I am trying to write a program which will use my 8800 GTS to convolve a 2d image, as in the seperable example, but instead of using a fixed convolution kernel for every data element, each data element must have its own individual convolution kernel. I’m not sure how to make this edit, but i imagine it will involve storing the kernels as a large array of kernels on the GPU and referencing the specific kernel to each data element. The kernels are likely to be fixed throughout the simulation, so i was hoping to be able to make only a single write to the GPU (i believe texture memory is the correct place for this?). If someone could please give me a little help in how to create, populate and reference such memory when programming the kernel function on the GPU, it would be much appreciated.
Thanks so much for your time, i’m racking my brains on how to complete this, its the last step i have to complete