Output details

Hypercubic storage layout and transforms in arbitrary dimensions using GPUs and CUDA

<25> This article explored how the spatial dimension itself could be encoded as a parameter of highly parallel simulation programs, enabling various models to be studied in arbitrarily high dimensions using the same software. Our rigorous performance measurements and explicit examples have inspired many citing users to exploit a similar approach, since it considerably lowers software development costs against the use of separate parallel programs for each dimension studied. The approach generalises to support many other applications using hyper-dimensional arrays such as voxel data analyses, and our novel algorithms make strong and highly cost-effective use of graphical processor unit architectures.