papi

PAPI, or Performance API, is an open-source library that provides a portable interface to hardware performance counters found in modern CPUs and other architectures. It standardizes access to events such as processor cycles, instructions retired, cache misses, branch mispredictions, and memory-level events, enabling performance tools and applications to collect timing and architectural data across platforms without bespoke code.

The library exposes a C API with Fortran and other language wrappers. Programmers create an event set, attach hardware counters, and use start, stop, read, and accumulate operations to measure counters over a code region. PAPI also supports multiplexing, allowing more events to be measured than there are counters by time-slicing across measurements.

Platform support spans Linux, Windows, and macOS, with back-end components for x86, ARM, POWER, SPARC, and other architectures. Each platform provides a PAPI component that maps the generic events to the native performance-counter names.

PAPI originated in the high-performance computing community in the late 1990s and is maintained by an open-source project with broad community involvement. It is released under a BSD-style license and distributed with source and prebuilt binaries for major operating systems.

PAPI is used by a variety of performance analysis tools, benchmarks, and research programs to gather consistent, architecture-aware measurements. It serves as a foundation for portable performance instrumentation and enables comparative studies across hardware generations.