A lightweight in-place implementation for software thread-level speculation