A High Performance Programming Model for LargeScale Molecular

We introduce a high performance programming model for large-scale molecular dynamics calculations on reconfigurable hardware. The programming model is based on a methodology to leverage legacy code and accelerate calculations running on reconfigurable supercomputers. We discuss the steps taken in the formulation of scalable, arbitrarily-sized, parallel 3-Dimensional FFT kernels implemented on networks of FPGAs. We achieve sustainable speedup and scaling by exchanging the concept of a network processing-based architecture with gather-scatter operations for a network of streams and accelerator kernels