Large Language Model Performance Benchmarking on Mobile Platforms: A Thorough Evaluation