how to use cooperative_group to sync a grid in cuda?

Question

I want to sync all threads in grid by cooperative_groups::this_grid().sync(), but It seems the code behind this sentence cannot be executed.

For example, I wrote a pi