Merge "ART: Introduce Uint8 compiler data type."
diff --git a/compiler/optimizing/code_generator_arm64.cc b/compiler/optimizing/code_generator_arm64.cc
index 42e9f68..468e93a 100644
--- a/compiler/optimizing/code_generator_arm64.cc
+++ b/compiler/optimizing/code_generator_arm64.cc
@@ -1929,17 +1929,18 @@
                               const MemOperand& src) {
   switch (type) {
     case DataType::Type::kBool:
+    case DataType::Type::kUint8:
       __ Ldrb(Register(dst), src);
       break;
     case DataType::Type::kInt8:
       __ Ldrsb(Register(dst), src);
       break;
-    case DataType::Type::kInt16:
-      __ Ldrsh(Register(dst), src);
-      break;
     case DataType::Type::kUint16:
       __ Ldrh(Register(dst), src);
       break;
+    case DataType::Type::kInt16:
+      __ Ldrsh(Register(dst), src);
+      break;
     case DataType::Type::kInt32:
     case DataType::Type::kReference:
     case DataType::Type::kInt64:
@@ -1972,14 +1973,7 @@
     MemOperand base = MemOperand(temp_base);
     switch (type) {
       case DataType::Type::kBool:
-        {
-          ExactAssemblyScope eas(masm, kInstructionSize, CodeBufferCheckScope::kExactSize);
-          __ ldarb(Register(dst), base);
-          if (needs_null_check) {
-            MaybeRecordImplicitNullCheck(instruction);
-          }
-        }
-        break;
+      case DataType::Type::kUint8:
       case DataType::Type::kInt8:
         {
           ExactAssemblyScope eas(masm, kInstructionSize, CodeBufferCheckScope::kExactSize);
@@ -1988,17 +1982,11 @@
             MaybeRecordImplicitNullCheck(instruction);
           }
         }
-        __ Sbfx(Register(dst), Register(dst), 0, DataType::Size(type) * kBitsPerByte);
-        break;
-      case DataType::Type::kUint16:
-        {
-          ExactAssemblyScope eas(masm, kInstructionSize, CodeBufferCheckScope::kExactSize);
-          __ ldarh(Register(dst), base);
-          if (needs_null_check) {
-            MaybeRecordImplicitNullCheck(instruction);
-          }
+        if (type == DataType::Type::kInt8) {
+          __ Sbfx(Register(dst), Register(dst), 0, DataType::Size(type) * kBitsPerByte);
         }
         break;
+      case DataType::Type::kUint16:
       case DataType::Type::kInt16:
         {
           ExactAssemblyScope eas(masm, kInstructionSize, CodeBufferCheckScope::kExactSize);
@@ -2007,7 +1995,9 @@
             MaybeRecordImplicitNullCheck(instruction);
           }
         }
-        __ Sbfx(Register(dst), Register(dst), 0, DataType::Size(type) * kBitsPerByte);
+        if (type == DataType::Type::kInt16) {
+          __ Sbfx(Register(dst), Register(dst), 0, DataType::Size(type) * kBitsPerByte);
+        }
         break;
       case DataType::Type::kInt32:
       case DataType::Type::kReference:
@@ -2048,6 +2038,7 @@
                                const MemOperand& dst) {
   switch (type) {
     case DataType::Type::kBool:
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
       __ Strb(Register(src), dst);
       break;
@@ -2087,6 +2078,7 @@
   // Ensure that between store and MaybeRecordImplicitNullCheck there are no pools emitted.
   switch (type) {
     case DataType::Type::kBool:
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
       {
         ExactAssemblyScope eas(masm, kInstructionSize, CodeBufferCheckScope::kExactSize);
@@ -3222,9 +3214,10 @@
   DataType::Type in_type = compare->InputAt(0)->GetType();
   switch (in_type) {
     case DataType::Type::kBool:
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
-    case DataType::Type::kInt16:
     case DataType::Type::kUint16:
+    case DataType::Type::kInt16:
     case DataType::Type::kInt32:
     case DataType::Type::kInt64: {
       locations->SetInAt(0, Location::RequiresRegister());
@@ -3255,9 +3248,10 @@
   // -1 if: left  < right
   switch (in_type) {
     case DataType::Type::kBool:
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
-    case DataType::Type::kInt16:
     case DataType::Type::kUint16:
+    case DataType::Type::kInt16:
     case DataType::Type::kInt32:
     case DataType::Type::kInt64: {
       Register result = OutputRegister(compare);
@@ -5744,7 +5738,8 @@
       new (GetGraph()->GetArena()) LocationSummary(conversion, LocationSummary::kNoCall);
   DataType::Type input_type = conversion->GetInputType();
   DataType::Type result_type = conversion->GetResultType();
-  DCHECK_NE(input_type, result_type);
+  DCHECK(!DataType::IsTypeConversionImplicit(input_type, result_type))
+      << input_type << " -> " << result_type;
   if ((input_type == DataType::Type::kReference) || (input_type == DataType::Type::kVoid) ||
       (result_type == DataType::Type::kReference) || (result_type == DataType::Type::kVoid)) {
     LOG(FATAL) << "Unexpected type conversion from " << input_type << " to " << result_type;
@@ -5767,7 +5762,8 @@
   DataType::Type result_type = conversion->GetResultType();
   DataType::Type input_type = conversion->GetInputType();
 
-  DCHECK_NE(input_type, result_type);
+  DCHECK(!DataType::IsTypeConversionImplicit(input_type, result_type))
+      << input_type << " -> " << result_type;
 
   if (DataType::IsIntegralType(result_type) && DataType::IsIntegralType(input_type)) {
     int result_size = DataType::Size(result_type);
@@ -5784,11 +5780,9 @@
       // 32bit input value as a 64bit value assuming that the top 32 bits are
       // zero.
       __ Mov(output.W(), source.W());
-    } else if (result_type == DataType::Type::kUint16 ||
-               (input_type == DataType::Type::kUint16 && input_size < result_size)) {
-      __ Ubfx(output,
-              output.IsX() ? source.X() : source.W(),
-              0, DataType::Size(DataType::Type::kUint16) * kBitsPerByte);
+    } else if (DataType::IsUnsignedType(result_type) ||
+               (DataType::IsUnsignedType(input_type) && input_size < result_size)) {
+      __ Ubfx(output, output.IsX() ? source.X() : source.W(), 0, result_size * kBitsPerByte);
     } else {
       __ Sbfx(output, output.IsX() ? source.X() : source.W(), 0, min_size * kBitsPerByte);
     }
diff --git a/compiler/optimizing/code_generator_arm_vixl.cc b/compiler/optimizing/code_generator_arm_vixl.cc
index 2b9e0fe..d4fb064 100644
--- a/compiler/optimizing/code_generator_arm_vixl.cc
+++ b/compiler/optimizing/code_generator_arm_vixl.cc
@@ -282,6 +282,58 @@
   return stack_offset;
 }
 
+static LoadOperandType GetLoadOperandType(DataType::Type type) {
+  switch (type) {
+    case DataType::Type::kReference:
+      return kLoadWord;
+    case DataType::Type::kBool:
+    case DataType::Type::kUint8:
+      return kLoadUnsignedByte;
+    case DataType::Type::kInt8:
+      return kLoadSignedByte;
+    case DataType::Type::kUint16:
+      return kLoadUnsignedHalfword;
+    case DataType::Type::kInt16:
+      return kLoadSignedHalfword;
+    case DataType::Type::kInt32:
+      return kLoadWord;
+    case DataType::Type::kInt64:
+      return kLoadWordPair;
+    case DataType::Type::kFloat32:
+      return kLoadSWord;
+    case DataType::Type::kFloat64:
+      return kLoadDWord;
+    default:
+      LOG(FATAL) << "Unreachable type " << type;
+      UNREACHABLE();
+  }
+}
+
+static StoreOperandType GetStoreOperandType(DataType::Type type) {
+  switch (type) {
+    case DataType::Type::kReference:
+      return kStoreWord;
+    case DataType::Type::kBool:
+    case DataType::Type::kUint8:
+    case DataType::Type::kInt8:
+      return kStoreByte;
+    case DataType::Type::kUint16:
+    case DataType::Type::kInt16:
+      return kStoreHalfword;
+    case DataType::Type::kInt32:
+      return kStoreWord;
+    case DataType::Type::kInt64:
+      return kStoreWordPair;
+    case DataType::Type::kFloat32:
+      return kStoreSWord;
+    case DataType::Type::kFloat64:
+      return kStoreDWord;
+    default:
+      LOG(FATAL) << "Unreachable type " << type;
+      UNREACHABLE();
+  }
+}
+
 void SlowPathCodeARMVIXL::SaveLiveRegisters(CodeGenerator* codegen, LocationSummary* locations) {
   size_t stack_offset = codegen->GetFirstRegisterSlotInSlowPath();
   size_t orig_offset = stack_offset;
@@ -2598,12 +2650,13 @@
 
 Location InvokeDexCallingConventionVisitorARMVIXL::GetNextLocation(DataType::Type type) {
   switch (type) {
+    case DataType::Type::kReference:
     case DataType::Type::kBool:
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
     case DataType::Type::kUint16:
     case DataType::Type::kInt16:
-    case DataType::Type::kInt32:
-    case DataType::Type::kReference: {
+    case DataType::Type::kInt32: {
       uint32_t index = gp_index_++;
       uint32_t stack_index = stack_index_++;
       if (index < calling_convention.GetNumberOfRegisters()) {
@@ -2674,12 +2727,13 @@
 
 Location InvokeDexCallingConventionVisitorARMVIXL::GetReturnLocation(DataType::Type type) const {
   switch (type) {
+    case DataType::Type::kReference:
     case DataType::Type::kBool:
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
     case DataType::Type::kUint16:
     case DataType::Type::kInt16:
-    case DataType::Type::kInt32:
-    case DataType::Type::kReference: {
+    case DataType::Type::kInt32: {
       return LocationFrom(r0);
     }
 
@@ -3728,7 +3782,8 @@
 void LocationsBuilderARMVIXL::VisitTypeConversion(HTypeConversion* conversion) {
   DataType::Type result_type = conversion->GetResultType();
   DataType::Type input_type = conversion->GetInputType();
-  DCHECK_NE(result_type, input_type);
+  DCHECK(!DataType::IsTypeConversionImplicit(input_type, result_type))
+      << input_type << " -> " << result_type;
 
   // The float-to-long, double-to-long and long-to-float type conversions
   // rely on a call to the runtime.
@@ -3741,67 +3796,30 @@
   LocationSummary* locations =
       new (GetGraph()->GetArena()) LocationSummary(conversion, call_kind);
 
-  // The Java language does not allow treating boolean as an integral type but
-  // our bit representation makes it safe.
-
   switch (result_type) {
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
-      switch (input_type) {
-        case DataType::Type::kInt64:
-          // Type conversion from long to byte is a result of code transformations.
-        case DataType::Type::kBool:
-          // Boolean input is a result of code transformations.
-        case DataType::Type::kInt16:
-        case DataType::Type::kInt32:
-        case DataType::Type::kUint16:
-          // Processing a Dex `int-to-byte' instruction.
-          locations->SetInAt(0, Location::RequiresRegister());
-          locations->SetOut(Location::RequiresRegister(), Location::kNoOutputOverlap);
-          break;
-
-        default:
-          LOG(FATAL) << "Unexpected type conversion from " << input_type
-                     << " to " << result_type;
-      }
-      break;
-
+    case DataType::Type::kUint16:
     case DataType::Type::kInt16:
-      switch (input_type) {
-        case DataType::Type::kInt64:
-          // Type conversion from long to short is a result of code transformations.
-        case DataType::Type::kBool:
-          // Boolean input is a result of code transformations.
-        case DataType::Type::kInt8:
-        case DataType::Type::kInt32:
-        case DataType::Type::kUint16:
-          // Processing a Dex `int-to-short' instruction.
-          locations->SetInAt(0, Location::RequiresRegister());
-          locations->SetOut(Location::RequiresRegister(), Location::kNoOutputOverlap);
-          break;
-
-        default:
-          LOG(FATAL) << "Unexpected type conversion from " << input_type
-                     << " to " << result_type;
-      }
+      DCHECK(DataType::IsIntegralType(input_type)) << input_type;
+      locations->SetInAt(0, Location::RequiresRegister());
+      locations->SetOut(Location::RequiresRegister(), Location::kNoOutputOverlap);
       break;
 
     case DataType::Type::kInt32:
       switch (input_type) {
         case DataType::Type::kInt64:
-          // Processing a Dex `long-to-int' instruction.
           locations->SetInAt(0, Location::Any());
           locations->SetOut(Location::RequiresRegister(), Location::kNoOutputOverlap);
           break;
 
         case DataType::Type::kFloat32:
-          // Processing a Dex `float-to-int' instruction.
           locations->SetInAt(0, Location::RequiresFpuRegister());
           locations->SetOut(Location::RequiresRegister());
           locations->AddTemp(Location::RequiresFpuRegister());
           break;
 
         case DataType::Type::kFloat64:
-          // Processing a Dex `double-to-int' instruction.
           locations->SetInAt(0, Location::RequiresFpuRegister());
           locations->SetOut(Location::RequiresRegister());
           locations->AddTemp(Location::RequiresFpuRegister());
@@ -3816,18 +3834,16 @@
     case DataType::Type::kInt64:
       switch (input_type) {
         case DataType::Type::kBool:
-          // Boolean input is a result of code transformations.
+        case DataType::Type::kUint8:
         case DataType::Type::kInt8:
+        case DataType::Type::kUint16:
         case DataType::Type::kInt16:
         case DataType::Type::kInt32:
-        case DataType::Type::kUint16:
-          // Processing a Dex `int-to-long' instruction.
           locations->SetInAt(0, Location::RequiresRegister());
           locations->SetOut(Location::RequiresRegister(), Location::kNoOutputOverlap);
           break;
 
         case DataType::Type::kFloat32: {
-          // Processing a Dex `float-to-long' instruction.
           InvokeRuntimeCallingConventionARMVIXL calling_convention;
           locations->SetInAt(0, LocationFrom(calling_convention.GetFpuRegisterAt(0)));
           locations->SetOut(LocationFrom(r0, r1));
@@ -3835,7 +3851,6 @@
         }
 
         case DataType::Type::kFloat64: {
-          // Processing a Dex `double-to-long' instruction.
           InvokeRuntimeCallingConventionARMVIXL calling_convention;
           locations->SetInAt(0, LocationFrom(calling_convention.GetFpuRegisterAt(0),
                                              calling_convention.GetFpuRegisterAt(1)));
@@ -3849,41 +3864,19 @@
       }
       break;
 
-    case DataType::Type::kUint16:
-      switch (input_type) {
-        case DataType::Type::kInt64:
-          // Type conversion from long to char is a result of code transformations.
-        case DataType::Type::kBool:
-          // Boolean input is a result of code transformations.
-        case DataType::Type::kInt8:
-        case DataType::Type::kInt16:
-        case DataType::Type::kInt32:
-          // Processing a Dex `int-to-char' instruction.
-          locations->SetInAt(0, Location::RequiresRegister());
-          locations->SetOut(Location::RequiresRegister(), Location::kNoOutputOverlap);
-          break;
-
-        default:
-          LOG(FATAL) << "Unexpected type conversion from " << input_type
-                     << " to " << result_type;
-      }
-      break;
-
     case DataType::Type::kFloat32:
       switch (input_type) {
         case DataType::Type::kBool:
-          // Boolean input is a result of code transformations.
+        case DataType::Type::kUint8:
         case DataType::Type::kInt8:
+        case DataType::Type::kUint16:
         case DataType::Type::kInt16:
         case DataType::Type::kInt32:
-        case DataType::Type::kUint16:
-          // Processing a Dex `int-to-float' instruction.
           locations->SetInAt(0, Location::RequiresRegister());
           locations->SetOut(Location::RequiresFpuRegister());
           break;
 
         case DataType::Type::kInt64: {
-          // Processing a Dex `long-to-float' instruction.
           InvokeRuntimeCallingConventionARMVIXL calling_convention;
           locations->SetInAt(0, LocationFrom(calling_convention.GetRegisterAt(0),
                                              calling_convention.GetRegisterAt(1)));
@@ -3892,7 +3885,6 @@
         }
 
         case DataType::Type::kFloat64:
-          // Processing a Dex `double-to-float' instruction.
           locations->SetInAt(0, Location::RequiresFpuRegister());
           locations->SetOut(Location::RequiresFpuRegister(), Location::kNoOutputOverlap);
           break;
@@ -3906,18 +3898,16 @@
     case DataType::Type::kFloat64:
       switch (input_type) {
         case DataType::Type::kBool:
-          // Boolean input is a result of code transformations.
+        case DataType::Type::kUint8:
         case DataType::Type::kInt8:
+        case DataType::Type::kUint16:
         case DataType::Type::kInt16:
         case DataType::Type::kInt32:
-        case DataType::Type::kUint16:
-          // Processing a Dex `int-to-double' instruction.
           locations->SetInAt(0, Location::RequiresRegister());
           locations->SetOut(Location::RequiresFpuRegister());
           break;
 
         case DataType::Type::kInt64:
-          // Processing a Dex `long-to-double' instruction.
           locations->SetInAt(0, Location::RequiresRegister());
           locations->SetOut(Location::RequiresFpuRegister());
           locations->AddTemp(Location::RequiresFpuRegister());
@@ -3925,7 +3915,6 @@
           break;
 
         case DataType::Type::kFloat32:
-          // Processing a Dex `float-to-double' instruction.
           locations->SetInAt(0, Location::RequiresFpuRegister());
           locations->SetOut(Location::RequiresFpuRegister(), Location::kNoOutputOverlap);
           break;
@@ -3948,22 +3937,55 @@
   Location in = locations->InAt(0);
   DataType::Type result_type = conversion->GetResultType();
   DataType::Type input_type = conversion->GetInputType();
-  DCHECK_NE(result_type, input_type);
+  DCHECK(!DataType::IsTypeConversionImplicit(input_type, result_type))
+      << input_type << " -> " << result_type;
   switch (result_type) {
-    case DataType::Type::kInt8:
+    case DataType::Type::kUint8:
       switch (input_type) {
-        case DataType::Type::kInt64:
-          // Type conversion from long to byte is a result of code transformations.
-          __ Sbfx(OutputRegister(conversion), LowRegisterFrom(in), 0, 8);
-          break;
-        case DataType::Type::kBool:
-          // Boolean input is a result of code transformations.
+        case DataType::Type::kInt8:
+        case DataType::Type::kUint16:
         case DataType::Type::kInt16:
         case DataType::Type::kInt32:
+          __ Ubfx(OutputRegister(conversion), InputRegisterAt(conversion, 0), 0, 8);
+          break;
+        case DataType::Type::kInt64:
+          __ Ubfx(OutputRegister(conversion), LowRegisterFrom(in), 0, 8);
+          break;
+
+        default:
+          LOG(FATAL) << "Unexpected type conversion from " << input_type
+                     << " to " << result_type;
+      }
+      break;
+
+    case DataType::Type::kInt8:
+      switch (input_type) {
+        case DataType::Type::kUint8:
         case DataType::Type::kUint16:
-          // Processing a Dex `int-to-byte' instruction.
+        case DataType::Type::kInt16:
+        case DataType::Type::kInt32:
           __ Sbfx(OutputRegister(conversion), InputRegisterAt(conversion, 0), 0, 8);
           break;
+        case DataType::Type::kInt64:
+          __ Sbfx(OutputRegister(conversion), LowRegisterFrom(in), 0, 8);
+          break;
+
+        default:
+          LOG(FATAL) << "Unexpected type conversion from " << input_type
+                     << " to " << result_type;
+      }
+      break;
+
+    case DataType::Type::kUint16:
+      switch (input_type) {
+        case DataType::Type::kInt8:
+        case DataType::Type::kInt16:
+        case DataType::Type::kInt32:
+          __ Ubfx(OutputRegister(conversion), InputRegisterAt(conversion, 0), 0, 16);
+          break;
+        case DataType::Type::kInt64:
+          __ Ubfx(OutputRegister(conversion), LowRegisterFrom(in), 0, 16);
+          break;
 
         default:
           LOG(FATAL) << "Unexpected type conversion from " << input_type
@@ -3973,18 +3995,13 @@
 
     case DataType::Type::kInt16:
       switch (input_type) {
-        case DataType::Type::kInt64:
-          // Type conversion from long to short is a result of code transformations.
-          __ Sbfx(OutputRegister(conversion), LowRegisterFrom(in), 0, 16);
-          break;
-        case DataType::Type::kBool:
-          // Boolean input is a result of code transformations.
-        case DataType::Type::kInt8:
-        case DataType::Type::kInt32:
         case DataType::Type::kUint16:
-          // Processing a Dex `int-to-short' instruction.
+        case DataType::Type::kInt32:
           __ Sbfx(OutputRegister(conversion), InputRegisterAt(conversion, 0), 0, 16);
           break;
+        case DataType::Type::kInt64:
+          __ Sbfx(OutputRegister(conversion), LowRegisterFrom(in), 0, 16);
+          break;
 
         default:
           LOG(FATAL) << "Unexpected type conversion from " << input_type
@@ -3995,7 +4012,6 @@
     case DataType::Type::kInt32:
       switch (input_type) {
         case DataType::Type::kInt64:
-          // Processing a Dex `long-to-int' instruction.
           DCHECK(out.IsRegister());
           if (in.IsRegisterPair()) {
             __ Mov(OutputRegister(conversion), LowRegisterFrom(in));
@@ -4013,7 +4029,6 @@
           break;
 
         case DataType::Type::kFloat32: {
-          // Processing a Dex `float-to-int' instruction.
           vixl32::SRegister temp = LowSRegisterFrom(locations->GetTemp(0));
           __ Vcvt(S32, F32, temp, InputSRegisterAt(conversion, 0));
           __ Vmov(OutputRegister(conversion), temp);
@@ -4021,7 +4036,6 @@
         }
 
         case DataType::Type::kFloat64: {
-          // Processing a Dex `double-to-int' instruction.
           vixl32::SRegister temp_s = LowSRegisterFrom(locations->GetTemp(0));
           __ Vcvt(S32, F64, temp_s, DRegisterFrom(in));
           __ Vmov(OutputRegister(conversion), temp_s);
@@ -4037,12 +4051,11 @@
     case DataType::Type::kInt64:
       switch (input_type) {
         case DataType::Type::kBool:
-          // Boolean input is a result of code transformations.
+        case DataType::Type::kUint8:
         case DataType::Type::kInt8:
+        case DataType::Type::kUint16:
         case DataType::Type::kInt16:
         case DataType::Type::kInt32:
-        case DataType::Type::kUint16:
-          // Processing a Dex `int-to-long' instruction.
           DCHECK(out.IsRegisterPair());
           DCHECK(in.IsRegister());
           __ Mov(LowRegisterFrom(out), InputRegisterAt(conversion, 0));
@@ -4051,13 +4064,11 @@
           break;
 
         case DataType::Type::kFloat32:
-          // Processing a Dex `float-to-long' instruction.
           codegen_->InvokeRuntime(kQuickF2l, conversion, conversion->GetDexPc());
           CheckEntrypointTypes<kQuickF2l, int64_t, float>();
           break;
 
         case DataType::Type::kFloat64:
-          // Processing a Dex `double-to-long' instruction.
           codegen_->InvokeRuntime(kQuickD2l, conversion, conversion->GetDexPc());
           CheckEntrypointTypes<kQuickD2l, int64_t, double>();
           break;
@@ -4068,49 +4079,24 @@
       }
       break;
 
-    case DataType::Type::kUint16:
-      switch (input_type) {
-        case DataType::Type::kInt64:
-          // Type conversion from long to char is a result of code transformations.
-          __ Ubfx(OutputRegister(conversion), LowRegisterFrom(in), 0, 16);
-          break;
-        case DataType::Type::kBool:
-          // Boolean input is a result of code transformations.
-        case DataType::Type::kInt8:
-        case DataType::Type::kInt16:
-        case DataType::Type::kInt32:
-          // Processing a Dex `int-to-char' instruction.
-          __ Ubfx(OutputRegister(conversion), InputRegisterAt(conversion, 0), 0, 16);
-          break;
-
-        default:
-          LOG(FATAL) << "Unexpected type conversion from " << input_type
-                     << " to " << result_type;
-      }
-      break;
-
     case DataType::Type::kFloat32:
       switch (input_type) {
         case DataType::Type::kBool:
-          // Boolean input is a result of code transformations.
+        case DataType::Type::kUint8:
         case DataType::Type::kInt8:
+        case DataType::Type::kUint16:
         case DataType::Type::kInt16:
         case DataType::Type::kInt32:
-        case DataType::Type::kUint16: {
-          // Processing a Dex `int-to-float' instruction.
           __ Vmov(OutputSRegister(conversion), InputRegisterAt(conversion, 0));
           __ Vcvt(F32, S32, OutputSRegister(conversion), OutputSRegister(conversion));
           break;
-        }
 
         case DataType::Type::kInt64:
-          // Processing a Dex `long-to-float' instruction.
           codegen_->InvokeRuntime(kQuickL2f, conversion, conversion->GetDexPc());
           CheckEntrypointTypes<kQuickL2f, float, int64_t>();
           break;
 
         case DataType::Type::kFloat64:
-          // Processing a Dex `double-to-float' instruction.
           __ Vcvt(F32, F64, OutputSRegister(conversion), DRegisterFrom(in));
           break;
 
@@ -4123,19 +4109,16 @@
     case DataType::Type::kFloat64:
       switch (input_type) {
         case DataType::Type::kBool:
-          // Boolean input is a result of code transformations.
+        case DataType::Type::kUint8:
         case DataType::Type::kInt8:
+        case DataType::Type::kUint16:
         case DataType::Type::kInt16:
         case DataType::Type::kInt32:
-        case DataType::Type::kUint16: {
-          // Processing a Dex `int-to-double' instruction.
           __ Vmov(LowSRegisterFrom(out), InputRegisterAt(conversion, 0));
           __ Vcvt(F64, S32, DRegisterFrom(out), LowSRegisterFrom(out));
           break;
-        }
 
         case DataType::Type::kInt64: {
-          // Processing a Dex `long-to-double' instruction.
           vixl32::Register low = LowRegisterFrom(in);
           vixl32::Register high = HighRegisterFrom(in);
           vixl32::SRegister out_s = LowSRegisterFrom(out);
@@ -4158,7 +4141,6 @@
         }
 
         case DataType::Type::kFloat32:
-          // Processing a Dex `float-to-double' instruction.
           __ Vcvt(F64, F32, DRegisterFrom(out), InputSRegisterAt(conversion, 0));
           break;
 
@@ -4760,6 +4742,7 @@
 
   switch (instruction->GetType()) {
     case DataType::Type::kBool:
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
     case DataType::Type::kUint16:
     case DataType::Type::kInt16:
@@ -5288,9 +5271,10 @@
       new (GetGraph()->GetArena()) LocationSummary(compare, LocationSummary::kNoCall);
   switch (compare->InputAt(0)->GetType()) {
     case DataType::Type::kBool:
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
-    case DataType::Type::kInt16:
     case DataType::Type::kUint16:
+    case DataType::Type::kInt16:
     case DataType::Type::kInt32:
     case DataType::Type::kInt64: {
       locations->SetInAt(0, Location::RequiresRegister());
@@ -5323,9 +5307,10 @@
   vixl32::Condition less_cond = vixl32::Condition(kNone);
   switch (type) {
     case DataType::Type::kBool:
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
-    case DataType::Type::kInt16:
     case DataType::Type::kUint16:
+    case DataType::Type::kInt16:
     case DataType::Type::kInt32: {
       // Emit move to `out` before the `Cmp`, as `Mov` might affect the status flags.
       __ Mov(out, 0);
@@ -5513,18 +5498,16 @@
 
   switch (field_type) {
     case DataType::Type::kBool:
-    case DataType::Type::kInt8: {
-      GetAssembler()->StoreToOffset(kStoreByte, RegisterFrom(value), base, offset);
-      break;
-    }
-
+    case DataType::Type::kUint8:
+    case DataType::Type::kInt8:
+    case DataType::Type::kUint16:
     case DataType::Type::kInt16:
-    case DataType::Type::kUint16: {
-      GetAssembler()->StoreToOffset(kStoreHalfword, RegisterFrom(value), base, offset);
+    case DataType::Type::kInt32: {
+      StoreOperandType operand_type = GetStoreOperandType(field_type);
+      GetAssembler()->StoreToOffset(operand_type, RegisterFrom(value), base, offset);
       break;
     }
 
-    case DataType::Type::kInt32:
     case DataType::Type::kReference: {
       if (kPoisonHeapReferences && needs_write_barrier) {
         // Note that in the case where `value` is a null reference,
@@ -5764,24 +5747,15 @@
 
   switch (field_type) {
     case DataType::Type::kBool:
-      GetAssembler()->LoadFromOffset(kLoadUnsignedByte, RegisterFrom(out), base, offset);
-      break;
-
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
-      GetAssembler()->LoadFromOffset(kLoadSignedByte, RegisterFrom(out), base, offset);
-      break;
-
-    case DataType::Type::kInt16:
-      GetAssembler()->LoadFromOffset(kLoadSignedHalfword, RegisterFrom(out), base, offset);
-      break;
-
     case DataType::Type::kUint16:
-      GetAssembler()->LoadFromOffset(kLoadUnsignedHalfword, RegisterFrom(out), base, offset);
+    case DataType::Type::kInt16:
+    case DataType::Type::kInt32: {
+      LoadOperandType operand_type = GetLoadOperandType(field_type);
+      GetAssembler()->LoadFromOffset(operand_type, RegisterFrom(out), base, offset);
       break;
-
-    case DataType::Type::kInt32:
-      GetAssembler()->LoadFromOffset(kLoadWord, RegisterFrom(out), base, offset);
-      break;
+    }
 
     case DataType::Type::kReference: {
       // /* HeapReference<Object> */ out = *(base + offset)
@@ -5995,56 +5969,6 @@
   codegen_->GenerateNullCheck(instruction);
 }
 
-static LoadOperandType GetLoadOperandType(DataType::Type type) {
-  switch (type) {
-    case DataType::Type::kReference:
-      return kLoadWord;
-    case DataType::Type::kBool:
-      return kLoadUnsignedByte;
-    case DataType::Type::kInt8:
-      return kLoadSignedByte;
-    case DataType::Type::kUint16:
-      return kLoadUnsignedHalfword;
-    case DataType::Type::kInt16:
-      return kLoadSignedHalfword;
-    case DataType::Type::kInt32:
-      return kLoadWord;
-    case DataType::Type::kInt64:
-      return kLoadWordPair;
-    case DataType::Type::kFloat32:
-      return kLoadSWord;
-    case DataType::Type::kFloat64:
-      return kLoadDWord;
-    default:
-      LOG(FATAL) << "Unreachable type " << type;
-      UNREACHABLE();
-  }
-}
-
-static StoreOperandType GetStoreOperandType(DataType::Type type) {
-  switch (type) {
-    case DataType::Type::kReference:
-      return kStoreWord;
-    case DataType::Type::kBool:
-    case DataType::Type::kInt8:
-      return kStoreByte;
-    case DataType::Type::kUint16:
-    case DataType::Type::kInt16:
-      return kStoreHalfword;
-    case DataType::Type::kInt32:
-      return kStoreWord;
-    case DataType::Type::kInt64:
-      return kStoreWordPair;
-    case DataType::Type::kFloat32:
-      return kStoreSWord;
-    case DataType::Type::kFloat64:
-      return kStoreDWord;
-    default:
-      LOG(FATAL) << "Unreachable type " << type;
-      UNREACHABLE();
-  }
-}
-
 void CodeGeneratorARMVIXL::LoadFromShiftedRegOffset(DataType::Type type,
                                                     Location out_loc,
                                                     vixl32::Register base,
@@ -6054,18 +5978,19 @@
   MemOperand mem_address(base, reg_index, vixl32::LSL, shift_count);
 
   switch (type) {
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
       __ Ldrsb(cond, RegisterFrom(out_loc), mem_address);
       break;
     case DataType::Type::kBool:
       __ Ldrb(cond, RegisterFrom(out_loc), mem_address);
       break;
-    case DataType::Type::kInt16:
-      __ Ldrsh(cond, RegisterFrom(out_loc), mem_address);
-      break;
     case DataType::Type::kUint16:
       __ Ldrh(cond, RegisterFrom(out_loc), mem_address);
       break;
+    case DataType::Type::kInt16:
+      __ Ldrsh(cond, RegisterFrom(out_loc), mem_address);
+      break;
     case DataType::Type::kReference:
     case DataType::Type::kInt32:
       __ Ldr(cond, RegisterFrom(out_loc), mem_address);
@@ -6089,12 +6014,13 @@
   MemOperand mem_address(base, reg_index, vixl32::LSL, shift_count);
 
   switch (type) {
-    case DataType::Type::kInt8:
     case DataType::Type::kBool:
+    case DataType::Type::kUint8:
+    case DataType::Type::kInt8:
       __ Strb(cond, RegisterFrom(loc), mem_address);
       break;
-    case DataType::Type::kInt16:
     case DataType::Type::kUint16:
+    case DataType::Type::kInt16:
       __ Strh(cond, RegisterFrom(loc), mem_address);
       break;
     case DataType::Type::kReference:
@@ -6182,9 +6108,10 @@
 
   switch (type) {
     case DataType::Type::kBool:
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
-    case DataType::Type::kInt16:
     case DataType::Type::kUint16:
+    case DataType::Type::kInt16:
     case DataType::Type::kInt32: {
       vixl32::Register length;
       if (maybe_compressed_char_at) {
@@ -6434,9 +6361,10 @@
 
   switch (value_type) {
     case DataType::Type::kBool:
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
-    case DataType::Type::kInt16:
     case DataType::Type::kUint16:
+    case DataType::Type::kInt16:
     case DataType::Type::kInt32: {
       if (index.IsConstant()) {
         int32_t const_index = Int32ConstantFrom(index);
diff --git a/compiler/optimizing/code_generator_mips.cc b/compiler/optimizing/code_generator_mips.cc
index a7c8557..70c8a5b 100644
--- a/compiler/optimizing/code_generator_mips.cc
+++ b/compiler/optimizing/code_generator_mips.cc
@@ -51,12 +51,13 @@
 
 Location MipsReturnLocation(DataType::Type return_type) {
   switch (return_type) {
+    case DataType::Type::kReference:
     case DataType::Type::kBool:
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
     case DataType::Type::kUint16:
     case DataType::Type::kInt16:
     case DataType::Type::kInt32:
-    case DataType::Type::kReference:
       return Location::RegisterLocation(V0);
 
     case DataType::Type::kInt64:
@@ -84,12 +85,13 @@
   Location next_location;
 
   switch (type) {
+    case DataType::Type::kReference:
     case DataType::Type::kBool:
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
     case DataType::Type::kUint16:
     case DataType::Type::kInt16:
-    case DataType::Type::kInt32:
-    case DataType::Type::kReference: {
+    case DataType::Type::kInt32: {
       uint32_t gp_index = gp_index_++;
       if (gp_index < calling_convention.GetNumberOfRegisters()) {
         next_location = Location::RegisterLocation(calling_convention.GetRegisterAt(gp_index));
@@ -2592,7 +2594,8 @@
   const bool maybe_compressed_char_at = mirror::kUseStringCompression &&
                                         instruction->IsStringCharAt();
   switch (type) {
-    case DataType::Type::kBool: {
+    case DataType::Type::kBool:
+    case DataType::Type::kUint8: {
       Register out = out_loc.AsRegister<Register>();
       if (index.IsConstant()) {
         size_t offset =
@@ -2618,19 +2621,6 @@
       break;
     }
 
-    case DataType::Type::kInt16: {
-      Register out = out_loc.AsRegister<Register>();
-      if (index.IsConstant()) {
-        size_t offset =
-            (index.GetConstant()->AsIntConstant()->GetValue() << TIMES_2) + data_offset;
-        __ LoadFromOffset(kLoadSignedHalfword, out, obj, offset, null_checker);
-      } else {
-        __ ShiftAndAdd(TMP, index.AsRegister<Register>(), obj, TIMES_2, TMP);
-        __ LoadFromOffset(kLoadSignedHalfword, out, TMP, data_offset, null_checker);
-      }
-      break;
-    }
-
     case DataType::Type::kUint16: {
       Register out = out_loc.AsRegister<Register>();
       if (maybe_compressed_char_at) {
@@ -2683,6 +2673,19 @@
       break;
     }
 
+    case DataType::Type::kInt16: {
+      Register out = out_loc.AsRegister<Register>();
+      if (index.IsConstant()) {
+        size_t offset =
+            (index.GetConstant()->AsIntConstant()->GetValue() << TIMES_2) + data_offset;
+        __ LoadFromOffset(kLoadSignedHalfword, out, obj, offset, null_checker);
+      } else {
+        __ ShiftAndAdd(TMP, index.AsRegister<Register>(), obj, TIMES_2, TMP);
+        __ LoadFromOffset(kLoadSignedHalfword, out, TMP, data_offset, null_checker);
+      }
+      break;
+    }
+
     case DataType::Type::kInt32: {
       DCHECK_EQ(sizeof(mirror::HeapReference<mirror::Object>), sizeof(int32_t));
       Register out = out_loc.AsRegister<Register>();
@@ -2880,6 +2883,7 @@
 
   switch (value_type) {
     case DataType::Type::kBool:
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8: {
       uint32_t data_offset = mirror::Array::DataOffset(sizeof(uint8_t)).Uint32Value();
       if (index.IsConstant()) {
@@ -2897,8 +2901,8 @@
       break;
     }
 
-    case DataType::Type::kInt16:
-    case DataType::Type::kUint16: {
+    case DataType::Type::kUint16:
+    case DataType::Type::kInt16: {
       uint32_t data_offset = mirror::Array::DataOffset(sizeof(uint16_t)).Uint32Value();
       if (index.IsConstant()) {
         data_offset += index.GetConstant()->AsIntConstant()->GetValue() << TIMES_2;
@@ -3390,9 +3394,10 @@
 
   switch (in_type) {
     case DataType::Type::kBool:
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
-    case DataType::Type::kInt16:
     case DataType::Type::kUint16:
+    case DataType::Type::kInt16:
     case DataType::Type::kInt32:
       locations->SetInAt(0, Location::RequiresRegister());
       locations->SetInAt(1, Location::RequiresRegister());
@@ -3429,9 +3434,10 @@
   // -1 if: left  < right
   switch (in_type) {
     case DataType::Type::kBool:
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
-    case DataType::Type::kInt16:
     case DataType::Type::kUint16:
+    case DataType::Type::kInt16:
     case DataType::Type::kInt32: {
       Register lhs = locations->InAt(0).AsRegister<Register>();
       Register rhs = locations->InAt(1).AsRegister<Register>();
@@ -3833,6 +3839,7 @@
 
   switch (type) {
     case DataType::Type::kBool:
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
     case DataType::Type::kUint16:
     case DataType::Type::kInt16:
@@ -6159,17 +6166,18 @@
 
   switch (type) {
     case DataType::Type::kBool:
+    case DataType::Type::kUint8:
       load_type = kLoadUnsignedByte;
       break;
     case DataType::Type::kInt8:
       load_type = kLoadSignedByte;
       break;
-    case DataType::Type::kInt16:
-      load_type = kLoadSignedHalfword;
-      break;
     case DataType::Type::kUint16:
       load_type = kLoadUnsignedHalfword;
       break;
+    case DataType::Type::kInt16:
+      load_type = kLoadSignedHalfword;
+      break;
     case DataType::Type::kInt32:
     case DataType::Type::kFloat32:
     case DataType::Type::kReference:
@@ -6312,11 +6320,12 @@
 
   switch (type) {
     case DataType::Type::kBool:
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
       store_type = kStoreByte;
       break;
-    case DataType::Type::kInt16:
     case DataType::Type::kUint16:
+    case DataType::Type::kInt16:
       store_type = kStoreHalfword;
       break;
     case DataType::Type::kInt32:
@@ -8601,7 +8610,8 @@
 void LocationsBuilderMIPS::VisitTypeConversion(HTypeConversion* conversion) {
   DataType::Type input_type = conversion->GetInputType();
   DataType::Type result_type = conversion->GetResultType();
-  DCHECK_NE(input_type, result_type);
+  DCHECK(!DataType::IsTypeConversionImplicit(input_type, result_type))
+      << input_type << " -> " << result_type;
   bool isR6 = codegen_->GetInstructionSetFeatures().IsR6();
 
   if ((input_type == DataType::Type::kReference) || (input_type == DataType::Type::kVoid) ||
@@ -8652,7 +8662,8 @@
   bool has_sign_extension = codegen_->GetInstructionSetFeatures().IsMipsIsaRevGreaterThanEqual2();
   bool isR6 = codegen_->GetInstructionSetFeatures().IsR6();
 
-  DCHECK_NE(input_type, result_type);
+  DCHECK(!DataType::IsTypeConversionImplicit(input_type, result_type))
+      << input_type << " -> " << result_type;
 
   if (result_type == DataType::Type::kInt64 && DataType::IsIntegralType(input_type)) {
     Register dst_high = locations->Out().AsRegisterPairHigh<Register>();
@@ -8670,8 +8681,8 @@
         : locations->InAt(0).AsRegister<Register>();
 
     switch (result_type) {
-      case DataType::Type::kUint16:
-        __ Andi(dst, src, 0xFFFF);
+      case DataType::Type::kUint8:
+        __ Andi(dst, src, 0xFF);
         break;
       case DataType::Type::kInt8:
         if (has_sign_extension) {
@@ -8681,6 +8692,9 @@
           __ Sra(dst, dst, 24);
         }
         break;
+      case DataType::Type::kUint16:
+        __ Andi(dst, src, 0xFFFF);
+        break;
       case DataType::Type::kInt16:
         if (has_sign_extension) {
           __ Seh(dst, src);
diff --git a/compiler/optimizing/code_generator_mips64.cc b/compiler/optimizing/code_generator_mips64.cc
index 7051cce..6877003 100644
--- a/compiler/optimizing/code_generator_mips64.cc
+++ b/compiler/optimizing/code_generator_mips64.cc
@@ -50,6 +50,7 @@
 Location Mips64ReturnLocation(DataType::Type return_type) {
   switch (return_type) {
     case DataType::Type::kBool:
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
     case DataType::Type::kUint16:
     case DataType::Type::kInt16:
@@ -2170,7 +2171,8 @@
   const bool maybe_compressed_char_at = mirror::kUseStringCompression &&
                                         instruction->IsStringCharAt();
   switch (type) {
-    case DataType::Type::kBool: {
+    case DataType::Type::kBool:
+    case DataType::Type::kUint8: {
       GpuRegister out = out_loc.AsRegister<GpuRegister>();
       if (index.IsConstant()) {
         size_t offset =
@@ -2196,19 +2198,6 @@
       break;
     }
 
-    case DataType::Type::kInt16: {
-      GpuRegister out = out_loc.AsRegister<GpuRegister>();
-      if (index.IsConstant()) {
-        size_t offset =
-            (index.GetConstant()->AsIntConstant()->GetValue() << TIMES_2) + data_offset;
-        __ LoadFromOffset(kLoadSignedHalfword, out, obj, offset, null_checker);
-      } else {
-        __ Dlsa(TMP, index.AsRegister<GpuRegister>(), obj, TIMES_2);
-        __ LoadFromOffset(kLoadSignedHalfword, out, TMP, data_offset, null_checker);
-      }
-      break;
-    }
-
     case DataType::Type::kUint16: {
       GpuRegister out = out_loc.AsRegister<GpuRegister>();
       if (maybe_compressed_char_at) {
@@ -2261,6 +2250,19 @@
       break;
     }
 
+    case DataType::Type::kInt16: {
+      GpuRegister out = out_loc.AsRegister<GpuRegister>();
+      if (index.IsConstant()) {
+        size_t offset =
+            (index.GetConstant()->AsIntConstant()->GetValue() << TIMES_2) + data_offset;
+        __ LoadFromOffset(kLoadSignedHalfword, out, obj, offset, null_checker);
+      } else {
+        __ Dlsa(TMP, index.AsRegister<GpuRegister>(), obj, TIMES_2);
+        __ LoadFromOffset(kLoadSignedHalfword, out, TMP, data_offset, null_checker);
+      }
+      break;
+    }
+
     case DataType::Type::kInt32: {
       DCHECK_EQ(sizeof(mirror::HeapReference<mirror::Object>), sizeof(int32_t));
       GpuRegister out = out_loc.AsRegister<GpuRegister>();
@@ -2460,6 +2462,7 @@
 
   switch (value_type) {
     case DataType::Type::kBool:
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8: {
       uint32_t data_offset = mirror::Array::DataOffset(sizeof(uint8_t)).Uint32Value();
       if (index.IsConstant()) {
@@ -2477,8 +2480,8 @@
       break;
     }
 
-    case DataType::Type::kInt16:
-    case DataType::Type::kUint16: {
+    case DataType::Type::kUint16:
+    case DataType::Type::kInt16: {
       uint32_t data_offset = mirror::Array::DataOffset(sizeof(uint16_t)).Uint32Value();
       if (index.IsConstant()) {
         data_offset += index.GetConstant()->AsIntConstant()->GetValue() << TIMES_2;
@@ -2969,9 +2972,10 @@
 
   switch (in_type) {
     case DataType::Type::kBool:
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
-    case DataType::Type::kInt16:
     case DataType::Type::kUint16:
+    case DataType::Type::kInt16:
     case DataType::Type::kInt32:
     case DataType::Type::kInt64:
       locations->SetInAt(0, Location::RequiresRegister());
@@ -3001,9 +3005,10 @@
   // -1 if: left  < right
   switch (in_type) {
     case DataType::Type::kBool:
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
-    case DataType::Type::kInt16:
     case DataType::Type::kUint16:
+    case DataType::Type::kInt16:
     case DataType::Type::kInt32:
     case DataType::Type::kInt64: {
       GpuRegister lhs = locations->InAt(0).AsRegister<GpuRegister>();
@@ -4681,17 +4686,18 @@
 
   switch (type) {
     case DataType::Type::kBool:
+    case DataType::Type::kUint8:
       load_type = kLoadUnsignedByte;
       break;
     case DataType::Type::kInt8:
       load_type = kLoadSignedByte;
       break;
-    case DataType::Type::kInt16:
-      load_type = kLoadSignedHalfword;
-      break;
     case DataType::Type::kUint16:
       load_type = kLoadUnsignedHalfword;
       break;
+    case DataType::Type::kInt16:
+      load_type = kLoadSignedHalfword;
+      break;
     case DataType::Type::kInt32:
     case DataType::Type::kFloat32:
       load_type = kLoadWord;
@@ -4779,11 +4785,12 @@
 
   switch (type) {
     case DataType::Type::kBool:
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
       store_type = kStoreByte;
       break;
-    case DataType::Type::kInt16:
     case DataType::Type::kUint16:
+    case DataType::Type::kInt16:
       store_type = kStoreHalfword;
       break;
     case DataType::Type::kInt32:
@@ -6767,7 +6774,8 @@
 void LocationsBuilderMIPS64::VisitTypeConversion(HTypeConversion* conversion) {
   DataType::Type input_type = conversion->GetInputType();
   DataType::Type result_type = conversion->GetResultType();
-  DCHECK_NE(input_type, result_type);
+  DCHECK(!DataType::IsTypeConversionImplicit(input_type, result_type))
+      << input_type << " -> " << result_type;
 
   if ((input_type == DataType::Type::kReference) || (input_type == DataType::Type::kVoid) ||
       (result_type == DataType::Type::kReference) || (result_type == DataType::Type::kVoid)) {
@@ -6794,15 +6802,16 @@
   DataType::Type result_type = conversion->GetResultType();
   DataType::Type input_type = conversion->GetInputType();
 
-  DCHECK_NE(input_type, result_type);
+  DCHECK(!DataType::IsTypeConversionImplicit(input_type, result_type))
+      << input_type << " -> " << result_type;
 
   if (DataType::IsIntegralType(result_type) && DataType::IsIntegralType(input_type)) {
     GpuRegister dst = locations->Out().AsRegister<GpuRegister>();
     GpuRegister src = locations->InAt(0).AsRegister<GpuRegister>();
 
     switch (result_type) {
-      case DataType::Type::kUint16:
-        __ Andi(dst, src, 0xFFFF);
+      case DataType::Type::kUint8:
+        __ Andi(dst, src, 0xFF);
         break;
       case DataType::Type::kInt8:
         if (input_type == DataType::Type::kInt64) {
@@ -6815,6 +6824,9 @@
           __ Seb(dst, src);
         }
         break;
+      case DataType::Type::kUint16:
+        __ Andi(dst, src, 0xFFFF);
+        break;
       case DataType::Type::kInt16:
         if (input_type == DataType::Type::kInt64) {
           // Type conversion from long to types narrower than int is a result of code
diff --git a/compiler/optimizing/code_generator_vector_arm64.cc b/compiler/optimizing/code_generator_vector_arm64.cc
index 5d5623b..b2aec1e 100644
--- a/compiler/optimizing/code_generator_vector_arm64.cc
+++ b/compiler/optimizing/code_generator_vector_arm64.cc
@@ -42,6 +42,7 @@
   HInstruction* input = instruction->InputAt(0);
   switch (instruction->GetPackedType()) {
     case DataType::Type::kBool:
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
     case DataType::Type::kUint16:
     case DataType::Type::kInt16:
@@ -73,6 +74,7 @@
   VRegister dst = VRegisterFrom(locations->Out());
   switch (instruction->GetPackedType()) {
     case DataType::Type::kBool:
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
       DCHECK_EQ(16u, instruction->GetVectorLength());
       if (src_loc.IsConstant()) {
@@ -132,6 +134,7 @@
   LocationSummary* locations = new (GetGraph()->GetArena()) LocationSummary(instruction);
   switch (instruction->GetPackedType()) {
     case DataType::Type::kBool:
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
     case DataType::Type::kUint16:
     case DataType::Type::kInt16:
@@ -185,6 +188,7 @@
                         instruction->IsVecNot() ? Location::kOutputOverlap
                                                 : Location::kNoOutputOverlap);
       break;
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
     case DataType::Type::kUint16:
     case DataType::Type::kInt16:
@@ -268,6 +272,7 @@
   VRegister src = VRegisterFrom(locations->InAt(0));
   VRegister dst = VRegisterFrom(locations->Out());
   switch (instruction->GetPackedType()) {
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
       DCHECK_EQ(16u, instruction->GetVectorLength());
       __ Neg(dst.V16B(), src.V16B());
@@ -312,7 +317,6 @@
       DCHECK_EQ(16u, instruction->GetVectorLength());
       __ Abs(dst.V16B(), src.V16B());
       break;
-    case DataType::Type::kUint16:
     case DataType::Type::kInt16:
       DCHECK_EQ(8u, instruction->GetVectorLength());
       __ Abs(dst.V8H(), src.V8H());
@@ -353,6 +357,7 @@
       __ Movi(dst.V16B(), 1);
       __ Eor(dst.V16B(), dst.V16B(), src.V16B());
       break;
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
     case DataType::Type::kUint16:
     case DataType::Type::kInt16:
@@ -371,6 +376,7 @@
   LocationSummary* locations = new (arena) LocationSummary(instruction);
   switch (instruction->GetPackedType()) {
     case DataType::Type::kBool:
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
     case DataType::Type::kUint16:
     case DataType::Type::kInt16:
@@ -398,6 +404,7 @@
   VRegister rhs = VRegisterFrom(locations->InAt(1));
   VRegister dst = VRegisterFrom(locations->Out());
   switch (instruction->GetPackedType()) {
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
       DCHECK_EQ(16u, instruction->GetVectorLength());
       __ Add(dst.V16B(), lhs.V16B(), rhs.V16B());
@@ -439,30 +446,29 @@
   VRegister rhs = VRegisterFrom(locations->InAt(1));
   VRegister dst = VRegisterFrom(locations->Out());
   switch (instruction->GetPackedType()) {
+    case DataType::Type::kUint8:
+      DCHECK_EQ(16u, instruction->GetVectorLength());
+      instruction->IsRounded()
+          ? __ Urhadd(dst.V16B(), lhs.V16B(), rhs.V16B())
+          : __ Uhadd(dst.V16B(), lhs.V16B(), rhs.V16B());
+      break;
     case DataType::Type::kInt8:
       DCHECK_EQ(16u, instruction->GetVectorLength());
-      if (instruction->IsUnsigned()) {
-        instruction->IsRounded()
-            ? __ Urhadd(dst.V16B(), lhs.V16B(), rhs.V16B())
-            : __ Uhadd(dst.V16B(), lhs.V16B(), rhs.V16B());
-      } else {
-        instruction->IsRounded()
-            ? __ Srhadd(dst.V16B(), lhs.V16B(), rhs.V16B())
-            : __ Shadd(dst.V16B(), lhs.V16B(), rhs.V16B());
-      }
+      instruction->IsRounded()
+          ? __ Srhadd(dst.V16B(), lhs.V16B(), rhs.V16B())
+          : __ Shadd(dst.V16B(), lhs.V16B(), rhs.V16B());
       break;
     case DataType::Type::kUint16:
+      DCHECK_EQ(8u, instruction->GetVectorLength());
+      instruction->IsRounded()
+          ? __ Urhadd(dst.V8H(), lhs.V8H(), rhs.V8H())
+          : __ Uhadd(dst.V8H(), lhs.V8H(), rhs.V8H());
+      break;
     case DataType::Type::kInt16:
       DCHECK_EQ(8u, instruction->GetVectorLength());
-      if (instruction->IsUnsigned()) {
-        instruction->IsRounded()
-            ? __ Urhadd(dst.V8H(), lhs.V8H(), rhs.V8H())
-            : __ Uhadd(dst.V8H(), lhs.V8H(), rhs.V8H());
-      } else {
-        instruction->IsRounded()
-            ? __ Srhadd(dst.V8H(), lhs.V8H(), rhs.V8H())
-            : __ Shadd(dst.V8H(), lhs.V8H(), rhs.V8H());
-      }
+      instruction->IsRounded()
+          ? __ Srhadd(dst.V8H(), lhs.V8H(), rhs.V8H())
+          : __ Shadd(dst.V8H(), lhs.V8H(), rhs.V8H());
       break;
     default:
       LOG(FATAL) << "Unsupported SIMD type";
@@ -480,6 +486,7 @@
   VRegister rhs = VRegisterFrom(locations->InAt(1));
   VRegister dst = VRegisterFrom(locations->Out());
   switch (instruction->GetPackedType()) {
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
       DCHECK_EQ(16u, instruction->GetVectorLength());
       __ Sub(dst.V16B(), lhs.V16B(), rhs.V16B());
@@ -521,6 +528,7 @@
   VRegister rhs = VRegisterFrom(locations->InAt(1));
   VRegister dst = VRegisterFrom(locations->Out());
   switch (instruction->GetPackedType()) {
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
       DCHECK_EQ(16u, instruction->GetVectorLength());
       __ Mul(dst.V16B(), lhs.V16B(), rhs.V16B());
@@ -582,22 +590,21 @@
   VRegister rhs = VRegisterFrom(locations->InAt(1));
   VRegister dst = VRegisterFrom(locations->Out());
   switch (instruction->GetPackedType()) {
+    case DataType::Type::kUint8:
+      DCHECK_EQ(16u, instruction->GetVectorLength());
+      __ Umin(dst.V16B(), lhs.V16B(), rhs.V16B());
+      break;
     case DataType::Type::kInt8:
       DCHECK_EQ(16u, instruction->GetVectorLength());
-      if (instruction->IsUnsigned()) {
-        __ Umin(dst.V16B(), lhs.V16B(), rhs.V16B());
-      } else {
-        __ Smin(dst.V16B(), lhs.V16B(), rhs.V16B());
-      }
+      __ Smin(dst.V16B(), lhs.V16B(), rhs.V16B());
       break;
     case DataType::Type::kUint16:
+      DCHECK_EQ(8u, instruction->GetVectorLength());
+      __ Umin(dst.V8H(), lhs.V8H(), rhs.V8H());
+      break;
     case DataType::Type::kInt16:
       DCHECK_EQ(8u, instruction->GetVectorLength());
-      if (instruction->IsUnsigned()) {
-        __ Umin(dst.V8H(), lhs.V8H(), rhs.V8H());
-      } else {
-        __ Smin(dst.V8H(), lhs.V8H(), rhs.V8H());
-      }
+      __ Smin(dst.V8H(), lhs.V8H(), rhs.V8H());
       break;
     case DataType::Type::kInt32:
       DCHECK_EQ(4u, instruction->GetVectorLength());
@@ -633,22 +640,21 @@
   VRegister rhs = VRegisterFrom(locations->InAt(1));
   VRegister dst = VRegisterFrom(locations->Out());
   switch (instruction->GetPackedType()) {
+    case DataType::Type::kUint8:
+      DCHECK_EQ(16u, instruction->GetVectorLength());
+      __ Umax(dst.V16B(), lhs.V16B(), rhs.V16B());
+      break;
     case DataType::Type::kInt8:
       DCHECK_EQ(16u, instruction->GetVectorLength());
-      if (instruction->IsUnsigned()) {
-        __ Umax(dst.V16B(), lhs.V16B(), rhs.V16B());
-      } else {
-        __ Smax(dst.V16B(), lhs.V16B(), rhs.V16B());
-      }
+      __ Smax(dst.V16B(), lhs.V16B(), rhs.V16B());
       break;
     case DataType::Type::kUint16:
+      DCHECK_EQ(8u, instruction->GetVectorLength());
+      __ Umax(dst.V8H(), lhs.V8H(), rhs.V8H());
+      break;
     case DataType::Type::kInt16:
       DCHECK_EQ(8u, instruction->GetVectorLength());
-      if (instruction->IsUnsigned()) {
-        __ Umax(dst.V8H(), lhs.V8H(), rhs.V8H());
-      } else {
-        __ Smax(dst.V8H(), lhs.V8H(), rhs.V8H());
-      }
+      __ Smax(dst.V8H(), lhs.V8H(), rhs.V8H());
       break;
     case DataType::Type::kInt32:
       DCHECK_EQ(4u, instruction->GetVectorLength());
@@ -675,6 +681,7 @@
 }
 
 void LocationsBuilderARM64::VisitVecAnd(HVecAnd* instruction) {
+  // TODO: Allow constants supported by BIC (vector, immediate).
   CreateVecBinOpLocations(GetGraph()->GetArena(), instruction);
 }
 
@@ -685,6 +692,7 @@
   VRegister dst = VRegisterFrom(locations->Out());
   switch (instruction->GetPackedType()) {
     case DataType::Type::kBool:
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
     case DataType::Type::kUint16:
     case DataType::Type::kInt16:
@@ -705,6 +713,7 @@
 }
 
 void InstructionCodeGeneratorARM64::VisitVecAndNot(HVecAndNot* instruction) {
+  // TODO: Use BIC (vector, register).
   LOG(FATAL) << "Unsupported SIMD instruction " << instruction->GetId();
 }
 
@@ -719,6 +728,7 @@
   VRegister dst = VRegisterFrom(locations->Out());
   switch (instruction->GetPackedType()) {
     case DataType::Type::kBool:
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
     case DataType::Type::kUint16:
     case DataType::Type::kInt16:
@@ -745,6 +755,7 @@
   VRegister dst = VRegisterFrom(locations->Out());
   switch (instruction->GetPackedType()) {
     case DataType::Type::kBool:
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
     case DataType::Type::kUint16:
     case DataType::Type::kInt16:
@@ -764,6 +775,7 @@
 static void CreateVecShiftLocations(ArenaAllocator* arena, HVecBinaryOperation* instruction) {
   LocationSummary* locations = new (arena) LocationSummary(instruction);
   switch (instruction->GetPackedType()) {
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
     case DataType::Type::kUint16:
     case DataType::Type::kInt16:
@@ -789,6 +801,7 @@
   VRegister dst = VRegisterFrom(locations->Out());
   int32_t value = locations->InAt(1).GetConstant()->AsIntConstant()->GetValue();
   switch (instruction->GetPackedType()) {
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
       DCHECK_EQ(16u, instruction->GetVectorLength());
       __ Shl(dst.V16B(), lhs.V16B(), value);
@@ -822,6 +835,7 @@
   VRegister dst = VRegisterFrom(locations->Out());
   int32_t value = locations->InAt(1).GetConstant()->AsIntConstant()->GetValue();
   switch (instruction->GetPackedType()) {
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
       DCHECK_EQ(16u, instruction->GetVectorLength());
       __ Sshr(dst.V16B(), lhs.V16B(), value);
@@ -855,6 +869,7 @@
   VRegister dst = VRegisterFrom(locations->Out());
   int32_t value = locations->InAt(1).GetConstant()->AsIntConstant()->GetValue();
   switch (instruction->GetPackedType()) {
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
       DCHECK_EQ(16u, instruction->GetVectorLength());
       __ Ushr(dst.V16B(), lhs.V16B(), value);
@@ -888,6 +903,7 @@
 
   switch (instruction->GetPackedType()) {
     case DataType::Type::kBool:
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
     case DataType::Type::kUint16:
     case DataType::Type::kInt16:
@@ -926,6 +942,7 @@
   // Set required elements.
   switch (instruction->GetPackedType()) {
     case DataType::Type::kBool:
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
       DCHECK_EQ(16u, instruction->GetVectorLength());
       __ Mov(dst.V16B(), 0, InputRegisterAt(instruction, 0));
@@ -953,6 +970,7 @@
 static void CreateVecAccumLocations(ArenaAllocator* arena, HVecOperation* instruction) {
   LocationSummary* locations = new (arena) LocationSummary(instruction);
   switch (instruction->GetPackedType()) {
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
     case DataType::Type::kUint16:
     case DataType::Type::kInt16:
@@ -985,6 +1003,7 @@
   DCHECK(locations->InAt(0).Equals(locations->Out()));
 
   switch (instruction->GetPackedType()) {
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
       DCHECK_EQ(16u, instruction->GetVectorLength());
       if (instruction->GetOpKind() == HInstruction::kAdd) {
@@ -1024,6 +1043,7 @@
   HVecOperation* b = instruction->InputAt(2)->AsVecOperation();
   DCHECK_EQ(a->GetPackedType(), b->GetPackedType());
   switch (a->GetPackedType()) {
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
       switch (instruction->GetPackedType()) {
         case DataType::Type::kInt64:
@@ -1069,10 +1089,10 @@
   HVecOperation* b = instruction->InputAt(2)->AsVecOperation();
   DCHECK_EQ(a->GetPackedType(), b->GetPackedType());
   switch (a->GetPackedType()) {
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
       DCHECK_EQ(16u, a->GetVectorLength());
       switch (instruction->GetPackedType()) {
-        case DataType::Type::kUint16:
         case DataType::Type::kInt16:
           DCHECK_EQ(8u, instruction->GetVectorLength());
           __ Sabal(acc.V8H(), left.V8B(), right.V8B());
@@ -1202,6 +1222,7 @@
   LocationSummary* locations = new (arena) LocationSummary(instruction);
   switch (instruction->GetPackedType()) {
     case DataType::Type::kBool:
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
     case DataType::Type::kUint16:
     case DataType::Type::kInt16:
@@ -1300,6 +1321,7 @@
       }
       FALLTHROUGH_INTENDED;
     case DataType::Type::kBool:
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
     case DataType::Type::kInt16:
     case DataType::Type::kInt32:
@@ -1329,6 +1351,7 @@
 
   switch (instruction->GetPackedType()) {
     case DataType::Type::kBool:
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
     case DataType::Type::kUint16:
     case DataType::Type::kInt16:
diff --git a/compiler/optimizing/code_generator_vector_arm_vixl.cc b/compiler/optimizing/code_generator_vector_arm_vixl.cc
index 333d108..df75752 100644
--- a/compiler/optimizing/code_generator_vector_arm_vixl.cc
+++ b/compiler/optimizing/code_generator_vector_arm_vixl.cc
@@ -36,6 +36,7 @@
   LocationSummary* locations = new (GetGraph()->GetArena()) LocationSummary(instruction);
   switch (instruction->GetPackedType()) {
     case DataType::Type::kBool:
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
     case DataType::Type::kUint16:
     case DataType::Type::kInt16:
@@ -54,6 +55,7 @@
   vixl32::DRegister dst = DRegisterFrom(locations->Out());
   switch (instruction->GetPackedType()) {
     case DataType::Type::kBool:
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
       DCHECK_EQ(8u, instruction->GetVectorLength());
       __ Vdup(Untyped8, dst, InputRegisterAt(instruction, 0));
@@ -91,6 +93,7 @@
                         instruction->IsVecNot() ? Location::kOutputOverlap
                                                 : Location::kNoOutputOverlap);
       break;
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
     case DataType::Type::kUint16:
     case DataType::Type::kInt16:
@@ -129,6 +132,7 @@
   vixl32::DRegister src = DRegisterFrom(locations->InAt(0));
   vixl32::DRegister dst = DRegisterFrom(locations->Out());
   switch (instruction->GetPackedType()) {
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
       DCHECK_EQ(8u, instruction->GetVectorLength());
       __ Vneg(DataTypeValue::S8, dst, src);
@@ -161,7 +165,6 @@
       DCHECK_EQ(8u, instruction->GetVectorLength());
       __ Vabs(DataTypeValue::S8, dst, src);
       break;
-    case DataType::Type::kUint16:
     case DataType::Type::kInt16:
       DCHECK_EQ(4u, instruction->GetVectorLength());
       __ Vabs(DataTypeValue::S16, dst, src);
@@ -190,6 +193,7 @@
       __ Vmov(I8, dst, 1);
       __ Veor(dst, dst, src);
       break;
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
     case DataType::Type::kUint16:
     case DataType::Type::kInt16:
@@ -207,6 +211,7 @@
   LocationSummary* locations = new (arena) LocationSummary(instruction);
   switch (instruction->GetPackedType()) {
     case DataType::Type::kBool:
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
     case DataType::Type::kUint16:
     case DataType::Type::kInt16:
@@ -231,6 +236,7 @@
   vixl32::DRegister rhs = DRegisterFrom(locations->InAt(1));
   vixl32::DRegister dst = DRegisterFrom(locations->Out());
   switch (instruction->GetPackedType()) {
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
       DCHECK_EQ(8u, instruction->GetVectorLength());
       __ Vadd(I8, dst, lhs, rhs);
@@ -260,30 +266,29 @@
   vixl32::DRegister rhs = DRegisterFrom(locations->InAt(1));
   vixl32::DRegister dst = DRegisterFrom(locations->Out());
   switch (instruction->GetPackedType()) {
+    case DataType::Type::kUint8:
+      DCHECK_EQ(8u, instruction->GetVectorLength());
+      instruction->IsRounded()
+          ? __ Vrhadd(DataTypeValue::U8, dst, lhs, rhs)
+          : __ Vhadd(DataTypeValue::U8, dst, lhs, rhs);
+      break;
     case DataType::Type::kInt8:
       DCHECK_EQ(8u, instruction->GetVectorLength());
-      if (instruction->IsUnsigned()) {
-        instruction->IsRounded()
-            ? __ Vrhadd(DataTypeValue::U8, dst, lhs, rhs)
-            : __ Vhadd(DataTypeValue::U8, dst, lhs, rhs);
-      } else {
-        instruction->IsRounded()
-            ? __ Vrhadd(DataTypeValue::S8, dst, lhs, rhs)
-            : __ Vhadd(DataTypeValue::S8, dst, lhs, rhs);
-      }
+      instruction->IsRounded()
+          ? __ Vrhadd(DataTypeValue::S8, dst, lhs, rhs)
+          : __ Vhadd(DataTypeValue::S8, dst, lhs, rhs);
       break;
     case DataType::Type::kUint16:
+      DCHECK_EQ(4u, instruction->GetVectorLength());
+      instruction->IsRounded()
+          ? __ Vrhadd(DataTypeValue::U16, dst, lhs, rhs)
+          : __ Vhadd(DataTypeValue::U16, dst, lhs, rhs);
+      break;
     case DataType::Type::kInt16:
       DCHECK_EQ(4u, instruction->GetVectorLength());
-      if (instruction->IsUnsigned()) {
-        instruction->IsRounded()
-            ? __ Vrhadd(DataTypeValue::U16, dst, lhs, rhs)
-            : __ Vhadd(DataTypeValue::U16, dst, lhs, rhs);
-      } else {
-        instruction->IsRounded()
-            ? __ Vrhadd(DataTypeValue::S16, dst, lhs, rhs)
-            : __ Vhadd(DataTypeValue::S16, dst, lhs, rhs);
-      }
+      instruction->IsRounded()
+          ? __ Vrhadd(DataTypeValue::S16, dst, lhs, rhs)
+          : __ Vhadd(DataTypeValue::S16, dst, lhs, rhs);
       break;
     default:
       LOG(FATAL) << "Unsupported SIMD type";
@@ -301,6 +306,7 @@
   vixl32::DRegister rhs = DRegisterFrom(locations->InAt(1));
   vixl32::DRegister dst = DRegisterFrom(locations->Out());
   switch (instruction->GetPackedType()) {
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
       DCHECK_EQ(8u, instruction->GetVectorLength());
       __ Vsub(I8, dst, lhs, rhs);
@@ -330,6 +336,7 @@
   vixl32::DRegister rhs = DRegisterFrom(locations->InAt(1));
   vixl32::DRegister dst = DRegisterFrom(locations->Out());
   switch (instruction->GetPackedType()) {
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
       DCHECK_EQ(8u, instruction->GetVectorLength());
       __ Vmul(I8, dst, lhs, rhs);
@@ -367,22 +374,21 @@
   vixl32::DRegister rhs = DRegisterFrom(locations->InAt(1));
   vixl32::DRegister dst = DRegisterFrom(locations->Out());
   switch (instruction->GetPackedType()) {
+    case DataType::Type::kUint8:
+      DCHECK_EQ(8u, instruction->GetVectorLength());
+      __ Vmin(DataTypeValue::U8, dst, lhs, rhs);
+      break;
     case DataType::Type::kInt8:
       DCHECK_EQ(8u, instruction->GetVectorLength());
-      if (instruction->IsUnsigned()) {
-        __ Vmin(DataTypeValue::U8, dst, lhs, rhs);
-      } else {
-        __ Vmin(DataTypeValue::S8, dst, lhs, rhs);
-      }
+      __ Vmin(DataTypeValue::S8, dst, lhs, rhs);
       break;
     case DataType::Type::kUint16:
+      DCHECK_EQ(4u, instruction->GetVectorLength());
+      __ Vmin(DataTypeValue::U16, dst, lhs, rhs);
+      break;
     case DataType::Type::kInt16:
       DCHECK_EQ(4u, instruction->GetVectorLength());
-      if (instruction->IsUnsigned()) {
-        __ Vmin(DataTypeValue::U16, dst, lhs, rhs);
-      } else {
-        __ Vmin(DataTypeValue::S16, dst, lhs, rhs);
-      }
+      __ Vmin(DataTypeValue::S16, dst, lhs, rhs);
       break;
     case DataType::Type::kInt32:
       DCHECK_EQ(2u, instruction->GetVectorLength());
@@ -408,22 +414,21 @@
   vixl32::DRegister rhs = DRegisterFrom(locations->InAt(1));
   vixl32::DRegister dst = DRegisterFrom(locations->Out());
   switch (instruction->GetPackedType()) {
+    case DataType::Type::kUint8:
+      DCHECK_EQ(8u, instruction->GetVectorLength());
+      __ Vmax(DataTypeValue::U8, dst, lhs, rhs);
+      break;
     case DataType::Type::kInt8:
       DCHECK_EQ(8u, instruction->GetVectorLength());
-      if (instruction->IsUnsigned()) {
-        __ Vmax(DataTypeValue::U8, dst, lhs, rhs);
-      } else {
-        __ Vmax(DataTypeValue::S8, dst, lhs, rhs);
-      }
+      __ Vmax(DataTypeValue::S8, dst, lhs, rhs);
       break;
     case DataType::Type::kUint16:
+      DCHECK_EQ(4u, instruction->GetVectorLength());
+      __ Vmax(DataTypeValue::U16, dst, lhs, rhs);
+      break;
     case DataType::Type::kInt16:
       DCHECK_EQ(4u, instruction->GetVectorLength());
-      if (instruction->IsUnsigned()) {
-        __ Vmax(DataTypeValue::U16, dst, lhs, rhs);
-      } else {
-        __ Vmax(DataTypeValue::S16, dst, lhs, rhs);
-      }
+      __ Vmax(DataTypeValue::S16, dst, lhs, rhs);
       break;
     case DataType::Type::kInt32:
       DCHECK_EQ(2u, instruction->GetVectorLength());
@@ -440,6 +445,7 @@
 }
 
 void LocationsBuilderARMVIXL::VisitVecAnd(HVecAnd* instruction) {
+  // TODO: Allow constants supported by VAND (immediate).
   CreateVecBinOpLocations(GetGraph()->GetArena(), instruction);
 }
 
@@ -450,6 +456,7 @@
   vixl32::DRegister dst = DRegisterFrom(locations->Out());
   switch (instruction->GetPackedType()) {
     case DataType::Type::kBool:
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
     case DataType::Type::kUint16:
     case DataType::Type::kInt16:
@@ -481,6 +488,7 @@
   vixl32::DRegister dst = DRegisterFrom(locations->Out());
   switch (instruction->GetPackedType()) {
     case DataType::Type::kBool:
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
     case DataType::Type::kUint16:
     case DataType::Type::kInt16:
@@ -504,6 +512,7 @@
   vixl32::DRegister dst = DRegisterFrom(locations->Out());
   switch (instruction->GetPackedType()) {
     case DataType::Type::kBool:
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
     case DataType::Type::kUint16:
     case DataType::Type::kInt16:
@@ -520,6 +529,7 @@
 static void CreateVecShiftLocations(ArenaAllocator* arena, HVecBinaryOperation* instruction) {
   LocationSummary* locations = new (arena) LocationSummary(instruction);
   switch (instruction->GetPackedType()) {
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
     case DataType::Type::kUint16:
     case DataType::Type::kInt16:
@@ -544,6 +554,7 @@
   vixl32::DRegister dst = DRegisterFrom(locations->Out());
   int32_t value = locations->InAt(1).GetConstant()->AsIntConstant()->GetValue();
   switch (instruction->GetPackedType()) {
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
       DCHECK_EQ(8u, instruction->GetVectorLength());
       __ Vshl(I8, dst, lhs, value);
@@ -573,6 +584,7 @@
   vixl32::DRegister dst = DRegisterFrom(locations->Out());
   int32_t value = locations->InAt(1).GetConstant()->AsIntConstant()->GetValue();
   switch (instruction->GetPackedType()) {
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
       DCHECK_EQ(8u, instruction->GetVectorLength());
       __ Vshr(DataTypeValue::S8, dst, lhs, value);
@@ -602,6 +614,7 @@
   vixl32::DRegister dst = DRegisterFrom(locations->Out());
   int32_t value = locations->InAt(1).GetConstant()->AsIntConstant()->GetValue();
   switch (instruction->GetPackedType()) {
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
       DCHECK_EQ(8u, instruction->GetVectorLength());
       __ Vshr(DataTypeValue::U8, dst, lhs, value);
@@ -633,6 +646,7 @@
 static void CreateVecAccumLocations(ArenaAllocator* arena, HVecOperation* instruction) {
   LocationSummary* locations = new (arena) LocationSummary(instruction);
   switch (instruction->GetPackedType()) {
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
     case DataType::Type::kUint16:
     case DataType::Type::kInt16:
@@ -678,6 +692,7 @@
   LocationSummary* locations = new (arena) LocationSummary(instruction);
   switch (instruction->GetPackedType()) {
     case DataType::Type::kBool:
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
     case DataType::Type::kUint16:
     case DataType::Type::kInt16:
@@ -764,6 +779,7 @@
 
   switch (instruction->GetPackedType()) {
     case DataType::Type::kBool:
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
       DCHECK_EQ(8u, instruction->GetVectorLength());
       if (IsWordAligned(instruction)) {
@@ -811,6 +827,7 @@
   vixl32::Register scratch;
   switch (instruction->GetPackedType()) {
     case DataType::Type::kBool:
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
       DCHECK_EQ(8u, instruction->GetVectorLength());
       if (IsWordAligned(instruction)) {
diff --git a/compiler/optimizing/code_generator_vector_mips.cc b/compiler/optimizing/code_generator_vector_mips.cc
index c25f5ac..e8c5157 100644
--- a/compiler/optimizing/code_generator_vector_mips.cc
+++ b/compiler/optimizing/code_generator_vector_mips.cc
@@ -27,6 +27,7 @@
   LocationSummary* locations = new (GetGraph()->GetArena()) LocationSummary(instruction);
   switch (instruction->GetPackedType()) {
     case DataType::Type::kBool:
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
     case DataType::Type::kUint16:
     case DataType::Type::kInt16:
@@ -51,6 +52,7 @@
   VectorRegister dst = VectorRegisterFrom(locations->Out());
   switch (instruction->GetPackedType()) {
     case DataType::Type::kBool:
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
       DCHECK_EQ(16u, instruction->GetVectorLength());
       __ FillB(dst, locations->InAt(0).AsRegister<Register>());
@@ -106,6 +108,7 @@
                         instruction->IsVecNot() ? Location::kOutputOverlap
                                                 : Location::kNoOutputOverlap);
       break;
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
     case DataType::Type::kUint16:
     case DataType::Type::kInt16:
@@ -160,6 +163,7 @@
   VectorRegister src = VectorRegisterFrom(locations->InAt(0));
   VectorRegister dst = VectorRegisterFrom(locations->Out());
   switch (instruction->GetPackedType()) {
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
       DCHECK_EQ(16u, instruction->GetVectorLength());
       __ FillB(dst, ZERO);
@@ -211,7 +215,6 @@
       __ FillB(dst, ZERO);       // all zeroes
       __ Add_aB(dst, dst, src);  // dst = abs(0) + abs(src)
       break;
-    case DataType::Type::kUint16:
     case DataType::Type::kInt16:
       DCHECK_EQ(8u, instruction->GetVectorLength());
       __ FillH(dst, ZERO);       // all zeroes
@@ -259,6 +262,7 @@
       __ LdiB(dst, 1);
       __ XorV(dst, dst, src);
       break;
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
     case DataType::Type::kUint16:
     case DataType::Type::kInt16:
@@ -281,6 +285,7 @@
   LocationSummary* locations = new (arena) LocationSummary(instruction);
   switch (instruction->GetPackedType()) {
     case DataType::Type::kBool:
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
     case DataType::Type::kUint16:
     case DataType::Type::kInt16:
@@ -308,6 +313,7 @@
   VectorRegister rhs = VectorRegisterFrom(locations->InAt(1));
   VectorRegister dst = VectorRegisterFrom(locations->Out());
   switch (instruction->GetPackedType()) {
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
       DCHECK_EQ(16u, instruction->GetVectorLength());
       __ AddvB(dst, lhs, rhs);
@@ -349,30 +355,29 @@
   VectorRegister rhs = VectorRegisterFrom(locations->InAt(1));
   VectorRegister dst = VectorRegisterFrom(locations->Out());
   switch (instruction->GetPackedType()) {
+    case DataType::Type::kUint8:
+      DCHECK_EQ(16u, instruction->GetVectorLength());
+      instruction->IsRounded()
+          ? __ Aver_uB(dst, lhs, rhs)
+          : __ Ave_uB(dst, lhs, rhs);
+      break;
     case DataType::Type::kInt8:
       DCHECK_EQ(16u, instruction->GetVectorLength());
-      if (instruction->IsUnsigned()) {
-        instruction->IsRounded()
-            ? __ Aver_uB(dst, lhs, rhs)
-            : __ Ave_uB(dst, lhs, rhs);
-      } else {
-        instruction->IsRounded()
-            ? __ Aver_sB(dst, lhs, rhs)
-            : __ Ave_sB(dst, lhs, rhs);
-      }
+      instruction->IsRounded()
+          ? __ Aver_sB(dst, lhs, rhs)
+          : __ Ave_sB(dst, lhs, rhs);
       break;
     case DataType::Type::kUint16:
+      DCHECK_EQ(8u, instruction->GetVectorLength());
+      instruction->IsRounded()
+          ? __ Aver_uH(dst, lhs, rhs)
+          : __ Ave_uH(dst, lhs, rhs);
+      break;
     case DataType::Type::kInt16:
       DCHECK_EQ(8u, instruction->GetVectorLength());
-      if (instruction->IsUnsigned()) {
-        instruction->IsRounded()
-            ? __ Aver_uH(dst, lhs, rhs)
-            : __ Ave_uH(dst, lhs, rhs);
-      } else {
-        instruction->IsRounded()
-            ? __ Aver_sH(dst, lhs, rhs)
-            : __ Ave_sH(dst, lhs, rhs);
-      }
+      instruction->IsRounded()
+          ? __ Aver_sH(dst, lhs, rhs)
+          : __ Ave_sH(dst, lhs, rhs);
       break;
     default:
       LOG(FATAL) << "Unsupported SIMD type";
@@ -390,6 +395,7 @@
   VectorRegister rhs = VectorRegisterFrom(locations->InAt(1));
   VectorRegister dst = VectorRegisterFrom(locations->Out());
   switch (instruction->GetPackedType()) {
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
       DCHECK_EQ(16u, instruction->GetVectorLength());
       __ SubvB(dst, lhs, rhs);
@@ -431,6 +437,7 @@
   VectorRegister rhs = VectorRegisterFrom(locations->InAt(1));
   VectorRegister dst = VectorRegisterFrom(locations->Out());
   switch (instruction->GetPackedType()) {
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
       DCHECK_EQ(16u, instruction->GetVectorLength());
       __ MulvB(dst, lhs, rhs);
@@ -496,22 +503,21 @@
   VectorRegister rhs = VectorRegisterFrom(locations->InAt(1));
   VectorRegister dst = VectorRegisterFrom(locations->Out());
   switch (instruction->GetPackedType()) {
+    case DataType::Type::kUint8:
+      DCHECK_EQ(16u, instruction->GetVectorLength());
+      __ Min_uB(dst, lhs, rhs);
+      break;
     case DataType::Type::kInt8:
       DCHECK_EQ(16u, instruction->GetVectorLength());
-      if (instruction->IsUnsigned()) {
-        __ Min_uB(dst, lhs, rhs);
-      } else {
-        __ Min_sB(dst, lhs, rhs);
-      }
+      __ Min_sB(dst, lhs, rhs);
       break;
     case DataType::Type::kUint16:
+      DCHECK_EQ(8u, instruction->GetVectorLength());
+      __ Min_uH(dst, lhs, rhs);
+      break;
     case DataType::Type::kInt16:
       DCHECK_EQ(8u, instruction->GetVectorLength());
-      if (instruction->IsUnsigned()) {
-        __ Min_uH(dst, lhs, rhs);
-      } else {
-        __ Min_sH(dst, lhs, rhs);
-      }
+      __ Min_sH(dst, lhs, rhs);
       break;
     case DataType::Type::kInt32:
       DCHECK_EQ(4u, instruction->GetVectorLength());
@@ -557,22 +563,21 @@
   VectorRegister rhs = VectorRegisterFrom(locations->InAt(1));
   VectorRegister dst = VectorRegisterFrom(locations->Out());
   switch (instruction->GetPackedType()) {
+    case DataType::Type::kUint8:
+      DCHECK_EQ(16u, instruction->GetVectorLength());
+      __ Max_uB(dst, lhs, rhs);
+      break;
     case DataType::Type::kInt8:
       DCHECK_EQ(16u, instruction->GetVectorLength());
-      if (instruction->IsUnsigned()) {
-        __ Max_uB(dst, lhs, rhs);
-      } else {
-        __ Max_sB(dst, lhs, rhs);
-      }
+      __ Max_sB(dst, lhs, rhs);
       break;
     case DataType::Type::kUint16:
+      DCHECK_EQ(8u, instruction->GetVectorLength());
+      __ Max_uH(dst, lhs, rhs);
+      break;
     case DataType::Type::kInt16:
       DCHECK_EQ(8u, instruction->GetVectorLength());
-      if (instruction->IsUnsigned()) {
-        __ Max_uH(dst, lhs, rhs);
-      } else {
-        __ Max_sH(dst, lhs, rhs);
-      }
+      __ Max_sH(dst, lhs, rhs);
       break;
     case DataType::Type::kInt32:
       DCHECK_EQ(4u, instruction->GetVectorLength());
@@ -619,6 +624,7 @@
   VectorRegister dst = VectorRegisterFrom(locations->Out());
   switch (instruction->GetPackedType()) {
     case DataType::Type::kBool:
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
     case DataType::Type::kUint16:
     case DataType::Type::kInt16:
@@ -655,6 +661,7 @@
   VectorRegister dst = VectorRegisterFrom(locations->Out());
   switch (instruction->GetPackedType()) {
     case DataType::Type::kBool:
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
     case DataType::Type::kUint16:
     case DataType::Type::kInt16:
@@ -683,6 +690,7 @@
   VectorRegister dst = VectorRegisterFrom(locations->Out());
   switch (instruction->GetPackedType()) {
     case DataType::Type::kBool:
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
     case DataType::Type::kUint16:
     case DataType::Type::kInt16:
@@ -704,6 +712,7 @@
 static void CreateVecShiftLocations(ArenaAllocator* arena, HVecBinaryOperation* instruction) {
   LocationSummary* locations = new (arena) LocationSummary(instruction);
   switch (instruction->GetPackedType()) {
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
     case DataType::Type::kUint16:
     case DataType::Type::kInt16:
@@ -729,6 +738,7 @@
   VectorRegister dst = VectorRegisterFrom(locations->Out());
   int32_t value = locations->InAt(1).GetConstant()->AsIntConstant()->GetValue();
   switch (instruction->GetPackedType()) {
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
       DCHECK_EQ(16u, instruction->GetVectorLength());
       __ SlliB(dst, lhs, value);
@@ -762,6 +772,7 @@
   VectorRegister dst = VectorRegisterFrom(locations->Out());
   int32_t value = locations->InAt(1).GetConstant()->AsIntConstant()->GetValue();
   switch (instruction->GetPackedType()) {
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
       DCHECK_EQ(16u, instruction->GetVectorLength());
       __ SraiB(dst, lhs, value);
@@ -795,6 +806,7 @@
   VectorRegister dst = VectorRegisterFrom(locations->Out());
   int32_t value = locations->InAt(1).GetConstant()->AsIntConstant()->GetValue();
   switch (instruction->GetPackedType()) {
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
       DCHECK_EQ(16u, instruction->GetVectorLength());
       __ SrliB(dst, lhs, value);
@@ -830,6 +842,7 @@
 static void CreateVecAccumLocations(ArenaAllocator* arena, HVecOperation* instruction) {
   LocationSummary* locations = new (arena) LocationSummary(instruction);
   switch (instruction->GetPackedType()) {
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
     case DataType::Type::kUint16:
     case DataType::Type::kInt16:
@@ -856,6 +869,7 @@
   VectorRegister left = VectorRegisterFrom(locations->InAt(1));
   VectorRegister right = VectorRegisterFrom(locations->InAt(2));
   switch (instruction->GetPackedType()) {
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
       DCHECK_EQ(16u, instruction->GetVectorLength());
       if (instruction->GetOpKind() == HInstruction::kAdd) {
@@ -911,6 +925,7 @@
   LocationSummary* locations = new (arena) LocationSummary(instruction);
   switch (instruction->GetPackedType()) {
     case DataType::Type::kBool:
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
     case DataType::Type::kUint16:
     case DataType::Type::kInt16:
@@ -976,6 +991,7 @@
   int32_t offset = VecAddress(locations, size, &base);
   switch (instruction->GetPackedType()) {
     case DataType::Type::kBool:
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
       DCHECK_EQ(16u, instruction->GetVectorLength());
       __ LdB(reg, base, offset);
@@ -1018,6 +1034,7 @@
   int32_t offset = VecAddress(locations, size, &base);
   switch (instruction->GetPackedType()) {
     case DataType::Type::kBool:
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
       DCHECK_EQ(16u, instruction->GetVectorLength());
       __ StB(reg, base, offset);
diff --git a/compiler/optimizing/code_generator_vector_mips64.cc b/compiler/optimizing/code_generator_vector_mips64.cc
index f60f708..7d69773 100644
--- a/compiler/optimizing/code_generator_vector_mips64.cc
+++ b/compiler/optimizing/code_generator_vector_mips64.cc
@@ -32,6 +32,7 @@
   LocationSummary* locations = new (GetGraph()->GetArena()) LocationSummary(instruction);
   switch (instruction->GetPackedType()) {
     case DataType::Type::kBool:
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
     case DataType::Type::kUint16:
     case DataType::Type::kInt16:
@@ -56,6 +57,7 @@
   VectorRegister dst = VectorRegisterFrom(locations->Out());
   switch (instruction->GetPackedType()) {
     case DataType::Type::kBool:
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
       DCHECK_EQ(16u, instruction->GetVectorLength());
       __ FillB(dst, locations->InAt(0).AsRegister<GpuRegister>());
@@ -109,6 +111,7 @@
                         instruction->IsVecNot() ? Location::kOutputOverlap
                                                 : Location::kNoOutputOverlap);
       break;
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
     case DataType::Type::kUint16:
     case DataType::Type::kInt16:
@@ -164,6 +167,7 @@
   VectorRegister src = VectorRegisterFrom(locations->InAt(0));
   VectorRegister dst = VectorRegisterFrom(locations->Out());
   switch (instruction->GetPackedType()) {
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
       DCHECK_EQ(16u, instruction->GetVectorLength());
       __ FillB(dst, ZERO);
@@ -215,7 +219,6 @@
       __ FillB(dst, ZERO);       // all zeroes
       __ Add_aB(dst, dst, src);  // dst = abs(0) + abs(src)
       break;
-    case DataType::Type::kUint16:
     case DataType::Type::kInt16:
       DCHECK_EQ(8u, instruction->GetVectorLength());
       __ FillH(dst, ZERO);       // all zeroes
@@ -263,6 +266,7 @@
       __ LdiB(dst, 1);
       __ XorV(dst, dst, src);
       break;
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
     case DataType::Type::kUint16:
     case DataType::Type::kInt16:
@@ -285,6 +289,7 @@
   LocationSummary* locations = new (arena) LocationSummary(instruction);
   switch (instruction->GetPackedType()) {
     case DataType::Type::kBool:
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
     case DataType::Type::kUint16:
     case DataType::Type::kInt16:
@@ -312,6 +317,7 @@
   VectorRegister rhs = VectorRegisterFrom(locations->InAt(1));
   VectorRegister dst = VectorRegisterFrom(locations->Out());
   switch (instruction->GetPackedType()) {
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
       DCHECK_EQ(16u, instruction->GetVectorLength());
       __ AddvB(dst, lhs, rhs);
@@ -353,30 +359,29 @@
   VectorRegister rhs = VectorRegisterFrom(locations->InAt(1));
   VectorRegister dst = VectorRegisterFrom(locations->Out());
   switch (instruction->GetPackedType()) {
+    case DataType::Type::kUint8:
+      DCHECK_EQ(16u, instruction->GetVectorLength());
+      instruction->IsRounded()
+          ? __ Aver_uB(dst, lhs, rhs)
+          : __ Ave_uB(dst, lhs, rhs);
+      break;
     case DataType::Type::kInt8:
       DCHECK_EQ(16u, instruction->GetVectorLength());
-      if (instruction->IsUnsigned()) {
-        instruction->IsRounded()
-            ? __ Aver_uB(dst, lhs, rhs)
-            : __ Ave_uB(dst, lhs, rhs);
-      } else {
-        instruction->IsRounded()
-            ? __ Aver_sB(dst, lhs, rhs)
-            : __ Ave_sB(dst, lhs, rhs);
-      }
+      instruction->IsRounded()
+          ? __ Aver_sB(dst, lhs, rhs)
+          : __ Ave_sB(dst, lhs, rhs);
       break;
     case DataType::Type::kUint16:
+      DCHECK_EQ(8u, instruction->GetVectorLength());
+      instruction->IsRounded()
+          ? __ Aver_uH(dst, lhs, rhs)
+          : __ Ave_uH(dst, lhs, rhs);
+      break;
     case DataType::Type::kInt16:
       DCHECK_EQ(8u, instruction->GetVectorLength());
-      if (instruction->IsUnsigned()) {
-        instruction->IsRounded()
-            ? __ Aver_uH(dst, lhs, rhs)
-            : __ Ave_uH(dst, lhs, rhs);
-      } else {
-        instruction->IsRounded()
-            ? __ Aver_sH(dst, lhs, rhs)
-            : __ Ave_sH(dst, lhs, rhs);
-      }
+      instruction->IsRounded()
+          ? __ Aver_sH(dst, lhs, rhs)
+          : __ Ave_sH(dst, lhs, rhs);
       break;
     default:
       LOG(FATAL) << "Unsupported SIMD type";
@@ -394,6 +399,7 @@
   VectorRegister rhs = VectorRegisterFrom(locations->InAt(1));
   VectorRegister dst = VectorRegisterFrom(locations->Out());
   switch (instruction->GetPackedType()) {
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
       DCHECK_EQ(16u, instruction->GetVectorLength());
       __ SubvB(dst, lhs, rhs);
@@ -435,6 +441,7 @@
   VectorRegister rhs = VectorRegisterFrom(locations->InAt(1));
   VectorRegister dst = VectorRegisterFrom(locations->Out());
   switch (instruction->GetPackedType()) {
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
       DCHECK_EQ(16u, instruction->GetVectorLength());
       __ MulvB(dst, lhs, rhs);
@@ -500,22 +507,21 @@
   VectorRegister rhs = VectorRegisterFrom(locations->InAt(1));
   VectorRegister dst = VectorRegisterFrom(locations->Out());
   switch (instruction->GetPackedType()) {
+    case DataType::Type::kUint8:
+      DCHECK_EQ(16u, instruction->GetVectorLength());
+      __ Min_uB(dst, lhs, rhs);
+      break;
     case DataType::Type::kInt8:
       DCHECK_EQ(16u, instruction->GetVectorLength());
-      if (instruction->IsUnsigned()) {
-        __ Min_uB(dst, lhs, rhs);
-      } else {
-        __ Min_sB(dst, lhs, rhs);
-      }
+      __ Min_sB(dst, lhs, rhs);
       break;
     case DataType::Type::kUint16:
+      DCHECK_EQ(8u, instruction->GetVectorLength());
+      __ Min_uH(dst, lhs, rhs);
+      break;
     case DataType::Type::kInt16:
       DCHECK_EQ(8u, instruction->GetVectorLength());
-      if (instruction->IsUnsigned()) {
-        __ Min_uH(dst, lhs, rhs);
-      } else {
-        __ Min_sH(dst, lhs, rhs);
-      }
+      __ Min_sH(dst, lhs, rhs);
       break;
     case DataType::Type::kInt32:
       DCHECK_EQ(4u, instruction->GetVectorLength());
@@ -561,22 +567,21 @@
   VectorRegister rhs = VectorRegisterFrom(locations->InAt(1));
   VectorRegister dst = VectorRegisterFrom(locations->Out());
   switch (instruction->GetPackedType()) {
+    case DataType::Type::kUint8:
+      DCHECK_EQ(16u, instruction->GetVectorLength());
+      __ Max_uB(dst, lhs, rhs);
+      break;
     case DataType::Type::kInt8:
       DCHECK_EQ(16u, instruction->GetVectorLength());
-      if (instruction->IsUnsigned()) {
-        __ Max_uB(dst, lhs, rhs);
-      } else {
-        __ Max_sB(dst, lhs, rhs);
-      }
+      __ Max_sB(dst, lhs, rhs);
       break;
     case DataType::Type::kUint16:
+      DCHECK_EQ(8u, instruction->GetVectorLength());
+      __ Max_uH(dst, lhs, rhs);
+      break;
     case DataType::Type::kInt16:
       DCHECK_EQ(8u, instruction->GetVectorLength());
-      if (instruction->IsUnsigned()) {
-        __ Max_uH(dst, lhs, rhs);
-      } else {
-        __ Max_sH(dst, lhs, rhs);
-      }
+      __ Max_sH(dst, lhs, rhs);
       break;
     case DataType::Type::kInt32:
       DCHECK_EQ(4u, instruction->GetVectorLength());
@@ -623,6 +628,7 @@
   VectorRegister dst = VectorRegisterFrom(locations->Out());
   switch (instruction->GetPackedType()) {
     case DataType::Type::kBool:
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
     case DataType::Type::kUint16:
     case DataType::Type::kInt16:
@@ -659,6 +665,7 @@
   VectorRegister dst = VectorRegisterFrom(locations->Out());
   switch (instruction->GetPackedType()) {
     case DataType::Type::kBool:
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
     case DataType::Type::kUint16:
     case DataType::Type::kInt16:
@@ -687,6 +694,7 @@
   VectorRegister dst = VectorRegisterFrom(locations->Out());
   switch (instruction->GetPackedType()) {
     case DataType::Type::kBool:
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
     case DataType::Type::kUint16:
     case DataType::Type::kInt16:
@@ -708,6 +716,7 @@
 static void CreateVecShiftLocations(ArenaAllocator* arena, HVecBinaryOperation* instruction) {
   LocationSummary* locations = new (arena) LocationSummary(instruction);
   switch (instruction->GetPackedType()) {
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
     case DataType::Type::kUint16:
     case DataType::Type::kInt16:
@@ -733,6 +742,7 @@
   VectorRegister dst = VectorRegisterFrom(locations->Out());
   int32_t value = locations->InAt(1).GetConstant()->AsIntConstant()->GetValue();
   switch (instruction->GetPackedType()) {
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
       DCHECK_EQ(16u, instruction->GetVectorLength());
       __ SlliB(dst, lhs, value);
@@ -766,6 +776,7 @@
   VectorRegister dst = VectorRegisterFrom(locations->Out());
   int32_t value = locations->InAt(1).GetConstant()->AsIntConstant()->GetValue();
   switch (instruction->GetPackedType()) {
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
       DCHECK_EQ(16u, instruction->GetVectorLength());
       __ SraiB(dst, lhs, value);
@@ -799,6 +810,7 @@
   VectorRegister dst = VectorRegisterFrom(locations->Out());
   int32_t value = locations->InAt(1).GetConstant()->AsIntConstant()->GetValue();
   switch (instruction->GetPackedType()) {
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
       DCHECK_EQ(16u, instruction->GetVectorLength());
       __ SrliB(dst, lhs, value);
@@ -834,6 +846,7 @@
 static void CreateVecAccumLocations(ArenaAllocator* arena, HVecOperation* instruction) {
   LocationSummary* locations = new (arena) LocationSummary(instruction);
   switch (instruction->GetPackedType()) {
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
     case DataType::Type::kUint16:
     case DataType::Type::kInt16:
@@ -860,6 +873,7 @@
   VectorRegister left = VectorRegisterFrom(locations->InAt(1));
   VectorRegister right = VectorRegisterFrom(locations->InAt(2));
   switch (instruction->GetPackedType()) {
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
       DCHECK_EQ(16u, instruction->GetVectorLength());
       if (instruction->GetOpKind() == HInstruction::kAdd) {
@@ -915,6 +929,7 @@
   LocationSummary* locations = new (arena) LocationSummary(instruction);
   switch (instruction->GetPackedType()) {
     case DataType::Type::kBool:
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
     case DataType::Type::kUint16:
     case DataType::Type::kInt16:
@@ -980,6 +995,7 @@
   int32_t offset = VecAddress(locations, size, &base);
   switch (instruction->GetPackedType()) {
     case DataType::Type::kBool:
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
       DCHECK_EQ(16u, instruction->GetVectorLength());
       __ LdB(reg, base, offset);
@@ -1022,6 +1038,7 @@
   int32_t offset = VecAddress(locations, size, &base);
   switch (instruction->GetPackedType()) {
     case DataType::Type::kBool:
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
       DCHECK_EQ(16u, instruction->GetVectorLength());
       __ StB(reg, base, offset);
diff --git a/compiler/optimizing/code_generator_vector_x86.cc b/compiler/optimizing/code_generator_vector_x86.cc
index 6515dbe..a2ef1b1 100644
--- a/compiler/optimizing/code_generator_vector_x86.cc
+++ b/compiler/optimizing/code_generator_vector_x86.cc
@@ -37,6 +37,7 @@
       }
       FALLTHROUGH_INTENDED;
     case DataType::Type::kBool:
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
     case DataType::Type::kUint16:
     case DataType::Type::kInt16:
@@ -70,6 +71,7 @@
 
   switch (instruction->GetPackedType()) {
     case DataType::Type::kBool:
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
       DCHECK_EQ(16u, instruction->GetVectorLength());
       __ movd(dst, locations->InAt(0).AsRegister<Register>());
@@ -122,6 +124,7 @@
       locations->AddTemp(Location::RequiresFpuRegister());
       FALLTHROUGH_INTENDED;
     case DataType::Type::kBool:
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
     case DataType::Type::kUint16:
     case DataType::Type::kInt16:
@@ -145,6 +148,7 @@
   XmmRegister src = locations->InAt(0).AsFpuRegister<XmmRegister>();
   switch (instruction->GetPackedType()) {
     case DataType::Type::kBool:
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
     case DataType::Type::kUint16:
     case DataType::Type::kInt16:  // TODO: up to here, and?
@@ -180,6 +184,7 @@
   LocationSummary* locations = new (arena) LocationSummary(instruction);
   switch (instruction->GetPackedType()) {
     case DataType::Type::kBool:
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
     case DataType::Type::kUint16:
     case DataType::Type::kInt16:
@@ -290,6 +295,7 @@
   XmmRegister src = locations->InAt(0).AsFpuRegister<XmmRegister>();
   XmmRegister dst = locations->Out().AsFpuRegister<XmmRegister>();
   switch (instruction->GetPackedType()) {
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
       DCHECK_EQ(16u, instruction->GetVectorLength());
       __ pxor(dst, dst);
@@ -390,6 +396,7 @@
       __ pxor(dst, src);
       break;
     }
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
     case DataType::Type::kUint16:
     case DataType::Type::kInt16:
@@ -421,6 +428,7 @@
   LocationSummary* locations = new (arena) LocationSummary(instruction);
   switch (instruction->GetPackedType()) {
     case DataType::Type::kBool:
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
     case DataType::Type::kUint16:
     case DataType::Type::kInt16:
@@ -448,6 +456,7 @@
   XmmRegister src = locations->InAt(1).AsFpuRegister<XmmRegister>();
   XmmRegister dst = locations->Out().AsFpuRegister<XmmRegister>();
   switch (instruction->GetPackedType()) {
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
       DCHECK_EQ(16u, instruction->GetVectorLength());
       __ paddb(dst, src);
@@ -490,15 +499,13 @@
   XmmRegister dst = locations->Out().AsFpuRegister<XmmRegister>();
 
   DCHECK(instruction->IsRounded());
-  DCHECK(instruction->IsUnsigned());
 
   switch (instruction->GetPackedType()) {
-    case DataType::Type::kInt8:
+    case DataType::Type::kUint8:
       DCHECK_EQ(16u, instruction->GetVectorLength());
      __ pavgb(dst, src);
      return;
     case DataType::Type::kUint16:
-    case DataType::Type::kInt16:
       DCHECK_EQ(8u, instruction->GetVectorLength());
       __ pavgw(dst, src);
       return;
@@ -518,6 +525,7 @@
   XmmRegister src = locations->InAt(1).AsFpuRegister<XmmRegister>();
   XmmRegister dst = locations->Out().AsFpuRegister<XmmRegister>();
   switch (instruction->GetPackedType()) {
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
       DCHECK_EQ(16u, instruction->GetVectorLength());
       __ psubb(dst, src);
@@ -616,22 +624,21 @@
   XmmRegister src = locations->InAt(1).AsFpuRegister<XmmRegister>();
   XmmRegister dst = locations->Out().AsFpuRegister<XmmRegister>();
   switch (instruction->GetPackedType()) {
+    case DataType::Type::kUint8:
+      DCHECK_EQ(16u, instruction->GetVectorLength());
+      __ pminub(dst, src);
+      break;
     case DataType::Type::kInt8:
       DCHECK_EQ(16u, instruction->GetVectorLength());
-      if (instruction->IsUnsigned()) {
-        __ pminub(dst, src);
-      } else {
-        __ pminsb(dst, src);
-      }
+      __ pminsb(dst, src);
       break;
     case DataType::Type::kUint16:
+      DCHECK_EQ(8u, instruction->GetVectorLength());
+      __ pminuw(dst, src);
+      break;
     case DataType::Type::kInt16:
       DCHECK_EQ(8u, instruction->GetVectorLength());
-      if (instruction->IsUnsigned()) {
-        __ pminuw(dst, src);
-      } else {
-        __ pminsw(dst, src);
-      }
+      __ pminsw(dst, src);
       break;
     case DataType::Type::kInt32:
       DCHECK_EQ(4u, instruction->GetVectorLength());
@@ -668,22 +675,21 @@
   XmmRegister src = locations->InAt(1).AsFpuRegister<XmmRegister>();
   XmmRegister dst = locations->Out().AsFpuRegister<XmmRegister>();
   switch (instruction->GetPackedType()) {
+    case DataType::Type::kUint8:
+      DCHECK_EQ(16u, instruction->GetVectorLength());
+      __ pmaxub(dst, src);
+      break;
     case DataType::Type::kInt8:
       DCHECK_EQ(16u, instruction->GetVectorLength());
-      if (instruction->IsUnsigned()) {
-        __ pmaxub(dst, src);
-      } else {
-        __ pmaxsb(dst, src);
-      }
+      __ pmaxsb(dst, src);
       break;
     case DataType::Type::kUint16:
+      DCHECK_EQ(8u, instruction->GetVectorLength());
+      __ pmaxuw(dst, src);
+      break;
     case DataType::Type::kInt16:
       DCHECK_EQ(8u, instruction->GetVectorLength());
-      if (instruction->IsUnsigned()) {
-        __ pmaxuw(dst, src);
-      } else {
-        __ pmaxsw(dst, src);
-      }
+      __ pmaxsw(dst, src);
       break;
     case DataType::Type::kInt32:
       DCHECK_EQ(4u, instruction->GetVectorLength());
@@ -721,6 +727,7 @@
   XmmRegister dst = locations->Out().AsFpuRegister<XmmRegister>();
   switch (instruction->GetPackedType()) {
     case DataType::Type::kBool:
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
     case DataType::Type::kUint16:
     case DataType::Type::kInt16:
@@ -755,6 +762,7 @@
   XmmRegister dst = locations->Out().AsFpuRegister<XmmRegister>();
   switch (instruction->GetPackedType()) {
     case DataType::Type::kBool:
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
     case DataType::Type::kUint16:
     case DataType::Type::kInt16:
@@ -789,6 +797,7 @@
   XmmRegister dst = locations->Out().AsFpuRegister<XmmRegister>();
   switch (instruction->GetPackedType()) {
     case DataType::Type::kBool:
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
     case DataType::Type::kUint16:
     case DataType::Type::kInt16:
@@ -823,6 +832,7 @@
   XmmRegister dst = locations->Out().AsFpuRegister<XmmRegister>();
   switch (instruction->GetPackedType()) {
     case DataType::Type::kBool:
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
     case DataType::Type::kUint16:
     case DataType::Type::kInt16:
@@ -963,6 +973,7 @@
       }
       FALLTHROUGH_INTENDED;
     case DataType::Type::kBool:
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
     case DataType::Type::kUint16:
     case DataType::Type::kInt16:
@@ -1000,6 +1011,7 @@
   // Set required elements.
   switch (instruction->GetPackedType()) {
     case DataType::Type::kBool:
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
     case DataType::Type::kUint16:
     case DataType::Type::kInt16:  // TODO: up to here, and?
@@ -1036,6 +1048,7 @@
 static void CreateVecAccumLocations(ArenaAllocator* arena, HVecOperation* instruction) {
   LocationSummary* locations = new (arena) LocationSummary(instruction);
   switch (instruction->GetPackedType()) {
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
     case DataType::Type::kUint16:
     case DataType::Type::kInt16:
@@ -1077,6 +1090,7 @@
   LocationSummary* locations = new (arena) LocationSummary(instruction);
   switch (instruction->GetPackedType()) {
     case DataType::Type::kBool:
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
     case DataType::Type::kUint16:
     case DataType::Type::kInt16:
@@ -1156,6 +1170,7 @@
       }
       FALLTHROUGH_INTENDED;
     case DataType::Type::kBool:
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
     case DataType::Type::kInt16:
     case DataType::Type::kInt32:
@@ -1190,6 +1205,7 @@
   bool is_aligned16 = instruction->GetAlignment().IsAlignedAt(16);
   switch (instruction->GetPackedType()) {
     case DataType::Type::kBool:
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
     case DataType::Type::kUint16:
     case DataType::Type::kInt16:
diff --git a/compiler/optimizing/code_generator_vector_x86_64.cc b/compiler/optimizing/code_generator_vector_x86_64.cc
index 4241042..2270f6b 100644
--- a/compiler/optimizing/code_generator_vector_x86_64.cc
+++ b/compiler/optimizing/code_generator_vector_x86_64.cc
@@ -31,6 +31,7 @@
   bool is_zero = IsZeroBitPattern(input);
   switch (instruction->GetPackedType()) {
     case DataType::Type::kBool:
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
     case DataType::Type::kUint16:
     case DataType::Type::kInt16:
@@ -65,6 +66,7 @@
 
   switch (instruction->GetPackedType()) {
     case DataType::Type::kBool:
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
       DCHECK_EQ(16u, instruction->GetVectorLength());
       __ movd(dst, locations->InAt(0).AsRegister<CpuRegister>(), /*64-bit*/ false);
@@ -109,6 +111,7 @@
   LocationSummary* locations = new (GetGraph()->GetArena()) LocationSummary(instruction);
   switch (instruction->GetPackedType()) {
     case DataType::Type::kBool:
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
     case DataType::Type::kUint16:
     case DataType::Type::kInt16:
@@ -133,6 +136,7 @@
   XmmRegister src = locations->InAt(0).AsFpuRegister<XmmRegister>();
   switch (instruction->GetPackedType()) {
     case DataType::Type::kBool:
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
     case DataType::Type::kUint16:
     case DataType::Type::kInt16:  // TODO: up to here, and?
@@ -163,6 +167,7 @@
   LocationSummary* locations = new (arena) LocationSummary(instruction);
   switch (instruction->GetPackedType()) {
     case DataType::Type::kBool:
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
     case DataType::Type::kUint16:
     case DataType::Type::kInt16:
@@ -273,6 +278,7 @@
   XmmRegister src = locations->InAt(0).AsFpuRegister<XmmRegister>();
   XmmRegister dst = locations->Out().AsFpuRegister<XmmRegister>();
   switch (instruction->GetPackedType()) {
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
       DCHECK_EQ(16u, instruction->GetVectorLength());
       __ pxor(dst, dst);
@@ -373,6 +379,7 @@
       __ pxor(dst, src);
       break;
     }
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
     case DataType::Type::kUint16:
     case DataType::Type::kInt16:
@@ -404,6 +411,7 @@
   LocationSummary* locations = new (arena) LocationSummary(instruction);
   switch (instruction->GetPackedType()) {
     case DataType::Type::kBool:
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
     case DataType::Type::kUint16:
     case DataType::Type::kInt16:
@@ -431,6 +439,7 @@
   XmmRegister src = locations->InAt(1).AsFpuRegister<XmmRegister>();
   XmmRegister dst = locations->Out().AsFpuRegister<XmmRegister>();
   switch (instruction->GetPackedType()) {
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
       DCHECK_EQ(16u, instruction->GetVectorLength());
       __ paddb(dst, src);
@@ -473,15 +482,13 @@
   XmmRegister dst = locations->Out().AsFpuRegister<XmmRegister>();
 
   DCHECK(instruction->IsRounded());
-  DCHECK(instruction->IsUnsigned());
 
   switch (instruction->GetPackedType()) {
-    case DataType::Type::kInt8:
+    case DataType::Type::kUint8:
       DCHECK_EQ(16u, instruction->GetVectorLength());
      __ pavgb(dst, src);
      return;
     case DataType::Type::kUint16:
-    case DataType::Type::kInt16:
       DCHECK_EQ(8u, instruction->GetVectorLength());
       __ pavgw(dst, src);
       return;
@@ -501,6 +508,7 @@
   XmmRegister src = locations->InAt(1).AsFpuRegister<XmmRegister>();
   XmmRegister dst = locations->Out().AsFpuRegister<XmmRegister>();
   switch (instruction->GetPackedType()) {
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
       DCHECK_EQ(16u, instruction->GetVectorLength());
       __ psubb(dst, src);
@@ -599,22 +607,21 @@
   XmmRegister src = locations->InAt(1).AsFpuRegister<XmmRegister>();
   XmmRegister dst = locations->Out().AsFpuRegister<XmmRegister>();
   switch (instruction->GetPackedType()) {
+    case DataType::Type::kUint8:
+      DCHECK_EQ(16u, instruction->GetVectorLength());
+      __ pminub(dst, src);
+      break;
     case DataType::Type::kInt8:
       DCHECK_EQ(16u, instruction->GetVectorLength());
-      if (instruction->IsUnsigned()) {
-        __ pminub(dst, src);
-      } else {
-        __ pminsb(dst, src);
-      }
+      __ pminsb(dst, src);
       break;
     case DataType::Type::kUint16:
+      DCHECK_EQ(8u, instruction->GetVectorLength());
+      __ pminuw(dst, src);
+      break;
     case DataType::Type::kInt16:
       DCHECK_EQ(8u, instruction->GetVectorLength());
-      if (instruction->IsUnsigned()) {
-        __ pminuw(dst, src);
-      } else {
-        __ pminsw(dst, src);
-      }
+      __ pminsw(dst, src);
       break;
     case DataType::Type::kInt32:
       DCHECK_EQ(4u, instruction->GetVectorLength());
@@ -651,22 +658,21 @@
   XmmRegister src = locations->InAt(1).AsFpuRegister<XmmRegister>();
   XmmRegister dst = locations->Out().AsFpuRegister<XmmRegister>();
   switch (instruction->GetPackedType()) {
+    case DataType::Type::kUint8:
+      DCHECK_EQ(16u, instruction->GetVectorLength());
+      __ pmaxub(dst, src);
+      break;
     case DataType::Type::kInt8:
       DCHECK_EQ(16u, instruction->GetVectorLength());
-      if (instruction->IsUnsigned()) {
-        __ pmaxub(dst, src);
-      } else {
-        __ pmaxsb(dst, src);
-      }
+      __ pmaxsb(dst, src);
       break;
     case DataType::Type::kUint16:
+      DCHECK_EQ(8u, instruction->GetVectorLength());
+      __ pmaxuw(dst, src);
+      break;
     case DataType::Type::kInt16:
       DCHECK_EQ(8u, instruction->GetVectorLength());
-      if (instruction->IsUnsigned()) {
-        __ pmaxuw(dst, src);
-      } else {
-        __ pmaxsw(dst, src);
-      }
+      __ pmaxsw(dst, src);
       break;
     case DataType::Type::kInt32:
       DCHECK_EQ(4u, instruction->GetVectorLength());
@@ -704,6 +710,7 @@
   XmmRegister dst = locations->Out().AsFpuRegister<XmmRegister>();
   switch (instruction->GetPackedType()) {
     case DataType::Type::kBool:
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
     case DataType::Type::kUint16:
     case DataType::Type::kInt16:
@@ -738,6 +745,7 @@
   XmmRegister dst = locations->Out().AsFpuRegister<XmmRegister>();
   switch (instruction->GetPackedType()) {
     case DataType::Type::kBool:
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
     case DataType::Type::kUint16:
     case DataType::Type::kInt16:
@@ -772,6 +780,7 @@
   XmmRegister dst = locations->Out().AsFpuRegister<XmmRegister>();
   switch (instruction->GetPackedType()) {
     case DataType::Type::kBool:
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
     case DataType::Type::kUint16:
     case DataType::Type::kInt16:
@@ -806,6 +815,7 @@
   XmmRegister dst = locations->Out().AsFpuRegister<XmmRegister>();
   switch (instruction->GetPackedType()) {
     case DataType::Type::kBool:
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
     case DataType::Type::kUint16:
     case DataType::Type::kInt16:
@@ -940,6 +950,7 @@
 
   switch (instruction->GetPackedType()) {
     case DataType::Type::kBool:
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
     case DataType::Type::kUint16:
     case DataType::Type::kInt16:
@@ -978,6 +989,7 @@
   // Set required elements.
   switch (instruction->GetPackedType()) {
     case DataType::Type::kBool:
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
     case DataType::Type::kUint16:
     case DataType::Type::kInt16:  // TODO: up to here, and?
@@ -1009,6 +1021,7 @@
 static void CreateVecAccumLocations(ArenaAllocator* arena, HVecOperation* instruction) {
   LocationSummary* locations = new (arena) LocationSummary(instruction);
   switch (instruction->GetPackedType()) {
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
     case DataType::Type::kUint16:
     case DataType::Type::kInt16:
@@ -1050,6 +1063,7 @@
   LocationSummary* locations = new (arena) LocationSummary(instruction);
   switch (instruction->GetPackedType()) {
     case DataType::Type::kBool:
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
     case DataType::Type::kUint16:
     case DataType::Type::kInt16:
@@ -1129,6 +1143,7 @@
       }
       FALLTHROUGH_INTENDED;
     case DataType::Type::kBool:
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
     case DataType::Type::kInt16:
     case DataType::Type::kInt32:
@@ -1163,6 +1178,7 @@
   bool is_aligned16 = instruction->GetAlignment().IsAlignedAt(16);
   switch (instruction->GetPackedType()) {
     case DataType::Type::kBool:
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
     case DataType::Type::kUint16:
     case DataType::Type::kInt16:
diff --git a/compiler/optimizing/code_generator_x86.cc b/compiler/optimizing/code_generator_x86.cc
index 70e270e..3515649 100644
--- a/compiler/optimizing/code_generator_x86.cc
+++ b/compiler/optimizing/code_generator_x86.cc
@@ -1131,12 +1131,13 @@
 
 Location InvokeDexCallingConventionVisitorX86::GetReturnLocation(DataType::Type type) const {
   switch (type) {
+    case DataType::Type::kReference:
     case DataType::Type::kBool:
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
     case DataType::Type::kUint16:
     case DataType::Type::kInt16:
     case DataType::Type::kInt32:
-    case DataType::Type::kReference:
       return Location::RegisterLocation(EAX);
 
     case DataType::Type::kInt64:
@@ -1159,12 +1160,13 @@
 
 Location InvokeDexCallingConventionVisitorX86::GetNextLocation(DataType::Type type) {
   switch (type) {
+    case DataType::Type::kReference:
     case DataType::Type::kBool:
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
     case DataType::Type::kUint16:
     case DataType::Type::kInt16:
-    case DataType::Type::kInt32:
-    case DataType::Type::kReference: {
+    case DataType::Type::kInt32: {
       uint32_t index = gp_index_++;
       stack_index_++;
       if (index < calling_convention.GetNumberOfRegisters()) {
@@ -2099,12 +2101,13 @@
   LocationSummary* locations =
       new (GetGraph()->GetArena()) LocationSummary(ret, LocationSummary::kNoCall);
   switch (ret->InputAt(0)->GetType()) {
+    case DataType::Type::kReference:
     case DataType::Type::kBool:
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
     case DataType::Type::kUint16:
     case DataType::Type::kInt16:
     case DataType::Type::kInt32:
-    case DataType::Type::kReference:
       locations->SetInAt(0, Location::RegisterLocation(EAX));
       break;
 
@@ -2127,12 +2130,13 @@
 void InstructionCodeGeneratorX86::VisitReturn(HReturn* ret) {
   if (kIsDebugBuild) {
     switch (ret->InputAt(0)->GetType()) {
+      case DataType::Type::kReference:
       case DataType::Type::kBool:
+      case DataType::Type::kUint8:
       case DataType::Type::kInt8:
       case DataType::Type::kUint16:
       case DataType::Type::kInt16:
       case DataType::Type::kInt32:
-      case DataType::Type::kReference:
         DCHECK_EQ(ret->GetLocations()->InAt(0).AsRegister<Register>(), EAX);
         break;
 
@@ -2408,7 +2412,8 @@
 void LocationsBuilderX86::VisitTypeConversion(HTypeConversion* conversion) {
   DataType::Type result_type = conversion->GetResultType();
   DataType::Type input_type = conversion->GetInputType();
-  DCHECK_NE(result_type, input_type);
+  DCHECK(!DataType::IsTypeConversionImplicit(input_type, result_type))
+      << input_type << " -> " << result_type;
 
   // The float-to-long and double-to-long type conversions rely on a
   // call to the runtime.
@@ -2420,14 +2425,21 @@
   LocationSummary* locations =
       new (GetGraph()->GetArena()) LocationSummary(conversion, call_kind);
 
-  // The Java language does not allow treating boolean as an integral type but
-  // our bit representation makes it safe.
-
   switch (result_type) {
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
       switch (input_type) {
+        case DataType::Type::kUint8:
+        case DataType::Type::kInt8:
+        case DataType::Type::kUint16:
+        case DataType::Type::kInt16:
+        case DataType::Type::kInt32:
+          locations->SetInAt(0, Location::ByteRegisterOrConstant(ECX, conversion->InputAt(0)));
+          // Make the output overlap to please the register allocator. This greatly simplifies
+          // the validation of the linear scan implementation
+          locations->SetOut(Location::RequiresRegister(), Location::kOutputOverlap);
+          break;
         case DataType::Type::kInt64: {
-          // Type conversion from long to byte is a result of code transformations.
           HInstruction* input = conversion->InputAt(0);
           Location input_location = input->IsConstant()
               ? Location::ConstantLocation(input->AsConstant())
@@ -2438,17 +2450,6 @@
           locations->SetOut(Location::RequiresRegister(), Location::kOutputOverlap);
           break;
         }
-        case DataType::Type::kBool:
-          // Boolean input is a result of code transformations.
-        case DataType::Type::kInt16:
-        case DataType::Type::kInt32:
-        case DataType::Type::kUint16:
-          // Processing a Dex `int-to-byte' instruction.
-          locations->SetInAt(0, Location::ByteRegisterOrConstant(ECX, conversion->InputAt(0)));
-          // Make the output overlap to please the register allocator. This greatly simplifies
-          // the validation of the linear scan implementation
-          locations->SetOut(Location::RequiresRegister(), Location::kOutputOverlap);
-          break;
 
         default:
           LOG(FATAL) << "Unexpected type conversion from " << input_type
@@ -2456,43 +2457,27 @@
       }
       break;
 
+    case DataType::Type::kUint16:
     case DataType::Type::kInt16:
-      switch (input_type) {
-        case DataType::Type::kInt64:
-          // Type conversion from long to short is a result of code transformations.
-        case DataType::Type::kBool:
-          // Boolean input is a result of code transformations.
-        case DataType::Type::kInt8:
-        case DataType::Type::kInt32:
-        case DataType::Type::kUint16:
-          // Processing a Dex `int-to-short' instruction.
-          locations->SetInAt(0, Location::Any());
-          locations->SetOut(Location::RequiresRegister(), Location::kNoOutputOverlap);
-          break;
-
-        default:
-          LOG(FATAL) << "Unexpected type conversion from " << input_type
-                     << " to " << result_type;
-      }
+      DCHECK(DataType::IsIntegralType(input_type)) << input_type;
+      locations->SetInAt(0, Location::Any());
+      locations->SetOut(Location::RequiresRegister(), Location::kNoOutputOverlap);
       break;
 
     case DataType::Type::kInt32:
       switch (input_type) {
         case DataType::Type::kInt64:
-          // Processing a Dex `long-to-int' instruction.
           locations->SetInAt(0, Location::Any());
           locations->SetOut(Location::RequiresRegister(), Location::kNoOutputOverlap);
           break;
 
         case DataType::Type::kFloat32:
-          // Processing a Dex `float-to-int' instruction.
           locations->SetInAt(0, Location::RequiresFpuRegister());
           locations->SetOut(Location::RequiresRegister());
           locations->AddTemp(Location::RequiresFpuRegister());
           break;
 
         case DataType::Type::kFloat64:
-          // Processing a Dex `double-to-int' instruction.
           locations->SetInAt(0, Location::RequiresFpuRegister());
           locations->SetOut(Location::RequiresRegister());
           locations->AddTemp(Location::RequiresFpuRegister());
@@ -2507,19 +2492,17 @@
     case DataType::Type::kInt64:
       switch (input_type) {
         case DataType::Type::kBool:
-          // Boolean input is a result of code transformations.
+        case DataType::Type::kUint8:
         case DataType::Type::kInt8:
+        case DataType::Type::kUint16:
         case DataType::Type::kInt16:
         case DataType::Type::kInt32:
-        case DataType::Type::kUint16:
-          // Processing a Dex `int-to-long' instruction.
           locations->SetInAt(0, Location::RegisterLocation(EAX));
           locations->SetOut(Location::RegisterPairLocation(EAX, EDX));
           break;
 
         case DataType::Type::kFloat32:
         case DataType::Type::kFloat64: {
-          // Processing a Dex `float-to-long' or 'double-to-long' instruction.
           InvokeRuntimeCallingConvention calling_convention;
           XmmRegister parameter = calling_convention.GetFpuRegisterAt(0);
           locations->SetInAt(0, Location::FpuRegisterLocation(parameter));
@@ -2535,47 +2518,24 @@
       }
       break;
 
-    case DataType::Type::kUint16:
-      switch (input_type) {
-        case DataType::Type::kInt64:
-          // Type conversion from long to char is a result of code transformations.
-        case DataType::Type::kBool:
-          // Boolean input is a result of code transformations.
-        case DataType::Type::kInt8:
-        case DataType::Type::kInt16:
-        case DataType::Type::kInt32:
-          // Processing a Dex `int-to-char' instruction.
-          locations->SetInAt(0, Location::Any());
-          locations->SetOut(Location::RequiresRegister(), Location::kNoOutputOverlap);
-          break;
-
-        default:
-          LOG(FATAL) << "Unexpected type conversion from " << input_type
-                     << " to " << result_type;
-      }
-      break;
-
     case DataType::Type::kFloat32:
       switch (input_type) {
         case DataType::Type::kBool:
-          // Boolean input is a result of code transformations.
+        case DataType::Type::kUint8:
         case DataType::Type::kInt8:
+        case DataType::Type::kUint16:
         case DataType::Type::kInt16:
         case DataType::Type::kInt32:
-        case DataType::Type::kUint16:
-          // Processing a Dex `int-to-float' instruction.
           locations->SetInAt(0, Location::RequiresRegister());
           locations->SetOut(Location::RequiresFpuRegister());
           break;
 
         case DataType::Type::kInt64:
-          // Processing a Dex `long-to-float' instruction.
           locations->SetInAt(0, Location::Any());
           locations->SetOut(Location::Any());
           break;
 
         case DataType::Type::kFloat64:
-          // Processing a Dex `double-to-float' instruction.
           locations->SetInAt(0, Location::RequiresFpuRegister());
           locations->SetOut(Location::RequiresFpuRegister(), Location::kNoOutputOverlap);
           break;
@@ -2589,24 +2549,21 @@
     case DataType::Type::kFloat64:
       switch (input_type) {
         case DataType::Type::kBool:
-          // Boolean input is a result of code transformations.
+        case DataType::Type::kUint8:
         case DataType::Type::kInt8:
+        case DataType::Type::kUint16:
         case DataType::Type::kInt16:
         case DataType::Type::kInt32:
-        case DataType::Type::kUint16:
-          // Processing a Dex `int-to-double' instruction.
           locations->SetInAt(0, Location::RequiresRegister());
           locations->SetOut(Location::RequiresFpuRegister());
           break;
 
         case DataType::Type::kInt64:
-          // Processing a Dex `long-to-double' instruction.
           locations->SetInAt(0, Location::Any());
           locations->SetOut(Location::Any());
           break;
 
         case DataType::Type::kFloat32:
-          // Processing a Dex `float-to-double' instruction.
           locations->SetInAt(0, Location::RequiresFpuRegister());
           locations->SetOut(Location::RequiresFpuRegister(), Location::kNoOutputOverlap);
           break;
@@ -2629,12 +2586,54 @@
   Location in = locations->InAt(0);
   DataType::Type result_type = conversion->GetResultType();
   DataType::Type input_type = conversion->GetInputType();
-  DCHECK_NE(result_type, input_type);
+  DCHECK(!DataType::IsTypeConversionImplicit(input_type, result_type))
+      << input_type << " -> " << result_type;
   switch (result_type) {
+    case DataType::Type::kUint8:
+      switch (input_type) {
+        case DataType::Type::kInt8:
+        case DataType::Type::kUint16:
+        case DataType::Type::kInt16:
+        case DataType::Type::kInt32:
+          if (in.IsRegister()) {
+            __ movzxb(out.AsRegister<Register>(), in.AsRegister<ByteRegister>());
+          } else {
+            DCHECK(in.GetConstant()->IsIntConstant());
+            int32_t value = in.GetConstant()->AsIntConstant()->GetValue();
+            __ movl(out.AsRegister<Register>(), Immediate(static_cast<uint8_t>(value)));
+          }
+          break;
+        case DataType::Type::kInt64:
+          if (in.IsRegisterPair()) {
+            __ movzxb(out.AsRegister<Register>(), in.AsRegisterPairLow<ByteRegister>());
+          } else {
+            DCHECK(in.GetConstant()->IsLongConstant());
+            int64_t value = in.GetConstant()->AsLongConstant()->GetValue();
+            __ movl(out.AsRegister<Register>(), Immediate(static_cast<uint8_t>(value)));
+          }
+          break;
+
+        default:
+          LOG(FATAL) << "Unexpected type conversion from " << input_type
+                     << " to " << result_type;
+      }
+      break;
+
     case DataType::Type::kInt8:
       switch (input_type) {
+        case DataType::Type::kUint8:
+        case DataType::Type::kUint16:
+        case DataType::Type::kInt16:
+        case DataType::Type::kInt32:
+          if (in.IsRegister()) {
+            __ movsxb(out.AsRegister<Register>(), in.AsRegister<ByteRegister>());
+          } else {
+            DCHECK(in.GetConstant()->IsIntConstant());
+            int32_t value = in.GetConstant()->AsIntConstant()->GetValue();
+            __ movl(out.AsRegister<Register>(), Immediate(static_cast<int8_t>(value)));
+          }
+          break;
         case DataType::Type::kInt64:
-          // Type conversion from long to byte is a result of code transformations.
           if (in.IsRegisterPair()) {
             __ movsxb(out.AsRegister<Register>(), in.AsRegisterPairLow<ByteRegister>());
           } else {
@@ -2643,18 +2642,37 @@
             __ movl(out.AsRegister<Register>(), Immediate(static_cast<int8_t>(value)));
           }
           break;
-        case DataType::Type::kBool:
-          // Boolean input is a result of code transformations.
+
+        default:
+          LOG(FATAL) << "Unexpected type conversion from " << input_type
+                     << " to " << result_type;
+      }
+      break;
+
+    case DataType::Type::kUint16:
+      switch (input_type) {
+        case DataType::Type::kInt8:
         case DataType::Type::kInt16:
         case DataType::Type::kInt32:
-        case DataType::Type::kUint16:
-          // Processing a Dex `int-to-byte' instruction.
           if (in.IsRegister()) {
-            __ movsxb(out.AsRegister<Register>(), in.AsRegister<ByteRegister>());
+            __ movzxw(out.AsRegister<Register>(), in.AsRegister<Register>());
+          } else if (in.IsStackSlot()) {
+            __ movzxw(out.AsRegister<Register>(), Address(ESP, in.GetStackIndex()));
           } else {
             DCHECK(in.GetConstant()->IsIntConstant());
             int32_t value = in.GetConstant()->AsIntConstant()->GetValue();
-            __ movl(out.AsRegister<Register>(), Immediate(static_cast<int8_t>(value)));
+            __ movl(out.AsRegister<Register>(), Immediate(static_cast<uint16_t>(value)));
+          }
+          break;
+        case DataType::Type::kInt64:
+          if (in.IsRegisterPair()) {
+            __ movzxw(out.AsRegister<Register>(), in.AsRegisterPairLow<Register>());
+          } else if (in.IsDoubleStackSlot()) {
+            __ movzxw(out.AsRegister<Register>(), Address(ESP, in.GetStackIndex()));
+          } else {
+            DCHECK(in.GetConstant()->IsLongConstant());
+            int64_t value = in.GetConstant()->AsLongConstant()->GetValue();
+            __ movl(out.AsRegister<Register>(), Immediate(static_cast<uint16_t>(value)));
           }
           break;
 
@@ -2666,24 +2684,8 @@
 
     case DataType::Type::kInt16:
       switch (input_type) {
-        case DataType::Type::kInt64:
-          // Type conversion from long to short is a result of code transformations.
-          if (in.IsRegisterPair()) {
-            __ movsxw(out.AsRegister<Register>(), in.AsRegisterPairLow<Register>());
-          } else if (in.IsDoubleStackSlot()) {
-            __ movsxw(out.AsRegister<Register>(), Address(ESP, in.GetStackIndex()));
-          } else {
-            DCHECK(in.GetConstant()->IsLongConstant());
-            int64_t value = in.GetConstant()->AsLongConstant()->GetValue();
-            __ movl(out.AsRegister<Register>(), Immediate(static_cast<int16_t>(value)));
-          }
-          break;
-        case DataType::Type::kBool:
-          // Boolean input is a result of code transformations.
-        case DataType::Type::kInt8:
-        case DataType::Type::kInt32:
         case DataType::Type::kUint16:
-          // Processing a Dex `int-to-short' instruction.
+        case DataType::Type::kInt32:
           if (in.IsRegister()) {
             __ movsxw(out.AsRegister<Register>(), in.AsRegister<Register>());
           } else if (in.IsStackSlot()) {
@@ -2694,6 +2696,17 @@
             __ movl(out.AsRegister<Register>(), Immediate(static_cast<int16_t>(value)));
           }
           break;
+        case DataType::Type::kInt64:
+          if (in.IsRegisterPair()) {
+            __ movsxw(out.AsRegister<Register>(), in.AsRegisterPairLow<Register>());
+          } else if (in.IsDoubleStackSlot()) {
+            __ movsxw(out.AsRegister<Register>(), Address(ESP, in.GetStackIndex()));
+          } else {
+            DCHECK(in.GetConstant()->IsLongConstant());
+            int64_t value = in.GetConstant()->AsLongConstant()->GetValue();
+            __ movl(out.AsRegister<Register>(), Immediate(static_cast<int16_t>(value)));
+          }
+          break;
 
         default:
           LOG(FATAL) << "Unexpected type conversion from " << input_type
@@ -2704,7 +2717,6 @@
     case DataType::Type::kInt32:
       switch (input_type) {
         case DataType::Type::kInt64:
-          // Processing a Dex `long-to-int' instruction.
           if (in.IsRegisterPair()) {
             __ movl(out.AsRegister<Register>(), in.AsRegisterPairLow<Register>());
           } else if (in.IsDoubleStackSlot()) {
@@ -2718,7 +2730,6 @@
           break;
 
         case DataType::Type::kFloat32: {
-          // Processing a Dex `float-to-int' instruction.
           XmmRegister input = in.AsFpuRegister<XmmRegister>();
           Register output = out.AsRegister<Register>();
           XmmRegister temp = locations->GetTemp(0).AsFpuRegister<XmmRegister>();
@@ -2743,7 +2754,6 @@
         }
 
         case DataType::Type::kFloat64: {
-          // Processing a Dex `double-to-int' instruction.
           XmmRegister input = in.AsFpuRegister<XmmRegister>();
           Register output = out.AsRegister<Register>();
           XmmRegister temp = locations->GetTemp(0).AsFpuRegister<XmmRegister>();
@@ -2776,12 +2786,11 @@
     case DataType::Type::kInt64:
       switch (input_type) {
         case DataType::Type::kBool:
-          // Boolean input is a result of code transformations.
+        case DataType::Type::kUint8:
         case DataType::Type::kInt8:
+        case DataType::Type::kUint16:
         case DataType::Type::kInt16:
         case DataType::Type::kInt32:
-        case DataType::Type::kUint16:
-          // Processing a Dex `int-to-long' instruction.
           DCHECK_EQ(out.AsRegisterPairLow<Register>(), EAX);
           DCHECK_EQ(out.AsRegisterPairHigh<Register>(), EDX);
           DCHECK_EQ(in.AsRegister<Register>(), EAX);
@@ -2789,13 +2798,11 @@
           break;
 
         case DataType::Type::kFloat32:
-          // Processing a Dex `float-to-long' instruction.
           codegen_->InvokeRuntime(kQuickF2l, conversion, conversion->GetDexPc());
           CheckEntrypointTypes<kQuickF2l, int64_t, float>();
           break;
 
         case DataType::Type::kFloat64:
-          // Processing a Dex `double-to-long' instruction.
           codegen_->InvokeRuntime(kQuickD2l, conversion, conversion->GetDexPc());
           CheckEntrypointTypes<kQuickD2l, int64_t, double>();
           break;
@@ -2806,57 +2813,18 @@
       }
       break;
 
-    case DataType::Type::kUint16:
-      switch (input_type) {
-        case DataType::Type::kInt64:
-          // Type conversion from long to short is a result of code transformations.
-          if (in.IsRegisterPair()) {
-            __ movzxw(out.AsRegister<Register>(), in.AsRegisterPairLow<Register>());
-          } else if (in.IsDoubleStackSlot()) {
-            __ movzxw(out.AsRegister<Register>(), Address(ESP, in.GetStackIndex()));
-          } else {
-            DCHECK(in.GetConstant()->IsLongConstant());
-            int64_t value = in.GetConstant()->AsLongConstant()->GetValue();
-            __ movl(out.AsRegister<Register>(), Immediate(static_cast<uint16_t>(value)));
-          }
-          break;
-        case DataType::Type::kBool:
-          // Boolean input is a result of code transformations.
-        case DataType::Type::kInt8:
-        case DataType::Type::kInt16:
-        case DataType::Type::kInt32:
-          // Processing a Dex `Process a Dex `int-to-char'' instruction.
-          if (in.IsRegister()) {
-            __ movzxw(out.AsRegister<Register>(), in.AsRegister<Register>());
-          } else if (in.IsStackSlot()) {
-            __ movzxw(out.AsRegister<Register>(), Address(ESP, in.GetStackIndex()));
-          } else {
-            DCHECK(in.GetConstant()->IsIntConstant());
-            int32_t value = in.GetConstant()->AsIntConstant()->GetValue();
-            __ movl(out.AsRegister<Register>(), Immediate(static_cast<uint16_t>(value)));
-          }
-          break;
-
-        default:
-          LOG(FATAL) << "Unexpected type conversion from " << input_type
-                     << " to " << result_type;
-      }
-      break;
-
     case DataType::Type::kFloat32:
       switch (input_type) {
         case DataType::Type::kBool:
-          // Boolean input is a result of code transformations.
+        case DataType::Type::kUint8:
         case DataType::Type::kInt8:
+        case DataType::Type::kUint16:
         case DataType::Type::kInt16:
         case DataType::Type::kInt32:
-        case DataType::Type::kUint16:
-          // Processing a Dex `int-to-float' instruction.
           __ cvtsi2ss(out.AsFpuRegister<XmmRegister>(), in.AsRegister<Register>());
           break;
 
         case DataType::Type::kInt64: {
-          // Processing a Dex `long-to-float' instruction.
           size_t adjustment = 0;
 
           // Create stack space for the call to
@@ -2886,7 +2854,6 @@
         }
 
         case DataType::Type::kFloat64:
-          // Processing a Dex `double-to-float' instruction.
           __ cvtsd2ss(out.AsFpuRegister<XmmRegister>(), in.AsFpuRegister<XmmRegister>());
           break;
 
@@ -2899,17 +2866,15 @@
     case DataType::Type::kFloat64:
       switch (input_type) {
         case DataType::Type::kBool:
-          // Boolean input is a result of code transformations.
+        case DataType::Type::kUint8:
         case DataType::Type::kInt8:
+        case DataType::Type::kUint16:
         case DataType::Type::kInt16:
         case DataType::Type::kInt32:
-        case DataType::Type::kUint16:
-          // Processing a Dex `int-to-double' instruction.
           __ cvtsi2sd(out.AsFpuRegister<XmmRegister>(), in.AsRegister<Register>());
           break;
 
         case DataType::Type::kInt64: {
-          // Processing a Dex `long-to-double' instruction.
           size_t adjustment = 0;
 
           // Create stack space for the call to
@@ -2939,7 +2904,6 @@
         }
 
         case DataType::Type::kFloat32:
-          // Processing a Dex `float-to-double' instruction.
           __ cvtss2sd(out.AsFpuRegister<XmmRegister>(), in.AsFpuRegister<XmmRegister>());
           break;
 
@@ -3832,6 +3796,7 @@
   LocationSummary* locations = codegen_->CreateThrowingSlowPathLocations(instruction);
   switch (instruction->GetType()) {
     case DataType::Type::kBool:
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
     case DataType::Type::kUint16:
     case DataType::Type::kInt16:
@@ -3860,6 +3825,7 @@
 
   switch (instruction->GetType()) {
     case DataType::Type::kBool:
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
     case DataType::Type::kUint16:
     case DataType::Type::kInt16:
@@ -4349,9 +4315,10 @@
       new (GetGraph()->GetArena()) LocationSummary(compare, LocationSummary::kNoCall);
   switch (compare->InputAt(0)->GetType()) {
     case DataType::Type::kBool:
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
-    case DataType::Type::kInt16:
     case DataType::Type::kUint16:
+    case DataType::Type::kInt16:
     case DataType::Type::kInt32:
     case DataType::Type::kInt64: {
       locations->SetInAt(0, Location::RequiresRegister());
@@ -4388,9 +4355,10 @@
 
   switch (compare->InputAt(0)->GetType()) {
     case DataType::Type::kBool:
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
-    case DataType::Type::kInt16:
     case DataType::Type::kUint16:
+    case DataType::Type::kInt16:
     case DataType::Type::kInt32: {
       codegen_->GenerateIntCompare(left, right);
       break;
@@ -4792,7 +4760,8 @@
   uint32_t offset = field_info.GetFieldOffset().Uint32Value();
 
   switch (field_type) {
-    case DataType::Type::kBool: {
+    case DataType::Type::kBool:
+    case DataType::Type::kUint8: {
       __ movzxb(out.AsRegister<Register>(), Address(base, offset));
       break;
     }
@@ -4802,13 +4771,13 @@
       break;
     }
 
-    case DataType::Type::kInt16: {
-      __ movsxw(out.AsRegister<Register>(), Address(base, offset));
+    case DataType::Type::kUint16: {
+      __ movzxw(out.AsRegister<Register>(), Address(base, offset));
       break;
     }
 
-    case DataType::Type::kUint16: {
-      __ movzxw(out.AsRegister<Register>(), Address(base, offset));
+    case DataType::Type::kInt16: {
+      __ movsxw(out.AsRegister<Register>(), Address(base, offset));
       break;
     }
 
@@ -4897,8 +4866,7 @@
   locations->SetInAt(0, Location::RequiresRegister());
   bool is_volatile = field_info.IsVolatile();
   DataType::Type field_type = field_info.GetFieldType();
-  bool is_byte_type = (field_type == DataType::Type::kBool)
-    || (field_type == DataType::Type::kInt8);
+  bool is_byte_type = DataType::Size(field_type) == 1u;
 
   // The register allocator does not support multiple
   // inputs that die at entry with one in a specific register.
@@ -4957,13 +4925,14 @@
 
   switch (field_type) {
     case DataType::Type::kBool:
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8: {
       __ movb(Address(base, offset), value.AsRegister<ByteRegister>());
       break;
     }
 
-    case DataType::Type::kInt16:
-    case DataType::Type::kUint16: {
+    case DataType::Type::kUint16:
+    case DataType::Type::kInt16: {
       if (value.IsConstant()) {
         __ movw(Address(base, offset),
                 Immediate(CodeGenerator::GetInt16ValueOf(value.GetConstant())));
@@ -5242,7 +5211,8 @@
 
   DataType::Type type = instruction->GetType();
   switch (type) {
-    case DataType::Type::kBool: {
+    case DataType::Type::kBool:
+    case DataType::Type::kUint8: {
       Register out = out_loc.AsRegister<Register>();
       __ movzxb(out, CodeGeneratorX86::ArrayAddress(obj, index, TIMES_1, data_offset));
       break;
@@ -5254,12 +5224,6 @@
       break;
     }
 
-    case DataType::Type::kInt16: {
-      Register out = out_loc.AsRegister<Register>();
-      __ movsxw(out, CodeGeneratorX86::ArrayAddress(obj, index, TIMES_2, data_offset));
-      break;
-    }
-
     case DataType::Type::kUint16: {
       Register out = out_loc.AsRegister<Register>();
       if (mirror::kUseStringCompression && instruction->IsStringCharAt()) {
@@ -5284,6 +5248,12 @@
       break;
     }
 
+    case DataType::Type::kInt16: {
+      Register out = out_loc.AsRegister<Register>();
+      __ movsxw(out, CodeGeneratorX86::ArrayAddress(obj, index, TIMES_2, data_offset));
+      break;
+    }
+
     case DataType::Type::kInt32: {
       Register out = out_loc.AsRegister<Register>();
       __ movl(out, CodeGeneratorX86::ArrayAddress(obj, index, TIMES_4, data_offset));
@@ -5368,8 +5338,7 @@
           LocationSummary::kCallOnSlowPath :
           LocationSummary::kNoCall);
 
-  bool is_byte_type = (value_type == DataType::Type::kBool)
-      || (value_type == DataType::Type::kInt8);
+  bool is_byte_type = DataType::Size(value_type) == 1u;
   // We need the inputs to be different than the output in case of long operation.
   // In case of a byte operation, the register allocator does not support multiple
   // inputs that die at entry with one in a specific register.
@@ -5407,6 +5376,7 @@
 
   switch (value_type) {
     case DataType::Type::kBool:
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8: {
       uint32_t offset = mirror::Array::DataOffset(sizeof(uint8_t)).Uint32Value();
       Address address = CodeGeneratorX86::ArrayAddress(array, index, TIMES_1, offset);
@@ -5419,8 +5389,8 @@
       break;
     }
 
-    case DataType::Type::kInt16:
-    case DataType::Type::kUint16: {
+    case DataType::Type::kUint16:
+    case DataType::Type::kInt16: {
       uint32_t offset = mirror::Array::DataOffset(sizeof(uint16_t)).Uint32Value();
       Address address = CodeGeneratorX86::ArrayAddress(array, index, TIMES_2, offset);
       if (value.IsRegister()) {
diff --git a/compiler/optimizing/code_generator_x86_64.cc b/compiler/optimizing/code_generator_x86_64.cc
index 42704e9..e8bfa66 100644
--- a/compiler/optimizing/code_generator_x86_64.cc
+++ b/compiler/optimizing/code_generator_x86_64.cc
@@ -1521,6 +1521,7 @@
   DataType::Type type = condition->InputAt(0)->GetType();
   switch (type) {
     case DataType::Type::kBool:
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
     case DataType::Type::kUint16:
     case DataType::Type::kInt16:
@@ -2036,9 +2037,10 @@
       new (GetGraph()->GetArena()) LocationSummary(compare, LocationSummary::kNoCall);
   switch (compare->InputAt(0)->GetType()) {
     case DataType::Type::kBool:
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
-    case DataType::Type::kInt16:
     case DataType::Type::kUint16:
+    case DataType::Type::kInt16:
     case DataType::Type::kInt32:
     case DataType::Type::kInt64: {
       locations->SetInAt(0, Location::RequiresRegister());
@@ -2070,9 +2072,10 @@
 
   switch (type) {
     case DataType::Type::kBool:
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
-    case DataType::Type::kInt16:
     case DataType::Type::kUint16:
+    case DataType::Type::kInt16:
     case DataType::Type::kInt32: {
       codegen_->GenerateIntCompare(left, right);
       break;
@@ -2207,12 +2210,13 @@
   LocationSummary* locations =
       new (GetGraph()->GetArena()) LocationSummary(ret, LocationSummary::kNoCall);
   switch (ret->InputAt(0)->GetType()) {
+    case DataType::Type::kReference:
     case DataType::Type::kBool:
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
     case DataType::Type::kUint16:
     case DataType::Type::kInt16:
     case DataType::Type::kInt32:
-    case DataType::Type::kReference:
     case DataType::Type::kInt64:
       locations->SetInAt(0, Location::RegisterLocation(RAX));
       break;
@@ -2230,12 +2234,13 @@
 void InstructionCodeGeneratorX86_64::VisitReturn(HReturn* ret) {
   if (kIsDebugBuild) {
     switch (ret->InputAt(0)->GetType()) {
+      case DataType::Type::kReference:
       case DataType::Type::kBool:
+      case DataType::Type::kUint8:
       case DataType::Type::kInt8:
       case DataType::Type::kUint16:
       case DataType::Type::kInt16:
       case DataType::Type::kInt32:
-      case DataType::Type::kReference:
       case DataType::Type::kInt64:
         DCHECK_EQ(ret->GetLocations()->InAt(0).AsRegister<CpuRegister>().AsRegister(), RAX);
         break;
@@ -2255,12 +2260,13 @@
 
 Location InvokeDexCallingConventionVisitorX86_64::GetReturnLocation(DataType::Type type) const {
   switch (type) {
+    case DataType::Type::kReference:
     case DataType::Type::kBool:
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
     case DataType::Type::kUint16:
     case DataType::Type::kInt16:
     case DataType::Type::kInt32:
-    case DataType::Type::kReference:
     case DataType::Type::kInt64:
       return Location::RegisterLocation(RAX);
 
@@ -2281,12 +2287,13 @@
 
 Location InvokeDexCallingConventionVisitorX86_64::GetNextLocation(DataType::Type type) {
   switch (type) {
+    case DataType::Type::kReference:
     case DataType::Type::kBool:
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
     case DataType::Type::kUint16:
     case DataType::Type::kInt16:
-    case DataType::Type::kInt32:
-    case DataType::Type::kReference: {
+    case DataType::Type::kInt32: {
       uint32_t index = gp_index_++;
       stack_index_++;
       if (index < calling_convention.GetNumberOfRegisters()) {
@@ -2536,68 +2543,32 @@
       new (GetGraph()->GetArena()) LocationSummary(conversion, LocationSummary::kNoCall);
   DataType::Type result_type = conversion->GetResultType();
   DataType::Type input_type = conversion->GetInputType();
-  DCHECK_NE(result_type, input_type);
-
-  // The Java language does not allow treating boolean as an integral type but
-  // our bit representation makes it safe.
+  DCHECK(!DataType::IsTypeConversionImplicit(input_type, result_type))
+      << input_type << " -> " << result_type;
 
   switch (result_type) {
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
-      switch (input_type) {
-        case DataType::Type::kInt64:
-          // Type conversion from long to byte is a result of code transformations.
-        case DataType::Type::kBool:
-          // Boolean input is a result of code transformations.
-        case DataType::Type::kInt16:
-        case DataType::Type::kInt32:
-        case DataType::Type::kUint16:
-          // Processing a Dex `int-to-byte' instruction.
-          locations->SetInAt(0, Location::Any());
-          locations->SetOut(Location::RequiresRegister(), Location::kNoOutputOverlap);
-          break;
-
-        default:
-          LOG(FATAL) << "Unexpected type conversion from " << input_type
-                     << " to " << result_type;
-      }
-      break;
-
+    case DataType::Type::kUint16:
     case DataType::Type::kInt16:
-      switch (input_type) {
-        case DataType::Type::kInt64:
-          // Type conversion from long to short is a result of code transformations.
-        case DataType::Type::kBool:
-          // Boolean input is a result of code transformations.
-        case DataType::Type::kInt8:
-        case DataType::Type::kInt32:
-        case DataType::Type::kUint16:
-          // Processing a Dex `int-to-short' instruction.
-          locations->SetInAt(0, Location::Any());
-          locations->SetOut(Location::RequiresRegister(), Location::kNoOutputOverlap);
-          break;
-
-        default:
-          LOG(FATAL) << "Unexpected type conversion from " << input_type
-                     << " to " << result_type;
-      }
+      DCHECK(DataType::IsIntegralType(input_type)) << input_type;
+      locations->SetInAt(0, Location::Any());
+      locations->SetOut(Location::RequiresRegister(), Location::kNoOutputOverlap);
       break;
 
     case DataType::Type::kInt32:
       switch (input_type) {
         case DataType::Type::kInt64:
-          // Processing a Dex `long-to-int' instruction.
           locations->SetInAt(0, Location::Any());
           locations->SetOut(Location::RequiresRegister(), Location::kNoOutputOverlap);
           break;
 
         case DataType::Type::kFloat32:
-          // Processing a Dex `float-to-int' instruction.
           locations->SetInAt(0, Location::RequiresFpuRegister());
           locations->SetOut(Location::RequiresRegister());
           break;
 
         case DataType::Type::kFloat64:
-          // Processing a Dex `double-to-int' instruction.
           locations->SetInAt(0, Location::RequiresFpuRegister());
           locations->SetOut(Location::RequiresRegister());
           break;
@@ -2611,12 +2582,11 @@
     case DataType::Type::kInt64:
       switch (input_type) {
         case DataType::Type::kBool:
-          // Boolean input is a result of code transformations.
+        case DataType::Type::kUint8:
         case DataType::Type::kInt8:
+        case DataType::Type::kUint16:
         case DataType::Type::kInt16:
         case DataType::Type::kInt32:
-        case DataType::Type::kUint16:
-          // Processing a Dex `int-to-long' instruction.
           // TODO: We would benefit from a (to-be-implemented)
           // Location::RegisterOrStackSlot requirement for this input.
           locations->SetInAt(0, Location::RequiresRegister());
@@ -2624,13 +2594,11 @@
           break;
 
         case DataType::Type::kFloat32:
-          // Processing a Dex `float-to-long' instruction.
           locations->SetInAt(0, Location::RequiresFpuRegister());
           locations->SetOut(Location::RequiresRegister());
           break;
 
         case DataType::Type::kFloat64:
-          // Processing a Dex `double-to-long' instruction.
           locations->SetInAt(0, Location::RequiresFpuRegister());
           locations->SetOut(Location::RequiresRegister());
           break;
@@ -2641,47 +2609,24 @@
       }
       break;
 
-    case DataType::Type::kUint16:
-      switch (input_type) {
-        case DataType::Type::kInt64:
-          // Type conversion from long to char is a result of code transformations.
-        case DataType::Type::kBool:
-          // Boolean input is a result of code transformations.
-        case DataType::Type::kInt8:
-        case DataType::Type::kInt16:
-        case DataType::Type::kInt32:
-          // Processing a Dex `int-to-char' instruction.
-          locations->SetInAt(0, Location::Any());
-          locations->SetOut(Location::RequiresRegister(), Location::kNoOutputOverlap);
-          break;
-
-        default:
-          LOG(FATAL) << "Unexpected type conversion from " << input_type
-                     << " to " << result_type;
-      }
-      break;
-
     case DataType::Type::kFloat32:
       switch (input_type) {
         case DataType::Type::kBool:
-          // Boolean input is a result of code transformations.
+        case DataType::Type::kUint8:
         case DataType::Type::kInt8:
+        case DataType::Type::kUint16:
         case DataType::Type::kInt16:
         case DataType::Type::kInt32:
-        case DataType::Type::kUint16:
-          // Processing a Dex `int-to-float' instruction.
           locations->SetInAt(0, Location::Any());
           locations->SetOut(Location::RequiresFpuRegister());
           break;
 
         case DataType::Type::kInt64:
-          // Processing a Dex `long-to-float' instruction.
           locations->SetInAt(0, Location::Any());
           locations->SetOut(Location::RequiresFpuRegister());
           break;
 
         case DataType::Type::kFloat64:
-          // Processing a Dex `double-to-float' instruction.
           locations->SetInAt(0, Location::Any());
           locations->SetOut(Location::RequiresFpuRegister(), Location::kNoOutputOverlap);
           break;
@@ -2695,24 +2640,21 @@
     case DataType::Type::kFloat64:
       switch (input_type) {
         case DataType::Type::kBool:
-          // Boolean input is a result of code transformations.
+        case DataType::Type::kUint8:
         case DataType::Type::kInt8:
+        case DataType::Type::kUint16:
         case DataType::Type::kInt16:
         case DataType::Type::kInt32:
-        case DataType::Type::kUint16:
-          // Processing a Dex `int-to-double' instruction.
           locations->SetInAt(0, Location::Any());
           locations->SetOut(Location::RequiresFpuRegister());
           break;
 
         case DataType::Type::kInt64:
-          // Processing a Dex `long-to-double' instruction.
           locations->SetInAt(0, Location::Any());
           locations->SetOut(Location::RequiresFpuRegister());
           break;
 
         case DataType::Type::kFloat32:
-          // Processing a Dex `float-to-double' instruction.
           locations->SetInAt(0, Location::Any());
           locations->SetOut(Location::RequiresFpuRegister(), Location::kNoOutputOverlap);
           break;
@@ -2735,18 +2677,40 @@
   Location in = locations->InAt(0);
   DataType::Type result_type = conversion->GetResultType();
   DataType::Type input_type = conversion->GetInputType();
-  DCHECK_NE(result_type, input_type);
+  DCHECK(!DataType::IsTypeConversionImplicit(input_type, result_type))
+      << input_type << " -> " << result_type;
   switch (result_type) {
-    case DataType::Type::kInt8:
+    case DataType::Type::kUint8:
       switch (input_type) {
-        case DataType::Type::kInt64:
-          // Type conversion from long to byte is a result of code transformations.
-        case DataType::Type::kBool:
-          // Boolean input is a result of code transformations.
+        case DataType::Type::kInt8:
+        case DataType::Type::kUint16:
         case DataType::Type::kInt16:
         case DataType::Type::kInt32:
+        case DataType::Type::kInt64:
+          if (in.IsRegister()) {
+            __ movzxb(out.AsRegister<CpuRegister>(), in.AsRegister<CpuRegister>());
+          } else if (in.IsStackSlot() || in.IsDoubleStackSlot()) {
+            __ movzxb(out.AsRegister<CpuRegister>(),
+                      Address(CpuRegister(RSP), in.GetStackIndex()));
+          } else {
+            __ movl(out.AsRegister<CpuRegister>(),
+                    Immediate(static_cast<uint8_t>(Int64FromConstant(in.GetConstant()))));
+          }
+          break;
+
+        default:
+          LOG(FATAL) << "Unexpected type conversion from " << input_type
+                     << " to " << result_type;
+      }
+      break;
+
+    case DataType::Type::kInt8:
+      switch (input_type) {
+        case DataType::Type::kUint8:
         case DataType::Type::kUint16:
-          // Processing a Dex `int-to-byte' instruction.
+        case DataType::Type::kInt16:
+        case DataType::Type::kInt32:
+        case DataType::Type::kInt64:
           if (in.IsRegister()) {
             __ movsxb(out.AsRegister<CpuRegister>(), in.AsRegister<CpuRegister>());
           } else if (in.IsStackSlot() || in.IsDoubleStackSlot()) {
@@ -2764,16 +2728,34 @@
       }
       break;
 
+    case DataType::Type::kUint16:
+      switch (input_type) {
+        case DataType::Type::kInt8:
+        case DataType::Type::kInt16:
+        case DataType::Type::kInt32:
+        case DataType::Type::kInt64:
+          if (in.IsRegister()) {
+            __ movzxw(out.AsRegister<CpuRegister>(), in.AsRegister<CpuRegister>());
+          } else if (in.IsStackSlot() || in.IsDoubleStackSlot()) {
+            __ movzxw(out.AsRegister<CpuRegister>(),
+                      Address(CpuRegister(RSP), in.GetStackIndex()));
+          } else {
+            __ movl(out.AsRegister<CpuRegister>(),
+                    Immediate(static_cast<uint16_t>(Int64FromConstant(in.GetConstant()))));
+          }
+          break;
+
+        default:
+          LOG(FATAL) << "Unexpected type conversion from " << input_type
+                     << " to " << result_type;
+      }
+      break;
+
     case DataType::Type::kInt16:
       switch (input_type) {
-        case DataType::Type::kInt64:
-          // Type conversion from long to short is a result of code transformations.
-        case DataType::Type::kBool:
-          // Boolean input is a result of code transformations.
-        case DataType::Type::kInt8:
-        case DataType::Type::kInt32:
         case DataType::Type::kUint16:
-          // Processing a Dex `int-to-short' instruction.
+        case DataType::Type::kInt32:
+        case DataType::Type::kInt64:
           if (in.IsRegister()) {
             __ movsxw(out.AsRegister<CpuRegister>(), in.AsRegister<CpuRegister>());
           } else if (in.IsStackSlot() || in.IsDoubleStackSlot()) {
@@ -2794,7 +2776,6 @@
     case DataType::Type::kInt32:
       switch (input_type) {
         case DataType::Type::kInt64:
-          // Processing a Dex `long-to-int' instruction.
           if (in.IsRegister()) {
             __ movl(out.AsRegister<CpuRegister>(), in.AsRegister<CpuRegister>());
           } else if (in.IsDoubleStackSlot()) {
@@ -2809,7 +2790,6 @@
           break;
 
         case DataType::Type::kFloat32: {
-          // Processing a Dex `float-to-int' instruction.
           XmmRegister input = in.AsFpuRegister<XmmRegister>();
           CpuRegister output = out.AsRegister<CpuRegister>();
           NearLabel done, nan;
@@ -2831,7 +2811,6 @@
         }
 
         case DataType::Type::kFloat64: {
-          // Processing a Dex `double-to-int' instruction.
           XmmRegister input = in.AsFpuRegister<XmmRegister>();
           CpuRegister output = out.AsRegister<CpuRegister>();
           NearLabel done, nan;
@@ -2862,18 +2841,16 @@
       switch (input_type) {
         DCHECK(out.IsRegister());
         case DataType::Type::kBool:
-          // Boolean input is a result of code transformations.
+        case DataType::Type::kUint8:
         case DataType::Type::kInt8:
+        case DataType::Type::kUint16:
         case DataType::Type::kInt16:
         case DataType::Type::kInt32:
-        case DataType::Type::kUint16:
-          // Processing a Dex `int-to-long' instruction.
           DCHECK(in.IsRegister());
           __ movsxd(out.AsRegister<CpuRegister>(), in.AsRegister<CpuRegister>());
           break;
 
         case DataType::Type::kFloat32: {
-          // Processing a Dex `float-to-long' instruction.
           XmmRegister input = in.AsFpuRegister<XmmRegister>();
           CpuRegister output = out.AsRegister<CpuRegister>();
           NearLabel done, nan;
@@ -2895,7 +2872,6 @@
         }
 
         case DataType::Type::kFloat64: {
-          // Processing a Dex `double-to-long' instruction.
           XmmRegister input = in.AsFpuRegister<XmmRegister>();
           CpuRegister output = out.AsRegister<CpuRegister>();
           NearLabel done, nan;
@@ -2922,42 +2898,14 @@
       }
       break;
 
-    case DataType::Type::kUint16:
-      switch (input_type) {
-        case DataType::Type::kInt64:
-          // Type conversion from long to char is a result of code transformations.
-        case DataType::Type::kBool:
-          // Boolean input is a result of code transformations.
-        case DataType::Type::kInt8:
-        case DataType::Type::kInt16:
-        case DataType::Type::kInt32:
-          // Processing a Dex `int-to-char' instruction.
-          if (in.IsRegister()) {
-            __ movzxw(out.AsRegister<CpuRegister>(), in.AsRegister<CpuRegister>());
-          } else if (in.IsStackSlot() || in.IsDoubleStackSlot()) {
-            __ movzxw(out.AsRegister<CpuRegister>(),
-                      Address(CpuRegister(RSP), in.GetStackIndex()));
-          } else {
-            __ movl(out.AsRegister<CpuRegister>(),
-                    Immediate(static_cast<uint16_t>(Int64FromConstant(in.GetConstant()))));
-          }
-          break;
-
-        default:
-          LOG(FATAL) << "Unexpected type conversion from " << input_type
-                     << " to " << result_type;
-      }
-      break;
-
     case DataType::Type::kFloat32:
       switch (input_type) {
         case DataType::Type::kBool:
-          // Boolean input is a result of code transformations.
+        case DataType::Type::kUint8:
         case DataType::Type::kInt8:
+        case DataType::Type::kUint16:
         case DataType::Type::kInt16:
         case DataType::Type::kInt32:
-        case DataType::Type::kUint16:
-          // Processing a Dex `int-to-float' instruction.
           if (in.IsRegister()) {
             __ cvtsi2ss(out.AsFpuRegister<XmmRegister>(), in.AsRegister<CpuRegister>(), false);
           } else if (in.IsConstant()) {
@@ -2971,7 +2919,6 @@
           break;
 
         case DataType::Type::kInt64:
-          // Processing a Dex `long-to-float' instruction.
           if (in.IsRegister()) {
             __ cvtsi2ss(out.AsFpuRegister<XmmRegister>(), in.AsRegister<CpuRegister>(), true);
           } else if (in.IsConstant()) {
@@ -2985,7 +2932,6 @@
           break;
 
         case DataType::Type::kFloat64:
-          // Processing a Dex `double-to-float' instruction.
           if (in.IsFpuRegister()) {
             __ cvtsd2ss(out.AsFpuRegister<XmmRegister>(), in.AsFpuRegister<XmmRegister>());
           } else if (in.IsConstant()) {
@@ -3007,12 +2953,11 @@
     case DataType::Type::kFloat64:
       switch (input_type) {
         case DataType::Type::kBool:
-          // Boolean input is a result of code transformations.
+        case DataType::Type::kUint8:
         case DataType::Type::kInt8:
+        case DataType::Type::kUint16:
         case DataType::Type::kInt16:
         case DataType::Type::kInt32:
-        case DataType::Type::kUint16:
-          // Processing a Dex `int-to-double' instruction.
           if (in.IsRegister()) {
             __ cvtsi2sd(out.AsFpuRegister<XmmRegister>(), in.AsRegister<CpuRegister>(), false);
           } else if (in.IsConstant()) {
@@ -3026,7 +2971,6 @@
           break;
 
         case DataType::Type::kInt64:
-          // Processing a Dex `long-to-double' instruction.
           if (in.IsRegister()) {
             __ cvtsi2sd(out.AsFpuRegister<XmmRegister>(), in.AsRegister<CpuRegister>(), true);
           } else if (in.IsConstant()) {
@@ -3040,7 +2984,6 @@
           break;
 
         case DataType::Type::kFloat32:
-          // Processing a Dex `float-to-double' instruction.
           if (in.IsFpuRegister()) {
             __ cvtss2sd(out.AsFpuRegister<XmmRegister>(), in.AsFpuRegister<XmmRegister>());
           } else if (in.IsConstant()) {
@@ -3883,6 +3826,7 @@
 
   switch (instruction->GetType()) {
     case DataType::Type::kBool:
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
     case DataType::Type::kUint16:
     case DataType::Type::kInt16:
@@ -4290,7 +4234,8 @@
   uint32_t offset = field_info.GetFieldOffset().Uint32Value();
 
   switch (field_type) {
-    case DataType::Type::kBool: {
+    case DataType::Type::kBool:
+    case DataType::Type::kUint8: {
       __ movzxb(out.AsRegister<CpuRegister>(), Address(base, offset));
       break;
     }
@@ -4300,13 +4245,13 @@
       break;
     }
 
-    case DataType::Type::kInt16: {
-      __ movsxw(out.AsRegister<CpuRegister>(), Address(base, offset));
+    case DataType::Type::kUint16: {
+      __ movzxw(out.AsRegister<CpuRegister>(), Address(base, offset));
       break;
     }
 
-    case DataType::Type::kUint16: {
-      __ movzxw(out.AsRegister<CpuRegister>(), Address(base, offset));
+    case DataType::Type::kInt16: {
+      __ movsxw(out.AsRegister<CpuRegister>(), Address(base, offset));
       break;
     }
 
@@ -4433,6 +4378,7 @@
 
   switch (field_type) {
     case DataType::Type::kBool:
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8: {
       if (value.IsConstant()) {
         __ movb(Address(base, offset),
@@ -4443,8 +4389,8 @@
       break;
     }
 
-    case DataType::Type::kInt16:
-    case DataType::Type::kUint16: {
+    case DataType::Type::kUint16:
+    case DataType::Type::kInt16: {
       if (value.IsConstant()) {
         __ movw(Address(base, offset),
                 Immediate(CodeGenerator::GetInt16ValueOf(value.GetConstant())));
@@ -4714,7 +4660,8 @@
 
   DataType::Type type = instruction->GetType();
   switch (type) {
-    case DataType::Type::kBool: {
+    case DataType::Type::kBool:
+    case DataType::Type::kUint8: {
       CpuRegister out = out_loc.AsRegister<CpuRegister>();
       __ movzxb(out, CodeGeneratorX86_64::ArrayAddress(obj, index, TIMES_1, data_offset));
       break;
@@ -4726,12 +4673,6 @@
       break;
     }
 
-    case DataType::Type::kInt16: {
-      CpuRegister out = out_loc.AsRegister<CpuRegister>();
-      __ movsxw(out, CodeGeneratorX86_64::ArrayAddress(obj, index, TIMES_2, data_offset));
-      break;
-    }
-
     case DataType::Type::kUint16: {
       CpuRegister out = out_loc.AsRegister<CpuRegister>();
       if (mirror::kUseStringCompression && instruction->IsStringCharAt()) {
@@ -4754,6 +4695,12 @@
       break;
     }
 
+    case DataType::Type::kInt16: {
+      CpuRegister out = out_loc.AsRegister<CpuRegister>();
+      __ movsxw(out, CodeGeneratorX86_64::ArrayAddress(obj, index, TIMES_2, data_offset));
+      break;
+    }
+
     case DataType::Type::kInt32: {
       CpuRegister out = out_loc.AsRegister<CpuRegister>();
       __ movl(out, CodeGeneratorX86_64::ArrayAddress(obj, index, TIMES_4, data_offset));
@@ -4865,6 +4812,7 @@
 
   switch (value_type) {
     case DataType::Type::kBool:
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8: {
       uint32_t offset = mirror::Array::DataOffset(sizeof(uint8_t)).Uint32Value();
       Address address = CodeGeneratorX86_64::ArrayAddress(array, index, TIMES_1, offset);
@@ -4877,8 +4825,8 @@
       break;
     }
 
-    case DataType::Type::kInt16:
-    case DataType::Type::kUint16: {
+    case DataType::Type::kUint16:
+    case DataType::Type::kInt16: {
       uint32_t offset = mirror::Array::DataOffset(sizeof(uint16_t)).Uint32Value();
       Address address = CodeGeneratorX86_64::ArrayAddress(array, index, TIMES_2, offset);
       if (value.IsRegister()) {
diff --git a/compiler/optimizing/data_type-inl.h b/compiler/optimizing/data_type-inl.h
index fbc0c12..e389bad 100644
--- a/compiler/optimizing/data_type-inl.h
+++ b/compiler/optimizing/data_type-inl.h
@@ -46,17 +46,19 @@
 
 constexpr char DataType::TypeId(DataType::Type type) {
   // Type id for visualizer.
+  // Types corresponding to Java types are given a lower-case version of their shorty character.
   switch (type) {
-    case DataType::Type::kBool: return 'z';
-    case DataType::Type::kInt8: return 'b';
-    case DataType::Type::kUint16: return 'c';
-    case DataType::Type::kInt16: return 's';
-    case DataType::Type::kInt32: return 'i';
-    case DataType::Type::kInt64: return 'j';
-    case DataType::Type::kFloat32: return 'f';
-    case DataType::Type::kFloat64: return 'd';
-    case DataType::Type::kReference: return 'l';
-    case DataType::Type::kVoid: return 'v';
+    case DataType::Type::kBool: return 'z';       // Java boolean (Z).
+    case DataType::Type::kUint8: return 'a';      // The character before Java byte's 'b'.
+    case DataType::Type::kInt8: return 'b';       // Java byte (B).
+    case DataType::Type::kUint16: return 'c';     // Java char (C).
+    case DataType::Type::kInt16: return 's';      // Java short (S).
+    case DataType::Type::kInt32: return 'i';      // Java int (I).
+    case DataType::Type::kInt64: return 'j';      // Java long (J).
+    case DataType::Type::kFloat32: return 'f';    // Java float (F).
+    case DataType::Type::kFloat64: return 'd';    // Java double (D).
+    case DataType::Type::kReference: return 'l';  // Java reference (L).
+    case DataType::Type::kVoid: return 'v';       // Java void (V).
   }
   LOG(FATAL) << "Unreachable";
   UNREACHABLE();
diff --git a/compiler/optimizing/data_type.cc b/compiler/optimizing/data_type.cc
index 6890617..3c99a76 100644
--- a/compiler/optimizing/data_type.cc
+++ b/compiler/optimizing/data_type.cc
@@ -21,6 +21,7 @@
 static const char* kTypeNames[] = {
     "Reference",
     "Bool",
+    "Uint8",
     "Int8",
     "Uint16",
     "Int16",
diff --git a/compiler/optimizing/data_type.h b/compiler/optimizing/data_type.h
index 08f9263..5a023ad 100644
--- a/compiler/optimizing/data_type.h
+++ b/compiler/optimizing/data_type.h
@@ -29,6 +29,7 @@
   enum class Type : uint8_t {
     kReference = 0,
     kBool,
+    kUint8,
     kInt8,
     kUint16,
     kInt16,
@@ -47,6 +48,7 @@
     switch (type) {
       case Type::kVoid:
       case Type::kBool:
+      case Type::kUint8:
       case Type::kInt8:
         return 0;
       case Type::kUint16:
@@ -71,6 +73,7 @@
       case Type::kVoid:
         return 0;
       case Type::kBool:
+      case Type::kUint8:
       case Type::kInt8:
         return 1;
       case Type::kUint16:
@@ -99,6 +102,7 @@
     // our bit representation makes it safe.
     switch (type) {
       case Type::kBool:
+      case Type::kUint8:
       case Type::kInt8:
       case Type::kUint16:
       case Type::kInt16:
@@ -118,10 +122,27 @@
     return type == Type::kInt64 || type == Type::kFloat64;
   }
 
+  static bool IsUnsignedType(Type type) {
+    return type == Type::kUint8 || type == Type::kUint16;
+  }
+
+  static Type ToSignedType(Type type) {
+    switch (type) {
+      case Type::kUint8:
+        return Type::kInt8;
+      case Type::kUint16:
+        return Type::kInt16;
+      default:
+        DCHECK(type != Type::kVoid && type != Type::kReference);
+        return type;
+    }
+  }
+
   // Return the general kind of `type`, fusing integer-like types as Type::kInt.
   static Type Kind(Type type) {
     switch (type) {
       case Type::kBool:
+      case Type::kUint8:
       case Type::kInt8:
       case Type::kInt16:
       case Type::kUint16:
@@ -136,6 +157,8 @@
     switch (type) {
       case Type::kBool:
         return std::numeric_limits<bool>::min();
+      case Type::kUint8:
+        return std::numeric_limits<uint8_t>::min();
       case Type::kInt8:
         return std::numeric_limits<int8_t>::min();
       case Type::kUint16:
@@ -156,6 +179,8 @@
     switch (type) {
       case Type::kBool:
         return std::numeric_limits<bool>::max();
+      case Type::kUint8:
+        return std::numeric_limits<uint8_t>::max();
       case Type::kInt8:
         return std::numeric_limits<int8_t>::max();
       case Type::kUint16:
@@ -172,6 +197,8 @@
     return 0;
   }
 
+  static bool IsTypeConversionImplicit(Type input_type, Type result_type);
+
   static const char* PrettyDescriptor(Type type);
 
  private:
@@ -179,6 +206,25 @@
 };
 std::ostream& operator<<(std::ostream& os, DataType::Type data_type);
 
+// Defined outside DataType to have the operator<< available for DCHECK_NE().
+inline bool DataType::IsTypeConversionImplicit(Type input_type, Type result_type) {
+  DCHECK_NE(DataType::Type::kVoid, result_type);
+  DCHECK_NE(DataType::Type::kVoid, input_type);
+
+  // Invariant: We should never generate a conversion to a Boolean value.
+  DCHECK_NE(DataType::Type::kBool, result_type);
+
+  // Besides conversion to the same type, integral conversions to non-Int64 types
+  // are implicit if the result value range covers the input value range, i.e.
+  // widening conversions that do not need to trim the sign bits.
+  return result_type == input_type ||
+         (result_type != Type::kInt64 &&
+          IsIntegralType(input_type) &&
+          IsIntegralType(result_type) &&
+          MinValueOfIntegralType(input_type) >= MinValueOfIntegralType(result_type) &&
+          MaxValueOfIntegralType(input_type) <= MaxValueOfIntegralType(result_type));
+}
+
 }  // namespace art
 
 #endif  // ART_COMPILER_OPTIMIZING_DATA_TYPE_H_
diff --git a/compiler/optimizing/data_type_test.cc b/compiler/optimizing/data_type_test.cc
index 927291a..3ce683a 100644
--- a/compiler/optimizing/data_type_test.cc
+++ b/compiler/optimizing/data_type_test.cc
@@ -18,6 +18,8 @@
 
 #include "data_type-inl.h"
 
+#include "base/array_ref.h"
+#include "base/macros.h"
 #include "primitive.h"
 
 namespace art {
@@ -57,4 +59,58 @@
 #undef CHECK_NAME
 }
 
+TEST(DataType, IsTypeConversionImplicit) {
+  static const DataType::Type kIntegralTypes[] = {
+      DataType::Type::kBool,
+      DataType::Type::kUint8,
+      DataType::Type::kInt8,
+      DataType::Type::kUint16,
+      DataType::Type::kInt16,
+      DataType::Type::kInt32,
+      DataType::Type::kInt64,
+  };
+  const ArrayRef<const DataType::Type> kIntegralInputTypes(kIntegralTypes);
+  // Note: kBool cannot be used as a result type.
+  DCHECK_EQ(kIntegralTypes[0], DataType::Type::kBool);
+  const ArrayRef<const DataType::Type> kIntegralResultTypes = kIntegralInputTypes.SubArray(1u);
+
+  static const bool kImplicitIntegralConversions[][arraysize(kIntegralTypes)] = {
+      //             Bool   Uint8   Int8 Uint16 Int16  Int32  Int64
+      { /*   Bool    N/A */  true,  true,  true,  true,  true, false },
+      { /*  Uint8    N/A */  true, false,  true,  true,  true, false },
+      { /*   Int8    N/A */ false,  true, false,  true,  true, false },
+      { /* Uint16    N/A */ false, false,  true, false,  true, false },
+      { /*  Int16    N/A */ false, false, false,  true,  true, false },
+      { /*  Int32    N/A */ false, false, false, false,  true, false },
+      { /*  Int64    N/A */ false, false, false, false, false,  true },
+  };
+  static_assert(arraysize(kIntegralTypes) == arraysize(kImplicitIntegralConversions), "size check");
+
+  for (size_t input_index = 0; input_index != kIntegralInputTypes.size(); ++input_index) {
+    DataType::Type input_type = kIntegralInputTypes[input_index];
+    for (size_t result_index = 1u; result_index != kIntegralResultTypes.size(); ++result_index) {
+      DataType::Type result_type = kIntegralResultTypes[result_index];
+      EXPECT_EQ(kImplicitIntegralConversions[input_index][result_index],
+                DataType::IsTypeConversionImplicit(input_type, result_type))
+          << input_type << " " << result_type;
+    }
+  }
+  for (DataType::Type input_type : kIntegralInputTypes) {
+    EXPECT_FALSE(DataType::IsTypeConversionImplicit(input_type, DataType::Type::kFloat32));
+    EXPECT_FALSE(DataType::IsTypeConversionImplicit(input_type, DataType::Type::kFloat64));
+  }
+  for (DataType::Type result_type : kIntegralResultTypes) {
+    EXPECT_FALSE(DataType::IsTypeConversionImplicit(DataType::Type::kFloat32, result_type));
+    EXPECT_FALSE(DataType::IsTypeConversionImplicit(DataType::Type::kFloat64, result_type));
+  }
+  EXPECT_TRUE(
+      DataType::IsTypeConversionImplicit(DataType::Type::kFloat32, DataType::Type::kFloat32));
+  EXPECT_FALSE(
+      DataType::IsTypeConversionImplicit(DataType::Type::kFloat32, DataType::Type::kFloat64));
+  EXPECT_FALSE(
+      DataType::IsTypeConversionImplicit(DataType::Type::kFloat64, DataType::Type::kFloat32));
+  EXPECT_TRUE(
+      DataType::IsTypeConversionImplicit(DataType::Type::kFloat64, DataType::Type::kFloat64));
+}
+
 }  // namespace art
diff --git a/compiler/optimizing/graph_visualizer.cc b/compiler/optimizing/graph_visualizer.cc
index 194f063..eccdccf 100644
--- a/compiler/optimizing/graph_visualizer.cc
+++ b/compiler/optimizing/graph_visualizer.cc
@@ -521,20 +521,28 @@
     StartAttributeStream("kind") << deoptimize->GetKind();
   }
 
+  void VisitVecOperation(HVecOperation* vec_operation) OVERRIDE {
+    StartAttributeStream("packed_type") << vec_operation->GetPackedType();
+  }
+
   void VisitVecHalvingAdd(HVecHalvingAdd* hadd) OVERRIDE {
+    VisitVecBinaryOperation(hadd);
     StartAttributeStream("unsigned") << std::boolalpha << hadd->IsUnsigned() << std::noboolalpha;
     StartAttributeStream("rounded") << std::boolalpha << hadd->IsRounded() << std::noboolalpha;
   }
 
   void VisitVecMin(HVecMin* min) OVERRIDE {
+    VisitVecBinaryOperation(min);
     StartAttributeStream("unsigned") << std::boolalpha << min->IsUnsigned() << std::noboolalpha;
   }
 
   void VisitVecMax(HVecMax* max) OVERRIDE {
+    VisitVecBinaryOperation(max);
     StartAttributeStream("unsigned") << std::boolalpha << max->IsUnsigned() << std::noboolalpha;
   }
 
   void VisitVecMultiplyAccumulate(HVecMultiplyAccumulate* instruction) OVERRIDE {
+    VisitVecOperation(instruction);
     StartAttributeStream("kind") << instruction->GetOpKind();
   }
 
diff --git a/compiler/optimizing/induction_var_analysis.cc b/compiler/optimizing/induction_var_analysis.cc
index fe286ab..eab17aa 100644
--- a/compiler/optimizing/induction_var_analysis.cc
+++ b/compiler/optimizing/induction_var_analysis.cc
@@ -59,14 +59,19 @@
 static bool IsNarrowingIntegralConversion(DataType::Type from, DataType::Type to) {
   switch (from) {
     case DataType::Type::kInt64:
-      return to == DataType::Type::kInt8 || to == DataType::Type::kInt16
-          || to == DataType::Type::kUint16 || to == DataType::Type::kInt32;
+      return to == DataType::Type::kUint8 ||
+             to == DataType::Type::kInt8 ||
+             to == DataType::Type::kUint16 ||
+             to == DataType::Type::kInt16 ||
+             to == DataType::Type::kInt32;
     case DataType::Type::kInt32:
-      return to == DataType::Type::kInt8 || to == DataType::Type::kInt16
-          || to == DataType::Type::kUint16;
+      return to == DataType::Type::kUint8 ||
+             to == DataType::Type::kInt8 ||
+             to == DataType::Type::kUint16 ||
+             to == DataType::Type::kInt16;
     case DataType::Type::kUint16:
     case DataType::Type::kInt16:
-      return to == DataType::Type::kInt8;
+      return to == DataType::Type::kUint8 || to == DataType::Type::kInt8;
     default:
       return false;
   }
@@ -77,10 +82,11 @@
  */
 static DataType::Type ImplicitConversion(DataType::Type type) {
   switch (type) {
-    case DataType::Type::kInt16:
-    case DataType::Type::kUint16:
-    case DataType::Type::kInt8:
     case DataType::Type::kBool:
+    case DataType::Type::kUint8:
+    case DataType::Type::kInt8:
+    case DataType::Type::kUint16:
+    case DataType::Type::kInt16:
       return DataType::Type::kInt32;
     default:
       return type;
@@ -1142,9 +1148,10 @@
 bool HInductionVarAnalysis::IsNarrowingLinear(InductionInfo* info) {
   return info != nullptr &&
       info->induction_class == kLinear &&
-      (info->type == DataType::Type::kInt8 ||
-       info->type == DataType::Type::kInt16 ||
+      (info->type == DataType::Type::kUint8 ||
+       info->type == DataType::Type::kInt8 ||
        info->type == DataType::Type::kUint16 ||
+       info->type == DataType::Type::kInt16 ||
        (info->type == DataType::Type::kInt32 && (info->op_a->type == DataType::Type::kInt64 ||
                                                  info->op_b->type == DataType::Type::kInt64)));
 }
diff --git a/compiler/optimizing/induction_var_range.cc b/compiler/optimizing/induction_var_range.cc
index 92b584c..ab6fbae 100644
--- a/compiler/optimizing/induction_var_range.cc
+++ b/compiler/optimizing/induction_var_range.cc
@@ -159,9 +159,10 @@
 /** Corrects a value for type to account for arithmetic wrap-around in lower precision. */
 static InductionVarRange::Value CorrectForType(InductionVarRange::Value v, DataType::Type type) {
   switch (type) {
-    case DataType::Type::kInt16:
+    case DataType::Type::kUint8:
+    case DataType::Type::kInt8:
     case DataType::Type::kUint16:
-    case DataType::Type::kInt8: {
+    case DataType::Type::kInt16: {
       // Constants within range only.
       // TODO: maybe some room for improvement, like allowing widening conversions
       int32_t min = DataType::MinValueOfIntegralType(type);
@@ -216,10 +217,11 @@
   // bounds check elimination, will have truncated higher precision induction
   // at their use point already).
   switch (info->type) {
-    case DataType::Type::kInt32:
-    case DataType::Type::kInt16:
-    case DataType::Type::kUint16:
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
+    case DataType::Type::kUint16:
+    case DataType::Type::kInt16:
+    case DataType::Type::kInt32:
       break;
     default:
       return false;
diff --git a/compiler/optimizing/instruction_simplifier.cc b/compiler/optimizing/instruction_simplifier.cc
index 3164ce1..a85354c 100644
--- a/compiler/optimizing/instruction_simplifier.cc
+++ b/compiler/optimizing/instruction_simplifier.cc
@@ -251,7 +251,8 @@
   InstructionSet isa = codegen_->GetInstructionSet();
   switch (isa) {
     case kArm64:
-      if (!(type == DataType::Type::kInt8 ||
+      if (!(type == DataType::Type::kUint8 ||
+            type == DataType::Type::kInt8 ||
             type == DataType::Type::kUint16 ||
             type == DataType::Type::kInt16 ||
             type == DataType::Type::kInt32)) {
@@ -260,7 +261,8 @@
       break;
     case kMips:
     case kMips64:
-      if (!(type == DataType::Type::kInt8 ||
+      if (!(type == DataType::Type::kUint8 ||
+            type == DataType::Type::kInt8 ||
             type == DataType::Type::kUint16 ||
             type == DataType::Type::kInt16 ||
             type == DataType::Type::kInt32 ||
@@ -876,10 +878,11 @@
   }
   DataType::Type type1 = a->GetType();
   DataType::Type type2 = b->GetType();
-  return (type1 == DataType::Type::kInt8  && type2 == DataType::Type::kInt8) ||
-         (type1 == DataType::Type::kInt16 && type2 == DataType::Type::kInt16) ||
-         (type1 == DataType::Type::kUint16  && type2 == DataType::Type::kUint16) ||
-         (type1 == DataType::Type::kInt32   && type2 == DataType::Type::kInt32 &&
+  return (type1 == DataType::Type::kUint8  && type2 == DataType::Type::kUint8) ||
+         (type1 == DataType::Type::kInt8   && type2 == DataType::Type::kInt8) ||
+         (type1 == DataType::Type::kInt16  && type2 == DataType::Type::kInt16) ||
+         (type1 == DataType::Type::kUint16 && type2 == DataType::Type::kUint16) ||
+         (type1 == DataType::Type::kInt32  && type2 == DataType::Type::kInt32 &&
           to_type == DataType::Type::kInt64);
 }
 
@@ -1036,30 +1039,13 @@
   }
 }
 
-static bool IsTypeConversionImplicit(DataType::Type input_type, DataType::Type result_type) {
-  // Invariant: We should never generate a conversion to a Boolean value.
-  DCHECK_NE(DataType::Type::kBool, result_type);
-
-  // Besides conversion to the same type, widening integral conversions are implicit,
-  // excluding conversions to long and the byte->char conversion where we need to
-  // clear the high 16 bits of the 32-bit sign-extended representation of byte.
-  return result_type == input_type ||
-      (result_type == DataType::Type::kInt32 && (input_type == DataType::Type::kBool ||
-                                                 input_type == DataType::Type::kInt8 ||
-                                                 input_type == DataType::Type::kInt16 ||
-                                                 input_type == DataType::Type::kUint16)) ||
-      (result_type == DataType::Type::kUint16 && input_type == DataType::Type::kBool) ||
-      (result_type == DataType::Type::kInt16 && (input_type == DataType::Type::kBool ||
-                                                 input_type == DataType::Type::kInt8)) ||
-      (result_type == DataType::Type::kInt8 && input_type == DataType::Type::kBool);
-}
-
 static bool IsTypeConversionLossless(DataType::Type input_type, DataType::Type result_type) {
   // The conversion to a larger type is loss-less with the exception of two cases,
-  //   - conversion to Uint16, the only unsigned type, where we may lose some bits, and
+  //   - conversion to the unsigned type Uint16, where we may lose some bits, and
   //   - conversion from float to long, the only FP to integral conversion with smaller FP type.
   // For integral to FP conversions this holds because the FP mantissa is large enough.
-  DCHECK_NE(input_type, result_type);
+  // Note: The size check excludes Uint8 as the result type.
+  DCHECK(!DataType::IsTypeConversionImplicit(input_type, result_type));
   return DataType::Size(result_type) > DataType::Size(input_type) &&
       result_type != DataType::Type::kUint16 &&
       !(result_type == DataType::Type::kInt64 && input_type == DataType::Type::kFloat32);
@@ -1069,7 +1055,7 @@
   HInstruction* input = instruction->GetInput();
   DataType::Type input_type = input->GetType();
   DataType::Type result_type = instruction->GetResultType();
-  if (IsTypeConversionImplicit(input_type, result_type)) {
+  if (DataType::IsTypeConversionImplicit(input_type, result_type)) {
     // Remove the implicit conversion; this includes conversion to the same type.
     instruction->ReplaceWith(input);
     instruction->GetBlock()->RemoveInstruction(instruction);
@@ -1098,7 +1084,7 @@
 
     if (is_first_conversion_lossless || integral_conversions_with_non_widening_second) {
       // If the merged conversion is implicit, do the simplification unconditionally.
-      if (IsTypeConversionImplicit(original_type, result_type)) {
+      if (DataType::IsTypeConversionImplicit(original_type, result_type)) {
         instruction->ReplaceWith(original_input);
         instruction->GetBlock()->RemoveInstruction(instruction);
         if (!input_conversion->HasUses()) {
@@ -1127,7 +1113,7 @@
       if (trailing_ones >= kBitsPerByte * DataType::Size(result_type)) {
         // The `HAnd` is useless, for example in `(byte) (x & 0xff)`, get rid of it.
         HInstruction* original_input = input_and->GetLeastConstantLeft();
-        if (IsTypeConversionImplicit(original_input->GetType(), result_type)) {
+        if (DataType::IsTypeConversionImplicit(original_input->GetType(), result_type)) {
           instruction->ReplaceWith(original_input);
           instruction->GetBlock()->RemoveInstruction(instruction);
           RecordSimplification();
@@ -2167,8 +2153,12 @@
   HBoundsCheck* bounds_check = new (arena) HBoundsCheck(
       index, length, dex_pc, invoke->GetDexMethodIndex());
   invoke->GetBlock()->InsertInstructionBefore(bounds_check, invoke);
-  HArrayGet* array_get = new (arena) HArrayGet(
-      str, bounds_check, DataType::Type::kUint16, dex_pc, /* is_string_char_at */ true);
+  HArrayGet* array_get = new (arena) HArrayGet(str,
+                                               bounds_check,
+                                               DataType::Type::kUint16,
+                                               SideEffects::None(),  // Strings are immutable.
+                                               dex_pc,
+                                               /* is_string_char_at */ true);
   invoke->GetBlock()->ReplaceAndRemoveInstructionWith(invoke, array_get);
   bounds_check->CopyEnvironmentFrom(invoke->GetEnvironment());
   GetGraph()->SetHasBoundsChecks(true);
diff --git a/compiler/optimizing/load_store_elimination.cc b/compiler/optimizing/load_store_elimination.cc
index bd14f2b..54c2d43 100644
--- a/compiler/optimizing/load_store_elimination.cc
+++ b/compiler/optimizing/load_store_elimination.cc
@@ -276,6 +276,7 @@
       case DataType::Type::kReference:
         return GetGraph()->GetNullConstant();
       case DataType::Type::kBool:
+      case DataType::Type::kUint8:
       case DataType::Type::kInt8:
       case DataType::Type::kUint16:
       case DataType::Type::kInt16:
diff --git a/compiler/optimizing/loop_optimization.cc b/compiler/optimizing/loop_optimization.cc
index fec64e2..2090a12 100644
--- a/compiler/optimizing/loop_optimization.cc
+++ b/compiler/optimizing/loop_optimization.cc
@@ -28,6 +28,46 @@
 
 namespace art {
 
+// TODO: Clean up the packed type detection so that we have the right type straight away
+// and do not need to go through this normalization.
+static inline void NormalizePackedType(/* inout */ DataType::Type* type,
+                                       /* inout */ bool* is_unsigned) {
+  switch (*type) {
+    case DataType::Type::kBool:
+      DCHECK(!*is_unsigned);
+      break;
+    case DataType::Type::kUint8:
+    case DataType::Type::kInt8:
+      if (*is_unsigned) {
+        *is_unsigned = false;
+        *type = DataType::Type::kUint8;
+      } else {
+        *type = DataType::Type::kInt8;
+      }
+      break;
+    case DataType::Type::kUint16:
+    case DataType::Type::kInt16:
+      if (*is_unsigned) {
+        *is_unsigned = false;
+        *type = DataType::Type::kUint16;
+      } else {
+        *type = DataType::Type::kInt16;
+      }
+      break;
+    case DataType::Type::kInt32:
+    case DataType::Type::kInt64:
+      // We do not have kUint32 and kUint64 at the moment.
+      break;
+    case DataType::Type::kFloat32:
+    case DataType::Type::kFloat64:
+      DCHECK(!*is_unsigned);
+      break;
+    default:
+      LOG(FATAL) << "Unexpected type " << *type;
+      UNREACHABLE();
+  }
+}
+
 // Enables vectorization (SIMDization) in the loop optimizer.
 static constexpr bool kEnableVectorization = true;
 
@@ -87,6 +127,7 @@
   int64_t value = 0;
   if (IsInt64AndGet(instruction, /*out*/ &value)) {
     switch (type) {
+      case DataType::Type::kUint8:
       case DataType::Type::kInt8:
         if (IsInt<8>(value)) {
           *operand = instruction;
@@ -151,6 +192,7 @@
   int64_t value = 0;
   if (IsInt64AndGet(instruction, /*out*/ &value)) {
     switch (type) {
+      case DataType::Type::kUint8:
       case DataType::Type::kInt8:
         if (IsUint<8>(value)) {
           *operand = instruction;
@@ -170,9 +212,13 @@
   }
   // An implicit widening conversion of any unsigned expression zero-extends.
   if (instruction->GetType() == type) {
-    if (type == DataType::Type::kUint16) {
-      *operand = instruction;
-      return true;
+    switch (type) {
+      case DataType::Type::kUint8:
+      case DataType::Type::kUint16:
+        *operand = instruction;
+        return true;
+      default:
+        return false;
     }
   }
   // A sign (or zero) extension followed by an explicit removal of just the
@@ -190,6 +236,7 @@
         (IsInt64AndGet(b, /*out*/ &mask) && (IsSignExtensionAndGet(a, type, /*out*/ operand) ||
                                              IsZeroExtensionAndGet(a, type, /*out*/ operand)))) {
       switch ((*operand)->GetType()) {
+        case DataType::Type::kUint8:
         case DataType::Type::kInt8:
           return mask == std::numeric_limits<uint8_t>::max();
         case DataType::Type::kUint16:
@@ -257,51 +304,10 @@
 
 // Compute relative vector length based on type difference.
 static size_t GetOtherVL(DataType::Type other_type, DataType::Type vector_type, size_t vl) {
-  switch (other_type) {
-    case DataType::Type::kBool:
-    case DataType::Type::kInt8:
-      switch (vector_type) {
-        case DataType::Type::kBool:
-        case DataType::Type::kInt8: return vl;
-        default: break;
-      }
-      return vl;
-    case DataType::Type::kUint16:
-    case DataType::Type::kInt16:
-      switch (vector_type) {
-        case DataType::Type::kBool:
-        case DataType::Type::kInt8: return vl >> 1;
-        case DataType::Type::kUint16:
-        case DataType::Type::kInt16: return vl;
-        default: break;
-      }
-      break;
-    case DataType::Type::kInt32:
-      switch (vector_type) {
-        case DataType::Type::kBool:
-        case DataType::Type::kInt8: return vl >> 2;
-        case DataType::Type::kUint16:
-        case DataType::Type::kInt16: return vl >> 1;
-        case DataType::Type::kInt32: return vl;
-        default: break;
-      }
-      break;
-    case DataType::Type::kInt64:
-      switch (vector_type) {
-        case DataType::Type::kBool:
-        case DataType::Type::kInt8: return vl >> 3;
-        case DataType::Type::kUint16:
-        case DataType::Type::kInt16: return vl >> 2;
-        case DataType::Type::kInt32: return vl >> 1;
-        case DataType::Type::kInt64: return vl;
-        default: break;
-      }
-      break;
-    default:
-      break;
-  }
-  LOG(FATAL) << "Unsupported idiom conversion";
-  UNREACHABLE();
+  DCHECK(DataType::IsIntegralType(other_type));
+  DCHECK(DataType::IsIntegralType(vector_type));
+  DCHECK_GE(DataType::SizeShift(other_type), DataType::SizeShift(vector_type));
+  return vl >> (DataType::SizeShift(other_type) - DataType::SizeShift(vector_type));
 }
 
 // Detect up to two instructions a and b, and an acccumulated constant c.
@@ -1105,19 +1111,19 @@
     return true;
   } else if (instruction->IsArrayGet()) {
     // Deal with vector restrictions.
-    if (instruction->AsArrayGet()->IsStringCharAt() &&
-        HasVectorRestrictions(restrictions, kNoStringCharAt)) {
+    bool is_string_char_at = instruction->AsArrayGet()->IsStringCharAt();
+    if (is_string_char_at && HasVectorRestrictions(restrictions, kNoStringCharAt)) {
       return false;
     }
     // Accept a right-hand-side array base[index] for
-    // (1) exact matching vector type,
+    // (1) matching vector type (exact match or signed/unsigned integral type of the same size),
     // (2) loop-invariant base,
     // (3) unit stride index,
     // (4) vectorizable right-hand-side value.
     HInstruction* base = instruction->InputAt(0);
     HInstruction* index = instruction->InputAt(1);
     HInstruction* offset = nullptr;
-    if (type == instruction->GetType() &&
+    if (DataType::ToSignedType(type) == DataType::ToSignedType(instruction->GetType()) &&
         node->loop_info->IsDefinedOutOfTheLoop(base) &&
         induction_range_.IsUnitStride(instruction, index, graph_, &offset)) {
       if (generate_code) {
@@ -1281,6 +1287,7 @@
         }
         if (VectorizeUse(node, r, generate_code, type, restrictions)) {
           if (generate_code) {
+            NormalizePackedType(&type, &is_unsigned);
             GenerateVecOp(instruction, vector_map_->Get(r), nullptr, type);
           }
           return true;
@@ -1340,6 +1347,7 @@
       // ARM 32-bit always supports advanced SIMD (64-bit SIMD).
       switch (type) {
         case DataType::Type::kBool:
+        case DataType::Type::kUint8:
         case DataType::Type::kInt8:
           *restrictions |= kNoDiv | kNoReduction;
           return TrySetVectorLength(8);
@@ -1359,6 +1367,7 @@
       // ARMv8 AArch64 always supports advanced SIMD (128-bit SIMD).
       switch (type) {
         case DataType::Type::kBool:
+        case DataType::Type::kUint8:
         case DataType::Type::kInt8:
           *restrictions |= kNoDiv;
           return TrySetVectorLength(16);
@@ -1387,6 +1396,7 @@
       if (features->AsX86InstructionSetFeatures()->HasSSE4_1()) {
         switch (type) {
           case DataType::Type::kBool:
+          case DataType::Type::kUint8:
           case DataType::Type::kInt8:
             *restrictions |=
                 kNoMul | kNoDiv | kNoShift | kNoAbs | kNoSignedHAdd | kNoUnroundedHAdd | kNoSAD;
@@ -1416,6 +1426,7 @@
       if (features->AsMipsInstructionSetFeatures()->HasMsa()) {
         switch (type) {
           case DataType::Type::kBool:
+          case DataType::Type::kUint8:
           case DataType::Type::kInt8:
             *restrictions |= kNoDiv | kNoReduction | kNoSAD;
             return TrySetVectorLength(16);
@@ -1444,6 +1455,7 @@
       if (features->AsMips64InstructionSetFeatures()->HasMsa()) {
         switch (type) {
           case DataType::Type::kBool:
+          case DataType::Type::kUint8:
           case DataType::Type::kInt8:
             *restrictions |= kNoDiv | kNoReduction | kNoSAD;
             return TrySetVectorLength(16);
@@ -1540,11 +1552,16 @@
     HInstruction* base = org->InputAt(0);
     if (opb != nullptr) {
       vector = new (global_allocator_) HVecStore(
-          global_allocator_, base, opa, opb, type, vector_length_);
+          global_allocator_, base, opa, opb, type, org->GetSideEffects(), vector_length_);
     } else  {
       bool is_string_char_at = org->AsArrayGet()->IsStringCharAt();
-      vector = new (global_allocator_) HVecLoad(
-          global_allocator_, base, opa, type, vector_length_, is_string_char_at);
+      vector = new (global_allocator_) HVecLoad(global_allocator_,
+                                                base,
+                                                opa,
+                                                type,
+                                                org->GetSideEffects(),
+                                                vector_length_,
+                                                is_string_char_at);
     }
     // Known dynamically enforced alignment?
     if (vector_peeling_candidate_ != nullptr &&
@@ -1556,11 +1573,12 @@
     // Scalar store or load.
     DCHECK(vector_mode_ == kSequential);
     if (opb != nullptr) {
-      vector = new (global_allocator_) HArraySet(org->InputAt(0), opa, opb, type, kNoDexPc);
+      vector = new (global_allocator_) HArraySet(
+          org->InputAt(0), opa, opb, type, org->GetSideEffects(), kNoDexPc);
     } else  {
       bool is_string_char_at = org->AsArrayGet()->IsStringCharAt();
       vector = new (global_allocator_) HArrayGet(
-          org->InputAt(0), opa, type, kNoDexPc, is_string_char_at);
+          org->InputAt(0), opa, type, org->GetSideEffects(), kNoDexPc, is_string_char_at);
     }
   }
   vector_map_->Put(org, vector);
@@ -1737,6 +1755,7 @@
           case Intrinsics::kMathMinLongLong:
           case Intrinsics::kMathMinFloatFloat:
           case Intrinsics::kMathMinDoubleDouble: {
+            NormalizePackedType(&type, &is_unsigned);
             vector = new (global_allocator_)
                 HVecMin(global_allocator_, opa, opb, type, vector_length_, is_unsigned);
             break;
@@ -1745,6 +1764,7 @@
           case Intrinsics::kMathMaxLongLong:
           case Intrinsics::kMathMaxFloatFloat:
           case Intrinsics::kMathMaxDoubleDouble: {
+            NormalizePackedType(&type, &is_unsigned);
             vector = new (global_allocator_)
                 HVecMax(global_allocator_, opa, opb, type, vector_length_, is_unsigned);
             break;
@@ -1857,14 +1877,15 @@
           VectorizeUse(node, s, generate_code, type, restrictions)) {
         if (generate_code) {
           if (vector_mode_ == kVector) {
+            NormalizePackedType(&type, &is_unsigned);
             vector_map_->Put(instruction, new (global_allocator_) HVecHalvingAdd(
                 global_allocator_,
                 vector_map_->Get(r),
                 vector_map_->Get(s),
                 type,
                 vector_length_,
-                is_unsigned,
-                is_rounded));
+                is_rounded,
+                is_unsigned));
             MaybeRecordStat(stats_, MethodCompilationStat::kLoopVectorizedIdiom);
           } else {
             GenerateVecOp(instruction, vector_map_->Get(r), vector_map_->Get(s), type);
@@ -1952,6 +1973,7 @@
       VectorizeUse(node, r, generate_code, sub_type, restrictions) &&
       VectorizeUse(node, s, generate_code, sub_type, restrictions)) {
     if (generate_code) {
+      NormalizePackedType(&reduction_type, &is_unsigned);
       if (vector_mode_ == kVector) {
         vector_map_->Put(instruction, new (global_allocator_) HVecSADAccumulate(
             global_allocator_,
diff --git a/compiler/optimizing/nodes.cc b/compiler/optimizing/nodes.cc
index 41ea998..cae5054 100644
--- a/compiler/optimizing/nodes.cc
+++ b/compiler/optimizing/nodes.cc
@@ -590,6 +590,7 @@
     case DataType::Type::kBool:
       DCHECK(IsUint<1>(value));
       FALLTHROUGH_INTENDED;
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
     case DataType::Type::kUint16:
     case DataType::Type::kInt16:
diff --git a/compiler/optimizing/nodes.h b/compiler/optimizing/nodes.h
index c49cee3..c75a512 100644
--- a/compiler/optimizing/nodes.h
+++ b/compiler/optimizing/nodes.h
@@ -1760,14 +1760,26 @@
   static constexpr uint64_t kAllReads =
       ((1ULL << (kLastBitForReads + 1 - kFieldReadOffset)) - 1) << kFieldReadOffset;
 
-  // Translates type to bit flag.
+  // Translates type to bit flag. The type must correspond to a Java type.
   static uint64_t TypeFlag(DataType::Type type, int offset) {
-    CHECK_NE(type, DataType::Type::kVoid);
-    const uint64_t one = 1;
-    const int shift = static_cast<int>(type);  // 0-based consecutive enum
+    int shift;
+    switch (type) {
+      case DataType::Type::kReference: shift = 0; break;
+      case DataType::Type::kBool:      shift = 1; break;
+      case DataType::Type::kInt8:      shift = 2; break;
+      case DataType::Type::kUint16:    shift = 3; break;
+      case DataType::Type::kInt16:     shift = 4; break;
+      case DataType::Type::kInt32:     shift = 5; break;
+      case DataType::Type::kInt64:     shift = 6; break;
+      case DataType::Type::kFloat32:   shift = 7; break;
+      case DataType::Type::kFloat64:   shift = 8; break;
+      default:
+        LOG(FATAL) << "Unexpected data type " << type;
+        UNREACHABLE();
+    }
     DCHECK_LE(kFieldWriteOffset, shift);
     DCHECK_LT(shift, kArrayWriteOffset);
-    return one << (shift + offset);
+    return UINT64_C(1) << (shift + offset);
   }
 
   // Private constructor on direct flags value.
@@ -5382,9 +5394,21 @@
   HArrayGet(HInstruction* array,
             HInstruction* index,
             DataType::Type type,
+            uint32_t dex_pc)
+     : HArrayGet(array,
+                 index,
+                 type,
+                 SideEffects::ArrayReadOfType(type),
+                 dex_pc,
+                 /* is_string_char_at */ false) {}
+
+  HArrayGet(HInstruction* array,
+            HInstruction* index,
+            DataType::Type type,
+            SideEffects side_effects,
             uint32_t dex_pc,
-            bool is_string_char_at = false)
-      : HExpression(type, SideEffects::ArrayReadOfType(type), dex_pc) {
+            bool is_string_char_at)
+      : HExpression(type, side_effects, dex_pc) {
     SetPackedFlag<kFlagIsStringCharAt>(is_string_char_at);
     SetRawInputAt(0, array);
     SetRawInputAt(1, index);
@@ -5453,7 +5477,21 @@
             HInstruction* value,
             DataType::Type expected_component_type,
             uint32_t dex_pc)
-      : HTemplateInstruction(SideEffects::None(), dex_pc) {
+      : HArraySet(array,
+                  index,
+                  value,
+                  expected_component_type,
+                  // Make a best guess for side effects now, may be refined during SSA building.
+                  ComputeSideEffects(GetComponentType(value->GetType(), expected_component_type)),
+                  dex_pc) {}
+
+  HArraySet(HInstruction* array,
+            HInstruction* index,
+            HInstruction* value,
+            DataType::Type expected_component_type,
+            SideEffects side_effects,
+            uint32_t dex_pc)
+      : HTemplateInstruction(side_effects, dex_pc) {
     SetPackedField<ExpectedComponentTypeField>(expected_component_type);
     SetPackedFlag<kFlagNeedsTypeCheck>(value->GetType() == DataType::Type::kReference);
     SetPackedFlag<kFlagValueCanBeNull>(true);
@@ -5461,8 +5499,6 @@
     SetRawInputAt(0, array);
     SetRawInputAt(1, index);
     SetRawInputAt(2, value);
-    // Make a best guess now, may be refined during SSA building.
-    ComputeSideEffects();
   }
 
   bool NeedsEnvironment() const OVERRIDE {
@@ -5501,24 +5537,26 @@
   HInstruction* GetValue() const { return InputAt(2); }
 
   DataType::Type GetComponentType() const {
+    return GetComponentType(GetValue()->GetType(), GetRawExpectedComponentType());
+  }
+
+  static DataType::Type GetComponentType(DataType::Type value_type,
+                                         DataType::Type expected_component_type) {
     // The Dex format does not type floating point index operations. Since the
-    // `expected_component_type_` is set during building and can therefore not
+    // `expected_component_type` comes from SSA building and can therefore not
     // be correct, we also check what is the value type. If it is a floating
     // point type, we must use that type.
-    DataType::Type value_type = GetValue()->GetType();
     return ((value_type == DataType::Type::kFloat32) || (value_type == DataType::Type::kFloat64))
         ? value_type
-        : GetRawExpectedComponentType();
+        : expected_component_type;
   }
 
   DataType::Type GetRawExpectedComponentType() const {
     return GetPackedField<ExpectedComponentTypeField>();
   }
 
-  void ComputeSideEffects() {
-    DataType::Type type = GetComponentType();
-    SetSideEffects(SideEffects::ArrayWriteOfType(type).Union(
-        SideEffectsForArchRuntimeCalls(type)));
+  static SideEffects ComputeSideEffects(DataType::Type type) {
+    return SideEffects::ArrayWriteOfType(type).Union(SideEffectsForArchRuntimeCalls(type));
   }
 
   static SideEffects SideEffectsForArchRuntimeCalls(DataType::Type value_type) {
diff --git a/compiler/optimizing/nodes_vector.h b/compiler/optimizing/nodes_vector.h
index 0aac260..d01f8c0 100644
--- a/compiler/optimizing/nodes_vector.h
+++ b/compiler/optimizing/nodes_vector.h
@@ -232,8 +232,10 @@
   DataType::Type input_type = input->AsVecOperation()->GetPackedType();
   switch (input_type) {
     case DataType::Type::kBool:
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
       return type == DataType::Type::kBool ||
+             type == DataType::Type::kUint8 ||
              type == DataType::Type::kInt8;
     case DataType::Type::kUint16:
     case DataType::Type::kInt16:
@@ -471,10 +473,14 @@
                  HInstruction* right,
                  DataType::Type packed_type,
                  size_t vector_length,
-                 bool is_unsigned,
                  bool is_rounded,
-                 uint32_t dex_pc = kNoDexPc)
-      : HVecBinaryOperation(arena, left, right, packed_type, vector_length, dex_pc) {
+                 bool is_unsigned = false)
+      : HVecBinaryOperation(arena, left, right, packed_type, vector_length, kNoDexPc) {
+    // The `is_unsigned` flag should be used exclusively with the Int32 or Int64.
+    // This flag is a temporary measure while we do not have the Uint32 and Uint64 data types.
+    DCHECK(!is_unsigned ||
+           packed_type == DataType::Type::kInt32 ||
+           packed_type == DataType::Type::kInt64) << packed_type;
     DCHECK(HasConsistentPackedTypes(left, packed_type));
     DCHECK(HasConsistentPackedTypes(right, packed_type));
     SetPackedFlag<kFieldHAddIsUnsigned>(is_unsigned);
@@ -584,9 +590,13 @@
           HInstruction* right,
           DataType::Type packed_type,
           size_t vector_length,
-          bool is_unsigned,
-          uint32_t dex_pc = kNoDexPc)
-      : HVecBinaryOperation(arena, left, right, packed_type, vector_length, dex_pc) {
+          bool is_unsigned = false)
+      : HVecBinaryOperation(arena, left, right, packed_type, vector_length, kNoDexPc) {
+    // The `is_unsigned` flag should be used exclusively with the Int32 or Int64.
+    // This flag is a temporary measure while we do not have the Uint32 and Uint64 data types.
+    DCHECK(!is_unsigned ||
+           packed_type == DataType::Type::kInt32 ||
+           packed_type == DataType::Type::kInt64) << packed_type;
     DCHECK(HasConsistentPackedTypes(left, packed_type));
     DCHECK(HasConsistentPackedTypes(right, packed_type));
     SetPackedFlag<kFieldMinOpIsUnsigned>(is_unsigned);
@@ -622,9 +632,13 @@
           HInstruction* right,
           DataType::Type packed_type,
           size_t vector_length,
-          bool is_unsigned,
-          uint32_t dex_pc = kNoDexPc)
-      : HVecBinaryOperation(arena, left, right, packed_type, vector_length, dex_pc) {
+          bool is_unsigned = false)
+      : HVecBinaryOperation(arena, left, right, packed_type, vector_length, kNoDexPc) {
+    // The `is_unsigned` flag should be used exclusively with the Int32 or Int64.
+    // This flag is a temporary measure while we do not have the Uint32 and Uint64 data types.
+    DCHECK(!is_unsigned ||
+           packed_type == DataType::Type::kInt32 ||
+           packed_type == DataType::Type::kInt64) << packed_type;
     DCHECK(HasConsistentPackedTypes(left, packed_type));
     DCHECK(HasConsistentPackedTypes(right, packed_type));
     SetPackedFlag<kFieldMaxOpIsUnsigned>(is_unsigned);
@@ -933,12 +947,13 @@
            HInstruction* base,
            HInstruction* index,
            DataType::Type packed_type,
+           SideEffects side_effects,
            size_t vector_length,
            bool is_string_char_at,
            uint32_t dex_pc = kNoDexPc)
       : HVecMemoryOperation(arena,
                             packed_type,
-                            SideEffects::ArrayReadOfType(packed_type),
+                            side_effects,
                             /* number_of_inputs */ 2,
                             vector_length,
                             dex_pc) {
@@ -977,11 +992,12 @@
             HInstruction* index,
             HInstruction* value,
             DataType::Type packed_type,
+            SideEffects side_effects,
             size_t vector_length,
             uint32_t dex_pc = kNoDexPc)
       : HVecMemoryOperation(arena,
                             packed_type,
-                            SideEffects::ArrayWriteOfType(packed_type),
+                            side_effects,
                             /* number_of_inputs */ 3,
                             vector_length,
                             dex_pc) {
diff --git a/compiler/optimizing/nodes_vector_test.cc b/compiler/optimizing/nodes_vector_test.cc
index 3acdb20..7dbfcda 100644
--- a/compiler/optimizing/nodes_vector_test.cc
+++ b/compiler/optimizing/nodes_vector_test.cc
@@ -47,6 +47,16 @@
                                                    0,
                                                    DataType::Type::kInt32);
     entry_block_->AddInstruction(parameter_);
+    int8_parameter_ = new (&allocator_) HParameterValue(graph_->GetDexFile(),
+                                                        dex::TypeIndex(1),
+                                                        0,
+                                                        DataType::Type::kInt8);
+    entry_block_->AddInstruction(int8_parameter_);
+    int16_parameter_ = new (&allocator_) HParameterValue(graph_->GetDexFile(),
+                                                         dex::TypeIndex(2),
+                                                         0,
+                                                         DataType::Type::kInt16);
+    entry_block_->AddInstruction(int16_parameter_);
   }
 
   // General building fields.
@@ -58,6 +68,8 @@
   HBasicBlock* exit_block_;
 
   HInstruction* parameter_;
+  HInstruction* int8_parameter_;
+  HInstruction* int16_parameter_;
 };
 
 //
@@ -126,8 +138,14 @@
       HVecReplicateScalar(&allocator_, parameter_, DataType::Type::kInt32, 2);
   HVecOperation* v3 = new (&allocator_)
       HVecReplicateScalar(&allocator_, parameter_, DataType::Type::kInt16, 4);
-  HVecOperation* v4 = new (&allocator_)
-      HVecStore(&allocator_, parameter_, parameter_, v0, DataType::Type::kInt32, 4);
+  HVecOperation* v4 = new (&allocator_) HVecStore(
+      &allocator_,
+      parameter_,
+      parameter_,
+      v0,
+      DataType::Type::kInt32,
+      SideEffects::ArrayWriteOfType(DataType::Type::kInt32),
+      4);
 
   EXPECT_TRUE(v0->Equals(v0));
   EXPECT_TRUE(v1->Equals(v1));
@@ -175,12 +193,27 @@
 }
 
 TEST_F(NodesVectorTest, VectorAlignmentAndStringCharAtMatterOnLoad) {
-  HVecLoad* v0 = new (&allocator_) HVecLoad(
-      &allocator_, parameter_, parameter_, DataType::Type::kInt32, 4, /*is_string_char_at*/ false);
-  HVecLoad* v1 = new (&allocator_) HVecLoad(
-      &allocator_, parameter_, parameter_, DataType::Type::kInt32, 4, /*is_string_char_at*/ false);
-  HVecLoad* v2 = new (&allocator_) HVecLoad(
-      &allocator_, parameter_, parameter_, DataType::Type::kInt32, 4, /*is_string_char_at*/ true);
+  HVecLoad* v0 = new (&allocator_) HVecLoad(&allocator_,
+                                            parameter_,
+                                            parameter_,
+                                            DataType::Type::kInt32,
+                                            SideEffects::ArrayReadOfType(DataType::Type::kInt32),
+                                            4,
+                                            /*is_string_char_at*/ false);
+  HVecLoad* v1 = new (&allocator_) HVecLoad(&allocator_,
+                                            parameter_,
+                                            parameter_,
+                                            DataType::Type::kInt32,
+                                            SideEffects::ArrayReadOfType(DataType::Type::kInt32),
+                                            4,
+                                            /*is_string_char_at*/ false);
+  HVecLoad* v2 = new (&allocator_) HVecLoad(&allocator_,
+                                            parameter_,
+                                            parameter_,
+                                            DataType::Type::kInt32,
+                                            SideEffects::ArrayReadOfType(DataType::Type::kInt32),
+                                            4,
+                                            /*is_string_char_at*/ true);
 
   EXPECT_TRUE(v0->CanBeMoved());
   EXPECT_TRUE(v1->CanBeMoved());
@@ -209,99 +242,155 @@
 }
 
 TEST_F(NodesVectorTest, VectorSignMattersOnMin) {
-  HVecOperation* v0 = new (&allocator_)
+  HVecOperation* p0 = new (&allocator_)
       HVecReplicateScalar(&allocator_, parameter_, DataType::Type::kInt32, 4);
+  HVecOperation* p1 = new (&allocator_)
+      HVecReplicateScalar(&allocator_, int8_parameter_, DataType::Type::kInt8, 4);
+  HVecOperation* p2 = new (&allocator_)
+      HVecReplicateScalar(&allocator_, int16_parameter_, DataType::Type::kInt16, 4);
 
-  HVecMin* v1 = new (&allocator_)
-      HVecMin(&allocator_, v0, v0, DataType::Type::kInt32, 4, /*is_unsigned*/ true);
-  HVecMin* v2 = new (&allocator_)
-      HVecMin(&allocator_, v0, v0, DataType::Type::kInt32, 4, /*is_unsigned*/ false);
-  HVecMin* v3 = new (&allocator_)
-      HVecMin(&allocator_, v0, v0, DataType::Type::kInt32, 2, /*is_unsigned*/ true);
+  HVecMin* v0 = new (&allocator_) HVecMin(
+      &allocator_, p0, p0, DataType::Type::kInt32, 4, /*is_unsigned*/ true);
+  HVecMin* v1 = new (&allocator_) HVecMin(
+      &allocator_, p0, p0, DataType::Type::kInt32, 4, /*is_unsigned*/ false);
+  HVecMin* v2 = new (&allocator_) HVecMin(
+      &allocator_, p0, p0, DataType::Type::kInt32, 2, /*is_unsigned*/ true);
+  HVecMin* v3 = new (&allocator_) HVecMin(&allocator_, p1, p1, DataType::Type::kUint8, 16);
+  HVecMin* v4 = new (&allocator_) HVecMin(&allocator_, p1, p1, DataType::Type::kInt8, 16);
+  HVecMin* v5 = new (&allocator_) HVecMin(&allocator_, p2, p2, DataType::Type::kUint16, 8);
+  HVecMin* v6 = new (&allocator_) HVecMin(&allocator_, p2, p2, DataType::Type::kInt16, 8);
+  HVecMin* min_insns[] = { v0, v1, v2, v3, v4, v5, v6 };
 
-  EXPECT_FALSE(v0->CanBeMoved());
-  EXPECT_TRUE(v1->CanBeMoved());
-  EXPECT_TRUE(v2->CanBeMoved());
-  EXPECT_TRUE(v3->CanBeMoved());
+  EXPECT_FALSE(p0->CanBeMoved());
+  EXPECT_FALSE(p1->CanBeMoved());
+  EXPECT_FALSE(p2->CanBeMoved());
 
-  EXPECT_TRUE(v1->IsUnsigned());
-  EXPECT_FALSE(v2->IsUnsigned());
-  EXPECT_TRUE(v3->IsUnsigned());
+  for (HVecMin* min_insn : min_insns) {
+    EXPECT_TRUE(min_insn->CanBeMoved());
+  }
 
-  EXPECT_TRUE(v1->Equals(v1));
-  EXPECT_TRUE(v2->Equals(v2));
-  EXPECT_TRUE(v3->Equals(v3));
+  // Deprecated; IsUnsigned() should be removed with the introduction of Uint32 and Uint64.
+  EXPECT_TRUE(v0->IsUnsigned());
+  EXPECT_FALSE(v1->IsUnsigned());
+  EXPECT_TRUE(v2->IsUnsigned());
 
-  EXPECT_FALSE(v1->Equals(v2));  // different signs
-  EXPECT_FALSE(v1->Equals(v3));  // different vector lengths
+  for (HVecMin* min_insn1 : min_insns) {
+    for (HVecMin* min_insn2 : min_insns) {
+      EXPECT_EQ(min_insn1 == min_insn2, min_insn1->Equals(min_insn2));
+    }
+  }
 }
 
 TEST_F(NodesVectorTest, VectorSignMattersOnMax) {
-  HVecOperation* v0 = new (&allocator_)
+  HVecOperation* p0 = new (&allocator_)
       HVecReplicateScalar(&allocator_, parameter_, DataType::Type::kInt32, 4);
+  HVecOperation* p1 = new (&allocator_)
+      HVecReplicateScalar(&allocator_, int8_parameter_, DataType::Type::kInt8, 4);
+  HVecOperation* p2 = new (&allocator_)
+      HVecReplicateScalar(&allocator_, int16_parameter_, DataType::Type::kInt16, 4);
 
-  HVecMax* v1 = new (&allocator_)
-      HVecMax(&allocator_, v0, v0, DataType::Type::kInt32, 4, /*is_unsigned*/ true);
-  HVecMax* v2 = new (&allocator_)
-      HVecMax(&allocator_, v0, v0, DataType::Type::kInt32, 4, /*is_unsigned*/ false);
-  HVecMax* v3 = new (&allocator_)
-      HVecMax(&allocator_, v0, v0, DataType::Type::kInt32, 2, /*is_unsigned*/ true);
+  HVecMax* v0 = new (&allocator_) HVecMax(
+      &allocator_, p0, p0, DataType::Type::kInt32, 4, /*is_unsigned*/ true);
+  HVecMax* v1 = new (&allocator_) HVecMax(
+      &allocator_, p0, p0, DataType::Type::kInt32, 4, /*is_unsigned*/ false);
+  HVecMax* v2 = new (&allocator_) HVecMax(
+      &allocator_, p0, p0, DataType::Type::kInt32, 2, /*is_unsigned*/ true);
+  HVecMax* v3 = new (&allocator_) HVecMax(&allocator_, p1, p1, DataType::Type::kUint8, 16);
+  HVecMax* v4 = new (&allocator_) HVecMax(&allocator_, p1, p1, DataType::Type::kInt8, 16);
+  HVecMax* v5 = new (&allocator_) HVecMax(&allocator_, p2, p2, DataType::Type::kUint16, 8);
+  HVecMax* v6 = new (&allocator_) HVecMax(&allocator_, p2, p2, DataType::Type::kInt16, 8);
+  HVecMax* max_insns[] = { v0, v1, v2, v3, v4, v5, v6 };
 
-  EXPECT_FALSE(v0->CanBeMoved());
-  EXPECT_TRUE(v1->CanBeMoved());
-  EXPECT_TRUE(v2->CanBeMoved());
-  EXPECT_TRUE(v3->CanBeMoved());
+  EXPECT_FALSE(p0->CanBeMoved());
+  EXPECT_FALSE(p1->CanBeMoved());
+  EXPECT_FALSE(p2->CanBeMoved());
 
-  EXPECT_TRUE(v1->IsUnsigned());
-  EXPECT_FALSE(v2->IsUnsigned());
-  EXPECT_TRUE(v3->IsUnsigned());
+  for (HVecMax* max_insn : max_insns) {
+    EXPECT_TRUE(max_insn->CanBeMoved());
+  }
 
-  EXPECT_TRUE(v1->Equals(v1));
-  EXPECT_TRUE(v2->Equals(v2));
-  EXPECT_TRUE(v3->Equals(v3));
+  // Deprecated; IsUnsigned() should be removed with the introduction of Uint32 and Uint64.
+  EXPECT_TRUE(v0->IsUnsigned());
+  EXPECT_FALSE(v1->IsUnsigned());
+  EXPECT_TRUE(v2->IsUnsigned());
 
-  EXPECT_FALSE(v1->Equals(v2));  // different signs
-  EXPECT_FALSE(v1->Equals(v3));  // different vector lengths
+  for (HVecMax* max_insn1 : max_insns) {
+    for (HVecMax* max_insn2 : max_insns) {
+      EXPECT_EQ(max_insn1 == max_insn2, max_insn1->Equals(max_insn2));
+    }
+  }
 }
 
 TEST_F(NodesVectorTest, VectorAttributesMatterOnHalvingAdd) {
-  HVecOperation* v0 = new (&allocator_)
+  HVecOperation* p0 = new (&allocator_)
       HVecReplicateScalar(&allocator_, parameter_, DataType::Type::kInt32, 4);
+  HVecOperation* p1 = new (&allocator_)
+      HVecReplicateScalar(&allocator_, int8_parameter_, DataType::Type::kInt8, 4);
+  HVecOperation* p2 = new (&allocator_)
+      HVecReplicateScalar(&allocator_, int16_parameter_, DataType::Type::kInt16, 4);
 
+  HVecHalvingAdd* v0 = new (&allocator_) HVecHalvingAdd(
+      &allocator_, p0, p0, DataType::Type::kInt32, 4, /*is_rounded*/ true, /*is_unsigned*/ true);
   HVecHalvingAdd* v1 = new (&allocator_) HVecHalvingAdd(
-      &allocator_, v0, v0, DataType::Type::kInt32, 4, /*is_unsigned*/ true, /*is_rounded*/ true);
+      &allocator_, p0, p0, DataType::Type::kInt32, 4, /*is_rounded*/ false, /*is_unsigned*/ true);
   HVecHalvingAdd* v2 = new (&allocator_) HVecHalvingAdd(
-      &allocator_, v0, v0, DataType::Type::kInt32, 4, /*is_unsigned*/ true, /*is_rounded*/ false);
+      &allocator_, p0, p0, DataType::Type::kInt32, 4, /*is_rounded*/ true, /*is_unsigned*/ false);
   HVecHalvingAdd* v3 = new (&allocator_) HVecHalvingAdd(
-      &allocator_, v0, v0, DataType::Type::kInt32, 4, /*is_unsigned*/ false, /*is_rounded*/ true);
+      &allocator_, p0, p0, DataType::Type::kInt32, 4, /*is_rounded*/ false, /*is_unsigned*/ false);
   HVecHalvingAdd* v4 = new (&allocator_) HVecHalvingAdd(
-      &allocator_, v0, v0, DataType::Type::kInt32, 4, /*is_unsigned*/ false, /*is_rounded*/ false);
+      &allocator_, p0, p0, DataType::Type::kInt32, 2, /*is_rounded*/ true, /*is_unsigned*/ true);
   HVecHalvingAdd* v5 = new (&allocator_) HVecHalvingAdd(
-      &allocator_, v0, v0, DataType::Type::kInt32, 2, /*is_unsigned*/ true, /*is_rounded*/ true);
+      &allocator_, p1, p1, DataType::Type::kUint8, 16, /*is_rounded*/ true);
+  HVecHalvingAdd* v6 = new (&allocator_) HVecHalvingAdd(
+      &allocator_, p1, p1, DataType::Type::kUint8, 16, /*is_rounded*/ false);
+  HVecHalvingAdd* v7 = new (&allocator_) HVecHalvingAdd(
+      &allocator_, p1, p1, DataType::Type::kInt8, 16, /*is_rounded*/ true);
+  HVecHalvingAdd* v8 = new (&allocator_) HVecHalvingAdd(
+      &allocator_, p1, p1, DataType::Type::kInt8, 16, /*is_rounded*/ false);
+  HVecHalvingAdd* v9 = new (&allocator_) HVecHalvingAdd(
+      &allocator_, p2, p2, DataType::Type::kUint16, 8, /*is_rounded*/ true);
+  HVecHalvingAdd* v10 = new (&allocator_) HVecHalvingAdd(
+      &allocator_, p2, p2, DataType::Type::kUint16, 8, /*is_rounded*/ false);
+  HVecHalvingAdd* v11 = new (&allocator_) HVecHalvingAdd(
+      &allocator_, p2, p2, DataType::Type::kInt16, 2, /*is_rounded*/ true);
+  HVecHalvingAdd* v12 = new (&allocator_) HVecHalvingAdd(
+      &allocator_, p2, p2, DataType::Type::kInt16, 2, /*is_rounded*/ false);
+  HVecHalvingAdd* hadd_insns[] = { v0, v1, v2, v3, v4, v5, v6, v7, v8, v9, v10, v11, v12 };
 
-  EXPECT_FALSE(v0->CanBeMoved());
-  EXPECT_TRUE(v1->CanBeMoved());
-  EXPECT_TRUE(v2->CanBeMoved());
-  EXPECT_TRUE(v3->CanBeMoved());
-  EXPECT_TRUE(v4->CanBeMoved());
-  EXPECT_TRUE(v5->CanBeMoved());
+  EXPECT_FALSE(p0->CanBeMoved());
+  EXPECT_FALSE(p1->CanBeMoved());
+  EXPECT_FALSE(p2->CanBeMoved());
 
-  EXPECT_TRUE(v1->Equals(v1));
-  EXPECT_TRUE(v2->Equals(v2));
-  EXPECT_TRUE(v3->Equals(v3));
-  EXPECT_TRUE(v4->Equals(v4));
-  EXPECT_TRUE(v5->Equals(v5));
+  for (HVecHalvingAdd* hadd_insn : hadd_insns) {
+    EXPECT_TRUE(hadd_insn->CanBeMoved());
+  }
 
-  EXPECT_TRUE(v1->IsUnsigned() && v1->IsRounded());
-  EXPECT_TRUE(v2->IsUnsigned() && !v2->IsRounded());
-  EXPECT_TRUE(!v3->IsUnsigned() && v3->IsRounded());
-  EXPECT_TRUE(!v4->IsUnsigned() && !v4->IsRounded());
-  EXPECT_TRUE(v5->IsUnsigned() && v5->IsRounded());
+  // Deprecated; IsUnsigned() should be removed with the introduction of Uint32 and Uint64.
+  EXPECT_TRUE(v0->IsUnsigned());
+  EXPECT_TRUE(v1->IsUnsigned());
+  EXPECT_TRUE(!v2->IsUnsigned());
+  EXPECT_TRUE(!v3->IsUnsigned());
+  EXPECT_TRUE(v4->IsUnsigned());
 
-  EXPECT_FALSE(v1->Equals(v2));  // different attributes
-  EXPECT_FALSE(v1->Equals(v3));  // different attributes
-  EXPECT_FALSE(v1->Equals(v4));  // different attributes
-  EXPECT_FALSE(v1->Equals(v5));  // different vector lengths
+  EXPECT_TRUE(v0->IsRounded());
+  EXPECT_TRUE(!v1->IsRounded());
+  EXPECT_TRUE(v2->IsRounded());
+  EXPECT_TRUE(!v3->IsRounded());
+  EXPECT_TRUE(v4->IsRounded());
+  EXPECT_TRUE(v5->IsRounded());
+  EXPECT_TRUE(!v6->IsRounded());
+  EXPECT_TRUE(v7->IsRounded());
+  EXPECT_TRUE(!v8->IsRounded());
+  EXPECT_TRUE(v9->IsRounded());
+  EXPECT_TRUE(!v10->IsRounded());
+  EXPECT_TRUE(v11->IsRounded());
+  EXPECT_TRUE(!v12->IsRounded());
+
+  for (HVecHalvingAdd* hadd_insn1 : hadd_insns) {
+    for (HVecHalvingAdd* hadd_insn2 : hadd_insns) {
+      EXPECT_EQ(hadd_insn1 == hadd_insn2, hadd_insn1->Equals(hadd_insn2));
+    }
+  }
 }
 
 TEST_F(NodesVectorTest, VectorOperationMattersOnMultiplyAccumulate) {
diff --git a/compiler/optimizing/register_allocation_resolver.cc b/compiler/optimizing/register_allocation_resolver.cc
index f0057c3..1786aa7 100644
--- a/compiler/optimizing/register_allocation_resolver.cc
+++ b/compiler/optimizing/register_allocation_resolver.cc
@@ -112,6 +112,7 @@
         case DataType::Type::kReference:
         case DataType::Type::kInt32:
         case DataType::Type::kUint16:
+        case DataType::Type::kUint8:
         case DataType::Type::kInt8:
         case DataType::Type::kBool:
         case DataType::Type::kInt16:
diff --git a/compiler/optimizing/register_allocator_graph_color.cc b/compiler/optimizing/register_allocator_graph_color.cc
index 4ff7315..33df607 100644
--- a/compiler/optimizing/register_allocator_graph_color.cc
+++ b/compiler/optimizing/register_allocator_graph_color.cc
@@ -1940,6 +1940,7 @@
         case DataType::Type::kReference:
         case DataType::Type::kInt32:
         case DataType::Type::kUint16:
+        case DataType::Type::kUint8:
         case DataType::Type::kInt8:
         case DataType::Type::kBool:
         case DataType::Type::kInt16:
diff --git a/compiler/optimizing/register_allocator_linear_scan.cc b/compiler/optimizing/register_allocator_linear_scan.cc
index 2012cd5..9803a7b 100644
--- a/compiler/optimizing/register_allocator_linear_scan.cc
+++ b/compiler/optimizing/register_allocator_linear_scan.cc
@@ -1116,6 +1116,7 @@
     case DataType::Type::kReference:
     case DataType::Type::kInt32:
     case DataType::Type::kUint16:
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
     case DataType::Type::kBool:
     case DataType::Type::kInt16:
diff --git a/compiler/optimizing/scheduler_arm.cc b/compiler/optimizing/scheduler_arm.cc
index 110db47..b3c8f10 100644
--- a/compiler/optimizing/scheduler_arm.cc
+++ b/compiler/optimizing/scheduler_arm.cc
@@ -514,9 +514,10 @@
   DataType::Type type = instr->InputAt(0)->GetType();
   switch (type) {
     case DataType::Type::kBool:
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
-    case DataType::Type::kInt16:
     case DataType::Type::kUint16:
+    case DataType::Type::kInt16:
     case DataType::Type::kInt32:
       last_visited_internal_latency_ = 2 * kArmIntegerOpLatency;
       break;
@@ -633,9 +634,10 @@
 
   switch (type) {
     case DataType::Type::kBool:
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
-    case DataType::Type::kInt16:
     case DataType::Type::kUint16:
+    case DataType::Type::kInt16:
     case DataType::Type::kInt32: {
       if (maybe_compressed_char_at) {
         last_visited_internal_latency_ += kArmMemoryLoadLatency;
@@ -733,9 +735,10 @@
 
   switch (value_type) {
     case DataType::Type::kBool:
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
-    case DataType::Type::kInt16:
     case DataType::Type::kUint16:
+    case DataType::Type::kInt16:
     case DataType::Type::kInt32: {
       if (index->IsConstant()) {
         last_visited_latency_ = kArmMemoryStoreLatency;
@@ -916,9 +919,10 @@
 
   switch (field_type) {
     case DataType::Type::kBool:
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
-    case DataType::Type::kInt16:
     case DataType::Type::kUint16:
+    case DataType::Type::kInt16:
     case DataType::Type::kInt32:
       last_visited_latency_ = kArmMemoryLoadLatency;
       break;
@@ -977,9 +981,10 @@
 
   switch (field_type) {
     case DataType::Type::kBool:
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
-    case DataType::Type::kInt16:
     case DataType::Type::kUint16:
+    case DataType::Type::kInt16:
       if (is_volatile) {
         last_visited_internal_latency_ = kArmMemoryBarrierLatency + kArmMemoryStoreLatency;
         last_visited_latency_ = kArmMemoryBarrierLatency;
@@ -1047,6 +1052,7 @@
   DataType::Type input_type = instr->GetInputType();
 
   switch (result_type) {
+    case DataType::Type::kUint8:
     case DataType::Type::kInt8:
     case DataType::Type::kUint16:
     case DataType::Type::kInt16:
@@ -1072,6 +1078,7 @@
     case DataType::Type::kInt64:
       switch (input_type) {
         case DataType::Type::kBool:
+        case DataType::Type::kUint8:
         case DataType::Type::kInt8:
         case DataType::Type::kUint16:
         case DataType::Type::kInt16:
@@ -1095,6 +1102,7 @@
     case DataType::Type::kFloat32:
       switch (input_type) {
         case DataType::Type::kBool:
+        case DataType::Type::kUint8:
         case DataType::Type::kInt8:
         case DataType::Type::kUint16:
         case DataType::Type::kInt16:
@@ -1118,6 +1126,7 @@
     case DataType::Type::kFloat64:
       switch (input_type) {
         case DataType::Type::kBool:
+        case DataType::Type::kUint8:
         case DataType::Type::kInt8:
         case DataType::Type::kUint16:
         case DataType::Type::kInt16:
diff --git a/compiler/optimizing/side_effects_test.cc b/compiler/optimizing/side_effects_test.cc
index ac5eb15..9731712 100644
--- a/compiler/optimizing/side_effects_test.cc
+++ b/compiler/optimizing/side_effects_test.cc
@@ -21,6 +21,19 @@
 
 namespace art {
 
+// Only runtime types other than void are allowed.
+static const DataType::Type kTestTypes[] = {
+    DataType::Type::kReference,
+    DataType::Type::kBool,
+    DataType::Type::kInt8,
+    DataType::Type::kUint16,
+    DataType::Type::kInt16,
+    DataType::Type::kInt32,
+    DataType::Type::kInt64,
+    DataType::Type::kFloat32,
+    DataType::Type::kFloat64,
+};
+
 /**
  * Tests for the SideEffects class.
  */
@@ -91,9 +104,7 @@
 
 TEST(SideEffectsTest, DependencesAndNoDependences) {
   // Apply test to each individual data type.
-  for (DataType::Type type = DataType::Type::kReference;
-       type < DataType::Type::kVoid;
-       type = static_cast<DataType::Type>(static_cast<uint8_t>(type) + 1u)) {
+  for (DataType::Type type : kTestTypes) {
     // Same data type and access type: proper write/read dep.
     testWriteAndReadDependence(
         SideEffects::FieldWriteOfType(type, false),
@@ -169,9 +180,7 @@
 TEST(SideEffectsTest, AllWritesAndReads) {
   SideEffects s = SideEffects::None();
   // Keep taking the union of different writes and reads.
-  for (DataType::Type type = DataType::Type::kReference;
-       type < DataType::Type::kVoid;
-       type = static_cast<DataType::Type>(static_cast<uint8_t>(type) + 1u)) {
+  for (DataType::Type type : kTestTypes) {
     s = s.Union(SideEffects::FieldWriteOfType(type, /* is_volatile */ false));
     s = s.Union(SideEffects::ArrayWriteOfType(type));
     s = s.Union(SideEffects::FieldReadOfType(type, /* is_volatile */ false));
diff --git a/compiler/optimizing/ssa_builder.cc b/compiler/optimizing/ssa_builder.cc
index 77b7a22..2356316 100644
--- a/compiler/optimizing/ssa_builder.cc
+++ b/compiler/optimizing/ssa_builder.cc
@@ -393,7 +393,7 @@
         }
         // Refine the side effects of this floating point aset. Note that we do this even if
         // no replacement occurs, since the right-hand-side may have been corrected already.
-        aset->ComputeSideEffects();
+        aset->SetSideEffects(HArraySet::ComputeSideEffects(aset->GetComponentType()));
       } else {
         // Array elements are integral and the value assigned to it initially
         // was integral too. Nothing to do.
diff --git a/test/646-checker-hadd-alt-byte/src/Main.java b/test/646-checker-hadd-alt-byte/src/Main.java
index 7be3151..87f7688 100644
--- a/test/646-checker-hadd-alt-byte/src/Main.java
+++ b/test/646-checker-hadd-alt-byte/src/Main.java
@@ -43,21 +43,21 @@
   /// CHECK-DAG: <<Phi:i\d+>>  Phi                                  loop:<<Loop:B\d+>> outer_loop:none
   /// CHECK-DAG: <<Get1:d\d+>> VecLoad                              loop:<<Loop>>      outer_loop:none
   /// CHECK-DAG: <<Get2:d\d+>> VecLoad                              loop:<<Loop>>      outer_loop:none
-  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] unsigned:false rounded:false loop:<<Loop>> outer_loop:none
+  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] packed_type:Int8 rounded:false loop:<<Loop>> outer_loop:none
   /// CHECK-DAG:               VecStore [{{l\d+}},<<Phi>>,<<HAdd>>] loop:<<Loop>>      outer_loop:none
   //
   /// CHECK-START-ARM64: void Main.halving_add_signed(byte[], byte[], byte[]) loop_optimization (after)
   /// CHECK-DAG: <<Phi:i\d+>>  Phi                                  loop:<<Loop:B\d+>> outer_loop:none
   /// CHECK-DAG: <<Get1:d\d+>> VecLoad                              loop:<<Loop>>      outer_loop:none
   /// CHECK-DAG: <<Get2:d\d+>> VecLoad                              loop:<<Loop>>      outer_loop:none
-  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] unsigned:false rounded:false loop:<<Loop>> outer_loop:none
+  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] packed_type:Int8 rounded:false loop:<<Loop>> outer_loop:none
   /// CHECK-DAG:               VecStore [{{l\d+}},<<Phi>>,<<HAdd>>] loop:<<Loop>>      outer_loop:none
   //
   /// CHECK-START-MIPS64: void Main.halving_add_signed(byte[], byte[], byte[]) loop_optimization (after)
   /// CHECK-DAG: <<Phi:i\d+>>  Phi                                  loop:<<Loop:B\d+>> outer_loop:none
   /// CHECK-DAG: <<Get1:d\d+>> VecLoad                              loop:<<Loop>>      outer_loop:none
   /// CHECK-DAG: <<Get2:d\d+>> VecLoad                              loop:<<Loop>>      outer_loop:none
-  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] unsigned:false rounded:false loop:<<Loop>> outer_loop:none
+  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] packed_type:Int8 rounded:false loop:<<Loop>> outer_loop:none
   /// CHECK-DAG:               VecStore [{{l\d+}},<<Phi>>,<<HAdd>>] loop:<<Loop>>      outer_loop:none
   private static void halving_add_signed(byte[] b1, byte[] b2, byte[] bo) {
     int min_length = Math.min(bo.length, Math.min(b1.length, b2.length));
@@ -83,21 +83,21 @@
   /// CHECK-DAG: <<Phi:i\d+>>  Phi                                  loop:<<Loop:B\d+>> outer_loop:none
   /// CHECK-DAG: <<Get1:d\d+>> VecLoad                              loop:<<Loop>>      outer_loop:none
   /// CHECK-DAG: <<Get2:d\d+>> VecLoad                              loop:<<Loop>>      outer_loop:none
-  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] unsigned:true rounded:false loop:<<Loop>> outer_loop:none
+  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] packed_type:Uint8 rounded:false loop:<<Loop>> outer_loop:none
   /// CHECK-DAG:               VecStore [{{l\d+}},<<Phi>>,<<HAdd>>] loop:<<Loop>>      outer_loop:none
   //
   /// CHECK-START-ARM64: void Main.halving_add_unsigned(byte[], byte[], byte[]) loop_optimization (after)
   /// CHECK-DAG: <<Phi:i\d+>>  Phi                                  loop:<<Loop:B\d+>> outer_loop:none
   /// CHECK-DAG: <<Get1:d\d+>> VecLoad                              loop:<<Loop>>      outer_loop:none
   /// CHECK-DAG: <<Get2:d\d+>> VecLoad                              loop:<<Loop>>      outer_loop:none
-  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] unsigned:true rounded:false loop:<<Loop>> outer_loop:none
+  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] packed_type:Uint8 rounded:false loop:<<Loop>> outer_loop:none
   /// CHECK-DAG:               VecStore [{{l\d+}},<<Phi>>,<<HAdd>>] loop:<<Loop>>      outer_loop:none
   //
   /// CHECK-START-MIPS64: void Main.halving_add_unsigned(byte[], byte[], byte[]) loop_optimization (after)
   /// CHECK-DAG: <<Phi:i\d+>>  Phi                                  loop:<<Loop:B\d+>> outer_loop:none
   /// CHECK-DAG: <<Get1:d\d+>> VecLoad                              loop:<<Loop>>      outer_loop:none
   /// CHECK-DAG: <<Get2:d\d+>> VecLoad                              loop:<<Loop>>      outer_loop:none
-  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] unsigned:true rounded:false loop:<<Loop>> outer_loop:none
+  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] packed_type:Uint8 rounded:false loop:<<Loop>> outer_loop:none
   /// CHECK-DAG:               VecStore [{{l\d+}},<<Phi>>,<<HAdd>>] loop:<<Loop>>      outer_loop:none
   private static void halving_add_unsigned(byte[] b1, byte[] b2, byte[] bo) {
     int min_length = Math.min(bo.length, Math.min(b1.length, b2.length));
@@ -121,21 +121,21 @@
   /// CHECK-DAG: <<Phi:i\d+>>  Phi                                  loop:<<Loop:B\d+>> outer_loop:none
   /// CHECK-DAG: <<Get1:d\d+>> VecLoad                              loop:<<Loop>>      outer_loop:none
   /// CHECK-DAG: <<Get2:d\d+>> VecLoad                              loop:<<Loop>>      outer_loop:none
-  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] unsigned:false rounded:true loop:<<Loop>> outer_loop:none
+  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] packed_type:Int8 rounded:true loop:<<Loop>> outer_loop:none
   /// CHECK-DAG:               VecStore [{{l\d+}},<<Phi>>,<<HAdd>>] loop:<<Loop>>      outer_loop:none
   //
   /// CHECK-START-ARM64: void Main.rounding_halving_add_signed(byte[], byte[], byte[]) loop_optimization (after)
   /// CHECK-DAG: <<Phi:i\d+>>  Phi                                  loop:<<Loop:B\d+>> outer_loop:none
   /// CHECK-DAG: <<Get1:d\d+>> VecLoad                              loop:<<Loop>>      outer_loop:none
   /// CHECK-DAG: <<Get2:d\d+>> VecLoad                              loop:<<Loop>>      outer_loop:none
-  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] unsigned:false rounded:true loop:<<Loop>> outer_loop:none
+  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] packed_type:Int8 rounded:true loop:<<Loop>> outer_loop:none
   /// CHECK-DAG:               VecStore [{{l\d+}},<<Phi>>,<<HAdd>>] loop:<<Loop>>      outer_loop:none
   //
   /// CHECK-START-MIPS64: void Main.rounding_halving_add_signed(byte[], byte[], byte[]) loop_optimization (after)
   /// CHECK-DAG: <<Phi:i\d+>>  Phi                                  loop:<<Loop:B\d+>> outer_loop:none
   /// CHECK-DAG: <<Get1:d\d+>> VecLoad                              loop:<<Loop>>      outer_loop:none
   /// CHECK-DAG: <<Get2:d\d+>> VecLoad                              loop:<<Loop>>      outer_loop:none
-  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] unsigned:false rounded:true loop:<<Loop>> outer_loop:none
+  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] packed_type:Int8 rounded:true loop:<<Loop>> outer_loop:none
   /// CHECK-DAG:               VecStore [{{l\d+}},<<Phi>>,<<HAdd>>] loop:<<Loop>>      outer_loop:none
   private static void rounding_halving_add_signed(byte[] b1, byte[] b2, byte[] bo) {
     int min_length = Math.min(bo.length, Math.min(b1.length, b2.length));
@@ -162,21 +162,21 @@
   /// CHECK-DAG: <<Phi:i\d+>>  Phi                                  loop:<<Loop:B\d+>> outer_loop:none
   /// CHECK-DAG: <<Get1:d\d+>> VecLoad                              loop:<<Loop>>      outer_loop:none
   /// CHECK-DAG: <<Get2:d\d+>> VecLoad                              loop:<<Loop>>      outer_loop:none
-  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>]  unsigned:true rounded:true loop:<<Loop>> outer_loop:none
+  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>]  packed_type:Uint8 rounded:true loop:<<Loop>> outer_loop:none
   /// CHECK-DAG:               VecStore [{{l\d+}},<<Phi>>,<<HAdd>>] loop:<<Loop>>      outer_loop:none
   //
   /// CHECK-START-ARM64: void Main.rounding_halving_add_unsigned(byte[], byte[], byte[]) loop_optimization (after)
   /// CHECK-DAG: <<Phi:i\d+>>  Phi                                  loop:<<Loop:B\d+>> outer_loop:none
   /// CHECK-DAG: <<Get1:d\d+>> VecLoad                              loop:<<Loop>>      outer_loop:none
   /// CHECK-DAG: <<Get2:d\d+>> VecLoad                              loop:<<Loop>>      outer_loop:none
-  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>]  unsigned:true rounded:true loop:<<Loop>> outer_loop:none
+  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>]  packed_type:Uint8 rounded:true loop:<<Loop>> outer_loop:none
   /// CHECK-DAG:               VecStore [{{l\d+}},<<Phi>>,<<HAdd>>] loop:<<Loop>>      outer_loop:none
   //
   /// CHECK-START-MIPS64: void Main.rounding_halving_add_unsigned(byte[], byte[], byte[]) loop_optimization (after)
   /// CHECK-DAG: <<Phi:i\d+>>  Phi                                  loop:<<Loop:B\d+>> outer_loop:none
   /// CHECK-DAG: <<Get1:d\d+>> VecLoad                              loop:<<Loop>>      outer_loop:none
   /// CHECK-DAG: <<Get2:d\d+>> VecLoad                              loop:<<Loop>>      outer_loop:none
-  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>]  unsigned:true rounded:true loop:<<Loop>> outer_loop:none
+  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>]  packed_type:Uint8 rounded:true loop:<<Loop>> outer_loop:none
   /// CHECK-DAG:               VecStore [{{l\d+}},<<Phi>>,<<HAdd>>] loop:<<Loop>>      outer_loop:none
   private static void rounding_halving_add_unsigned(byte[] b1, byte[] b2, byte[] bo) {
     int min_length = Math.min(bo.length, Math.min(b1.length, b2.length));
@@ -200,7 +200,7 @@
   /// CHECK-DAG: <<Repl:d\d+>> VecReplicateScalar [<<I127>>]        loop:none
   /// CHECK-DAG: <<Phi:i\d+>>  Phi                                  loop:<<Loop:B\d+>> outer_loop:none
   /// CHECK-DAG: <<Get:d\d+>>  VecLoad                              loop:<<Loop>>      outer_loop:none
-  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get>>,<<Repl>>] unsigned:false rounded:false loop:<<Loop>> outer_loop:none
+  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get>>,<<Repl>>] packed_type:Int8 rounded:false loop:<<Loop>> outer_loop:none
   /// CHECK-DAG:               VecStore [{{l\d+}},<<Phi>>,<<HAdd>>] loop:<<Loop>>      outer_loop:none
   //
   /// CHECK-START-ARM64: void Main.halving_add_signed_constant(byte[], byte[]) loop_optimization (after)
@@ -208,7 +208,7 @@
   /// CHECK-DAG: <<Repl:d\d+>> VecReplicateScalar [<<I127>>]        loop:none
   /// CHECK-DAG: <<Phi:i\d+>>  Phi                                  loop:<<Loop:B\d+>> outer_loop:none
   /// CHECK-DAG: <<Get:d\d+>>  VecLoad                              loop:<<Loop>>      outer_loop:none
-  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get>>,<<Repl>>] unsigned:false rounded:false loop:<<Loop>> outer_loop:none
+  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get>>,<<Repl>>] packed_type:Int8 rounded:false loop:<<Loop>> outer_loop:none
   /// CHECK-DAG:               VecStore [{{l\d+}},<<Phi>>,<<HAdd>>] loop:<<Loop>>      outer_loop:none
   //
   /// CHECK-START-MIPS64: void Main.halving_add_signed_constant(byte[], byte[]) loop_optimization (after)
@@ -216,7 +216,7 @@
   /// CHECK-DAG: <<Repl:d\d+>> VecReplicateScalar [<<I127>>]        loop:none
   /// CHECK-DAG: <<Phi:i\d+>>  Phi                                  loop:<<Loop:B\d+>> outer_loop:none
   /// CHECK-DAG: <<Get:d\d+>>  VecLoad                              loop:<<Loop>>      outer_loop:none
-  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get>>,<<Repl>>] unsigned:false rounded:false loop:<<Loop>> outer_loop:none
+  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get>>,<<Repl>>] packed_type:Int8 rounded:false loop:<<Loop>> outer_loop:none
   /// CHECK-DAG:               VecStore [{{l\d+}},<<Phi>>,<<HAdd>>] loop:<<Loop>>      outer_loop:none
   private static void halving_add_signed_constant(byte[] b1, byte[] bo) {
     int min_length = Math.min(bo.length, b1.length);
@@ -241,7 +241,7 @@
   /// CHECK-DAG: <<Repl:d\d+>> VecReplicateScalar [<<I255>>]        loop:none
   /// CHECK-DAG: <<Phi:i\d+>>  Phi                                  loop:<<Loop:B\d+>> outer_loop:none
   /// CHECK-DAG: <<Get:d\d+>>  VecLoad                              loop:<<Loop>>      outer_loop:none
-  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get>>,<<Repl>>] unsigned:true rounded:false loop:<<Loop>> outer_loop:none
+  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get>>,<<Repl>>] packed_type:Uint8 rounded:false loop:<<Loop>> outer_loop:none
   /// CHECK-DAG:               VecStore [{{l\d+}},<<Phi>>,<<HAdd>>] loop:<<Loop>>      outer_loop:none
   //
   /// CHECK-START-ARM64: void Main.halving_add_unsigned_constant(byte[], byte[]) loop_optimization (after)
@@ -249,7 +249,7 @@
   /// CHECK-DAG: <<Repl:d\d+>> VecReplicateScalar [<<I255>>]        loop:none
   /// CHECK-DAG: <<Phi:i\d+>>  Phi                                  loop:<<Loop:B\d+>> outer_loop:none
   /// CHECK-DAG: <<Get:d\d+>>  VecLoad                              loop:<<Loop>>      outer_loop:none
-  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get>>,<<Repl>>] unsigned:true rounded:false loop:<<Loop>> outer_loop:none
+  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get>>,<<Repl>>] packed_type:Uint8 rounded:false loop:<<Loop>> outer_loop:none
   /// CHECK-DAG:               VecStore [{{l\d+}},<<Phi>>,<<HAdd>>] loop:<<Loop>>      outer_loop:none
   //
   /// CHECK-START-MIPS64: void Main.halving_add_unsigned_constant(byte[], byte[]) loop_optimization (after)
@@ -257,7 +257,7 @@
   /// CHECK-DAG: <<Repl:d\d+>> VecReplicateScalar [<<I255>>]        loop:none
   /// CHECK-DAG: <<Phi:i\d+>>  Phi                                  loop:<<Loop:B\d+>> outer_loop:none
   /// CHECK-DAG: <<Get:d\d+>>  VecLoad                              loop:<<Loop>>      outer_loop:none
-  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get>>,<<Repl>>] unsigned:true rounded:false loop:<<Loop>> outer_loop:none
+  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get>>,<<Repl>>] packed_type:Uint8 rounded:false loop:<<Loop>> outer_loop:none
   /// CHECK-DAG:               VecStore [{{l\d+}},<<Phi>>,<<HAdd>>] loop:<<Loop>>      outer_loop:none
   private static void halving_add_unsigned_constant(byte[] b1, byte[] bo) {
     int min_length = Math.min(bo.length, b1.length);
diff --git a/test/646-checker-hadd-alt-char/src/Main.java b/test/646-checker-hadd-alt-char/src/Main.java
index 2799ea7..292ea1b 100644
--- a/test/646-checker-hadd-alt-char/src/Main.java
+++ b/test/646-checker-hadd-alt-char/src/Main.java
@@ -43,21 +43,21 @@
   /// CHECK-DAG: <<Phi:i\d+>>  Phi                                  loop:<<Loop:B\d+>> outer_loop:none
   /// CHECK-DAG: <<Get1:d\d+>> VecLoad                              loop:<<Loop>>      outer_loop:none
   /// CHECK-DAG: <<Get2:d\d+>> VecLoad                              loop:<<Loop>>      outer_loop:none
-  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] unsigned:true rounded:false loop:<<Loop>> outer_loop:none
+  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] packed_type:Uint16 rounded:false loop:<<Loop>> outer_loop:none
   /// CHECK-DAG:               VecStore [{{l\d+}},<<Phi>>,<<HAdd>>] loop:<<Loop>>      outer_loop:none
   //
   /// CHECK-START-ARM64: void Main.halving_add_unsigned(char[], char[], char[]) loop_optimization (after)
   /// CHECK-DAG: <<Phi:i\d+>>  Phi                                  loop:<<Loop:B\d+>> outer_loop:none
   /// CHECK-DAG: <<Get1:d\d+>> VecLoad                              loop:<<Loop>>      outer_loop:none
   /// CHECK-DAG: <<Get2:d\d+>> VecLoad                              loop:<<Loop>>      outer_loop:none
-  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] unsigned:true rounded:false loop:<<Loop>> outer_loop:none
+  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] packed_type:Uint16 rounded:false loop:<<Loop>> outer_loop:none
   /// CHECK-DAG:               VecStore [{{l\d+}},<<Phi>>,<<HAdd>>] loop:<<Loop>>      outer_loop:none
   //
   /// CHECK-START-MIPS64: void Main.halving_add_unsigned(char[], char[], char[]) loop_optimization (after)
   /// CHECK-DAG: <<Phi:i\d+>>  Phi                                  loop:<<Loop:B\d+>> outer_loop:none
   /// CHECK-DAG: <<Get1:d\d+>> VecLoad                              loop:<<Loop>>      outer_loop:none
   /// CHECK-DAG: <<Get2:d\d+>> VecLoad                              loop:<<Loop>>      outer_loop:none
-  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] unsigned:true rounded:false loop:<<Loop>> outer_loop:none
+  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] packed_type:Uint16 rounded:false loop:<<Loop>> outer_loop:none
   /// CHECK-DAG:               VecStore [{{l\d+}},<<Phi>>,<<HAdd>>] loop:<<Loop>>      outer_loop:none
   private static void halving_add_unsigned(char[] b1, char[] b2, char[] bo) {
     int min_length = Math.min(bo.length, Math.min(b1.length, b2.length));
@@ -83,21 +83,21 @@
   /// CHECK-DAG: <<Phi:i\d+>>  Phi                                  loop:<<Loop:B\d+>> outer_loop:none
   /// CHECK-DAG: <<Get1:d\d+>> VecLoad                              loop:<<Loop>>      outer_loop:none
   /// CHECK-DAG: <<Get2:d\d+>> VecLoad                              loop:<<Loop>>      outer_loop:none
-  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] unsigned:true rounded:false loop:<<Loop>> outer_loop:none
+  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] packed_type:Uint16 rounded:false loop:<<Loop>> outer_loop:none
   /// CHECK-DAG:               VecStore [{{l\d+}},<<Phi>>,<<HAdd>>] loop:<<Loop>>      outer_loop:none
   //
   /// CHECK-START-ARM64: void Main.halving_add_also_unsigned(char[], char[], char[]) loop_optimization (after)
   /// CHECK-DAG: <<Phi:i\d+>>  Phi                                  loop:<<Loop:B\d+>> outer_loop:none
   /// CHECK-DAG: <<Get1:d\d+>> VecLoad                              loop:<<Loop>>      outer_loop:none
   /// CHECK-DAG: <<Get2:d\d+>> VecLoad                              loop:<<Loop>>      outer_loop:none
-  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] unsigned:true rounded:false loop:<<Loop>> outer_loop:none
+  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] packed_type:Uint16 rounded:false loop:<<Loop>> outer_loop:none
   /// CHECK-DAG:               VecStore [{{l\d+}},<<Phi>>,<<HAdd>>] loop:<<Loop>>      outer_loop:none
   //
   /// CHECK-START-MIPS64: void Main.halving_add_also_unsigned(char[], char[], char[]) loop_optimization (after)
   /// CHECK-DAG: <<Phi:i\d+>>  Phi                                  loop:<<Loop:B\d+>> outer_loop:none
   /// CHECK-DAG: <<Get1:d\d+>> VecLoad                              loop:<<Loop>>      outer_loop:none
   /// CHECK-DAG: <<Get2:d\d+>> VecLoad                              loop:<<Loop>>      outer_loop:none
-  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] unsigned:true rounded:false loop:<<Loop>> outer_loop:none
+  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] packed_type:Uint16 rounded:false loop:<<Loop>> outer_loop:none
   /// CHECK-DAG:               VecStore [{{l\d+}},<<Phi>>,<<HAdd>>] loop:<<Loop>>      outer_loop:none
   //
   // Note: HAnd has no impact (already a zero extension).
@@ -124,21 +124,21 @@
   /// CHECK-DAG: <<Phi:i\d+>>  Phi                                  loop:<<Loop:B\d+>> outer_loop:none
   /// CHECK-DAG: <<Get1:d\d+>> VecLoad                              loop:<<Loop>>      outer_loop:none
   /// CHECK-DAG: <<Get2:d\d+>> VecLoad                              loop:<<Loop>>      outer_loop:none
-  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] unsigned:true rounded:true loop:<<Loop>> outer_loop:none
+  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] packed_type:Uint16 rounded:true loop:<<Loop>> outer_loop:none
   /// CHECK-DAG:               VecStore [{{l\d+}},<<Phi>>,<<HAdd>>] loop:<<Loop>>      outer_loop:none
   //
   /// CHECK-START-ARM64: void Main.rounding_halving_add_unsigned(char[], char[], char[]) loop_optimization (after)
   /// CHECK-DAG: <<Phi:i\d+>>  Phi                                  loop:<<Loop:B\d+>> outer_loop:none
   /// CHECK-DAG: <<Get1:d\d+>> VecLoad                              loop:<<Loop>>      outer_loop:none
   /// CHECK-DAG: <<Get2:d\d+>> VecLoad                              loop:<<Loop>>      outer_loop:none
-  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] unsigned:true rounded:true loop:<<Loop>> outer_loop:none
+  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] packed_type:Uint16 rounded:true loop:<<Loop>> outer_loop:none
   /// CHECK-DAG:               VecStore [{{l\d+}},<<Phi>>,<<HAdd>>] loop:<<Loop>>      outer_loop:none
   //
   /// CHECK-START-MIPS64: void Main.rounding_halving_add_unsigned(char[], char[], char[]) loop_optimization (after)
   /// CHECK-DAG: <<Phi:i\d+>>  Phi                                  loop:<<Loop:B\d+>> outer_loop:none
   /// CHECK-DAG: <<Get1:d\d+>> VecLoad                              loop:<<Loop>>      outer_loop:none
   /// CHECK-DAG: <<Get2:d\d+>> VecLoad                              loop:<<Loop>>      outer_loop:none
-  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] unsigned:true rounded:true loop:<<Loop>> outer_loop:none
+  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] packed_type:Uint16 rounded:true loop:<<Loop>> outer_loop:none
   /// CHECK-DAG:               VecStore [{{l\d+}},<<Phi>>,<<HAdd>>] loop:<<Loop>>      outer_loop:none
   private static void rounding_halving_add_unsigned(char[] b1, char[] b2, char[] bo) {
     int min_length = Math.min(bo.length, Math.min(b1.length, b2.length));
@@ -165,21 +165,21 @@
   /// CHECK-DAG: <<Phi:i\d+>>  Phi                                  loop:<<Loop:B\d+>> outer_loop:none
   /// CHECK-DAG: <<Get1:d\d+>> VecLoad                              loop:<<Loop>>      outer_loop:none
   /// CHECK-DAG: <<Get2:d\d+>> VecLoad                              loop:<<Loop>>      outer_loop:none
-  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] unsigned:true rounded:true loop:<<Loop>> outer_loop:none
+  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] packed_type:Uint16 rounded:true loop:<<Loop>> outer_loop:none
   /// CHECK-DAG:               VecStore [{{l\d+}},<<Phi>>,<<HAdd>>] loop:<<Loop>>      outer_loop:none
   //
   /// CHECK-START-ARM64: void Main.rounding_halving_add_also_unsigned(char[], char[], char[]) loop_optimization (after)
   /// CHECK-DAG: <<Phi:i\d+>>  Phi                                  loop:<<Loop:B\d+>> outer_loop:none
   /// CHECK-DAG: <<Get1:d\d+>> VecLoad                              loop:<<Loop>>      outer_loop:none
   /// CHECK-DAG: <<Get2:d\d+>> VecLoad                              loop:<<Loop>>      outer_loop:none
-  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] unsigned:true rounded:true loop:<<Loop>> outer_loop:none
+  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] packed_type:Uint16 rounded:true loop:<<Loop>> outer_loop:none
   /// CHECK-DAG:               VecStore [{{l\d+}},<<Phi>>,<<HAdd>>] loop:<<Loop>>      outer_loop:none
   //
   /// CHECK-START-MIPS64: void Main.rounding_halving_add_also_unsigned(char[], char[], char[]) loop_optimization (after)
   /// CHECK-DAG: <<Phi:i\d+>>  Phi                                  loop:<<Loop:B\d+>> outer_loop:none
   /// CHECK-DAG: <<Get1:d\d+>> VecLoad                              loop:<<Loop>>      outer_loop:none
   /// CHECK-DAG: <<Get2:d\d+>> VecLoad                              loop:<<Loop>>      outer_loop:none
-  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] unsigned:true rounded:true loop:<<Loop>> outer_loop:none
+  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] packed_type:Uint16 rounded:true loop:<<Loop>> outer_loop:none
   /// CHECK-DAG:               VecStore [{{l\d+}},<<Phi>>,<<HAdd>>] loop:<<Loop>>      outer_loop:none
   //
   // Note: HAnd has no impact (already a zero extension).
@@ -206,7 +206,7 @@
   /// CHECK-DAG: <<Repl:d\d+>> VecReplicateScalar [<<UMAX>>]        loop:none
   /// CHECK-DAG: <<Phi:i\d+>>  Phi                                  loop:<<Loop:B\d+>> outer_loop:none
   /// CHECK-DAG: <<Get:d\d+>>  VecLoad                              loop:<<Loop>>      outer_loop:none
-  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get>>,<<Repl>>] unsigned:true rounded:false loop:<<Loop>> outer_loop:none
+  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get>>,<<Repl>>] packed_type:Uint16 rounded:false loop:<<Loop>> outer_loop:none
   /// CHECK-DAG:               VecStore [{{l\d+}},<<Phi>>,<<HAdd>>] loop:<<Loop>>      outer_loop:none
   //
   /// CHECK-START-ARM64: void Main.halving_add_unsigned_constant(char[], char[]) loop_optimization (after)
@@ -214,7 +214,7 @@
   /// CHECK-DAG: <<Repl:d\d+>> VecReplicateScalar [<<UMAX>>]        loop:none
   /// CHECK-DAG: <<Phi:i\d+>>  Phi                                  loop:<<Loop:B\d+>> outer_loop:none
   /// CHECK-DAG: <<Get:d\d+>>  VecLoad                              loop:<<Loop>>      outer_loop:none
-  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get>>,<<Repl>>] unsigned:true rounded:false loop:<<Loop>> outer_loop:none
+  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get>>,<<Repl>>] packed_type:Uint16 rounded:false loop:<<Loop>> outer_loop:none
   /// CHECK-DAG:               VecStore [{{l\d+}},<<Phi>>,<<HAdd>>] loop:<<Loop>>      outer_loop:none
   //
   /// CHECK-START-MIPS64: void Main.halving_add_unsigned_constant(char[], char[]) loop_optimization (after)
@@ -222,7 +222,7 @@
   /// CHECK-DAG: <<Repl:d\d+>> VecReplicateScalar [<<UMAX>>]        loop:none
   /// CHECK-DAG: <<Phi:i\d+>>  Phi                                  loop:<<Loop:B\d+>> outer_loop:none
   /// CHECK-DAG: <<Get:d\d+>>  VecLoad                              loop:<<Loop>>      outer_loop:none
-  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get>>,<<Repl>>] unsigned:true rounded:false loop:<<Loop>> outer_loop:none
+  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get>>,<<Repl>>] packed_type:Uint16 rounded:false loop:<<Loop>> outer_loop:none
   /// CHECK-DAG:               VecStore [{{l\d+}},<<Phi>>,<<HAdd>>] loop:<<Loop>>      outer_loop:none
   private static void halving_add_unsigned_constant(char[] b1, char[] bo) {
     int min_length = Math.min(bo.length, b1.length);
@@ -247,7 +247,7 @@
   /// CHECK-DAG: <<Repl:d\d+>> VecReplicateScalar [<<UMAX>>]        loop:none
   /// CHECK-DAG: <<Phi:i\d+>>  Phi                                  loop:<<Loop:B\d+>> outer_loop:none
   /// CHECK-DAG: <<Get:d\d+>>  VecLoad                              loop:<<Loop>>      outer_loop:none
-  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get>>,<<Repl>>] unsigned:true rounded:false loop:<<Loop>> outer_loop:none
+  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get>>,<<Repl>>] packed_type:Uint16 rounded:false loop:<<Loop>> outer_loop:none
   /// CHECK-DAG:               VecStore [{{l\d+}},<<Phi>>,<<HAdd>>] loop:<<Loop>>      outer_loop:none
   //
   /// CHECK-START-MIPS64: void Main.halving_add_also_unsigned_constant(char[], char[]) loop_optimization (after)
@@ -255,7 +255,7 @@
   /// CHECK-DAG: <<Repl:d\d+>> VecReplicateScalar [<<UMAX>>]        loop:none
   /// CHECK-DAG: <<Phi:i\d+>>  Phi                                  loop:<<Loop:B\d+>> outer_loop:none
   /// CHECK-DAG: <<Get:d\d+>>  VecLoad                              loop:<<Loop>>      outer_loop:none
-  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get>>,<<Repl>>] unsigned:true rounded:false loop:<<Loop>> outer_loop:none
+  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get>>,<<Repl>>] packed_type:Uint16 rounded:false loop:<<Loop>> outer_loop:none
   /// CHECK-DAG:               VecStore [{{l\d+}},<<Phi>>,<<HAdd>>] loop:<<Loop>>      outer_loop:none
   //
   // Note: HAnd has no impact (already a zero extension).
diff --git a/test/646-checker-hadd-alt-short/src/Main.java b/test/646-checker-hadd-alt-short/src/Main.java
index 6cd102f..da94829 100644
--- a/test/646-checker-hadd-alt-short/src/Main.java
+++ b/test/646-checker-hadd-alt-short/src/Main.java
@@ -43,21 +43,21 @@
   /// CHECK-DAG: <<Phi:i\d+>>  Phi                                  loop:<<Loop:B\d+>> outer_loop:none
   /// CHECK-DAG: <<Get1:d\d+>> VecLoad                              loop:<<Loop>>      outer_loop:none
   /// CHECK-DAG: <<Get2:d\d+>> VecLoad                              loop:<<Loop>>      outer_loop:none
-  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] unsigned:false rounded:false loop:<<Loop>> outer_loop:none
+  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] packed_type:Int16 rounded:false loop:<<Loop>> outer_loop:none
   /// CHECK-DAG:               VecStore [{{l\d+}},<<Phi>>,<<HAdd>>] loop:<<Loop>>      outer_loop:none
   //
   /// CHECK-START-ARM64: void Main.halving_add_signed(short[], short[], short[]) loop_optimization (after)
   /// CHECK-DAG: <<Phi:i\d+>>  Phi                                  loop:<<Loop:B\d+>> outer_loop:none
   /// CHECK-DAG: <<Get1:d\d+>> VecLoad                              loop:<<Loop>>      outer_loop:none
   /// CHECK-DAG: <<Get2:d\d+>> VecLoad                              loop:<<Loop>>      outer_loop:none
-  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] unsigned:false rounded:false loop:<<Loop>> outer_loop:none
+  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] packed_type:Int16 rounded:false loop:<<Loop>> outer_loop:none
   /// CHECK-DAG:               VecStore [{{l\d+}},<<Phi>>,<<HAdd>>] loop:<<Loop>>      outer_loop:none
   //
   /// CHECK-START-MIPS64: void Main.halving_add_signed(short[], short[], short[]) loop_optimization (after)
   /// CHECK-DAG: <<Phi:i\d+>>  Phi                                  loop:<<Loop:B\d+>> outer_loop:none
   /// CHECK-DAG: <<Get1:d\d+>> VecLoad                              loop:<<Loop>>      outer_loop:none
   /// CHECK-DAG: <<Get2:d\d+>> VecLoad                              loop:<<Loop>>      outer_loop:none
-  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] unsigned:false rounded:false loop:<<Loop>> outer_loop:none
+  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] packed_type:Int16 rounded:false loop:<<Loop>> outer_loop:none
   /// CHECK-DAG:               VecStore [{{l\d+}},<<Phi>>,<<HAdd>>] loop:<<Loop>>      outer_loop:none
   private static void halving_add_signed(short[] b1, short[] b2, short[] bo) {
     int min_length = Math.min(bo.length, Math.min(b1.length, b2.length));
@@ -83,21 +83,21 @@
   /// CHECK-DAG: <<Phi:i\d+>>  Phi                                  loop:<<Loop:B\d+>> outer_loop:none
   /// CHECK-DAG: <<Get1:d\d+>> VecLoad                              loop:<<Loop>>      outer_loop:none
   /// CHECK-DAG: <<Get2:d\d+>> VecLoad                              loop:<<Loop>>      outer_loop:none
-  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] unsigned:true rounded:false loop:<<Loop>> outer_loop:none
+  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] packed_type:Uint16 rounded:false loop:<<Loop>> outer_loop:none
   /// CHECK-DAG:               VecStore [{{l\d+}},<<Phi>>,<<HAdd>>] loop:<<Loop>>      outer_loop:none
   //
   /// CHECK-START-ARM64: void Main.halving_add_unsigned(short[], short[], short[]) loop_optimization (after)
   /// CHECK-DAG: <<Phi:i\d+>>  Phi                                  loop:<<Loop:B\d+>> outer_loop:none
   /// CHECK-DAG: <<Get1:d\d+>> VecLoad                              loop:<<Loop>>      outer_loop:none
   /// CHECK-DAG: <<Get2:d\d+>> VecLoad                              loop:<<Loop>>      outer_loop:none
-  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] unsigned:true rounded:false loop:<<Loop>> outer_loop:none
+  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] packed_type:Uint16 rounded:false loop:<<Loop>> outer_loop:none
   /// CHECK-DAG:               VecStore [{{l\d+}},<<Phi>>,<<HAdd>>] loop:<<Loop>>      outer_loop:none
   //
   /// CHECK-START-MIPS64: void Main.halving_add_unsigned(short[], short[], short[]) loop_optimization (after)
   /// CHECK-DAG: <<Phi:i\d+>>  Phi                                  loop:<<Loop:B\d+>> outer_loop:none
   /// CHECK-DAG: <<Get1:d\d+>> VecLoad                              loop:<<Loop>>      outer_loop:none
   /// CHECK-DAG: <<Get2:d\d+>> VecLoad                              loop:<<Loop>>      outer_loop:none
-  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] unsigned:true rounded:false loop:<<Loop>> outer_loop:none
+  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] packed_type:Uint16 rounded:false loop:<<Loop>> outer_loop:none
   /// CHECK-DAG:               VecStore [{{l\d+}},<<Phi>>,<<HAdd>>] loop:<<Loop>>      outer_loop:none
   private static void halving_add_unsigned(short[] b1, short[] b2, short[] bo) {
     int min_length = Math.min(bo.length, Math.min(b1.length, b2.length));
@@ -121,21 +121,21 @@
   /// CHECK-DAG: <<Phi:i\d+>>  Phi                                  loop:<<Loop:B\d+>> outer_loop:none
   /// CHECK-DAG: <<Get1:d\d+>> VecLoad                              loop:<<Loop>>      outer_loop:none
   /// CHECK-DAG: <<Get2:d\d+>> VecLoad                              loop:<<Loop>>      outer_loop:none
-  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] unsigned:false rounded:true loop:<<Loop>> outer_loop:none
+  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] packed_type:Int16 rounded:true loop:<<Loop>> outer_loop:none
   /// CHECK-DAG:               VecStore [{{l\d+}},<<Phi>>,<<HAdd>>] loop:<<Loop>>      outer_loop:none
   //
   /// CHECK-START-ARM64: void Main.rounding_halving_add_signed(short[], short[], short[]) loop_optimization (after)
   /// CHECK-DAG: <<Phi:i\d+>>  Phi                                  loop:<<Loop:B\d+>> outer_loop:none
   /// CHECK-DAG: <<Get1:d\d+>> VecLoad                              loop:<<Loop>>      outer_loop:none
   /// CHECK-DAG: <<Get2:d\d+>> VecLoad                              loop:<<Loop>>      outer_loop:none
-  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] unsigned:false rounded:true loop:<<Loop>> outer_loop:none
+  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] packed_type:Int16 rounded:true loop:<<Loop>> outer_loop:none
   /// CHECK-DAG:               VecStore [{{l\d+}},<<Phi>>,<<HAdd>>] loop:<<Loop>>      outer_loop:none
   //
   /// CHECK-START-MIPS64: void Main.rounding_halving_add_signed(short[], short[], short[]) loop_optimization (after)
   /// CHECK-DAG: <<Phi:i\d+>>  Phi                                  loop:<<Loop:B\d+>> outer_loop:none
   /// CHECK-DAG: <<Get1:d\d+>> VecLoad                              loop:<<Loop>>      outer_loop:none
   /// CHECK-DAG: <<Get2:d\d+>> VecLoad                              loop:<<Loop>>      outer_loop:none
-  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] unsigned:false rounded:true loop:<<Loop>> outer_loop:none
+  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] packed_type:Int16 rounded:true loop:<<Loop>> outer_loop:none
   /// CHECK-DAG:               VecStore [{{l\d+}},<<Phi>>,<<HAdd>>] loop:<<Loop>>      outer_loop:none
   private static void rounding_halving_add_signed(short[] b1, short[] b2, short[] bo) {
     int min_length = Math.min(bo.length, Math.min(b1.length, b2.length));
@@ -162,21 +162,21 @@
   /// CHECK-DAG: <<Phi:i\d+>>  Phi                                  loop:<<Loop:B\d+>> outer_loop:none
   /// CHECK-DAG: <<Get1:d\d+>> VecLoad                              loop:<<Loop>>      outer_loop:none
   /// CHECK-DAG: <<Get2:d\d+>> VecLoad                              loop:<<Loop>>      outer_loop:none
-  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] unsigned:true rounded:true loop:<<Loop>> outer_loop:none
+  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] packed_type:Uint16 rounded:true loop:<<Loop>> outer_loop:none
   /// CHECK-DAG:               VecStore [{{l\d+}},<<Phi>>,<<HAdd>>] loop:<<Loop>>      outer_loop:none
   //
   /// CHECK-START-ARM64: void Main.rounding_halving_add_unsigned(short[], short[], short[]) loop_optimization (after)
   /// CHECK-DAG: <<Phi:i\d+>>  Phi                                  loop:<<Loop:B\d+>> outer_loop:none
   /// CHECK-DAG: <<Get1:d\d+>> VecLoad                              loop:<<Loop>>      outer_loop:none
   /// CHECK-DAG: <<Get2:d\d+>> VecLoad                              loop:<<Loop>>      outer_loop:none
-  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] unsigned:true rounded:true loop:<<Loop>> outer_loop:none
+  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] packed_type:Uint16 rounded:true loop:<<Loop>> outer_loop:none
   /// CHECK-DAG:               VecStore [{{l\d+}},<<Phi>>,<<HAdd>>] loop:<<Loop>>      outer_loop:none
   //
   /// CHECK-START-MIPS64: void Main.rounding_halving_add_unsigned(short[], short[], short[]) loop_optimization (after)
   /// CHECK-DAG: <<Phi:i\d+>>  Phi                                  loop:<<Loop:B\d+>> outer_loop:none
   /// CHECK-DAG: <<Get1:d\d+>> VecLoad                              loop:<<Loop>>      outer_loop:none
   /// CHECK-DAG: <<Get2:d\d+>> VecLoad                              loop:<<Loop>>      outer_loop:none
-  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] unsigned:true rounded:true loop:<<Loop>> outer_loop:none
+  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] packed_type:Uint16 rounded:true loop:<<Loop>> outer_loop:none
   /// CHECK-DAG:               VecStore [{{l\d+}},<<Phi>>,<<HAdd>>] loop:<<Loop>>      outer_loop:none
   private static void rounding_halving_add_unsigned(short[] b1, short[] b2, short[] bo) {
     int min_length = Math.min(bo.length, Math.min(b1.length, b2.length));
@@ -200,7 +200,7 @@
   /// CHECK-DAG: <<Repl:d\d+>> VecReplicateScalar [<<SMAX>>]        loop:none
   /// CHECK-DAG: <<Phi:i\d+>>  Phi                                  loop:<<Loop:B\d+>> outer_loop:none
   /// CHECK-DAG: <<Get:d\d+>>  VecLoad                              loop:<<Loop>>      outer_loop:none
-  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get>>,<<Repl>>] unsigned:false rounded:false loop:<<Loop>> outer_loop:none
+  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get>>,<<Repl>>] packed_type:Int16 rounded:false loop:<<Loop>> outer_loop:none
   /// CHECK-DAG:               VecStore [{{l\d+}},<<Phi>>,<<HAdd>>] loop:<<Loop>>      outer_loop:none
   //
   /// CHECK-START-ARM64: void Main.halving_add_signed_constant(short[], short[]) loop_optimization (after)
@@ -208,7 +208,7 @@
   /// CHECK-DAG: <<Repl:d\d+>> VecReplicateScalar [<<SMAX>>]        loop:none
   /// CHECK-DAG: <<Phi:i\d+>>  Phi                                  loop:<<Loop:B\d+>> outer_loop:none
   /// CHECK-DAG: <<Get:d\d+>>  VecLoad                              loop:<<Loop>>      outer_loop:none
-  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get>>,<<Repl>>] unsigned:false rounded:false loop:<<Loop>> outer_loop:none
+  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get>>,<<Repl>>] packed_type:Int16 rounded:false loop:<<Loop>> outer_loop:none
   /// CHECK-DAG:               VecStore [{{l\d+}},<<Phi>>,<<HAdd>>] loop:<<Loop>>      outer_loop:none
   //
   /// CHECK-START-MIPS64: void Main.halving_add_signed_constant(short[], short[]) loop_optimization (after)
@@ -216,7 +216,7 @@
   /// CHECK-DAG: <<Repl:d\d+>> VecReplicateScalar [<<SMAX>>]        loop:none
   /// CHECK-DAG: <<Phi:i\d+>>  Phi                                  loop:<<Loop:B\d+>> outer_loop:none
   /// CHECK-DAG: <<Get:d\d+>>  VecLoad                              loop:<<Loop>>      outer_loop:none
-  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get>>,<<Repl>>] unsigned:false rounded:false loop:<<Loop>> outer_loop:none
+  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get>>,<<Repl>>] packed_type:Int16 rounded:false loop:<<Loop>> outer_loop:none
   /// CHECK-DAG:               VecStore [{{l\d+}},<<Phi>>,<<HAdd>>] loop:<<Loop>>      outer_loop:none
   private static void halving_add_signed_constant(short[] b1, short[] bo) {
     int min_length = Math.min(bo.length, b1.length);
@@ -241,7 +241,7 @@
   /// CHECK-DAG: <<Repl:d\d+>> VecReplicateScalar [<<UMAX>>]        loop:none
   /// CHECK-DAG: <<Phi:i\d+>>  Phi                                  loop:<<Loop:B\d+>> outer_loop:none
   /// CHECK-DAG: <<Get:d\d+>>  VecLoad                              loop:<<Loop>>      outer_loop:none
-  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get>>,<<Repl>>] unsigned:true rounded:false loop:<<Loop>> outer_loop:none
+  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get>>,<<Repl>>] packed_type:Uint16 rounded:false loop:<<Loop>> outer_loop:none
   /// CHECK-DAG:               VecStore [{{l\d+}},<<Phi>>,<<HAdd>>] loop:<<Loop>>      outer_loop:none
   //
   /// CHECK-START-ARM64: void Main.halving_add_unsigned_constant(short[], short[]) loop_optimization (after)
@@ -249,7 +249,7 @@
   /// CHECK-DAG: <<Repl:d\d+>> VecReplicateScalar [<<UMAX>>]        loop:none
   /// CHECK-DAG: <<Phi:i\d+>>  Phi                                  loop:<<Loop:B\d+>> outer_loop:none
   /// CHECK-DAG: <<Get:d\d+>>  VecLoad                              loop:<<Loop>>      outer_loop:none
-  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get>>,<<Repl>>] unsigned:true rounded:false loop:<<Loop>> outer_loop:none
+  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get>>,<<Repl>>] packed_type:Uint16 rounded:false loop:<<Loop>> outer_loop:none
   /// CHECK-DAG:               VecStore [{{l\d+}},<<Phi>>,<<HAdd>>] loop:<<Loop>>      outer_loop:none
   //
   /// CHECK-START-MIPS64: void Main.halving_add_unsigned_constant(short[], short[]) loop_optimization (after)
@@ -257,7 +257,7 @@
   /// CHECK-DAG: <<Repl:d\d+>> VecReplicateScalar [<<UMAX>>]        loop:none
   /// CHECK-DAG: <<Phi:i\d+>>  Phi                                  loop:<<Loop:B\d+>> outer_loop:none
   /// CHECK-DAG: <<Get:d\d+>>  VecLoad                              loop:<<Loop>>      outer_loop:none
-  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get>>,<<Repl>>] unsigned:true rounded:false loop:<<Loop>> outer_loop:none
+  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get>>,<<Repl>>] packed_type:Uint16 rounded:false loop:<<Loop>> outer_loop:none
   /// CHECK-DAG:               VecStore [{{l\d+}},<<Phi>>,<<HAdd>>] loop:<<Loop>>      outer_loop:none
   private static void halving_add_unsigned_constant(short[] b1, short[] bo) {
     int min_length = Math.min(bo.length, b1.length);
diff --git a/test/646-checker-hadd-byte/src/Main.java b/test/646-checker-hadd-byte/src/Main.java
index a9e844c..f0adca3 100644
--- a/test/646-checker-hadd-byte/src/Main.java
+++ b/test/646-checker-hadd-byte/src/Main.java
@@ -40,21 +40,21 @@
   /// CHECK-DAG: <<Phi:i\d+>>  Phi                                  loop:<<Loop:B\d+>> outer_loop:none
   /// CHECK-DAG: <<Get1:d\d+>> VecLoad                              loop:<<Loop>>      outer_loop:none
   /// CHECK-DAG: <<Get2:d\d+>> VecLoad                              loop:<<Loop>>      outer_loop:none
-  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] unsigned:false rounded:false loop:<<Loop>> outer_loop:none
+  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] packed_type:Int8 rounded:false loop:<<Loop>> outer_loop:none
   /// CHECK-DAG:               VecStore [{{l\d+}},<<Phi>>,<<HAdd>>] loop:<<Loop>>      outer_loop:none
   //
   /// CHECK-START-ARM64: void Main.halving_add_signed(byte[], byte[], byte[]) loop_optimization (after)
   /// CHECK-DAG: <<Phi:i\d+>>  Phi                                  loop:<<Loop:B\d+>> outer_loop:none
   /// CHECK-DAG: <<Get1:d\d+>> VecLoad                              loop:<<Loop>>      outer_loop:none
   /// CHECK-DAG: <<Get2:d\d+>> VecLoad                              loop:<<Loop>>      outer_loop:none
-  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] unsigned:false rounded:false loop:<<Loop>> outer_loop:none
+  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] packed_type:Int8 rounded:false loop:<<Loop>> outer_loop:none
   /// CHECK-DAG:               VecStore [{{l\d+}},<<Phi>>,<<HAdd>>] loop:<<Loop>>      outer_loop:none
   //
   /// CHECK-START-MIPS64: void Main.halving_add_signed(byte[], byte[], byte[]) loop_optimization (after)
   /// CHECK-DAG: <<Phi:i\d+>>  Phi                                  loop:<<Loop:B\d+>> outer_loop:none
   /// CHECK-DAG: <<Get1:d\d+>> VecLoad                              loop:<<Loop>>      outer_loop:none
   /// CHECK-DAG: <<Get2:d\d+>> VecLoad                              loop:<<Loop>>      outer_loop:none
-  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] unsigned:false rounded:false loop:<<Loop>> outer_loop:none
+  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] packed_type:Int8 rounded:false loop:<<Loop>> outer_loop:none
   /// CHECK-DAG:               VecStore [{{l\d+}},<<Phi>>,<<HAdd>>] loop:<<Loop>>      outer_loop:none
   private static void halving_add_signed(byte[] b1, byte[] b2, byte[] bo) {
     int min_length = Math.min(bo.length, Math.min(b1.length, b2.length));
@@ -80,21 +80,21 @@
   /// CHECK-DAG: <<Phi:i\d+>>  Phi                                  loop:<<Loop:B\d+>> outer_loop:none
   /// CHECK-DAG: <<Get1:d\d+>> VecLoad                              loop:<<Loop>>      outer_loop:none
   /// CHECK-DAG: <<Get2:d\d+>> VecLoad                              loop:<<Loop>>      outer_loop:none
-  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] unsigned:true rounded:false loop:<<Loop>> outer_loop:none
+  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] packed_type:Uint8 rounded:false loop:<<Loop>> outer_loop:none
   /// CHECK-DAG:               VecStore [{{l\d+}},<<Phi>>,<<HAdd>>] loop:<<Loop>>      outer_loop:none
   //
   /// CHECK-START-ARM64: void Main.halving_add_unsigned(byte[], byte[], byte[]) loop_optimization (after)
   /// CHECK-DAG: <<Phi:i\d+>>  Phi                                  loop:<<Loop:B\d+>> outer_loop:none
   /// CHECK-DAG: <<Get1:d\d+>> VecLoad                              loop:<<Loop>>      outer_loop:none
   /// CHECK-DAG: <<Get2:d\d+>> VecLoad                              loop:<<Loop>>      outer_loop:none
-  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] unsigned:true rounded:false loop:<<Loop>> outer_loop:none
+  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] packed_type:Uint8 rounded:false loop:<<Loop>> outer_loop:none
   /// CHECK-DAG:               VecStore [{{l\d+}},<<Phi>>,<<HAdd>>] loop:<<Loop>>      outer_loop:none
   //
   /// CHECK-START-MIPS64: void Main.halving_add_unsigned(byte[], byte[], byte[]) loop_optimization (after)
   /// CHECK-DAG: <<Phi:i\d+>>  Phi                                  loop:<<Loop:B\d+>> outer_loop:none
   /// CHECK-DAG: <<Get1:d\d+>> VecLoad                              loop:<<Loop>>      outer_loop:none
   /// CHECK-DAG: <<Get2:d\d+>> VecLoad                              loop:<<Loop>>      outer_loop:none
-  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] unsigned:true rounded:false loop:<<Loop>> outer_loop:none
+  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] packed_type:Uint8 rounded:false loop:<<Loop>> outer_loop:none
   /// CHECK-DAG:               VecStore [{{l\d+}},<<Phi>>,<<HAdd>>] loop:<<Loop>>      outer_loop:none
   private static void halving_add_unsigned(byte[] b1, byte[] b2, byte[] bo) {
     int min_length = Math.min(bo.length, Math.min(b1.length, b2.length));
@@ -118,21 +118,21 @@
   /// CHECK-DAG: <<Phi:i\d+>>  Phi                                  loop:<<Loop:B\d+>> outer_loop:none
   /// CHECK-DAG: <<Get1:d\d+>> VecLoad                              loop:<<Loop>>      outer_loop:none
   /// CHECK-DAG: <<Get2:d\d+>> VecLoad                              loop:<<Loop>>      outer_loop:none
-  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] unsigned:false rounded:true loop:<<Loop>> outer_loop:none
+  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] packed_type:Int8 rounded:true loop:<<Loop>> outer_loop:none
   /// CHECK-DAG:               VecStore [{{l\d+}},<<Phi>>,<<HAdd>>] loop:<<Loop>>      outer_loop:none
   //
   /// CHECK-START-ARM64: void Main.rounding_halving_add_signed(byte[], byte[], byte[]) loop_optimization (after)
   /// CHECK-DAG: <<Phi:i\d+>>  Phi                                  loop:<<Loop:B\d+>> outer_loop:none
   /// CHECK-DAG: <<Get1:d\d+>> VecLoad                              loop:<<Loop>>      outer_loop:none
   /// CHECK-DAG: <<Get2:d\d+>> VecLoad                              loop:<<Loop>>      outer_loop:none
-  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] unsigned:false rounded:true loop:<<Loop>> outer_loop:none
+  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] packed_type:Int8 rounded:true loop:<<Loop>> outer_loop:none
   /// CHECK-DAG:               VecStore [{{l\d+}},<<Phi>>,<<HAdd>>] loop:<<Loop>>      outer_loop:none
   //
   /// CHECK-START-MIPS64: void Main.rounding_halving_add_signed(byte[], byte[], byte[]) loop_optimization (after)
   /// CHECK-DAG: <<Phi:i\d+>>  Phi                                  loop:<<Loop:B\d+>> outer_loop:none
   /// CHECK-DAG: <<Get1:d\d+>> VecLoad                              loop:<<Loop>>      outer_loop:none
   /// CHECK-DAG: <<Get2:d\d+>> VecLoad                              loop:<<Loop>>      outer_loop:none
-  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] unsigned:false rounded:true loop:<<Loop>> outer_loop:none
+  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] packed_type:Int8 rounded:true loop:<<Loop>> outer_loop:none
   /// CHECK-DAG:               VecStore [{{l\d+}},<<Phi>>,<<HAdd>>] loop:<<Loop>>      outer_loop:none
   private static void rounding_halving_add_signed(byte[] b1, byte[] b2, byte[] bo) {
     int min_length = Math.min(bo.length, Math.min(b1.length, b2.length));
@@ -159,21 +159,21 @@
   /// CHECK-DAG: <<Phi:i\d+>>  Phi                                  loop:<<Loop:B\d+>> outer_loop:none
   /// CHECK-DAG: <<Get1:d\d+>> VecLoad                              loop:<<Loop>>      outer_loop:none
   /// CHECK-DAG: <<Get2:d\d+>> VecLoad                              loop:<<Loop>>      outer_loop:none
-  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>]  unsigned:true rounded:true loop:<<Loop>> outer_loop:none
+  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>]  packed_type:Uint8 rounded:true loop:<<Loop>> outer_loop:none
   /// CHECK-DAG:               VecStore [{{l\d+}},<<Phi>>,<<HAdd>>] loop:<<Loop>>      outer_loop:none
   //
   /// CHECK-START-ARM64: void Main.rounding_halving_add_unsigned(byte[], byte[], byte[]) loop_optimization (after)
   /// CHECK-DAG: <<Phi:i\d+>>  Phi                                  loop:<<Loop:B\d+>> outer_loop:none
   /// CHECK-DAG: <<Get1:d\d+>> VecLoad                              loop:<<Loop>>      outer_loop:none
   /// CHECK-DAG: <<Get2:d\d+>> VecLoad                              loop:<<Loop>>      outer_loop:none
-  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>]  unsigned:true rounded:true loop:<<Loop>> outer_loop:none
+  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>]  packed_type:Uint8 rounded:true loop:<<Loop>> outer_loop:none
   /// CHECK-DAG:               VecStore [{{l\d+}},<<Phi>>,<<HAdd>>] loop:<<Loop>>      outer_loop:none
   //
   /// CHECK-START-MIPS64: void Main.rounding_halving_add_unsigned(byte[], byte[], byte[]) loop_optimization (after)
   /// CHECK-DAG: <<Phi:i\d+>>  Phi                                  loop:<<Loop:B\d+>> outer_loop:none
   /// CHECK-DAG: <<Get1:d\d+>> VecLoad                              loop:<<Loop>>      outer_loop:none
   /// CHECK-DAG: <<Get2:d\d+>> VecLoad                              loop:<<Loop>>      outer_loop:none
-  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>]  unsigned:true rounded:true loop:<<Loop>> outer_loop:none
+  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>]  packed_type:Uint8 rounded:true loop:<<Loop>> outer_loop:none
   /// CHECK-DAG:               VecStore [{{l\d+}},<<Phi>>,<<HAdd>>] loop:<<Loop>>      outer_loop:none
   private static void rounding_halving_add_unsigned(byte[] b1, byte[] b2, byte[] bo) {
     int min_length = Math.min(bo.length, Math.min(b1.length, b2.length));
@@ -197,7 +197,7 @@
   /// CHECK-DAG: <<Repl:d\d+>> VecReplicateScalar [<<I127>>]        loop:none
   /// CHECK-DAG: <<Phi:i\d+>>  Phi                                  loop:<<Loop:B\d+>> outer_loop:none
   /// CHECK-DAG: <<Get:d\d+>>  VecLoad                              loop:<<Loop>>      outer_loop:none
-  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get>>,<<Repl>>] unsigned:false rounded:false loop:<<Loop>> outer_loop:none
+  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get>>,<<Repl>>] packed_type:Int8 rounded:false loop:<<Loop>> outer_loop:none
   /// CHECK-DAG:               VecStore [{{l\d+}},<<Phi>>,<<HAdd>>] loop:<<Loop>>      outer_loop:none
   //
   /// CHECK-START-ARM64: void Main.halving_add_signed_constant(byte[], byte[]) loop_optimization (after)
@@ -205,7 +205,7 @@
   /// CHECK-DAG: <<Repl:d\d+>> VecReplicateScalar [<<I127>>]        loop:none
   /// CHECK-DAG: <<Phi:i\d+>>  Phi                                  loop:<<Loop:B\d+>> outer_loop:none
   /// CHECK-DAG: <<Get:d\d+>>  VecLoad                              loop:<<Loop>>      outer_loop:none
-  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get>>,<<Repl>>] unsigned:false rounded:false loop:<<Loop>> outer_loop:none
+  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get>>,<<Repl>>] packed_type:Int8 rounded:false loop:<<Loop>> outer_loop:none
   /// CHECK-DAG:               VecStore [{{l\d+}},<<Phi>>,<<HAdd>>] loop:<<Loop>>      outer_loop:none
   //
   /// CHECK-START-MIPS64: void Main.halving_add_signed_constant(byte[], byte[]) loop_optimization (after)
@@ -213,7 +213,7 @@
   /// CHECK-DAG: <<Repl:d\d+>> VecReplicateScalar [<<I127>>]        loop:none
   /// CHECK-DAG: <<Phi:i\d+>>  Phi                                  loop:<<Loop:B\d+>> outer_loop:none
   /// CHECK-DAG: <<Get:d\d+>>  VecLoad                              loop:<<Loop>>      outer_loop:none
-  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get>>,<<Repl>>] unsigned:false rounded:false loop:<<Loop>> outer_loop:none
+  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get>>,<<Repl>>] packed_type:Int8 rounded:false loop:<<Loop>> outer_loop:none
   /// CHECK-DAG:               VecStore [{{l\d+}},<<Phi>>,<<HAdd>>] loop:<<Loop>>      outer_loop:none
   private static void halving_add_signed_constant(byte[] b1, byte[] bo) {
     int min_length = Math.min(bo.length, b1.length);
@@ -238,7 +238,7 @@
   /// CHECK-DAG: <<Repl:d\d+>> VecReplicateScalar [<<I255>>]        loop:none
   /// CHECK-DAG: <<Phi:i\d+>>  Phi                                  loop:<<Loop:B\d+>> outer_loop:none
   /// CHECK-DAG: <<Get:d\d+>>  VecLoad                              loop:<<Loop>>      outer_loop:none
-  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get>>,<<Repl>>] unsigned:true rounded:false loop:<<Loop>> outer_loop:none
+  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get>>,<<Repl>>] packed_type:Uint8 rounded:false loop:<<Loop>> outer_loop:none
   /// CHECK-DAG:               VecStore [{{l\d+}},<<Phi>>,<<HAdd>>] loop:<<Loop>>      outer_loop:none
   //
   /// CHECK-START-ARM64: void Main.halving_add_unsigned_constant(byte[], byte[]) loop_optimization (after)
@@ -246,7 +246,7 @@
   /// CHECK-DAG: <<Repl:d\d+>> VecReplicateScalar [<<I255>>]        loop:none
   /// CHECK-DAG: <<Phi:i\d+>>  Phi                                  loop:<<Loop:B\d+>> outer_loop:none
   /// CHECK-DAG: <<Get:d\d+>>  VecLoad                              loop:<<Loop>>      outer_loop:none
-  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get>>,<<Repl>>] unsigned:true rounded:false loop:<<Loop>> outer_loop:none
+  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get>>,<<Repl>>] packed_type:Uint8 rounded:false loop:<<Loop>> outer_loop:none
   /// CHECK-DAG:               VecStore [{{l\d+}},<<Phi>>,<<HAdd>>] loop:<<Loop>>      outer_loop:none
   //
   /// CHECK-START-MIPS64: void Main.halving_add_unsigned_constant(byte[], byte[]) loop_optimization (after)
@@ -254,7 +254,7 @@
   /// CHECK-DAG: <<Repl:d\d+>> VecReplicateScalar [<<I255>>]        loop:none
   /// CHECK-DAG: <<Phi:i\d+>>  Phi                                  loop:<<Loop:B\d+>> outer_loop:none
   /// CHECK-DAG: <<Get:d\d+>>  VecLoad                              loop:<<Loop>>      outer_loop:none
-  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get>>,<<Repl>>] unsigned:true rounded:false loop:<<Loop>> outer_loop:none
+  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get>>,<<Repl>>] packed_type:Uint8 rounded:false loop:<<Loop>> outer_loop:none
   /// CHECK-DAG:               VecStore [{{l\d+}},<<Phi>>,<<HAdd>>] loop:<<Loop>>      outer_loop:none
   private static void halving_add_unsigned_constant(byte[] b1, byte[] bo) {
     int min_length = Math.min(bo.length, b1.length);
diff --git a/test/646-checker-hadd-char/src/Main.java b/test/646-checker-hadd-char/src/Main.java
index 22eb7cb..94030cc 100644
--- a/test/646-checker-hadd-char/src/Main.java
+++ b/test/646-checker-hadd-char/src/Main.java
@@ -40,21 +40,21 @@
   /// CHECK-DAG: <<Phi:i\d+>>  Phi                                  loop:<<Loop:B\d+>> outer_loop:none
   /// CHECK-DAG: <<Get1:d\d+>> VecLoad                              loop:<<Loop>>      outer_loop:none
   /// CHECK-DAG: <<Get2:d\d+>> VecLoad                              loop:<<Loop>>      outer_loop:none
-  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] unsigned:true rounded:false loop:<<Loop>> outer_loop:none
+  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] packed_type:Uint16 rounded:false loop:<<Loop>> outer_loop:none
   /// CHECK-DAG:               VecStore [{{l\d+}},<<Phi>>,<<HAdd>>] loop:<<Loop>>      outer_loop:none
   //
   /// CHECK-START-ARM64: void Main.halving_add_unsigned(char[], char[], char[]) loop_optimization (after)
   /// CHECK-DAG: <<Phi:i\d+>>  Phi                                  loop:<<Loop:B\d+>> outer_loop:none
   /// CHECK-DAG: <<Get1:d\d+>> VecLoad                              loop:<<Loop>>      outer_loop:none
   /// CHECK-DAG: <<Get2:d\d+>> VecLoad                              loop:<<Loop>>      outer_loop:none
-  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] unsigned:true rounded:false loop:<<Loop>> outer_loop:none
+  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] packed_type:Uint16 rounded:false loop:<<Loop>> outer_loop:none
   /// CHECK-DAG:               VecStore [{{l\d+}},<<Phi>>,<<HAdd>>] loop:<<Loop>>      outer_loop:none
   //
   /// CHECK-START-MIPS64: void Main.halving_add_unsigned(char[], char[], char[]) loop_optimization (after)
   /// CHECK-DAG: <<Phi:i\d+>>  Phi                                  loop:<<Loop:B\d+>> outer_loop:none
   /// CHECK-DAG: <<Get1:d\d+>> VecLoad                              loop:<<Loop>>      outer_loop:none
   /// CHECK-DAG: <<Get2:d\d+>> VecLoad                              loop:<<Loop>>      outer_loop:none
-  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] unsigned:true rounded:false loop:<<Loop>> outer_loop:none
+  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] packed_type:Uint16 rounded:false loop:<<Loop>> outer_loop:none
   /// CHECK-DAG:               VecStore [{{l\d+}},<<Phi>>,<<HAdd>>] loop:<<Loop>>      outer_loop:none
   private static void halving_add_unsigned(char[] b1, char[] b2, char[] bo) {
     int min_length = Math.min(bo.length, Math.min(b1.length, b2.length));
@@ -80,21 +80,21 @@
   /// CHECK-DAG: <<Phi:i\d+>>  Phi                                  loop:<<Loop:B\d+>> outer_loop:none
   /// CHECK-DAG: <<Get1:d\d+>> VecLoad                              loop:<<Loop>>      outer_loop:none
   /// CHECK-DAG: <<Get2:d\d+>> VecLoad                              loop:<<Loop>>      outer_loop:none
-  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] unsigned:true rounded:false loop:<<Loop>> outer_loop:none
+  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] packed_type:Uint16 rounded:false loop:<<Loop>> outer_loop:none
   /// CHECK-DAG:               VecStore [{{l\d+}},<<Phi>>,<<HAdd>>] loop:<<Loop>>      outer_loop:none
   //
   /// CHECK-START-ARM64: void Main.halving_add_also_unsigned(char[], char[], char[]) loop_optimization (after)
   /// CHECK-DAG: <<Phi:i\d+>>  Phi                                  loop:<<Loop:B\d+>> outer_loop:none
   /// CHECK-DAG: <<Get1:d\d+>> VecLoad                              loop:<<Loop>>      outer_loop:none
   /// CHECK-DAG: <<Get2:d\d+>> VecLoad                              loop:<<Loop>>      outer_loop:none
-  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] unsigned:true rounded:false loop:<<Loop>> outer_loop:none
+  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] packed_type:Uint16 rounded:false loop:<<Loop>> outer_loop:none
   /// CHECK-DAG:               VecStore [{{l\d+}},<<Phi>>,<<HAdd>>] loop:<<Loop>>      outer_loop:none
   //
   /// CHECK-START-MIPS64: void Main.halving_add_also_unsigned(char[], char[], char[]) loop_optimization (after)
   /// CHECK-DAG: <<Phi:i\d+>>  Phi                                  loop:<<Loop:B\d+>> outer_loop:none
   /// CHECK-DAG: <<Get1:d\d+>> VecLoad                              loop:<<Loop>>      outer_loop:none
   /// CHECK-DAG: <<Get2:d\d+>> VecLoad                              loop:<<Loop>>      outer_loop:none
-  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] unsigned:true rounded:false loop:<<Loop>> outer_loop:none
+  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] packed_type:Uint16 rounded:false loop:<<Loop>> outer_loop:none
   /// CHECK-DAG:               VecStore [{{l\d+}},<<Phi>>,<<HAdd>>] loop:<<Loop>>      outer_loop:none
   //
   // Note: HAnd has no impact (already a zero extension).
@@ -121,21 +121,21 @@
   /// CHECK-DAG: <<Phi:i\d+>>  Phi                                  loop:<<Loop:B\d+>> outer_loop:none
   /// CHECK-DAG: <<Get1:d\d+>> VecLoad                              loop:<<Loop>>      outer_loop:none
   /// CHECK-DAG: <<Get2:d\d+>> VecLoad                              loop:<<Loop>>      outer_loop:none
-  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] unsigned:true rounded:true loop:<<Loop>> outer_loop:none
+  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] packed_type:Uint16 rounded:true loop:<<Loop>> outer_loop:none
   /// CHECK-DAG:               VecStore [{{l\d+}},<<Phi>>,<<HAdd>>] loop:<<Loop>>      outer_loop:none
   //
   /// CHECK-START-ARM64: void Main.rounding_halving_add_unsigned(char[], char[], char[]) loop_optimization (after)
   /// CHECK-DAG: <<Phi:i\d+>>  Phi                                  loop:<<Loop:B\d+>> outer_loop:none
   /// CHECK-DAG: <<Get1:d\d+>> VecLoad                              loop:<<Loop>>      outer_loop:none
   /// CHECK-DAG: <<Get2:d\d+>> VecLoad                              loop:<<Loop>>      outer_loop:none
-  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] unsigned:true rounded:true loop:<<Loop>> outer_loop:none
+  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] packed_type:Uint16 rounded:true loop:<<Loop>> outer_loop:none
   /// CHECK-DAG:               VecStore [{{l\d+}},<<Phi>>,<<HAdd>>] loop:<<Loop>>      outer_loop:none
   //
   /// CHECK-START-MIPS64: void Main.rounding_halving_add_unsigned(char[], char[], char[]) loop_optimization (after)
   /// CHECK-DAG: <<Phi:i\d+>>  Phi                                  loop:<<Loop:B\d+>> outer_loop:none
   /// CHECK-DAG: <<Get1:d\d+>> VecLoad                              loop:<<Loop>>      outer_loop:none
   /// CHECK-DAG: <<Get2:d\d+>> VecLoad                              loop:<<Loop>>      outer_loop:none
-  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] unsigned:true rounded:true loop:<<Loop>> outer_loop:none
+  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] packed_type:Uint16 rounded:true loop:<<Loop>> outer_loop:none
   /// CHECK-DAG:               VecStore [{{l\d+}},<<Phi>>,<<HAdd>>] loop:<<Loop>>      outer_loop:none
   private static void rounding_halving_add_unsigned(char[] b1, char[] b2, char[] bo) {
     int min_length = Math.min(bo.length, Math.min(b1.length, b2.length));
@@ -162,21 +162,21 @@
   /// CHECK-DAG: <<Phi:i\d+>>  Phi                                  loop:<<Loop:B\d+>> outer_loop:none
   /// CHECK-DAG: <<Get1:d\d+>> VecLoad                              loop:<<Loop>>      outer_loop:none
   /// CHECK-DAG: <<Get2:d\d+>> VecLoad                              loop:<<Loop>>      outer_loop:none
-  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] unsigned:true rounded:true loop:<<Loop>> outer_loop:none
+  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] packed_type:Uint16 rounded:true loop:<<Loop>> outer_loop:none
   /// CHECK-DAG:               VecStore [{{l\d+}},<<Phi>>,<<HAdd>>] loop:<<Loop>>      outer_loop:none
   //
   /// CHECK-START-ARM64: void Main.rounding_halving_add_also_unsigned(char[], char[], char[]) loop_optimization (after)
   /// CHECK-DAG: <<Phi:i\d+>>  Phi                                  loop:<<Loop:B\d+>> outer_loop:none
   /// CHECK-DAG: <<Get1:d\d+>> VecLoad                              loop:<<Loop>>      outer_loop:none
   /// CHECK-DAG: <<Get2:d\d+>> VecLoad                              loop:<<Loop>>      outer_loop:none
-  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] unsigned:true rounded:true loop:<<Loop>> outer_loop:none
+  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] packed_type:Uint16 rounded:true loop:<<Loop>> outer_loop:none
   /// CHECK-DAG:               VecStore [{{l\d+}},<<Phi>>,<<HAdd>>] loop:<<Loop>>      outer_loop:none
   //
   /// CHECK-START-MIPS64: void Main.rounding_halving_add_also_unsigned(char[], char[], char[]) loop_optimization (after)
   /// CHECK-DAG: <<Phi:i\d+>>  Phi                                  loop:<<Loop:B\d+>> outer_loop:none
   /// CHECK-DAG: <<Get1:d\d+>> VecLoad                              loop:<<Loop>>      outer_loop:none
   /// CHECK-DAG: <<Get2:d\d+>> VecLoad                              loop:<<Loop>>      outer_loop:none
-  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] unsigned:true rounded:true loop:<<Loop>> outer_loop:none
+  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] packed_type:Uint16 rounded:true loop:<<Loop>> outer_loop:none
   /// CHECK-DAG:               VecStore [{{l\d+}},<<Phi>>,<<HAdd>>] loop:<<Loop>>      outer_loop:none
   //
   // Note: HAnd has no impact (already a zero extension).
@@ -203,7 +203,7 @@
   /// CHECK-DAG: <<Repl:d\d+>> VecReplicateScalar [<<UMAX>>]        loop:none
   /// CHECK-DAG: <<Phi:i\d+>>  Phi                                  loop:<<Loop:B\d+>> outer_loop:none
   /// CHECK-DAG: <<Get:d\d+>>  VecLoad                              loop:<<Loop>>      outer_loop:none
-  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get>>,<<Repl>>] unsigned:true rounded:false loop:<<Loop>> outer_loop:none
+  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get>>,<<Repl>>] packed_type:Uint16 rounded:false loop:<<Loop>> outer_loop:none
   /// CHECK-DAG:               VecStore [{{l\d+}},<<Phi>>,<<HAdd>>] loop:<<Loop>>      outer_loop:none
   //
   /// CHECK-START-ARM64: void Main.halving_add_unsigned_constant(char[], char[]) loop_optimization (after)
@@ -211,7 +211,7 @@
   /// CHECK-DAG: <<Repl:d\d+>> VecReplicateScalar [<<UMAX>>]        loop:none
   /// CHECK-DAG: <<Phi:i\d+>>  Phi                                  loop:<<Loop:B\d+>> outer_loop:none
   /// CHECK-DAG: <<Get:d\d+>>  VecLoad                              loop:<<Loop>>      outer_loop:none
-  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get>>,<<Repl>>] unsigned:true rounded:false loop:<<Loop>> outer_loop:none
+  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get>>,<<Repl>>] packed_type:Uint16 rounded:false loop:<<Loop>> outer_loop:none
   /// CHECK-DAG:               VecStore [{{l\d+}},<<Phi>>,<<HAdd>>] loop:<<Loop>>      outer_loop:none
   //
   /// CHECK-START-MIPS64: void Main.halving_add_unsigned_constant(char[], char[]) loop_optimization (after)
@@ -219,7 +219,7 @@
   /// CHECK-DAG: <<Repl:d\d+>> VecReplicateScalar [<<UMAX>>]        loop:none
   /// CHECK-DAG: <<Phi:i\d+>>  Phi                                  loop:<<Loop:B\d+>> outer_loop:none
   /// CHECK-DAG: <<Get:d\d+>>  VecLoad                              loop:<<Loop>>      outer_loop:none
-  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get>>,<<Repl>>] unsigned:true rounded:false loop:<<Loop>> outer_loop:none
+  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get>>,<<Repl>>] packed_type:Uint16 rounded:false loop:<<Loop>> outer_loop:none
   /// CHECK-DAG:               VecStore [{{l\d+}},<<Phi>>,<<HAdd>>] loop:<<Loop>>      outer_loop:none
   private static void halving_add_unsigned_constant(char[] b1, char[] bo) {
     int min_length = Math.min(bo.length, b1.length);
@@ -244,7 +244,7 @@
   /// CHECK-DAG: <<Repl:d\d+>> VecReplicateScalar [<<UMAX>>]        loop:none
   /// CHECK-DAG: <<Phi:i\d+>>  Phi                                  loop:<<Loop:B\d+>> outer_loop:none
   /// CHECK-DAG: <<Get:d\d+>>  VecLoad                              loop:<<Loop>>      outer_loop:none
-  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get>>,<<Repl>>] unsigned:true rounded:false loop:<<Loop>> outer_loop:none
+  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get>>,<<Repl>>] packed_type:Uint16 rounded:false loop:<<Loop>> outer_loop:none
   /// CHECK-DAG:               VecStore [{{l\d+}},<<Phi>>,<<HAdd>>] loop:<<Loop>>      outer_loop:none
   //
   /// CHECK-START-ARM64: void Main.halving_add_also_unsigned_constant(char[], char[]) loop_optimization (after)
@@ -252,7 +252,7 @@
   /// CHECK-DAG: <<Repl:d\d+>> VecReplicateScalar [<<UMAX>>]        loop:none
   /// CHECK-DAG: <<Phi:i\d+>>  Phi                                  loop:<<Loop:B\d+>> outer_loop:none
   /// CHECK-DAG: <<Get:d\d+>>  VecLoad                              loop:<<Loop>>      outer_loop:none
-  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get>>,<<Repl>>] unsigned:true rounded:false loop:<<Loop>> outer_loop:none
+  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get>>,<<Repl>>] packed_type:Uint16 rounded:false loop:<<Loop>> outer_loop:none
   /// CHECK-DAG:               VecStore [{{l\d+}},<<Phi>>,<<HAdd>>] loop:<<Loop>>      outer_loop:none
   //
   /// CHECK-START-MIPS64: void Main.halving_add_also_unsigned_constant(char[], char[]) loop_optimization (after)
@@ -260,7 +260,7 @@
   /// CHECK-DAG: <<Repl:d\d+>> VecReplicateScalar [<<UMAX>>]        loop:none
   /// CHECK-DAG: <<Phi:i\d+>>  Phi                                  loop:<<Loop:B\d+>> outer_loop:none
   /// CHECK-DAG: <<Get:d\d+>>  VecLoad                              loop:<<Loop>>      outer_loop:none
-  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get>>,<<Repl>>] unsigned:true rounded:false loop:<<Loop>> outer_loop:none
+  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get>>,<<Repl>>] packed_type:Uint16 rounded:false loop:<<Loop>> outer_loop:none
   /// CHECK-DAG:               VecStore [{{l\d+}},<<Phi>>,<<HAdd>>] loop:<<Loop>>      outer_loop:none
   //
   // Note: HAnd has no impact (already a zero extension).
diff --git a/test/646-checker-hadd-short/src/Main.java b/test/646-checker-hadd-short/src/Main.java
index 756f8a8..4ed2356 100644
--- a/test/646-checker-hadd-short/src/Main.java
+++ b/test/646-checker-hadd-short/src/Main.java
@@ -40,21 +40,21 @@
   /// CHECK-DAG: <<Phi:i\d+>>  Phi                                  loop:<<Loop:B\d+>> outer_loop:none
   /// CHECK-DAG: <<Get1:d\d+>> VecLoad                              loop:<<Loop>>      outer_loop:none
   /// CHECK-DAG: <<Get2:d\d+>> VecLoad                              loop:<<Loop>>      outer_loop:none
-  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] unsigned:false rounded:false loop:<<Loop>> outer_loop:none
+  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] packed_type:Int16 rounded:false loop:<<Loop>> outer_loop:none
   /// CHECK-DAG:               VecStore [{{l\d+}},<<Phi>>,<<HAdd>>] loop:<<Loop>>      outer_loop:none
   //
   /// CHECK-START-ARM64: void Main.halving_add_signed(short[], short[], short[]) loop_optimization (after)
   /// CHECK-DAG: <<Phi:i\d+>>  Phi                                  loop:<<Loop:B\d+>> outer_loop:none
   /// CHECK-DAG: <<Get1:d\d+>> VecLoad                              loop:<<Loop>>      outer_loop:none
   /// CHECK-DAG: <<Get2:d\d+>> VecLoad                              loop:<<Loop>>      outer_loop:none
-  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] unsigned:false rounded:false loop:<<Loop>> outer_loop:none
+  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] packed_type:Int16 rounded:false loop:<<Loop>> outer_loop:none
   /// CHECK-DAG:               VecStore [{{l\d+}},<<Phi>>,<<HAdd>>] loop:<<Loop>>      outer_loop:none
   //
   /// CHECK-START-MIPS64: void Main.halving_add_signed(short[], short[], short[]) loop_optimization (after)
   /// CHECK-DAG: <<Phi:i\d+>>  Phi                                  loop:<<Loop:B\d+>> outer_loop:none
   /// CHECK-DAG: <<Get1:d\d+>> VecLoad                              loop:<<Loop>>      outer_loop:none
   /// CHECK-DAG: <<Get2:d\d+>> VecLoad                              loop:<<Loop>>      outer_loop:none
-  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] unsigned:false rounded:false loop:<<Loop>> outer_loop:none
+  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] packed_type:Int16 rounded:false loop:<<Loop>> outer_loop:none
   /// CHECK-DAG:               VecStore [{{l\d+}},<<Phi>>,<<HAdd>>] loop:<<Loop>>      outer_loop:none
   private static void halving_add_signed(short[] b1, short[] b2, short[] bo) {
     int min_length = Math.min(bo.length, Math.min(b1.length, b2.length));
@@ -81,21 +81,21 @@
   /// CHECK-DAG: <<Phi:i\d+>>  Phi                                  loop:<<Loop:B\d+>> outer_loop:none
   /// CHECK-DAG: <<Get1:d\d+>> VecLoad                              loop:<<Loop>>      outer_loop:none
   /// CHECK-DAG: <<Get2:d\d+>> VecLoad                              loop:<<Loop>>      outer_loop:none
-  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] unsigned:false rounded:false loop:<<Loop>> outer_loop:none
+  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] packed_type:Int16 rounded:false loop:<<Loop>> outer_loop:none
   /// CHECK-DAG:               VecStore [{{l\d+}},<<Phi>>,<<HAdd>>] loop:<<Loop>>      outer_loop:none
   //
   /// CHECK-START-ARM64: void Main.halving_add_signed_alt(short[], short[], short[]) loop_optimization (after)
   /// CHECK-DAG: <<Phi:i\d+>>  Phi                                  loop:<<Loop:B\d+>> outer_loop:none
   /// CHECK-DAG: <<Get1:d\d+>> VecLoad                              loop:<<Loop>>      outer_loop:none
   /// CHECK-DAG: <<Get2:d\d+>> VecLoad                              loop:<<Loop>>      outer_loop:none
-  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] unsigned:false rounded:false loop:<<Loop>> outer_loop:none
+  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] packed_type:Int16 rounded:false loop:<<Loop>> outer_loop:none
   /// CHECK-DAG:               VecStore [{{l\d+}},<<Phi>>,<<HAdd>>] loop:<<Loop>>      outer_loop:none
   //
   /// CHECK-START-MIPS64: void Main.halving_add_signed_alt(short[], short[], short[]) loop_optimization (after)
   /// CHECK-DAG: <<Phi:i\d+>>  Phi                                  loop:<<Loop:B\d+>> outer_loop:none
   /// CHECK-DAG: <<Get1:d\d+>> VecLoad                              loop:<<Loop>>      outer_loop:none
   /// CHECK-DAG: <<Get2:d\d+>> VecLoad                              loop:<<Loop>>      outer_loop:none
-  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] unsigned:false rounded:false loop:<<Loop>> outer_loop:none
+  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] packed_type:Int16 rounded:false loop:<<Loop>> outer_loop:none
   /// CHECK-DAG:               VecStore [{{l\d+}},<<Phi>>,<<HAdd>>] loop:<<Loop>>      outer_loop:none
   private static void halving_add_signed_alt(short[] b1, short[] b2, short[] bo) {
     int min_length = Math.min(bo.length, Math.min(b1.length, b2.length));
@@ -122,21 +122,21 @@
   /// CHECK-DAG: <<Phi:i\d+>>  Phi                                  loop:<<Loop:B\d+>> outer_loop:none
   /// CHECK-DAG: <<Get1:d\d+>> VecLoad                              loop:<<Loop>>      outer_loop:none
   /// CHECK-DAG: <<Get2:d\d+>> VecLoad                              loop:<<Loop>>      outer_loop:none
-  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] unsigned:true rounded:false loop:<<Loop>> outer_loop:none
+  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] packed_type:Uint16 rounded:false loop:<<Loop>> outer_loop:none
   /// CHECK-DAG:               VecStore [{{l\d+}},<<Phi>>,<<HAdd>>] loop:<<Loop>>      outer_loop:none
   //
   /// CHECK-START-ARM64: void Main.halving_add_unsigned(short[], short[], short[]) loop_optimization (after)
   /// CHECK-DAG: <<Phi:i\d+>>  Phi                                  loop:<<Loop:B\d+>> outer_loop:none
   /// CHECK-DAG: <<Get1:d\d+>> VecLoad                              loop:<<Loop>>      outer_loop:none
   /// CHECK-DAG: <<Get2:d\d+>> VecLoad                              loop:<<Loop>>      outer_loop:none
-  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] unsigned:true rounded:false loop:<<Loop>> outer_loop:none
+  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] packed_type:Uint16 rounded:false loop:<<Loop>> outer_loop:none
   /// CHECK-DAG:               VecStore [{{l\d+}},<<Phi>>,<<HAdd>>] loop:<<Loop>>      outer_loop:none
   //
   /// CHECK-START-MIPS64: void Main.halving_add_unsigned(short[], short[], short[]) loop_optimization (after)
   /// CHECK-DAG: <<Phi:i\d+>>  Phi                                  loop:<<Loop:B\d+>> outer_loop:none
   /// CHECK-DAG: <<Get1:d\d+>> VecLoad                              loop:<<Loop>>      outer_loop:none
   /// CHECK-DAG: <<Get2:d\d+>> VecLoad                              loop:<<Loop>>      outer_loop:none
-  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] unsigned:true rounded:false loop:<<Loop>> outer_loop:none
+  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] packed_type:Uint16 rounded:false loop:<<Loop>> outer_loop:none
   /// CHECK-DAG:               VecStore [{{l\d+}},<<Phi>>,<<HAdd>>] loop:<<Loop>>      outer_loop:none
   private static void halving_add_unsigned(short[] b1, short[] b2, short[] bo) {
     int min_length = Math.min(bo.length, Math.min(b1.length, b2.length));
@@ -160,21 +160,21 @@
   /// CHECK-DAG: <<Phi:i\d+>>  Phi                                  loop:<<Loop:B\d+>> outer_loop:none
   /// CHECK-DAG: <<Get1:d\d+>> VecLoad                              loop:<<Loop>>      outer_loop:none
   /// CHECK-DAG: <<Get2:d\d+>> VecLoad                              loop:<<Loop>>      outer_loop:none
-  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] unsigned:false rounded:true loop:<<Loop>> outer_loop:none
+  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] packed_type:Int16 rounded:true loop:<<Loop>> outer_loop:none
   /// CHECK-DAG:               VecStore [{{l\d+}},<<Phi>>,<<HAdd>>] loop:<<Loop>>      outer_loop:none
   //
   /// CHECK-START-ARM64: void Main.rounding_halving_add_signed(short[], short[], short[]) loop_optimization (after)
   /// CHECK-DAG: <<Phi:i\d+>>  Phi                                  loop:<<Loop:B\d+>> outer_loop:none
   /// CHECK-DAG: <<Get1:d\d+>> VecLoad                              loop:<<Loop>>      outer_loop:none
   /// CHECK-DAG: <<Get2:d\d+>> VecLoad                              loop:<<Loop>>      outer_loop:none
-  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] unsigned:false rounded:true loop:<<Loop>> outer_loop:none
+  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] packed_type:Int16 rounded:true loop:<<Loop>> outer_loop:none
   /// CHECK-DAG:               VecStore [{{l\d+}},<<Phi>>,<<HAdd>>] loop:<<Loop>>      outer_loop:none
   //
   /// CHECK-START-MIPS64: void Main.rounding_halving_add_signed(short[], short[], short[]) loop_optimization (after)
   /// CHECK-DAG: <<Phi:i\d+>>  Phi                                  loop:<<Loop:B\d+>> outer_loop:none
   /// CHECK-DAG: <<Get1:d\d+>> VecLoad                              loop:<<Loop>>      outer_loop:none
   /// CHECK-DAG: <<Get2:d\d+>> VecLoad                              loop:<<Loop>>      outer_loop:none
-  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] unsigned:false rounded:true loop:<<Loop>> outer_loop:none
+  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] packed_type:Int16 rounded:true loop:<<Loop>> outer_loop:none
   /// CHECK-DAG:               VecStore [{{l\d+}},<<Phi>>,<<HAdd>>] loop:<<Loop>>      outer_loop:none
   private static void rounding_halving_add_signed(short[] b1, short[] b2, short[] bo) {
     int min_length = Math.min(bo.length, Math.min(b1.length, b2.length));
@@ -198,21 +198,21 @@
   /// CHECK-DAG: <<Phi:i\d+>>  Phi                                  loop:<<Loop:B\d+>> outer_loop:none
   /// CHECK-DAG: <<Get1:d\d+>> VecLoad                              loop:<<Loop>>      outer_loop:none
   /// CHECK-DAG: <<Get2:d\d+>> VecLoad                              loop:<<Loop>>      outer_loop:none
-  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] unsigned:false rounded:true loop:<<Loop>> outer_loop:none
+  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] packed_type:Int16 rounded:true loop:<<Loop>> outer_loop:none
   /// CHECK-DAG:               VecStore [{{l\d+}},<<Phi>>,<<HAdd>>] loop:<<Loop>>      outer_loop:none
   //
   /// CHECK-START-ARM64: void Main.rounding_halving_add_signed_alt(short[], short[], short[]) loop_optimization (after)
   /// CHECK-DAG: <<Phi:i\d+>>  Phi                                  loop:<<Loop:B\d+>> outer_loop:none
   /// CHECK-DAG: <<Get1:d\d+>> VecLoad                              loop:<<Loop>>      outer_loop:none
   /// CHECK-DAG: <<Get2:d\d+>> VecLoad                              loop:<<Loop>>      outer_loop:none
-  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] unsigned:false rounded:true loop:<<Loop>> outer_loop:none
+  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] packed_type:Int16 rounded:true loop:<<Loop>> outer_loop:none
   /// CHECK-DAG:               VecStore [{{l\d+}},<<Phi>>,<<HAdd>>] loop:<<Loop>>      outer_loop:none
   //
   /// CHECK-START-MIPS64: void Main.rounding_halving_add_signed_alt(short[], short[], short[]) loop_optimization (after)
   /// CHECK-DAG: <<Phi:i\d+>>  Phi                                  loop:<<Loop:B\d+>> outer_loop:none
   /// CHECK-DAG: <<Get1:d\d+>> VecLoad                              loop:<<Loop>>      outer_loop:none
   /// CHECK-DAG: <<Get2:d\d+>> VecLoad                              loop:<<Loop>>      outer_loop:none
-  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] unsigned:false rounded:true loop:<<Loop>> outer_loop:none
+  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] packed_type:Int16 rounded:true loop:<<Loop>> outer_loop:none
   /// CHECK-DAG:               VecStore [{{l\d+}},<<Phi>>,<<HAdd>>] loop:<<Loop>>      outer_loop:none
   private static void rounding_halving_add_signed_alt(short[] b1, short[] b2, short[] bo) {
     int min_length = Math.min(bo.length, Math.min(b1.length, b2.length));
@@ -240,21 +240,21 @@
   /// CHECK-DAG: <<Phi:i\d+>>  Phi                                  loop:<<Loop:B\d+>> outer_loop:none
   /// CHECK-DAG: <<Get1:d\d+>> VecLoad                              loop:<<Loop>>      outer_loop:none
   /// CHECK-DAG: <<Get2:d\d+>> VecLoad                              loop:<<Loop>>      outer_loop:none
-  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] unsigned:false rounded:true loop:<<Loop>> outer_loop:none
+  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] packed_type:Int16 rounded:true loop:<<Loop>> outer_loop:none
   /// CHECK-DAG:               VecStore [{{l\d+}},<<Phi>>,<<HAdd>>] loop:<<Loop>>      outer_loop:none
   //
   /// CHECK-START-ARM64: void Main.rounding_halving_add_signed_alt2(short[], short[], short[]) loop_optimization (after)
   /// CHECK-DAG: <<Phi:i\d+>>  Phi                                  loop:<<Loop:B\d+>> outer_loop:none
   /// CHECK-DAG: <<Get1:d\d+>> VecLoad                              loop:<<Loop>>      outer_loop:none
   /// CHECK-DAG: <<Get2:d\d+>> VecLoad                              loop:<<Loop>>      outer_loop:none
-  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] unsigned:false rounded:true loop:<<Loop>> outer_loop:none
+  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] packed_type:Int16 rounded:true loop:<<Loop>> outer_loop:none
   /// CHECK-DAG:               VecStore [{{l\d+}},<<Phi>>,<<HAdd>>] loop:<<Loop>>      outer_loop:none
   //
   /// CHECK-START-MIPS64: void Main.rounding_halving_add_signed_alt2(short[], short[], short[]) loop_optimization (after)
   /// CHECK-DAG: <<Phi:i\d+>>  Phi                                  loop:<<Loop:B\d+>> outer_loop:none
   /// CHECK-DAG: <<Get1:d\d+>> VecLoad                              loop:<<Loop>>      outer_loop:none
   /// CHECK-DAG: <<Get2:d\d+>> VecLoad                              loop:<<Loop>>      outer_loop:none
-  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] unsigned:false rounded:true loop:<<Loop>> outer_loop:none
+  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] packed_type:Int16 rounded:true loop:<<Loop>> outer_loop:none
   /// CHECK-DAG:               VecStore [{{l\d+}},<<Phi>>,<<HAdd>>] loop:<<Loop>>      outer_loop:none
   private static void rounding_halving_add_signed_alt2(short[] b1, short[] b2, short[] bo) {
     int min_length = Math.min(bo.length, Math.min(b1.length, b2.length));
@@ -282,21 +282,21 @@
   /// CHECK-DAG: <<Phi:i\d+>>  Phi                                  loop:<<Loop:B\d+>> outer_loop:none
   /// CHECK-DAG: <<Get1:d\d+>> VecLoad                              loop:<<Loop>>      outer_loop:none
   /// CHECK-DAG: <<Get2:d\d+>> VecLoad                              loop:<<Loop>>      outer_loop:none
-  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] unsigned:true rounded:true loop:<<Loop>> outer_loop:none
+  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] packed_type:Uint16 rounded:true loop:<<Loop>> outer_loop:none
   /// CHECK-DAG:               VecStore [{{l\d+}},<<Phi>>,<<HAdd>>] loop:<<Loop>>      outer_loop:none
   //
   /// CHECK-START-ARM64: void Main.rounding_halving_add_unsigned(short[], short[], short[]) loop_optimization (after)
   /// CHECK-DAG: <<Phi:i\d+>>  Phi                                  loop:<<Loop:B\d+>> outer_loop:none
   /// CHECK-DAG: <<Get1:d\d+>> VecLoad                              loop:<<Loop>>      outer_loop:none
   /// CHECK-DAG: <<Get2:d\d+>> VecLoad                              loop:<<Loop>>      outer_loop:none
-  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] unsigned:true rounded:true loop:<<Loop>> outer_loop:none
+  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] packed_type:Uint16 rounded:true loop:<<Loop>> outer_loop:none
   /// CHECK-DAG:               VecStore [{{l\d+}},<<Phi>>,<<HAdd>>] loop:<<Loop>>      outer_loop:none
   //
   /// CHECK-START-MIPS64: void Main.rounding_halving_add_unsigned(short[], short[], short[]) loop_optimization (after)
   /// CHECK-DAG: <<Phi:i\d+>>  Phi                                  loop:<<Loop:B\d+>> outer_loop:none
   /// CHECK-DAG: <<Get1:d\d+>> VecLoad                              loop:<<Loop>>      outer_loop:none
   /// CHECK-DAG: <<Get2:d\d+>> VecLoad                              loop:<<Loop>>      outer_loop:none
-  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] unsigned:true rounded:true loop:<<Loop>> outer_loop:none
+  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] packed_type:Uint16 rounded:true loop:<<Loop>> outer_loop:none
   /// CHECK-DAG:               VecStore [{{l\d+}},<<Phi>>,<<HAdd>>] loop:<<Loop>>      outer_loop:none
   private static void rounding_halving_add_unsigned(short[] b1, short[] b2, short[] bo) {
     int min_length = Math.min(bo.length, Math.min(b1.length, b2.length));
@@ -323,21 +323,21 @@
   /// CHECK-DAG: <<Phi:i\d+>>  Phi                                  loop:<<Loop:B\d+>> outer_loop:none
   /// CHECK-DAG: <<Get1:d\d+>> VecLoad                              loop:<<Loop>>      outer_loop:none
   /// CHECK-DAG: <<Get2:d\d+>> VecLoad                              loop:<<Loop>>      outer_loop:none
-  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] unsigned:true rounded:true loop:<<Loop>> outer_loop:none
+  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] packed_type:Uint16 rounded:true loop:<<Loop>> outer_loop:none
   /// CHECK-DAG:               VecStore [{{l\d+}},<<Phi>>,<<HAdd>>] loop:<<Loop>>      outer_loop:none
   //
   /// CHECK-START-ARM64: void Main.rounding_halving_add_unsigned_alt(short[], short[], short[]) loop_optimization (after)
   /// CHECK-DAG: <<Phi:i\d+>>  Phi                                  loop:<<Loop:B\d+>> outer_loop:none
   /// CHECK-DAG: <<Get1:d\d+>> VecLoad                              loop:<<Loop>>      outer_loop:none
   /// CHECK-DAG: <<Get2:d\d+>> VecLoad                              loop:<<Loop>>      outer_loop:none
-  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] unsigned:true rounded:true loop:<<Loop>> outer_loop:none
+  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] packed_type:Uint16 rounded:true loop:<<Loop>> outer_loop:none
   /// CHECK-DAG:               VecStore [{{l\d+}},<<Phi>>,<<HAdd>>] loop:<<Loop>>      outer_loop:none
   //
   /// CHECK-START-MIPS64: void Main.rounding_halving_add_unsigned_alt(short[], short[], short[]) loop_optimization (after)
   /// CHECK-DAG: <<Phi:i\d+>>  Phi                                  loop:<<Loop:B\d+>> outer_loop:none
   /// CHECK-DAG: <<Get1:d\d+>> VecLoad                              loop:<<Loop>>      outer_loop:none
   /// CHECK-DAG: <<Get2:d\d+>> VecLoad                              loop:<<Loop>>      outer_loop:none
-  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] unsigned:true rounded:true loop:<<Loop>> outer_loop:none
+  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get1>>,<<Get2>>] packed_type:Uint16 rounded:true loop:<<Loop>> outer_loop:none
   /// CHECK-DAG:               VecStore [{{l\d+}},<<Phi>>,<<HAdd>>] loop:<<Loop>>      outer_loop:none
   private static void rounding_halving_add_unsigned_alt(short[] b1, short[] b2, short[] bo) {
     int min_length = Math.min(bo.length, Math.min(b1.length, b2.length));
@@ -362,7 +362,7 @@
   /// CHECK-DAG: <<Repl:d\d+>> VecReplicateScalar [<<SMAX>>]        loop:none
   /// CHECK-DAG: <<Phi:i\d+>>  Phi                                  loop:<<Loop:B\d+>> outer_loop:none
   /// CHECK-DAG: <<Get:d\d+>>  VecLoad                              loop:<<Loop>>      outer_loop:none
-  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get>>,<<Repl>>] unsigned:false rounded:false loop:<<Loop>> outer_loop:none
+  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get>>,<<Repl>>] packed_type:Int16 rounded:false loop:<<Loop>> outer_loop:none
   /// CHECK-DAG:               VecStore [{{l\d+}},<<Phi>>,<<HAdd>>] loop:<<Loop>>      outer_loop:none
   //
   /// CHECK-START-ARM64: void Main.halving_add_signed_constant(short[], short[]) loop_optimization (after)
@@ -370,7 +370,7 @@
   /// CHECK-DAG: <<Repl:d\d+>> VecReplicateScalar [<<SMAX>>]        loop:none
   /// CHECK-DAG: <<Phi:i\d+>>  Phi                                  loop:<<Loop:B\d+>> outer_loop:none
   /// CHECK-DAG: <<Get:d\d+>>  VecLoad                              loop:<<Loop>>      outer_loop:none
-  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get>>,<<Repl>>] unsigned:false rounded:false loop:<<Loop>> outer_loop:none
+  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get>>,<<Repl>>] packed_type:Int16 rounded:false loop:<<Loop>> outer_loop:none
   /// CHECK-DAG:               VecStore [{{l\d+}},<<Phi>>,<<HAdd>>] loop:<<Loop>>      outer_loop:none
   //
   /// CHECK-START-MIPS64: void Main.halving_add_signed_constant(short[], short[]) loop_optimization (after)
@@ -378,7 +378,7 @@
   /// CHECK-DAG: <<Repl:d\d+>> VecReplicateScalar [<<SMAX>>]        loop:none
   /// CHECK-DAG: <<Phi:i\d+>>  Phi                                  loop:<<Loop:B\d+>> outer_loop:none
   /// CHECK-DAG: <<Get:d\d+>>  VecLoad                              loop:<<Loop>>      outer_loop:none
-  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get>>,<<Repl>>] unsigned:false rounded:false loop:<<Loop>> outer_loop:none
+  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get>>,<<Repl>>] packed_type:Int16 rounded:false loop:<<Loop>> outer_loop:none
   /// CHECK-DAG:               VecStore [{{l\d+}},<<Phi>>,<<HAdd>>] loop:<<Loop>>      outer_loop:none
   private static void halving_add_signed_constant(short[] b1, short[] bo) {
     int min_length = Math.min(bo.length, b1.length);
@@ -403,7 +403,7 @@
   /// CHECK-DAG: <<Repl:d\d+>> VecReplicateScalar [<<UMAX>>]        loop:none
   /// CHECK-DAG: <<Phi:i\d+>>  Phi                                  loop:<<Loop:B\d+>> outer_loop:none
   /// CHECK-DAG: <<Get:d\d+>>  VecLoad                              loop:<<Loop>>      outer_loop:none
-  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get>>,<<Repl>>] unsigned:true rounded:false loop:<<Loop>> outer_loop:none
+  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get>>,<<Repl>>] packed_type:Uint16 rounded:false loop:<<Loop>> outer_loop:none
   /// CHECK-DAG:               VecStore [{{l\d+}},<<Phi>>,<<HAdd>>] loop:<<Loop>>      outer_loop:none
   //
   /// CHECK-START-ARM64: void Main.halving_add_unsigned_constant(short[], short[]) loop_optimization (after)
@@ -411,7 +411,7 @@
   /// CHECK-DAG: <<Repl:d\d+>> VecReplicateScalar [<<UMAX>>]        loop:none
   /// CHECK-DAG: <<Phi:i\d+>>  Phi                                  loop:<<Loop:B\d+>> outer_loop:none
   /// CHECK-DAG: <<Get:d\d+>>  VecLoad                              loop:<<Loop>>      outer_loop:none
-  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get>>,<<Repl>>] unsigned:true rounded:false loop:<<Loop>> outer_loop:none
+  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get>>,<<Repl>>] packed_type:Uint16 rounded:false loop:<<Loop>> outer_loop:none
   /// CHECK-DAG:               VecStore [{{l\d+}},<<Phi>>,<<HAdd>>] loop:<<Loop>>      outer_loop:none
   //
   /// CHECK-START-MIPS64: void Main.halving_add_unsigned_constant(short[], short[]) loop_optimization (after)
@@ -419,7 +419,7 @@
   /// CHECK-DAG: <<Repl:d\d+>> VecReplicateScalar [<<UMAX>>]        loop:none
   /// CHECK-DAG: <<Phi:i\d+>>  Phi                                  loop:<<Loop:B\d+>> outer_loop:none
   /// CHECK-DAG: <<Get:d\d+>>  VecLoad                              loop:<<Loop>>      outer_loop:none
-  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get>>,<<Repl>>] unsigned:true rounded:false loop:<<Loop>> outer_loop:none
+  /// CHECK-DAG: <<HAdd:d\d+>> VecHalvingAdd [<<Get>>,<<Repl>>] packed_type:Uint16 rounded:false loop:<<Loop>> outer_loop:none
   /// CHECK-DAG:               VecStore [{{l\d+}},<<Phi>>,<<HAdd>>] loop:<<Loop>>      outer_loop:none
   private static void halving_add_unsigned_constant(short[] b1, short[] bo) {
     int min_length = Math.min(bo.length, b1.length);
diff --git a/test/651-checker-byte-simd-minmax/src/Main.java b/test/651-checker-byte-simd-minmax/src/Main.java
index 9643b90..44472a8 100644
--- a/test/651-checker-byte-simd-minmax/src/Main.java
+++ b/test/651-checker-byte-simd-minmax/src/Main.java
@@ -31,21 +31,21 @@
   /// CHECK-DAG: <<Phi:i\d+>>  Phi                                 loop:<<Loop:B\d+>> outer_loop:none
   /// CHECK-DAG: <<Get1:d\d+>> VecLoad                             loop:<<Loop>>      outer_loop:none
   /// CHECK-DAG: <<Get2:d\d+>> VecLoad                             loop:<<Loop>>      outer_loop:none
-  /// CHECK-DAG: <<Min:d\d+>>  VecMin [<<Get1>>,<<Get2>>] unsigned:false loop:<<Loop>> outer_loop:none
+  /// CHECK-DAG: <<Min:d\d+>>  VecMin [<<Get1>>,<<Get2>>] packed_type:Int8 loop:<<Loop>> outer_loop:none
   /// CHECK-DAG:               VecStore [{{l\d+}},<<Phi>>,<<Min>>] loop:<<Loop>>      outer_loop:none
   //
   /// CHECK-START-ARM64: void Main.doitMin(byte[], byte[], byte[]) loop_optimization (after)
   /// CHECK-DAG: <<Phi:i\d+>>  Phi                                 loop:<<Loop:B\d+>> outer_loop:none
   /// CHECK-DAG: <<Get1:d\d+>> VecLoad                             loop:<<Loop>>      outer_loop:none
   /// CHECK-DAG: <<Get2:d\d+>> VecLoad                             loop:<<Loop>>      outer_loop:none
-  /// CHECK-DAG: <<Min:d\d+>>  VecMin [<<Get1>>,<<Get2>>] unsigned:false loop:<<Loop>> outer_loop:none
+  /// CHECK-DAG: <<Min:d\d+>>  VecMin [<<Get1>>,<<Get2>>] packed_type:Int8 loop:<<Loop>> outer_loop:none
   /// CHECK-DAG:               VecStore [{{l\d+}},<<Phi>>,<<Min>>] loop:<<Loop>>      outer_loop:none
   //
   /// CHECK-START-MIPS64: void Main.doitMin(byte[], byte[], byte[]) loop_optimization (after)
   /// CHECK-DAG: <<Phi:i\d+>>  Phi                                 loop:<<Loop:B\d+>> outer_loop:none
   /// CHECK-DAG: <<Get1:d\d+>> VecLoad                             loop:<<Loop>>      outer_loop:none
   /// CHECK-DAG: <<Get2:d\d+>> VecLoad                             loop:<<Loop>>      outer_loop:none
-  /// CHECK-DAG: <<Min:d\d+>>  VecMin [<<Get1>>,<<Get2>>] unsigned:false loop:<<Loop>> outer_loop:none
+  /// CHECK-DAG: <<Min:d\d+>>  VecMin [<<Get1>>,<<Get2>>] packed_type:Int8 loop:<<Loop>> outer_loop:none
   /// CHECK-DAG:               VecStore [{{l\d+}},<<Phi>>,<<Min>>] loop:<<Loop>>      outer_loop:none
   private static void doitMin(byte[] x, byte[] y, byte[] z) {
     int min = Math.min(x.length, Math.min(y.length, z.length));
@@ -69,21 +69,21 @@
   /// CHECK-DAG: <<Phi:i\d+>>  Phi                                 loop:<<Loop:B\d+>> outer_loop:none
   /// CHECK-DAG: <<Get1:d\d+>> VecLoad                             loop:<<Loop>>      outer_loop:none
   /// CHECK-DAG: <<Get2:d\d+>> VecLoad                             loop:<<Loop>>      outer_loop:none
-  /// CHECK-DAG: <<Min:d\d+>>  VecMin [<<Get1>>,<<Get2>>] unsigned:true loop:<<Loop>> outer_loop:none
+  /// CHECK-DAG: <<Min:d\d+>>  VecMin [<<Get1>>,<<Get2>>] packed_type:Uint8 loop:<<Loop>> outer_loop:none
   /// CHECK-DAG:               VecStore [{{l\d+}},<<Phi>>,<<Min>>] loop:<<Loop>>      outer_loop:none
   //
   /// CHECK-START-ARM64: void Main.doitMinUnsigned(byte[], byte[], byte[]) loop_optimization (after)
   /// CHECK-DAG: <<Phi:i\d+>>  Phi                                 loop:<<Loop:B\d+>> outer_loop:none
   /// CHECK-DAG: <<Get1:d\d+>> VecLoad                             loop:<<Loop>>      outer_loop:none
   /// CHECK-DAG: <<Get2:d\d+>> VecLoad                             loop:<<Loop>>      outer_loop:none
-  /// CHECK-DAG: <<Min:d\d+>>  VecMin [<<Get1>>,<<Get2>>] unsigned:true loop:<<Loop>> outer_loop:none
+  /// CHECK-DAG: <<Min:d\d+>>  VecMin [<<Get1>>,<<Get2>>] packed_type:Uint8 loop:<<Loop>> outer_loop:none
   /// CHECK-DAG:               VecStore [{{l\d+}},<<Phi>>,<<Min>>] loop:<<Loop>>      outer_loop:none
   //
   /// CHECK-START-MIPS64: void Main.doitMinUnsigned(byte[], byte[], byte[]) loop_optimization (after)
   /// CHECK-DAG: <<Phi:i\d+>>  Phi                                 loop:<<Loop:B\d+>> outer_loop:none
   /// CHECK-DAG: <<Get1:d\d+>> VecLoad                             loop:<<Loop>>      outer_loop:none
   /// CHECK-DAG: <<Get2:d\d+>> VecLoad                             loop:<<Loop>>      outer_loop:none
-  /// CHECK-DAG: <<Min:d\d+>>  VecMin [<<Get1>>,<<Get2>>] unsigned:true loop:<<Loop>> outer_loop:none
+  /// CHECK-DAG: <<Min:d\d+>>  VecMin [<<Get1>>,<<Get2>>] packed_type:Uint8 loop:<<Loop>> outer_loop:none
   /// CHECK-DAG:               VecStore [{{l\d+}},<<Phi>>,<<Min>>] loop:<<Loop>>      outer_loop:none
   private static void doitMinUnsigned(byte[] x, byte[] y, byte[] z) {
     int min = Math.min(x.length, Math.min(y.length, z.length));
@@ -104,21 +104,21 @@
   /// CHECK-DAG: <<Phi:i\d+>>  Phi                                 loop:<<Loop:B\d+>> outer_loop:none
   /// CHECK-DAG: <<Get1:d\d+>> VecLoad                             loop:<<Loop>>      outer_loop:none
   /// CHECK-DAG: <<Get2:d\d+>> VecLoad                             loop:<<Loop>>      outer_loop:none
-  /// CHECK-DAG: <<Max:d\d+>>  VecMax [<<Get1>>,<<Get2>>] unsigned:false loop:<<Loop>> outer_loop:none
+  /// CHECK-DAG: <<Max:d\d+>>  VecMax [<<Get1>>,<<Get2>>] packed_type:Int8 loop:<<Loop>> outer_loop:none
   /// CHECK-DAG:               VecStore [{{l\d+}},<<Phi>>,<<Max>>] loop:<<Loop>>      outer_loop:none
   //
   /// CHECK-START-ARM64: void Main.doitMax(byte[], byte[], byte[]) loop_optimization (after)
   /// CHECK-DAG: <<Phi:i\d+>>  Phi                                 loop:<<Loop:B\d+>> outer_loop:none
   /// CHECK-DAG: <<Get1:d\d+>> VecLoad                             loop:<<Loop>>      outer_loop:none
   /// CHECK-DAG: <<Get2:d\d+>> VecLoad                             loop:<<Loop>>      outer_loop:none
-  /// CHECK-DAG: <<Max:d\d+>>  VecMax [<<Get1>>,<<Get2>>] unsigned:false loop:<<Loop>> outer_loop:none
+  /// CHECK-DAG: <<Max:d\d+>>  VecMax [<<Get1>>,<<Get2>>] packed_type:Int8 loop:<<Loop>> outer_loop:none
   /// CHECK-DAG:               VecStore [{{l\d+}},<<Phi>>,<<Max>>] loop:<<Loop>>      outer_loop:none
   //
   /// CHECK-START-MIPS64: void Main.doitMax(byte[], byte[], byte[]) loop_optimization (after)
   /// CHECK-DAG: <<Phi:i\d+>>  Phi                                 loop:<<Loop:B\d+>> outer_loop:none
   /// CHECK-DAG: <<Get1:d\d+>> VecLoad                             loop:<<Loop>>      outer_loop:none
   /// CHECK-DAG: <<Get2:d\d+>> VecLoad                             loop:<<Loop>>      outer_loop:none
-  /// CHECK-DAG: <<Max:d\d+>>  VecMax [<<Get1>>,<<Get2>>] unsigned:false loop:<<Loop>> outer_loop:none
+  /// CHECK-DAG: <<Max:d\d+>>  VecMax [<<Get1>>,<<Get2>>] packed_type:Int8 loop:<<Loop>> outer_loop:none
   /// CHECK-DAG:               VecStore [{{l\d+}},<<Phi>>,<<Max>>] loop:<<Loop>>      outer_loop:none
   private static void doitMax(byte[] x, byte[] y, byte[] z) {
     int min = Math.min(x.length, Math.min(y.length, z.length));
@@ -142,21 +142,21 @@
   /// CHECK-DAG: <<Phi:i\d+>>  Phi                                 loop:<<Loop:B\d+>> outer_loop:none
   /// CHECK-DAG: <<Get1:d\d+>> VecLoad                             loop:<<Loop>>      outer_loop:none
   /// CHECK-DAG: <<Get2:d\d+>> VecLoad                             loop:<<Loop>>      outer_loop:none
-  /// CHECK-DAG: <<Max:d\d+>>  VecMax [<<Get1>>,<<Get2>>] unsigned:true loop:<<Loop>> outer_loop:none
+  /// CHECK-DAG: <<Max:d\d+>>  VecMax [<<Get1>>,<<Get2>>] packed_type:Uint8 loop:<<Loop>> outer_loop:none
   /// CHECK-DAG:               VecStore [{{l\d+}},<<Phi>>,<<Max>>] loop:<<Loop>>      outer_loop:none
   //
   /// CHECK-START-ARM64: void Main.doitMaxUnsigned(byte[], byte[], byte[]) loop_optimization (after)
   /// CHECK-DAG: <<Phi:i\d+>>  Phi                                 loop:<<Loop:B\d+>> outer_loop:none
   /// CHECK-DAG: <<Get1:d\d+>> VecLoad                             loop:<<Loop>>      outer_loop:none
   /// CHECK-DAG: <<Get2:d\d+>> VecLoad                             loop:<<Loop>>      outer_loop:none
-  /// CHECK-DAG: <<Max:d\d+>>  VecMax [<<Get1>>,<<Get2>>] unsigned:true loop:<<Loop>> outer_loop:none
+  /// CHECK-DAG: <<Max:d\d+>>  VecMax [<<Get1>>,<<Get2>>] packed_type:Uint8 loop:<<Loop>> outer_loop:none
   /// CHECK-DAG:               VecStore [{{l\d+}},<<Phi>>,<<Max>>] loop:<<Loop>>      outer_loop:none
   //
   /// CHECK-START-MIPS64: void Main.doitMaxUnsigned(byte[], byte[], byte[]) loop_optimization (after)
   /// CHECK-DAG: <<Phi:i\d+>>  Phi                                 loop:<<Loop:B\d+>> outer_loop:none
   /// CHECK-DAG: <<Get1:d\d+>> VecLoad                             loop:<<Loop>>      outer_loop:none
   /// CHECK-DAG: <<Get2:d\d+>> VecLoad                             loop:<<Loop>>      outer_loop:none
-  /// CHECK-DAG: <<Max:d\d+>>  VecMax [<<Get1>>,<<Get2>>] unsigned:true loop:<<Loop>> outer_loop:none
+  /// CHECK-DAG: <<Max:d\d+>>  VecMax [<<Get1>>,<<Get2>>] packed_type:Uint8 loop:<<Loop>> outer_loop:none
   /// CHECK-DAG:               VecStore [{{l\d+}},<<Phi>>,<<Max>>] loop:<<Loop>>      outer_loop:none
   private static void doitMaxUnsigned(byte[] x, byte[] y, byte[] z) {
     int min = Math.min(x.length, Math.min(y.length, z.length));
@@ -178,7 +178,7 @@
   /// CHECK-DAG: <<Repl:d\d+>> VecReplicateScalar [<<I100>>]       loop:none
   /// CHECK-DAG: <<Phi:i\d+>>  Phi                                 loop:<<Loop:B\d+>> outer_loop:none
   /// CHECK-DAG: <<Get:d\d+>>  VecLoad                             loop:<<Loop>>      outer_loop:none
-  /// CHECK-DAG: <<Min:d\d+>>  VecMin [<<Get>>,<<Repl>>] unsigned:false loop:<<Loop>> outer_loop:none
+  /// CHECK-DAG: <<Min:d\d+>>  VecMin [<<Get>>,<<Repl>>] packed_type:Int8 loop:<<Loop>> outer_loop:none
   /// CHECK-DAG:               VecStore [{{l\d+}},<<Phi>>,<<Min>>] loop:<<Loop>>      outer_loop:none
   private static void doitMin100(byte[] x, byte[] y) {
     int min = Math.min(x.length, y.length);
diff --git a/test/651-checker-char-simd-minmax/src/Main.java b/test/651-checker-char-simd-minmax/src/Main.java
index 8a0262c..93f21f8 100644
--- a/test/651-checker-char-simd-minmax/src/Main.java
+++ b/test/651-checker-char-simd-minmax/src/Main.java
@@ -31,21 +31,21 @@
   /// CHECK-DAG: <<Phi:i\d+>>  Phi                                 loop:<<Loop:B\d+>> outer_loop:none
   /// CHECK-DAG: <<Get1:d\d+>> VecLoad                             loop:<<Loop>>      outer_loop:none
   /// CHECK-DAG: <<Get2:d\d+>> VecLoad                             loop:<<Loop>>      outer_loop:none
-  /// CHECK-DAG: <<Min:d\d+>>  VecMin [<<Get1>>,<<Get2>>] unsigned:true loop:<<Loop>> outer_loop:none
+  /// CHECK-DAG: <<Min:d\d+>>  VecMin [<<Get1>>,<<Get2>>] packed_type:Uint16 loop:<<Loop>> outer_loop:none
   /// CHECK-DAG:               VecStore [{{l\d+}},<<Phi>>,<<Min>>] loop:<<Loop>>      outer_loop:none
   //
   /// CHECK-START-ARM64: void Main.doitMin(char[], char[], char[]) loop_optimization (after)
   /// CHECK-DAG: <<Phi:i\d+>>  Phi                                 loop:<<Loop:B\d+>> outer_loop:none
   /// CHECK-DAG: <<Get1:d\d+>> VecLoad                             loop:<<Loop>>      outer_loop:none
   /// CHECK-DAG: <<Get2:d\d+>> VecLoad                             loop:<<Loop>>      outer_loop:none
-  /// CHECK-DAG: <<Min:d\d+>>  VecMin [<<Get1>>,<<Get2>>] unsigned:true loop:<<Loop>> outer_loop:none
+  /// CHECK-DAG: <<Min:d\d+>>  VecMin [<<Get1>>,<<Get2>>] packed_type:Uint16 loop:<<Loop>> outer_loop:none
   /// CHECK-DAG:               VecStore [{{l\d+}},<<Phi>>,<<Min>>] loop:<<Loop>>      outer_loop:none
   //
   /// CHECK-START-MIPS64: void Main.doitMin(char[], char[], char[]) loop_optimization (after)
   /// CHECK-DAG: <<Phi:i\d+>>  Phi                                 loop:<<Loop:B\d+>> outer_loop:none
   /// CHECK-DAG: <<Get1:d\d+>> VecLoad                             loop:<<Loop>>      outer_loop:none
   /// CHECK-DAG: <<Get2:d\d+>> VecLoad                             loop:<<Loop>>      outer_loop:none
-  /// CHECK-DAG: <<Min:d\d+>>  VecMin [<<Get1>>,<<Get2>>] unsigned:true loop:<<Loop>> outer_loop:none
+  /// CHECK-DAG: <<Min:d\d+>>  VecMin [<<Get1>>,<<Get2>>] packed_type:Uint16 loop:<<Loop>> outer_loop:none
   /// CHECK-DAG:               VecStore [{{l\d+}},<<Phi>>,<<Min>>] loop:<<Loop>>      outer_loop:none
   private static void doitMin(char[] x, char[] y, char[] z) {
     int min = Math.min(x.length, Math.min(y.length, z.length));
@@ -66,21 +66,21 @@
   /// CHECK-DAG: <<Phi:i\d+>>  Phi                                 loop:<<Loop:B\d+>> outer_loop:none
   /// CHECK-DAG: <<Get1:d\d+>> VecLoad                             loop:<<Loop>>      outer_loop:none
   /// CHECK-DAG: <<Get2:d\d+>> VecLoad                             loop:<<Loop>>      outer_loop:none
-  /// CHECK-DAG: <<Max:d\d+>>  VecMax [<<Get1>>,<<Get2>>] unsigned:true loop:<<Loop>> outer_loop:none
+  /// CHECK-DAG: <<Max:d\d+>>  VecMax [<<Get1>>,<<Get2>>] packed_type:Uint16 loop:<<Loop>> outer_loop:none
   /// CHECK-DAG:               VecStore [{{l\d+}},<<Phi>>,<<Max>>] loop:<<Loop>>      outer_loop:none
   //
   /// CHECK-START-ARM64: void Main.doitMax(char[], char[], char[]) loop_optimization (after)
   /// CHECK-DAG: <<Phi:i\d+>>  Phi                                 loop:<<Loop:B\d+>> outer_loop:none
   /// CHECK-DAG: <<Get1:d\d+>> VecLoad                             loop:<<Loop>>      outer_loop:none
   /// CHECK-DAG: <<Get2:d\d+>> VecLoad                             loop:<<Loop>>      outer_loop:none
-  /// CHECK-DAG: <<Max:d\d+>>  VecMax [<<Get1>>,<<Get2>>] unsigned:true loop:<<Loop>> outer_loop:none
+  /// CHECK-DAG: <<Max:d\d+>>  VecMax [<<Get1>>,<<Get2>>] packed_type:Uint16 loop:<<Loop>> outer_loop:none
   /// CHECK-DAG:               VecStore [{{l\d+}},<<Phi>>,<<Max>>] loop:<<Loop>>      outer_loop:none
   //
   /// CHECK-START-MIPS64: void Main.doitMax(char[], char[], char[]) loop_optimization (after)
   /// CHECK-DAG: <<Phi:i\d+>>  Phi                                 loop:<<Loop:B\d+>> outer_loop:none
   /// CHECK-DAG: <<Get1:d\d+>> VecLoad                             loop:<<Loop>>      outer_loop:none
   /// CHECK-DAG: <<Get2:d\d+>> VecLoad                             loop:<<Loop>>      outer_loop:none
-  /// CHECK-DAG: <<Max:d\d+>>  VecMax [<<Get1>>,<<Get2>>] unsigned:true loop:<<Loop>> outer_loop:none
+  /// CHECK-DAG: <<Max:d\d+>>  VecMax [<<Get1>>,<<Get2>>] packed_type:Uint16 loop:<<Loop>> outer_loop:none
   /// CHECK-DAG:               VecStore [{{l\d+}},<<Phi>>,<<Max>>] loop:<<Loop>>      outer_loop:none
   private static void doitMax(char[] x, char[] y, char[] z) {
     int min = Math.min(x.length, Math.min(y.length, z.length));
@@ -102,7 +102,7 @@
   /// CHECK-DAG: <<Repl:d\d+>> VecReplicateScalar [<<I100>>]       loop:none
   /// CHECK-DAG: <<Phi:i\d+>>  Phi                                 loop:<<Loop:B\d+>> outer_loop:none
   /// CHECK-DAG: <<Get:d\d+>>  VecLoad                             loop:<<Loop>>      outer_loop:none
-  /// CHECK-DAG: <<Min:d\d+>>  VecMin [<<Get>>,<<Repl>>] unsigned:true loop:<<Loop>>  outer_loop:none
+  /// CHECK-DAG: <<Min:d\d+>>  VecMin [<<Get>>,<<Repl>>] packed_type:Uint16 loop:<<Loop>>  outer_loop:none
   /// CHECK-DAG:               VecStore [{{l\d+}},<<Phi>>,<<Min>>] loop:<<Loop>>      outer_loop:none
   private static void doitMin100(char[] x, char[] y) {
     int min = Math.min(x.length, y.length);
diff --git a/test/651-checker-short-simd-minmax/src/Main.java b/test/651-checker-short-simd-minmax/src/Main.java
index ffbf73b..00569e4 100644
--- a/test/651-checker-short-simd-minmax/src/Main.java
+++ b/test/651-checker-short-simd-minmax/src/Main.java
@@ -31,21 +31,21 @@
   /// CHECK-DAG: <<Phi:i\d+>>  Phi                                 loop:<<Loop:B\d+>> outer_loop:none
   /// CHECK-DAG: <<Get1:d\d+>> VecLoad                             loop:<<Loop>>      outer_loop:none
   /// CHECK-DAG: <<Get2:d\d+>> VecLoad                             loop:<<Loop>>      outer_loop:none
-  /// CHECK-DAG: <<Min:d\d+>>  VecMin [<<Get1>>,<<Get2>>] unsigned:false loop:<<Loop>> outer_loop:none
+  /// CHECK-DAG: <<Min:d\d+>>  VecMin [<<Get1>>,<<Get2>>] packed_type:Int16 loop:<<Loop>> outer_loop:none
   /// CHECK-DAG:               VecStore [{{l\d+}},<<Phi>>,<<Min>>] loop:<<Loop>>      outer_loop:none
   //
   /// CHECK-START-ARM64: void Main.doitMin(short[], short[], short[]) loop_optimization (after)
   /// CHECK-DAG: <<Phi:i\d+>>  Phi                                 loop:<<Loop:B\d+>> outer_loop:none
   /// CHECK-DAG: <<Get1:d\d+>> VecLoad                             loop:<<Loop>>      outer_loop:none
   /// CHECK-DAG: <<Get2:d\d+>> VecLoad                             loop:<<Loop>>      outer_loop:none
-  /// CHECK-DAG: <<Min:d\d+>>  VecMin [<<Get1>>,<<Get2>>] unsigned:false loop:<<Loop>> outer_loop:none
+  /// CHECK-DAG: <<Min:d\d+>>  VecMin [<<Get1>>,<<Get2>>] packed_type:Int16 loop:<<Loop>> outer_loop:none
   /// CHECK-DAG:               VecStore [{{l\d+}},<<Phi>>,<<Min>>] loop:<<Loop>>      outer_loop:none
   //
   /// CHECK-START-MIPS64: void Main.doitMin(short[], short[], short[]) loop_optimization (after)
   /// CHECK-DAG: <<Phi:i\d+>>  Phi                                 loop:<<Loop:B\d+>> outer_loop:none
   /// CHECK-DAG: <<Get1:d\d+>> VecLoad                             loop:<<Loop>>      outer_loop:none
   /// CHECK-DAG: <<Get2:d\d+>> VecLoad                             loop:<<Loop>>      outer_loop:none
-  /// CHECK-DAG: <<Min:d\d+>>  VecMin [<<Get1>>,<<Get2>>] unsigned:false loop:<<Loop>> outer_loop:none
+  /// CHECK-DAG: <<Min:d\d+>>  VecMin [<<Get1>>,<<Get2>>] packed_type:Int16 loop:<<Loop>> outer_loop:none
   /// CHECK-DAG:               VecStore [{{l\d+}},<<Phi>>,<<Min>>] loop:<<Loop>>      outer_loop:none
   private static void doitMin(short[] x, short[] y, short[] z) {
     int min = Math.min(x.length, Math.min(y.length, z.length));
@@ -69,21 +69,21 @@
   /// CHECK-DAG: <<Phi:i\d+>>  Phi                                 loop:<<Loop:B\d+>> outer_loop:none
   /// CHECK-DAG: <<Get1:d\d+>> VecLoad                             loop:<<Loop>>      outer_loop:none
   /// CHECK-DAG: <<Get2:d\d+>> VecLoad                             loop:<<Loop>>      outer_loop:none
-  /// CHECK-DAG: <<Min:d\d+>>  VecMin [<<Get1>>,<<Get2>>] unsigned:true loop:<<Loop>> outer_loop:none
+  /// CHECK-DAG: <<Min:d\d+>>  VecMin [<<Get1>>,<<Get2>>] packed_type:Uint16 loop:<<Loop>> outer_loop:none
   /// CHECK-DAG:               VecStore [{{l\d+}},<<Phi>>,<<Min>>] loop:<<Loop>>      outer_loop:none
   //
   /// CHECK-START-ARM64: void Main.doitMinUnsigned(short[], short[], short[]) loop_optimization (after)
   /// CHECK-DAG: <<Phi:i\d+>>  Phi                                 loop:<<Loop:B\d+>> outer_loop:none
   /// CHECK-DAG: <<Get1:d\d+>> VecLoad                             loop:<<Loop>>      outer_loop:none
   /// CHECK-DAG: <<Get2:d\d+>> VecLoad                             loop:<<Loop>>      outer_loop:none
-  /// CHECK-DAG: <<Min:d\d+>>  VecMin [<<Get1>>,<<Get2>>] unsigned:true loop:<<Loop>> outer_loop:none
+  /// CHECK-DAG: <<Min:d\d+>>  VecMin [<<Get1>>,<<Get2>>] packed_type:Uint16 loop:<<Loop>> outer_loop:none
   /// CHECK-DAG:               VecStore [{{l\d+}},<<Phi>>,<<Min>>] loop:<<Loop>>      outer_loop:none
   //
   /// CHECK-START-MIPS64: void Main.doitMinUnsigned(short[], short[], short[]) loop_optimization (after)
   /// CHECK-DAG: <<Phi:i\d+>>  Phi                                 loop:<<Loop:B\d+>> outer_loop:none
   /// CHECK-DAG: <<Get1:d\d+>> VecLoad                             loop:<<Loop>>      outer_loop:none
   /// CHECK-DAG: <<Get2:d\d+>> VecLoad                             loop:<<Loop>>      outer_loop:none
-  /// CHECK-DAG: <<Min:d\d+>>  VecMin [<<Get1>>,<<Get2>>] unsigned:true loop:<<Loop>> outer_loop:none
+  /// CHECK-DAG: <<Min:d\d+>>  VecMin [<<Get1>>,<<Get2>>] packed_type:Uint16 loop:<<Loop>> outer_loop:none
   /// CHECK-DAG:               VecStore [{{l\d+}},<<Phi>>,<<Min>>] loop:<<Loop>>      outer_loop:none
   private static void doitMinUnsigned(short[] x, short[] y, short[] z) {
     int min = Math.min(x.length, Math.min(y.length, z.length));
@@ -104,21 +104,21 @@
   /// CHECK-DAG: <<Phi:i\d+>>  Phi                                 loop:<<Loop:B\d+>> outer_loop:none
   /// CHECK-DAG: <<Get1:d\d+>> VecLoad                             loop:<<Loop>>      outer_loop:none
   /// CHECK-DAG: <<Get2:d\d+>> VecLoad                             loop:<<Loop>>      outer_loop:none
-  /// CHECK-DAG: <<Max:d\d+>>  VecMax [<<Get1>>,<<Get2>>] unsigned:false loop:<<Loop>> outer_loop:none
+  /// CHECK-DAG: <<Max:d\d+>>  VecMax [<<Get1>>,<<Get2>>] packed_type:Int16 loop:<<Loop>> outer_loop:none
   /// CHECK-DAG:               VecStore [{{l\d+}},<<Phi>>,<<Max>>] loop:<<Loop>>      outer_loop:none
   //
   /// CHECK-START-ARM64: void Main.doitMax(short[], short[], short[]) loop_optimization (after)
   /// CHECK-DAG: <<Phi:i\d+>>  Phi                                 loop:<<Loop:B\d+>> outer_loop:none
   /// CHECK-DAG: <<Get1:d\d+>> VecLoad                             loop:<<Loop>>      outer_loop:none
   /// CHECK-DAG: <<Get2:d\d+>> VecLoad                             loop:<<Loop>>      outer_loop:none
-  /// CHECK-DAG: <<Max:d\d+>>  VecMax [<<Get1>>,<<Get2>>] unsigned:false loop:<<Loop>> outer_loop:none
+  /// CHECK-DAG: <<Max:d\d+>>  VecMax [<<Get1>>,<<Get2>>] packed_type:Int16 loop:<<Loop>> outer_loop:none
   /// CHECK-DAG:               VecStore [{{l\d+}},<<Phi>>,<<Max>>] loop:<<Loop>>      outer_loop:none
   //
   /// CHECK-START-MIPS64: void Main.doitMax(short[], short[], short[]) loop_optimization (after)
   /// CHECK-DAG: <<Phi:i\d+>>  Phi                                 loop:<<Loop:B\d+>> outer_loop:none
   /// CHECK-DAG: <<Get1:d\d+>> VecLoad                             loop:<<Loop>>      outer_loop:none
   /// CHECK-DAG: <<Get2:d\d+>> VecLoad                             loop:<<Loop>>      outer_loop:none
-  /// CHECK-DAG: <<Max:d\d+>>  VecMax [<<Get1>>,<<Get2>>] unsigned:false loop:<<Loop>> outer_loop:none
+  /// CHECK-DAG: <<Max:d\d+>>  VecMax [<<Get1>>,<<Get2>>] packed_type:Int16 loop:<<Loop>> outer_loop:none
   /// CHECK-DAG:               VecStore [{{l\d+}},<<Phi>>,<<Max>>] loop:<<Loop>>      outer_loop:none
   private static void doitMax(short[] x, short[] y, short[] z) {
     int min = Math.min(x.length, Math.min(y.length, z.length));
@@ -142,21 +142,21 @@
   /// CHECK-DAG: <<Phi:i\d+>>  Phi                                 loop:<<Loop:B\d+>> outer_loop:none
   /// CHECK-DAG: <<Get1:d\d+>> VecLoad                             loop:<<Loop>>      outer_loop:none
   /// CHECK-DAG: <<Get2:d\d+>> VecLoad                             loop:<<Loop>>      outer_loop:none
-  /// CHECK-DAG: <<Max:d\d+>>  VecMax [<<Get1>>,<<Get2>>] unsigned:true loop:<<Loop>> outer_loop:none
+  /// CHECK-DAG: <<Max:d\d+>>  VecMax [<<Get1>>,<<Get2>>] packed_type:Uint16 loop:<<Loop>> outer_loop:none
   /// CHECK-DAG:               VecStore [{{l\d+}},<<Phi>>,<<Max>>] loop:<<Loop>>      outer_loop:none
   //
   /// CHECK-START-ARM64: void Main.doitMaxUnsigned(short[], short[], short[]) loop_optimization (after)
   /// CHECK-DAG: <<Phi:i\d+>>  Phi                                 loop:<<Loop:B\d+>> outer_loop:none
   /// CHECK-DAG: <<Get1:d\d+>> VecLoad                             loop:<<Loop>>      outer_loop:none
   /// CHECK-DAG: <<Get2:d\d+>> VecLoad                             loop:<<Loop>>      outer_loop:none
-  /// CHECK-DAG: <<Max:d\d+>>  VecMax [<<Get1>>,<<Get2>>] unsigned:true loop:<<Loop>> outer_loop:none
+  /// CHECK-DAG: <<Max:d\d+>>  VecMax [<<Get1>>,<<Get2>>] packed_type:Uint16 loop:<<Loop>> outer_loop:none
   /// CHECK-DAG:               VecStore [{{l\d+}},<<Phi>>,<<Max>>] loop:<<Loop>>      outer_loop:none
   //
   /// CHECK-START-MIPS64: void Main.doitMaxUnsigned(short[], short[], short[]) loop_optimization (after)
   /// CHECK-DAG: <<Phi:i\d+>>  Phi                                 loop:<<Loop:B\d+>> outer_loop:none
   /// CHECK-DAG: <<Get1:d\d+>> VecLoad                             loop:<<Loop>>      outer_loop:none
   /// CHECK-DAG: <<Get2:d\d+>> VecLoad                             loop:<<Loop>>      outer_loop:none
-  /// CHECK-DAG: <<Max:d\d+>>  VecMax [<<Get1>>,<<Get2>>] unsigned:true loop:<<Loop>> outer_loop:none
+  /// CHECK-DAG: <<Max:d\d+>>  VecMax [<<Get1>>,<<Get2>>] packed_type:Uint16 loop:<<Loop>> outer_loop:none
   /// CHECK-DAG:               VecStore [{{l\d+}},<<Phi>>,<<Max>>] loop:<<Loop>>      outer_loop:none
   private static void doitMaxUnsigned(short[] x, short[] y, short[] z) {
     int min = Math.min(x.length, Math.min(y.length, z.length));
@@ -178,7 +178,7 @@
   /// CHECK-DAG: <<Repl:d\d+>> VecReplicateScalar [<<I100>>]       loop:none
   /// CHECK-DAG: <<Phi:i\d+>>  Phi                                 loop:<<Loop:B\d+>> outer_loop:none
   /// CHECK-DAG: <<Get:d\d+>>  VecLoad                             loop:<<Loop>>      outer_loop:none
-  /// CHECK-DAG: <<Min:d\d+>>  VecMin [<<Get>>,<<Repl>>] unsigned:false loop:<<Loop>> outer_loop:none
+  /// CHECK-DAG: <<Min:d\d+>>  VecMin [<<Get>>,<<Repl>>] packed_type:Int16 loop:<<Loop>> outer_loop:none
   /// CHECK-DAG:               VecStore [{{l\d+}},<<Phi>>,<<Min>>] loop:<<Loop>>      outer_loop:none
   private static void doitMin100(short[] x, short[] y) {
     int min = Math.min(x.length, y.length);