ACM SRC poster: a portable implementation of the integral histogram in starss

Parallel programming models converge on key concepts. Program syntax avoids explicitly parallel constructs like threads and data dependence guides the computation, as opposed to resource-centric models like MPI or OpenMP. Aside from StarSs, StarPU and recently the MAGMA and PLASMA projects encapsulate computation on data blocks in tasks. These are scheduled dynamically via a TDG. We intent to demonstrate the use of StarSs in the development of the Integral Histogram (IH) and analyze the application on SMP, Cell/B.E. and GPU. As opposed to the applications for StarPU, MAGMA and PLASMA, which so far have been limited to numerical linear algebra kernels. IH is a recently proposed preprocessing technique that constructs the histogram for rectangular regions in constant time (for e.g. object recognition, content-based image retrieval, segmentation, detection and tracking). To the best of our knowledge, IH in StarSs is the first parallel implementation of this algorithm in the literature.