Multidimensional fast Fourier transform acceleration method based on matrix operation

The invention belongs to the field of computing power optimization, and relates to a multi-dimensional fast Fourier transform acceleration method based on matrix operation. The method comprises the following steps: S1, applying for computing resources, copying computing data to a global memory of equipment by a host end, and generating a DFT matrix and an input data matrix; s2, reading DFT matrix data and an input data matrix from the global memory to a register and a shared memory; s3, a calculation task of matrix element multiplication of the input data matrix and the twiddle factor matrix is completed in parallel through multiple threads, and intermediate calculation result data is obtained; s4, calling a matrix operation unit to perform matrix multiplication on the intermediate calculation result data obtained in the step S3 and a DFT matrix, and outputting final calculation result data; and S5, the host end copies the final calculation result data from the global memory of the equipment to a host memory,