question about lambda_vec

[miaohua]

I am looking at the following motion search code :
t = d_mv_bits(... )
sad += (data->lambda16 * t * sad) >> 10

Would someone explain to me how the vectors
xvid_me_lambda_vec16 and xvid_me_lambda_vec8 are
generated and how and why they are used to adjust the
SAD value?

Thanks,