High performance computing is essential for complex fusion simulations, especially for guiding-center drift-orbit integration, which accurately describes important physical phenomena in Tokamak devices, such as high-energy particle transport, current actuation, and energy deposition. In this paper, we develop a two-level parallel strategy for the Tokamak guiding-center drift-orbit integration program (ORBIT) on the Sunway Bluelight II supercomputer that combines thread-level and process-level optimizations. Thread-level improvements include computation data optimization, out-of-order output optimization, and particle tracking optimization, while process-level optimization methods are based on particle independence to develop new communication domain and data processing methods to reduce latency and improve scalability. On a single core group, thread-level optimization provides up to 12.02x speedup. When expanded to 2,048 processes (133,120 cores), the two-level optimized program has a strong scaling efficiency of 66.6% and a weak scaling efficiency of over 98% on a 16-process benchmark. This approach greatly accelerates ORBIT and enriches the application ecosystem of the Sunway system.