OpenACC provides users with a directive-based way to write programs that will be able to take advantage of accelerators such as the NVIDIA Tesla and Fermi GPUs. Similar to the popular OpenMP directives for multithreading applications to take advantage of multicore, OpenACC provides users with a way to work incrementally to add acceleration to an already existing C/C++ or Fortran program. OpenACC is supported by Rogue Wave partners: NVIDIA, PGI, and Cray.
TotalView provides full support for debugging across both the host and accelerator in OpenACC programs compiled with the Cray CCE 8 compiler. We encourage developers to try this out and report your experiences. We plan to offer official support for this feature in the near future.
Features supported include:
- Setting breakpoints in OpenACC outlined code
- Stepping at the warp level in OpenACC outlined code
- Display of OpenACC variables
- Ability to navigate the device threads by device, sm, warp, and lane