OpenACC, Intel OneAPI and Nvidia HPC Toolchain Support

New Features

• Allow the user to unset the CUDA and MPI compilers by setting them to None by @khsa1 in #406
• Intel OneAPI toolchain support implemented by @khsa1 in #403
• Support for profiling OpenACC codes and using the Nvidia HPC toolchain implemented by @vikram8128 in #416

Changes

• Fix release --all in setup.py by @zbeekman in #407
• Fix libdwarf requirements by @PlatinumCD in #411
• Add Analytica grant acknowledgement to README by @nchaimov in #408

Release Assets

For a complete set of ARCH/OS package combinations please visit https://taucommander.paratools.com/downloads

Full Changelog: v1.5.0...v1.6.0

Python 3 migration and SQLite support

New Features

• Support for TAU 2.31.
• Support for sqlite backend database as an alternative to TinyDB via setting the environment variable __TAUCMDR_DB_BACKEND__=sqlite or initializing a project with tau init --backend sqlite.
• Support for generating sqlite-format profiles by setting the profile attribute of a Measurement to sqlite.
• Software managed by TAU/TAU Commander is now mirrored on ParaTools servers and in the event of a failed download TAU Commander will fall back to the ParaTools hosted mirror.

Changes

• Internally TAU Commander uses python 3 instead of python 2, but can still be used for analysis of python 2.7 applications

Release Assets

For a complete set of ARCH/OS package combinations please visit https://taucommander.paratools.com/downloads

SC'19 Release

New features

• Support for TAU 2.29.
• Support for profiling of OpenMP work sharing regions using the OpenMP Tools Interface (OMPT) TR6.
• Support for profiling of Python codes.
• Support for stack unwinding when sampling.
• Support for profiling of Coarray Fortran codes.
• Support for memory footprint tracking.
• Support for armclang, armclang++, and armflang compilers

Changes

• Significant performance improvements when operating on a Project containing many Trials.

SC'17 Release

New features

• Runtime selective instrumentation.
• Support load tracking with TAU_TRACK_LOAD.
• Support jumbled command lines, e.g. tau list measurement is the same as tau measurement list.
• Specify OpenMP measurement library on tau init, e.g. tau init --openmp=ompt.
• New default measurements:
  • baseline: generate a profile file showing application wall clock time without instrumentation of any kind, i.e. zero overhead.
  • source-inst: automatic source-based instrumentation only.
  • compiler-inst: automatic compiler-based instrumentation only.

Changes

• "profile" and "trace" measurements to fall back to compiler-based inst
• tau gcc -h no longer shows TAU environment variables on stdout (compatibility with GNU autotools)
• Change --mpi-compilers to --mpi-wrappers
• Many bugs fixed.

Improved MPMD support and stability fixes.

Added support for sampling MPI MPMD codes and fixed several bugs.

Updated build/install procedure.

Improved build/install based on feedback from LLNL and Argonne.

Compiler-based instrumentation for CUDA

Developed in part at the 2017 ORNL Hackathon at NASA with help from the NASA High Performance Computing Incubator and the FUN3D team.

The 2017 ORNL Hackathon at NASA was a collaboration between and used resources
of both the National Aeronautics and Space Administration and the Oak Ridge Leadership
Computing Facility at Oak Ridge National Laboratory. Oak Ridge Nation Laboratory is
supported by the Office of Science of the U.S. Department of Energy under Contract No. DE-AC05-00OR22725.

Support for CORAL systems and IBM pre-release XL compilers.

Tested on SummitDev at ORNL, Ray at LLNL, and Theta at ALCF.

Native OTF2 and improved Cray KNL support.

v1.1.0

Initialize trace measurements to OTF2 by default.

v1.0.2

Fixes for stdout writers