[wasm] Implement 128-bit endian swap for simd type

src/compiler/simd-scalar-lowering.cc

Index: src/compiler/simd-scalar-lowering.cc
diff --git a/src/compiler/simd-scalar-lowering.cc b/src/compiler/simd-scalar-lowering.cc
index 10c43f1aa38f1b077f7fe042478b51532b5577c0..af7dae91b9c4f3c92e87486b6e49ca6a8d7df48a 100644
--- a/src/compiler/simd-scalar-lowering.cc
+++ b/src/compiler/simd-scalar-lowering.cc
@@ -199,12 +199,17 @@ static int GetReturnCountAfterLowering(
   return result;
 }
 
+constexpr int SimdScalarLowering::kLaneOffsets[];
+
 void SimdScalarLowering::GetIndexNodes(Node* index, Node** new_indices) {
-  new_indices[0] = index;
+  int laneIndex = kLaneOffsets[0];
+  new_indices[laneIndex] = index;
   for (size_t i = 1; i < kMaxLanes; ++i) {
-    new_indices[i] = graph()->NewNode(machine()->Int32Add(), index,
-                                      graph()->NewNode(common()->Int32Constant(
-                                          static_cast<int>(i) * kLaneWidth)));
+    laneIndex = kLaneOffsets[i];
+    new_indices[laneIndex] = graph()->NewNode(
+        machine()->Int32Add(), index,
+        graph()->NewNode(
+            common()->Int32Constant(static_cast<int>(i) * kLaneWidth)));
   }
 }
 
@@ -229,6 +234,7 @@ void SimdScalarLowering::LowerLoadOp(MachineRepresentation rep, Node* node,
       rep_nodes[1] = graph()->NewNode(load_op, base, indices[1], rep_nodes[2],
                                       control_input);
       rep_nodes[0]->ReplaceInput(2, rep_nodes[1]);
+      rep_nodes[0]->ReplaceInput(1, indices[0]);
     } else {
       for (size_t i = 1; i < kMaxLanes; ++i) {
         rep_nodes[i] = graph()->NewNode(load_op, base, indices[i]);
@@ -267,6 +273,7 @@ void SimdScalarLowering::LowerStoreOp(MachineRepresentation rep, Node* node,
       rep_nodes[1] = graph()->NewNode(store_op, base, indices[1], rep_inputs[1],
                                       rep_nodes[2], control_input);
       rep_nodes[0]->ReplaceInput(3, rep_nodes[1]);
+      rep_nodes[0]->ReplaceInput(1, indices[0]);
 
     } else {
       for (size_t i = 1; i < kMaxLanes; ++i) {