Numerical Representation of Directed Acyclic Graphs for Efficient Dataflow Embedded Resource Allocation

Abstract

Stream processing applications running on Heterogeneous Multi-Processor Systems on Chips (sHMPSoCs) require efficient resource allocation and management, both at compile-time and at runtime. To cope with modern adaptive applications whose behavior can not be exhaustively predicted at compile-time, runtime managers must be able to take resource allocation decisions on-the-fly, with a minimum overhead on application performance. Resource allocation algorithms often rely on an internal modeling of an application. Directed Acyclic Graphs (sDAGs) are the most commonly used models for capturing control and data dependencies between tasks. DAGs are notably often used as an intermediate representation for deploying applications modeled with a dataflow Model of Computation (MoC) on HMPSoCs. Building such intermediate representation at runtime for massively parallel applications is costly both in terms of computation and memory overhead. In this paper, an intermediate representation of DAGs for resource allocation is presented. This new representation shows improved performance for run-time analysis of dataflow graphs with less overhead in both computation time and memory footprint. The performances of the proposed representation are evaluated on a set of computer vision and machine learning applications.

Runtime multi-versioning and specialization inside a memoized speculative loop optimizer

“…The benchmarks were run on two hardware platforms: The set of benchmark programs has been built from two different benchmarks suites: the Apollo benchmarks 4 and a selection of loop kernels from the Polybench benchmark suite 5 . In total, 20 different benchmark programs have been used to evaluate our framework: 11 from Polybench and 9 from Apollo, including spmatmat with two different configurations related to the shape of the input matrices.…”

“…This kind of applications can be specified with a dataflow Model of Computation (MoC), which is a directed graph of nodes representing computations, the so-called actors, and arcs, which represent First-In First-Out (FIFO) queues where data tokens interchanged between actors are stored [18]. Due to their natural expressivity of task parallelism and analyzability, dataflow MoCs are increasingly popular [4]. Dataflow MoCs are especially useful for applications performing the same processing to a stream of data; consequently, they run on a loop, and hence they invoke actor kernels iteratively.…”