static const unsigned int ncores = 8;//std::thread::hardware_concurrency();

      auto doWork = [&](unsigned int part) -> void {
        std::uint32_t N = lhs.height()/ncores;
        std::uint32_t end = (part+1)*N;
        typename SA::Scalar partSum = 0.0;
        for (std::uint32_t i = part*N; i < end; i++){
          processRow(i, partSum);
        }
        std::lock_guard<std::mutex> guard(sumMutex);
        sum += partSum;
      };

      ADD_ELAPSED_TIME_AND_RESET_MARK( profiler, START);
      std::vector<std::thread> threads;
      for (unsigned int i=0; i<ncores; ++i) {
        threads.emplace_back( doWork, i );
        ADD_ELAPSED_TIME_AND_RESET_MARK( profiler, EMPLACE);
      }
      for (unsigned int i=0; i<ncores; ++i) {
        threads.at(i).join();
        ADD_ELAPSED_TIME_AND_RESET_MARK( profiler, JOIN);
      }