It is clear that massively parallel computers offer the best path to the supercomputers of the future. Unfortunately, they are very difficult to program efficiently due to the lack of compiler support for the high level languages that exist. The research undertaken in this proposal will focus on one such language, Fortran 90, with the intent of providing techniques to enhance compilation for distributed memory computers. Fortran 90 offers two advantages: a rich set of array operations that enable the programmer to address complex applications, and a language that, as an extension of Fortran 77, is familiar to the broad scientific community. The research will focus on three key compilation problems: automatic array alignment, parameterized data layout, and generation of collective communications. Emphasis will be placed on the development of general principles of compilation and the generation of efficient code.
