Performance tuning is a hard problem, especially for
high-performance scientific programs. ParallelAccelerator
is a step toward bringing good parallel performance to
Julia by utilizing both domain-specific and general compiler and run-time
optimization techniques without much effort from the user. However,
eventually there will be bottlenecks, and not all optimizations work in
favor of each other. ParallelAccelerator is still a proof-of-concept
at this stage, and feedback from users will help us improve the system.
We welcome bug reports and code contributions.
Please feel free to use our system, fork the project, and
file bug reports on our GitHub issue tracker.