Dynamic partial reconfiguration for pipelined digital systems— A Case study using a color space conversion engine

In digital hardware design, reconfigurable devices such as Field Programmable Gate Arrays (FPGAs) allow for a unique feature called partial reconfiguration PR). This refers to the reprogramming of a subset of the reconfigurable logic during active operation. PR allows multiple hardware blocks to be consolidated into a single partition, which can be reprogrammed at run-time as desired. This may reduce the logic circuit (and silicon area) requirements and greatly extend functionality. Furthermore, dynamic partial reconfiguration (DPR) refers to PR that does not halt the system during reprogramming. This allows for configuration to overlap with normal processing, potentially achieving better system performance than a static(halting) PR implementation. This work has investigated the advantages and trade-offs of DPR as applied to an existing color space conversion(CSC) engine provided by Hewlett-Packard (HP). Two versions were created: a single-pipeline engine, which can only overlap tasks in specific sequences; and a dual-pipeline engine, which can overlap any consecutive tasks. These were implemented in a Virtex-6 FPGA. Data communication occurs over the PCI Express (PCIe) interface. Test results show improvements in execution speed and resource utilization, though some are minor due to intrinsic characteristics of the CSC engine pipeline. The dual-pipeline version outperformed the single-pipeline in most test cases. Therefore, future work will focus on multiple-pipeline architectures