Dynamic block prefetching for tree search

Tree search algorithm typically involves long latency for the main memory access due to irregular and unpredictable data access during the tree traversal. Data prefetching is one method to bridge the memory gap and improve cache performance. In this paper, we extend our previous work and propose a novel data prefetching approach for tree search, called Dynamic Block Prefetching. Dynamic block prefetching can be used in combination with other optimized methods, such as search branch unwinding. This approach can be used to improve the performance in general-purpose processors which own hardware cache and heterogeneous acceleration cores which not provide hardware supported memory coherence or have limited on chip storage. The experiments show that dynamic block prefetching performs as high as 27% better than no prefetching search on Intel Core i7. Also, it is capable of improving performance over no prefetching search and static buffer scheme by 26% and 11% on Cell SPE.