This paper addresses the joint ride-sharing matching and route planning problem, in which passenger pick-up and drop-off points are adjustable to reduce vehicle detours. We first formulate the problem as a mixed-integer linear programming (MILP) model and demonstrate its NP-hardness. The model incorporates practical constraints, such as passenger travel-time windows, and simultaneously optimizes ride-sharing matches, pick-up and drop-off points, and vehicle routes to minimize total system costs (including vehicle activation costs, vehicle travel costs, passenger travel-time window penalties, and passenger time cost). Given the inefficiency of commercial solvers in solving the MILP model and the need for efficient computation in real-world cases, we introduce the concept of "spatial-temporal order similarity" to decompose large-order demands into data clusters for parallel computing. For each data cluster, we develop a two-stage memory search-dynamic programming algorithm. In the first stage, an improved memory search algorithm is proposed to address ride-sharing matching and route planning with fixed pick-up and drop-off points. In the second stage, a specialized dynamic programming algorithm is developed to select the optimized pick-up and drop-off points for ride-sharing. A feedback mechanism is introduced between the two stages, whereby if the second stage fails to reduce total system costs, it will return to the first stage to re-optimize ride-sharing matching and route planning. We apply the MILP model and algorithms within the areas of the Beijing Fifth Ring Road. Experimental results demonstrate a 51.10% reduction in vehicle travel distance and a 1124.53 kg decrease in CO2 emissions compared to conventional ride-hailing services. Furthermore, the MILP model reduces total platform operating costs by 2.08% compared to fixed pick-up/drop-off systems, underscoring its economic and environmental advantages.