CsWinRT overhead on hot paths

[ekalchev]

"native interop that bypasses CsWinRT overhead on hot paths"

Can someone elaborate on this statement. How you bypass CsWinRT?

[codemonkeychris]

where did you see this? I suspect this was out of date from a past experiment - we tried using XamlDirect style invocation where we get a raw object pointer to the underlying WinRT object and do flat API style invocation of methods, so you avoid creating an RCW (runtime callable wrapper). CLR is pretty effective at caching the RCW and is relatively efficient, but there are some patterns in WinRT that make the interop layer slow. In the experiment we saw 3%-4% framerate improvement, but it meant manually doing ref counting, keeping your own cache of objects. The folks with long history with WinUI said that the XamlDirect model had proven to be more trouble than it was worth, and recommended avoiding it.

We removed the experiment and focused on other areas to improve performance (less WinRT updates, smaller/faster virtual DOM nodes, better diffing optimizations, improved hooks)...

[ekalchev]

its in README.md

microsoft-ui-reactor/README.md

Line 129 in c1cb896

- **Performance.** Element pooling, render coalescing, skip-unchanged optimization, native interop that bypasses CsWinRT overhead on hot paths.

[codemonkeychris]

yeah, that's my bad. stale content... will fix