Parallel implementation of Aggressive PNN method for devices with GPUs

摘要

We implement the Aggressive PNN method on GPUs by using CUDA to generate codebooks for VQ compression and the speedup of up to 4.01 is achieved compared with the tau PNN method on the CPU. Our second method enhances the parallelism by using indirect vectors to reduce idle threads. We also improved the algorithm by about 20% by using indirect vectors.