Yeah, to be more specific the problem with doing Bresenham on GPU is that, in the traditional GPU pipeline, the main way to create tasks is to use the fixed-fun

Yeah, to be more specific the problem with doing Bresenham on GPU is that, in the traditional GPU pipeline, the main way to create tasks is to use the fixed-function triangle rasterization hardware. I guess one could technically do something like figuring out the number of pixels spanned on CPU and then rasterizing point primitives. In your vertex shader you'd have each 1x1 pixel point figure out, in parallel, where it's supposed to go. This is of course bound to be horrendously slow and would never be practical :)