cuDNN high performance NHWC support

Hello! When (if it’s planned at all) can we expect high performance NHWC kernels in cuDNN?