Abstract:

A novel workload-distribution based CPU/GPU MTF compensation approach for high resolution satellite images is proposed in this paper. First, the basic GPU implementation issues are addressed; Next, three performance tuning methods – the execution configuration optimization, memory access optimization and instruction optimization – are applied to further improve the performance. We test our approach with the GF-1 panchromatic image in the CPU/GPU system that consists of an Intel Xeon E5650 CPU and a NVIDIA Tesla C2050 GPU. The experimental result shows that the speedup ratio is up to 42.80 times. Furthermore, the CPU/GPU workload distribution strategy is presented to fully exploit CPU’s computing horsepower. With this strategy, the speedup ratio of MTF compensation finally reaches to 47.82 times (the corresponding processing time is 1.62s), which could meet the requirement of near real-time MTF compensation for high resolution satellite images.

Key words: CPU/GPU, MTF compensation, execution configuration optimization, memory access optimization, instruction optimization, workload distribution strategy