FPGA based Custom GPU Design

Abstract

Due to increase in the demand of graphics based technology, image processing has become a significant subject. Results gets better with the decrease in time required for processing. For this particular purpose a dedicated processor for image processing is a key component. There are many processors available that work dedicatedly for image processing. But more research is required when we have special requirements like speed, flexibility and reliability with real time processing. In this particular project the optimized solution with greater speed and better parallel processing for images has been presented using FPGA. Customized ISA for dedicated GPU has been developed on Xilinx with forty nine different instructions. Instructions include the regular arithmetic and logical instructions, control branch instructions with vector instructions that are designed especially for fast image processing. To process instructions in parallel, technique of C-slow is being used. To further enhance the processing speed of the instructions, micro pipelining has been in-cooperated. Further to test the capabilities of this custom GPU, an application of template matching on .NET of has been developed. The application runs successfully on the personalized processor thus fulfilling the objective of area and time effective architecture for a standalone processor with essential built in resources for template matching application.