The rapid evolution of high performance computing (HPC) has driven the synergistic development of hardware architectures and parallel programming models. However, with the breakdown of Moore’s Law and the end of Dennard Scaling, relying solely on hardware iterations can no longer significantly improve attainable application performance, leading to a widening gap between measured application performance and peak hardware performance. Consequently, deep optimization at the software level has emerged as a crucial pathway to unlock hardware potential and enhance computational efficiency. Over decades of development, a comprehensive workflow encompassing performance data collection, performance analysis, and performance optimization has been established within the field. This study systematically reviews performance data collection and analysis methods, along with their technical foundations, across multi-scale. It provides an in-depth analysis of the advantages and limitations of existing methods and tools at different scales, including the instruction, thread, and process levels. Furthermore, this study highlights the profound impact of artificial intelligence (AI) in its dual role as both an emerging computational workload and an intelligent analysis approach, with the aim of providing cutting-edge perspectives and valuable references for research and practice in high-performance computing performance engineering.