This is particularly useful, for example, when creating pipelines in many threads. Although access to the pipeline cache is thread-safe, implementations may internally take locks to prevent concurrent write access to multiple caches. If you instead create multiple pipeline caches—one for each thread—and use them during initial creation of your pipelines, any per-cache locks taken by the implementation will be uncontested, speeding access to them. Later, when all pipelines are created, you can merge the pipelines in order to save their data in one large resource.
Performance improvement: Interesting hints
Share this highlighthttp://www.safaribooksonline.com/a/vulkantm-programming-guide/12795210/