The proposed project approaches the problem of library development by a fundamentally different approach -- one in which the knowledge abstractions at an appropriate level are represented and then this knowledge is converted into implementations optimized for different target platforms and circumstances. The primary intellectual merit here is the automation of high quality numerical code generation.
The broader impact is that abstraction driven automation for library developers can significantly cut down the "software gap" -- delay between development of new architectures and new software that maximizes its use. The impact of this will be broad across all disciplines that use HPC.
