Currently G1 calculates the total non-copy time for young gen incrementally as regions are added to the young gen. The reason for this is that previously getting remembered set size has been expensive (i.e. part of non-copy time).

That is not the case any more since JDK 16 or so.

Further, actually in the calculation the remset work time does not need to be calculated on a per-region basis at all; first, it is incorrect as it does not remove duplicates, and additionally there is/has always been a total remembered set prediction anyway. (Which means the remembered set scan/merge time is considered multiple times).

So all this code can be removed.