Efficient Data Supply for Parallel Heterogeneous Architectures

Abstract

Decoupling techniques have been proposed to reduce the amount of memory latency exposed to highperformance accelerators as they fetch data. Although decoupled access-execute (DAE) and more recent decoupled data supply approaches offer promising single-threaded performance improvements, little work has considered how to extend them into parallel scenarios. This article explores the opportunities and challenges of designing parallel, high-performance, resource-efficient decoupled data supply systems. We propose Mercury, a parallel decoupled data supply system that utilizes thread-level parallelism for high-throughput data supply with good portability attributes. Additionally, we introduce some microarchitectural improvements for data supply units to efficiently handle long-latency indirect loads. CCS Concepts: • Computer systems organization → Heterogeneous (hybrid) systems; Parallel architectures;

Tiny but mighty

“…area-constrained environments tend to use simple in-order cores. (3) Techniques that rely on ISA extensions [23] or ISA-specific instructions have limited applicability and portability problems, especially in the context of heterogeneous-ISA architectures [7,34]. (4) Hardware-only techniques like Slipstream [52,54] or hardware prefetching [1,26] often require costly structures for book-keeping, detection, and prediction.…”

“…This also allows a process to decide at runtime which MAPLE unit to target. As we introduced before, previous approaches [22,23,49] do not offer this software programmability of decouplinghardware resources, as these are tightly connected to specific cores.…”