use vulkano::buffer::{BufferUsage, CpuAccessibleBuffer, DeviceLocalBuffer, ImmutableBuffer, BufferAccess};
use vulkano::command_buffer::{AutoCommandBufferBuilder, DynamicState};
use vulkano::descriptor::descriptor_set::{PersistentDescriptorSet, StdDescriptorPoolAlloc};
use vulkano::device::{Device, DeviceExtensions, QueuesIter, Queue};
use vulkano::framebuffer::{Framebuffer, FramebufferAbstract, RenderPassAbstract, Subpass};
use vulkano::instance::{Instance, InstanceExtensions, PhysicalDevice, QueueFamily};
use vulkano::pipeline::{ComputePipeline, GraphicsPipeline};
use vulkano::pipeline::viewport::Viewport;
use vulkano::sync::{FlushError, GpuFuture};
use vulkano::sync;
use vulkano::image::SwapchainImage;
use vulkano::swapchain::{AcquireError, PresentMode, SurfaceTransform, Swapchain, SwapchainCreationError};
use vulkano::swapchain;
use std::time::SystemTime;
use std::sync::Arc;
use std::ffi::CStr;
use std::path::PathBuf;
use shade_runner as sr;
use image::{DynamicImage, GenericImage, GenericImageView, ImageBuffer};
use vulkano::descriptor::pipeline_layout::PipelineLayout;
use shade_runner::{ComputeLayout, CompileError};
use vulkano::descriptor::descriptor_set::PersistentDescriptorSetBuf;
use shaderc::CompileOptions;
use winit::{Event, EventsLoop, Window, WindowBuilder, WindowEvent};
use vulkano_win::VkSurfaceBuild;
use vulkano::SafeDeref;



fn main() {
    let instance = {
        let extensions = vulkano_win::required_extensions();
        Instance::new(None, &extensions, None).unwrap()
    };

    let physical = PhysicalDevice::enumerate(&instance).next().unwrap();

    // The objective of this example is to draw a triangle on a window. To do so, we first need to
    // create the window.
    //
    // This is done by creating a `WindowBuilder` from the `winit` crate, then calling the
    // `build_vk_surface` method provided by the `VkSurfaceBuild` trait from `vulkano_win`. If you
    // ever get an error about `build_vk_surface` being undefined in one of your projects, this
    // probably means that you forgot to import this trait.
    //
    // This returns a `vulkano::swapchain::Surface` object that contains both a cross-platform winit
    // window and a cross-platform Vulkan surface that represents the surface of the window.
    let mut events_loop = EventsLoop::new();

    let surface = WindowBuilder::new().build_vk_surface(&events_loop, instance.clone()).unwrap();
    let window = surface.window();

    let queue_family = physical.queue_families().find(|&q| {
        // We take the first queue that supports drawing to our window.
        q.supports_graphics() &&
            surface.is_supported(q).unwrap_or(false) &&
            q.supports_compute()
    }).unwrap();

    let device_ext = DeviceExtensions { khr_swapchain: true, ..DeviceExtensions::none() };
    let (device, mut queues) = Device::new(physical, physical.supported_features(), &device_ext,
                                           [(queue_family, 0.5)].iter().cloned()).unwrap();

    let queue = queues.next().unwrap();

    // Before we can draw on the surface, we have to create what is called a swapchain. Creating
    // a swapchain allocates the color buffers that will contain the image that will ultimately
    // be visible on the screen. These images are returned alongside with the swapchain.
    let (mut swapchain, images) = {
        // Querying the capabilities of the surface. When we create the swapchain we can only
        // pass values that are allowed by the capabilities.
        let capabilities = surface.capabilities(physical).unwrap();

        let usage = capabilities.supported_usage_flags;

        // The alpha mode indicates how the alpha value of the final image will behave. For example
        // you can choose whether the window will be opaque or transparent.
        let alpha = capabilities.supported_composite_alpha.iter().next().unwrap();

        // Choosing the internal format that the images will have.
        let format = capabilities.supported_formats[0].0;

        // The dimensions of the window, only used to initially setup the swapchain.
        // NOTE:
        // On some drivers the swapchain dimensions are specified by `caps.current_extent` and the
        // swapchain size must use these dimensions.
        // These dimensions are always the same as the window dimensions
        //
        // However other drivers dont specify a value i.e. `caps.current_extent` is `None`
        // These drivers will allow anything but the only sensible value is the window dimensions.
        //
        // Because for both of these cases, the swapchain needs to be the window dimensions, we just use that.
        let initial_dimensions = if let Some(dimensions) = window.get_inner_size() {
            // convert to physical pixels
            let dimensions: (u32, u32) = dimensions.to_physical(window.get_hidpi_factor()).into();
            [dimensions.0, dimensions.1]
        } else {
            // The window no longer exists so exit the application.
            return;
        };

        // Please take a look at the docs for the meaning of the parameters we didn't mention.
        Swapchain::new(device.clone(), surface.clone(), capabilities.min_image_count, format,
                       initial_dimensions, 1, usage, &queue, SurfaceTransform::Identity, alpha,
                       PresentMode::Fifo, true, None).unwrap()
    };


    // We now create a buffer that will store the shape of our triangle.
    let vertex_buffer = {
        #[derive(Default, Debug, Clone)]
        struct Vertex { position: [f32; 2] }
        vulkano::impl_vertex!(Vertex, position);

        CpuAccessibleBuffer::from_iter(device.clone(), BufferUsage::all(), [
            Vertex { position: [-0.5, -0.25] },
            Vertex { position: [0.0, 0.5] },
            Vertex { position: [0.25, -0.1] }
        ].iter().cloned()).unwrap()
    };

    mod vs {
        vulkano_shaders::shader! {
            ty: "vertex",
            src: "
#version 450

layout(location = 0) in vec2 position;

void main() {
    gl_Position = vec4(position, 0.0, 1.0);
}"
        }
    }

    mod fs {
        vulkano_shaders::shader! {
            ty: "fragment",
            src: "
#version 450

layout(location = 0) out vec4 f_color;

void main() {
    f_color = vec4(1.0, 0.0, 0.0, 1.0);
}
"
        }
    }

    let vs = vs::Shader::load(device.clone()).unwrap();
    let fs = fs::Shader::load(device.clone()).unwrap();

    // The next step is to create a *render pass*, which is an object that describes where the
    // output of the graphics pipeline will go. It describes the layout of the images
    // where the colors, depth and/or stencil information will be written.
    let render_pass = Arc::new(vulkano::single_pass_renderpass!(
        device.clone(),
        attachments: {
            // `color` is a custom name we give to the first and only attachment.
            color: {
                // `load: Clear` means that we ask the GPU to clear the content of this
                // attachment at the start of the drawing.
                load: Clear,
                // `store: Store` means that we ask the GPU to store the output of the draw
                // in the actual image. We could also ask it to discard the result.
                store: Store,
                // `format: <ty>` indicates the type of the format of the image. This has to
                // be one of the types of the `vulkano::format` module (or alternatively one
                // of your structs that implements the `FormatDesc` trait). Here we use the
                // same format as the swapchain.
                format: swapchain.format(),
                // TODO:
                samples: 1,
            }
        },
        pass: {
            // We use the attachment named `color` as the one and only color attachment.
            color: [color],
            // No depth-stencil attachment is indicated with empty brackets.
            depth_stencil: {}
        }
    ).unwrap());

    // Before we draw we have to create what is called a pipeline. This is similar to an OpenGL
    // program, but much more specific.
    let pipeline = Arc::new(GraphicsPipeline::start()
        // We need to indicate the layout of the vertices.
        // The type `SingleBufferDefinition` actually contains a template parameter corresponding
        // to the type of each vertex. But in this code it is automatically inferred.
        .vertex_input_single_buffer()
        // A Vulkan shader can in theory contain multiple entry points, so we have to specify
        // which one. The `main` word of `main_entry_point` actually corresponds to the name of
        // the entry point.
        .vertex_shader(vs.main_entry_point(), ())
        // The content of the vertex buffer describes a list of triangles.
        .triangle_list()
        // Use a resizable viewport set to draw over the entire window
        .viewports_dynamic_scissors_irrelevant(1)
        // See `vertex_shader`.
        .fragment_shader(fs.main_entry_point(), ())
        // We have to indicate which subpass of which render pass this pipeline is going to be used
        // in. The pipeline will only be usable from this particular subpass.
        .render_pass(Subpass::from(render_pass.clone(), 0).unwrap())
        // Now that our builder is filled, we call `build()` to obtain an actual pipeline.
        .build(device.clone())
        .unwrap());

    // Dynamic viewports allow us to recreate just the viewport when the window is resized
    // Otherwise we would have to recreate the whole pipeline.
    let mut dynamic_state = DynamicState { line_width: None, viewports: None, scissors: None };

    // The render pass we created above only describes the layout of our framebuffers. Before we
    // can draw we also need to create the actual framebuffers.
    //
    // Since we need to draw to multiple images, we are going to create a different framebuffer for
    // each image.
    let mut framebuffers = window_size_dependent_setup(&images, render_pass.clone(), &mut dynamic_state);

    // Initialization is finally finished!

    // In some situations, the swapchain will become invalid by itself. This includes for example
    // when the window is resized (as the images of the swapchain will no longer match the
    // window's) or, on Android, when the application went to the background and goes back to the
    // foreground.
    //
    // In this situation, acquiring a swapchain image or presenting it will return an error.
    // Rendering to an image of that swapchain will not produce any error, but may or may not work.
    // To continue rendering, we need to recreate the swapchain by creating a new swapchain.
    // Here, we remember that we need to do this for the next loop iteration.
    let mut recreate_swapchain = false;

    // In the loop below we are going to submit commands to the GPU. Submitting a command produces
    // an object that implements the `GpuFuture` trait, which holds the resources for as long as
    // they are in use by the GPU.
    //
    // Destroying the `GpuFuture` blocks until the GPU is finished executing it. In order to avoid
    // that, we store the submission of the previous frame here.
    let mut previous_frame_end = Box::new(sync::now(device.clone())) as Box<dyn GpuFuture>;

    loop {
        // It is important to call this function from time to time, otherwise resources will keep
        // accumulating and you will eventually reach an out of memory error.
        // Calling this function polls various fences in order to determine what the GPU has
        // already processed, and frees the resources that are no longer needed.
        previous_frame_end.cleanup_finished();

        // Whenever the window resizes we need to recreate everything dependent on the window size.
        // In this example that includes the swapchain, the framebuffers and the dynamic state viewport.
        if recreate_swapchain {
            // Get the new dimensions of the window.
            let dimensions = if let Some(dimensions) = window.get_inner_size() {
                let dimensions: (u32, u32) = dimensions.to_physical(window.get_hidpi_factor()).into();
                [dimensions.0, dimensions.1]
            } else {
                return;
            };

            let (new_swapchain, new_images) = match swapchain.recreate_with_dimension(dimensions) {
                Ok(r) => r,
                // This error tends to happen when the user is manually resizing the window.
                // Simply restarting the loop is the easiest way to fix this issue.
                Err(SwapchainCreationError::UnsupportedDimensions) => continue,
                Err(err) => panic!("{:?}", err)
            };

            swapchain = new_swapchain;
            // Because framebuffers contains an Arc on the old swapchain, we need to
            // recreate framebuffers as well.
            framebuffers = window_size_dependent_setup(&new_images, render_pass.clone(), &mut dynamic_state);

            recreate_swapchain = false;
        }

        // Before we can draw on the output, we have to *acquire* an image from the swapchain. If
        // no image is available (which happens if you submit draw commands too quickly), then the
        // function will block.
        // This operation returns the index of the image that we are allowed to draw upon.
        //
        // This function can block if no image is available. The parameter is an optional timeout
        // after which the function call will return an error.
        let (image_num, acquire_future) = match swapchain::acquire_next_image(swapchain.clone(), None) {
            Ok(r) => r,
            Err(AcquireError::OutOfDate) => {
                recreate_swapchain = true;
                continue;
            }
            Err(err) => panic!("{:?}", err)
        };

        // Specify the color to clear the framebuffer with i.e. blue
        let clear_values = vec!([0.0, 0.0, 1.0, 1.0].into());


        {
            let project_root =
                std::env::current_dir()
                    .expect("failed to get root directory");

            let mut compute_path = project_root.clone();
            compute_path.push(PathBuf::from("resources/shaders/"));
            compute_path.push(PathBuf::from("simple-edge.compute"));


            let mut options = CompileOptions::new().ok_or(CompileError::CreateCompiler).unwrap();
            options.add_macro_definition("SETTING_POS_X", Some("0"));
            options.add_macro_definition("SETTING_POS_Y", Some("1"));
            options.add_macro_definition("SETTING_BUCKETS_START", Some("2"));
            options.add_macro_definition("SETTING_BUCKETS_LEN", Some("2"));

            let shader =
                shade_runner::load_compute_with_options(compute_path, options)
                    .expect("Failed to compile");

            let vulkano_entry =
                shade_runner::parse_compute(&shader)
                    .expect("failed to parse");

            let x = unsafe {
                vulkano::pipeline::shader::ShaderModule::from_words(device.clone(), &shader.compute)
            }.unwrap();

            let c_pipeline = Arc::new({
                unsafe {
                    ComputePipeline::new(device.clone(), &x.compute_entry_point(
                        CStr::from_bytes_with_nul_unchecked(b"main\0"),
                        vulkano_entry.compute_layout), &(),
                    ).unwrap()
                }
            });

            let project_root =
                std::env::current_dir()
                    .expect("failed to get root directory");

            let mut compute_path = project_root.clone();
            compute_path.push(PathBuf::from("resources/images/"));
            compute_path.push(PathBuf::from("funky-bird.jpg"));

            let img = image::open(compute_path).expect("Couldn't find image");

            let xy = img.dimensions();

            let data_length = xy.0 * xy.1 * 4;
            let pixel_count = img.raw_pixels().len();
            println!("Pixel count {}", pixel_count);

            let mut image_buffer = Vec::new();
            if pixel_count != data_length as usize {
                println!("Creating apha channel...");
                for i in img.raw_pixels().iter() {
                    if (image_buffer.len() + 1) % 4 == 0 {
                        image_buffer.push(255);
                    }
                    image_buffer.push(*i);
                }
                image_buffer.push(255);
            } else {
                image_buffer = img.raw_pixels();
            }

            println!("Buffer length {}", image_buffer.len());
            println!("Size {:?}", xy);

            println!("Allocating Buffers...");

            // Pull out the image data and place it in a buffer for the kernel to write to and for us to read from
            let write_buffer = {
                let mut buff = image_buffer.iter();
                let data_iter = (0..data_length).map(|n| *(buff.next().unwrap()));
                CpuAccessibleBuffer::from_iter(device.clone(), BufferUsage::all(), data_iter).unwrap()
            };


            // Pull out the image data and place it in a buffer for the kernel to read from
            let read_buffer = {
                let mut buff = image_buffer.iter();
                let data_iter = (0..data_length).map(|n| *(buff.next().unwrap()));
                CpuAccessibleBuffer::from_iter(device.clone(), BufferUsage::all(), data_iter).unwrap()
            };


            // A buffer to hold many i32 values to use as settings
            let settings_buffer = {
                let vec = vec![xy.0, xy.1];
                let mut buff = vec.iter();
                let data_iter =
                    (0..2).map(|n| *(buff.next().unwrap()));
                CpuAccessibleBuffer::from_iter(device.clone(),
                                               BufferUsage::all(),
                                               data_iter).unwrap()
            };

            println!("Done");

            // Create the data descriptor set for our previously created shader pipeline
            let mut set =
                PersistentDescriptorSet::start(c_pipeline.clone(), 0)
                    .add_buffer(write_buffer.clone()).unwrap()
                    .add_buffer(read_buffer.clone()).unwrap()
                    .add_buffer(settings_buffer.clone()).unwrap();

            let mut set = Arc::new(set.build().unwrap());




            // In order to draw, we have to build a *command buffer*. The command buffer object holds
            // the list of commands that are going to be executed.
            //
            // Building a command buffer is an expensive operation (usually a few hundred
            // microseconds), but it is known to be a hot path in the driver and is expected to be
            // optimized.
            //
            // Note that we have to pass a queue family when we create the command buffer. The command
            // buffer will only be executable on that given queue family.
            let command_buffer =
                AutoCommandBufferBuilder::primary_one_time_submit(device.clone(), queue.family())
                    .unwrap()

                    .dispatch([xy.0, xy.1, 1],
                              c_pipeline.clone(),
                              set.clone(), ()).unwrap()
                    // Before we can draw, we have to *enter a render pass*. There are two methods to do
                    // this: `draw_inline` and `draw_secondary`. The latter is a bit more advanced and is
                    // not covered here.
                    //
                    // The third parameter builds the list of values to clear the attachments with. The API
                    // is similar to the list of attachments when building the framebuffers, except that
                    // only the attachments that use `load: Clear` appear in the list.
                    .begin_render_pass(framebuffers[image_num].clone(), false, clear_values)
                    .unwrap()


                    // We are now inside the first subpass of the render pass. We add a draw command.
                    //
                    // The last two parameters contain the list of resources to pass to the shaders.
                    // Since we used an `EmptyPipeline` object, the objects have to be `()`.
                    .draw(pipeline.clone(), &dynamic_state, vertex_buffer.clone(), (), ())
                    .unwrap()

                    // We leave the render pass by calling `draw_end`. Note that if we had multiple
                    // subpasses we could have called `next_inline` (or `next_secondary`) to jump to the
                    // next subpass.
                    .end_render_pass()
                    .unwrap()

                    // Finish building the command buffer by calling `build`.
                    .build().unwrap();

            let future = previous_frame_end.join(acquire_future)
                .then_execute(queue.clone(), command_buffer).unwrap()

                // The color output is now expected to contain our triangle. But in order to show it on
                // the screen, we have to *present* the image by calling `present`.
                //
                // This function does not actually present the image immediately. Instead it submits a
                // present command at the end of the queue. This means that it will only be presented once
                // the GPU has finished executing the command buffer that draws the triangle.
                .then_swapchain_present(queue.clone(), swapchain.clone(), image_num)
                .then_signal_fence_and_flush();

            match future {
                Ok(future) => {
                    previous_frame_end = Box::new(future) as Box<_>;
                }
                Err(FlushError::OutOfDate) => {
                    recreate_swapchain = true;
                    previous_frame_end = Box::new(sync::now(device.clone())) as Box<_>;
                }
                Err(e) => {
                    println!("{:?}", e);
                    previous_frame_end = Box::new(sync::now(device.clone())) as Box<_>;
                }
            }
        }
        // Note that in more complex programs it is likely that one of `acquire_next_image`,
        // `command_buffer::submit`, or `present` will block for some time. This happens when the
        // GPU's queue is full and the driver has to wait until the GPU finished some work.
        //
        // Unfortunately the Vulkan API doesn't provide any way to not wait or to detect when a
        // wait would happen. Blocking may be the desired behavior, but if you don't want to
        // block you should spawn a separate thread dedicated to submissions.

        // Handling the window events in order to close the program when the user wants to close
        // it.
        let mut done = false;
        events_loop.poll_events(|ev| {
            match ev {
                Event::WindowEvent { event: WindowEvent::CloseRequested, .. } => done = true,
                Event::WindowEvent { event: WindowEvent::Resized(_), .. } => recreate_swapchain = true,
                _ => ()
            }
        });
        if done { return; }
    }
}

/// This method is called once during initialization, then again whenever the window is resized
fn window_size_dependent_setup(
    images: &[Arc<SwapchainImage<Window>>],
    render_pass: Arc<dyn RenderPassAbstract + Send + Sync>,
    dynamic_state: &mut DynamicState,
) -> Vec<Arc<dyn FramebufferAbstract + Send + Sync>> {
    let dimensions = images[0].dimensions();

    let viewport = Viewport {
        origin: [0.0, 0.0],
        dimensions: [dimensions[0] as f32, dimensions[1] as f32],
        depth_range: 0.0..1.0,
    };
    dynamic_state.viewports = Some(vec!(viewport));

    images.iter().map(|image| {
        Arc::new(
            Framebuffer::start(render_pass.clone())
                .add(image.clone()).unwrap()
                .build().unwrap()
        ) as Arc<dyn FramebufferAbstract + Send + Sync>
    }).collect::<Vec<_>>()
}