In the optimizing compilation process of instruction level parallelism, the memory allocation of data elements is an unavoidable key problem, which has a direct effect on the efficiency of the final results of optimizing compilation. In this paper, the first part briefly presents the background of the paper. Part 1 is the principles of a brand new software pipelining algorithm—ILSP(interlaced inner and outer loop software pipelining). In part 2, taking the example of matrix multiplication, the authors expound the characteristic of date accessing under ILSP algorithm,and make a deep theoretical analysis on that characteristic and conclude a general algorithm on memory allocation of data elements in the nested loops.The experiment is presented in part 4.And part 5,the conclusion.