I’m trying to do some image processing using Cuda. I’ve read the articles regarding convolution in Cuda and am following most of the strategies. However, My images are not 4-bytes per pixel (like all the examples I’m seeing) but instead 8-bit per pixel grayscales. This seems like it should be beneficial as it will allow a greater number of pixels into each block’s shared memory. My strategy then is to copy the 8-bit pixels into shared memory and then cast them as floats as they are accessed and operated upon in the kernel. Is this okay, or is there a problem doing casts in Cuda? Thanks in advance.