How to implement SIFT descriptors if the only data available are the keypoints from VLFEAT?

So I used VLFEAT library in order to determine keypoints in an image (which is just one line of code). Now given the keypoints, I need to implement my own SIFT descriptor.

Here is what I have done for keypoint detection:

a=single(rgb2gray(imread('image1.jpg')));
kpl = vl_sift(a)

Now I need to implement a SIFT descriptor based on the given keypoints kp1. Please help me with this....