Mixed precision has made many advances in deep learning and precision tuning and optimization. Extensive research shows that the mixed precision for stencil computation is also very challenging. At the same time, a series of research of polyhedral model in the field of automatic parallelization show that the model provides a good mathematical abstraction for loop nesting, a series of loop transformations can be performed on the basis of it. This paper designs and implements an automatic mixed precision optimizer for Stencil computations based on the polyhedral compilation. By performing iterative domain partitioning, data flow analysis and scheduling tree transformation on the intermediate presentation, we implement the source-to-source automatic generation of mixed precision code for stencil computation for the first time. The experiments show that the code after automatic mixed precision can give full play to its potential parallelism and improve the performance of the program on the basis of reducing precision redundancy. Taking high-precision computing as benchmark, the maximum speedup is 1.76 and the geometric average speedup is 1.15 on the X86 architecture. On the new-generation Sunway architecture, the maximum speedup is 1.64 and the geometric average speedup is 1.20.