| This thesis presents techniques, and a software architecture, developed by the author to improve the execution performance of MATLAB 1. While previous approaches to this problem, notably the FALCON project, translated MATLAB code into equivalent Fortran or C source code suitable for batch compilation, MaJIC provides an interactive front end that looks like MATLAB and attempts to compile/optimize code behind the scenes in real time. To achieve the dual and contradictory requirements of high performance and low response time, MaJIC employs two techniques: just-in-time analysis and compilation, and ahead-of-time speculative analysis coupled with native compilation. Performance results show that just-in-time compilation provides near-zero response time and performance that supersedes MATLAB's own compiler; speculative compilation masks at least a part of the response time due to native compilation and provides performance comparable to the best existing batch compiler.; 1MATLAB is a trademark of The Mathworks, Inc. |
