No abstract

Abstract. OpenMP includes in its latest 4.0 specification the accelerator model. In this paper we present a partial implementation of this specification in the OmpSs programming model developed at the Barcelona Supercomputing Center with the aim of identifying which should be the roles of the programmer, the compiler and the runtime system in order to facilitate the asynchronous execution of tasks in architectures with multiple accelerator devices and processors. The design of OmpSs is highly biassed to delegate most of the decisions to the runtime system, which based on the task graph built at runtime (depend clauses) is able to schedule tasks in a data flow way to the available processors and accelerator devices and orchestrate data transfers and reuse among multiple address spaces. For this reason our implementation is partial, just considering from 4.0 those directives that enable the compiler the generation of the so called "kernels" to be executed on the target device. Several extensions to the current specification are also presented, such as the specification of tasks in "native" CUDA and OpenCL or how to specify the device and data privatization in the target construct. Finally, the paper also discusses some challenges found in code generation and a preliminary performance evaluation with some kernel applications.

No 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.