Fast convolution can be computed using the FFT by taking the FFTs of the input sequences, multiplying them point-wise in the frequency domain, and taking the inverse FFT. This is more efficient than direct convolution when the sequences are long. For long sequences, fast convolution is done using a block processing approach with either overlap-save or overlap-add methods. Overlap-save discards samples at the block edges while overlap-add overlaps and adds samples at the block edges to combine the results.