(Astro)physical supercomputing: Ada95 as a safe, object oriented alternative

Abstract

Due to a variety of policy errors of the guardians of Ada in the past, the scientific community has never come to appreciate Ada83; Fortran still holds virtually absolute sway over computational (astro)physics. The advent of Ada95 may have changed this unhappy situation. Extensive exploration of the potential of Ada in the field of spectral line synthesis over the past years has revealed its suitability for numerically intensive (astro)physical modelling. Genericity, hierarchical libraries and programming by extension are among the most attractive features, facilitating code maintenance and extension and achieving a high degree of software reuse. In addition, Ada tasking together with protected objects allows straightforward parallelisation with very few modifications to sequential code, yielding linear scaling of performance with the number of processors, and almost perfect load balance, whereas the data-parallel approach of High Performance Fortran turns out to be of little use. Numerous code examples illustrate how easy it becomes to write safe, reusable and massively parallel scientific codes in Ada95 without incurring the “abstraction penalty”.