Implementing new languages on supercomputers is animportanttask, but it is also adifficultone. It is an important task because language implementation is necessary for research in language design, and language design has often been a source of new ideas for hardware design. Language implementation for supercomputers is a difficult task for many reasons. Perhaps the most significant reason is the sheer complexity of the more popular supercomputers. This paper argues that a useful and practical approach to implementing new languages on common supercomputers in a research context is to write atranslator, which translates programs into Fortran source code rather than compiling them directly into assembly language. This approach has many advantages, particularly in a research environment. The paper also includes a call to supercomputer designers for greater simplicity in the programmers' view of supercomputer architectures.
展开▼