Should record all_reduce duration time in profile report
Created by: Yancey1989
profile of all_reduce is 0
thread0::increment 40 0.677888 0.014336 0.023552 0.0169472 7.0255e-05
thread0::create_double_buffer_reader 40 0.293888 0.00512 0.012288 0.0073472 3.0458e-05
thread0::all_reduce 805 0 0 0 0 0