Knowledge-based support for scientific programming

The author considers how knowledge-based tools can support scientific computing, with a focus on the support of code generation. Scientific computing traditionally is carried out by mathematical modelers who write their own Fortran programs. As faster machines and new architectures make more complex problems computationally feasible, programming becomes more cost effective. Scientific computing comprises a variety of activities including model formulation, coding, and interpretation. The SINAPSE program synthesis system illustrates the use of knowledge-based techniques in model solution. SINAPSE is designed to be part of a problem-solving environment for forward modeling that includes several additional scientific programming activities. SINAPSE generates code by applying refinements and optimizing transformations, first producing an algorithm description and then an array-based, high-level language program before generating target code. SINAPSE generates implementations targeted toward a specified architecture.<
>