vulkano/examples/specialization-constants/main.rs

// TODO: Give a paragraph about what specialization are and what problems they solve.

use std::sync::Arc;
use vulkano::{
    buffer::{Buffer, BufferCreateInfo, BufferUsage},
    command_buffer::{
        allocator::StandardCommandBufferAllocator, CommandBufferUsage, RecordingCommandBuffer,
    },
    descriptor_set::{
        allocator::StandardDescriptorSetAllocator, DescriptorSet, WriteDescriptorSet,
    },
    device::{
        physical::PhysicalDeviceType, Device, DeviceCreateInfo, DeviceExtensions, QueueCreateInfo,
        QueueFlags,
    },
    instance::{Instance, InstanceCreateFlags, InstanceCreateInfo},
    memory::allocator::{AllocationCreateInfo, MemoryTypeFilter, StandardMemoryAllocator},
    pipeline::{
        compute::ComputePipelineCreateInfo, layout::PipelineDescriptorSetLayoutCreateInfo,
        ComputePipeline, Pipeline, PipelineBindPoint, PipelineLayout,
        PipelineShaderStageCreateInfo,
    },
    sync::{self, GpuFuture},
    VulkanLibrary,
};

fn main() {
    let library = VulkanLibrary::new().unwrap();
    let instance = Instance::new(
        library,
        InstanceCreateInfo {
            flags: InstanceCreateFlags::ENUMERATE_PORTABILITY,
            ..Default::default()
        },
    )
    .unwrap();

    let device_extensions = DeviceExtensions {
        khr_storage_buffer_storage_class: true,
        ..DeviceExtensions::empty()
    };
    let (physical_device, queue_family_index) = instance
        .enumerate_physical_devices()
        .unwrap()
        .filter(|p| p.supported_extensions().contains(&device_extensions))
        .filter_map(|p| {
            p.queue_family_properties()
                .iter()
                .position(|q| q.queue_flags.intersects(QueueFlags::COMPUTE))
                .map(|i| (p, i as u32))
        })
        .min_by_key(|(p, _)| match p.properties().device_type {
            PhysicalDeviceType::DiscreteGpu => 0,
            PhysicalDeviceType::IntegratedGpu => 1,
            PhysicalDeviceType::VirtualGpu => 2,
            PhysicalDeviceType::Cpu => 3,
            PhysicalDeviceType::Other => 4,
            _ => 5,
        })
        .unwrap();

    println!(
        "using device: {} (type: {:?})",
        physical_device.properties().device_name,
        physical_device.properties().device_type,
    );

    let (device, mut queues) = Device::new(
        physical_device,
        DeviceCreateInfo {
            enabled_extensions: device_extensions,
            queue_create_infos: vec![QueueCreateInfo {
                queue_family_index,
                ..Default::default()
            }],
            ..Default::default()
        },
    )
    .unwrap();

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

    mod cs {
        vulkano_shaders::shader! {
            ty: "compute",
            src: r"
                #version 450

                layout(local_size_x = 64, local_size_y = 1, local_size_z = 1) in;

                layout(constant_id = 0) const int multiple = 64;
                layout(constant_id = 1) const float addend = 64;
                layout(constant_id = 2) const bool enable = true;

                layout(set = 0, binding = 0) buffer Data {
                    uint data[];
                };

                void main() {
                    uint idx = gl_GlobalInvocationID.x;
                    if (enable) {
                        data[idx] *= multiple;
                        data[idx] += uint(addend);
                    }
                }
            ",
        }
    }

    let pipeline = {
        let cs = cs::load(device.clone())
            .unwrap()
            .specialize(
                [(0, 1i32.into()), (1, 1.0f32.into()), (2, true.into())]
                    .into_iter()
                    .collect(),
            )
            .unwrap()
            .entry_point("main")
            .unwrap();
        let stage = PipelineShaderStageCreateInfo::new(cs);
        let layout = PipelineLayout::new(
            device.clone(),
            PipelineDescriptorSetLayoutCreateInfo::from_stages([&stage])
                .into_pipeline_layout_create_info(device.clone())
                .unwrap(),
        )
        .unwrap();

        ComputePipeline::new(
            device.clone(),
            None,
            ComputePipelineCreateInfo::stage_layout(stage, layout),
        )
        .unwrap()
    };

    let memory_allocator = Arc::new(StandardMemoryAllocator::new_default(device.clone()));
    let descriptor_set_allocator = Arc::new(StandardDescriptorSetAllocator::new(
        device.clone(),
        Default::default(),
    ));
    let command_buffer_allocator = Arc::new(StandardCommandBufferAllocator::new(
        device.clone(),
        Default::default(),
    ));

    let data_buffer = Buffer::from_iter(
        memory_allocator,
        BufferCreateInfo {
            usage: BufferUsage::STORAGE_BUFFER,
            ..Default::default()
        },
        AllocationCreateInfo {
            memory_type_filter: MemoryTypeFilter::PREFER_DEVICE
                | MemoryTypeFilter::HOST_RANDOM_ACCESS,
            ..Default::default()
        },
        0..65536u32,
    )
    .unwrap();

    let layout = &pipeline.layout().set_layouts()[0];
    let descriptor_set = DescriptorSet::new(
        descriptor_set_allocator,
        layout.clone(),
        [WriteDescriptorSet::buffer(0, data_buffer.clone())],
        [],
    )
    .unwrap();

    let mut builder = RecordingCommandBuffer::primary(
        command_buffer_allocator,
        queue.queue_family_index(),
        CommandBufferUsage::OneTimeSubmit,
    )
    .unwrap();

    builder
        .bind_pipeline_compute(pipeline.clone())
        .unwrap()
        .bind_descriptor_sets(
            PipelineBindPoint::Compute,
            pipeline.layout().clone(),
            0,
            descriptor_set,
        )
        .unwrap();

    unsafe {
        builder.dispatch([1024, 1, 1]).unwrap();
    }

    let command_buffer = builder.end().unwrap();

    let future = sync::now(device)
        .then_execute(queue, command_buffer)
        .unwrap()
        .then_signal_fence_and_flush()
        .unwrap();

    future.wait(None).unwrap();

    let data_buffer_content = data_buffer.read().unwrap();
    for n in 0..65536u32 {
        assert_eq!(data_buffer_content[n as usize], n + 1);
    }
    println!("Success");
}