[Turbofan] Add native ARM support for basic SIMD 32x4 operations.

src/compiler/arm/instruction-selector-arm.cc

 // Copyright 2014 the V8 project authors. All rights reserved.
 // Use of this source code is governed by a BSD-style license that can be
 // found in the LICENSE file.
 
 #include "src/base/adapters.h"
 #include "src/base/bits.h"
 #include "src/compiler/instruction-selector-impl.h"
 #include "src/compiler/node-matchers.h"
 #include "src/compiler/node-properties.h"
 
@@ skipping 2240 matching lines @@
   AddressingMode addressing_mode = kMode_Offset_RR;
   InstructionOperand inputs[4];
   size_t input_count = 0;
   inputs[input_count++] = g.UseUniqueRegister(base);
   inputs[input_count++] = g.UseUniqueRegister(index);
   inputs[input_count++] = g.UseUniqueRegister(value);
   InstructionCode code = opcode | AddressingModeField::encode(addressing_mode);
   Emit(code, 0, nullptr, input_count, inputs);
 }
 
+void InstructionSelector::VisitCreateFloat32x4(Node* node) {
+  ArmOperandGenerator g(this);
+  Emit(kArmFloat32x4Splat, g.DefineAsRegister(node), g.Use(node->InputAt(0)));
+}
+
+void InstructionSelector::VisitFloat32x4ExtractLane(Node* node) {
+  ArmOperandGenerator g(this);
+  int32_t lane = OpParameter<int32_t>(node);
+  Emit(kArmFloat32x4ExtractLane, g.DefineAsRegister(node),
+       g.UseRegister(node->InputAt(0)), g.UseImmediate(lane));
+}
+
+void InstructionSelector::VisitFloat32x4ReplaceLane(Node* node) {
+  ArmOperandGenerator g(this);
+  int32_t lane = OpParameter<int32_t>(node);
+  Emit(kArmFloat32x4ReplaceLane, g.DefineAsRegister(node),
+       g.UseRegister(node->InputAt(0)), g.UseImmediate(lane),
+       g.Use(node->InputAt(1)));
+}
+
+void InstructionSelector::VisitFloat32x4FromInt32x4(Node* node) {
+  ArmOperandGenerator g(this);
+  Emit(kArmFloat32x4FromInt32x4, g.DefineAsRegister(node),
+       g.UseRegister(node->InputAt(0)));
+}
+
+void InstructionSelector::VisitFloat32x4FromUint32x4(Node* node) {
+  ArmOperandGenerator g(this);
+  Emit(kArmFloat32x4FromUint32x4, g.DefineAsRegister(node),
+       g.UseRegister(node->InputAt(0)));
+}
+
+void InstructionSelector::VisitFloat32x4Add(Node* node) {
+  ArmOperandGenerator g(this);
+  Emit(kArmFloat32x4Add, g.DefineAsRegister(node),
+       g.UseRegister(node->InputAt(0)), g.UseRegister(node->InputAt(1)));
+}
+
+void InstructionSelector::VisitFloat32x4Sub(Node* node) {
+  ArmOperandGenerator g(this);
+  Emit(kArmFloat32x4Sub, g.DefineAsRegister(node),
+       g.UseRegister(node->InputAt(0)), g.UseRegister(node->InputAt(1)));
+}
+
+void InstructionSelector::VisitCreateInt32x4(Node* node) {
+  ArmOperandGenerator g(this);
+  Emit(kArmInt32x4Splat, g.DefineAsRegister(node), g.Use(node->InputAt(0)));
+}
+
+void InstructionSelector::VisitInt32x4ExtractLane(Node* node) {
+  ArmOperandGenerator g(this);
+  int32_t lane = OpParameter<int32_t>(node);
+  Emit(kArmInt32x4ExtractLane, g.DefineAsRegister(node),
+       g.UseRegister(node->InputAt(0)), g.UseImmediate(lane));
+}
+
+void InstructionSelector::VisitInt32x4ReplaceLane(Node* node) {
+  ArmOperandGenerator g(this);
+  int32_t lane = OpParameter<int32_t>(node);
+  Emit(kArmInt32x4ReplaceLane, g.DefineAsRegister(node),
+       g.UseRegister(node->InputAt(0)), g.UseImmediate(lane),
+       g.Use(node->InputAt(1)));
+}
+
+void InstructionSelector::VisitInt32x4FromFloat32x4(Node* node) {
+  ArmOperandGenerator g(this);
+  Emit(kArmInt32x4FromFloat32x4, g.DefineAsRegister(node),
+       g.UseRegister(node->InputAt(0)));
+}
+
+void InstructionSelector::VisitUint32x4FromFloat32x4(Node* node) {
+  ArmOperandGenerator g(this);
+  Emit(kArmUint32x4FromFloat32x4, g.DefineAsRegister(node),
+       g.UseRegister(node->InputAt(0)));
+}
+
+void InstructionSelector::VisitInt32x4Add(Node* node) {
+  ArmOperandGenerator g(this);
+  Emit(kArmInt32x4Add, g.DefineAsRegister(node),
+       g.UseRegister(node->InputAt(0)), g.UseRegister(node->InputAt(1)));
+}
+
+void InstructionSelector::VisitInt32x4Sub(Node* node) {
+  ArmOperandGenerator g(this);
+  Emit(kArmInt32x4Sub, g.DefineAsRegister(node),
+       g.UseRegister(node->InputAt(0)), g.UseRegister(node->InputAt(1)));
+}
+
+void InstructionSelector::VisitInt32x4Equal(Node* node) {
+  ArmOperandGenerator g(this);
+  Emit(kArmInt32x4Eq, g.DefineAsRegister(node), g.UseRegister(node->InputAt(0)),
+       g.UseRegister(node->InputAt(1)));
+}
+
+void InstructionSelector::VisitInt32x4NotEqual(Node* node) {
+  ArmOperandGenerator g(this);
+  Emit(kArmInt32x4Ne, g.DefineAsRegister(node), g.UseRegister(node->InputAt(0)),
+       g.UseRegister(node->InputAt(1)));
+}
+
+void InstructionSelector::VisitSimd32x4Select(Node* node) {
+  ArmOperandGenerator g(this);
+  Emit(kArmSimd32x4Select, g.DefineAsRegister(node),
+       g.UseRegister(node->InputAt(0)), g.UseRegister(node->InputAt(1)),
+       g.UseRegister(node->InputAt(2)));
+}
+
 // static
 MachineOperatorBuilder::Flags
 InstructionSelector::SupportedMachineOperatorFlags() {
   MachineOperatorBuilder::Flags flags;
   if (CpuFeatures::IsSupported(SUDIV)) {
     // The sdiv and udiv instructions correctly return 0 if the divisor is 0,
     // but the fall-back implementation does not.
     flags |= MachineOperatorBuilder::kInt32DivIsSafe |
              MachineOperatorBuilder::kUint32DivIsSafe;
   }
@@ skipping 20 matching lines @@
   Vector<MachineType> req_aligned = Vector<MachineType>::New(2);
   req_aligned[0] = MachineType::Float32();
   req_aligned[1] = MachineType::Float64();
   return MachineOperatorBuilder::AlignmentRequirements::
       SomeUnalignedAccessUnsupported(req_aligned, req_aligned);
 }
 
 }  // namespace compiler
 }  // namespace internal
 }  // namespace v8