Multiple Target Task Sharing Support for the OpenMP Accelerator Model

Abstract

The use of GPU accelerators is becoming common in HPC platforms due to the their effective performance and energy efficiency. In addition, new generations of multicore processors are being designed with wider vector units and/or larger hardware thread counts, also contributing to the peak performance of the whole system. Although current directive-based paradigms, such as OpenMP or OpenACC, support both accelerators and multicore-based hosts, they do not provide an effective and efficient way to concurrently use them, usually resulting in accelerated programs in which the potential computational performance of the host is not exploited. In this paper we propose an extension to the OpenMP 4.5 directive-based programming model to support the specification and execution of multiple instances of task regions on different devices (i.e. accelerators in conjunction with the vector and heavily multithreaded capabilities in multicore processors). The compiler is responsible for the generation of device-specific code for each device kind, delegating to the runtime system the dynamic schedule of the tasks to the available devices. The new proposed clause conveys useful insight to guide the scheduler while keeping a clean, abstract and machine independent programmer interface. The potential of the proposal is analyzed in a prototype implementation in the OmpSs compiler and runtime infrastructure. Performance evaluation is done using three kernels (N-Body, tiled matrix multiply and Stream) on different GPU-capable systems based on ARM, Intel x86 and IBM Power8. From the evaluation we observe speed-ups in the 8-20% range compared to versions in which only the GPU is used, reaching 96% of the additional peak performance thanks to the reduction of data transfers and the benefits introduced by the OmpSs NUMA-aware scheduler.

A Novel Set of Directives for Multi-device Programming with OpenMP

“…A. target device(any) and if device(...) Similar to our approach, previous work has been done in order to extend the target directive in order to enhance OpenMP to support execution in multiple devices [19]. It was implemented on top of the MACC compiler and the OmpSs runtime [20].…”

The OpenMP Cluster Programming Model

“…Duran et al introduced another directive-based programming language, called OmpSs [13], to program heterogeneous multi-core architectures. OmpSs was successfully used to design applications for multi-CPU (multi-socket) and multi-GPU architectures [28]. OmpSs was also extended to support a distributed architecture [9] based on GAS-Net [6], a software infrastructure for PGAS languages commonly used in supercomputing.…”

Cluster Programming using the OpenMP Accelerator Model

“…Next, Duran et al introduced another directive-based programming language, called OmpSs [60], to program heterogeneous multi-core architectures. In fact, experiments was performed using OmpSs on multi-CPU (several sockets) / multi-GPU architectures [61]. OmpSs have also been extended to support distributed architecture [62,63] using GAS-Net [64], a software infrastructure for Partitioned Global Address Space (PGAS) languages over high-performance networks, generally used in supercomputing infrastructure.…”