A question on forcing fp32 in the code

[zjuliangxun]

Thanks for your outstanding work！However, when I was analyzing the code，I found some "@force_fp32" operations hard to understand. For example, in line of 92 of the file "projects/mmdet3d_plugin/bevformer/modules/encoder.py", the function "point_sampling" requires certain parameters to be float32.

# This function must use fp32!!!
    @force_fp32(apply_to=('reference_points', 'img_metas'))
    def point_sampling(self, reference_points, pc_range,  img_metas):

Would you please tell me what will happen if they are float16?