Streaming irregular arrays

Abstract

Previous work has demonstrated that it is possible to generate efficient and highly parallel code for multicore CPUs and GPUs from combinator-based array languages for a range of applications. That work, however, has been limited to operating on flat, rectangular structures without any facilities for irregularity or nesting. In this paper, we show that even a limited form of nesting provides substantial benefits both in terms of the expressiveness of the language (increasing modularity and providing support for simple irregular structures) and the portability of the code (increasing portability across resource-constrained devices, such as GPUs). Specifically, we generalise Blelloch's flattening transformation along two lines: (1) we explicitly distinguish between definitely regular and potentially irregular computations; and (2) we handle multidimensional arrays. We demonstrate the utility of this generalisation by an extension of the embedded array language Accelerate to include irregular streams of multidimensional arrays. We discuss code generation, optimisation, and irregular stream scheduling as well as a range of benchmarks on both multicore CPUs and GPUs.

Embedded Pattern Matching

“…We implemented embedded pattern matching in Accelerate [Chakravarty et al 2011;Clifton-Everest et al 2017;McDonell et al 2015McDonell et al , 2013van den Haak et al 2020], an open source language embedded in Haskell for data-parallel array computations. Accelerate code embedded in Haskell is not compiled by the Haskell compiler: instead, the Accelerate library includes a runtime compiler which generates and compiles parallel SIMD or GPU code at application runtime.…”

Generating fast sparse matrix vector multiplication from a high level generic functional IR

“…Generic high-level code generators have to stick to generic data representations. In some high performance code generators, such as Accelerate [15], runtime information is injected in the data structure directly to model irregularity [7]. While this approach is very flexible, it introduces runtime computation that could be avoided, and it limits somewhat the range of possible sparse formats expressed.…”

“…Streaming irregular arrays [7] represent a solution for expressing sparse data structure in the high-level code generator Accelerate [15]. Similarly to our approach, it use typing information to model the shape of the sparse data structure, but is more reliant on runtime support.…”