Conclusions and Future Directions

Computer architecture is moving toward specialization, and we desperately need new tools, techniques, and languages to enable productivity in this new era. We have argued that to move forward, it is not sufficient to build more sophisticated compilers to automatically parallelize and retarget code (although such tools are certainly useful and welcome!). Instead, it will be more effective to create new languages optimized for particular domains, and write compilers that leverage domain knowledge about particularly good design patterns to produce great designs. Specialized hardware engines exploit unique features of the application to be as efficient as possible; our programming languages ought to do this too.