high performance computing on graphics processing units: hgpu.org

This paper examines an algorithm using dual OpenCL image buffers to optimize data streaming for ensemble processing and visualization. Image buffers were utilized because they allow cached memory access, unlike simple data buffers, which are more commonly used. OpenCL image object performance was improved by allowing upload and mapping into one buffer to occur concurrently with mapping and/or processing of data in another buffer. This technique was applied in an interactive application allowing multiple flood extent maps to be combined into a single image, and allowing users to vary input image sets in real time. The efficiency of this technique was tested by varying both dimensions of input images and number of iterations; computation scaled linearly with number of input images, with best results achieved using ~4k images. Tests were performed to determine the rate at which data could be moved from data buffers to image buffers, examining a large range of possible image buffer dimensions. Additional tests examined kernel runtimes with different image and buffer variants. Limitations of the algorithm and possible applications are discussed.