ESTIMATING THE PERFORMANCE OF MULTIPLE FUNCTIONAL UNIT AND PIPELINED PROCESSOR USING STATISTICAL INSTRUCTION DEPENDENCIES

This paper considers the processor with the parallel operation and loo kahead control functions based on the real-time system program. The re lation between the logical dependencies among instructions and the per formance improvement is analyzed. With respect to the dependency, the number of dynamic steps that are processed between logically dependent instructions is considered, and the histogram is examined in the prog ram actually run. A method is devised wherein the behavior of the proc essor executing the instructions following the written order is repres ented in terms of the probability based on the statistical data. The e ffects of the dependency and the number of functional units on the pro cessor also is evaluated by a simulation. The validity of the proposed method is verified and the effect is evaluated of performance improve ment when the order of instruction execution is made variable.