[turbofan] Switch passing of new.target to register.

src/runtime/runtime-function.cc

Index: src/runtime/runtime-function.cc
diff --git a/src/runtime/runtime-function.cc b/src/runtime/runtime-function.cc
index d6a4f34dd086b156e458cca054c6eec32ad24b8a..2df297b1d8d61f79192e82c73af04b008e251ace 100644
--- a/src/runtime/runtime-function.cc
+++ b/src/runtime/runtime-function.cc
@@ -590,16 +590,9 @@ RUNTIME_FUNCTION(Runtime_GetNewTarget) {
   DCHECK(args.length() == 0);
   JavaScriptFrameIterator it(isolate);
   JavaScriptFrame* frame = it.frame();
-  // TODO(4544): Currently we never inline any [[Construct]] calls where the
-  // actual target differs from the new target. Fix this soon!
-  if (frame->HasInlinedFrames()) {
-    HandleScope scope(isolate);
-    List<FrameSummary> frames(FLAG_max_inlining_levels + 1);
-    it.frame()->Summarize(&frames);
-    FrameSummary& summary = frames.last();
-    return summary.is_constructor() ? Object::cast(*summary.function())
-                                    : isolate->heap()->undefined_value();
-  }
+  // TODO(4544): By now the runtime function is only used by the interpreter,
+  // get rid of the entire runtime function once the interpreter is switched.
+  DCHECK(!frame->is_optimized() && !frame->HasInlinedFrames());
   return frame->IsConstructor() ? frame->GetNewTarget()
                                 : isolate->heap()->undefined_value();
 }