peterbell10

Yeah, for some reason the windows build was linking both the static and dynamic cuda runtimes in this PR but not #82695 so I just squashed them.

The name is meant to complement `cFastNBT` and `cParsedNBT`.

As far as I remember this works as-is. It just never got reviewed.

It's not going to increase the line count if it's only changing the same variable names.

I'm not sure I'd trust that regex... e.g. `lua_tostring` would match.

Just checked, "a_[a-z]" matches over 21,000 times in Bindings.cpp alone.

I think ignoring Bindings.cpp would be best, for now anyway.

Posting this here because I want it to be visible after #3874 is merged but it's very much in response to that discussion. I've written a little benchmark to help...

In the context of cuberite there are quite a few virtual calls and functions defined in other translation units etc. that will prevent inlining so it's not completely irrelevant. Although...

When profiling the lighting code I noticed that `cPieceGeneratorBFSTree` spends a lot of time in `cCuboid::DoesIntersect` and that's just the first that comes to mind. I do agree it's certainly...