Suppose two consecutive convolutional layers are widened by a factor , i.e., for the second layer, both its number of input channels and its number of output filters are therefore multiplied by , while its kernel size and spatial dimensions remain fixed. How does the second layer's computation scale approximately?