Merge "Test case for getting annotations of renamed enums."
diff --git a/compiler/Android.mk b/compiler/Android.mk
index b05f479..564bd7e 100644
--- a/compiler/Android.mk
+++ b/compiler/Android.mk
@@ -81,6 +81,7 @@
 	optimizing/load_store_elimination.cc \
 	optimizing/locations.cc \
 	optimizing/nodes.cc \
+	optimizing/nodes_arm64.cc \
 	optimizing/optimization.cc \
 	optimizing/optimizing_compiler.cc \
 	optimizing/parallel_move_resolver.cc \
diff --git a/compiler/dwarf/debug_info_entry_writer.h b/compiler/dwarf/debug_info_entry_writer.h
index aa31036..a551e4b 100644
--- a/compiler/dwarf/debug_info_entry_writer.h
+++ b/compiler/dwarf/debug_info_entry_writer.h
@@ -112,6 +112,12 @@
     this->PushData(ptr, num_bytes);
   }
 
+  void WriteExprLoc(Attribute attrib, const void* ptr, size_t num_bytes) {
+    AddAbbrevAttribute(attrib, DW_FORM_exprloc);
+    this->PushUleb128(dchecked_integral_cast<uint32_t>(num_bytes));
+    this->PushData(ptr, num_bytes);
+  }
+
   void WriteData1(Attribute attrib, uint8_t value) {
     AddAbbrevAttribute(attrib, DW_FORM_data1);
     this->PushUint8(value);
@@ -132,6 +138,11 @@
     this->PushUint64(value);
   }
 
+  void WriteSecOffset(Attribute attrib, uint32_t offset) {
+    AddAbbrevAttribute(attrib, DW_FORM_sec_offset);
+    this->PushUint32(offset);
+  }
+
   void WriteSdata(Attribute attrib, int value) {
     AddAbbrevAttribute(attrib, DW_FORM_sdata);
     this->PushSleb128(value);
diff --git a/compiler/dwarf/headers.h b/compiler/dwarf/headers.h
index c75aeac..f76f76f 100644
--- a/compiler/dwarf/headers.h
+++ b/compiler/dwarf/headers.h
@@ -134,7 +134,7 @@
   Writer<> writer(debug_info);
   size_t start = writer.data()->size();
   writer.PushUint32(0);  // Length placeholder.
-  writer.PushUint16(3);  // Version.
+  writer.PushUint16(4);  // Version.
   writer.PushUint32(debug_abbrev_offset);
   writer.PushUint8(entries.Is64bit() ? 8 : 4);
   size_t entries_offset = writer.data()->size();
@@ -167,10 +167,7 @@
   Writer<> writer(debug_line);
   size_t header_start = writer.data()->size();
   writer.PushUint32(0);  // Section-length placeholder.
-  // Claim DWARF-2 version even though we use some DWARF-3 features.
-  // DWARF-2 consumers will ignore the unknown opcodes.
-  // This is what clang currently does.
-  writer.PushUint16(2);  // .debug_line version.
+  writer.PushUint16(3);  // .debug_line version.
   size_t header_length_pos = writer.data()->size();
   writer.PushUint32(0);  // Header-length placeholder.
   writer.PushUint8(1 << opcodes.GetCodeFactorBits());
diff --git a/compiler/dwarf/register.h b/compiler/dwarf/register.h
index 7045237..b67e8dd 100644
--- a/compiler/dwarf/register.h
+++ b/compiler/dwarf/register.h
@@ -35,9 +35,10 @@
   //   Arm64 mapping is correct since we already do this there.
   //   libunwind might struggle with the new mapping as well.
 
-  static Reg ArmCore(int num) { return Reg(num); }
+  static Reg ArmCore(int num) { return Reg(num); }  // R0-R15.
   static Reg ArmFp(int num) { return Reg(64 + num); }  // S0–S31.
-  static Reg Arm64Core(int num) { return Reg(num); }
+  static Reg ArmDp(int num) { return Reg(256 + num); }  // D0–D31.
+  static Reg Arm64Core(int num) { return Reg(num); }  // X0-X31.
   static Reg Arm64Fp(int num) { return Reg(64 + num); }  // V0-V31.
   static Reg MipsCore(int num) { return Reg(num); }
   static Reg Mips64Core(int num) { return Reg(num); }
diff --git a/compiler/elf_writer_debug.cc b/compiler/elf_writer_debug.cc
index 5e2a8bf..81f574f 100644
--- a/compiler/elf_writer_debug.cc
+++ b/compiler/elf_writer_debug.cc
@@ -29,10 +29,49 @@
 #include "elf_builder.h"
 #include "oat_writer.h"
 #include "utils.h"
+#include "stack_map.h"
 
 namespace art {
 namespace dwarf {
 
+static Reg GetDwarfCoreReg(InstructionSet isa, int machine_reg) {
+  switch (isa) {
+    case kArm:
+    case kThumb2:
+      return Reg::ArmCore(machine_reg);
+    case kArm64:
+      return Reg::Arm64Core(machine_reg);
+    case kX86:
+      return Reg::X86Core(machine_reg);
+    case kX86_64:
+      return Reg::X86_64Core(machine_reg);
+    case kMips:
+      return Reg::MipsCore(machine_reg);
+    case kMips64:
+      return Reg::Mips64Core(machine_reg);
+    default:
+      LOG(FATAL) << "Unknown instruction set: " << isa;
+      UNREACHABLE();
+  }
+}
+
+static Reg GetDwarfFpReg(InstructionSet isa, int machine_reg) {
+  switch (isa) {
+    case kArm:
+    case kThumb2:
+      return Reg::ArmFp(machine_reg);
+    case kArm64:
+      return Reg::Arm64Fp(machine_reg);
+    case kX86:
+      return Reg::X86Fp(machine_reg);
+    case kX86_64:
+      return Reg::X86_64Fp(machine_reg);
+    default:
+      LOG(FATAL) << "Unknown instruction set: " << isa;
+      UNREACHABLE();
+  }
+}
+
 static void WriteCIE(InstructionSet isa,
                      CFIFormat format,
                      std::vector<uint8_t>* buffer) {
@@ -280,8 +319,8 @@
       info_.WriteStrp(DW_AT_producer, owner_->WriteString("Android dex2oat"));
       info_.WriteData1(DW_AT_language, DW_LANG_Java);
       info_.WriteAddr(DW_AT_low_pc, text_address + compilation_unit.low_pc_);
-      info_.WriteAddr(DW_AT_high_pc, text_address + compilation_unit.high_pc_);
-      info_.WriteData4(DW_AT_stmt_list, compilation_unit.debug_line_offset_);
+      info_.WriteUdata(DW_AT_high_pc, compilation_unit.high_pc_ - compilation_unit.low_pc_);
+      info_.WriteSecOffset(DW_AT_stmt_list, compilation_unit.debug_line_offset_);
 
       const char* last_dex_class_desc = nullptr;
       for (auto mi : compilation_unit.methods_) {
@@ -321,18 +360,41 @@
         info_.StartTag(DW_TAG_subprogram);
         WriteName(dex->GetMethodName(dex_method));
         info_.WriteAddr(DW_AT_low_pc, text_address + mi->low_pc_);
-        info_.WriteAddr(DW_AT_high_pc, text_address + mi->high_pc_);
+        info_.WriteUdata(DW_AT_high_pc, mi->high_pc_ - mi->low_pc_);
+        uint8_t frame_base[] = { DW_OP_call_frame_cfa };
+        info_.WriteExprLoc(DW_AT_frame_base, &frame_base, sizeof(frame_base));
         WriteLazyType(dex->GetReturnTypeDescriptor(dex_proto));
         if (dex_params != nullptr) {
+          uint32_t vreg = mi->code_item_ == nullptr ? 0 :
+              mi->code_item_->registers_size_ - mi->code_item_->ins_size_;
+          if ((mi->access_flags_ & kAccStatic) == 0) {
+            info_.StartTag(DW_TAG_formal_parameter);
+            WriteName("this");
+            info_.WriteFlag(DW_AT_artificial, true);
+            WriteLazyType(dex_class_desc);
+            const bool is64bitValue = false;
+            WriteRegLocation(mi, vreg, is64bitValue, compilation_unit.low_pc_);
+            vreg++;
+            info_.EndTag();
+          }
           for (uint32_t i = 0; i < dex_params->Size(); ++i) {
             info_.StartTag(DW_TAG_formal_parameter);
             // Parameter names may not be always available.
             if (i < param_names.size() && param_names[i] != nullptr) {
               WriteName(param_names[i]);
             }
-            WriteLazyType(dex->StringByTypeIdx(dex_params->GetTypeItem(i).type_idx_));
+            // Write the type.
+            const char* type_desc = dex->StringByTypeIdx(dex_params->GetTypeItem(i).type_idx_);
+            WriteLazyType(type_desc);
+            // Write the stack location of the parameter.
+            const bool is64bitValue = type_desc[0] == 'D' || type_desc[0] == 'J';
+            WriteRegLocation(mi, vreg, is64bitValue, compilation_unit.low_pc_);
+            vreg += is64bitValue ? 2 : 1;
             info_.EndTag();
           }
+          if (mi->code_item_ != nullptr) {
+            CHECK_EQ(vreg, mi->code_item_->registers_size_);
+          }
         }
         info_.EndTag();
         CHECK_EQ(info_.Depth(), start_depth);  // Balanced start/end.
@@ -352,6 +414,158 @@
       owner_->builder_->GetDebugInfo()->WriteFully(buffer.data(), buffer.size());
     }
 
+    // Write table into .debug_loc which describes location of dex register.
+    // The dex register might be valid only at some points and it might
+    // move between machine registers and stack.
+    void WriteRegLocation(const OatWriter::DebugInfo* method_info, uint16_t vreg,
+                          bool is64bitValue, uint32_t compilation_unit_low_pc) {
+      using Kind = DexRegisterLocation::Kind;
+      bool is_optimizing = method_info->compiled_method_->GetQuickCode().size() > 0 &&
+                           method_info->compiled_method_->GetVmapTable().size() > 0 &&
+                           method_info->compiled_method_->GetGcMap().size() == 0 &&
+                           method_info->code_item_ != nullptr;
+      if (!is_optimizing) {
+        return;
+      }
+
+      Writer<> writer(&owner_->debug_loc_);
+      info_.WriteSecOffset(DW_AT_location, writer.size());
+
+      const InstructionSet isa = owner_->builder_->GetIsa();
+      const bool is64bit = Is64BitInstructionSet(isa);
+      const CodeInfo code_info(method_info->compiled_method_->GetVmapTable().data());
+      const StackMapEncoding encoding = code_info.ExtractEncoding();
+      DexRegisterLocation last_reg_lo = DexRegisterLocation::None();
+      DexRegisterLocation last_reg_hi = DexRegisterLocation::None();
+      size_t offset_of_last_end_address = 0;
+      for (uint32_t s = 0; s < code_info.GetNumberOfStackMaps(); s++) {
+        StackMap stack_map = code_info.GetStackMapAt(s, encoding);
+        DCHECK(stack_map.IsValid());
+
+        // Find the location of the dex register.
+        DexRegisterLocation reg_lo = DexRegisterLocation::None();
+        DexRegisterLocation reg_hi = DexRegisterLocation::None();
+        if (stack_map.HasDexRegisterMap(encoding)) {
+          DexRegisterMap dex_register_map = code_info.GetDexRegisterMapOf(
+              stack_map, encoding, method_info->code_item_->registers_size_);
+          reg_lo = dex_register_map.GetDexRegisterLocation(
+              vreg, method_info->code_item_->registers_size_, code_info, encoding);
+          if (is64bitValue) {
+            reg_hi = dex_register_map.GetDexRegisterLocation(
+                vreg + 1, method_info->code_item_->registers_size_, code_info, encoding);
+          }
+        }
+        if ((reg_lo == last_reg_lo && reg_hi == last_reg_hi) ||
+            reg_lo.GetKind() == Kind::kNone) {
+          // Skip identical or undefined locations.
+          continue;
+        }
+        last_reg_lo = reg_lo;
+        last_reg_hi = reg_hi;
+
+        // Translate dex register location to DWARF expression.
+        // Note that 64-bit value might be split to two distinct locations.
+        // (for example, two 32-bit machine registers, or even stack and register)
+        uint8_t buffer[64];
+        uint8_t* pos = buffer;
+        for (int piece = 0; piece < (is64bitValue ? 2 : 1); piece++) {
+          DexRegisterLocation reg_loc = (piece == 0 ? reg_lo : reg_hi);
+          const Kind kind = reg_loc.GetKind();
+          const int32_t value = reg_loc.GetValue();
+          if (kind == Kind::kInStack) {
+            const size_t frame_size = method_info->compiled_method_->GetFrameSizeInBytes();
+            *(pos++) = DW_OP_fbreg;
+            // The stack offset is relative to SP. Make it relative to CFA.
+            pos = EncodeSignedLeb128(pos, value - frame_size);
+            if (piece == 0 && reg_hi.GetKind() == Kind::kInStack &&
+                reg_hi.GetValue() == value + 4) {
+              break;  // the high word is correctly implied by the low word.
+            }
+          } else if (kind == Kind::kInRegister) {
+            pos = WriteOpReg(pos, GetDwarfCoreReg(isa, value).num());
+            if (piece == 0 && reg_hi.GetKind() == Kind::kInRegisterHigh &&
+                reg_hi.GetValue() == value) {
+              break;  // the high word is correctly implied by the low word.
+            }
+          } else if (kind == Kind::kInFpuRegister) {
+            if ((isa == kArm || isa == kThumb2) &&
+                piece == 0 && reg_hi.GetKind() == Kind::kInFpuRegister &&
+                reg_hi.GetValue() == value + 1 && value % 2 == 0) {
+              // Translate S register pair to D register (e.g. S4+S5 to D2).
+              pos = WriteOpReg(pos, Reg::ArmDp(value / 2).num());
+              break;
+            }
+            if (isa == kMips || isa == kMips64) {
+              // TODO: Find what the DWARF floating point register numbers are on MIPS.
+              break;
+            }
+            pos = WriteOpReg(pos, GetDwarfFpReg(isa, value).num());
+            if (piece == 0 && reg_hi.GetKind() == Kind::kInFpuRegisterHigh &&
+                reg_hi.GetValue() == reg_lo.GetValue()) {
+              break;  // the high word is correctly implied by the low word.
+            }
+          } else if (kind == Kind::kConstant) {
+            *(pos++) = DW_OP_consts;
+            pos = EncodeSignedLeb128(pos, value);
+            *(pos++) = DW_OP_stack_value;
+          } else if (kind == Kind::kNone) {
+            break;
+          } else {
+            // kInStackLargeOffset and kConstantLargeValue are hidden by GetKind().
+            // kInRegisterHigh and kInFpuRegisterHigh should be handled by
+            // the special cases above and they should not occur alone.
+            LOG(ERROR) << "Unexpected register location kind: "
+                       << DexRegisterLocation::PrettyDescriptor(kind);
+            break;
+          }
+          if (is64bitValue) {
+            // Write the marker which is needed by split 64-bit values.
+            // This code is skipped by the special cases.
+            *(pos++) = DW_OP_piece;
+            pos = EncodeUnsignedLeb128(pos, 4);
+          }
+        }
+
+        // Write end address for previous entry.
+        const uint32_t pc = method_info->low_pc_ + stack_map.GetNativePcOffset(encoding);
+        if (offset_of_last_end_address != 0) {
+          if (is64bit) {
+            writer.UpdateUint64(offset_of_last_end_address, pc - compilation_unit_low_pc);
+          } else {
+            writer.UpdateUint32(offset_of_last_end_address, pc - compilation_unit_low_pc);
+          }
+        }
+        offset_of_last_end_address = 0;
+
+        DCHECK_LE(static_cast<size_t>(pos - buffer), sizeof(buffer));
+        if (pos > buffer) {
+          // Write start/end address.
+          if (is64bit) {
+            writer.PushUint64(pc - compilation_unit_low_pc);
+            offset_of_last_end_address = writer.size();
+            writer.PushUint64(method_info->high_pc_ - compilation_unit_low_pc);
+          } else {
+            writer.PushUint32(pc - compilation_unit_low_pc);
+            offset_of_last_end_address = writer.size();
+            writer.PushUint32(method_info->high_pc_ - compilation_unit_low_pc);
+          }
+          // Write the expression.
+          writer.PushUint16(pos - buffer);
+          writer.PushData(buffer, pos - buffer);
+        } else {
+          // Otherwise leave the address range undefined.
+        }
+      }
+      // Write end-of-list entry.
+      if (is64bit) {
+        writer.PushUint64(0);
+        writer.PushUint64(0);
+      } else {
+        writer.PushUint32(0);
+        writer.PushUint32(0);
+      }
+    }
+
     // Some types are difficult to define as we go since they need
     // to be enclosed in the right set of namespaces. Therefore we
     // just define all types lazily at the end of compilation unit.
@@ -375,6 +589,17 @@
       info_.WriteStrp(DW_AT_name, owner_->WriteString(name));
     }
 
+    // Helper which writes DWARF expression referencing a register.
+    static uint8_t* WriteOpReg(uint8_t* buffer, uint32_t dwarf_reg_num) {
+      if (dwarf_reg_num < 32) {
+        *(buffer++) = DW_OP_reg0 + dwarf_reg_num;
+      } else {
+        *(buffer++) = DW_OP_regx;
+        buffer = EncodeUnsignedLeb128(buffer, dwarf_reg_num);
+      }
+      return buffer;
+    }
+
     // Convert dex type descriptor to DWARF.
     // Returns offset in the compilation unit.
     size_t WriteType(const char* desc) {
@@ -398,22 +623,60 @@
       } else {
         // Primitive types.
         const char* name;
+        uint32_t encoding;
+        uint32_t byte_size;
         switch (*desc) {
-        case 'B': name = "byte"; break;
-        case 'C': name = "char"; break;
-        case 'D': name = "double"; break;
-        case 'F': name = "float"; break;
-        case 'I': name = "int"; break;
-        case 'J': name = "long"; break;
-        case 'S': name = "short"; break;
-        case 'Z': name = "boolean"; break;
-        case 'V': name = "void"; break;
+        case 'B':
+          name = "byte";
+          encoding = DW_ATE_signed;
+          byte_size = 1;
+          break;
+        case 'C':
+          name = "char";
+          encoding = DW_ATE_UTF;
+          byte_size = 2;
+          break;
+        case 'D':
+          name = "double";
+          encoding = DW_ATE_float;
+          byte_size = 8;
+          break;
+        case 'F':
+          name = "float";
+          encoding = DW_ATE_float;
+          byte_size = 4;
+          break;
+        case 'I':
+          name = "int";
+          encoding = DW_ATE_signed;
+          byte_size = 4;
+          break;
+        case 'J':
+          name = "long";
+          encoding = DW_ATE_signed;
+          byte_size = 8;
+          break;
+        case 'S':
+          name = "short";
+          encoding = DW_ATE_signed;
+          byte_size = 2;
+          break;
+        case 'Z':
+          name = "boolean";
+          encoding = DW_ATE_boolean;
+          byte_size = 1;
+          break;
+        case 'V':
+          LOG(FATAL) << "Void type should not be encoded";
+          UNREACHABLE();
         default:
           LOG(FATAL) << "Unknown dex type descriptor: " << desc;
           UNREACHABLE();
         }
         offset = info_.StartTag(DW_TAG_base_type);
         WriteName(name);
+        info_.WriteData1(DW_AT_encoding, encoding);
+        info_.WriteData1(DW_AT_byte_size, byte_size);
         info_.EndTag();
       }
 
@@ -480,6 +743,7 @@
     builder_->WritePatches(".debug_info.oat_patches", &debug_info_patches_);
     builder_->WriteSection(".debug_abbrev", &debug_abbrev_.Data());
     builder_->WriteSection(".debug_str", &debug_str_.Data());
+    builder_->WriteSection(".debug_loc", &debug_loc_);
   }
 
  private:
@@ -491,6 +755,7 @@
   std::vector<uintptr_t> debug_info_patches_;
   DedupVector debug_abbrev_;
   DedupVector debug_str_;
+  std::vector<uint8_t> debug_loc_;
 
   std::unordered_set<const char*> defined_dex_classes_;  // For CHECKs only.
 };
diff --git a/compiler/image_writer.cc b/compiler/image_writer.cc
index 9cac8b4..341742e 100644
--- a/compiler/image_writer.cc
+++ b/compiler/image_writer.cc
@@ -592,6 +592,17 @@
 }
 
 bool ImageWriter::ContainsBootClassLoaderNonImageClass(mirror::Class* klass) {
+  bool early_exit = false;
+  std::unordered_set<mirror::Class*> visited;
+  return ContainsBootClassLoaderNonImageClassInternal(klass, &early_exit, &visited);
+}
+
+bool ImageWriter::ContainsBootClassLoaderNonImageClassInternal(
+    mirror::Class* klass,
+    bool* early_exit,
+    std::unordered_set<mirror::Class*>* visited) {
+  DCHECK(early_exit != nullptr);
+  DCHECK(visited != nullptr);
   if (klass == nullptr) {
     return false;
   }
@@ -600,14 +611,22 @@
     // Already computed, return the found value.
     return found->second;
   }
-  // Place holder value to prevent infinite recursion.
-  prune_class_memo_.emplace(klass, false);
+  // Circular dependencies, return false but do not store the result in the memoization table.
+  if (visited->find(klass) != visited->end()) {
+    *early_exit = true;
+    return false;
+  }
+  visited->emplace(klass);
   bool result = IsBootClassLoaderNonImageClass(klass);
+  bool my_early_exit = false;  // Only for ourselves, ignore caller.
   if (!result) {
     // Check interfaces since these wont be visited through VisitReferences.)
     mirror::IfTable* if_table = klass->GetIfTable();
     for (size_t i = 0, num_interfaces = klass->GetIfTableCount(); i < num_interfaces; ++i) {
-      result = result || ContainsBootClassLoaderNonImageClass(if_table->GetInterface(i));
+      result = result || ContainsBootClassLoaderNonImageClassInternal(
+          if_table->GetInterface(i),
+          &my_early_exit,
+          visited);
     }
   }
   // Check static fields and their classes.
@@ -621,16 +640,38 @@
       mirror::Object* ref = klass->GetFieldObject<mirror::Object>(field_offset);
       if (ref != nullptr) {
         if (ref->IsClass()) {
-          result = result || ContainsBootClassLoaderNonImageClass(ref->AsClass());
+          result = result ||
+                   ContainsBootClassLoaderNonImageClassInternal(
+                       ref->AsClass(),
+                       &my_early_exit,
+                       visited);
         }
-        result = result || ContainsBootClassLoaderNonImageClass(ref->GetClass());
+        result = result ||
+                 ContainsBootClassLoaderNonImageClassInternal(
+                     ref->GetClass(),
+                     &my_early_exit,
+                     visited);
       }
       field_offset = MemberOffset(field_offset.Uint32Value() +
                                   sizeof(mirror::HeapReference<mirror::Object>));
     }
   }
-  result = result || ContainsBootClassLoaderNonImageClass(klass->GetSuperClass());
-  prune_class_memo_[klass] = result;
+  result = result ||
+           ContainsBootClassLoaderNonImageClassInternal(
+               klass->GetSuperClass(),
+               &my_early_exit,
+               visited);
+  // Erase the element we stored earlier since we are exiting the function.
+  auto it = visited->find(klass);
+  DCHECK(it != visited->end());
+  visited->erase(it);
+  // Only store result if it is true or none of the calls early exited due to circular
+  // dependencies. If visited is empty then we are the root caller, in this case the cycle was in
+  // a child call and we can remember the result.
+  if (result == true || !my_early_exit || visited->empty()) {
+    prune_class_memo_[klass] = result;
+  }
+  *early_exit |= my_early_exit;
   return result;
 }
 
diff --git a/compiler/image_writer.h b/compiler/image_writer.h
index 22cb91a..889cd10 100644
--- a/compiler/image_writer.h
+++ b/compiler/image_writer.h
@@ -343,6 +343,12 @@
   bool ContainsBootClassLoaderNonImageClass(mirror::Class* klass)
       SHARED_REQUIRES(Locks::mutator_lock_);
 
+  // early_exit is true if we had a cyclic dependency anywhere down the chain.
+  bool ContainsBootClassLoaderNonImageClassInternal(mirror::Class* klass,
+                                                    bool* early_exit,
+                                                    std::unordered_set<mirror::Class*>* visited)
+      SHARED_REQUIRES(Locks::mutator_lock_);
+
   static Bin BinTypeForNativeRelocationType(NativeObjectRelocationType type);
 
   uintptr_t NativeOffsetInImage(void* obj);
diff --git a/compiler/optimizing/code_generator_arm64.cc b/compiler/optimizing/code_generator_arm64.cc
index db2defb..ac16268 100644
--- a/compiler/optimizing/code_generator_arm64.cc
+++ b/compiler/optimizing/code_generator_arm64.cc
@@ -1889,6 +1889,82 @@
   HandleBinaryOp(instruction);
 }
 
+void LocationsBuilderARM64::VisitArm64DataProcWithShifterOp(
+    HArm64DataProcWithShifterOp* instruction) {
+  DCHECK(instruction->GetType() == Primitive::kPrimInt ||
+         instruction->GetType() == Primitive::kPrimLong);
+  LocationSummary* locations =
+      new (GetGraph()->GetArena()) LocationSummary(instruction, LocationSummary::kNoCall);
+  if (instruction->GetInstrKind() == HInstruction::kNeg) {
+    locations->SetInAt(0, Location::ConstantLocation(instruction->InputAt(0)->AsConstant()));
+  } else {
+    locations->SetInAt(0, Location::RequiresRegister());
+  }
+  locations->SetInAt(1, Location::RequiresRegister());
+  locations->SetOut(Location::RequiresRegister(), Location::kNoOutputOverlap);
+}
+
+void InstructionCodeGeneratorARM64::VisitArm64DataProcWithShifterOp(
+    HArm64DataProcWithShifterOp* instruction) {
+  Primitive::Type type = instruction->GetType();
+  HInstruction::InstructionKind kind = instruction->GetInstrKind();
+  DCHECK(type == Primitive::kPrimInt || type == Primitive::kPrimLong);
+  Register out = OutputRegister(instruction);
+  Register left;
+  if (kind != HInstruction::kNeg) {
+    left = InputRegisterAt(instruction, 0);
+  }
+  // If this `HArm64DataProcWithShifterOp` was created by merging a type conversion as the
+  // shifter operand operation, the IR generating `right_reg` (input to the type
+  // conversion) can have a different type from the current instruction's type,
+  // so we manually indicate the type.
+  Register right_reg = RegisterFrom(instruction->GetLocations()->InAt(1), type);
+  int64_t shift_amount = (type == Primitive::kPrimInt)
+    ? static_cast<uint32_t>(instruction->GetShiftAmount() & kMaxIntShiftValue)
+    : static_cast<uint32_t>(instruction->GetShiftAmount() & kMaxLongShiftValue);
+
+  Operand right_operand(0);
+
+  HArm64DataProcWithShifterOp::OpKind op_kind = instruction->GetOpKind();
+  if (HArm64DataProcWithShifterOp::IsExtensionOp(op_kind)) {
+    right_operand = Operand(right_reg, helpers::ExtendFromOpKind(op_kind));
+  } else {
+    right_operand = Operand(right_reg, helpers::ShiftFromOpKind(op_kind), shift_amount);
+  }
+
+  // Logical binary operations do not support extension operations in the
+  // operand. Note that VIXL would still manage if it was passed by generating
+  // the extension as a separate instruction.
+  // `HNeg` also does not support extension. See comments in `ShifterOperandSupportsExtension()`.
+  DCHECK(!right_operand.IsExtendedRegister() ||
+         (kind != HInstruction::kAnd && kind != HInstruction::kOr && kind != HInstruction::kXor &&
+          kind != HInstruction::kNeg));
+  switch (kind) {
+    case HInstruction::kAdd:
+      __ Add(out, left, right_operand);
+      break;
+    case HInstruction::kAnd:
+      __ And(out, left, right_operand);
+      break;
+    case HInstruction::kNeg:
+      DCHECK(instruction->InputAt(0)->AsConstant()->IsZero());
+      __ Neg(out, right_operand);
+      break;
+    case HInstruction::kOr:
+      __ Orr(out, left, right_operand);
+      break;
+    case HInstruction::kSub:
+      __ Sub(out, left, right_operand);
+      break;
+    case HInstruction::kXor:
+      __ Eor(out, left, right_operand);
+      break;
+    default:
+      LOG(FATAL) << "Unexpected operation kind: " << kind;
+      UNREACHABLE();
+  }
+}
+
 void LocationsBuilderARM64::VisitArm64IntermediateAddress(HArm64IntermediateAddress* instruction) {
   LocationSummary* locations =
       new (GetGraph()->GetArena()) LocationSummary(instruction, LocationSummary::kNoCall);
@@ -4389,9 +4465,7 @@
     int min_size = std::min(result_size, input_size);
     Register output = OutputRegister(conversion);
     Register source = InputRegisterAt(conversion, 0);
-    if ((result_type == Primitive::kPrimChar) && (input_size < result_size)) {
-      __ Ubfx(output, source, 0, result_size * kBitsPerByte);
-    } else if (result_type == Primitive::kPrimInt && input_type == Primitive::kPrimLong) {
+    if (result_type == Primitive::kPrimInt && input_type == Primitive::kPrimLong) {
       // 'int' values are used directly as W registers, discarding the top
       // bits, so we don't need to sign-extend and can just perform a move.
       // We do not pass the `kDiscardForSameWReg` argument to force clearing the
@@ -4400,9 +4474,11 @@
       // 32bit input value as a 64bit value assuming that the top 32 bits are
       // zero.
       __ Mov(output.W(), source.W());
-    } else if ((result_type == Primitive::kPrimChar) ||
-               ((input_type == Primitive::kPrimChar) && (result_size > input_size))) {
-      __ Ubfx(output, output.IsX() ? source.X() : source.W(), 0, min_size * kBitsPerByte);
+    } else if (result_type == Primitive::kPrimChar ||
+               (input_type == Primitive::kPrimChar && input_size < result_size)) {
+      __ Ubfx(output,
+              output.IsX() ? source.X() : source.W(),
+              0, Primitive::ComponentSize(Primitive::kPrimChar) * kBitsPerByte);
     } else {
       __ Sbfx(output, output.IsX() ? source.X() : source.W(), 0, min_size * kBitsPerByte);
     }
diff --git a/compiler/optimizing/common_arm64.h b/compiler/optimizing/common_arm64.h
index e1a8c9c..af8b8b5 100644
--- a/compiler/optimizing/common_arm64.h
+++ b/compiler/optimizing/common_arm64.h
@@ -17,6 +17,7 @@
 #ifndef ART_COMPILER_OPTIMIZING_COMMON_ARM64_H_
 #define ART_COMPILER_OPTIMIZING_COMMON_ARM64_H_
 
+#include "code_generator.h"
 #include "locations.h"
 #include "nodes.h"
 #include "utils/arm64/assembler_arm64.h"
@@ -255,6 +256,67 @@
   return true;
 }
 
+static inline vixl::Shift ShiftFromOpKind(HArm64DataProcWithShifterOp::OpKind op_kind) {
+  switch (op_kind) {
+    case HArm64DataProcWithShifterOp::kASR: return vixl::ASR;
+    case HArm64DataProcWithShifterOp::kLSL: return vixl::LSL;
+    case HArm64DataProcWithShifterOp::kLSR: return vixl::LSR;
+    default:
+      LOG(FATAL) << "Unexpected op kind " << op_kind;
+      UNREACHABLE();
+      return vixl::NO_SHIFT;
+  }
+}
+
+static inline vixl::Extend ExtendFromOpKind(HArm64DataProcWithShifterOp::OpKind op_kind) {
+  switch (op_kind) {
+    case HArm64DataProcWithShifterOp::kUXTB: return vixl::UXTB;
+    case HArm64DataProcWithShifterOp::kUXTH: return vixl::UXTH;
+    case HArm64DataProcWithShifterOp::kUXTW: return vixl::UXTW;
+    case HArm64DataProcWithShifterOp::kSXTB: return vixl::SXTB;
+    case HArm64DataProcWithShifterOp::kSXTH: return vixl::SXTH;
+    case HArm64DataProcWithShifterOp::kSXTW: return vixl::SXTW;
+    default:
+      LOG(FATAL) << "Unexpected op kind " << op_kind;
+      UNREACHABLE();
+      return vixl::NO_EXTEND;
+  }
+}
+
+static inline bool CanFitInShifterOperand(HInstruction* instruction) {
+  if (instruction->IsTypeConversion()) {
+    HTypeConversion* conversion = instruction->AsTypeConversion();
+    Primitive::Type result_type = conversion->GetResultType();
+    Primitive::Type input_type = conversion->GetInputType();
+    // We don't expect to see the same type as input and result.
+    return Primitive::IsIntegralType(result_type) && Primitive::IsIntegralType(input_type) &&
+        (result_type != input_type);
+  } else {
+    return (instruction->IsShl() && instruction->AsShl()->InputAt(1)->IsIntConstant()) ||
+        (instruction->IsShr() && instruction->AsShr()->InputAt(1)->IsIntConstant()) ||
+        (instruction->IsUShr() && instruction->AsUShr()->InputAt(1)->IsIntConstant());
+  }
+}
+
+static inline bool HasShifterOperand(HInstruction* instr) {
+  // `neg` instructions are an alias of `sub` using the zero register as the
+  // first register input.
+  bool res = instr->IsAdd() || instr->IsAnd() || instr->IsNeg() ||
+      instr->IsOr() || instr->IsSub() || instr->IsXor();
+  return res;
+}
+
+static inline bool ShifterOperandSupportsExtension(HInstruction* instruction) {
+  DCHECK(HasShifterOperand(instruction));
+  // Although the `neg` instruction is an alias of the `sub` instruction, `HNeg`
+  // does *not* support extension. This is because the `extended register` form
+  // of the `sub` instruction interprets the left register with code 31 as the
+  // stack pointer and not the zero register. (So does the `immediate` form.) In
+  // the other form `shifted register, the register with code 31 is interpreted
+  // as the zero register.
+  return instruction->IsAdd() || instruction->IsSub();
+}
+
 }  // namespace helpers
 }  // namespace arm64
 }  // namespace art
diff --git a/compiler/optimizing/graph_checker.cc b/compiler/optimizing/graph_checker.cc
index 5814d75..b3b09d2 100644
--- a/compiler/optimizing/graph_checker.cc
+++ b/compiler/optimizing/graph_checker.cc
@@ -735,26 +735,29 @@
     }
   }
 
-  // Test phi equivalents. There should not be two of the same type and they
-  // should only be created for constants which were untyped in DEX.
-  for (HInstructionIterator phi_it(phi->GetBlock()->GetPhis()); !phi_it.Done(); phi_it.Advance()) {
-    HPhi* other_phi = phi_it.Current()->AsPhi();
-    if (phi != other_phi && phi->GetRegNumber() == other_phi->GetRegNumber()) {
-      if (phi->GetType() == other_phi->GetType()) {
-        std::stringstream type_str;
-        type_str << phi->GetType();
-        AddError(StringPrintf("Equivalent phi (%d) found for VReg %d with type: %s.",
-                              phi->GetId(),
-                              phi->GetRegNumber(),
-                              type_str.str().c_str()));
-      } else {
-        ArenaBitVector visited(GetGraph()->GetArena(), 0, /* expandable */ true);
-        if (!IsConstantEquivalent(phi, other_phi, &visited)) {
-          AddError(StringPrintf("Two phis (%d and %d) found for VReg %d but they "
-                                "are not equivalents of constants.",
+  // Test phi equivalents. There should not be two of the same type and they should only be
+  // created for constants which were untyped in DEX. Note that this test can be skipped for
+  // a synthetic phi (indicated by lack of a virtual register).
+  if (phi->GetRegNumber() != kNoRegNumber) {
+    for (HInstructionIterator phi_it(phi->GetBlock()->GetPhis()); !phi_it.Done(); phi_it.Advance()) {
+      HPhi* other_phi = phi_it.Current()->AsPhi();
+      if (phi != other_phi && phi->GetRegNumber() == other_phi->GetRegNumber()) {
+        if (phi->GetType() == other_phi->GetType()) {
+          std::stringstream type_str;
+          type_str << phi->GetType();
+          AddError(StringPrintf("Equivalent phi (%d) found for VReg %d with type: %s.",
                                 phi->GetId(),
-                                other_phi->GetId(),
-                                phi->GetRegNumber()));
+                                phi->GetRegNumber(),
+                                type_str.str().c_str()));
+        } else {
+          ArenaBitVector visited(GetGraph()->GetArena(), 0, /* expandable */ true);
+          if (!IsConstantEquivalent(phi, other_phi, &visited)) {
+            AddError(StringPrintf("Two phis (%d and %d) found for VReg %d but they "
+                                  "are not equivalents of constants.",
+                                  phi->GetId(),
+                                  other_phi->GetId(),
+                                  phi->GetRegNumber()));
+          }
         }
       }
     }
diff --git a/compiler/optimizing/graph_visualizer.cc b/compiler/optimizing/graph_visualizer.cc
index 4438190..48bcd10 100644
--- a/compiler/optimizing/graph_visualizer.cc
+++ b/compiler/optimizing/graph_visualizer.cc
@@ -423,6 +423,13 @@
   }
 
 #ifdef ART_ENABLE_CODEGEN_arm64
+  void VisitArm64DataProcWithShifterOp(HArm64DataProcWithShifterOp* instruction) OVERRIDE {
+    StartAttributeStream("kind") << instruction->GetInstrKind() << "+" << instruction->GetOpKind();
+    if (HArm64DataProcWithShifterOp::IsShiftOp(instruction->GetOpKind())) {
+      StartAttributeStream("shift") << instruction->GetShiftAmount();
+    }
+  }
+
   void VisitArm64MultiplyAccumulate(HArm64MultiplyAccumulate* instruction) OVERRIDE {
     StartAttributeStream("kind") << instruction->GetOpKind();
   }
diff --git a/compiler/optimizing/instruction_simplifier_arm64.cc b/compiler/optimizing/instruction_simplifier_arm64.cc
index 54dd2cc..6a34b13 100644
--- a/compiler/optimizing/instruction_simplifier_arm64.cc
+++ b/compiler/optimizing/instruction_simplifier_arm64.cc
@@ -16,11 +16,16 @@
 
 #include "instruction_simplifier_arm64.h"
 
+#include "common_arm64.h"
 #include "mirror/array-inl.h"
 
 namespace art {
 namespace arm64 {
 
+using helpers::CanFitInShifterOperand;
+using helpers::HasShifterOperand;
+using helpers::ShifterOperandSupportsExtension;
+
 void InstructionSimplifierArm64Visitor::TryExtractArrayAccessAddress(HInstruction* access,
                                                                      HInstruction* array,
                                                                      HInstruction* index,
@@ -62,6 +67,108 @@
   RecordSimplification();
 }
 
+bool InstructionSimplifierArm64Visitor::TryMergeIntoShifterOperand(HInstruction* use,
+                                                                   HInstruction* bitfield_op,
+                                                                   bool do_merge) {
+  DCHECK(HasShifterOperand(use));
+  DCHECK(use->IsBinaryOperation() || use->IsNeg());
+  DCHECK(CanFitInShifterOperand(bitfield_op));
+  DCHECK(!bitfield_op->HasEnvironmentUses());
+
+  Primitive::Type type = use->GetType();
+  if (type != Primitive::kPrimInt && type != Primitive::kPrimLong) {
+    return false;
+  }
+
+  HInstruction* left;
+  HInstruction* right;
+  if (use->IsBinaryOperation()) {
+    left = use->InputAt(0);
+    right = use->InputAt(1);
+  } else {
+    DCHECK(use->IsNeg());
+    right = use->AsNeg()->InputAt(0);
+    left = GetGraph()->GetConstant(right->GetType(), 0);
+  }
+  DCHECK(left == bitfield_op || right == bitfield_op);
+
+  if (left == right) {
+    // TODO: Handle special transformations in this situation?
+    // For example should we transform `(x << 1) + (x << 1)` into `(x << 2)`?
+    // Or should this be part of a separate transformation logic?
+    return false;
+  }
+
+  bool is_commutative = use->IsBinaryOperation() && use->AsBinaryOperation()->IsCommutative();
+  HInstruction* other_input;
+  if (bitfield_op == right) {
+    other_input = left;
+  } else {
+    if (is_commutative) {
+      other_input = right;
+    } else {
+      return false;
+    }
+  }
+
+  HArm64DataProcWithShifterOp::OpKind op_kind;
+  int shift_amount = 0;
+  HArm64DataProcWithShifterOp::GetOpInfoFromInstruction(bitfield_op, &op_kind, &shift_amount);
+
+  if (HArm64DataProcWithShifterOp::IsExtensionOp(op_kind) &&
+      !ShifterOperandSupportsExtension(use)) {
+    return false;
+  }
+
+  if (do_merge) {
+    HArm64DataProcWithShifterOp* alu_with_op =
+        new (GetGraph()->GetArena()) HArm64DataProcWithShifterOp(use,
+                                                                 other_input,
+                                                                 bitfield_op->InputAt(0),
+                                                                 op_kind,
+                                                                 shift_amount,
+                                                                 use->GetDexPc());
+    use->GetBlock()->ReplaceAndRemoveInstructionWith(use, alu_with_op);
+    if (bitfield_op->GetUses().IsEmpty()) {
+      bitfield_op->GetBlock()->RemoveInstruction(bitfield_op);
+    }
+    RecordSimplification();
+  }
+
+  return true;
+}
+
+// Merge a bitfield move instruction into its uses if it can be merged in all of them.
+bool InstructionSimplifierArm64Visitor::TryMergeIntoUsersShifterOperand(HInstruction* bitfield_op) {
+  DCHECK(CanFitInShifterOperand(bitfield_op));
+
+  if (bitfield_op->HasEnvironmentUses()) {
+    return false;
+  }
+
+  const HUseList<HInstruction*>& uses = bitfield_op->GetUses();
+
+  // Check whether we can merge the instruction in all its users' shifter operand.
+  for (HUseIterator<HInstruction*> it_use(uses); !it_use.Done(); it_use.Advance()) {
+    HInstruction* use = it_use.Current()->GetUser();
+    if (!HasShifterOperand(use)) {
+      return false;
+    }
+    if (!CanMergeIntoShifterOperand(use, bitfield_op)) {
+      return false;
+    }
+  }
+
+  // Merge the instruction into its uses.
+  for (HUseIterator<HInstruction*> it_use(uses); !it_use.Done(); it_use.Advance()) {
+    HInstruction* use = it_use.Current()->GetUser();
+    bool merged = MergeIntoShifterOperand(use, bitfield_op);
+    DCHECK(merged);
+  }
+
+  return true;
+}
+
 bool InstructionSimplifierArm64Visitor::TrySimpleMultiplyAccumulatePatterns(
     HMul* mul, HBinaryOperation* input_binop, HInstruction* input_other) {
   DCHECK(Primitive::IsIntOrLongType(mul->GetType()));
@@ -210,5 +317,37 @@
   }
 }
 
+void InstructionSimplifierArm64Visitor::VisitShl(HShl* instruction) {
+  if (instruction->InputAt(1)->IsConstant()) {
+    TryMergeIntoUsersShifterOperand(instruction);
+  }
+}
+
+void InstructionSimplifierArm64Visitor::VisitShr(HShr* instruction) {
+  if (instruction->InputAt(1)->IsConstant()) {
+    TryMergeIntoUsersShifterOperand(instruction);
+  }
+}
+
+void InstructionSimplifierArm64Visitor::VisitTypeConversion(HTypeConversion* instruction) {
+  Primitive::Type result_type = instruction->GetResultType();
+  Primitive::Type input_type = instruction->GetInputType();
+
+  if (input_type == result_type) {
+    // We let the arch-independent code handle this.
+    return;
+  }
+
+  if (Primitive::IsIntegralType(result_type) && Primitive::IsIntegralType(input_type)) {
+    TryMergeIntoUsersShifterOperand(instruction);
+  }
+}
+
+void InstructionSimplifierArm64Visitor::VisitUShr(HUShr* instruction) {
+  if (instruction->InputAt(1)->IsConstant()) {
+    TryMergeIntoUsersShifterOperand(instruction);
+  }
+}
+
 }  // namespace arm64
 }  // namespace art
diff --git a/compiler/optimizing/instruction_simplifier_arm64.h b/compiler/optimizing/instruction_simplifier_arm64.h
index eed2276..b7f490b 100644
--- a/compiler/optimizing/instruction_simplifier_arm64.h
+++ b/compiler/optimizing/instruction_simplifier_arm64.h
@@ -39,6 +39,17 @@
                                     HInstruction* array,
                                     HInstruction* index,
                                     int access_size);
+  bool TryMergeIntoUsersShifterOperand(HInstruction* instruction);
+  bool TryMergeIntoShifterOperand(HInstruction* use,
+                                  HInstruction* bitfield_op,
+                                  bool do_merge);
+  bool CanMergeIntoShifterOperand(HInstruction* use, HInstruction* bitfield_op) {
+    return TryMergeIntoShifterOperand(use, bitfield_op, false);
+  }
+  bool MergeIntoShifterOperand(HInstruction* use, HInstruction* bitfield_op) {
+    DCHECK(CanMergeIntoShifterOperand(use, bitfield_op));
+    return TryMergeIntoShifterOperand(use, bitfield_op, true);
+  }
 
   bool TrySimpleMultiplyAccumulatePatterns(HMul* mul,
                                            HBinaryOperation* input_binop,
@@ -48,6 +59,10 @@
   void VisitArrayGet(HArrayGet* instruction) OVERRIDE;
   void VisitArraySet(HArraySet* instruction) OVERRIDE;
   void VisitMul(HMul* instruction) OVERRIDE;
+  void VisitShl(HShl* instruction) OVERRIDE;
+  void VisitShr(HShr* instruction) OVERRIDE;
+  void VisitTypeConversion(HTypeConversion* instruction) OVERRIDE;
+  void VisitUShr(HUShr* instruction) OVERRIDE;
 
   OptimizingCompilerStats* stats_;
 };
diff --git a/compiler/optimizing/nodes.cc b/compiler/optimizing/nodes.cc
index 890598d..b5ac773 100644
--- a/compiler/optimizing/nodes.cc
+++ b/compiler/optimizing/nodes.cc
@@ -1890,7 +1890,7 @@
  *             |
  *          if_block
  *           /    \
- *  dummy_block   deopt_block
+ *  true_block   false_block
  *           \    /
  *       new_pre_header
  *             |
@@ -1898,62 +1898,73 @@
  */
 void HGraph::TransformLoopHeaderForBCE(HBasicBlock* header) {
   DCHECK(header->IsLoopHeader());
-  HBasicBlock* pre_header = header->GetDominator();
+  HBasicBlock* old_pre_header = header->GetDominator();
 
-  // Need this to avoid critical edge.
+  // Need extra block to avoid critical edge.
   HBasicBlock* if_block = new (arena_) HBasicBlock(this, header->GetDexPc());
-  // Need this to avoid critical edge.
-  HBasicBlock* dummy_block = new (arena_) HBasicBlock(this, header->GetDexPc());
-  HBasicBlock* deopt_block = new (arena_) HBasicBlock(this, header->GetDexPc());
+  HBasicBlock* true_block = new (arena_) HBasicBlock(this, header->GetDexPc());
+  HBasicBlock* false_block = new (arena_) HBasicBlock(this, header->GetDexPc());
   HBasicBlock* new_pre_header = new (arena_) HBasicBlock(this, header->GetDexPc());
   AddBlock(if_block);
-  AddBlock(dummy_block);
-  AddBlock(deopt_block);
+  AddBlock(true_block);
+  AddBlock(false_block);
   AddBlock(new_pre_header);
 
-  header->ReplacePredecessor(pre_header, new_pre_header);
-  pre_header->successors_.clear();
-  pre_header->dominated_blocks_.clear();
+  header->ReplacePredecessor(old_pre_header, new_pre_header);
+  old_pre_header->successors_.clear();
+  old_pre_header->dominated_blocks_.clear();
 
-  pre_header->AddSuccessor(if_block);
-  if_block->AddSuccessor(dummy_block);  // True successor
-  if_block->AddSuccessor(deopt_block);  // False successor
-  dummy_block->AddSuccessor(new_pre_header);
-  deopt_block->AddSuccessor(new_pre_header);
+  old_pre_header->AddSuccessor(if_block);
+  if_block->AddSuccessor(true_block);  // True successor
+  if_block->AddSuccessor(false_block);  // False successor
+  true_block->AddSuccessor(new_pre_header);
+  false_block->AddSuccessor(new_pre_header);
 
-  pre_header->dominated_blocks_.push_back(if_block);
-  if_block->SetDominator(pre_header);
-  if_block->dominated_blocks_.push_back(dummy_block);
-  dummy_block->SetDominator(if_block);
-  if_block->dominated_blocks_.push_back(deopt_block);
-  deopt_block->SetDominator(if_block);
+  old_pre_header->dominated_blocks_.push_back(if_block);
+  if_block->SetDominator(old_pre_header);
+  if_block->dominated_blocks_.push_back(true_block);
+  true_block->SetDominator(if_block);
+  if_block->dominated_blocks_.push_back(false_block);
+  false_block->SetDominator(if_block);
   if_block->dominated_blocks_.push_back(new_pre_header);
   new_pre_header->SetDominator(if_block);
   new_pre_header->dominated_blocks_.push_back(header);
   header->SetDominator(new_pre_header);
 
+  // Fix reverse post order.
   size_t index_of_header = IndexOfElement(reverse_post_order_, header);
   MakeRoomFor(&reverse_post_order_, 4, index_of_header - 1);
   reverse_post_order_[index_of_header++] = if_block;
-  reverse_post_order_[index_of_header++] = dummy_block;
-  reverse_post_order_[index_of_header++] = deopt_block;
+  reverse_post_order_[index_of_header++] = true_block;
+  reverse_post_order_[index_of_header++] = false_block;
   reverse_post_order_[index_of_header++] = new_pre_header;
 
-  HLoopInformation* info = pre_header->GetLoopInformation();
-  if (info != nullptr) {
-    if_block->SetLoopInformation(info);
-    dummy_block->SetLoopInformation(info);
-    deopt_block->SetLoopInformation(info);
-    new_pre_header->SetLoopInformation(info);
-    for (HLoopInformationOutwardIterator loop_it(*pre_header);
+  // Fix loop information.
+  HLoopInformation* loop_info = old_pre_header->GetLoopInformation();
+  if (loop_info != nullptr) {
+    if_block->SetLoopInformation(loop_info);
+    true_block->SetLoopInformation(loop_info);
+    false_block->SetLoopInformation(loop_info);
+    new_pre_header->SetLoopInformation(loop_info);
+    // Add blocks to all enveloping loops.
+    for (HLoopInformationOutwardIterator loop_it(*old_pre_header);
          !loop_it.Done();
          loop_it.Advance()) {
       loop_it.Current()->Add(if_block);
-      loop_it.Current()->Add(dummy_block);
-      loop_it.Current()->Add(deopt_block);
+      loop_it.Current()->Add(true_block);
+      loop_it.Current()->Add(false_block);
       loop_it.Current()->Add(new_pre_header);
     }
   }
+
+  // Fix try/catch information.
+  TryCatchInformation* try_catch_info = old_pre_header->IsTryBlock()
+      ? old_pre_header->GetTryCatchInformation()
+      : nullptr;
+  if_block->SetTryCatchInformation(try_catch_info);
+  true_block->SetTryCatchInformation(try_catch_info);
+  false_block->SetTryCatchInformation(try_catch_info);
+  new_pre_header->SetTryCatchInformation(try_catch_info);
 }
 
 void HInstruction::SetReferenceTypeInfo(ReferenceTypeInfo rti) {
diff --git a/compiler/optimizing/nodes.h b/compiler/optimizing/nodes.h
index d3f30cb..d5110a7 100644
--- a/compiler/optimizing/nodes.h
+++ b/compiler/optimizing/nodes.h
@@ -1101,6 +1101,7 @@
 #define FOR_EACH_CONCRETE_INSTRUCTION_ARM64(M)
 #else
 #define FOR_EACH_CONCRETE_INSTRUCTION_ARM64(M)                          \
+  M(Arm64DataProcWithShifterOp, Instruction)                            \
   M(Arm64IntermediateAddress, Instruction)                              \
   M(Arm64MultiplyAccumulate, Instruction)
 #endif
diff --git a/compiler/optimizing/nodes_arm64.cc b/compiler/optimizing/nodes_arm64.cc
new file mode 100644
index 0000000..ac2f093
--- /dev/null
+++ b/compiler/optimizing/nodes_arm64.cc
@@ -0,0 +1,84 @@
+/*
+ * 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 "common_arm64.h"
+#include "nodes.h"
+
+namespace art {
+
+using arm64::helpers::CanFitInShifterOperand;
+
+void HArm64DataProcWithShifterOp::GetOpInfoFromInstruction(HInstruction* instruction,
+                                                           /*out*/OpKind* op_kind,
+                                                           /*out*/int* shift_amount) {
+  DCHECK(CanFitInShifterOperand(instruction));
+  if (instruction->IsShl()) {
+    *op_kind = kLSL;
+    *shift_amount = instruction->AsShl()->GetRight()->AsIntConstant()->GetValue();
+  } else if (instruction->IsShr()) {
+    *op_kind = kASR;
+    *shift_amount = instruction->AsShr()->GetRight()->AsIntConstant()->GetValue();
+  } else if (instruction->IsUShr()) {
+    *op_kind = kLSR;
+    *shift_amount = instruction->AsUShr()->GetRight()->AsIntConstant()->GetValue();
+  } else {
+    DCHECK(instruction->IsTypeConversion());
+    Primitive::Type result_type = instruction->AsTypeConversion()->GetResultType();
+    Primitive::Type input_type = instruction->AsTypeConversion()->GetInputType();
+    int result_size = Primitive::ComponentSize(result_type);
+    int input_size = Primitive::ComponentSize(input_type);
+    int min_size = std::min(result_size, input_size);
+    // This follows the logic in
+    // `InstructionCodeGeneratorARM64::VisitTypeConversion()`.
+    if (result_type == Primitive::kPrimInt && input_type == Primitive::kPrimLong) {
+      // There is actually nothing to do. The register will be used as a W
+      // register, discarding the top bits. This is represented by the default
+      // encoding 'LSL 0'.
+      *op_kind = kLSL;
+      *shift_amount = 0;
+    } else if (result_type == Primitive::kPrimChar ||
+               (input_type == Primitive::kPrimChar && input_size < result_size)) {
+      *op_kind = kUXTH;
+    } else {
+      switch (min_size) {
+        case 1: *op_kind = kSXTB; break;
+        case 2: *op_kind = kSXTH; break;
+        case 4: *op_kind = kSXTW; break;
+        default:
+          LOG(FATAL) << "Unexpected min size " << min_size;
+      }
+    }
+  }
+}
+
+std::ostream& operator<<(std::ostream& os, const HArm64DataProcWithShifterOp::OpKind op) {
+  switch (op) {
+    case HArm64DataProcWithShifterOp::kLSL:  return os << "LSL";
+    case HArm64DataProcWithShifterOp::kLSR:  return os << "LSR";
+    case HArm64DataProcWithShifterOp::kASR:  return os << "ASR";
+    case HArm64DataProcWithShifterOp::kUXTB: return os << "UXTB";
+    case HArm64DataProcWithShifterOp::kUXTH: return os << "UXTH";
+    case HArm64DataProcWithShifterOp::kUXTW: return os << "UXTW";
+    case HArm64DataProcWithShifterOp::kSXTB: return os << "SXTB";
+    case HArm64DataProcWithShifterOp::kSXTH: return os << "SXTH";
+    case HArm64DataProcWithShifterOp::kSXTW: return os << "SXTW";
+    default:
+      LOG(FATAL) << "Invalid OpKind " << static_cast<int>(op);
+      UNREACHABLE();
+  }
+}
+
+}  // namespace art
diff --git a/compiler/optimizing/nodes_arm64.h b/compiler/optimizing/nodes_arm64.h
index d07f019..e843935 100644
--- a/compiler/optimizing/nodes_arm64.h
+++ b/compiler/optimizing/nodes_arm64.h
@@ -19,6 +19,79 @@
 
 namespace art {
 
+class HArm64DataProcWithShifterOp : public HExpression<2> {
+ public:
+  enum OpKind {
+    kLSL,   // Logical shift left.
+    kLSR,   // Logical shift right.
+    kASR,   // Arithmetic shift right.
+    kUXTB,  // Unsigned extend byte.
+    kUXTH,  // Unsigned extend half-word.
+    kUXTW,  // Unsigned extend word.
+    kSXTB,  // Signed extend byte.
+    kSXTH,  // Signed extend half-word.
+    kSXTW,  // Signed extend word.
+
+    // Aliases.
+    kFirstShiftOp = kLSL,
+    kLastShiftOp = kASR,
+    kFirstExtensionOp = kUXTB,
+    kLastExtensionOp = kSXTW
+  };
+  HArm64DataProcWithShifterOp(HInstruction* instr,
+                              HInstruction* left,
+                              HInstruction* right,
+                              OpKind op,
+                              // The shift argument is unused if the operation
+                              // is an extension.
+                              int shift = 0,
+                              uint32_t dex_pc = kNoDexPc)
+      : HExpression(instr->GetType(), SideEffects::None(), dex_pc),
+        instr_kind_(instr->GetKind()), op_kind_(op), shift_amount_(shift) {
+    DCHECK(!instr->HasSideEffects());
+    SetRawInputAt(0, left);
+    SetRawInputAt(1, right);
+  }
+
+  bool CanBeMoved() const OVERRIDE { return true; }
+  bool InstructionDataEquals(HInstruction* other_instr) const OVERRIDE {
+    HArm64DataProcWithShifterOp* other = other_instr->AsArm64DataProcWithShifterOp();
+    return instr_kind_ == other->instr_kind_ &&
+        op_kind_ == other->op_kind_ &&
+        shift_amount_ == other->shift_amount_;
+  }
+
+  static bool IsShiftOp(OpKind op_kind) {
+    return kFirstShiftOp <= op_kind && op_kind <= kLastShiftOp;
+  }
+
+  static bool IsExtensionOp(OpKind op_kind) {
+    return kFirstExtensionOp <= op_kind && op_kind <= kLastExtensionOp;
+  }
+
+  // Find the operation kind and shift amount from a bitfield move instruction.
+  static void GetOpInfoFromInstruction(HInstruction* bitfield_op,
+                                       /*out*/OpKind* op_kind,
+                                       /*out*/int* shift_amount);
+
+  InstructionKind GetInstrKind() const { return instr_kind_; }
+  OpKind GetOpKind() const { return op_kind_; }
+  int GetShiftAmount() const { return shift_amount_; }
+
+  DECLARE_INSTRUCTION(Arm64DataProcWithShifterOp);
+
+ private:
+  InstructionKind instr_kind_;
+  OpKind op_kind_;
+  int shift_amount_;
+
+  friend std::ostream& operator<<(std::ostream& os, OpKind op);
+
+  DISALLOW_COPY_AND_ASSIGN(HArm64DataProcWithShifterOp);
+};
+
+std::ostream& operator<<(std::ostream& os, const HArm64DataProcWithShifterOp::OpKind op);
+
 // This instruction computes an intermediate address pointing in the 'middle' of an object. The
 // result pointer cannot be handled by GC, so extra care is taken to make sure that this value is
 // never used across anything that can trigger GC.
diff --git a/runtime/arch/stub_test.cc b/runtime/arch/stub_test.cc
index afa48cd..8746bad 100644
--- a/runtime/arch/stub_test.cc
+++ b/runtime/arch/stub_test.cc
@@ -431,9 +431,9 @@
           [referrer] "r"(referrer), [hidden] "r"(hidden)
         : "at", "v0", "v1", "t0", "t1", "t2", "t3", "s0", "s1", "s2", "s3", "s4", "s5", "s6", "s7",
           "t8", "t9", "k0", "k1", "fp", "ra",
-          "f0", "f1", "f2", "f3", "f4", "f5", "f6", "f7", "f8", "f9", "f10", "f11", "f12", "f13",
-          "f14", "f15", "f16", "f17", "f18", "f19", "f20", "f21", "f22", "f23", "f24", "f25", "f26",
-          "f27", "f28", "f29", "f30", "f31",
+          "$f0", "$f1", "$f2", "$f3", "$f4", "$f5", "$f6", "$f7", "$f8", "$f9", "$f10", "$f11",
+          "$f12", "$f13", "$f14", "$f15", "$f16", "$f17", "$f18", "$f19", "$f20", "$f21", "$f22",
+          "$f23", "$f24", "$f25", "$f26", "$f27", "$f28", "$f29", "$f30", "$f31",
           "memory");  // clobber.
 #elif defined(__x86_64__) && !defined(__APPLE__)
 #define PUSH(reg) "pushq " # reg "\n\t .cfi_adjust_cfa_offset 8\n\t"
diff --git a/runtime/jit/jit.cc b/runtime/jit/jit.cc
index 4660426..27a0e2d 100644
--- a/runtime/jit/jit.cc
+++ b/runtime/jit/jit.cc
@@ -81,7 +81,7 @@
     return nullptr;
   }
   jit->offline_profile_info_.reset(nullptr);
-  if (true) {
+  if (options->GetSaveProfilingInfo()) {
     jit->offline_profile_info_.reset(new OfflineProfilingInfo());
   }
   LOG(INFO) << "JIT created with initial_capacity="
diff --git a/test/551-checker-shifter-operand/build b/test/551-checker-shifter-operand/build
new file mode 100644
index 0000000..18e8c59
--- /dev/null
+++ b/test/551-checker-shifter-operand/build
@@ -0,0 +1,212 @@
+#!/bin/bash
+#
+# Copyright (C) 2008 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.
+
+
+# This is an almost exact copy of `art/test/etc/default-build`. Only the parsing
+# of `dx` option has been overriden.
+
+# Stop if something fails.
+set -e
+
+# Set default values for directories.
+if [ -d smali ]; then
+  HAS_SMALI=true
+else
+  HAS_SMALI=false
+fi
+
+if [ -d src ]; then
+  HAS_SRC=true
+else
+  HAS_SRC=false
+fi
+
+if [ -d src2 ]; then
+  HAS_SRC2=true
+else
+  HAS_SRC2=false
+fi
+
+if [ -d src-multidex ]; then
+  HAS_SRC_MULTIDEX=true
+else
+  HAS_SRC_MULTIDEX=false
+fi
+
+if [ -d src-ex ]; then
+  HAS_SRC_EX=true
+else
+  HAS_SRC_EX=false
+fi
+
+DX_FLAGS=""
+SKIP_DX_MERGER="false"
+EXPERIMENTAL=""
+
+# Setup experimental flag mappings in a bash associative array.
+declare -A JACK_EXPERIMENTAL_ARGS
+JACK_EXPERIMENTAL_ARGS["default-methods"]="-D jack.java.source.version=1.8"
+JACK_EXPERIMENTAL_ARGS["lambdas"]="-D jack.java.source.version=1.8"
+
+while true; do
+  if [ "x$1" = "x--dx-option" ]; then
+    shift
+    option="$1"
+    # Make sure we run this test *with* `dx` optimizations.
+    if [ "x$option" != "x--no-optimize" ]; then
+      DX_FLAGS="${DX_FLAGS} $option"
+    fi
+    shift
+  elif [ "x$1" = "x--jvm" ]; then
+    shift
+  elif [ "x$1" = "x--no-src" ]; then
+    HAS_SRC=false
+    shift
+  elif [ "x$1" = "x--no-src2" ]; then
+    HAS_SRC2=false
+    shift
+  elif [ "x$1" = "x--no-src-multidex" ]; then
+    HAS_SRC_MULTIDEX=false
+    shift
+  elif [ "x$1" = "x--no-src-ex" ]; then
+    HAS_SRC_EX=false
+    shift
+  elif [ "x$1" = "x--no-smali" ]; then
+    HAS_SMALI=false
+    shift
+  elif [ "x$1" = "x--experimental" ]; then
+    shift
+    EXPERIMENTAL="${EXPERIMENTAL} $1"
+    shift
+  elif expr "x$1" : "x--" >/dev/null 2>&1; then
+    echo "unknown $0 option: $1" 1>&2
+    exit 1
+  else
+    break
+  fi
+done
+
+# Add args from the experimental mappings.
+for experiment in ${EXPERIMENTAL}; do
+  JACK_ARGS="${JACK_ARGS} ${JACK_EXPERIMENTAL_ARGS[${experiment}]}"
+done
+
+if [ -e classes.dex ]; then
+  zip $TEST_NAME.jar classes.dex
+  exit 0
+fi
+
+if ! [ "${HAS_SRC}" = "true" ] && ! [ "${HAS_SRC2}" = "true" ]; then
+  # No src directory? Then forget about trying to run dx.
+  SKIP_DX_MERGER="true"
+fi
+
+if [ "${HAS_SRC_MULTIDEX}" = "true" ]; then
+  # Jack does not support this configuration unless we specify how to partition the DEX file
+  # with a .jpp file.
+  USE_JACK="false"
+fi
+
+if [ ${USE_JACK} = "true" ]; then
+  # Jack toolchain
+  if [ "${HAS_SRC}" = "true" ]; then
+    ${JACK} ${JACK_ARGS} --output-jack src.jack src
+    imported_jack_files="--import src.jack"
+  fi
+
+  if [ "${HAS_SRC2}" = "true" ]; then
+    ${JACK} ${JACK_ARGS} --output-jack src2.jack src2
+    imported_jack_files="--import src2.jack ${imported_jack_files}"
+  fi
+
+  # Compile jack files into a DEX file. We set jack.import.type.policy=keep-first to consider
+  # class definitions from src2 first.
+  if [ "${HAS_SRC}" = "true" ] || [ "${HAS_SRC2}" = "true" ]; then
+    ${JACK} ${JACK_ARGS} ${imported_jack_files} -D jack.import.type.policy=keep-first --output-dex .
+  fi
+else
+  # Legacy toolchain with javac+dx
+  if [ "${HAS_SRC}" = "true" ]; then
+    mkdir classes
+    ${JAVAC} ${JAVAC_ARGS} -implicit:none -classpath src-multidex -d classes `find src -name '*.java'`
+  fi
+
+  if [ "${HAS_SRC_MULTIDEX}" = "true" ]; then
+    mkdir classes2
+    ${JAVAC} -implicit:none -classpath src -d classes2 `find src-multidex -name '*.java'`
+    if [ ${NEED_DEX} = "true" ]; then
+      ${DX} -JXmx256m --debug --dex --dump-to=classes2.lst --output=classes2.dex \
+        --dump-width=1000 ${DX_FLAGS} classes2
+    fi
+  fi
+
+  if [ "${HAS_SRC2}" = "true" ]; then
+    mkdir -p classes
+    ${JAVAC} ${JAVAC_ARGS} -d classes `find src2 -name '*.java'`
+  fi
+
+  if [ "${HAS_SRC}" = "true" ] || [ "${HAS_SRC2}" = "true" ]; then
+    if [ ${NEED_DEX} = "true" -a ${SKIP_DX_MERGER} = "false" ]; then
+      ${DX} -JXmx256m --debug --dex --dump-to=classes.lst --output=classes.dex \
+        --dump-width=1000 ${DX_FLAGS} classes
+    fi
+  fi
+fi
+
+if [ "${HAS_SMALI}" = "true" ]; then
+  # Compile Smali classes
+  ${SMALI} -JXmx512m ${SMALI_ARGS} --output smali_classes.dex `find smali -name '*.smali'`
+
+  # Don't bother with dexmerger if we provide our own main function in a smali file.
+  if [ ${SKIP_DX_MERGER} = "false" ]; then
+    ${DXMERGER} classes.dex classes.dex smali_classes.dex
+  else
+    mv smali_classes.dex classes.dex
+  fi
+fi
+
+if [ ${HAS_SRC_EX} = "true" ]; then
+  if [ ${USE_JACK} = "true" ]; then
+      # Rename previous "classes.dex" so it is not overwritten.
+      mv classes.dex classes-1.dex
+      #TODO find another way to append src.jack to the jack classpath
+      ${JACK}:src.jack ${JACK_ARGS} --output-dex . src-ex
+      zip $TEST_NAME-ex.jar classes.dex
+      # Restore previous "classes.dex" so it can be zipped.
+      mv classes-1.dex classes.dex
+  else
+    mkdir classes-ex
+    ${JAVAC} ${JAVAC_ARGS} -d classes-ex -cp classes `find src-ex -name '*.java'`
+    if [ ${NEED_DEX} = "true" ]; then
+      ${DX} -JXmx256m --debug --dex --dump-to=classes-ex.lst --output=classes-ex.dex \
+        --dump-width=1000 ${DX_FLAGS} classes-ex
+
+      # quick shuffle so that the stored name is "classes.dex"
+      mv classes.dex classes-1.dex
+      mv classes-ex.dex classes.dex
+      zip $TEST_NAME-ex.jar classes.dex
+      mv classes.dex classes-ex.dex
+      mv classes-1.dex classes.dex
+    fi
+  fi
+fi
+
+# Create a single jar with two dex files for multidex.
+if [ ${HAS_SRC_MULTIDEX} = "true" ]; then
+  zip $TEST_NAME.jar classes.dex classes2.dex
+elif [ ${NEED_DEX} = "true" ]; then
+  zip $TEST_NAME.jar classes.dex
+fi
diff --git a/test/551-checker-shifter-operand/expected.txt b/test/551-checker-shifter-operand/expected.txt
new file mode 100644
index 0000000..e69de29
--- /dev/null
+++ b/test/551-checker-shifter-operand/expected.txt
diff --git a/test/551-checker-shifter-operand/info.txt b/test/551-checker-shifter-operand/info.txt
new file mode 100644
index 0000000..10e998c
--- /dev/null
+++ b/test/551-checker-shifter-operand/info.txt
@@ -0,0 +1 @@
+Test the merging of instructions into the shifter operand on arm64.
diff --git a/test/551-checker-shifter-operand/src/Main.java b/test/551-checker-shifter-operand/src/Main.java
new file mode 100644
index 0000000..decdd1f
--- /dev/null
+++ b/test/551-checker-shifter-operand/src/Main.java
@@ -0,0 +1,678 @@
+/*
+* 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.
+*/
+
+public class Main {
+
+  // A dummy value to defeat inlining of these routines.
+  static boolean doThrow = false;
+
+  public static void assertByteEquals(byte expected, byte result) {
+    if (expected != result) {
+      throw new Error("Expected: " + expected + ", found: " + result);
+    }
+  }
+
+  public static void assertCharEquals(char expected, char result) {
+    if (expected != result) {
+      throw new Error("Expected: " + expected + ", found: " + result);
+    }
+  }
+
+  public static void assertShortEquals(short expected, short result) {
+    if (expected != result) {
+      throw new Error("Expected: " + expected + ", found: " + result);
+    }
+  }
+
+  public static void assertIntEquals(int expected, int result) {
+    if (expected != result) {
+      throw new Error("Expected: " + expected + ", found: " + result);
+    }
+  }
+
+  public static void assertLongEquals(long expected, long result) {
+    if (expected != result) {
+      throw new Error("Expected: " + expected + ", found: " + result);
+    }
+  }
+
+  // Non-inlinable type-casting helpers.
+  static  char $noinline$byteToChar   (byte v) { if (doThrow) throw new Error(); return  (char)v; }
+  static short $noinline$byteToShort  (byte v) { if (doThrow) throw new Error(); return (short)v; }
+  static   int $noinline$byteToInt    (byte v) { if (doThrow) throw new Error(); return   (int)v; }
+  static  long $noinline$byteToLong   (byte v) { if (doThrow) throw new Error(); return  (long)v; }
+  static  byte $noinline$charToByte   (char v) { if (doThrow) throw new Error(); return  (byte)v; }
+  static short $noinline$charToShort  (char v) { if (doThrow) throw new Error(); return (short)v; }
+  static   int $noinline$charToInt    (char v) { if (doThrow) throw new Error(); return   (int)v; }
+  static  long $noinline$charToLong   (char v) { if (doThrow) throw new Error(); return  (long)v; }
+  static  byte $noinline$shortToByte (short v) { if (doThrow) throw new Error(); return  (byte)v; }
+  static  char $noinline$shortToChar (short v) { if (doThrow) throw new Error(); return  (char)v; }
+  static   int $noinline$shortToInt  (short v) { if (doThrow) throw new Error(); return   (int)v; }
+  static  long $noinline$shortToLong (short v) { if (doThrow) throw new Error(); return  (long)v; }
+  static  byte $noinline$intToByte     (int v) { if (doThrow) throw new Error(); return  (byte)v; }
+  static  char $noinline$intToChar     (int v) { if (doThrow) throw new Error(); return  (char)v; }
+  static short $noinline$intToShort    (int v) { if (doThrow) throw new Error(); return (short)v; }
+  static  long $noinline$intToLong     (int v) { if (doThrow) throw new Error(); return  (long)v; }
+  static  byte $noinline$longToByte   (long v) { if (doThrow) throw new Error(); return  (byte)v; }
+  static  char $noinline$longToChar   (long v) { if (doThrow) throw new Error(); return  (char)v; }
+  static short $noinline$longToShort  (long v) { if (doThrow) throw new Error(); return (short)v; }
+  static   int $noinline$longToInt    (long v) { if (doThrow) throw new Error(); return   (int)v; }
+
+  /**
+   * Basic test merging a bitfield move operation (here a type conversion) into
+   * the shifter operand.
+   */
+
+  /// CHECK-START-ARM64: long Main.$opt$noinline$translate(long, byte) instruction_simplifier_arm64 (before)
+  /// CHECK-DAG:   <<l:j\d+>>           ParameterValue
+  /// CHECK-DAG:   <<b:b\d+>>           ParameterValue
+  /// CHECK:       <<tmp:j\d+>>         TypeConversion [<<b>>]
+  /// CHECK:                            Sub [<<l>>,<<tmp>>]
+
+  /// CHECK-START-ARM64: long Main.$opt$noinline$translate(long, byte) instruction_simplifier_arm64 (after)
+  /// CHECK-DAG:   <<l:j\d+>>           ParameterValue
+  /// CHECK-DAG:   <<b:b\d+>>           ParameterValue
+  /// CHECK:                            Arm64DataProcWithShifterOp [<<l>>,<<b>>] kind:Sub+SXTB
+
+  /// CHECK-START-ARM64: long Main.$opt$noinline$translate(long, byte) instruction_simplifier_arm64 (after)
+  /// CHECK-NOT:                        TypeConversion
+  /// CHECK-NOT:                        Sub
+
+  /// CHECK-START-ARM64: long Main.$opt$noinline$translate(long, byte) disassembly (after)
+  /// CHECK:                            sub x{{\d+}}, x{{\d+}}, w{{\d+}}, sxtb
+
+  public static long $opt$noinline$translate(long l, byte b) {
+    if (doThrow) throw new Error();
+    long tmp = (long)b;
+    return l - tmp;
+  }
+
+
+  /**
+   * Test that we do not merge into the shifter operand when the left and right
+   * inputs are the the IR.
+   */
+
+  /// CHECK-START-ARM64: int Main.$opt$noinline$sameInput(int) instruction_simplifier_arm64 (before)
+  /// CHECK:       <<a:i\d+>>           ParameterValue
+  /// CHECK:       <<Const2:i\d+>>      IntConstant 2
+  /// CHECK:       <<tmp:i\d+>>         Shl [<<a>>,<<Const2>>]
+  /// CHECK:                            Add [<<tmp>>,<<tmp>>]
+
+  /// CHECK-START-ARM64: int Main.$opt$noinline$sameInput(int) instruction_simplifier_arm64 (after)
+  /// CHECK-DAG:   <<a:i\d+>>           ParameterValue
+  /// CHECK-DAG:   <<Const2:i\d+>>      IntConstant 2
+  /// CHECK:       <<Shl:i\d+>>         Shl [<<a>>,<<Const2>>]
+  /// CHECK:                            Add [<<Shl>>,<<Shl>>]
+
+  /// CHECK-START-ARM64: int Main.$opt$noinline$sameInput(int) instruction_simplifier_arm64 (after)
+  /// CHECK-NOT:                        Arm64DataProcWithShifterOp
+
+  public static int $opt$noinline$sameInput(int a) {
+    if (doThrow) throw new Error();
+    int tmp = a << 2;
+    return tmp + tmp;
+  }
+
+  /**
+   * Check that we perform the merge for multiple uses.
+   */
+
+  /// CHECK-START-ARM64: int Main.$opt$noinline$multipleUses(int) instruction_simplifier_arm64 (before)
+  /// CHECK:       <<arg:i\d+>>         ParameterValue
+  /// CHECK:       <<Const23:i\d+>>     IntConstant 23
+  /// CHECK:       <<tmp:i\d+>>         Shl [<<arg>>,<<Const23>>]
+  /// CHECK:                            Add [<<tmp>>,{{i\d+}}]
+  /// CHECK:                            Add [<<tmp>>,{{i\d+}}]
+  /// CHECK:                            Add [<<tmp>>,{{i\d+}}]
+  /// CHECK:                            Add [<<tmp>>,{{i\d+}}]
+  /// CHECK:                            Add [<<tmp>>,{{i\d+}}]
+
+  /// CHECK-START-ARM64: int Main.$opt$noinline$multipleUses(int) instruction_simplifier_arm64 (after)
+  /// CHECK:       <<arg:i\d+>>         ParameterValue
+  /// CHECK:                            Arm64DataProcWithShifterOp [{{i\d+}},<<arg>>] kind:Add+LSL shift:23
+  /// CHECK:                            Arm64DataProcWithShifterOp [{{i\d+}},<<arg>>] kind:Add+LSL shift:23
+  /// CHECK:                            Arm64DataProcWithShifterOp [{{i\d+}},<<arg>>] kind:Add+LSL shift:23
+  /// CHECK:                            Arm64DataProcWithShifterOp [{{i\d+}},<<arg>>] kind:Add+LSL shift:23
+  /// CHECK:                            Arm64DataProcWithShifterOp [{{i\d+}},<<arg>>] kind:Add+LSL shift:23
+
+  /// CHECK-START-ARM64: int Main.$opt$noinline$multipleUses(int) instruction_simplifier_arm64 (after)
+  /// CHECK-NOT:                        Shl
+  /// CHECK-NOT:                        Add
+
+  public static int $opt$noinline$multipleUses(int arg) {
+    if (doThrow) throw new Error();
+    int tmp = arg << 23;
+    switch (arg) {
+      case 1:  return (arg | 1) + tmp;
+      case 2:  return (arg | 2) + tmp;
+      case 3:  return (arg | 3) + tmp;
+      case 4:  return (arg | 4) + tmp;
+      case (1 << 20):  return (arg | 5) + tmp;
+      default: return 0;
+    }
+  }
+
+  /**
+   * Logical instructions cannot take 'extend' operations into the shift
+   * operand, so test that only the shifts are merged.
+   */
+
+  /// CHECK-START-ARM64: void Main.$opt$noinline$testAnd(long, long) instruction_simplifier_arm64 (after)
+  /// CHECK:                            Arm64DataProcWithShifterOp
+  /// CHECK-NOT:                        Arm64DataProcWithShifterOp
+
+  /// CHECK-START-ARM64: void Main.$opt$noinline$testAnd(long, long) disassembly (after)
+  /// CHECK:                            and lsl
+  /// CHECK:                            sxtb
+  /// CHECK:                            and
+
+  static void $opt$noinline$testAnd(long a, long b) {
+    if (doThrow) throw new Error();
+    assertLongEquals((a & $noinline$LongShl(b, 5)) | (a & $noinline$longToByte(b)),
+                     (a & (b << 5)) | (a & (byte)b));
+  }
+
+  /// CHECK-START-ARM64: void Main.$opt$noinline$testOr(int, int) instruction_simplifier_arm64 (after)
+  /// CHECK:                            Arm64DataProcWithShifterOp
+  /// CHECK-NOT:                        Arm64DataProcWithShifterOp
+
+  /// CHECK-START-ARM64: void Main.$opt$noinline$testOr(int, int) disassembly (after)
+  /// CHECK:                            orr asr
+  /// CHECK:                            uxth
+  /// CHECK:                            orr
+
+  static void $opt$noinline$testOr(int a, int b) {
+    if (doThrow) throw new Error();
+    assertIntEquals((a | $noinline$IntShr(b, 6)) | (a | $noinline$intToChar(b)),
+                    (a | (b >> 6)) | (a | (char)b));
+  }
+
+  /// CHECK-START-ARM64: void Main.$opt$noinline$testXor(long, long) instruction_simplifier_arm64 (after)
+  /// CHECK:                            Arm64DataProcWithShifterOp
+  /// CHECK-NOT:                        Arm64DataProcWithShifterOp
+
+  /// CHECK-START-ARM64: void Main.$opt$noinline$testXor(long, long) disassembly (after)
+  /// CHECK:                            eor lsr
+  /// CHECK:                            sxtw
+  /// CHECK:                            eor
+
+  static void $opt$noinline$testXor(long a, long b) {
+    if (doThrow) throw new Error();
+    assertLongEquals((a ^ $noinline$LongUshr(b, 7)) | (a ^ $noinline$longToInt(b)),
+                     (a ^ (b >>> 7)) | (a ^ (int)b));
+  }
+
+  /// CHECK-START-ARM64: void Main.$opt$noinline$testNeg(int) instruction_simplifier_arm64 (after)
+  /// CHECK:                            Arm64DataProcWithShifterOp
+  /// CHECK-NOT:                        Arm64DataProcWithShifterOp
+
+  /// CHECK-START-ARM64: void Main.$opt$noinline$testNeg(int) disassembly (after)
+  /// CHECK:                            neg lsl
+  /// CHECK:                            sxth
+  /// CHECK:                            neg
+
+  static void $opt$noinline$testNeg(int a) {
+    if (doThrow) throw new Error();
+    assertIntEquals(-$noinline$IntShl(a, 8) | -$noinline$intToShort(a),
+                    (-(a << 8)) | (-(short)a));
+  }
+
+  /**
+   * The functions below are used to compare the result of optimized operations
+   * to non-optimized operations.
+   * On the left-hand side we use a non-inlined function call to ensure the
+   * optimization does not occur. The checker tests ensure that the optimization
+   * does occur on the right-hand.
+   */
+
+  /// CHECK-START-ARM64: void Main.$opt$validateExtendByteInt1(int, byte) instruction_simplifier_arm64 (after)
+  /// CHECK:                            Arm64DataProcWithShifterOp
+  /// CHECK:                            Arm64DataProcWithShifterOp
+
+  /// CHECK-START-ARM64: void Main.$opt$validateExtendByteInt1(int, byte) instruction_simplifier_arm64 (after)
+  /// CHECK-NOT:                        TypeConversion
+
+  public static void $opt$validateExtendByteInt1(int a, byte b) {
+    assertIntEquals(a + $noinline$byteToChar (b), a +  (char)b);
+    assertIntEquals(a + $noinline$byteToShort(b), a + (short)b);
+  }
+
+  /// CHECK-START-ARM64: void Main.$opt$validateExtendByteInt2(int, byte) instruction_simplifier_arm64 (after)
+  /// CHECK-NOT:                        Arm64DataProcWithShifterOp
+  /// CHECK-NOT:                        Arm64DataProcWithShifterOp
+
+  public static void $opt$validateExtendByteInt2(int a, byte b) {
+    // The conversion to `int` has been optimized away, so there is nothing to merge.
+    assertIntEquals (a + $noinline$byteToInt (b), a +  (int)b);
+    // There is an environment use for `(long)b`, preventing the merge.
+    assertLongEquals(a + $noinline$byteToLong(b), a + (long)b);
+  }
+
+  /// CHECK-START-ARM64: void Main.$opt$validateExtendByteLong(long, byte) instruction_simplifier_arm64 (after)
+  /// CHECK:                            Arm64DataProcWithShifterOp
+  /// CHECK:                            Arm64DataProcWithShifterOp
+  /// CHECK:                            Arm64DataProcWithShifterOp
+
+  /// CHECK-START-ARM64: void Main.$opt$validateExtendByteLong(long, byte) instruction_simplifier_arm64 (after)
+  /// CHECK:                            TypeConversion
+  /// CHECK:                            TypeConversion
+  /// CHECK-NOT:                        TypeConversion
+
+  public static void $opt$validateExtendByteLong(long a, byte b) {
+    // The first two tests have a type conversion.
+    assertLongEquals(a + $noinline$byteToChar (b), a +  (char)b);
+    assertLongEquals(a + $noinline$byteToShort(b), a + (short)b);
+    // This test does not because the conversion to `int` is optimized away.
+    assertLongEquals(a + $noinline$byteToInt  (b), a +  (int)b);
+  }
+
+  public static void $opt$validateExtendByte(long a, byte b) {
+    $opt$validateExtendByteInt1((int)a, b);
+    $opt$validateExtendByteInt2((int)a, b);
+    $opt$validateExtendByteLong(a, b);
+  }
+
+  /// CHECK-START-ARM64: void Main.$opt$validateExtendCharInt1(int, char) instruction_simplifier_arm64 (after)
+  /// CHECK:                            Arm64DataProcWithShifterOp
+  /// CHECK:                            Arm64DataProcWithShifterOp
+
+  /// CHECK-START-ARM64: void Main.$opt$validateExtendCharInt1(int, char) instruction_simplifier_arm64 (after)
+  /// CHECK-NOT:                        TypeConversion
+
+  public static void $opt$validateExtendCharInt1(int a, char b) {
+    assertIntEquals(a + $noinline$charToByte (b), a +  (byte)b);
+    assertIntEquals(a + $noinline$charToShort(b), a + (short)b);
+  }
+
+  /// CHECK-START-ARM64: void Main.$opt$validateExtendCharInt2(int, char) instruction_simplifier_arm64 (after)
+  /// CHECK-NOT:                        Arm64DataProcWithShifterOp
+  /// CHECK-NOT:                        Arm64DataProcWithShifterOp
+
+  public static void $opt$validateExtendCharInt2(int a, char b) {
+    // The conversion to `int` has been optimized away, so there is nothing to merge.
+    assertIntEquals (a + $noinline$charToInt (b), a +  (int)b);
+    // There is an environment use for `(long)b`, preventing the merge.
+    assertLongEquals(a + $noinline$charToLong(b), a + (long)b);
+  }
+
+  /// CHECK-START-ARM64: void Main.$opt$validateExtendCharLong(long, char) instruction_simplifier_arm64 (after)
+  /// CHECK:                            Arm64DataProcWithShifterOp
+  /// CHECK:                            Arm64DataProcWithShifterOp
+  /// CHECK:                            Arm64DataProcWithShifterOp
+  /// CHECK:                            Arm64DataProcWithShifterOp
+
+  /// CHECK-START-ARM64: void Main.$opt$validateExtendCharLong(long, char) instruction_simplifier_arm64 (after)
+  /// CHECK:                            TypeConversion
+  /// CHECK:                            TypeConversion
+  /// CHECK-NOT:                        TypeConversion
+
+  public static void $opt$validateExtendCharLong(long a, char b) {
+    // The first two tests have a type conversion.
+    assertLongEquals(a + $noinline$charToByte (b), a +  (byte)b);
+    assertLongEquals(a + $noinline$charToShort(b), a + (short)b);
+    // This test does not because the conversion to `int` is optimized away.
+    assertLongEquals(a + $noinline$charToInt  (b), a +   (int)b);
+  }
+
+  public static void $opt$validateExtendChar(long a, char b) {
+    $opt$validateExtendCharInt1((int)a, b);
+    $opt$validateExtendCharInt2((int)a, b);
+    $opt$validateExtendCharLong(a, b);
+  }
+
+  /// CHECK-START-ARM64: void Main.$opt$validateExtendShortInt1(int, short) instruction_simplifier_arm64 (after)
+  /// CHECK:                            Arm64DataProcWithShifterOp
+  /// CHECK:                            Arm64DataProcWithShifterOp
+
+  /// CHECK-START-ARM64: void Main.$opt$validateExtendShortInt1(int, short) instruction_simplifier_arm64 (after)
+  /// CHECK-NOT:                        TypeConversion
+
+  public static void $opt$validateExtendShortInt1(int a, short b) {
+    assertIntEquals(a + $noinline$shortToByte (b), a + (byte)b);
+    assertIntEquals(a + $noinline$shortToChar (b), a + (char)b);
+  }
+
+  /// CHECK-START-ARM64: void Main.$opt$validateExtendShortInt2(int, short) instruction_simplifier_arm64 (after)
+  /// CHECK-NOT:                        Arm64DataProcWithShifterOp
+  /// CHECK-NOT:                        Arm64DataProcWithShifterOp
+
+  public static void $opt$validateExtendShortInt2(int a, short b) {
+    // The conversion to `int` has been optimized away, so there is nothing to merge.
+    assertIntEquals (a + $noinline$shortToInt  (b), a +  (int)b);
+    // There is an environment use for `(long)b`, preventing the merge.
+    assertLongEquals(a + $noinline$shortToLong (b), a + (long)b);
+  }
+
+  /// CHECK-START-ARM64: void Main.$opt$validateExtendShortLong(long, short) instruction_simplifier_arm64 (after)
+  /// CHECK:                            Arm64DataProcWithShifterOp
+  /// CHECK:                            Arm64DataProcWithShifterOp
+  /// CHECK:                            Arm64DataProcWithShifterOp
+
+  /// CHECK-START-ARM64: void Main.$opt$validateExtendShortLong(long, short) instruction_simplifier_arm64 (after)
+  /// CHECK:                            TypeConversion
+  /// CHECK:                            TypeConversion
+  /// CHECK-NOT:                        TypeConversion
+
+  public static void $opt$validateExtendShortLong(long a, short b) {
+    // The first two tests have a type conversion.
+    assertLongEquals(a + $noinline$shortToByte(b), a + (byte)b);
+    assertLongEquals(a + $noinline$shortToChar(b), a + (char)b);
+    // This test does not because the conversion to `int` is optimized away.
+    assertLongEquals(a + $noinline$shortToInt (b), a +  (int)b);
+  }
+
+  public static void $opt$validateExtendShort(long a, short b) {
+    $opt$validateExtendShortInt1((int)a, b);
+    $opt$validateExtendShortInt2((int)a, b);
+    $opt$validateExtendShortLong(a, b);
+  }
+
+  /// CHECK-START-ARM64: void Main.$opt$validateExtendInt(long, int) instruction_simplifier_arm64 (after)
+  /// CHECK:                            Arm64DataProcWithShifterOp
+  /// CHECK:                            Arm64DataProcWithShifterOp
+  /// CHECK:                            Arm64DataProcWithShifterOp
+  /// CHECK:                            Arm64DataProcWithShifterOp
+
+  /// CHECK-START-ARM64: void Main.$opt$validateExtendInt(long, int) instruction_simplifier_arm64 (after)
+  /// CHECK:                            TypeConversion
+  /// CHECK:                            TypeConversion
+  /// CHECK:                            TypeConversion
+  /// CHECK-NOT:                        TypeConversion
+
+  public static void $opt$validateExtendInt(long a, int b) {
+    // All tests have a conversion to `long`. The first three tests also have a
+    // conversion from `int` to the specified type. For each test the conversion
+    // to `long` is merged into the shifter operand.
+    assertLongEquals(a + $noinline$intToByte (b), a +  (byte)b);
+    assertLongEquals(a + $noinline$intToChar (b), a +  (char)b);
+    assertLongEquals(a + $noinline$intToShort(b), a + (short)b);
+    assertLongEquals(a + $noinline$intToLong (b), a +  (long)b);
+  }
+
+  /// CHECK-START-ARM64: void Main.$opt$validateExtendLong(long, long) instruction_simplifier_arm64 (after)
+  /// CHECK:                            Arm64DataProcWithShifterOp
+  /// CHECK:                            Arm64DataProcWithShifterOp
+  /// CHECK:                            Arm64DataProcWithShifterOp
+  /// CHECK:                            Arm64DataProcWithShifterOp
+
+  /// CHECK-START-ARM64: void Main.$opt$validateExtendLong(long, long) instruction_simplifier_arm64 (after)
+  /// CHECK:                            TypeConversion
+  /// CHECK:                            TypeConversion
+  /// CHECK:                            TypeConversion
+  /// CHECK:                            TypeConversion
+  /// CHECK-NOT:                        TypeConversion
+
+  public static void $opt$validateExtendLong(long a, long b) {
+    // Each test has two conversions, from `long` and then back to `long`. The
+    // conversions to `long` are merged.
+    assertLongEquals(a + $noinline$longToByte (b), a +  (byte)b);
+    assertLongEquals(a + $noinline$longToChar (b), a +  (char)b);
+    assertLongEquals(a + $noinline$longToShort(b), a + (short)b);
+    assertLongEquals(a + $noinline$longToInt  (b), a +   (int)b);
+  }
+
+
+  static int $noinline$IntShl(int b, int c) {
+    if (doThrow) throw new Error();
+    return b << c;
+  }
+  static int $noinline$IntShr(int b, int c) {
+    if (doThrow) throw new Error();
+    return b >> c;
+  }
+  static int $noinline$IntUshr(int b, int c) {
+    if (doThrow) throw new Error();
+    return b >>> c;
+  }
+
+
+  // Each test line below should see one merge.
+  /// CHECK-START-ARM64: void Main.$opt$validateShiftInt(int, int) instruction_simplifier_arm64 (after)
+  /// CHECK:                            Arm64DataProcWithShifterOp
+  /// CHECK:                            Arm64DataProcWithShifterOp
+  /// CHECK:                            Arm64DataProcWithShifterOp
+  /// CHECK:                            Arm64DataProcWithShifterOp
+  /// CHECK:                            Arm64DataProcWithShifterOp
+  /// CHECK:                            Arm64DataProcWithShifterOp
+  /// CHECK:                            Arm64DataProcWithShifterOp
+  /// CHECK:                            Arm64DataProcWithShifterOp
+  /// CHECK:                            Arm64DataProcWithShifterOp
+  /// CHECK:                            Arm64DataProcWithShifterOp
+  /// CHECK:                            Arm64DataProcWithShifterOp
+  /// CHECK:                            Arm64DataProcWithShifterOp
+  /// CHECK:                            Arm64DataProcWithShifterOp
+  /// CHECK:                            Arm64DataProcWithShifterOp
+  /// CHECK:                            Arm64DataProcWithShifterOp
+  /// CHECK:                            Arm64DataProcWithShifterOp
+  /// CHECK:                            Arm64DataProcWithShifterOp
+  /// CHECK:                            Arm64DataProcWithShifterOp
+  /// CHECK:                            Arm64DataProcWithShifterOp
+  /// CHECK:                            Arm64DataProcWithShifterOp
+  /// CHECK:                            Arm64DataProcWithShifterOp
+  /// CHECK:                            Arm64DataProcWithShifterOp
+  /// CHECK:                            Arm64DataProcWithShifterOp
+  /// CHECK:                            Arm64DataProcWithShifterOp
+  /// CHECK:                            Arm64DataProcWithShifterOp
+  /// CHECK:                            Arm64DataProcWithShifterOp
+  /// CHECK:                            Arm64DataProcWithShifterOp
+  /// CHECK:                            Arm64DataProcWithShifterOp
+  /// CHECK:                            Arm64DataProcWithShifterOp
+  /// CHECK:                            Arm64DataProcWithShifterOp
+  /// CHECK:                            Arm64DataProcWithShifterOp
+  /// CHECK:                            Arm64DataProcWithShifterOp
+  /// CHECK:                            Arm64DataProcWithShifterOp
+  /// CHECK:                            Arm64DataProcWithShifterOp
+  /// CHECK:                            Arm64DataProcWithShifterOp
+  /// CHECK:                            Arm64DataProcWithShifterOp
+
+  /// CHECK-START-ARM64: void Main.$opt$validateShiftInt(int, int) instruction_simplifier_arm64 (after)
+  /// CHECK-NOT:                        Shl
+  /// CHECK-NOT:                        Shr
+  /// CHECK-NOT:                        UShr
+
+  public static void $opt$validateShiftInt(int a, int b) {
+    assertIntEquals(a + $noinline$IntShl(b, 1),   a + (b <<  1));
+    assertIntEquals(a + $noinline$IntShl(b, 6),   a + (b <<  6));
+    assertIntEquals(a + $noinline$IntShl(b, 7),   a + (b <<  7));
+    assertIntEquals(a + $noinline$IntShl(b, 8),   a + (b <<  8));
+    assertIntEquals(a + $noinline$IntShl(b, 14),  a + (b << 14));
+    assertIntEquals(a + $noinline$IntShl(b, 15),  a + (b << 15));
+    assertIntEquals(a + $noinline$IntShl(b, 16),  a + (b << 16));
+    assertIntEquals(a + $noinline$IntShl(b, 30),  a + (b << 30));
+    assertIntEquals(a + $noinline$IntShl(b, 31),  a + (b << 31));
+    assertIntEquals(a + $noinline$IntShl(b, 32),  a + (b << 32));
+    assertIntEquals(a + $noinline$IntShl(b, 62),  a + (b << 62));
+    assertIntEquals(a + $noinline$IntShl(b, 63),  a + (b << 63));
+
+    assertIntEquals(a - $noinline$IntShr(b, 1),   a - (b >>  1));
+    assertIntEquals(a - $noinline$IntShr(b, 6),   a - (b >>  6));
+    assertIntEquals(a - $noinline$IntShr(b, 7),   a - (b >>  7));
+    assertIntEquals(a - $noinline$IntShr(b, 8),   a - (b >>  8));
+    assertIntEquals(a - $noinline$IntShr(b, 14),  a - (b >> 14));
+    assertIntEquals(a - $noinline$IntShr(b, 15),  a - (b >> 15));
+    assertIntEquals(a - $noinline$IntShr(b, 16),  a - (b >> 16));
+    assertIntEquals(a - $noinline$IntShr(b, 30),  a - (b >> 30));
+    assertIntEquals(a - $noinline$IntShr(b, 31),  a - (b >> 31));
+    assertIntEquals(a - $noinline$IntShr(b, 32),  a - (b >> 32));
+    assertIntEquals(a - $noinline$IntShr(b, 62),  a - (b >> 62));
+    assertIntEquals(a - $noinline$IntShr(b, 63),  a - (b >> 63));
+
+    assertIntEquals(a ^ $noinline$IntUshr(b, 1),   a ^ (b >>>  1));
+    assertIntEquals(a ^ $noinline$IntUshr(b, 6),   a ^ (b >>>  6));
+    assertIntEquals(a ^ $noinline$IntUshr(b, 7),   a ^ (b >>>  7));
+    assertIntEquals(a ^ $noinline$IntUshr(b, 8),   a ^ (b >>>  8));
+    assertIntEquals(a ^ $noinline$IntUshr(b, 14),  a ^ (b >>> 14));
+    assertIntEquals(a ^ $noinline$IntUshr(b, 15),  a ^ (b >>> 15));
+    assertIntEquals(a ^ $noinline$IntUshr(b, 16),  a ^ (b >>> 16));
+    assertIntEquals(a ^ $noinline$IntUshr(b, 30),  a ^ (b >>> 30));
+    assertIntEquals(a ^ $noinline$IntUshr(b, 31),  a ^ (b >>> 31));
+    assertIntEquals(a ^ $noinline$IntUshr(b, 32),  a ^ (b >>> 32));
+    assertIntEquals(a ^ $noinline$IntUshr(b, 62),  a ^ (b >>> 62));
+    assertIntEquals(a ^ $noinline$IntUshr(b, 63),  a ^ (b >>> 63));
+  }
+
+
+  static long $noinline$LongShl(long b, long c) {
+    if (doThrow) throw new Error();
+    return b << c;
+  }
+  static long $noinline$LongShr(long b, long c) {
+    if (doThrow) throw new Error();
+    return b >> c;
+  }
+  static long $noinline$LongUshr(long b, long c) {
+    if (doThrow) throw new Error();
+    return b >>> c;
+  }
+
+  // Each test line below should see one merge.
+  /// CHECK-START-ARM64: void Main.$opt$validateShiftLong(long, long) instruction_simplifier_arm64 (after)
+  /// CHECK:                            Arm64DataProcWithShifterOp
+  /// CHECK:                            Arm64DataProcWithShifterOp
+  /// CHECK:                            Arm64DataProcWithShifterOp
+  /// CHECK:                            Arm64DataProcWithShifterOp
+  /// CHECK:                            Arm64DataProcWithShifterOp
+  /// CHECK:                            Arm64DataProcWithShifterOp
+  /// CHECK:                            Arm64DataProcWithShifterOp
+  /// CHECK:                            Arm64DataProcWithShifterOp
+  /// CHECK:                            Arm64DataProcWithShifterOp
+  /// CHECK:                            Arm64DataProcWithShifterOp
+  /// CHECK:                            Arm64DataProcWithShifterOp
+  /// CHECK:                            Arm64DataProcWithShifterOp
+  /// CHECK:                            Arm64DataProcWithShifterOp
+  /// CHECK:                            Arm64DataProcWithShifterOp
+  /// CHECK:                            Arm64DataProcWithShifterOp
+  /// CHECK:                            Arm64DataProcWithShifterOp
+  /// CHECK:                            Arm64DataProcWithShifterOp
+  /// CHECK:                            Arm64DataProcWithShifterOp
+  /// CHECK:                            Arm64DataProcWithShifterOp
+  /// CHECK:                            Arm64DataProcWithShifterOp
+  /// CHECK:                            Arm64DataProcWithShifterOp
+  /// CHECK:                            Arm64DataProcWithShifterOp
+  /// CHECK:                            Arm64DataProcWithShifterOp
+  /// CHECK:                            Arm64DataProcWithShifterOp
+  /// CHECK:                            Arm64DataProcWithShifterOp
+  /// CHECK:                            Arm64DataProcWithShifterOp
+  /// CHECK:                            Arm64DataProcWithShifterOp
+  /// CHECK:                            Arm64DataProcWithShifterOp
+  /// CHECK:                            Arm64DataProcWithShifterOp
+  /// CHECK:                            Arm64DataProcWithShifterOp
+  /// CHECK:                            Arm64DataProcWithShifterOp
+  /// CHECK:                            Arm64DataProcWithShifterOp
+  /// CHECK:                            Arm64DataProcWithShifterOp
+  /// CHECK:                            Arm64DataProcWithShifterOp
+  /// CHECK:                            Arm64DataProcWithShifterOp
+  /// CHECK:                            Arm64DataProcWithShifterOp
+
+  /// CHECK-START-ARM64: void Main.$opt$validateShiftLong(long, long) instruction_simplifier_arm64 (after)
+  /// CHECK-NOT:                        Shl
+  /// CHECK-NOT:                        Shr
+  /// CHECK-NOT:                        UShr
+
+  public static void $opt$validateShiftLong(long a, long b) {
+    assertLongEquals(a + $noinline$LongShl(b, 1),   a + (b <<  1));
+    assertLongEquals(a + $noinline$LongShl(b, 6),   a + (b <<  6));
+    assertLongEquals(a + $noinline$LongShl(b, 7),   a + (b <<  7));
+    assertLongEquals(a + $noinline$LongShl(b, 8),   a + (b <<  8));
+    assertLongEquals(a + $noinline$LongShl(b, 14),  a + (b << 14));
+    assertLongEquals(a + $noinline$LongShl(b, 15),  a + (b << 15));
+    assertLongEquals(a + $noinline$LongShl(b, 16),  a + (b << 16));
+    assertLongEquals(a + $noinline$LongShl(b, 30),  a + (b << 30));
+    assertLongEquals(a + $noinline$LongShl(b, 31),  a + (b << 31));
+    assertLongEquals(a + $noinline$LongShl(b, 32),  a + (b << 32));
+    assertLongEquals(a + $noinline$LongShl(b, 62),  a + (b << 62));
+    assertLongEquals(a + $noinline$LongShl(b, 63),  a + (b << 63));
+
+    assertLongEquals(a - $noinline$LongShr(b, 1),   a - (b >>  1));
+    assertLongEquals(a - $noinline$LongShr(b, 6),   a - (b >>  6));
+    assertLongEquals(a - $noinline$LongShr(b, 7),   a - (b >>  7));
+    assertLongEquals(a - $noinline$LongShr(b, 8),   a - (b >>  8));
+    assertLongEquals(a - $noinline$LongShr(b, 14),  a - (b >> 14));
+    assertLongEquals(a - $noinline$LongShr(b, 15),  a - (b >> 15));
+    assertLongEquals(a - $noinline$LongShr(b, 16),  a - (b >> 16));
+    assertLongEquals(a - $noinline$LongShr(b, 30),  a - (b >> 30));
+    assertLongEquals(a - $noinline$LongShr(b, 31),  a - (b >> 31));
+    assertLongEquals(a - $noinline$LongShr(b, 32),  a - (b >> 32));
+    assertLongEquals(a - $noinline$LongShr(b, 62),  a - (b >> 62));
+    assertLongEquals(a - $noinline$LongShr(b, 63),  a - (b >> 63));
+
+    assertLongEquals(a ^ $noinline$LongUshr(b, 1),   a ^ (b >>>  1));
+    assertLongEquals(a ^ $noinline$LongUshr(b, 6),   a ^ (b >>>  6));
+    assertLongEquals(a ^ $noinline$LongUshr(b, 7),   a ^ (b >>>  7));
+    assertLongEquals(a ^ $noinline$LongUshr(b, 8),   a ^ (b >>>  8));
+    assertLongEquals(a ^ $noinline$LongUshr(b, 14),  a ^ (b >>> 14));
+    assertLongEquals(a ^ $noinline$LongUshr(b, 15),  a ^ (b >>> 15));
+    assertLongEquals(a ^ $noinline$LongUshr(b, 16),  a ^ (b >>> 16));
+    assertLongEquals(a ^ $noinline$LongUshr(b, 30),  a ^ (b >>> 30));
+    assertLongEquals(a ^ $noinline$LongUshr(b, 31),  a ^ (b >>> 31));
+    assertLongEquals(a ^ $noinline$LongUshr(b, 32),  a ^ (b >>> 32));
+    assertLongEquals(a ^ $noinline$LongUshr(b, 62),  a ^ (b >>> 62));
+    assertLongEquals(a ^ $noinline$LongUshr(b, 63),  a ^ (b >>> 63));
+  }
+
+
+  public static void main(String[] args) {
+    assertLongEquals(10000L - 3L, $opt$noinline$translate(10000L, (byte)3));
+    assertLongEquals(-10000L - -3L, $opt$noinline$translate(-10000L, (byte)-3));
+
+    assertIntEquals(4096, $opt$noinline$sameInput(512));
+    assertIntEquals(-8192, $opt$noinline$sameInput(-1024));
+
+    assertIntEquals(((1 << 23) | 1), $opt$noinline$multipleUses(1));
+    assertIntEquals(((1 << 20) | 5), $opt$noinline$multipleUses(1 << 20));
+
+    long inputs[] = {
+      -((1L <<  7) - 1L), -((1L <<  7)), -((1L <<  7) + 1L),
+      -((1L << 15) - 1L), -((1L << 15)), -((1L << 15) + 1L),
+      -((1L << 16) - 1L), -((1L << 16)), -((1L << 16) + 1L),
+      -((1L << 31) - 1L), -((1L << 31)), -((1L << 31) + 1L),
+      -((1L << 32) - 1L), -((1L << 32)), -((1L << 32) + 1L),
+      -((1L << 63) - 1L), -((1L << 63)), -((1L << 63) + 1L),
+      -42L, -314L, -2718281828L, -0x123456789L, -0x987654321L,
+      -1L, -20L, -300L, -4000L, -50000L, -600000L, -7000000L, -80000000L,
+      0L,
+      1L, 20L, 300L, 4000L, 50000L, 600000L, 7000000L, 80000000L,
+      42L,  314L,  2718281828L,  0x123456789L,  0x987654321L,
+      (1L <<  7) - 1L, (1L <<  7), (1L <<  7) + 1L,
+      (1L <<  8) - 1L, (1L <<  8), (1L <<  8) + 1L,
+      (1L << 15) - 1L, (1L << 15), (1L << 15) + 1L,
+      (1L << 16) - 1L, (1L << 16), (1L << 16) + 1L,
+      (1L << 31) - 1L, (1L << 31), (1L << 31) + 1L,
+      (1L << 32) - 1L, (1L << 32), (1L << 32) + 1L,
+      (1L << 63) - 1L, (1L << 63), (1L << 63) + 1L,
+      Long.MIN_VALUE, Long.MAX_VALUE
+    };
+    for (int i = 0; i < inputs.length; i++) {
+      $opt$noinline$testNeg((int)inputs[i]);
+      for (int j = 0; j < inputs.length; j++) {
+        $opt$noinline$testAnd(inputs[i], inputs[j]);
+        $opt$noinline$testOr((int)inputs[i], (int)inputs[j]);
+        $opt$noinline$testXor(inputs[i], inputs[j]);
+
+        $opt$validateExtendByte(inputs[i], (byte)inputs[j]);
+        $opt$validateExtendChar(inputs[i], (char)inputs[j]);
+        $opt$validateExtendShort(inputs[i], (short)inputs[j]);
+        $opt$validateExtendInt(inputs[i], (int)inputs[j]);
+        $opt$validateExtendLong(inputs[i], inputs[j]);
+
+        $opt$validateShiftInt((int)inputs[i], (int)inputs[j]);
+        $opt$validateShiftLong(inputs[i], inputs[j]);
+      }
+    }
+
+  }
+}