high performance computing on graphics processing units: hgpu.org

We present methods for parallelizing the transport map construction for multi-core processors and for Graphics Processing Units (GPUs). We provide an efficient implementation of the transport map construction. We describe a method for multi-core processors using the OpenMP framework which brings performance improvement over the serial version of the map construction. We developed a novel and efficient algorithm for multivariate polynomial multiplication for GPUs and we implemented it using the CUDA framework. We show the benefits of using the multivariate polynomial multiplication algorithm for GPUs in the map composition operation for high orders. Finally, we present an algorithm for map composition for GPUs.