Current Work and Future Plans

Current work is being carried out by developers at AMD. Contributors can pick action items from the Future Plans list in which case they can open an issue with an "enhancement" tag. Contributions for optimizing performance of Bolt routines are welcome.

Current Work

1. Performance optimizations for routines.
   • Sort family
   • Scan family
2. C++ AMP support to achieve parity with OpenCL routines.
3. Iterators Support:
   • Permutation iterators
   • Zip iterators

Future Plans

1. Support for Mac OS.
2. List more functions to be implemented. Currently in BOLT we group functions in the following family of routines: Prefix Sum, Reductions, Sorting, Transformations, Copying, Searching, Merging, Re-ordering etc. These group are not functionally complete yet and will be taken up in the future releases, to achieve functional parity with thrust library.
3. Multi Device support. AMD is a Founder member of the HSA(Heterogeneous System Architecture) foundation and is considering to support load balancing across various devices in the platform. Feedback on this feature is being sought from the developer community.
4. OpenCL kernel binaries Creating a method to save Bolt OpenCL kernel binaries, such that Bolt functions can run without kernel compiles.
5. Asynchronous Execution.