Variable optimization in code migration from GPU to multi-core

Variable analysis and optimization could improve the performance of code migration from CUDA to multi-core. Based on memory mapping, a temporary arrays pruning algorithm is proposed in this paper. By analyzing the live range and def-use relation, the temporary arrays which could be pruned are identified. A function map between the definitions and use of temporary arrays is constructed to replace the original arrays. Some conditions of the optimization algorithm are discussed and the test result shows that the performance could improve by 3.4% with the optimization.