Selection schemes for dynamically microcoding fortran programs

Abstract

The objective of the present study has been to investigate possible methods to reduce a program's execution time by detecting and converting automatically the more frequently executed program parts, mostly inner loops, into microcode. The methods proposed were static loading of inner loops, selective loading of inner loops, over-lay of inner loops, and user-aided scheme. Using Fortran programs as the test programs, a simulation program was written to measure the gain achieved by each method. A final gain between 1.587 and 4.76 was achieved by the proposed methods for memory speed ratios between 3 and 8. It was found that 90% of the final gain of the test programs could be obtained with writable control memory requirements that were less than 40% of the final requirement.