Update cases for gradweight ff_tiling.
Allow DOUBLE_CONFIG for all fft_tiling computations in cuDNN V6.