When comparing QuartzCore's implementation of back to front buffer copies to our implementation of the copy in RemoteLayerBackingStore, there are two optimizations that we are missing. First, QuartzCore masks the back to front buffer copy by the dirty shape. Secondly, it avoids re-clearing a surface that was just allocated. Implement these optimizations in RemoteLayerBackingStore to see if they improve our memory bandwidth regression.
<rdar://problem/109179068>
Pull request: https://github.com/WebKit/WebKit/pull/13729
Pull request: https://github.com/WebKit/WebKit/pull/13770
Committed 263994@main (c26d9ad7d08d): <https://commits.webkit.org/263994@main> Reviewed commits have been landed. Closing PR #13770 and removing active labels.