), "bnezc", "Sb" },
+  { kITypeMask | (1 << 25), (62u << kOpcodeShift) | (1 << 25), "bnezc", "Sb" },
   { kITypeMask, 63u << kOpcodeShift, "sd", "TO", },
 
   // Floating point.
@@ -241,7 +338,7 @@
       opcode = gMipsInstructions[i].name;
       for (const char* args_fmt = gMipsInstructions[i].args_fmt; *args_fmt; ++args_fmt) {
         switch (*args_fmt) {
-          case 'A':  // sa (shift amount).
+          case 'A':  // sa (shift amount or [d]ext position).
             args << sa;
             break;
           case 'B':  // Branch offset.
@@ -253,6 +350,15 @@
                    << StringPrintf("  ; %+d", offset);
             }
             break;
+          case 'b':  // 21-bit branch offset.
+            {
+              int32_t offset = (instruction & 0x1fffff) - ((instruction & 0x100000) << 1);
+              offset <<= 2;
+              offset += 4;  // Delay slot.
+              args << FormatInstructionPointer(instr_ptr + offset)
+                   << StringPrintf("  ; %+d", offset);
+            }
+            break;
           case 'D': args << 'r' << rd; break;
           case 'd': args << 'f' << rd; break;
           case 'a': args << 'f' << sa; break;
@@ -302,6 +408,7 @@
           case 's': args << 'f' << rs; break;
           case 'T': args << 'r' << rt; break;
           case 't': args << 'f' << rt; break;
+          case 'Z': args << rd; break;   // sz ([d]ext size).
         }
         if (*(args_fmt + 1)) {
           args << ", ";
@@ -311,9 +418,36 @@
     }
   }
 
+  // Special cases for sequences of:
+  //   pc-relative +/- 2GB branch:
+  //     auipc  reg, imm
+  //     jic    reg, imm
+  //   pc-relative +/- 2GB branch and link:
+  //     auipc  reg, imm
+  //     daddiu reg, reg, imm
+  //     jialc  reg, 0
+  if (((op == 0x36 && rs == 0 && rt != 0) ||  // jic
+       (op == 0x19 && rs == rt && rt != 0)) &&  // daddiu
+      last_ptr_ && (intptr_t)instr_ptr - (intptr_t)last_ptr_ == 4 &&
+      (last_instr_ & 0xFC1F0000) == 0xEC1E0000 &&  // auipc
+      ((last_instr_ >> 21) & 0x1F) == rt) {
+    uint32_t offset = (last_instr_ << 16) | (instruction & 0xFFFF);
+    offset -= (offset & 0x8000) << 1;
+    offset -= 4;
+    if (op == 0x36) {
+      args << "  ; b ";
+    } else {
+      args << "  ; move r" << rt << ", ";
+    }
+    args << FormatInstructionPointer(instr_ptr + (int32_t)offset);
+    args << StringPrintf("  ; %+d", (int32_t)offset);
+  }
+
   os << FormatInstructionPointer(instr_ptr)
      << StringPrintf(": %08x\t%-7s ", instruction, opcode.c_str())
      << args.str() << '\n';
+  last_ptr_ = instr_ptr;
+  last_instr_ = instruction;
   return 4;
 }
 
diff --git a/disassembler/disassembler_mips.h b/disassembler/disassembler_mips.h
index 67c3fcb..4f70a9b 100644
--- a/disassembler/disassembler_mips.h
+++ b/disassembler/disassembler_mips.h
@@ -27,7 +27,9 @@
 class DisassemblerMips FINAL : public Disassembler {
  public:
   explicit DisassemblerMips(DisassemblerOptions* options, bool is64bit) : Disassembler(options),
-      is64bit_(is64bit) {}
+      is64bit_(is64bit),
+      last_ptr_(nullptr),
+      last_instr_(0) {}
 
   size_t Dump(std::ostream& os, const uint8_t* begin) OVERRIDE;
   void Dump(std::ostream& os, const uint8_t* begin, const uint8_t* end) OVERRIDE;
@@ -35,6 +37,11 @@
  private:
   const bool is64bit_;
 
+  // Address and encoding of the last disassembled instruction.
+  // Needed to produce more readable disassembly of certain 2-instruction sequences.
+  const uint8_t* last_ptr_;
+  uint32_t last_instr_;
+
   DISALLOW_COPY_AND_ASSIGN(DisassemblerMips);
 };
 
diff --git a/test/507-referrer/expected.txt b/test/507-referrer/expected.txt
new file mode 100644
index 0000000..e69de29
--- /dev/null
+++ b/test/507-referrer/expected.txt
diff --git a/test/507-referrer/info.txt b/test/507-referrer/info.txt
new file mode 100644
index 0000000..1335994
--- /dev/null
+++ b/test/507-referrer/info.txt
@@ -0,0 +1,2 @@
+Regression test for the optimizing compiler, which used
+to do incorrect access checks on static fields when inlining.
diff --git a/test/507-referrer/src/Main.java b/test/507-referrer/src/Main.java
new file mode 100644
index 0000000..6393f39
--- /dev/null
+++ b/test/507-referrer/src/Main.java
@@ -0,0 +1,26 @@
+/*
+ * Copyright (C) 2015 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.
+ */
+
+import p1.InPackage;
+
+public class Main {
+  public static void main(String[] args) {
+    int result = InPackage.$inline$foo();
+    if (result != 42) {
+      throw new Error("Expected 42, got " + result);
+    }
+  }
+}
diff --git a/test/507-referrer/src/p1/InPackage.java b/test/507-referrer/src/p1/InPackage.java
new file mode 100644
index 0000000..162f055
--- /dev/null
+++ b/test/507-referrer/src/p1/InPackage.java
@@ -0,0 +1,25 @@
+/*
+ * Copyright (C) 2015 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.
+ */
+
+package p1;
+
+public class InPackage {
+  public static int $inline$foo() {
+    return myField;
+  }
+
+  protected static int myField = 42;
+}
diff --git a/test/Android.run-test.mk b/test/Android.run-test.mk
index c2380cc..60165d9 100644
--- a/test/Android.run-test.mk
+++ b/test/Android.run-test.mk
@@ -398,6 +398,37 @@
 
 TEST_ART_BROKEN_OPTIMIZING_ARM64_RUN_TESTS :=
 
+# Known broken tests for the MIPS64 optimizing compiler backend in 64-bit mode.  b/21555893
+TEST_ART_BROKEN_OPTIMIZING_MIPS64_64BIT_RUN_TESTS := \
+  004-SignalTest \
+  018-stack-overflow \
+  107-int-math2 \
+  449-checker-bce
+
+ifeq ($(TARGET_ARCH),mips64)
+  ifneq (,$(filter optimizing,$(COMPILER_TYPES)))
+    ART_TEST_KNOWN_BROKEN += $(call all-run-test-names,target,$(RUN_TYPES),$(PREBUILD_TYPES), \
+        optimizing,$(RELOCATE_TYPES),$(TRACE_TYPES),$(GC_TYPES),$(JNI_TYPES), \
+        $(IMAGE_TYPES),$(PICTEST_TYPES),$(DEBUGGABLE_TYPES),$(TEST_ART_BROKEN_OPTIMIZING_MIPS64_64BIT_RUN_TESTS),64)
+  endif
+endif
+
+TEST_ART_BROKEN_OPTIMIZING_MIPS64_64BIT_RUN_TESTS :=
+
+# Known broken tests for the MIPS64 optimizing compiler backend in 32-bit mode.  b/21555893
+TEST_ART_BROKEN_OPTIMIZING_MIPS64_32BIT_RUN_TESTS := \
+  496-checker-inlining-and-class-loader
+
+ifeq ($(TARGET_ARCH),mips64)
+  ifneq (,$(filter optimizing,$(COMPILER_TYPES)))
+    ART_TEST_KNOWN_BROKEN += $(call all-run-test-names,target,$(RUN_TYPES),$(PREBUILD_TYPES), \
+        optimizing,$(RELOCATE_TYPES),$(TRACE_TYPES),$(GC_TYPES),$(JNI_TYPES), \
+        $(IMAGE_TYPES),$(PICTEST_TYPES),$(DEBUGGABLE_TYPES),$(TEST_ART_BROKEN_OPTIMIZING_MIPS64_32BIT_RUN_TESTS),32)
+  endif
+endif
+
+TEST_ART_BROKEN_OPTIMIZING_MIPS64_32BIT_RUN_TESTS :=
+
 # Known broken tests for the optimizing compiler.
 TEST_ART_BROKEN_OPTIMIZING_RUN_TESTS :=