Publications by Mr Michael Vollmer

Recent work showed that compiling functional programs to use dense, serialized memory representations for recursive algebraic datatypes can yield significant constant-factor speedups for sequential programs. But serializing data in a maximally dense format consequently serializes the processing of that data, yielding a tension between density and parallelism. This paper shows that a disciplined, practical compromise is possible. We present Parallel Gibbon, a compiler that obtains the benefits of dense data formats and parallelism. We formalize the semantics of the parallel location calculus underpinning this novel implementation strategy, and show that it is type-safe. Parallel Gibbon exceeds the parallel performance of existing compilers for purely functional programs that use recursive algebraic datatypes, including, notably, abstract-syntax-tree traversals as in compilers.

A core, but often neglected, aspect of a programming language design is its memory (consistency) model. Sequential consistency~(SC) is the most intuitive memory model for programmers as it guarantees sequential composition of instructions and provides a simple abstraction of shared memory as a single global store with atomic read and writes. Unfortunately, SC is widely considered to be impractical due to its associated performance overheads. Perhaps contrary to popular opinion, this paper demonstrates that SC is achievable with acceptable performance overheads for mainstream languages that minimize mutable shared heap. In particular, we modify the Glasgow Haskell Compiler to insert fences on all writes to shared mutable memory accessed in nonfunctional parts of the program. For a benchmark suite containing 1,279 programs, SC adds a geomean overhead of less than 0.4 on an x86 machine. The efficiency of SC arises primarily due to the isolation provided by the Haskell type system between purely functional and thread-local imperative computations on the one hand, and imperative computations on the global heap on the other. We show how to use new programming idioms to further reduce the SC overhead; these create a virtuous cycle of less overhead and even stronger semantic guarantees (static data-race freedom).

While programmers know that the low-level memory representation of data structures can have significant effects on performance, compiler support to optimize the layout of those structures is an under-explored field. Prior work has optimized the layout of individual, non-recursive structures without considering how collections of those objects in linked or recursive data structures are laid out. This work introduces Marmoset, a compiler that optimizes the layouts of algebraic datatypes, with a special focus on producing highly optimized, packed data layouts where recursive structures can be traversed with minimal pointer chasing. Marmoset performs an analysis of how a recursive ADT is used across functions to choose a global layout that promotes simple, strided access for that ADT in memory. It does so by building and solving a constraint system to minimize an abstract cost model, yielding a predicted efficient layout for the ADT. Marmoset then builds on top of Gibbon, a prior compiler for packed, mostly-serial representations, to synthesize optimized ADTs. We show experimentally that Marmoset is able to choose optimal layouts across a series of microbenchmarks and case studies, outperforming both Gibbons baseline approach, as well as MLton, a Standard ML compiler that uses traditional pointer-heavy representations.

Over the years, traditional tracing garbage collectors have accumulated assumptions that may not hold in new language designs. For instance, we usually assume that run-time objects do not hold addressable sub-parts and have a size of at least one pointer. These fail in systems striving to eliminate pointers and represent data in a dense, serialized form, such as the Gibbon compiler. We propose a new memory management strategy for language runtimes with mostly serialized heaps. It uses a hybrid, generational collector, where regions are bump-allocated into the young generation and objects are bump-allocated within those regions. Minor collections copy data into larger regions in the old generation, compacting it further. The old generation uses region-level reference counting. The resulting system maintains high performance for data traversal programs, while significantly improving performance on other kinds of allocation patterns.

Generic programming with recursion schemes provides a powerful abstraction for structuring recursion while ensuring termination and providing reasoning about program equivalences, as well as deriving optimisations which have been successfully applied to functional programming. Formalising recursion schemes in a type theory offers additional termination guarantees, but it often requires compromises affecting the resulting code, such as imposing performance penalties, requiring the assumption of additional axioms, or introducing unsafe casts into extracted code (e.g. Obj.magic in OCaml). To the best of our knowledge, this paper presents the first Coq formalisation of a recursion scheme, called the hylomorphism, along with its algebraic laws allowing for the mechanisation of all recognised (terminating) recursive algorithms. The key contribution of this paper is that this formalisation is fully axiom-free allowing for the extraction of safe, idiomatic OCaml code. We exemplify the framework by formalising a series of algorithms based on different recursive paradigms such as divide-and conquer, dynamic programming, and mutual recursion and demonstrate that the extracted OCaml code for the programs formalised in our framework is efficient, resembles code that a human programmer would write, and contains no occurrences of Obj.magic. We also present a machine-checked proof of the well-known short-cut fusion optimisation.

Memory safety issues are a serious concern in systems programming. Rust is a systems language that provides memory safety through a combination of a static checks embodied in the type system and ad hoc dynamic checks inserted where this analysis becomes impractical. The Morello prototype architecture from ARM uses capabilities, fat pointers augmented with object bounds information, to catch failures of memory safety. This paper presents a compiler from Rust to the Morello architecture, together with a comparison of the performance of Rust's runtime safety checks and the hardware-supported checks of Morello. The cost of Morello's always-on memory safety guarantees is 39% in our 19 benchmark suites from the Rust crates repository (comprising 870 total benchmarks). For this cost, Morello's capabilities ensure that even unsafe Rust code benefits from memory safety guarantees.

Substructural type systems are growing in popularity because they allow for a resourceful interpretation of data which can be used to rule out various software bugs. Indeed, substructurality is finally taking hold in modern programming; Haskell now has linear types roughly based on Girard's linear logic but integrated via graded function arrows, Clean has uniqueness types designed to ensure that values have at most a single reference to them, and Rust has an intricate ownership system for guaranteeing memory safety. But despite this broad range of resourceful type systems, there is comparatively little understanding of their relative strengths and weaknesses or whether their underlying frameworks can be unified. There is often confusion about whether linearity and uniqueness are essentially the same, or are instead 'dual' to one another, or somewhere in between. This paper formalises the relationship between these two well-studied but rarely contrasted ideas, building on two distinct bodies of literature, showing that it is possible and advantageous to have both linear and unique types in the same type system. We study the guarantees of the resulting system and provide a practical implementation in the graded modal setting of the Granule language, adding a third kind of modality alongside coeffect and effect modalities. We then demonstrate via a benchmark that our implementation benefits from expected efficiency gains enabled by adding uniqueness to a language that already has a linear basis.

In a typical data-processing program, the representation of data in memory is distinct from its representation in a serialized form on disk. The former has pointers and arbitrary, sparse layout, facilitating easy manipulation by a program, while the latter is packed contiguously, facilitating easy I/O. We propose a language, LoCal, to unify in-memory and serialized formats. LoCal extends a region calculus into a location calculus, employing a type system that tracks the byte-addressed layout of all heap values. We formalize LoCal and prove type safety, and show how LoCal programs can be inferred from unannotated source terms. We transform the existing Gibbon compiler to use LoCal as an intermediate language, with the goal of achieving a balance between code speed and data compactness by introducing just enough indirection into heap layouts, preserving the asymptotic complexity of traditional representations, but working with mostly or completely serialized data. We show that our approach yields significant performance improvement over prior approaches to operating on packed data, without abandoning idiomatic programming with recursive functions.

When written idiomatically in most programming languages, programs that traverse and construct trees operate over pointer-based data structures, using one heap object per-leaf and per-node. This representation is efficient for random access and shape-changing modifications, but for traversals, such as compiler passes, that process most or all of a tree in bulk, it can be inefficient. In this work we instead compile tree traversals to operate on pointer-free pre-order serializations of trees. On modern architectures such programs often run significantly faster than their pointer-based counterparts, and additionally are directly suited to storage and transmission without requiring marshaling. We present a prototype compiler, Gibbon, that compiles a small first-order, purely functional language sufficient for tree traversals. The compiler transforms this language into intermediate representation with explicit pointers into input and output buffers for packed data. The key compiler technologies include an effect system for capturing traversal behavior, combined with an algorithm to insert destination cursors. We evaluate our compiler on tree transformations over a real-world dataset of source-code syntax trees. For traversals touching the whole tree, such as maps and folds, packed data allows speedups of over 2x compared to a highly-optimized pointer-based baseline.

High-level domain-specific languages for array processing on the GPU are increasingly common, but they typically only run on a single GPU. As computational power is distributed across more devices, languages must target multiple devices simultaneously. To this end, we present a compositional translation that fissions data-parallel programs in the Accelerate language, allowing subsequent compiler and runtime stages to map computations onto multiple devices for improved performance---even programs that begin as a single data-parallel kernel.

Writing high performance GPGPU code is often difficult and time-consuming, potentially requiring laborious manual tuning of low-level details. Despite these challenges, the cost in ignoring GPUs in high performance computing is increasingly large. Auto-tuning is a potential solution to the problem of tedious manual tuning. We present a framework for auto-tuning GPU kernels which are expressed in an embedded DSL, and which expose compile-time parameters for tuning. Our framework allows for kernels to be polymorphic over what search strategy will tune them, and allows search strategies to be implemented in the same meta-language as the kernel-generation code (Haskell). Further, we show how to use functional programming abstractions to enforce regular (hyper-rectangular) search spaces. We also evaluate several common search strategies on a variety of kernels, and demonstrate that the framework can tune both EDSL and ordinary CUDA code.