Check and adjust the device load when dealing with transcoding pipelines. For example running the Intel® Quick Sync Video encoding might reduce benefits of using the Intel® Graphics for some OpenCL™ frame preprocessing. The reason is that the Intel® Quick Sync Video encoding already loads the Intel® Graphics units quite substantially. In some cases using the CPU device for OpenCL tasks reduces the burden and improves the overall performance. Consider experimenting to find the best solution.
For more complete information about compiler optimizations, see our Optimization Notice.