MATLAB Extensions and Compiler Techniques for High-Performance Computing

点击翻译按钮获取中文摘要

MATLAB has become the most popular rapid prototyping language forscientific computing. Highlevel mathematical languages like MATLABdramatically simplify program development. Unfortunately, thissimplification often has come at the expense of efficient implementation.Today, MATLAB programs are usually interpreted on sequential machines.Thus, the performance of these programs suffers in comparison with Fortranprograms compiled on parallel high-performance computers. The objective ofthis project is to develop and evaluate language extensions, run-timelibraries, and tools for compiling MATLAB and similar interactive languagesfor efficient execution on classes of target machines ranging fromuniprocessor workstations to highly-parallel distributed systems.The technical thrust of this project is to develop restructuring compilertechnology that will enable the translation of MATLAB codes into sequentialor parallel object code that is competitive with handwritten code producedby skillful FORTRAN programmers. The work will build on previous work bythe investigators and will take advantage of existing software modules fromthe Falcon system (developed at Illinois). The compiler will acceptparallel extensions and will be able to automatically translate sequentialMATLAB loops into parallel form. The compiler will transform the code toenhance locality and generate efficient sparse code from densecomputations. This transformation is necessary because MATLAB codes usuallyrepresent sparse computations in dense form. The investigators will alsoconsider incorporating some of the compiler modules into the interpreterfor just-in-time restructuring.