cacuda

CaCUDA - Unleash the Power of Cactus on Hybrid Systems Introduction We present a massive data parallel computational framework that can be used to develop large scale scientific applications on such petascale/exascale hybrid systems. The framework is built upon the highly scalable Cactus computational framework that has been used by scientists and engineers from various fields to solve a wide spectrum of real world scientific problems. Within Cactus, we further explore the performance and scalability of hybrid systems by making use of the computing power of the accelerating devices, in particular GPU"s connected to many-core CPU"s. ### Kernel Definition ### The task of simplifying the generation of CUDA code for a finite differencing code is not a straightforward one. Shared arrays with appropriate stencil sizes have to be carefully managed, and data needed by the stencil has to be streamed in while calculations proceed