Université Paris Diderot
The authors propose a methodology for the modeling, instrumentation and design space exploration with prototyping of complex designs exhibiting difficult to anticipate performance caveats. It relies on monadic functional programming as a design specification means. Its advantages lie in its Model-of-Computation (MoC) mixing capabilities, its suitability for both human-assisted and automatic design space exploration, as well as its relative ease of use compared to other functional approaches. They show those aspects on a case study involving memory fetches from a set of tomographic backprojection coprocessors working in parallel.