Making gcc use SSE directly for float->int conversion when available