Tracks performance issues of inner persistent scheduler #107
Description
The results of const int64_t warp_size = std::min(warp_size_based_on_l1, warp_size_based_on_l2); in innerPersistentHeuristiccan be a odd number such as 15. which seems not right. Prefer 16 or 32. Noticed while working on this PR #68 (review)