I wonder if I could send a framebuffer image to a compute shader without loading the image from the framebuffer in the CPU side and then binding it to the compute pipeline again. Can I do it just through the GPU side? or can I just bind the framebuffer image or swapchain image to the compute shader? Is that legal?
When you say “framebuffer image” do you mean swapchain image, i.e. ones retrieved with
Whether sampling from a swapchain image is allowed depends on the
VkSurfaceCapabilitiesKHR.supportedUsageFlags that you can query with
OK, I just added the swapchain imageview to my descriptor binding. sampling from swap chain didn’t work for me on an integrated intel gpu, but I was able to do it on a good AMD GPU.