[wasm]implement simd lowering for simple F32x4 and I32x4 unops

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 ecb1987e56e193b57a6aa9ab58eb0e790f47ad80..b8f94239206638acf94248d842d4da9cf4d6104f 100644
--- a/src/compiler/wasm-compiler.cc
+++ b/src/compiler/wasm-compiler.cc
@@ -3383,6 +3383,8 @@ Node* WasmGraphBuilder::SimdOp(wasm::WasmOpcode opcode,
       return graph()->NewNode(jsgraph()->machine()->Float32x4Abs(), inputs[0]);
     case wasm::kExprF32x4Neg:
       return graph()->NewNode(jsgraph()->machine()->Float32x4Neg(), inputs[0]);
+    case wasm::kExprF32x4Sqrt:
+      return graph()->NewNode(jsgraph()->machine()->Float32x4Sqrt(), inputs[0]);

[bbudge, 2017/03/01 17:36:42]

We're rethinking sqrt support. On some platforms (ARM32) this has to be
simulated by a very slow sequence of scalar square root instructions which can't
be rescheduled. Users would be much better off using the recip and rsqrt
approximation and refinement operations instead, since they can schedule those
to avoid stalls.

     case wasm::kExprF32x4Add:
       return graph()->NewNode(jsgraph()->machine()->Float32x4Add(), inputs[0],
                               inputs[1]);