μprofiler:: Profiling user-level threads in a shared-memory programming environment

A profiler is an important tool for understanding the dynamic behaviour of concurrent programs to locate problems and optimize performance The best way to improve profiling capabilities and reduce the time to analyze a concurrent program is to use a target-specific profiler that understands the underlying concurrent runtime environment. A profiler for understanding execution of user and kernel level threads is presented, which is target-specific for the mu C++ concurrency system. This allows the insertion of hooks into the mu C++ data structures and runtime kernel to ensure crucial operations are monitored exactly. Because the profiler is written in mu C++ and has an extendible design, it is easy for users to write new metrics and incorporate them into the profiler.