So for a candidate submatrix dimension m-by-n, multiply the elements of an FFT of an m-by-n matrix of ones with the FFT of A, then compute the inverse 2D FFT of the result ... that is wrapped around A ...

Abstract: Despite C/C++ and Python ... syntax. NumPy's potential, however, is limited by its inability to run on graphics processing units (GPUs), processors specialized for handling computations. On ...