Entanglement detection with near-zero cost

Recent research on parallel functional programming has culminated in a provably efficient (in work and space) memory manager, which been incorporated into the MPL (MaPLe) compiler for Parallel ML shown to deliver practical efficiency scalability. The manager exploits property of programs called disentanglement, restricts computations from accessing concurrently allocated objects. Disentanglement is closely related race-freedom, but subtly differs it. Unlike however, no known techniques exists ensuring leaving task entirely programmer. This challenging task, because it requires reasoning about low-level operations (e.g., allocations accesses), especially difficult languages. In this paper, we present detecting entanglement dynamically, while program running. We first dynamic semantics language with references that checks by consulting sequential dependency relations program. Notably, mutable objects only. prove soundness several realizing efficiently, particular pruning away large number checks. also provide bounds space our techniques. show detection are implementing them ML. Considering variety benchmarks, an evaluation measure time overheads less than 5% average up 72 cores. These results negligible cost can therefore remain deployed little or impact efficiency, scalability, space.