An Inference Mechanism for the Compilation of Interactive Array Languages

Interactive array languages are powerful programming tools for the development of programs for numerical computation. They provide an environment that tends to increase productivity in software development. The trade-off is that in order to provide this nicer programming environment, array languages are usually interpreted, with the resulting negative effect on performance. The ideal programming environment for the development of programs for numerical computation should use an interactive array language for the conception and development phase of the program and a compiler for the production phase. Performance can be further improved with the exploitation of parallelism. This paper presents an inference mechanism that is used for the compilation of Matlab programs into Fortran 90. This approach combines static and dynamic inference methods for type, rank, and shape inference. This work was supported by NIST under Grant No. MDA904-92-C-2312 and the CSRD Affiliates under grants from the U.S. National Security Agency and the Motorola Corporation.