Performance Measurement with Score-P
Score-P is the primary code instrumentation and run-time measurement framework for Vampir 9, and also works natively with Scalasca, TAU, and Periscope.
It supports an extensive set of events such as function and library calls, communication events, and hardware counters. Score-P supports various instrumentation methods, including instrumentation at source level and at compile/link time. Recently, experimental support for sampling has been added. Vampir and Score-P provide a performance tool framework with special focus on highly-parallel applications. Performance data is collected from multi-process (MPI, SHMEM), thread-parallel (OpenMP, Pthreads), as well as accelerator-based paradigms (CUDA, OpenCL, OpenACC).
What's New in Score-P 7.0
- Added support for recording calls to OpenCL 2.1/2.2 functions.
- Added ability to generate an initial filter file based on “max buffer size” and “time per visit” parameters.
- Compile-time filtering via `scorep –instrument-filter` is now also available for builds using Intel compilers.
- Added support for Kokkos through the Kokkos Tools Interface