The fast Fourier transform (FFT) plays a vital role in a wide range of applications including image and video processing, data compression, and wireless communication systems. The output samples of the FFT are generated in a bit-reversed order, which necessitates the implementation of a reordering scheme. However, this introduces additional latency, limits the overall throughput, and increases the hardware complexity and cost. In this paper, a novel parallel and flexible architecture for computing FFT with arbitrary lengths is proposed. The key idea is to leverage the in-memory computing (IMC) technique to eliminate the need for a dedicated reordering scheme, while enabling ultra high-throughput and low-latency FFT computation.