JLP uses assembly instructions to perform GPU computations, also overall process of jumping is pretty good optimized in my opinion. Can you please give an example of non-optimal computations in JLP code, by optimizing which we can get a such increase in speed?