ARM64: Fix LCodeGen::ToOperand32.

This fixes the following generated code sequence:
  movn w1, #0     // Synthesize -1.
  cmp w0, w1

With a properly-constructed Operand, the MacroAssembler can optimize it
as follows:
  cmn w0, #1

BUG=
R=ulan@chromium.org

Review URL: https://codereview.chromium.org/253513003

git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@20989 ce2b1a6d-e550-0410-aec6-3dcde31c8c00

src/arm64/lithium-codegen-arm64.cc

@@ -1235,9 +1235,9 @@ Operand LCodeGen::ToOperand32(LOperand* op, IntegerSignedness signedness) {
     Representation r = chunk_->LookupLiteralRepresentation(const_op);
     if (r.IsInteger32()) {
       ASSERT(constant->HasInteger32Value());
-      return Operand(signedness == SIGNED_INT32
-                     ? constant->Integer32Value()
-                     : static_cast<uint32_t>(constant->Integer32Value()));
+      return (signedness == SIGNED_INT32)
+          ? Operand(constant->Integer32Value())
+          : Operand(static_cast<uint32_t>(constant->Integer32Value()));
     } else {
       // Other constants not implemented.
       Abort(kToOperand32UnsupportedImmediate);