ICFP 2022
Sun 11 - Fri 16 September 2022 Ljubljana, Slovenia
Thu 15 Sep 2022 09:00 - 09:50 at Štih - Language Design Chair(s): Benoît Montagu

Researchers have argued for decades that functional programming simplifies parallel programming, in particular by helping programmers avoid difficult concurrency bugs arising from destructive in-place updates. However, parallel functional languages have historically underperformed in comparison to parallel programs written in lower-level languages. The difficulty is that functional programs have high demand for memory, and this demand only grows with parallelism, causing traditional parallel memory management techniques to buckle under the increased pressure.

Recent work has made progress on this problem by identifying a broadly applicable memory property called disentanglement. To exploit disentanglement for improved efficiency and scalability, we show how to partition memory into a tree of heaps, mirroring the dynamic nesting of parallel tasks. This design allows for task-local allocations and garbage collections to proceed independently and in parallel. The result is a provably efficient parallel memory manager.

These ideas have been incorporated into the MPL (“maple") compiler for Parallel ML, which offers practical efficiency and scalability for parallel functional programs. Our empirical evaluations show that, at scale (on 72 processors), MPL outperforms modern implementations of both functional and imperative languages, including Java and Go. Additionally, we show that MPL is competitive with low-level, memory-unsafe languages such as C++, in terms of both space and time.

Thu 15 Sep

Displayed time zone: Belgrade, Bratislava, Budapest, Ljubljana, Prague change

09:00 - 10:30
Language DesignML at Štih
Chair(s): Benoît Montagu Inria
09:00
50m
Keynote
Efficient and Scalable Parallel Functional Programming Through Disentanglement
ML
Sam Westrick Carnegie Mellon University
09:50
20m
Talk
Towards Algebraic Subtyping for Extensible Records
ML
Rodrigo Marques Universidade do Porto, Mário Florido Universidade do Porto, Pedro Vasconcelos LIACC, Universidade do Porto, Porto, Portugal
10:10
20m
Talk
The Ultimate Conditional SyntaxVirtual
ML
Lionel Parreaux The Hong Kong University of Science and Technology (HKUST)
Pre-print File Attached