Introduction - If you have any usage issues, please Google them yourself
Analysis of how to create a simple 7x7 filter is applied to all OpenCL GPU. This case study will focus on improving the performance of the time convolution convolution kernel with a small image, in general, based on FFT convolution kernel and the subsequent 30x30 faster [1], but in fact it is not common to use a large window size. . We will first implement the convolution image, and then use the filter performance improved through the use of memory and cache __constant OpenCL image information in the __local space.