Implicitly parallel task-based runtime systems often perform dynamic analysis to discover dependencies in and extract parallelism from sequential programs. Dependence analysis becomes expensive as task granularity drops below a threshold. Tracing techniques have been developed where programmers annotate repeated program fragments (traces) issued by the application, and the runtime system memoizes the dependence analysis for those fragments, greatly reducing overhead when the fragments are executed again. However, manual trace annotation can be brittle and not easily applicable to complex programs built through the composition of independent components. We introduce Apophenia, a system that automatically traces the dependence analysis of task-based runtime systems, removing the burden of manual annotations from programmers and enabling new and complex programs to be traced. Apophenia identifies traces dynamically through a series of dynamic string analyses, which find repeated program fragments in the stream of tasks issued to the runtime system. We show that Apophenia is able to come between 0.92x--1.03x the performance of manually traced programs, and is able to effectively trace previously untraced programs to yield speedups of between 0.91x--2.82x on the Perlmutter and Eos supercomputers.

We introduce Diffuse, a system that dynamically performs task and kernel fusion in distributed, task-based runtime systems. The key component of Diffuse is an intermediate representation of distributed computation that enables the necessary analyses for the fusion of distributed tasks to be performed in a scalable manner. We pair task fusion with a JIT compiler to fuse together the kernels within fused tasks. We show empirically that Diffuse's intermediate representation is general enough to be a target for two real-world, task-based libraries (cuPyNumeric and Legate Sparse), letting Diffuse find optimization opportunities across function and library boundaries. Diffuse accelerates unmodified applications developed by composing task-based libraries by 1.86x on average (geo-mean), and by between 0.93x--10.7x on up to 128 GPUs. Diffuse also finds optimization opportunities missed by the original application developers, enabling high-level Python programs to match or exceed the performance of an explicitly parallel MPI library.

Modern scientific discovery increasingly relies on high-performance computing for complex modeling and simulation. A key challenge in improving parallel program performance is efficiently mapping tasks to processors and data to memory, a process dictated by intricate, low-level system code known as mappers. Developing high-performance mappers demands days of manual tuning, posing a significant barrier for domain scientists without systems expertise. We introduce a framework that automates mapper development with generative optimization, leveraging richer feedback beyond scalar performance metrics. Our approach features the Agent-System Interface, which includes a Domain-Specific Language (DSL) to abstract away the low-level complexity of system code and define a structured search space, as well as AutoGuide, a mechanism that interprets raw execution output into actionable feedback. Unlike traditional reinforcement learning methods such as OpenTuner, which rely solely on scalar feedback, our method finds superior mappers in far fewer iterations. With just 10 iterations, it outperforms OpenTuner even after 1000 iterations, achieving $3.8\times$ faster performance. Our approach finds mappers that surpass expert-written mappers by up to $1.34\times$ speedup across nine benchmarks while reducing tuning time from days to minutes.

Domain-specific, fixed-function units are becoming increasingly common in modern processors. As the computational demands of applications evolve, the capabilities and programming interfaces of these fixed-function units continue to change. NVIDIA’s Hopper GPU architecture contains multiple fixed-function units per compute unit, including an asynchronous data movement unit (TMA) and an asynchronous matrix multiplication unit (Tensor Core). Efficiently utilizing these units requires a fundamentally different programming style than previous architectures; programmers must now develop warp-specialized kernels that orchestrate producer consumer pipelines between the asynchronous units. To manage the complexity of programming these new architectures, we introduce Cypress, a task-based programming model with sequential semantics. Cypress programs are a set of designated functions called tasks that operate on tensors and are free of communication and synchronization. Cypress programs are bound to the target machine through a mapping specification that describes where tasks should run and in which memories tensors should be materialized. We present a compiler architecture that lowers Cypress programs into CUDA programs that perform competitively with expert-written codes. Cypress achieves 0.88x-1.06x the performance of cuBLAS on GEMM, and between 0.80x-0.98x the performance of the currently best-known Flash Attention implementation while eliminating all aspects of explicit data movement and asynchronous computation from application code.

In a parallel and distributed application, a mapping is a selection of a processor for each computation or task and memories for the data collections that each task accesses. Finding high-performance mappings is challenging, particularly on heterogeneous hardware with multiple choices for processors and memories. We show that fast mappings are sensitive to the machine, application, and input. Porting to a new machine, modifying the application, or using a different input size may necessitate re-tuning the mapping to maintain the best possible performance.

The sparse module of the popular SciPy Python library is widely used across applications in scientific computing, data analysis and machine learning. The standard implementation of SciPy is restricted to a single CPU and cannot take advantage of modern distributed and accelerated computing resources. We introduce Legate Sparse, a system that transparently distributes and accelerates unmodified sparse matrix-based SciPy programs across clusters of CPUs and GPUs, and composes with cuNumeric, a distributed NumPy library. Legate Sparse uses a combination of static and dynamic techniques to efficiently compose independently written sparse and dense array programming libraries, providing a unified Python interface for distributed sparse and dense array computations. We show that Legate Sparse is competitive with single-GPU libraries like CuPy and achieves 65% of the performance of PETSc on up to 1280 CPU cores and 192 GPUs of the Summit supercomputer, while offering the productivity benefits of idiomatic SciPy and NumPy.

We introduce DISTAL, a compiler for dense tensor algebra that targets modern distributed and heterogeneous systems. DISTAL lets users independently describe how tensors and computation map onto target machines through separate format and scheduling languages. The combination of choices for data and computation distribution creates a large design space that includes many algorithms from both the past (e.g., Cannon’s algorithm) and the present (e.g., COSMA). DISTAL compiles a tensor algebra domain specific language to a distributed task-based runtime system and supports nodes with multi-core CPUs and multiple GPUs. Code generated by is competitive with optimized codes for matrix multiply on 256 nodes of the Lassen supercomputer and outperforms existing systems by between 1.8x to 3.7x (with a 45.7x outlier) on higher order tensor operations.

We introduce SpDISTAL, a compiler for sparse tensor algebra that targets distributed systems. SpDISTAL combines separate descriptions of tensor algebra expressions, sparse data structures, data distribution, and computation distribution. Thus, it enables distributed execution of sparse tensor algebra expressions with a wide variety of sparse data structures and data distributions. SpDISTAL is implemented as a C++ library that targets a distributed task-based runtime system and can generate code for nodes with both multi-core CPUs and multiple GPUs. SpDISTAL generates distributed code that achieves performance competitive with hand-written distributed functions for specific sparse tensor algebra expressions and that outperforms general interpretation-based systems by one to two orders of magnitude.

This paper shows how to compile sparse array programming languages. A sparse array programming language is an array programming language that supports element-wise application, reduction, and broadcasting of arbitrary functions over dense and sparse arrays with any fill value. Such a language has great expressive power and can express sparse and dense linear and tensor algebra, functions over images, exclusion and inclusion filters, and even graph algorithms. Our compiler strategy generalizes prior work in the literature on sparse tensor algebra compilation to support any function applied to sparse arrays, instead of only addition and multiplication. To achieve this, we generalize the notion of sparse iteration spaces beyond intersections and unions. These iteration spaces are automatically derived by considering how algebraic properties annotated onto functions interact with the fill values of the arrays. We then show how to compile these iteration spaces to efficient code. When compared with two widely-used Python sparse array packages, our evaluation shows that we generate built-in sparse array library features with a performance of 1.4× to 53.7× when measured against PyData/Sparse for user-defined functions and between 0.98× and 5.53× when measured against SciPy/Sparse for sparse array slicing. Our technique outperforms PyData/Sparse by 6.58× to 70.3×, and (where applicable) performs between 0.96× and 28.9× that of a dense NumPy implementation, on end-to-end sparse array applications. We also implement graph linear algebra kernels in our system with a performance of between 0.56× and 3.50× compared to that of the hand-optimized SuiteSparse:GraphBLAS library.

Nested parallelism has proved to be a popular approach for programming the rapidly expanding range of multicore computers. It allows programmers to express parallelism at a high level and relies on a run-time system and a scheduler to deliver efficiency and scalability. As a result, many programming languages and extensions that support nested parallelism have been developed, including in C/C++, Java, Haskell, and ML. Yet, writing efficient and scalable nested parallel programs remains challenging, primarily due to difficult concurrency bugs arising from destructive updates or effects. For decades, researchers have argued that functional programming can simplify writing parallel programs by allowing more control over effects but functional programs continue to underperform in comparison to parallel programs written in lower-level languages. The fundamental difficulty with functional languages is that they have high demand for memory, and this demand only grows with parallelism. In this paper, we identify a memory property, called disentanglement, of nested-parallel programs, and propose memory management techniques for improved efficiency and scalability. Disentanglement allows for (destructive) effects as long as concurrently executing threads do not gain knowledge of the memory objects allocated by each other. We formally define disentanglement by considering an ML-like higher-order language with mutable references and presenting a dynamic semantics for it that enables reasoning about computation graphs of nested parallel programs. Based on this graph semantics, we formalize a classic correctness property---determinacy race freedom---and prove that it implies disentanglement. This establishes that disentanglement applies to a relatively broad class of parallel programs. We then propose memory management techniques for nested-parallel programs that take advantage of disentanglement for improved efficiency and scalability. We show that these techniques are practical by extending the MLton compiler for Standard ML to support this form of nested parallelism. Our empirical evaluation shows that our techniques are efficient and scale well.

Subgraph isomorphism is a fundamental property of graphs that requires checking whether the network structure of one graph can be found (embedded) within another graph. It has numerous applications and is a computationally challenging problem: it is NP-complete and known algorithms explore an exponentially large search space. Even though it has been studied extensively, relatively little is known about whether subgraph isomorphism accepts a theoretically and practically efficient parallel solution. In this paper, we present our ongoing work on designing a parallel algorithm for the subgraph (and graph) isomorphism problem, which addresses challenges commonly faced when attempting to obtain a parallel algorithm for isomorphism. Our algorithm appears to scale well up to the 70 cores of our empirical machine.