Measurement tools

Hardware Performance Counter Measurement Tools

Low Level

• Perf: The main interface in the Linux kernel and a corresponding user-space tool to measure hardware counters
• PAPI (Performance-API): A generic API for applications to measure different aspects of the system. For hardware performance counters it uses the perf backend for measurements. Other plugins for GPU and other components exist
• PCM (Performance Counter Monitor): A higher level tool and API that provides common metrics like memory bandwidth and NUMA traffic. The API also provides access to any hardware counter event
• PMU-Tools: A set of Python scripts that use the perf backend
• LIKWID: Command line applications and API to measure hardware events which can use perf as backend but also provides other backends to be independent of the kernel version

High Level

• ARM Performance Report: A tool that provides a simple one page HTML report that highlights processor, memory, communication and I/O issues and offers advice on how to improve the performance.
• Vampir: A scalable framework for performance analysis using PAPI as backend
• TAU: Utilities to sample or instrument code for hardware counters and other metrics
• HPC-Toolkit: Toolkit to sample timers and hardware performance counters for serial and parallel applications
• Intel Advisor: A vectorization and threading optimization tool
• Intel VTune: A performance profiling tool to analyse algorithms and hardware usage for serial and parallel applications
• Scalasca Trace Tools: A performance optimisation tool for runtime behaviour measurement and analysis of parallel programs
• Score-P: A Scalable Performance Measurement Infrastructure for Parallel Codes
• Intel Trace Collector/Analyzer: Powerful tools that acquire/display information on the communication behavior of MPI programs
• Oracle Sampling Collector and Performance Analyzer: Pair of tools that can collect and analyze performance data for serial or parallel applications

Links and Further Information

• Virtual Institute - High Productivity Supercomputing (VI-HPS)