This paper proposes an adaptive approach for reducing the computational load of computer-generated holograms (CGHs). Instead of using the whole hologram plate resolution or carrying out point-wise judgment, this approach pre-divides the object space into subspaces and calculates an effective hologram plate region for each subspace according to interference fringe spatial frequency and grating diffraction energy distribution. As both the very low and very high spatial frequency portions are filtered out, the quality of the reconstructed image is even better in terms of sharpness. This space subdivision based approach also makes parallel CGH computing more straightforward. A simple load-balancing strategy is given.