Fast wave-front reconstruction methods are becoming increasingly important, for example, in large astronomical adaptive optics systems and high spatial resolution shear interferometry, where pseudoinverse matrix methods scale poorly with problem size. Wave-front reconstruction from difference measurements can be achieved by use of fast implementations of the discrete Fourier transform (DFT), obtaining performance comparable with that of the pseudoinverse in terms of the noise propagation coefficient. Existing methods that are based on the use of the DFT give exact results (in the absence of noise) only for the particular case in which the shear is a divisor of the number of samples to be reconstructed. We present two alternate solutions for the more general case when the shear is any integer. In the first solution the dimensions of the problem are enlarged, and in the second the problem is subdivided into a set of smaller problems with shear amplitude equal to one. We also show that the retrieved solutions have minimum norm and calculate the noise propagation coefficient for both methods. The proposed algorithms are implemented and timed against pseudoinverse multiplication. The results show a speed increase by a factor of 50 over the pseudoinverse multiplication for a grid with N = 3 x 10(3) samples.