No API

A Vulkan implementation/interpretation of Sebastian Aaltonen's blog post: "No Graphics API".

Usage

The usage is pretty straightforward and as close as it could be to what described in Aaltonen's blog post, the only major difference is located in the texture heap allocation as vulkan requires the backing buffer size to be that of the descriptors.

main :: (){
	set_working_directory(#filepath);
	window := create_window(1280, 720, "Window Creation");
	quit := false;

	gpu_init(1, {.GENERAL, 1});
	gpu_create_swapchain(window, 1280, 720, .FIFO);
	queue := gpu_create_queue(.GENERAL);
	
	compute_ir := gpu_generate_shader_ir("res/compute.spv");
	pipeline := gpu_create_compute_pipeline(compute_ir);
	backbuffer := gpu_create_texture({1280, 720, .R8G8B8A8_UNORM, .STORAGE | .COLOR_ATTACHMENT | .TRANSFER_SRC | .TRANSFER_DST });

	rw_texture_heap := gpu_create_rw_texture_heap();
	gpu_create_rw_texture_view_descriptor(backbuffer, rw_texture_heap, 0);

	while !quit {
        update_window_events();
		for get_window_resizes() {
            gpu_resize_swapchain(it.width, it.height);
		}
		
		command_buffer := gpu_start_command_recording(*queue);
		
		gpu_bind_pipeline(*command_buffer, *pipeline);
		gpu_bind_texture_heap(command_buffer, null, rw_texture_heap);
		gpu_dispatch(command_buffer, null, 1280 / 16, 720 / 16, 1);
	
		if gpu_present(command_buffer, *queue, *backbuffer) then break;
		gpu_end_command_recording(command_buffer);
		
		gpu_submit(*queue, frame_semaphore, frame + 1, command_buffer);
    }
}

Problems

• As vulkan is still a pretty much buffer-centric API most calls dealing with memory need to go through a tree lookup to transform buffer + offset into a VkBuffer.