Image computing in OpenCL (image2D or array)

Hi all,

I want to make a convolution operation on an image. So my question is regarding the use of Image2D from OpenCL. Is it better to convert the image that I read in the host to an array and send this array to gpu to convolve it?.
I am wondering which way will be faster?.



Best way would be openGL-openCL interop. If the image is already loaded in openGL you should look for this for best performance. I have tried both for very lage convolution operations (200² kernelsize) an image access was slower.