cuda::experimental::coop::lane_synchronizer

namespace cuda::experimental::coop {

class lane_synchronizer
{
public:
  explicit lane_synchronizer() = default;

  template <typename Unit, typename ParentGroup, typename MappingResult>
  [[nodiscard]] __device__
  auto make_instance(const Unit&, const ParentGroup&, const MappingResult&) const noexcept;
};

} // namespace cuda::experimental::coop

Overview

cuda::experimental::coop::lane_synchronizer is a synchronizer that uses __syncwarp(lane_mask) to synchronize the group, where lane_mask is the final group mapping result’s lane mask. It can only be used with groups that group threads within the warp level.

The type is empty and is explicitly default constructible.