Abstract. In the field of HPC, the current hardware trend is to design multiprocessor architectures that feature heterogeneous technologies such as specialized coprocessors (e.g., Cell/BE SPUs) or data-parallel accelerators (e.g., GPGPUs). Approaching the theoretical performance of these architectures is a complex issue. Indeed, substantial efforts have already been devoted to efficiently offload parts of the computations. However, designing an execution model that unifies all computing units and associated embedded memory remains a main challenge. We have thus designed STARPU, an original runtime system providing a highlevel, unified execution model tightly coupled with an expressive data management library. The main goal of STARPU is to provide numerical kernel designers with a convenient way to generate parallel tasks over heterogeneous hardware on the one hand, and easily develop and tune powerful scheduling algorithms on the other hand. We have developed several strategies that can be selected seamlessly at run time, and we have demonstrated their efficiency by analyzing the impact of those scheduling policies on several classical linear algebra algorithms that take advantage of multiple cores and GPUs at the same time. In addition to substantial improvements regarding execution times, we obtained consistent superlinear parallelism by actually exploiting the heterogeneous nature of the machine.

Abstract. In the field of HPC, the current hardware trend is to design multiprocessor architectures that feature heterogeneous technologies such as specialized coprocessors (e.g., Cell/BE SPUs) or data-parallel accelerators (e.g., GPGPUs). Approaching the theoretical performance of these architectures is a complex issue. Indeed, substantial efforts have already been devoted to efficiently offload parts of the computations. However, designing an execution model that unifies all computing units and associated embedded memory remains a main challenge. We have thus designed STARPU, an original runtime system providing a highlevel, unified execution model tightly coupled with an expressive data management library. The main goal of STARPU is to provide numerical kernel designers with a convenient way to generate parallel tasks over heterogeneous hardware on the one hand, and easily develop and tune powerful scheduling algorithms on the other hand. We have developed several strategies that can be selected seamlessly at run time, and we have demonstrated their efficiency by analyzing the impact of those scheduling policies on several classical linear algebra algorithms that take advantage of multiple cores and GPUs at the same time. In addition to substantial improvements regarding execution times, we obtained consistent superlinear parallelism by actually exploiting the heterogeneous nature of the machine.

Abstract-The increasing numbers of cores, shared caches and memory nodes within machines introduces a complex hardware topology. High-performance computing applications now have to carefully adapt their placement and behavior according to the underlying hierarchy of hardware resources and their software affinities.We introduce the Hardware Locality (hwloc) software which gathers hardware information about processors, caches, memory nodes and more, and exposes it to applications and runtime systems in a abstracted and portable hierarchical manner. hwloc may significantly help performance by having runtime systems place their tasks or adapt their communication strategies depending on hardware affinities.We show that hwloc can already be used by popular highperformance OPENMP or MPI software. Indeed, scheduling OPENMP threads according to their affinities or placing MPI processes according to their communication patterns shows interesting performance improvement thanks to hwloc. An optimized MPI communication strategy may also be dynamically chosen according to the location of the communicating processes in the machine and its hardware characteristics.

No abstract

The emergence of accelerators as standard computing resources on supercomputers and the subsequent architectural complexity increase revived the need for high-level parallel programming paradigms. Sequential task-based programming model has been shown to efficiently meet this challenge on a single multicore node possibly enhanced with accelerators, which motivated its support in the OpenMP 4.0 standard. In this paper, we show that this paradigm can also be employed to achieve high performance on modern supercomputers composed of multiple such nodes, with extremely limited changes in the user code. To prove this claim, we have extended the StarPU runtime system with an advanced inter-node data management layer that supports this model by posting communications automatically. We illustrate our discussion with the task-based tile Cholesky algorithm that we implemented on top of this new runtime system layer. We show that it enables very high productivity while achieving a performance competitive with both the pure Message Passing Interface (MPI)-based ScaLAPACK Cholesky reference implementation and the DPLASMA Cholesky code, which implements another (non-sequential) task-based programming paradigm.

One of the major trends in the design of exascale architectures is the use of multicore nodes enhanced with GPU accelerators. Exploiting all resources of a hybrid acceleratorsbased node at their maximum potential is thus a fundamental step towards exascale computing. In this article, we present the design of a highly efficient QR factorization for such a node. Our method is in three steps. The first step consists of expressing the QR factorization as a sequence of tasks of well chosen granularity that will aim at being executed on a CPU core or a GPU. We show that we can efficiently adapt high-level algorithms from the literature that were initially designed for homogeneous multicore architectures. The second step consists of designing the kernels that implement each individual task. We use CPU kernels from previous work and present new kernels for GPUs that complement kernels already available in the MAGMA library. We show the impact on performance of these GPU kernels. In particular, we present the benefits of new hybrid CPU/GPU kernels. The last step consists of scheduling these tasks on the computational units. We present two alternative approaches, respectively based on static and dynamic scheduling. In the case of static scheduling, we exploit the a priori knowledge of the schedule to perform successive optimizations leading to very high performance. We, however, highlight the lack of portability of this approach and its limitations to relatively simple algorithms on relatively homogeneous nodes. Alternatively, by relying on an efficient runtime system, StarPU, in charge of ensuring data availability and coherency, we can schedule more complex algorithms on complex heterogeneous nodes with much higher productivity. In this latter case, we show that we can achieve high performance in a portable way thanks to a fine interaction between the application and the runtime system. We demonstrate that the obtained performance is very close to the theoretical upper bounds that we obtained using Linear Programming.

No abstract

To fully tap into the potential of heterogeneous machines composed of multicore processors and multiple accelerators, simple offloading approaches in which the main trunk of the application runs on regular cores while only specific parts are offloaded on accelerators are not sufficient. The real challenge is to build systems where the application would permanently spread across the entire machine, that is, where parallel tasks would be dynamically scheduled over the full set of available processing units. To face this challenge, we previously proposed StarPU, a runtime system capable of scheduling tasks over multicore machines equipped with GPU accelerators. StarPU uses a software virtual shared memory (VSM) that provides a highlevel programming interface and automates data transfers between processing units so as to enable a dynamic scheduling of tasks. We now present how we have extended StarPU to minimize the cost of transfers between processing units in order to efficiently cope with multi-GPU hardware configurations. To this end, our runtime system implements data prefetching based on asynchronous data transfers, and uses data transfer cost prediction to influence the decisions taken by the task scheduler. We demonstrate the relevance of our approach by benchmarking two parallel numerical algorithms using our runtime system. We obtain significant speedups and high efficiency over multicore machines equipped with multiple accelerators. We also evaluate the behaviour of these applications over clusters featuring multiple GPUs per node, showing how our runtime system can combine with MPI.