13_2_CUDAstream_kernel_copy_overlap.cu Too many blocks allocated

[Yucheng-HO]

line 44
应该用 nums_per_stream
int s = ceil((size_per_stream + bs - 1.) / bs);