[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..8653820b3876f9becc53e7cc0c6160200dbfcde2 100644
--- a/src/compiler/simd-scalar-lowering.cc
+++ b/src/compiler/simd-scalar-lowering.cc
@@ -200,12 +200,22 @@ static int GetReturnCountAfterLowering(
 }
 
 void SimdScalarLowering::GetIndexNodes(Node* index, Node** new_indices) {
+#if defined(V8_TARGET_BIG_ENDIAN)

[Clemens Hammacher, 2017/04/28 09:45:47]

It looks like this code would get a lot cleaner if we introduce something like
this in the header:

#if defined(V8_TARGET_BIG_ENDIAN)                                               
   
  static constexpr int kLaneOffsets[] = {3, 2, 1, 0};                           

#else                                                                           
   
  static constexpr int kLaneOffsets[] = {0, 1, 2, 3};                           
   
#endif                                                                          
   

+  new_indices[3] = index;
+  for (size_t i = 0; i < kMaxLanes - 1; ++i) {
+    new_indices[i] = graph()->NewNode(
+        machine()->Int32Add(), index,
+        graph()->NewNode(common()->Int32Constant(
+            static_cast<int>(kMaxLanes - i - 1) * kLaneWidth)));
+  }
+#else
   new_indices[0] = 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)));
   }
+#endif
 }
 
 void SimdScalarLowering::LowerLoadOp(MachineRepresentation rep, Node* node,
@@ -229,6 +239,7 @@ 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 +278,7 @@ 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) {