Creating Customized CGRAs for Scientific Applications

Executing complex scientific applications on Coarse Grain Reconfigurable Arrays (CGRAs) offers improvements in the execution time and/or energy consumption when compared to optimized software implementations or even fully customized hardware solutions. In this work, we explore the potential of application analysis methods in such customized hardware solutions. We offer analysis metrics from various scientific applications and tailor the results that are to be used by MC-Def, a novel Mixed-CGRA Definition Framework targeting a Mixed-CGRA architecture that leverages the advantages of CGRAs and those of FPGAs by utilizing a customized cell-array along, with a separate LUT array being used for adaptability. Additionally, we present the implementation results regarding the VHDL-created hardware implementations of our CGRA cell concerning various scientific applications.

Automated summary

Executing complex scientific applications on Coarse Grain Reconfigurable Arrays (CGRAs) can improve execution time and energy consumption compared to software implementations or fully customized hardware solutions. This work explores the potential of application analysis methods in customized hardware solutions and presents the MC-Def, a Mixed-CGRA Definition Framework targeting a Mixed-CGRA architecture that leverages the advantages of CGRAs and FPGAs. Results show improved performance in executing various scientific applications with VHDL-created hardware implementations of CGRA cells.