[wasm] Implement simd lowering for F32x4 and I32x4 compare ops

src/compiler/wasm-compiler.cc

Index: src/compiler/wasm-compiler.cc
diff --git a/src/compiler/wasm-compiler.cc b/src/compiler/wasm-compiler.cc
index 443680891e27fce555e9c0702f0ab32e5afa1f7c..9b287cb3c3e2bdbbf39991d2fc7a55b892759f20 100644
--- a/src/compiler/wasm-compiler.cc
+++ b/src/compiler/wasm-compiler.cc
@@ -3407,6 +3407,19 @@ Node* WasmGraphBuilder::SimdOp(wasm::WasmOpcode opcode,
     case wasm::kExprF32x4Ne:
       return graph()->NewNode(jsgraph()->machine()->Float32x4NotEqual(),
                               inputs[0], inputs[1]);
+    case wasm::kExprF32x4Gt:
+      return graph()->NewNode(jsgraph()->machine()->Float32x4GreaterThan(),
+                              inputs[0], inputs[1]);
+    case wasm::kExprF32x4Ge:
+      return graph()->NewNode(
+          jsgraph()->machine()->Float32x4GreaterThanOrEqual(), inputs[0],
+          inputs[1]);
+    case wasm::kExprF32x4Lt:
+      return graph()->NewNode(jsgraph()->machine()->Float32x4LessThan(),
+                              inputs[0], inputs[1]);
+    case wasm::kExprF32x4Le:
+      return graph()->NewNode(jsgraph()->machine()->Float32x4LessThanOrEqual(),
+                              inputs[0], inputs[1]);

[bbudge, 2017/03/02 18:29:09]

Follow the pattern below where Lt and Le are implemented with Ge and Gt,
switching the order of the inputs. We should probably also remove the redundant
machine operators.

[bbudge, 2017/03/02 23:10:59]

Just add a TODO to make the machine operators minimal (lt and le).

     case wasm::kExprI32x4Splat:
       return graph()->NewNode(jsgraph()->machine()->Int32x4Splat(), inputs[0]);
     case wasm::kExprI32x4SConvertF32x4: