Breakout Groups

Day 1 (Aug/4)

Day 2 (Aug/5)

Charge Questions

1. What are the current areas of focus and successes in program synthesis within the research community and industry?
2. How should these focus areas and successes be expanded to encompass the needs of scientific computing and HPC?
3. What are the challenges in applying program synthesis to problems in scientific computing and HPC?
4. What opportunities exist for applying existing successful approaches in scientific computing and HPC?
5. What are aggressive short term (1-2 years), medium term (3-5 years), and long term (> 5 years) goals?

Charge Questions (applications sections during breakout 1):

1. What are the most labor-intensive aspects of programming scientific applications/libraries?
2. What are the most technically difficult aspects of programming scientific applications/libraries?
3. What capabilities would a software system need to possess in order to largely, or completely, automate these tasks? What are the required inputs? How would the system interact with the human programmers?

Charge Questions (compiler and tools sections during breakout 1):

1. What opportunities exist to improve compilers and program analysis tools to better support scientific applications and HPC?
2. How might higher-level information be leveraged to capture those opportunities?
3. How might compiler technology be improved to better integrate with program-synthesis systems?

Additional Prompt Questions

In addition to these general charge questions, here are a number of challenge questions highlighted by our attendees in the pre-workshop questionnaire that might serve as useful points of discussion in the breakout sessions:

1. How can we make program-synthesis tools scalable?
2. How can we make program-synthesis tools composible?
3. How can we make use of natural language in program-synthesis systems in this area?
4. How can high-level specifications drive program synthesis in this area?
5. How can data-driven program synthesis be applied to this area?
6. How can we make use of deep learning in program-synthesis systems?
7. How should program-synthesis systems handle flawed and ambiguous specifications?
8. How can we make program-synthesis systems integrate with IDEs and otherwise have useful interfaces?
9. How can we construct general program synthesis systems (i.e., systems not restricted to a small, domain-specific set of primitives)?
10. How can we most-effectively use domain-specific languages with program-synthesis systems?
11. How can we access the productivity improvements from program-synthesis systems and use that data to improve those systems?
12. How can we create program-synthesis systems that can create parallelization strategies and address algorithmic-complexity metrics?
13. How can we select test cases, or create test cases, useful for training and benchmarking program-synthesis systems?
14. How can program-synthesis systems assist with hardware/software co-design?
15. How can program-synthesis systems complement programming environments featuring runtime specialization (e.g., just-in-time compilation)?
16. How can program-synthesis systems help address performance-portability requirements?
17. How can program-synthesis systems contribute to increasing software reliability?
18. How can program-synthesis systems most-usefully provide formal correctness guarantees?