Jamake: A java compiler environment

Abstract

Alanguage of choice for general-purpose programming, Javais quickly becoming popular in more specialized areas, such as scientificcomputing. However, even though the compilation technologies have significantlyimproved Java execution, performance is still the main obstacleto the use of Java for scientific applications. Although good JavaVirtual Machine implementations are approaching the performance ofFortran on similarly-coded applications, significant performance problemsremain because of the power of the object-oriented programmingparadigm. Our experiments show that full use of polymorphic, objectorientedprogramming can result in performance penalties of up to twoorders of magnitude. To address this performance difficulty, the authorshave developed the JaMake Java transformation system, which uses advancedprogram analysis and transformation techniques to allow programmersto create extensible and maintainable programs using objectorienteddesign, while generating Java programs whose performance approachesthat of hand-optimized, Fortran-style code. Experiments on ourcollection of object-oriented scientific programs have shown that transformationby JaMake can yield speed-ups of a factor of ten or more,bringing the performance of these object-oriented programs to within75% of hand-optimized, Fortran-style code.