This article was accepted into the corpus but its outbound wikilinks were never NER-processed — typical at the deepest BFS hop or when the run's entity cap was reached. No expansion funnel to show.
| TAU (software) | |
|---|---|
| Name | TAU |
TAU (software) is a performance analysis and profiling toolkit used for high-performance computing, designed to instrument and measure parallel applications on supercomputers and clusters. It provides measurement, tracing, and visualization capabilities aimed at optimizing codes written for platforms that include distributed and shared memory systems. TAU is commonly used by researchers and engineers working with scientific applications on systems associated with organizations and projects in computational science.
TAU originated to support performance engineering for large-scale simulation codes developed by researchers at Argonne National Laboratory, Oak Ridge National Laboratory, Lawrence Berkeley National Laboratory, and universities associated with the Department of Energy and the National Science Foundation. The toolkit targets applications using programming models such as MPI (Message Passing Interface), OpenMP, CUDA, HIP (computing), and runtime systems employed in projects funded by agencies like DARPA and programs at Los Alamos National Laboratory. TAU integrates with analysis and visualization tools developed at institutions such as University of Illinois Urbana–Champaign and leverages formats and standards promoted by consortia including the TOP500 community.
TAU provides a suite of components for measurement, including source-code instrumentation, binary instrumentation, and sampling frameworks used alongside tools and formats like PAPI, HPCToolkit, Score-P, VTune, and Perf. Its components include timers, counters, trace recorders, and profile aggregators that work with visualization systems such as ParaView, VisIt, and flame graph generators influenced by work at Brendan Gregg. TAU supports performance metrics collection for hardware counters exposed by vendors like Intel Corporation, NVIDIA Corporation, AMD, and systems from vendors such as Cray (company), Hewlett Packard Enterprise, and IBM.
The architecture of TAU is modular, separating front-end instrumentation layers from back-end data aggregation and storage subsystems developed to operate on systems like Fermilab clusters, National Energy Research Scientific Computing Center, and regional XSEDE resources. TAU's design accommodates language front-ends for C++, Fortran, and Python (programming language), and integrates with middleware used in projects at Sandia National Laboratories and Lawrence Livermore National Laboratory. Its plug-in and adapter model enables interoperability with trace formats and collectors from initiatives such as OpenTelemetry-styled efforts and archives maintained by Software Carpentry-affiliated groups.
TAU's measurement strategies include instrumentation APIs, automatic compiler-directed instrumentation, and event-based sampling comparable to methods used in gprof, Valgrind, and research from Stanford University and MIT. Analysis workflows produced by TAU generate profiles, timelines, and call-path summaries that help diagnose issues like load imbalance on systems exemplified by Summit (supercomputer), Sierra (supercomputer), and clusters in academic centers such as National Center for Supercomputing Applications. TAU's output formats are consumable by visualization and statistical packages developed at places like Georgia Institute of Technology and ETH Zurich, enabling performance scientists to correlate hotspots with hardware events captured by ecosystems including LIKWID.
TAU is applied across domains such as climate modeling at institutions like National Center for Atmospheric Research, computational fluid dynamics studied at NASA Ames Research Center, astrophysics simulations from groups at Princeton University and California Institute of Technology, and materials science research affiliated with MIT Lincoln Laboratory. It is used to optimize codes implementing solvers from libraries such as PETSc, Trilinos, and application frameworks like FLASH (software) and AMReX, often in collaborations with teams at Argonne and Oak Ridge.
TAU integrates with build systems and toolchains common to supercomputing environments, including CMake, Autotools, and vendor compilers like those from GNU Project, Intel Corporation, and NVIDIA Corporation. It interoperates with job schedulers and resource managers such as Slurm Workload Manager, PBS (software), and LSF (software), and connects to continuous integration services used in academic software projects coordinated by organizations like GitHub, GitLab, and Bitbucket. TAU's outputs are often visualized through pipelines involving ParaView, VisIt, and tools developed in collaboration with teams at Lawrence Berkeley National Laboratory.
TAU's development has involved collaborations among national laboratories, universities, and vendor partners, with contributions from researchers at Argonne National Laboratory, University of Oregon, University of Wisconsin–Madison, and groups funded by DOE Office of Science. Its evolution reflects advances in parallel programming models like MPI (Message Passing Interface), OpenMP, and GPU programming initiatives led by NVIDIA Corporation and standards bodies such as the OpenACC and Khronos Group. TAU's project history includes adoption in benchmark studies reported at conferences such as SC (conference), IEEE International Conference on Cluster Computing, and workshops hosted by ACM SIGARCH.
TAU is distributed under licensing terms developed to support research use and collaborations with national laboratories, universities, and commercial partners including IBM, Cray (company), and Hewlett Packard Enterprise. Binaries and source distributions have been packaged for use on platforms ranging from departmental clusters at universities to leadership-class systems listed on the TOP500 and are typically available through institutional repositories and collaborative code-hosting platforms maintained by organizations such as GitHub and SourceForge.
Category:Profilers