The invention belongs to the technical field of high performance computing, in particular to a processing method for cyclic tilt transformation optimization. Firstly, dependency analysis is performed, a program is analyzed and changed to be suitable for cyclic tilt optimization; secondly, array re-layout is performed for cyclic tilting, and data locality is improved to facilitate vectorized operation; finally, non-full-load vector operation is achieved. The problems of o data discretization, poor data locality and difficult vectorization of the parallel execution part in cyclic tilt transformation are solved. Data dependence analysis, cyclic tilt transformation, data reconstitution and non-full-load vector operation are achieved according to cyclic tilt change, the operating efficiency and executing performance of a cyclic tilt transformation program are effectively improved, and cyclic tilt transformation optimization is achieved. Finally, testing is performed by selecting a wavefront program, and average 10.73 times of acceleration effect can be obtained through program calculation after optimization.