We test the relative performances of two different approaches to the computation of forces for molecular dynamics simulations on graphics processing units. A "vertex-based" approach, where a computing thread is started per particle, is compared to an "edge-based" approach, where a thread is started per each potentially non-zero interaction. We find that the former is more efficient for systems with many simple interactions per particle while the latter is more efficient if the system has more complicated interactions or fewer of them. By comparing computation times on more and less recent graphics processing unit technology, we predict that, if the current trend of increasing the number of processing cores--as opposed to their computing power--remains, the "edge-based" approach will gradually become the most efficient choice in an increasing number of cases.

The great potential of computational diffusion MRI (dMRI) relies on indirect inference of tissue microstructure and brain connections, since modelling and tractography frameworks map diffusion measurements to neuroanatomical features. This mapping however can be computationally highly expensive, particularly given the trend of increasing dataset sizes and the complexity in biophysical modelling. Limitations on computing resources can restrict data exploration and methodology development. A step forward is to take advantage of the computational power offered by recent parallel computing architectures, especially Graphics Processing Units (GPUs). GPUs are massive parallel processors that offer trillions of floating point operations per second, and have made possible the solution of computationally-intensive scientific problems that were intractable before. However, they are not inherently suited for all problems. Here, we present two different frameworks for accelerating dMRI computations using GPUs that cover the most typical dMRI applications: a framework for performing biophysical modelling and microstructure estimation, and a second framework for performing tractography and long-range connectivity estimation. The former provides a front-end and automatically generates a GPU executable file from a user-specified biophysical model, allowing accelerated non-linear model fitting in both deterministic and stochastic ways (Bayesian inference). The latter performs probabilistic tractography, it can generate whole-brain connectomes and supports new functionality for imposing anatomical constraints, such as inherent consideration of surface meshes (GIFTI files) along with volumetric images. We validate the frameworks against well-established CPU-based implementations and we show that despite the very different challenges for parallelising these problems, a single GPU achieves better performances than 200 CPU cores thanks to our parallel designs.

OP2 is an "active" library framework for the solution of unstructured mesh-based applications. It utilizes sourceto-source translation and compilation so that a single application code written using the OP2 API can be transformed into different parallel implementations for execution on different back-end hardware platforms. In this paper we present the design of the current OP2 library, and investigate its capabilities in achieving performance portability, near-optimal performance, and scaling on modern multi-core and many-core processor based systems. A key feature of this work is OP2's recent extension facilitating the development and execution of applications on a distributed memory cluster of GPUs.We discuss the main design issues in parallelizing unstructured mesh based applications on heterogeneous platforms. These include handling data dependencies in accessing indirectly referenced data, the impact of unstructured mesh data layouts (array of structs vs. struct of arrays) and design considerations in generating code for execution on a cluster of GPUs. A representative CFD application written using the OP2 framework is utilized to provide a contrasting benchmarking and performance analysis study on a range of multicore/many-core systems. These include multi-core CPUs from Intel (Westmere and Sandy Bridge) and AMD (MagnyCours), GPUs from NVIDIA (GTX560Ti, Tesla C2070), a distributed memory CPU cluster (Cray XE6) and a distributed memory GPU cluster (Tesla C2050 GPUs with InfiniBand). OP2's design choices are explored with quantitative insights into their contributions to performance. We demonstrate that an application written once at a high-level using the OP2 API can be easily portable across a wide range of contrasting platforms and is capable of achieving near-optimal performance without the intervention of the domain application programmer.

The solution of large sparse linear systems arises in many applications, such as computational fluid dynamics and oil reservoir simulation. In realistic cases the matrices are often so large that they require large scale distributed parallel computing to obtain the solution of interest in a reasonable time. In this paper we discuss the design and implementation of the AmgX library, which provides drop-in GPU acceleration of distributed algebraic multigrid (AMG) and preconditioned iterative methods. The AmgX library implements both classical and aggregation-based AMG methods with different selector and interpolation strategies, along with a variety of smoothers and preconditioners, including block-Jacobi, Gauss-Seidel, and incomplete-LU factorization. The library contains many of the standard and flexible preconditioned Krylov subspace iterative methods, which can be combined with any of the available multigrid methods or simpler preconditioners. The parallelism in the aggregation scheme exploits parallel graph matching techniques, while the smoothers and preconditioners often rely on parallel graph coloring algorithms. The AMG algorithm implemented in the AmgX library achieves 2-5× speedup on a single GPU against a competitive implementation on the CPU. As will be shown in the numerical experiments section, both setup and solve phases scale well across multiple nodes, sustaining this performance advantage.

Code maintainability, performance portability and future proofing are some of the key challenges in this era of rapid change in High Performance Computing. Domain Specific Languages and Active Libraries address these challenges by focusing on a single application domain and providing a high-level programming approach, and then subsequently using domain knowledge to deliver high performance on various hardware.In this paper, we introduce the OPS high-level abstraction and active library aimed at multi-block structured grid computations, and discuss some of its key design points; we demonstrate how OPS can be embedded in C/C++ and the API made to look like a traditional library, and how through a combination of simple text manipulation and back-end logic we can enable execution on a diverse range of hardware using different parallel programming approaches.Relying on the access-execute description of the OPS abstraction, we introduce a number of automated execution techniques that enable distributed memory parallelization, optimization of communication patterns, checkpointing and cache-blocking. Using performance results from CloverLeaf from the Mantevo suite of benchmarks, we demonstrate the utility of OPS.