The use of GPU (Graphics Processor Unit) technology is wide spread in the IT industry and has a significant place in image and video processing. This technology has also found a place outside of the IT industry and appears everywhere there are vision and video processing requirements. The approach taken to programming and implementing GPU is in vision and video processing is different to that required for FPGA design implementation.
FPGA SoC devices often contain a GPU ‘hard IP’ block that can be used as part of the overall design. Omnitek have a broad experience of using GPU technology as part of image processing designs as an alternative to using other FPGA resources such as DPLs and LUTs.
Omnitek’s indepth knowledge and daily use of FPGA SoC devices in its customer’s designs, means that it can provide vision and video processing solutions that hardness the power of either FPGA or DPU technologies.
Using GPU programming languages such as Cg and CUDA it is possible to use GPU functionality to perform sophisticated image processing functions such sizing, rotation and warping. The pixel shader functionality within the GPU can be used to compare multiple images simultaneously on a pixel by pixel basis.
Example applications where GPU hardened IP on MPSoC devices has been used include camera Image Signal Processing (to provide de-bayer, image sharpening and level correct) and Image Stitch (to compare camera overlap areas). In these applications it is more effective to use the resource available in the GPU than trying to emulate software recursive operations using FPGA gates.