First attempt at a GPU runner

Currently failing with an error I don't understand:

```
wgpu error: Validation Error

Caused by:
  In Device::create_shader_module, label = '...\image_binary.spv'

Shader '...\image_binary.spv' parsing error: InvalidTypeWidth(1)
```

examples/image-runner/Cargo.toml

@@ -0,0 +1,27 @@
+[package]
+name = "image-runner"
+version.workspace = true
+authors.workspace = true
+edition.workspace = true
+license.workspace = true
+repository.workspace = true
+
+[dependencies]
+enkou-shaders = { path = "../../enkou-shaders" }
+image-binary = { path = "../image-binary", features = ["wgpu"] }
+
+anyhow.workspace = true
+bytemuck.workspace = true
+futures.workspace = true
+glam = { workspace = true, features = ["u8"] }
+image.workspace = true
+tempfile.workspace = true
+wgpu.workspace = true
+xflags.workspace = true
+
+[build-dependencies]
+anyhow.workspace = true
+cargo-gpu-install.workspace = true
+
+[lints]
+workspace = true

examples/image-runner/build.rs

@@ -0,0 +1,28 @@
+use cargo_gpu_install::install::Install;
+use cargo_gpu_install::spirv_builder::{Capability, ShaderPanicStrategy, SpirvMetadata};
+use std::path::PathBuf;
+
+pub fn main() -> anyhow::Result<()> {
+    let manifest_dir = env!("CARGO_MANIFEST_DIR");
+    let crate_path = [manifest_dir, "..", "image-binary"]
+        .iter()
+        .copied()
+        .collect::<PathBuf>();
+
+    let install = Install::from_shader_crate(crate_path.clone())
+        .within_build_script()
+        .run()?;
+    let mut builder = install.to_spirv_builder(crate_path, "spirv-unknown-vulkan1.3");
+    builder.build_script.defaults = true;
+    builder.shader_panic_strategy = ShaderPanicStrategy::SilentExit;
+    builder.spirv_metadata = SpirvMetadata::Full;
+    builder.capabilities = vec![Capability::Int8, Capability::Int16, Capability::Int64];
+
+    let compile_result = builder.build()?;
+    let spv_path = compile_result.module.unwrap_single();
+    println!(
+        "cargo::rustc-env=SHADER_SPV_PATH_IMAGE_BINARY={}",
+        spv_path.display()
+    );
+    Ok(())
+}

examples/image-runner/src/image_binary.rs

@@ -0,0 +1,210 @@
+use enkou_shaders::Coefficients2;
+use enkou_shaders::transform::Transform;
+use enkou_shaders::variation::Variation;
+use futures::channel::oneshot;
+use futures::executor::block_on;
+use glam::{uvec2, Affine2, UVec2, UVec4, Vec2};
+use image::{Rgba, RgbaImage};
+use image_binary::{main_image_binary, BIND_GROUP_IMAGE_BINARY, BGLE_IMAGE_DIMENSIONS, BGLE_TRANSFORMS, BGLE_WEIGHTS, BGLE_CAMERA, BGLE_IMAGE_BUFFER, BGLE_VARIATIONS};
+use std::path::Path;
+use wgpu::util::DeviceExt;
+use enkou_shaders::camera::Camera;
+
+fn transforms() -> [Transform; 3] {
+    [
+        {
+            // F_0: (x / 2, y / 2)
+            let coefficients = Affine2::from_coefficients(0.5, 0.0, 0.0, 0.0, 0.5, 0.0);
+            Transform::new(coefficients, Affine2::IDENTITY, uvec2(0, 1))
+        },
+        {
+            // F_1: ((x + 1) / 2, y / 2)
+            let coefficients = Affine2::from_coefficients(0.5, 0.0, 0.5, 0.0, 0.5, 0.0);
+            Transform::new(coefficients, Affine2::IDENTITY, uvec2(0, 1))
+        },
+        {
+            // F_2: (x / 2, (y + 1) / 2)
+            let coefficients = Affine2::from_coefficients(0.5, 0.0, 0.0, 0.0, 0.5, 0.5);
+            Transform::new(coefficients, Affine2::IDENTITY, uvec2(0, 1))
+        },
+    ]
+}
+
+fn weights() -> [f32; 3] {
+    [1.0 / 3.0; 3]
+}
+
+fn variations() -> [Variation; 1] {
+    [Variation::IDENTITY]
+}
+
+fn camera(image_dimensions: UVec2) -> Camera {
+    Camera::new(
+        image_dimensions,
+        Vec2::ONE * 0.5,
+        0.0,
+        Vec2::ZERO,
+        Vec2::splat(image_dimensions.min_element() as f32),
+    )
+}
+
+pub(crate) fn main_cpu(image_dimensions: UVec2, output_path: &Path) -> Result<(), anyhow::Error> {
+    let mut image_buffer = Vec::<UVec4>::new();
+    image_buffer.resize(image_dimensions.element_product() as usize, UVec4::ZERO);
+
+    main_image_binary(
+        &image_dimensions,
+        &transforms(),
+        &weights(),
+        &variations(),
+        &camera(image_dimensions),
+        &mut image_buffer,
+    );
+
+    let mut image = RgbaImage::new(image_dimensions.x, image_dimensions.y);
+    for (i, color) in image_buffer.into_iter().enumerate() {
+        let image_x = i as u32 % image_dimensions.x;
+        let image_y = i as u32 / image_dimensions.x;
+
+        image.put_pixel(image_x, image_y, color.as_u8vec4().to_array().into());
+    }
+
+    image.save(output_path)?;
+    Ok(())
+}
+
+const SHADER_MODULE: wgpu::ShaderModuleDescriptor = wgpu::include_spirv!(env!("SHADER_SPV_PATH_IMAGE_BINARY"));
+
+fn bge<'a>(entry: &'a wgpu::BindGroupLayoutEntry, buffer: &'a wgpu::Buffer) -> wgpu::BindGroupEntry<'a> {
+    wgpu::BindGroupEntry {
+        binding: entry.binding,
+        resource: buffer.as_entire_binding(),
+    }
+}
+
+pub(crate) fn main_gpu(device: &wgpu::Device, queue: &wgpu::Queue, image_dimensions: UVec2, output_path: &Path) -> Result<(), anyhow::Error> {
+
+    let bind_group_layout = device.create_bind_group_layout(&BIND_GROUP_IMAGE_BINARY);
+
+    let image_dimensions_buffer = device.create_buffer_init(&wgpu::util::BufferInitDescriptor {
+        label: Some("image_dimensions"),
+        contents: bytemuck::bytes_of(&image_dimensions),
+        usage: wgpu::BufferUsages::COPY_DST | wgpu::BufferUsages::STORAGE,
+    });
+
+    let transforms_buffer = device.create_buffer_init(&wgpu::util::BufferInitDescriptor {
+        label: Some("transforms"),
+        contents: bytemuck::cast_slice(&transforms()),
+        usage: wgpu::BufferUsages::COPY_DST | wgpu::BufferUsages::STORAGE,
+    });
+
+    let weights_buffer = device.create_buffer_init(&wgpu::util::BufferInitDescriptor {
+        label: Some("weights"),
+        contents: bytemuck::cast_slice(&weights()),
+        usage: wgpu::BufferUsages::COPY_DST | wgpu::BufferUsages::STORAGE,
+    });
+
+    let variations_buffer = device.create_buffer_init(&wgpu::util::BufferInitDescriptor {
+        label: Some("variations"),
+        contents: bytemuck::cast_slice(&variations()),
+        usage: wgpu::BufferUsages::COPY_DST | wgpu::BufferUsages::STORAGE,
+    });
+
+    let camera_buffer = device.create_buffer_init(&wgpu::util::BufferInitDescriptor {
+        label: Some("camera"),
+        contents: bytemuck::bytes_of(&camera(image_dimensions)),
+        usage: wgpu::BufferUsages::COPY_DST | wgpu::BufferUsages::STORAGE,
+    });
+
+    let image_buffer_elements = image_dimensions.element_product() as u64;
+    let image_buffer_size = image_buffer_elements * size_of::<UVec4>() as u64;
+    let image_buffer = device.create_buffer(&wgpu::BufferDescriptor {
+        label: Some("image_buffer"),
+        size: image_buffer_size,
+        usage: wgpu::BufferUsages::COPY_SRC | wgpu::BufferUsages::STORAGE,
+        mapped_at_creation: false,
+    });
+
+    let image_staging = device.create_buffer(&wgpu::BufferDescriptor {
+        label: Some("image_buffer_staging"),
+        size: image_buffer_size,
+        usage: wgpu::BufferUsages::MAP_READ | wgpu::BufferUsages::COPY_DST,
+        mapped_at_creation: false,
+    });
+
+    let bind_group = device.create_bind_group(&wgpu::BindGroupDescriptor {
+        label: Some("main_image_binary"),
+        layout: &bind_group_layout,
+        entries: &[
+            bge(&BGLE_IMAGE_DIMENSIONS, &image_dimensions_buffer),
+            bge(&BGLE_TRANSFORMS, &transforms_buffer),
+            bge(&BGLE_WEIGHTS, &weights_buffer),
+            bge(&BGLE_VARIATIONS, &variations_buffer),
+            bge(&BGLE_CAMERA, &camera_buffer),
+            bge(&BGLE_IMAGE_BUFFER, &image_buffer),
+        ],
+    });
+
+    let layout = device.create_pipeline_layout(&wgpu::PipelineLayoutDescriptor {
+        label: Some("main_image_binary"),
+        bind_group_layouts: &[Some(&bind_group_layout)],
+        immediate_size: 0,
+    });
+
+    let module = device.create_shader_module(SHADER_MODULE);
+
+    let compute_pipeline = device.create_compute_pipeline(&wgpu::ComputePipelineDescriptor {
+        label: Some("main_image_binary"),
+        layout: Some(&layout),
+        module: &module,
+        entry_point: Some("main_image_binary"),
+        compilation_options: Default::default(),
+        cache: None,
+    });
+
+    let mut encoder = device.create_command_encoder(&wgpu::CommandEncoderDescriptor {
+        label: Some("main_image_binary"),
+    });
+
+    {
+        let mut compute_pass = encoder.begin_compute_pass(&wgpu::ComputePassDescriptor {
+            label: Some("main_image_binary"),
+            timestamp_writes: None,
+        });
+
+        compute_pass.set_pipeline(&compute_pipeline);
+        compute_pass.set_bind_group(0, &bind_group, &[]);
+    }
+
+    encoder.copy_buffer_to_buffer(&image_buffer, 0, &image_staging, 0, Some(image_buffer_size));
+
+    let (sender, receiver) = oneshot::channel();
+    let image_staging_capturable = image_staging.clone();
+    encoder.map_buffer_on_submit(&image_buffer, wgpu::MapMode::Read, .., move |result| {
+        result.expect("unable to map buffer");
+
+        let staging_buffer_view = image_staging_capturable.get_mapped_range(..).expect("Unable to map staging buffer");
+
+        let mut image = RgbaImage::new(image_dimensions.x, image_dimensions.y);
+        let image_buffer_elements = bytemuck::cast_slice::<u8, UVec4>(staging_buffer_view.as_ref());
+        for (i, element) in image_buffer_elements.iter().enumerate() {
+            let image_x = i as u32 % image_dimensions.x;
+            let image_y = i as u32 / image_dimensions.x;
+
+            let pixel_colors = element.as_u8vec4();
+            image.put_pixel(image_x, image_y, Rgba(*pixel_colors.as_ref()))
+        }
+
+        sender.send(image).expect("Unable to send image");
+    });
+
+    queue.submit(Some(encoder.finish()));
+    device.poll(wgpu::PollType::wait_indefinitely())?;
+    let image = block_on(receiver)?;
+
+    image_staging.unmap();
+
+    image.save(output_path)?;
+
+    Ok(())
+}

examples/image-runner/src/main.rs

@@ -0,0 +1,77 @@
+use glam::{uvec2};
+use std::mem;
+use std::process::Command;
+use std::path::PathBuf;
+use futures::executor::block_on;
+use tempfile::NamedTempFile;
+
+mod image_binary;
+
+fn main() -> Result<(), anyhow::Error> {
+    let instance_future = wgpu::util::new_instance_with_webgpu_detection(wgpu::InstanceDescriptor {
+        backends: Default::default(),
+        flags: Default::default(),
+        memory_budget_thresholds: Default::default(),
+        backend_options: Default::default(),
+        display: None,
+    });
+    let instance = block_on(instance_future);
+
+    let adapter_future = instance.request_adapter(&wgpu::RequestAdapterOptions {
+        power_preference: Default::default(),
+        force_fallback_adapter: false,
+        compatible_surface: None,
+        apply_limit_buckets: false,
+    });
+    let adapter = block_on(adapter_future)?;
+
+    let device_future = adapter.request_device(&wgpu::DeviceDescriptor {
+        label: Some("image-runner"),
+        required_features: Default::default(),
+        required_limits: Default::default(),
+        experimental_features: Default::default(),
+        memory_hints: Default::default(),
+        trace: Default::default(),
+    });
+    let (device, queue) = block_on(device_future)?;
+
+    let flags = xflags::parse_or_exit! {
+        /// Image dimensions to output, as `width,height`
+        optional -d, --dimensions dimensions: String
+
+        /// Output pathname to use
+        optional -o, --output output: PathBuf
+
+        /// Image type to generate
+        required image: String
+    };
+
+    let dimensions = if let Some(dimensions) = flags.dimensions {
+        let (width_str, height_str) = dimensions.split_once(",").ok_or(anyhow::anyhow!("Invalid format for image dimensions"))?;
+        uvec2(width_str.parse()?, height_str.parse()?)
+    } else {
+        uvec2(1600, 900)
+    };
+
+    let output = if let Some(output) = flags.output { output } else {
+        let path = NamedTempFile::with_suffix(".png")?;
+        let pathbuf: PathBuf = path.path().into();
+        mem::forget(path);
+        pathbuf
+    };
+
+    match flags.image.as_ref() {
+        "binary_cpu" => image_binary::main_cpu(dimensions, output.as_ref()),
+        "binary_gpu" => image_binary::main_gpu(&device, &queue, dimensions, output.as_ref()),
+        _ => Err(anyhow::anyhow!("Unrecognized image type"))
+    }?;
+
+    let mut command = cfg_select! {
+        unix => Command::new("xdg-open").arg(temp.path()).spawn(),
+        windows => Command::new("PowerShell").arg("-Command").arg(format!("start {}", output.display())).spawn(),
+        _ => Err(anyhow::anyhow!("No available program to open images"))?
+    }?;
+    command.wait()?;
+
+    Ok(())
+}