RePar


By Greg Gibeling

RePar

Reconfigurable parallel computing research, including programming languages and high level synthesis.

RePar was the unbrella project for my PhD research into High Level Synthesis. As a compiler RePar could transform a computation kernel in optimized Verilog, C/C++, CUDA or Java code, including useable I/O mechanisms. While the project was not completed to any usable form, performance results were excellent. The canonical example was the compilation of a recursive LCS implementation to Verilog, including automated memo-ization and hierarchical memory blocking.