DivL

DivL is a programming language and runtime designed to support scalable divide-and-conquer parallelism. It provides language constructs to express partitioning of data and work, and a runtime that schedules tasks across cores or cluster nodes, aiming to minimize data movement and synchronization overhead.

DivL was created by the DivL Project at the University of Tech in 2018, with a first public release in 2019. The project released an open-source compiler and runtime, and the name is a backronym for Divide and Learn, reflecting its emphasis on splitting tasks and combining results.

The language is statically typed and functional in style, with explicit parallel constructs such as partitioning primitives and futures. It includes a work-stealing scheduler, distributed execution support, and a memory model designed for locality. The standard library provides parallel collections and higher-order combinators, and there are foreign function interfaces to C and Python to facilitate integration with existing codebases.

The compiler translates DivL programs into a task graph, which the runtime executes across available hardware. DivL has been used in academic settings and in research prototypes to explore parallelism patterns that arise from dividing problems into smaller tasks. Benchmarking and case studies reported by the project focus on workloads with significant parallelism and data partitioning requirements.

Reception to DivL notes its potential for clear expression of divide-and-conquer strategies and locality-aware execution, while highlighting a relatively small ecosystem, a steeper learning curve, and debugging tools that are less mature than those for more established languages.

See also: List of parallel programming languages, dataflow languages, task-based parallelism.