high performance computing on graphics processing units: hgpu.org

We consider the recovery of sparse signals from a limited number of noisy observations using the AMP algorithm. In this paper, we present two fast implementations of this algorithm that run on a CPU and on a GPU and which can either be used for arbitrary unstructured measurement matrices or take advantage of the structure of a DCT matrix to give an even faster implementation. Our results show that for small problem sizes, the CPU based implementation is the fastest, but for large problem sizes, a GPU based implementation has the highest throughput.