High performance flow field visualization with high-order access dependencies

We present a novel model based on high-order access dependencies for high performance pathline computation in flow field. The high-order access dependencies are defined as transition probabilities from one data block to other blocks based on a few historical data accesses. Compared with existing methods which employed first-order access dependencies, our approach takes the advantages of high order access dependencies with higher accuracy and reliability in data access prediction. In our work, high-order access dependencies are calculated by tracing densely-seeded pathlines. The efficiency of our proposed approach is demonstrated through a parallel particle tracing framework with high-order data prefetching. Results show that our method can achieve higher data locality than the first-order access dependencies based method, thereby reducing the I/O requests and improving the efficiency of pathline computation in various applications.