Conflict-free parallel access scheme for mixed-radix FFT supporting I/O permutations

FFT algorithms are inherently highly parallel and often require frequent access to memory indicating need for high memory bandwidth. Unfortunately, in-place FFT algorithms access data in specific data patterns, thus conflict-free parallel access calls for specialized access schemes. In this paper, we propose a conflict-free parallel access scheme for mixed-radix FFT computations, which supports not only the kernel computations but also I/O permutations of the FFT. The scheme supports all the sequence sizes of power-of-two and all the power-of-two numbers of memory modules. The implementation of the address generation unit is simple requiring only XOR gates and hardwiring.