Solution 4.5:
have a local cell_list per thread block. can reduce synchronization overhead by a constant factor.
paracon
@-o4, do you mean have multiple local cell_lists? (One for each cell or a higher granularity). I think there would be overhead of even more number of merges.
Solution 4.5: have a local
cell_list
per thread block. can reduce synchronization overhead by a constant factor.@-o4, do you mean have multiple local cell_lists? (One for each cell or a higher granularity). I think there would be overhead of even more number of merges.