Memory Access Optimization in Recurrent Image Processing Algorithms with CUDA

摘要

The present paper deals with the algorithms of image processing using CUDA technology. Mem- ory optimizations are the most important area for performance of a CUDA application. This is especially urgent for the recurrent algorithms of data processing. The paper proposes a new approach for recurrent data processing on the GPU. The effectiveness of this approach is shown for the moving average algorithm. The proposed algorithms were used to develop image processing software system. Some results obtained for this system are given in the paper.