- Open action Items
- Topics for 2024
- Release v2.0 (Kris)
- Documentation
- Concurrency examples
- Examples of low/mixed-precision usage in CUDA, HIP, SYCL, etc.
- Document existing (host and kernel) macros
- GitHub discussion thread for feedback on feature test macros
- Create/find examples of atomic usage in CUDA, HIP, SYCL, etc.
- Create a mailing list
- Walkthrough of new OKL transpiler
- Walkthrough of new OCCA Tools API
- Mini-apps (more involved than existing examples, but less than full applications)
- Demonstrate some common parallel/solver patterns
- Tim existing examples from courses, workshops
- Q: Where should these examples live?
- Custom data types (for use in OKL kernels)
- Currently need to explicitly manage all device memory in custom data types
- Ideally users could register non-trivial data types and OCCA runtime would manage
- Integrating new features of backend programming models
- Device-side library for warp/block level collectives
- Matrix/tensor core instructions
- Q: Should we form a WG for some of these topics?
- User survey (e.g., which backends are used?)
- How many users of downstream applications?
- How many write OKL kernels?
- How many developers of OCCA applications?
- How many involved in OCCA development?
- How many users of downstream applications?
- Papers?