Efficient imageToColumn implementation at CUDA

Question

All

Is there any efficient implementation of ImageToColumn that converts the feature map of convolution operation to GEMM at the CUDA level? Because I found