GPU-based spatially divided predictive partitioned vector quantization for gifts ultraspectral data compression

Predictive partitioned vector quantization (PPVQ) has been proven to be an effective lossless compression scheme for ultraspectral sounder data. In previous work, we have identified the two most time-consuming stages of PPVQ for implementation on GPU. By using 4 GPUs and a spectral division design in sharing the workload, we showed a 42x speedup on NASA´s Geostationary Imaging Fourier Transform Spectrometer (GIFTS) dataset compared to its original single-threaded CPU code. In this paper, an alternative spatial division design is developed to run on 4 GPUs. The experiment on the GIFTS dataset shows that a 72x speedup can be further achieved by this new design of the GPU-based PPVQ compression scheme.