(axis == 0 == -4) 100; 100x3; 100x3x40; 100x3x40x60
(axis == 1 == -3) 3; 3x40; 3x40x60
(axis == 2 == -2) 40; 40x60
(axis == 3 == -1) 60
并且,bottom[1] 的shape可以为空,表示scale是一个标量(这时axis就没有意义)