There should be a method to delay the flushing of stream and kernel requests to the GPU. By default, assume a "flush immediately" model to maintain compatibility but give the option to have lazy flushing of requests.
Log in to post a comment.
