Hierarchical Performance Analysis Methodology for Ampere Processors

This tutorial is a guide that describes a hierarchical performance analysis methodology aimed at enabling developers to deliver high-performance applications on Ampere Altra and Ampere Altra Max 64-bit multi-core processors. It outlines a systematic approach to identify performance bottlenecks, starting with platform-level performance measurements to ascertain if the platform can effectively run the application. The methodology emphasizes the importance of characterizing CPU, memory, network utilization, and operating system performance metrics before delving into application-specific performance metrics. The guide details the iterative nature of performance analysis, where performance data is collected, code changes are made, and results are re-evaluated. It also discusses the significance of selecting workloads that are measurable, repeatable, and representative. The tutorial concludes by providing references to various performance tools and metrics that can assist in characterizing application performance effectively.