Fortress Abstractions in X10 Framework

Fortress provides a nice set of abstractions used widely in scientific computing. The use of such abstractions enhances the productivity of programmers/users. Also, in scientific computations, boilerplate code has extensive usage. Keeping this in view, we embed Fortress abstractions in an X10 environment so that we can get better productivity without losing performance. In this paper, we transform Fortress into X10 through a transcompilation system. We describe compilation strategies for a few important constructs and discuss the performance of the generated X10 code with respect to the original Fortress code. The translated X10 code outperforms the original Fortress code with a maximum of 206x speedup achieved in the best case. The system also supports the multiresolution language approach that simplifies parallel programming by allowing domain scientists to write programs in the Fortress syntax that is closer to the mathematical notation. The translated X10 code, which can further be compiled to either C++ or Java, implicitly assures performance and may further be optimized for performance by utilizing the low-level features of X10 (or C++/Java).