TAU Commander Initialize:

To begin Initialize TAU Commander, enter: tau initialize or simply
tau init.
This first initialization will take quite a bit of time. Not only is this
command creating a project it is also downloading and building the
TAU Performance System® and associated libraries that it depends on.
Let this run and check for successful completion. When it completes it will
display the tau dashboard (which can be viewed at anytime by entering:
'tau dashboard').

Many parameters may be defined at initialization. The full list of options is
displayed below this help section. It is easier to define options at
initialization (eg. tau init --MPI T) than it is to edit application and
measurement later.


Command Line Usage

tau initialize [arguments]

Optional Arguments Description
--bare [T/F] Initialize project storage but don't configure anything.
-h, --help Show this help message and exit.
Application Arguments Description
--application-name name Name of the new application configuration.
--cuda [T/F] Application uses NVIDIA CUDA.
--linkage linkage Application linkage.
- linkage: static, dynamic
--mpc [T/F] Application uses MPC.
--mpi [T/F] Application uses MPI.
--opencl [T/F] Application uses OpenCL.
--openmp [T/F] Application uses OpenMP.
--pthreads [T/F] Application uses pthreads.
--select-file path Specify selective instrumentation file.
--shmem [T/F] Application uses SHMEM.
--tbb [T/F] Application uses Thread Building Blocks (TBB).
Measurement Arguments Description
--baseline [T/F] Completely disable all instrumentation and measure wall clock time via the OS.
--callpath [depth] Maximum depth for callpath recording.
--callsite [T/F] Record event callsites.
--comm-matrix [T/F] Record the point-to-point communication matrix.
--compiler-inst [mode] Use compiler-generated callbacks to gather performance data.
- mode: always, fallback, never
--force-tau-options option [option ...] Forcibly set the TAU_OPTIONS environment variable (not recommended).
--heap-usage [T/F] Measure heap memory usage.
--io [T/F] Measure time spent in POSIX I/O calls.
--keep-inst-files [T/F] Don't remove instrumented files after compilation.
--memory-alloc [T/F] Record memory allocation/deallocation events and detect leaks.
--metadata-merge [T/F] Merge metadata of TAU profiles.
--metrics metric [metric ...] Performance metrics to gather, e.g. TIME, PAPI_FP_INS.
--profile [format] Generate application profiles.
- format: tau, merged, cubex, none
--reuse-inst-files [T/F] Reuse and preserve instrumented files after compilation.
--sample [T/F] Use event-based sampling to gather performance data.
--source-inst [mode] Use hooks inserted into the application source code to gather performance data.
- mode: automatic, manual, never
--throttle [T/F] Throttle lightweight events to reduce overhead.
--throttle-num-calls [count] Lightweight event call count threshold.
--throttle-per-call [us] Lightweight event duration threshold in microseconds.
--trace [format] Generate application traces.
- format: slog2, otf2, none
Project Arguments Description
--project-name name Name of the new project.
--storage-level levels Location of installation directory.
- levels: project, user, system
Target Arguments Description
--arch arch Host architecture.
- arch: x86_64, KNL, ppc64le, arm32, ppc64, BGL, KNC, BGP, BGQ, arm64, ibm64
--binutils (path|url|download|None) Path or URL to a GNU binutils installation or archive file.
--cc command Host C compiler command.
--compilers family Select all host compilers automatically from the given family, ignored if at least one host compiler is specified.
- family: Apple, BlueGene, Cray, GNU, IBM, Intel, PGI, System
--cuda-compilers family Select all CUDA compilers automatically from the given family, ignored if at least one CUDA compiler is specified.
- family: IBM, NVIDIA
--cuda-cxx command CUDA compiler command.
--cuda-fc command CUDA Fortran compiler command.
--cuda-toolkit path Path to NVIDIA CUDA Toolkit (enables OpenCL support).
--cxx command Host C++ compiler command.
--fc command Host Fortran compiler command.
--from-tau-makefile path Populate target configuration from a TAU Makefile (WARNING: Overrides safety checks).
--libunwind (path|url|download|None) Path or URL to a libunwind installation or archive file.
--mpi-cc command MPI C compiler command.
--mpi-compilers family Select all MPI compilers automatically from the given family, ignored if at least one MPI compiler is specified.
- family: Cray, IBM, Intel, System
--mpi-cxx command MPI C++ compiler command.
--mpi-fc command MPI Fortran compiler command.
--mpi-libraries flag [flag ...] Libraries to link to when building MPI applications.
--ompt (path|url|download|None) Path or URL to OMPT installation or archive file.
--os os Host operating system.
- os: CNL, Darwin, CNK, Android, Linux
--otf (path|url|download|None) Path or URL to libotf2 installation or archive file.
--papi (path|url|download|None) Path or URL to a PAPI installation or archive file.
--pdt (path|url|download|None) Path or URL to a PDT installation or archive file.
--scorep (path|url|download|None) Path or URL to a Score-P installation or archive file.
--shmem-cc command SHMEM C compiler command.
--shmem-compilers family Select all SHMEM compilers automatically from the given family, ignored if at least one SHMEM compiler is specified.
- family: Cray, OpenSHMEM, SOS
--shmem-cxx command SHMEM C++ compiler command.
--shmem-fc command SHMEM Fortran compiler command.
--shmem-libraries flag [flag ...] Libraries to link to when building SHMEM applications.
--target-name name Name of the new target configuration.
--tau (path|url|download|nightly) Path or URL to a TAU installation or archive file.
--upc command Universal Parallel C compiler command.

Previous: TAU Commander Help
Manual Home TOC
Next: TAU Commander Measurement