We work every day to make parallel programming faster and easier, so we are excited to introduce you to the new features of Parallware Trainer 0.5 version. Take a look at the big changes!
- User can now select among three implementations of parallel scalar/sparse reductions:
- Built-in reductions (e.g. OpenMP/OpenACC reduction clause)
- Atomic access (e.g. OpenMP/OpenACC atomic construct)
- Explicit privatization and synchronization of reduction variables (e.g. combination of OpenMP private, barrier and critical/atomic)
- Improved GUI with new User Action LIst to support OpenMP/OpenACC-directive parallel templates
- New follow-on actions to fill in the template (e.g. CPU-GPU transfer ranges using OpenMP map to/map from and OpenACC copyin/copyout).
- New requirements to guarantee correctness (e.g. assert non-aliasing with restrict)
- New suggestions to improve performance (e.g. remove atomic if not needed)
- Improved GUI with a new dialog window for the creation of Parallware pragmas
- Interactive on-the-fly syntax validation
- Clang-based reengineering of the Parallware driver for C/C++
- GCC/Clang-compatible compilation flags (e.g. -D ,-I, -std)
- Clang-based reengineering of Parallware front-end for C/C++
- Improved support for preprocessor macros
- Improved validation of Parallware pragma parameters
- New documentation for the specification of Parallware pragmas
- Bugfixes in GUI
In order to try Parallware Trainer, you must register for the Early Access Program.
Users will have full and free access to the tool for a month and will be able to learn parallel programming while improving their code. We look forward to receiving feedback from the participants on the usability and parallelization capabilities of Parallware Trainer.
✉ Get all our updates by subscribing to our newsletter.