I am evaluating the MKL for creating a signal processing library. My specific needs are to make use of the 1/2/3-d dft functions. Testing the 1d real dft function gives expected results. Testing the 2d real dft vs. the 2d complex dft yields different answers (ignoring the complex conjugate symmetry). The 3x3 matrix is a linear, contiguous array, so the strides are 0,N,1. Other than memory allocation, and loading real vs. complex data, the real and complex code is identical. Is there a bug here, or am I missing something? I can display code if necessary.