Tau

Installed versions

Resource	Version
Dardel/cpe24.11	2.34

General information

The University of Oregon’s TAU is an open-source portable profiling and tracing toolkit for the performance analysis of parallel programs written in Fortran, C, C++, UPC, Java, Python and other languages. The instrumentation of programs can be done by binary code rewriting, manual compiler directives, or automatic source code transformation. TAU can support many parallel programming interfaces, including MPI, OpenMP, pthreads, and ROCm. TAU includes paraprof which is a profile visualization tool, and generated execution traces can be displayed by the Vampir, Paraver or JumpShot (included) visualization tools.

https://www.cs.uoregon.edu/research/tau/docs.phpx

How to use

Load the following modules

$ module load PrgEnv-gnu
$ module load PDC
$ module load tau

Compile your code using tau wrappers, such as tau_cc, tau_cxx, tau_f90/tau_f77. For instance, to compile a C++ code, the following command can be used:

tau_cxx.sh <myProgran.cpp> -o <myProgramEx>

Tracing and profiling are turned on by default when the module is loaded. If you'd like to disable one of them, the environment variables export TAU_TRACE=0 and/or export TAU_PROFILE=0 can be set this way respectively. Further, the location for storing the profiling, tracing data can be set using:

export PROFILEDIR=<path-to-profiling-data-directory>
export TRACEDIR=<path-to-tracing-data-directory>

Then, run the program binary as follows:

srun <flags-as-usual> tau_exec <binary>

After the program completes its run successfully, navigate to the profile (cd $PROFILEDIR)data directory and run the following command pprof to generate a text output, or paraprof to generate graphical output.

NOTE: If your program uses OpenMP based parallelism, make sure to use --hint=nomultithread along with the srun command