DPP-PMRF: Rethinking Optimization for a Probabilistic Graphical Model\n Using Data-Parallel Primitives

We present a new parallel algorithm for probabilistic graphical model\noptimization. The algorithm relies on data-parallel primitives (DPPs), which\nprovide portable performance over hardware architecture. We evaluate results on\nCPUs and GPUs for an image segmentation problem. Compared to a serial baseline,\nwe observe runtime speedups of up to 13X (CPU) and 44X (GPU). We also compare\nour performance to a reference, OpenMP-based algorithm, and find speedups of up\nto 7X (CPU).\n