High Performance Computing on Cross-correlation Functions
Produced CUDA parallel programs to compute cross-correlation functions of large-scale seismic data on GPU
Calculation of noise cross-correlation functions (NCF) plays an important role in ambient noise imaging which is a vital seismic method to obtain Earth inner structures. To raise the resolution of the imaging results, we need more seismic data for imaging. However, as the size of seismic data increases, the serial algorithm for NCF calculation becomes much more time-consuming. Thus, how to accelerate the NCF calculation becomes a key problem in ambient noise imaging. Based on the analysis of serial algorithm, we proposed a new parallel algorithm for NCF calculation using NVIDIA GPU platform. In addition, we improved reading and writing strategy to reduce I/O consumption. Experimental results on real seismic data show the effectiveness of our method. The parallel program achives about 1861 times speedup.
