Adding upstream version 115.7.0esr.upstream/115.7.0esr

Signed-off-by: Daniel Baumann <daniel.baumann@progress-linux.org>

4 files changed, 2586 insertions, 0 deletions

diff --git a/gfx/wgpu_bindings/src/client.rs b/gfx/wgpu_bindings/src/client.rs
new file mode 100644
index 0000000000..78f91ec365
--- /dev/null
+++ b/gfx/wgpu_bindings/src/client.rs
@@ -0,0 +1,1150 @@
+/* This Source Code Form is subject to the terms of the Mozilla Public
+ * License, v. 2.0. If a copy of the MPL was not distributed with this
+ * file, You can obtain one at http://mozilla.org/MPL/2.0/. */
+
+use crate::{
+    cow_label, wgpu_string, AdapterInformation, ByteBuf, CommandEncoderAction, DeviceAction,
+    DropAction, ImageDataLayout, ImplicitLayout, QueueWriteAction, RawString, TextureAction,
+};
+
+use wgc::{hub::IdentityManager, id};
+use wgt::{Backend, TextureFormat};
+
+pub use wgc::command::{compute_ffi::*, render_ffi::*};
+
+use parking_lot::Mutex;
+
+use nsstring::{nsACString, nsString};
+
+use std::{borrow::Cow, ptr};
+
+// we can't call `from_raw_parts` unconditionally because the caller
+// may not even have a valid pointer (e.g. NULL) if the `length` is zero.
+fn make_slice<'a, T>(pointer: *const T, length: usize) -> &'a [T] {
+    if length == 0 {
+        &[]
+    } else {
+        unsafe { std::slice::from_raw_parts(pointer, length) }
+    }
+}
+
+fn make_byte_buf<T: serde::Serialize>(data: &T) -> ByteBuf {
+    let vec = bincode::serialize(data).unwrap();
+    ByteBuf::from_vec(vec)
+}
+
+#[repr(C)]
+pub struct ProgrammableStageDescriptor {
+    module: id::ShaderModuleId,
+    entry_point: RawString,
+}
+
+impl ProgrammableStageDescriptor {
+    fn to_wgpu(&self) -> wgc::pipeline::ProgrammableStageDescriptor {
+        wgc::pipeline::ProgrammableStageDescriptor {
+            module: self.module,
+            entry_point: cow_label(&self.entry_point).unwrap(),
+        }
+    }
+}
+
+#[repr(C)]
+pub struct ComputePipelineDescriptor {
+    label: RawString,
+    layout: Option<id::PipelineLayoutId>,
+    stage: ProgrammableStageDescriptor,
+}
+
+#[repr(C)]
+pub struct VertexBufferLayout {
+    array_stride: wgt::BufferAddress,
+    step_mode: wgt::VertexStepMode,
+    attributes: *const wgt::VertexAttribute,
+    attributes_length: usize,
+}
+
+#[repr(C)]
+pub struct VertexState {
+    stage: ProgrammableStageDescriptor,
+    buffers: *const VertexBufferLayout,
+    buffers_length: usize,
+}
+
+impl VertexState {
+    fn to_wgpu(&self) -> wgc::pipeline::VertexState {
+        let buffer_layouts = make_slice(self.buffers, self.buffers_length)
+            .iter()
+            .map(|vb| wgc::pipeline::VertexBufferLayout {
+                array_stride: vb.array_stride,
+                step_mode: vb.step_mode,
+                attributes: Cow::Borrowed(make_slice(vb.attributes, vb.attributes_length)),
+            })
+            .collect();
+        wgc::pipeline::VertexState {
+            stage: self.stage.to_wgpu(),
+            buffers: Cow::Owned(buffer_layouts),
+        }
+    }
+}
+
+#[repr(C)]
+pub struct ColorTargetState<'a> {
+    format: wgt::TextureFormat,
+    blend: Option<&'a wgt::BlendState>,
+    write_mask: wgt::ColorWrites,
+}
+
+#[repr(C)]
+pub struct FragmentState<'a> {
+    stage: ProgrammableStageDescriptor,
+    targets: *const ColorTargetState<'a>,
+    targets_length: usize,
+}
+
+impl FragmentState<'_> {
+    fn to_wgpu(&self) -> wgc::pipeline::FragmentState {
+        let color_targets = make_slice(self.targets, self.targets_length)
+            .iter()
+            .map(|ct| {
+                Some(wgt::ColorTargetState {
+                    format: ct.format,
+                    blend: ct.blend.cloned(),
+                    write_mask: ct.write_mask,
+                })
+            })
+            .collect();
+        wgc::pipeline::FragmentState {
+            stage: self.stage.to_wgpu(),
+            targets: Cow::Owned(color_targets),
+        }
+    }
+}
+
+#[repr(C)]
+pub struct PrimitiveState<'a> {
+    topology: wgt::PrimitiveTopology,
+    strip_index_format: Option<&'a wgt::IndexFormat>,
+    front_face: wgt::FrontFace,
+    cull_mode: Option<&'a wgt::Face>,
+    polygon_mode: wgt::PolygonMode,
+    unclipped_depth: bool,
+}
+
+impl PrimitiveState<'_> {
+    fn to_wgpu(&self) -> wgt::PrimitiveState {
+        wgt::PrimitiveState {
+            topology: self.topology,
+            strip_index_format: self.strip_index_format.cloned(),
+            front_face: self.front_face.clone(),
+            cull_mode: self.cull_mode.cloned(),
+            polygon_mode: self.polygon_mode,
+            unclipped_depth: self.unclipped_depth,
+            conservative: false,
+        }
+    }
+}
+
+#[repr(C)]
+pub struct RenderPipelineDescriptor<'a> {
+    label: Option<&'a nsACString>,
+    layout: Option<id::PipelineLayoutId>,
+    vertex: &'a VertexState,
+    primitive: PrimitiveState<'a>,
+    fragment: Option<&'a FragmentState<'a>>,
+    depth_stencil: Option<&'a wgt::DepthStencilState>,
+    multisample: wgt::MultisampleState,
+}
+
+#[repr(C)]
+pub enum RawTextureSampleType {
+    Float,
+    UnfilterableFloat,
+    Uint,
+    Sint,
+    Depth,
+}
+
+#[repr(C)]
+pub enum RawBindingType {
+    UniformBuffer,
+    StorageBuffer,
+    ReadonlyStorageBuffer,
+    Sampler,
+    SampledTexture,
+    ReadonlyStorageTexture,
+    WriteonlyStorageTexture,
+}
+
+#[repr(C)]
+pub struct BindGroupLayoutEntry<'a> {
+    binding: u32,
+    visibility: wgt::ShaderStages,
+    ty: RawBindingType,
+    has_dynamic_offset: bool,
+    min_binding_size: Option<wgt::BufferSize>,
+    view_dimension: Option<&'a wgt::TextureViewDimension>,
+    texture_sample_type: Option<&'a RawTextureSampleType>,
+    multisampled: bool,
+    storage_texture_format: Option<&'a wgt::TextureFormat>,
+    sampler_filter: bool,
+    sampler_compare: bool,
+}
+
+#[repr(C)]
+pub struct BindGroupLayoutDescriptor<'a> {
+    label: Option<&'a nsACString>,
+    entries: *const BindGroupLayoutEntry<'a>,
+    entries_length: usize,
+}
+
+#[repr(C)]
+#[derive(Debug)]
+pub struct BindGroupEntry {
+    binding: u32,
+    buffer: Option<id::BufferId>,
+    offset: wgt::BufferAddress,
+    size: Option<wgt::BufferSize>,
+    sampler: Option<id::SamplerId>,
+    texture_view: Option<id::TextureViewId>,
+}
+
+#[repr(C)]
+pub struct BindGroupDescriptor<'a> {
+    label: Option<&'a nsACString>,
+    layout: id::BindGroupLayoutId,
+    entries: *const BindGroupEntry,
+    entries_length: usize,
+}
+
+#[repr(C)]
+pub struct PipelineLayoutDescriptor<'a> {
+    label: Option<&'a nsACString>,
+    bind_group_layouts: *const id::BindGroupLayoutId,
+    bind_group_layouts_length: usize,
+}
+
+#[repr(C)]
+pub struct SamplerDescriptor<'a> {
+    label: Option<&'a nsACString>,
+    address_modes: [wgt::AddressMode; 3],
+    mag_filter: wgt::FilterMode,
+    min_filter: wgt::FilterMode,
+    mipmap_filter: wgt::FilterMode,
+    lod_min_clamp: f32,
+    lod_max_clamp: f32,
+    compare: Option<&'a wgt::CompareFunction>,
+    anisotropy_clamp: Option<&'a u16>,
+}
+
+#[repr(C)]
+pub struct TextureViewDescriptor<'a> {
+    label: Option<&'a nsACString>,
+    format: Option<&'a wgt::TextureFormat>,
+    dimension: Option<&'a wgt::TextureViewDimension>,
+    aspect: wgt::TextureAspect,
+    base_mip_level: u32,
+    mip_level_count: Option<&'a u32>,
+    base_array_layer: u32,
+    array_layer_count: Option<&'a u32>,
+}
+
+#[repr(C)]
+pub struct RenderBundleEncoderDescriptor<'a> {
+    label: Option<&'a nsACString>,
+    color_formats: *const wgt::TextureFormat,
+    color_formats_length: usize,
+    depth_stencil_format: Option<&'a wgt::TextureFormat>,
+    depth_read_only: bool,
+    stencil_read_only: bool,
+    sample_count: u32,
+}
+
+#[derive(Debug, Default)]
+struct IdentityHub {
+    adapters: IdentityManager,
+    devices: IdentityManager,
+    buffers: IdentityManager,
+    command_buffers: IdentityManager,
+    render_bundles: IdentityManager,
+    bind_group_layouts: IdentityManager,
+    pipeline_layouts: IdentityManager,
+    bind_groups: IdentityManager,
+    shader_modules: IdentityManager,
+    compute_pipelines: IdentityManager,
+    render_pipelines: IdentityManager,
+    textures: IdentityManager,
+    texture_views: IdentityManager,
+    samplers: IdentityManager,
+}
+
+impl ImplicitLayout<'_> {
+    fn new(identities: &mut IdentityHub, backend: Backend) -> Self {
+        ImplicitLayout {
+            pipeline: identities.pipeline_layouts.alloc(backend),
+            bind_groups: Cow::Owned(
+                (0..8) // hal::MAX_BIND_GROUPS
+                    .map(|_| identities.bind_group_layouts.alloc(backend))
+                    .collect(),
+            ),
+        }
+    }
+}
+
+#[derive(Debug, Default)]
+struct Identities {
+    vulkan: IdentityHub,
+    #[cfg(any(target_os = "ios", target_os = "macos"))]
+    metal: IdentityHub,
+    #[cfg(windows)]
+    dx12: IdentityHub,
+}
+
+impl Identities {
+    fn select(&mut self, backend: Backend) -> &mut IdentityHub {
+        match backend {
+            Backend::Vulkan => &mut self.vulkan,
+            #[cfg(any(target_os = "ios", target_os = "macos"))]
+            Backend::Metal => &mut self.metal,
+            #[cfg(windows)]
+            Backend::Dx12 => &mut self.dx12,
+            _ => panic!("Unexpected backend: {:?}", backend),
+        }
+    }
+}
+
+#[derive(Debug)]
+pub struct Client {
+    identities: Mutex<Identities>,
+}
+
+#[no_mangle]
+pub unsafe extern "C" fn wgpu_client_drop_action(client: &mut Client, byte_buf: &ByteBuf) {
+    let mut cursor = std::io::Cursor::new(byte_buf.as_slice());
+    let mut identities = client.identities.lock();
+    while let Ok(action) = bincode::deserialize_from(&mut cursor) {
+        match action {
+            DropAction::Adapter(id) => identities.select(id.backend()).adapters.free(id),
+            DropAction::Device(id) => identities.select(id.backend()).devices.free(id),
+            DropAction::ShaderModule(id) => identities.select(id.backend()).shader_modules.free(id),
+            DropAction::PipelineLayout(id) => {
+                identities.select(id.backend()).pipeline_layouts.free(id)
+            }
+            DropAction::BindGroupLayout(id) => {
+                identities.select(id.backend()).bind_group_layouts.free(id)
+            }
+            DropAction::BindGroup(id) => identities.select(id.backend()).bind_groups.free(id),
+            DropAction::CommandBuffer(id) => {
+                identities.select(id.backend()).command_buffers.free(id)
+            }
+            DropAction::RenderBundle(id) => identities.select(id.backend()).render_bundles.free(id),
+            DropAction::RenderPipeline(id) => {
+                identities.select(id.backend()).render_pipelines.free(id)
+            }
+            DropAction::ComputePipeline(id) => {
+                identities.select(id.backend()).compute_pipelines.free(id)
+            }
+            DropAction::Buffer(id) => identities.select(id.backend()).buffers.free(id),
+            DropAction::Texture(id) => identities.select(id.backend()).textures.free(id),
+            DropAction::TextureView(id) => identities.select(id.backend()).texture_views.free(id),
+            DropAction::Sampler(id) => identities.select(id.backend()).samplers.free(id),
+        }
+    }
+}
+
+#[no_mangle]
+pub extern "C" fn wgpu_client_kill_device_id(client: &Client, id: id::DeviceId) {
+    client
+        .identities
+        .lock()
+        .select(id.backend())
+        .devices
+        .free(id)
+}
+
+#[repr(C)]
+#[derive(Debug)]
+pub struct Infrastructure {
+    pub client: *mut Client,
+    pub error: *const u8,
+}
+
+#[no_mangle]
+pub extern "C" fn wgpu_client_new() -> Infrastructure {
+    log::info!("Initializing WGPU client");
+    let client = Box::new(Client {
+        identities: Mutex::new(Identities::default()),
+    });
+    Infrastructure {
+        client: Box::into_raw(client),
+        error: ptr::null(),
+    }
+}
+
+/// # Safety
+///
+/// This function is unsafe because improper use may lead to memory
+/// problems. For example, a double-free may occur if the function is called
+/// twice on the same raw pointer.
+#[no_mangle]
+pub unsafe extern "C" fn wgpu_client_delete(client: *mut Client) {
+    log::info!("Terminating WGPU client");
+    let _client = Box::from_raw(client);
+}
+
+/// # Safety
+///
+/// This function is unsafe as there is no guarantee that the given pointer is
+/// valid for `id_length` elements.
+#[no_mangle]
+pub unsafe extern "C" fn wgpu_client_make_adapter_ids(
+    client: &Client,
+    ids: *mut id::AdapterId,
+    id_length: usize,
+) -> usize {
+    let mut identities = client.identities.lock();
+    assert_ne!(id_length, 0);
+    let mut ids = std::slice::from_raw_parts_mut(ids, id_length).iter_mut();
+
+    *ids.next().unwrap() = identities.vulkan.adapters.alloc(Backend::Vulkan);
+
+    #[cfg(any(target_os = "ios", target_os = "macos"))]
+    {
+        *ids.next().unwrap() = identities.metal.adapters.alloc(Backend::Metal);
+    }
+    #[cfg(windows)]
+    {
+        *ids.next().unwrap() = identities.dx12.adapters.alloc(Backend::Dx12);
+    }
+
+    id_length - ids.len()
+}
+
+#[no_mangle]
+pub extern "C" fn wgpu_client_fill_default_limits(limits: &mut wgt::Limits) {
+    *limits = wgt::Limits::default();
+}
+
+#[no_mangle]
+pub extern "C" fn wgpu_client_adapter_extract_info(
+    byte_buf: &ByteBuf,
+    info: &mut AdapterInformation<nsString>,
+) {
+    let AdapterInformation {
+        backend,
+        device_type,
+        device,
+        driver_info,
+        driver,
+        features,
+        id,
+        limits,
+        name,
+        vendor,
+    } = bincode::deserialize::<AdapterInformation<String>>(unsafe { byte_buf.as_slice() }).unwrap();
+
+    let nss = |s: &str| {
+        let mut ns_string = nsString::new();
+        ns_string.assign_str(s);
+        ns_string
+    };
+    *info = AdapterInformation {
+        backend,
+        device_type,
+        device,
+        driver_info: nss(&driver_info),
+        driver: nss(&driver),
+        features,
+        id,
+        limits,
+        name: nss(&name),
+        vendor,
+    };
+}
+
+#[no_mangle]
+pub extern "C" fn wgpu_client_serialize_device_descriptor(
+    desc: &wgt::DeviceDescriptor<Option<&nsACString>>,
+    bb: &mut ByteBuf,
+) {
+    let label = wgpu_string(desc.label);
+    *bb = make_byte_buf(&desc.map_label(|_| label));
+}
+
+#[no_mangle]
+pub extern "C" fn wgpu_client_make_device_id(
+    client: &Client,
+    adapter_id: id::AdapterId,
+) -> id::DeviceId {
+    let backend = adapter_id.backend();
+    client
+        .identities
+        .lock()
+        .select(backend)
+        .devices
+        .alloc(backend)
+}
+
+#[no_mangle]
+pub extern "C" fn wgpu_client_make_buffer_id(
+    client: &Client,
+    device_id: id::DeviceId,
+) -> id::BufferId {
+    let backend = device_id.backend();
+    client
+        .identities
+        .lock()
+        .select(backend)
+        .buffers
+        .alloc(backend)
+}
+
+#[no_mangle]
+pub extern "C" fn wgpu_client_create_texture(
+    client: &Client,
+    device_id: id::DeviceId,
+    desc: &wgt::TextureDescriptor<Option<&nsACString>, crate::FfiSlice<TextureFormat>>,
+    bb: &mut ByteBuf,
+) -> id::TextureId {
+    let label = wgpu_string(desc.label);
+
+    let backend = device_id.backend();
+    let id = client
+        .identities
+        .lock()
+        .select(backend)
+        .textures
+        .alloc(backend);
+
+    let view_formats = unsafe { desc.view_formats.as_slice() }.to_vec();
+
+    let action = DeviceAction::CreateTexture(
+        id,
+        desc.map_label_and_view_formats(|_| label, |_| view_formats),
+    );
+    *bb = make_byte_buf(&action);
+
+    id
+}
+
+#[no_mangle]
+pub extern "C" fn wgpu_client_create_texture_view(
+    client: &Client,
+    device_id: id::DeviceId,
+    desc: &TextureViewDescriptor,
+    bb: &mut ByteBuf,
+) -> id::TextureViewId {
+    let label = wgpu_string(desc.label);
+
+    let backend = device_id.backend();
+    let id = client
+        .identities
+        .lock()
+        .select(backend)
+        .texture_views
+        .alloc(backend);
+
+    let wgpu_desc = wgc::resource::TextureViewDescriptor {
+        label: label,
+        format: desc.format.cloned(),
+        dimension: desc.dimension.cloned(),
+        range: wgt::ImageSubresourceRange {
+            aspect: desc.aspect,
+            base_mip_level: desc.base_mip_level,
+            mip_level_count: desc.mip_level_count.map(|ptr| *ptr),
+            base_array_layer: desc.base_array_layer,
+            array_layer_count: desc.array_layer_count.map(|ptr| *ptr),
+        },
+    };
+
+    let action = TextureAction::CreateView(id, wgpu_desc);
+    *bb = make_byte_buf(&action);
+    id
+}
+
+#[no_mangle]
+pub extern "C" fn wgpu_client_create_sampler(
+    client: &Client,
+    device_id: id::DeviceId,
+    desc: &SamplerDescriptor,
+    bb: &mut ByteBuf,
+) -> id::SamplerId {
+    let label = wgpu_string(desc.label);
+
+    let backend = device_id.backend();
+    let id = client
+        .identities
+        .lock()
+        .select(backend)
+        .samplers
+        .alloc(backend);
+
+    let wgpu_desc = wgc::resource::SamplerDescriptor {
+        label: label,
+        address_modes: desc.address_modes,
+        mag_filter: desc.mag_filter,
+        min_filter: desc.min_filter,
+        mipmap_filter: desc.mipmap_filter,
+        lod_min_clamp: desc.lod_min_clamp,
+        lod_max_clamp: desc.lod_max_clamp,
+        compare: desc.compare.cloned(),
+        anisotropy_clamp: *desc.anisotropy_clamp.unwrap_or(&1),
+        border_color: None,
+    };
+    let action = DeviceAction::CreateSampler(id, wgpu_desc);
+    *bb = make_byte_buf(&action);
+    id
+}
+
+#[no_mangle]
+pub extern "C" fn wgpu_client_make_encoder_id(
+    client: &Client,
+    device_id: id::DeviceId,
+) -> id::CommandEncoderId {
+    let backend = device_id.backend();
+    client
+        .identities
+        .lock()
+        .select(backend)
+        .command_buffers
+        .alloc(backend)
+}
+
+#[no_mangle]
+pub extern "C" fn wgpu_client_create_command_encoder(
+    client: &Client,
+    device_id: id::DeviceId,
+    desc: &wgt::CommandEncoderDescriptor<Option<&nsACString>>,
+    bb: &mut ByteBuf,
+) -> id::CommandEncoderId {
+    let label = wgpu_string(desc.label);
+
+    let backend = device_id.backend();
+    let id = client
+        .identities
+        .lock()
+        .select(backend)
+        .command_buffers
+        .alloc(backend);
+
+    let action = DeviceAction::CreateCommandEncoder(id, desc.map_label(|_| label));
+    *bb = make_byte_buf(&action);
+    id
+}
+
+#[no_mangle]
+pub extern "C" fn wgpu_device_create_render_bundle_encoder(
+    device_id: id::DeviceId,
+    desc: &RenderBundleEncoderDescriptor,
+    bb: &mut ByteBuf,
+) -> *mut wgc::command::RenderBundleEncoder {
+    let label = wgpu_string(desc.label);
+
+    let color_formats: Vec<_> = make_slice(desc.color_formats, desc.color_formats_length)
+        .iter()
+        .map(|format| Some(format.clone()))
+        .collect();
+    let descriptor = wgc::command::RenderBundleEncoderDescriptor {
+        label: label,
+        color_formats: Cow::Owned(color_formats),
+        depth_stencil: desc
+            .depth_stencil_format
+            .map(|&format| wgt::RenderBundleDepthStencil {
+                format,
+                depth_read_only: desc.depth_read_only,
+                stencil_read_only: desc.stencil_read_only,
+            }),
+        sample_count: desc.sample_count,
+        multiview: None,
+    };
+    match wgc::command::RenderBundleEncoder::new(&descriptor, device_id, None) {
+        Ok(encoder) => Box::into_raw(Box::new(encoder)),
+        Err(e) => {
+            let message = format!("Error in Device::create_render_bundle_encoder: {}", e);
+            let action = DeviceAction::Error(message);
+            *bb = make_byte_buf(&action);
+            ptr::null_mut()
+        }
+    }
+}
+
+#[no_mangle]
+pub unsafe extern "C" fn wgpu_render_bundle_encoder_destroy(
+    pass: *mut wgc::command::RenderBundleEncoder,
+) {
+    // The RB encoder is just a boxed Rust struct, it doesn't have any API primitives
+    // associated with it right now, but in the future it will.
+    let _ = Box::from_raw(pass);
+}
+
+#[no_mangle]
+pub unsafe extern "C" fn wgpu_client_create_render_bundle(
+    client: &Client,
+    encoder: *mut wgc::command::RenderBundleEncoder,
+    device_id: id::DeviceId,
+    desc: &wgt::RenderBundleDescriptor<Option<&nsACString>>,
+    bb: &mut ByteBuf,
+) -> id::RenderBundleId {
+    let label = wgpu_string(desc.label);
+
+    let backend = device_id.backend();
+    let id = client
+        .identities
+        .lock()
+        .select(backend)
+        .render_bundles
+        .alloc(backend);
+
+    let action =
+        DeviceAction::CreateRenderBundle(id, *Box::from_raw(encoder), desc.map_label(|_| label));
+    *bb = make_byte_buf(&action);
+    id
+}
+
+#[repr(C)]
+pub struct ComputePassDescriptor<'a> {
+    pub label: Option<&'a nsACString>,
+}
+
+#[no_mangle]
+pub unsafe extern "C" fn wgpu_command_encoder_begin_compute_pass(
+    encoder_id: id::CommandEncoderId,
+    desc: &ComputePassDescriptor,
+) -> *mut wgc::command::ComputePass {
+    let label = wgpu_string(desc.label);
+
+    let pass = wgc::command::ComputePass::new(
+        encoder_id,
+        &wgc::command::ComputePassDescriptor { label: label },
+    );
+    Box::into_raw(Box::new(pass))
+}
+
+#[no_mangle]
+pub unsafe extern "C" fn wgpu_compute_pass_finish(
+    pass: *mut wgc::command::ComputePass,
+    output: &mut ByteBuf,
+) {
+    let command = Box::from_raw(pass).into_command();
+    *output = make_byte_buf(&command);
+}
+
+#[no_mangle]
+pub unsafe extern "C" fn wgpu_compute_pass_destroy(pass: *mut wgc::command::ComputePass) {
+    let _ = Box::from_raw(pass);
+}
+
+#[repr(C)]
+pub struct RenderPassDescriptor<'a> {
+    pub label: Option<&'a nsACString>,
+    pub color_attachments: *const wgc::command::RenderPassColorAttachment,
+    pub color_attachments_length: usize,
+    pub depth_stencil_attachment: *const wgc::command::RenderPassDepthStencilAttachment,
+}
+
+#[no_mangle]
+pub unsafe extern "C" fn wgpu_command_encoder_begin_render_pass(
+    encoder_id: id::CommandEncoderId,
+    desc: &RenderPassDescriptor,
+) -> *mut wgc::command::RenderPass {
+    let label = wgpu_string(desc.label);
+
+    let color_attachments: Vec<_> =
+        make_slice(desc.color_attachments, desc.color_attachments_length)
+            .iter()
+            .map(|format| Some(format.clone()))
+            .collect();
+    let pass = wgc::command::RenderPass::new(
+        encoder_id,
+        &wgc::command::RenderPassDescriptor {
+            label: label,
+            color_attachments: Cow::Owned(color_attachments),
+            depth_stencil_attachment: desc.depth_stencil_attachment.as_ref(),
+        },
+    );
+    Box::into_raw(Box::new(pass))
+}
+
+#[no_mangle]
+pub unsafe extern "C" fn wgpu_render_pass_finish(
+    pass: *mut wgc::command::RenderPass,
+    output: &mut ByteBuf,
+) {
+    let command = Box::from_raw(pass).into_command();
+    *output = make_byte_buf(&command);
+}
+
+#[no_mangle]
+pub unsafe extern "C" fn wgpu_render_pass_destroy(pass: *mut wgc::command::RenderPass) {
+    let _ = Box::from_raw(pass);
+}
+
+#[no_mangle]
+pub unsafe extern "C" fn wgpu_client_create_bind_group_layout(
+    client: &Client,
+    device_id: id::DeviceId,
+    desc: &BindGroupLayoutDescriptor,
+    bb: &mut ByteBuf,
+) -> id::BindGroupLayoutId {
+    let label = wgpu_string(desc.label);
+
+    let backend = device_id.backend();
+    let id = client
+        .identities
+        .lock()
+        .select(backend)
+        .bind_group_layouts
+        .alloc(backend);
+
+    let mut entries = Vec::with_capacity(desc.entries_length);
+    for entry in make_slice(desc.entries, desc.entries_length) {
+        entries.push(wgt::BindGroupLayoutEntry {
+            binding: entry.binding,
+            visibility: entry.visibility,
+            count: None,
+            ty: match entry.ty {
+                RawBindingType::UniformBuffer => wgt::BindingType::Buffer {
+                    ty: wgt::BufferBindingType::Uniform,
+                    has_dynamic_offset: entry.has_dynamic_offset,
+                    min_binding_size: entry.min_binding_size,
+                },
+                RawBindingType::StorageBuffer => wgt::BindingType::Buffer {
+                    ty: wgt::BufferBindingType::Storage { read_only: false },
+                    has_dynamic_offset: entry.has_dynamic_offset,
+                    min_binding_size: entry.min_binding_size,
+                },
+                RawBindingType::ReadonlyStorageBuffer => wgt::BindingType::Buffer {
+                    ty: wgt::BufferBindingType::Storage { read_only: true },
+                    has_dynamic_offset: entry.has_dynamic_offset,
+                    min_binding_size: entry.min_binding_size,
+                },
+                RawBindingType::Sampler => wgt::BindingType::Sampler(if entry.sampler_compare {
+                    wgt::SamplerBindingType::Comparison
+                } else if entry.sampler_filter {
+                    wgt::SamplerBindingType::Filtering
+                } else {
+                    wgt::SamplerBindingType::NonFiltering
+                }),
+                RawBindingType::SampledTexture => wgt::BindingType::Texture {
+                    //TODO: the spec has a bug here
+                    view_dimension: *entry
+                        .view_dimension
+                        .unwrap_or(&wgt::TextureViewDimension::D2),
+                    sample_type: match entry.texture_sample_type {
+                        None | Some(RawTextureSampleType::Float) => {
+                            wgt::TextureSampleType::Float { filterable: true }
+                        }
+                        Some(RawTextureSampleType::UnfilterableFloat) => {
+                            wgt::TextureSampleType::Float { filterable: false }
+                        }
+                        Some(RawTextureSampleType::Uint) => wgt::TextureSampleType::Uint,
+                        Some(RawTextureSampleType::Sint) => wgt::TextureSampleType::Sint,
+                        Some(RawTextureSampleType::Depth) => wgt::TextureSampleType::Depth,
+                    },
+                    multisampled: entry.multisampled,
+                },
+                RawBindingType::ReadonlyStorageTexture => wgt::BindingType::StorageTexture {
+                    access: wgt::StorageTextureAccess::ReadOnly,
+                    view_dimension: *entry.view_dimension.unwrap(),
+                    format: *entry.storage_texture_format.unwrap(),
+                },
+                RawBindingType::WriteonlyStorageTexture => wgt::BindingType::StorageTexture {
+                    access: wgt::StorageTextureAccess::WriteOnly,
+                    view_dimension: *entry.view_dimension.unwrap(),
+                    format: *entry.storage_texture_format.unwrap(),
+                },
+            },
+        });
+    }
+    let wgpu_desc = wgc::binding_model::BindGroupLayoutDescriptor {
+        label: label,
+        entries: Cow::Owned(entries),
+    };
+
+    let action = DeviceAction::CreateBindGroupLayout(id, wgpu_desc);
+    *bb = make_byte_buf(&action);
+    id
+}
+
+#[no_mangle]
+pub unsafe extern "C" fn wgpu_client_create_pipeline_layout(
+    client: &Client,
+    device_id: id::DeviceId,
+    desc: &PipelineLayoutDescriptor,
+    bb: &mut ByteBuf,
+) -> id::PipelineLayoutId {
+    let label = wgpu_string(desc.label);
+
+    let backend = device_id.backend();
+    let id = client
+        .identities
+        .lock()
+        .select(backend)
+        .pipeline_layouts
+        .alloc(backend);
+
+    let wgpu_desc = wgc::binding_model::PipelineLayoutDescriptor {
+        label: label,
+        bind_group_layouts: Cow::Borrowed(make_slice(
+            desc.bind_group_layouts,
+            desc.bind_group_layouts_length,
+        )),
+        push_constant_ranges: Cow::Borrowed(&[]),
+    };
+
+    let action = DeviceAction::CreatePipelineLayout(id, wgpu_desc);
+    *bb = make_byte_buf(&action);
+    id
+}
+
+#[no_mangle]
+pub unsafe extern "C" fn wgpu_client_create_bind_group(
+    client: &Client,
+    device_id: id::DeviceId,
+    desc: &BindGroupDescriptor,
+    bb: &mut ByteBuf,
+) -> id::BindGroupId {
+    let label = wgpu_string(desc.label);
+
+    let backend = device_id.backend();
+    let id = client
+        .identities
+        .lock()
+        .select(backend)
+        .bind_groups
+        .alloc(backend);
+
+    let mut entries = Vec::with_capacity(desc.entries_length);
+    for entry in make_slice(desc.entries, desc.entries_length) {
+        entries.push(wgc::binding_model::BindGroupEntry {
+            binding: entry.binding,
+            resource: if let Some(id) = entry.buffer {
+                wgc::binding_model::BindingResource::Buffer(wgc::binding_model::BufferBinding {
+                    buffer_id: id,
+                    offset: entry.offset,
+                    size: entry.size,
+                })
+            } else if let Some(id) = entry.sampler {
+                wgc::binding_model::BindingResource::Sampler(id)
+            } else if let Some(id) = entry.texture_view {
+                wgc::binding_model::BindingResource::TextureView(id)
+            } else {
+                panic!("Unexpected binding entry {:?}", entry);
+            },
+        });
+    }
+    let wgpu_desc = wgc::binding_model::BindGroupDescriptor {
+        label: label,
+        layout: desc.layout,
+        entries: Cow::Owned(entries),
+    };
+
+    let action = DeviceAction::CreateBindGroup(id, wgpu_desc);
+    *bb = make_byte_buf(&action);
+    id
+}
+
+#[no_mangle]
+pub extern "C" fn wgpu_client_make_shader_module_id(
+    client: &Client,
+    device_id: id::DeviceId,
+) -> id::ShaderModuleId {
+    let backend = device_id.backend();
+    client
+        .identities
+        .lock()
+        .select(backend)
+        .shader_modules
+        .alloc(backend)
+}
+
+#[no_mangle]
+pub unsafe extern "C" fn wgpu_client_create_compute_pipeline(
+    client: &Client,
+    device_id: id::DeviceId,
+    desc: &ComputePipelineDescriptor,
+    bb: &mut ByteBuf,
+    implicit_pipeline_layout_id: *mut Option<id::PipelineLayoutId>,
+    implicit_bind_group_layout_ids: *mut Option<id::BindGroupLayoutId>,
+) -> id::ComputePipelineId {
+    let backend = device_id.backend();
+    let mut identities = client.identities.lock();
+    let id = identities.select(backend).compute_pipelines.alloc(backend);
+
+    let wgpu_desc = wgc::pipeline::ComputePipelineDescriptor {
+        label: cow_label(&desc.label),
+        layout: desc.layout,
+        stage: desc.stage.to_wgpu(),
+    };
+
+    let implicit = match desc.layout {
+        Some(_) => None,
+        None => {
+            let implicit = ImplicitLayout::new(identities.select(backend), backend);
+            ptr::write(implicit_pipeline_layout_id, Some(implicit.pipeline));
+            for (i, bgl_id) in implicit.bind_groups.iter().enumerate() {
+                *implicit_bind_group_layout_ids.add(i) = Some(*bgl_id);
+            }
+            Some(implicit)
+        }
+    };
+
+    let action = DeviceAction::CreateComputePipeline(id, wgpu_desc, implicit);
+    *bb = make_byte_buf(&action);
+    id
+}
+
+#[no_mangle]
+pub unsafe extern "C" fn wgpu_client_create_render_pipeline(
+    client: &Client,
+    device_id: id::DeviceId,
+    desc: &RenderPipelineDescriptor,
+    bb: &mut ByteBuf,
+    implicit_pipeline_layout_id: *mut Option<id::PipelineLayoutId>,
+    implicit_bind_group_layout_ids: *mut Option<id::BindGroupLayoutId>,
+) -> id::RenderPipelineId {
+    let label = wgpu_string(desc.label);
+
+    let backend = device_id.backend();
+    let mut identities = client.identities.lock();
+    let id = identities.select(backend).render_pipelines.alloc(backend);
+
+    let wgpu_desc = wgc::pipeline::RenderPipelineDescriptor {
+        label: label,
+        layout: desc.layout,
+        vertex: desc.vertex.to_wgpu(),
+        fragment: desc.fragment.map(FragmentState::to_wgpu),
+        primitive: desc.primitive.to_wgpu(),
+        depth_stencil: desc.depth_stencil.cloned(),
+        multisample: desc.multisample.clone(),
+        multiview: None,
+    };
+
+    let implicit = match desc.layout {
+        Some(_) => None,
+        None => {
+            let implicit = ImplicitLayout::new(identities.select(backend), backend);
+            ptr::write(implicit_pipeline_layout_id, Some(implicit.pipeline));
+            for (i, bgl_id) in implicit.bind_groups.iter().enumerate() {
+                *implicit_bind_group_layout_ids.add(i) = Some(*bgl_id);
+            }
+            Some(implicit)
+        }
+    };
+
+    let action = DeviceAction::CreateRenderPipeline(id, wgpu_desc, implicit);
+    *bb = make_byte_buf(&action);
+    id
+}
+
+#[no_mangle]
+pub unsafe extern "C" fn wgpu_command_encoder_copy_buffer_to_buffer(
+    src: id::BufferId,
+    src_offset: wgt::BufferAddress,
+    dst: id::BufferId,
+    dst_offset: wgt::BufferAddress,
+    size: wgt::BufferAddress,
+    bb: &mut ByteBuf,
+) {
+    let action = CommandEncoderAction::CopyBufferToBuffer {
+        src,
+        src_offset,
+        dst,
+        dst_offset,
+        size,
+    };
+    *bb = make_byte_buf(&action);
+}
+
+#[no_mangle]
+pub unsafe extern "C" fn wgpu_command_encoder_copy_texture_to_buffer(
+    src: wgc::command::ImageCopyTexture,
+    dst_buffer: wgc::id::BufferId,
+    dst_layout: &ImageDataLayout,
+    size: wgt::Extent3d,
+    bb: &mut ByteBuf,
+) {
+    let action = CommandEncoderAction::CopyTextureToBuffer {
+        src,
+        dst: wgc::command::ImageCopyBuffer {
+            buffer: dst_buffer,
+            layout: dst_layout.into_wgt(),
+        },
+        size,
+    };
+    *bb = make_byte_buf(&action);
+}
+
+#[no_mangle]
+pub unsafe extern "C" fn wgpu_command_encoder_copy_buffer_to_texture(
+    src_buffer: wgc::id::BufferId,
+    src_layout: &ImageDataLayout,
+    dst: wgc::command::ImageCopyTexture,
+    size: wgt::Extent3d,
+    bb: &mut ByteBuf,
+) {
+    let action = CommandEncoderAction::CopyBufferToTexture {
+        src: wgc::command::ImageCopyBuffer {
+            buffer: src_buffer,
+            layout: src_layout.into_wgt(),
+        },
+        dst,
+        size,
+    };
+    *bb = make_byte_buf(&action);
+}
+
+#[no_mangle]
+pub unsafe extern "C" fn wgpu_command_encoder_copy_texture_to_texture(
+    src: wgc::command::ImageCopyTexture,
+    dst: wgc::command::ImageCopyTexture,
+    size: wgt::Extent3d,
+    bb: &mut ByteBuf,
+) {
+    let action = CommandEncoderAction::CopyTextureToTexture { src, dst, size };
+    *bb = make_byte_buf(&action);
+}
+
+#[no_mangle]
+pub extern "C" fn wgpu_command_encoder_push_debug_group(marker: &nsACString, bb: &mut ByteBuf) {
+    let string = marker.to_string();
+    let action = CommandEncoderAction::PushDebugGroup(string);
+    *bb = make_byte_buf(&action);
+}
+
+#[no_mangle]
+pub unsafe extern "C" fn wgpu_command_encoder_pop_debug_group(bb: &mut ByteBuf) {
+    let action = CommandEncoderAction::PopDebugGroup;
+    *bb = make_byte_buf(&action);
+}
+
+#[no_mangle]
+pub unsafe extern "C" fn wgpu_command_encoder_insert_debug_marker(
+    marker: &nsACString,
+    bb: &mut ByteBuf,
+) {
+    let string = marker.to_string();
+    let action = CommandEncoderAction::InsertDebugMarker(string);
+    *bb = make_byte_buf(&action);
+}
+
+#[no_mangle]
+pub unsafe extern "C" fn wgpu_queue_write_buffer(
+    dst: id::BufferId,
+    offset: wgt::BufferAddress,
+    bb: &mut ByteBuf,
+) {
+    let action = QueueWriteAction::Buffer { dst, offset };
+    *bb = make_byte_buf(&action);
+}
+
+#[no_mangle]
+pub unsafe extern "C" fn wgpu_queue_write_texture(
+    dst: wgt::ImageCopyTexture<id::TextureId>,
+    layout: ImageDataLayout,
+    size: wgt::Extent3d,
+    bb: &mut ByteBuf,
+) {
+    let layout = layout.into_wgt();
+    let action = QueueWriteAction::Texture { dst, layout, size };
+    *bb = make_byte_buf(&action);
+}
diff --git a/gfx/wgpu_bindings/src/identity.rs b/gfx/wgpu_bindings/src/identity.rs
new file mode 100644
index 0000000000..7d608b275b
--- /dev/null
+++ b/gfx/wgpu_bindings/src/identity.rs
@@ -0,0 +1,225 @@
+/* This Source Code Form is subject to the terms of the Mozilla Public
+ * License, v. 2.0. If a copy of the MPL was not distributed with this
+ * file, You can obtain one at http://mozilla.org/MPL/2.0/. */
+
+use wgc::id;
+
+pub type FactoryParam = *mut std::ffi::c_void;
+
+#[derive(Debug)]
+pub struct IdentityRecycler<I> {
+    fun: extern "C" fn(I, FactoryParam),
+    param: FactoryParam,
+    kind: &'static str,
+}
+
+impl<I: id::TypedId + Clone + std::fmt::Debug> wgc::hub::IdentityHandler<I>
+    for IdentityRecycler<I>
+{
+    type Input = I;
+    fn process(&self, id: I, _backend: wgt::Backend) -> I {
+        log::debug!("process {} {:?}", self.kind, id);
+        //debug_assert_eq!(id.unzip().2, backend);
+        id
+    }
+    fn free(&self, id: I) {
+        log::debug!("free {} {:?}", self.kind, id);
+        (self.fun)(id, self.param);
+    }
+}
+
+//TODO: remove this in favor of `DropAction` that could be sent over IPC.
+#[repr(C)]
+pub struct IdentityRecyclerFactory {
+    param: FactoryParam,
+    free_adapter: extern "C" fn(id::AdapterId, FactoryParam),
+    free_device: extern "C" fn(id::DeviceId, FactoryParam),
+    free_pipeline_layout: extern "C" fn(id::PipelineLayoutId, FactoryParam),
+    free_shader_module: extern "C" fn(id::ShaderModuleId, FactoryParam),
+    free_bind_group_layout: extern "C" fn(id::BindGroupLayoutId, FactoryParam),
+    free_bind_group: extern "C" fn(id::BindGroupId, FactoryParam),
+    free_command_buffer: extern "C" fn(id::CommandBufferId, FactoryParam),
+    free_render_bundle: extern "C" fn(id::RenderBundleId, FactoryParam),
+    free_render_pipeline: extern "C" fn(id::RenderPipelineId, FactoryParam),
+    free_compute_pipeline: extern "C" fn(id::ComputePipelineId, FactoryParam),
+    free_query_set: extern "C" fn(id::QuerySetId, FactoryParam),
+    free_buffer: extern "C" fn(id::BufferId, FactoryParam),
+    free_staging_buffer: extern "C" fn(id::StagingBufferId, FactoryParam),
+    free_texture: extern "C" fn(id::TextureId, FactoryParam),
+    free_texture_view: extern "C" fn(id::TextureViewId, FactoryParam),
+    free_sampler: extern "C" fn(id::SamplerId, FactoryParam),
+    free_surface: extern "C" fn(id::SurfaceId, FactoryParam),
+}
+
+impl wgc::hub::IdentityHandlerFactory<id::AdapterId> for IdentityRecyclerFactory {
+    type Filter = IdentityRecycler<id::AdapterId>;
+    fn spawn(&self) -> Self::Filter {
+        IdentityRecycler {
+            fun: self.free_adapter,
+            param: self.param,
+            kind: "adapter",
+        }
+    }
+}
+impl wgc::hub::IdentityHandlerFactory<id::DeviceId> for IdentityRecyclerFactory {
+    type Filter = IdentityRecycler<id::DeviceId>;
+    fn spawn(&self) -> Self::Filter {
+        IdentityRecycler {
+            fun: self.free_device,
+            param: self.param,
+            kind: "device",
+        }
+    }
+}
+impl wgc::hub::IdentityHandlerFactory<id::PipelineLayoutId> for IdentityRecyclerFactory {
+    type Filter = IdentityRecycler<id::PipelineLayoutId>;
+    fn spawn(&self) -> Self::Filter {
+        IdentityRecycler {
+            fun: self.free_pipeline_layout,
+            param: self.param,
+            kind: "pipeline_layout",
+        }
+    }
+}
+impl wgc::hub::IdentityHandlerFactory<id::ShaderModuleId> for IdentityRecyclerFactory {
+    type Filter = IdentityRecycler<id::ShaderModuleId>;
+    fn spawn(&self) -> Self::Filter {
+        IdentityRecycler {
+            fun: self.free_shader_module,
+            param: self.param,
+            kind: "shader_module",
+        }
+    }
+}
+impl wgc::hub::IdentityHandlerFactory<id::BindGroupLayoutId> for IdentityRecyclerFactory {
+    type Filter = IdentityRecycler<id::BindGroupLayoutId>;
+    fn spawn(&self) -> Self::Filter {
+        IdentityRecycler {
+            fun: self.free_bind_group_layout,
+            param: self.param,
+            kind: "bind_group_layout",
+        }
+    }
+}
+impl wgc::hub::IdentityHandlerFactory<id::BindGroupId> for IdentityRecyclerFactory {
+    type Filter = IdentityRecycler<id::BindGroupId>;
+    fn spawn(&self) -> Self::Filter {
+        IdentityRecycler {
+            fun: self.free_bind_group,
+            param: self.param,
+            kind: "bind_group",
+        }
+    }
+}
+impl wgc::hub::IdentityHandlerFactory<id::CommandBufferId> for IdentityRecyclerFactory {
+    type Filter = IdentityRecycler<id::CommandBufferId>;
+    fn spawn(&self) -> Self::Filter {
+        IdentityRecycler {
+            fun: self.free_command_buffer,
+            param: self.param,
+            kind: "command_buffer",
+        }
+    }
+}
+impl wgc::hub::IdentityHandlerFactory<id::RenderBundleId> for IdentityRecyclerFactory {
+    type Filter = IdentityRecycler<id::RenderBundleId>;
+    fn spawn(&self) -> Self::Filter {
+        IdentityRecycler {
+            fun: self.free_render_bundle,
+            param: self.param,
+            kind: "render_bundle",
+        }
+    }
+}
+impl wgc::hub::IdentityHandlerFactory<id::RenderPipelineId> for IdentityRecyclerFactory {
+    type Filter = IdentityRecycler<id::RenderPipelineId>;
+    fn spawn(&self) -> Self::Filter {
+        IdentityRecycler {
+            fun: self.free_render_pipeline,
+            param: self.param,
+            kind: "render_pipeline",
+        }
+    }
+}
+impl wgc::hub::IdentityHandlerFactory<id::ComputePipelineId> for IdentityRecyclerFactory {
+    type Filter = IdentityRecycler<id::ComputePipelineId>;
+    fn spawn(&self) -> Self::Filter {
+        IdentityRecycler {
+            fun: self.free_compute_pipeline,
+            param: self.param,
+            kind: "compute_pipeline",
+        }
+    }
+}
+impl wgc::hub::IdentityHandlerFactory<id::QuerySetId> for IdentityRecyclerFactory {
+    type Filter = IdentityRecycler<id::QuerySetId>;
+    fn spawn(&self) -> Self::Filter {
+        IdentityRecycler {
+            fun: self.free_query_set,
+            param: self.param,
+            kind: "query_set",
+        }
+    }
+}
+impl wgc::hub::IdentityHandlerFactory<id::BufferId> for IdentityRecyclerFactory {
+    type Filter = IdentityRecycler<id::BufferId>;
+    fn spawn(&self) -> Self::Filter {
+        IdentityRecycler {
+            fun: self.free_buffer,
+            param: self.param,
+            kind: "buffer",
+        }
+    }
+}
+impl wgc::hub::IdentityHandlerFactory<id::StagingBufferId> for IdentityRecyclerFactory {
+    type Filter = IdentityRecycler<id::StagingBufferId>;
+    fn spawn(&self) -> Self::Filter {
+        IdentityRecycler {
+            fun: self.free_staging_buffer,
+            param: self.param,
+            kind: "staging buffer",
+        }
+    }
+}
+impl wgc::hub::IdentityHandlerFactory<id::TextureId> for IdentityRecyclerFactory {
+    type Filter = IdentityRecycler<id::TextureId>;
+    fn spawn(&self) -> Self::Filter {
+        IdentityRecycler {
+            fun: self.free_texture,
+            param: self.param,
+            kind: "texture",
+        }
+    }
+}
+impl wgc::hub::IdentityHandlerFactory<id::TextureViewId> for IdentityRecyclerFactory {
+    type Filter = IdentityRecycler<id::TextureViewId>;
+    fn spawn(&self) -> Self::Filter {
+        IdentityRecycler {
+            fun: self.free_texture_view,
+            param: self.param,
+            kind: "texture_view",
+        }
+    }
+}
+impl wgc::hub::IdentityHandlerFactory<id::SamplerId> for IdentityRecyclerFactory {
+    type Filter = IdentityRecycler<id::SamplerId>;
+    fn spawn(&self) -> Self::Filter {
+        IdentityRecycler {
+            fun: self.free_sampler,
+            param: self.param,
+            kind: "sampler",
+        }
+    }
+}
+impl wgc::hub::IdentityHandlerFactory<id::SurfaceId> for IdentityRecyclerFactory {
+    type Filter = IdentityRecycler<id::SurfaceId>;
+    fn spawn(&self) -> Self::Filter {
+        IdentityRecycler {
+            fun: self.free_surface,
+            param: self.param,
+            kind: "surface",
+        }
+    }
+}
+
+impl wgc::hub::GlobalIdentityHandlerFactory for IdentityRecyclerFactory {}
diff --git a/gfx/wgpu_bindings/src/lib.rs b/gfx/wgpu_bindings/src/lib.rs
new file mode 100644
index 0000000000..de34649c95
--- /dev/null
+++ b/gfx/wgpu_bindings/src/lib.rs
@@ -0,0 +1,218 @@
+/* This Source Code Form is subject to the terms of the Mozilla Public
+ * License, v. 2.0. If a copy of the MPL was not distributed with this
+ * file, You can obtain one at http://mozilla.org/MPL/2.0/. */
+
+use wgc::id;
+
+pub use wgc::command::{compute_ffi::*, render_ffi::*};
+
+pub mod client;
+pub mod identity;
+pub mod server;
+
+pub use wgc::device::trace::Command as CommandEncoderAction;
+
+use std::marker::PhantomData;
+use std::{borrow::Cow, mem, slice};
+
+use nsstring::nsACString;
+
+type RawString = *const std::os::raw::c_char;
+
+//TODO: figure out why 'a and 'b have to be different here
+//TODO: remove this
+fn cow_label<'a, 'b>(raw: &'a RawString) -> Option<Cow<'b, str>> {
+    if raw.is_null() {
+        None
+    } else {
+        let cstr = unsafe { std::ffi::CStr::from_ptr(*raw) };
+        cstr.to_str().ok().map(Cow::Borrowed)
+    }
+}
+
+// Hides the repeated boilerplate of turning a `Option<&nsACString>` into a `Option<Cow<str>`.
+pub fn wgpu_string(gecko_string: Option<&nsACString>) -> Option<Cow<str>> {
+    gecko_string.map(|s| s.to_utf8())
+}
+
+/// An equivalent of `&[T]` for ffi structures and function parameters.
+#[repr(C)]
+pub struct FfiSlice<'a, T> {
+    // `data` may be null.
+    pub data: *const T,
+    pub length: usize,
+    pub _marker: PhantomData<&'a T>,
+}
+
+impl<'a, T> FfiSlice<'a, T> {
+    pub unsafe fn as_slice(&self) -> &'a [T] {
+        if self.data.is_null() {
+            // It is invalid to construct a rust slice with a null pointer.
+            return &[];
+        }
+
+        std::slice::from_raw_parts(self.data, self.length)
+    }
+}
+
+impl<'a, T> Copy for FfiSlice<'a, T> {}
+impl<'a, T> Clone for FfiSlice<'a, T> {
+    fn clone(&self) -> Self {
+        *self
+    }
+}
+
+#[repr(C)]
+pub struct ByteBuf {
+    data: *const u8,
+    len: usize,
+    capacity: usize,
+}
+
+impl ByteBuf {
+    fn from_vec(vec: Vec<u8>) -> Self {
+        if vec.is_empty() {
+            ByteBuf {
+                data: std::ptr::null(),
+                len: 0,
+                capacity: 0,
+            }
+        } else {
+            let bb = ByteBuf {
+                data: vec.as_ptr(),
+                len: vec.len(),
+                capacity: vec.capacity(),
+            };
+            mem::forget(vec);
+            bb
+        }
+    }
+
+    unsafe fn as_slice(&self) -> &[u8] {
+        slice::from_raw_parts(self.data, self.len)
+    }
+}
+
+#[repr(C)]
+#[derive(serde::Serialize, serde::Deserialize)]
+pub struct AdapterInformation<S> {
+    id: id::AdapterId,
+    limits: wgt::Limits,
+    features: wgt::Features,
+    name: S,
+    vendor: u32,
+    device: u32,
+    device_type: wgt::DeviceType,
+    driver: S,
+    driver_info: S,
+    backend: wgt::Backend,
+}
+
+#[derive(serde::Serialize, serde::Deserialize)]
+struct ImplicitLayout<'a> {
+    pipeline: id::PipelineLayoutId,
+    bind_groups: Cow<'a, [id::BindGroupLayoutId]>,
+}
+
+#[derive(serde::Serialize, serde::Deserialize)]
+enum DeviceAction<'a> {
+    CreateTexture(id::TextureId, wgc::resource::TextureDescriptor<'a>),
+    CreateSampler(id::SamplerId, wgc::resource::SamplerDescriptor<'a>),
+    CreateBindGroupLayout(
+        id::BindGroupLayoutId,
+        wgc::binding_model::BindGroupLayoutDescriptor<'a>,
+    ),
+    CreatePipelineLayout(
+        id::PipelineLayoutId,
+        wgc::binding_model::PipelineLayoutDescriptor<'a>,
+    ),
+    CreateBindGroup(id::BindGroupId, wgc::binding_model::BindGroupDescriptor<'a>),
+    CreateShaderModule(
+        id::ShaderModuleId,
+        wgc::pipeline::ShaderModuleDescriptor<'a>,
+        Cow<'a, str>,
+    ),
+    CreateComputePipeline(
+        id::ComputePipelineId,
+        wgc::pipeline::ComputePipelineDescriptor<'a>,
+        Option<ImplicitLayout<'a>>,
+    ),
+    CreateRenderPipeline(
+        id::RenderPipelineId,
+        wgc::pipeline::RenderPipelineDescriptor<'a>,
+        Option<ImplicitLayout<'a>>,
+    ),
+    CreateRenderBundle(
+        id::RenderBundleId,
+        wgc::command::RenderBundleEncoder,
+        wgc::command::RenderBundleDescriptor<'a>,
+    ),
+    CreateCommandEncoder(
+        id::CommandEncoderId,
+        wgt::CommandEncoderDescriptor<wgc::Label<'a>>,
+    ),
+    Error(String),
+}
+
+#[derive(serde::Serialize, serde::Deserialize)]
+enum QueueWriteAction {
+    Buffer {
+        dst: id::BufferId,
+        offset: wgt::BufferAddress,
+    },
+    Texture {
+        dst: wgt::ImageCopyTexture<id::TextureId>,
+        layout: wgt::ImageDataLayout,
+        size: wgt::Extent3d,
+    },
+}
+
+#[derive(serde::Serialize, serde::Deserialize)]
+enum TextureAction<'a> {
+    CreateView(id::TextureViewId, wgc::resource::TextureViewDescriptor<'a>),
+}
+
+#[repr(C)]
+#[derive(serde::Serialize, serde::Deserialize)]
+enum DropAction {
+    Adapter(id::AdapterId),
+    Device(id::DeviceId),
+    ShaderModule(id::ShaderModuleId),
+    PipelineLayout(id::PipelineLayoutId),
+    BindGroupLayout(id::BindGroupLayoutId),
+    BindGroup(id::BindGroupId),
+    CommandBuffer(id::CommandBufferId),
+    RenderBundle(id::RenderBundleId),
+    RenderPipeline(id::RenderPipelineId),
+    ComputePipeline(id::ComputePipelineId),
+    Buffer(id::BufferId),
+    Texture(id::TextureId),
+    TextureView(id::TextureViewId),
+    Sampler(id::SamplerId),
+}
+
+impl DropAction {
+    // helper function to construct byte bufs
+    fn to_byte_buf(&self) -> ByteBuf {
+        let mut data = Vec::new();
+        bincode::serialize_into(&mut data, self).unwrap();
+        ByteBuf::from_vec(data)
+    }
+}
+
+#[repr(C)]
+pub struct ImageDataLayout<'a> {
+    pub offset: wgt::BufferAddress,
+    pub bytes_per_row: Option<&'a u32>,
+    pub rows_per_image: Option<&'a u32>,
+}
+
+impl<'a> ImageDataLayout<'a> {
+    fn into_wgt(&self) -> wgt::ImageDataLayout {
+        wgt::ImageDataLayout {
+            offset: self.offset,
+            bytes_per_row: self.bytes_per_row.map(|bpr| *bpr),
+            rows_per_image: self.rows_per_image.map(|rpi| *rpi),
+        }
+    }
+}
diff --git a/gfx/wgpu_bindings/src/server.rs b/gfx/wgpu_bindings/src/server.rs
new file mode 100644
index 0000000000..6fb170761b
--- /dev/null
+++ b/gfx/wgpu_bindings/src/server.rs
@@ -0,0 +1,993 @@
+/* This Source Code Form is subject to the terms of the Mozilla Public
+ * License, v. 2.0. If a copy of the MPL was not distributed with this
+ * file, You can obtain one at http://mozilla.org/MPL/2.0/. */
+
+use crate::{
+    identity::IdentityRecyclerFactory, wgpu_string, AdapterInformation, ByteBuf,
+    CommandEncoderAction, DeviceAction, DropAction, QueueWriteAction, TextureAction,
+};
+
+use nsstring::{nsACString, nsCString, nsString};
+
+use wgc::pipeline::CreateShaderModuleError;
+use wgc::{gfx_select, id};
+
+use std::borrow::Cow;
+use std::sync::atomic::{AtomicU32, Ordering};
+use std::{error::Error, os::raw::c_char, ptr, slice};
+
+/// We limit the size of buffer allocations for stability reason.
+/// We can reconsider this limit in the future. Note that some drivers (mesa for example),
+/// have issues when the size of a buffer, mapping or copy command does not fit into a
+/// signed 32 bits integer, so beyond a certain size, large allocations will need some form
+/// of driver allow/blocklist.
+const MAX_BUFFER_SIZE: wgt::BufferAddress = 1 << 30;
+// Mesa has issues with height/depth that don't fit in a 16 bits signed integers.
+const MAX_TEXTURE_EXTENT: u32 = std::i16::MAX as u32;
+
+/// A fixed-capacity, null-terminated error buffer owned by C++.
+///
+/// This type points to space owned by a C++ `mozilla::webgpu::ErrorBuffer`
+/// object, owned by our callers in `WebGPUParent.cpp`. If we catch a
+/// `Result::Err` here, we convert the error to a string, copy as much of that
+/// string as fits into this buffer, and null-terminate it. The caller
+/// determines whether a error occurred by simply checking if there's any text
+/// before the first null byte.
+///
+/// C++ callers of Rust functions that expect one of these structs can create a
+/// `mozilla::webgpu::ErrorBuffer` object, and call its `ToFFI` method to
+/// construct a value of this type, available to C++ as
+/// `mozilla::webgpu::ffi::WGPUErrorBuffer`.
+#[repr(C)]
+pub struct ErrorBuffer {
+    string: *mut c_char,
+    capacity: usize,
+}
+
+impl ErrorBuffer {
+    /// Fill this buffer with the textual representation of `error`.
+    ///
+    /// If the error message is too long, truncate it as needed. In either case,
+    /// the error message is always terminated by a zero byte.
+    ///
+    /// Note that there is no explicit indication of the message's length, only
+    /// the terminating zero byte. If the textual form of `error` itself
+    /// includes a zero byte (as Rust strings can), then the C++ code receiving
+    /// this error message has no way to distinguish that from the terminating
+    /// zero byte, and will see the message as shorter than it is.
+    fn init(&mut self, error: impl Error) {
+        use std::fmt::Write;
+
+        let mut string = format!("{}", error);
+        let mut e = error.source();
+        while let Some(source) = e {
+            write!(string, ", caused by: {}", source).unwrap();
+            e = source.source();
+        }
+
+        self.init_str(&string);
+    }
+
+    fn init_str(&mut self, message: &str) {
+        assert_ne!(self.capacity, 0);
+        let length = if message.len() >= self.capacity {
+            log::warn!(
+                "Error length {} reached capacity {}",
+                message.len(),
+                self.capacity
+            );
+            self.capacity - 1
+        } else {
+            message.len()
+        };
+        unsafe {
+            ptr::copy_nonoverlapping(message.as_ptr(), self.string as *mut u8, length);
+            *self.string.add(length) = 0;
+        }
+    }
+}
+
+// hide wgc's global in private
+pub struct Global(wgc::hub::Global<IdentityRecyclerFactory>);
+
+impl std::ops::Deref for Global {
+    type Target = wgc::hub::Global<IdentityRecyclerFactory>;
+    fn deref(&self) -> &Self::Target {
+        &self.0
+    }
+}
+
+#[no_mangle]
+pub extern "C" fn wgpu_server_new(factory: IdentityRecyclerFactory) -> *mut Global {
+    log::info!("Initializing WGPU server");
+    let backends_pref = static_prefs::pref!("dom.webgpu.wgpu-backend").to_string();
+    let backends = if backends_pref.is_empty() {
+        wgt::Backends::PRIMARY
+    } else {
+        log::info!(
+            "Selecting backends based on dom.webgpu.wgpu-backend pref: {:?}",
+            backends_pref
+        );
+        wgc::instance::parse_backends_from_comma_list(&backends_pref)
+    };
+    let global = Global(wgc::hub::Global::new(
+        "wgpu",
+        factory,
+        wgt::InstanceDescriptor {
+            backends,
+            dx12_shader_compiler: wgt::Dx12Compiler::Fxc,
+        },
+    ));
+    Box::into_raw(Box::new(global))
+}
+
+/// # Safety
+///
+/// This function is unsafe because improper use may lead to memory
+/// problems. For example, a double-free may occur if the function is called
+/// twice on the same raw pointer.
+#[no_mangle]
+pub unsafe extern "C" fn wgpu_server_delete(global: *mut Global) {
+    log::info!("Terminating WGPU server");
+    let _ = Box::from_raw(global);
+}
+
+#[no_mangle]
+pub extern "C" fn wgpu_server_poll_all_devices(global: &Global, force_wait: bool) {
+    global.poll_all_devices(force_wait).unwrap();
+}
+
+/// Request an adapter according to the specified options.
+/// Provide the list of IDs to pick from.
+///
+/// Returns the index in this list, or -1 if unable to pick.
+///
+/// # Safety
+///
+/// This function is unsafe as there is no guarantee that the given pointer is
+/// valid for `id_length` elements.
+#[no_mangle]
+pub unsafe extern "C" fn wgpu_server_instance_request_adapter(
+    global: &Global,
+    desc: &wgc::instance::RequestAdapterOptions,
+    ids: *const id::AdapterId,
+    id_length: usize,
+    mut error_buf: ErrorBuffer,
+) -> i8 {
+    let ids = slice::from_raw_parts(ids, id_length);
+    match global.request_adapter(
+        desc,
+        wgc::instance::AdapterInputs::IdSet(ids, |i| i.backend()),
+    ) {
+        Ok(id) => ids.iter().position(|&i| i == id).unwrap() as i8,
+        Err(e) => {
+            error_buf.init(e);
+            -1
+        }
+    }
+}
+
+#[no_mangle]
+pub unsafe extern "C" fn wgpu_server_adapter_pack_info(
+    global: &Global,
+    self_id: Option<id::AdapterId>,
+    byte_buf: &mut ByteBuf,
+) {
+    let mut data = Vec::new();
+    match self_id {
+        Some(id) => {
+            let wgt::AdapterInfo {
+                name,
+                vendor,
+                device,
+                device_type,
+                driver,
+                driver_info,
+                backend,
+            } = gfx_select!(id => global.adapter_get_info(id)).unwrap();
+
+            let info = AdapterInformation {
+                id,
+                limits: gfx_select!(id => global.adapter_limits(id)).unwrap(),
+                features: gfx_select!(id => global.adapter_features(id)).unwrap(),
+                name,
+                vendor,
+                device,
+                device_type,
+                driver,
+                driver_info,
+                backend,
+            };
+            bincode::serialize_into(&mut data, &info).unwrap();
+        }
+        None => {
+            bincode::serialize_into(&mut data, &0u64).unwrap();
+        }
+    }
+    *byte_buf = ByteBuf::from_vec(data);
+}
+
+static TRACE_IDX: AtomicU32 = AtomicU32::new(0);
+
+#[no_mangle]
+pub unsafe extern "C" fn wgpu_server_adapter_request_device(
+    global: &Global,
+    self_id: id::AdapterId,
+    byte_buf: &ByteBuf,
+    new_id: id::DeviceId,
+    mut error_buf: ErrorBuffer,
+) {
+    let desc: wgc::device::DeviceDescriptor = bincode::deserialize(byte_buf.as_slice()).unwrap();
+    let trace_string = std::env::var("WGPU_TRACE").ok().map(|s| {
+        let idx = TRACE_IDX.fetch_add(1, Ordering::Relaxed);
+        let path = format!("{}/{}/", s, idx);
+
+        if std::fs::create_dir_all(&path).is_err() {
+            log::warn!("Failed to create directory {:?} for wgpu recording.", path);
+        }
+
+        path
+    });
+    let trace_path = trace_string
+        .as_ref()
+        .map(|string| std::path::Path::new(string.as_str()));
+    let (_, error) =
+        gfx_select!(self_id => global.adapter_request_device(self_id, &desc, trace_path, new_id));
+    if let Some(err) = error {
+        error_buf.init(err);
+    }
+}
+
+#[no_mangle]
+pub extern "C" fn wgpu_server_adapter_drop(global: &Global, adapter_id: id::AdapterId) {
+    gfx_select!(adapter_id => global.adapter_drop(adapter_id))
+}
+
+#[no_mangle]
+pub extern "C" fn wgpu_server_device_drop(global: &Global, self_id: id::DeviceId) {
+    gfx_select!(self_id => global.device_drop(self_id))
+}
+
+impl ShaderModuleCompilationMessage {
+    fn set_error(&mut self, error: &CreateShaderModuleError, source: &str) {
+        // The WebGPU spec says that if the message doesn't point to a particular position in
+        // the source, the line number, position, offset and lengths should be zero.
+        self.line_number = 0;
+        self.line_pos = 0;
+        self.utf16_offset = 0;
+        self.utf16_length = 0;
+
+        if let Some(location) = error.location(source) {
+            self.line_number = location.line_number as u64;
+            self.line_pos = location.line_position as u64;
+
+            let start = location.offset as usize;
+            let end = start + location.length as usize;
+            self.utf16_offset = source[0..start].chars().map(|c| c.len_utf16() as u64).sum();
+            self.utf16_length = source[start..end]
+                .chars()
+                .map(|c| c.len_utf16() as u64)
+                .sum();
+        }
+
+        let error_string = error.to_string();
+
+        if !error_string.is_empty() {
+            self.message = nsString::from(&error_string[..]);
+        }
+    }
+}
+
+/// A compilation message representation for the ffi boundary.
+/// the message is immediately copied into an equivalent C++
+/// structure that owns its strings.
+#[repr(C)]
+#[derive(Clone)]
+pub struct ShaderModuleCompilationMessage {
+    pub line_number: u64,
+    pub line_pos: u64,
+    pub utf16_offset: u64,
+    pub utf16_length: u64,
+    pub message: nsString,
+}
+
+/// Creates a shader module and returns an object describing the errors if any.
+///
+/// If there was no error, the returned pointer is nil.
+#[no_mangle]
+pub extern "C" fn wgpu_server_device_create_shader_module(
+    global: &Global,
+    self_id: id::DeviceId,
+    module_id: id::ShaderModuleId,
+    label: Option<&nsACString>,
+    code: &nsCString,
+    out_message: &mut ShaderModuleCompilationMessage,
+) -> bool {
+    let utf8_label = label.map(|utf16| utf16.to_string());
+    let label = utf8_label.as_ref().map(|s| Cow::from(&s[..]));
+
+    let source_str = code.to_utf8();
+
+    let source = wgc::pipeline::ShaderModuleSource::Wgsl(Cow::from(&source_str[..]));
+
+    let desc = wgc::pipeline::ShaderModuleDescriptor {
+        label,
+        shader_bound_checks: wgt::ShaderBoundChecks::new(),
+    };
+
+    let (_, error) = gfx_select!(
+        self_id => global.device_create_shader_module(
+            self_id, &desc, source, module_id
+        )
+    );
+
+    if let Some(err) = error {
+        out_message.set_error(&err, &source_str[..]);
+        return false;
+    }
+
+    // Avoid allocating the structure that holds errors in the common case (no errors).
+    return true;
+}
+
+#[no_mangle]
+pub extern "C" fn wgpu_server_device_create_buffer(
+    global: &Global,
+    self_id: id::DeviceId,
+    buffer_id: id::BufferId,
+    label: Option<&nsACString>,
+    size: wgt::BufferAddress,
+    usage: u32,
+    mapped_at_creation: bool,
+    mut error_buf: ErrorBuffer,
+) {
+    let utf8_label = label.map(|utf16| utf16.to_string());
+    let label = utf8_label.as_ref().map(|s| Cow::from(&s[..]));
+    let usage = wgt::BufferUsages::from_bits_retain(usage);
+
+    // Don't trust the graphics driver with buffer sizes larger than our conservative max texture size.
+    if size > MAX_BUFFER_SIZE {
+        error_buf.init_str("Out of memory");
+        gfx_select!(self_id => global.create_buffer_error(buffer_id, label));
+        return;
+    }
+
+    let desc = wgc::resource::BufferDescriptor {
+        label,
+        size,
+        usage,
+        mapped_at_creation,
+    };
+    let (_, error) = gfx_select!(self_id => global.device_create_buffer(self_id, &desc, buffer_id));
+    if let Some(err) = error {
+        error_buf.init(err);
+    }
+}
+
+/// # Safety
+///
+/// Callers are responsible for ensuring `callback` is well-formed.
+#[no_mangle]
+pub unsafe extern "C" fn wgpu_server_buffer_map(
+    global: &Global,
+    buffer_id: id::BufferId,
+    start: wgt::BufferAddress,
+    size: wgt::BufferAddress,
+    map_mode: wgc::device::HostMap,
+    callback: wgc::resource::BufferMapCallbackC,
+) {
+    let callback = wgc::resource::BufferMapCallback::from_c(callback);
+    let operation = wgc::resource::BufferMapOperation {
+        host: map_mode,
+        callback,
+    };
+    // All errors are also exposed to the mapping callback, so we handle them there and ignore
+    // the the returned value of buffer_map_async.
+    let _ = gfx_select!(buffer_id => global.buffer_map_async(
+        buffer_id,
+        start .. start + size,
+        operation
+    ));
+}
+
+#[repr(C)]
+pub struct MappedBufferSlice {
+    pub ptr: *mut u8,
+    pub length: u64,
+}
+
+/// # Safety
+///
+/// This function is unsafe as there is no guarantee that the given pointer is
+/// valid for `size` elements.
+#[no_mangle]
+pub unsafe extern "C" fn wgpu_server_buffer_get_mapped_range(
+    global: &Global,
+    buffer_id: id::BufferId,
+    start: wgt::BufferAddress,
+    size: wgt::BufferAddress,
+) -> MappedBufferSlice {
+    let result = gfx_select!(buffer_id => global.buffer_get_mapped_range(
+        buffer_id,
+        start,
+        Some(size)
+    ));
+
+    // TODO: error reporting.
+
+    result
+        .map(|(ptr, length)| MappedBufferSlice { ptr, length })
+        .unwrap_or(MappedBufferSlice {
+            ptr: std::ptr::null_mut(),
+            length: 0,
+        })
+}
+
+#[no_mangle]
+pub extern "C" fn wgpu_server_buffer_unmap(
+    global: &Global,
+    buffer_id: id::BufferId,
+    mut error_buf: ErrorBuffer,
+) {
+    if let Err(e) = gfx_select!(buffer_id => global.buffer_unmap(buffer_id)) {
+        error_buf.init(e);
+    }
+}
+
+#[no_mangle]
+pub extern "C" fn wgpu_server_buffer_destroy(global: &Global, self_id: id::BufferId) {
+    // Per spec, there is no need for the buffer or even device to be in a valid state,
+    // even calling calling destroy multiple times is fine, so no error to push into
+    // an error scope.
+    let _ = gfx_select!(self_id => global.buffer_destroy(self_id));
+}
+
+#[no_mangle]
+pub extern "C" fn wgpu_server_buffer_drop(global: &Global, self_id: id::BufferId) {
+    gfx_select!(self_id => global.buffer_drop(self_id, false));
+}
+
+impl Global {
+    fn device_action<A: wgc::hub::HalApi>(
+        &self,
+        self_id: id::DeviceId,
+        action: DeviceAction,
+        mut error_buf: ErrorBuffer,
+    ) {
+        match action {
+            DeviceAction::CreateTexture(id, desc) => {
+                let max = MAX_TEXTURE_EXTENT;
+                if desc.size.width > max
+                    || desc.size.height > max
+                    || desc.size.depth_or_array_layers > max
+                {
+                    gfx_select!(self_id => self.create_texture_error(id, desc.label));
+                    error_buf.init_str("Out of memory");
+                    return;
+                }
+                let (_, error) = self.device_create_texture::<A>(self_id, &desc, id);
+                if let Some(err) = error {
+                    error_buf.init(err);
+                }
+            }
+            DeviceAction::CreateSampler(id, desc) => {
+                let (_, error) = self.device_create_sampler::<A>(self_id, &desc, id);
+                if let Some(err) = error {
+                    error_buf.init(err);
+                }
+            }
+            DeviceAction::CreateBindGroupLayout(id, desc) => {
+                let (_, error) = self.device_create_bind_group_layout::<A>(self_id, &desc, id);
+                if let Some(err) = error {
+                    error_buf.init(err);
+                }
+            }
+            DeviceAction::CreatePipelineLayout(id, desc) => {
+                let (_, error) = self.device_create_pipeline_layout::<A>(self_id, &desc, id);
+                if let Some(err) = error {
+                    error_buf.init(err);
+                }
+            }
+            DeviceAction::CreateBindGroup(id, desc) => {
+                let (_, error) = self.device_create_bind_group::<A>(self_id, &desc, id);
+                if let Some(err) = error {
+                    error_buf.init(err);
+                }
+            }
+            DeviceAction::CreateShaderModule(id, desc, code) => {
+                let source = wgc::pipeline::ShaderModuleSource::Wgsl(code);
+                let (_, error) = self.device_create_shader_module::<A>(self_id, &desc, source, id);
+                if let Some(err) = error {
+                    error_buf.init(err);
+                }
+            }
+            DeviceAction::CreateComputePipeline(id, desc, implicit) => {
+                let implicit_ids = implicit
+                    .as_ref()
+                    .map(|imp| wgc::device::ImplicitPipelineIds {
+                        root_id: imp.pipeline,
+                        group_ids: &imp.bind_groups,
+                    });
+                let (_, error) =
+                    self.device_create_compute_pipeline::<A>(self_id, &desc, id, implicit_ids);
+                if let Some(err) = error {
+                    error_buf.init(err);
+                }
+            }
+            DeviceAction::CreateRenderPipeline(id, desc, implicit) => {
+                let implicit_ids = implicit
+                    .as_ref()
+                    .map(|imp| wgc::device::ImplicitPipelineIds {
+                        root_id: imp.pipeline,
+                        group_ids: &imp.bind_groups,
+                    });
+                let (_, error) =
+                    self.device_create_render_pipeline::<A>(self_id, &desc, id, implicit_ids);
+                if let Some(err) = error {
+                    error_buf.init(err);
+                }
+            }
+            DeviceAction::CreateRenderBundle(id, encoder, desc) => {
+                let (_, error) = self.render_bundle_encoder_finish::<A>(encoder, &desc, id);
+                if let Some(err) = error {
+                    error_buf.init(err);
+                }
+            }
+            DeviceAction::CreateCommandEncoder(id, desc) => {
+                let (_, error) = self.device_create_command_encoder::<A>(self_id, &desc, id);
+                if let Some(err) = error {
+                    error_buf.init(err);
+                }
+            }
+            DeviceAction::Error(message) => {
+                error_buf.init_str(&message);
+            }
+        }
+    }
+
+    fn texture_action<A: wgc::hub::HalApi>(
+        &self,
+        self_id: id::TextureId,
+        action: TextureAction,
+        mut error_buf: ErrorBuffer,
+    ) {
+        match action {
+            TextureAction::CreateView(id, desc) => {
+                let (_, error) = self.texture_create_view::<A>(self_id, &desc, id);
+                if let Some(err) = error {
+                    error_buf.init(err);
+                }
+            }
+        }
+    }
+
+    fn command_encoder_action<A: wgc::hub::HalApi>(
+        &self,
+        self_id: id::CommandEncoderId,
+        action: CommandEncoderAction,
+        mut error_buf: ErrorBuffer,
+    ) {
+        match action {
+            CommandEncoderAction::CopyBufferToBuffer {
+                src,
+                src_offset,
+                dst,
+                dst_offset,
+                size,
+            } => {
+                if let Err(err) = self.command_encoder_copy_buffer_to_buffer::<A>(
+                    self_id, src, src_offset, dst, dst_offset, size,
+                ) {
+                    error_buf.init(err);
+                }
+            }
+            CommandEncoderAction::CopyBufferToTexture { src, dst, size } => {
+                if let Err(err) =
+                    self.command_encoder_copy_buffer_to_texture::<A>(self_id, &src, &dst, &size)
+                {
+                    error_buf.init(err);
+                }
+            }
+            CommandEncoderAction::CopyTextureToBuffer { src, dst, size } => {
+                if let Err(err) =
+                    self.command_encoder_copy_texture_to_buffer::<A>(self_id, &src, &dst, &size)
+                {
+                    error_buf.init(err);
+                }
+            }
+            CommandEncoderAction::CopyTextureToTexture { src, dst, size } => {
+                if let Err(err) =
+                    self.command_encoder_copy_texture_to_texture::<A>(self_id, &src, &dst, &size)
+                {
+                    error_buf.init(err);
+                }
+            }
+            CommandEncoderAction::RunComputePass { base } => {
+                if let Err(err) =
+                    self.command_encoder_run_compute_pass_impl::<A>(self_id, base.as_ref())
+                {
+                    error_buf.init(err);
+                }
+            }
+            CommandEncoderAction::WriteTimestamp {
+                query_set_id,
+                query_index,
+            } => {
+                if let Err(err) =
+                    self.command_encoder_write_timestamp::<A>(self_id, query_set_id, query_index)
+                {
+                    error_buf.init(err);
+                }
+            }
+            CommandEncoderAction::ResolveQuerySet {
+                query_set_id,
+                start_query,
+                query_count,
+                destination,
+                destination_offset,
+            } => {
+                if let Err(err) = self.command_encoder_resolve_query_set::<A>(
+                    self_id,
+                    query_set_id,
+                    start_query,
+                    query_count,
+                    destination,
+                    destination_offset,
+                ) {
+                    error_buf.init(err);
+                }
+            }
+            CommandEncoderAction::RunRenderPass {
+                base,
+                target_colors,
+                target_depth_stencil,
+            } => {
+                if let Err(err) = self.command_encoder_run_render_pass_impl::<A>(
+                    self_id,
+                    base.as_ref(),
+                    &target_colors,
+                    target_depth_stencil.as_ref(),
+                ) {
+                    error_buf.init(err);
+                }
+            }
+            CommandEncoderAction::ClearBuffer { dst, offset, size } => {
+                if let Err(err) = self.command_encoder_clear_buffer::<A>(self_id, dst, offset, size)
+                {
+                    error_buf.init(err);
+                }
+            }
+            CommandEncoderAction::ClearTexture {
+                dst,
+                ref subresource_range,
+            } => {
+                if let Err(err) =
+                    self.command_encoder_clear_texture::<A>(self_id, dst, subresource_range)
+                {
+                    error_buf.init(err);
+                }
+            }
+            CommandEncoderAction::PushDebugGroup(marker) => {
+                if let Err(err) = self.command_encoder_push_debug_group::<A>(self_id, &marker) {
+                    error_buf.init(err);
+                }
+            }
+            CommandEncoderAction::PopDebugGroup => {
+                if let Err(err) = self.command_encoder_pop_debug_group::<A>(self_id) {
+                    error_buf.init(err);
+                }
+            }
+            CommandEncoderAction::InsertDebugMarker(marker) => {
+                if let Err(err) = self.command_encoder_insert_debug_marker::<A>(self_id, &marker) {
+                    error_buf.init(err);
+                }
+            }
+        }
+    }
+}
+
+#[no_mangle]
+pub unsafe extern "C" fn wgpu_server_device_action(
+    global: &Global,
+    self_id: id::DeviceId,
+    byte_buf: &ByteBuf,
+    error_buf: ErrorBuffer,
+) {
+    let action = bincode::deserialize(byte_buf.as_slice()).unwrap();
+    gfx_select!(self_id => global.device_action(self_id, action, error_buf));
+}
+
+#[no_mangle]
+pub unsafe extern "C" fn wgpu_server_texture_action(
+    global: &Global,
+    self_id: id::TextureId,
+    byte_buf: &ByteBuf,
+    error_buf: ErrorBuffer,
+) {
+    let action = bincode::deserialize(byte_buf.as_slice()).unwrap();
+    gfx_select!(self_id => global.texture_action(self_id, action, error_buf));
+}
+
+#[no_mangle]
+pub unsafe extern "C" fn wgpu_server_command_encoder_action(
+    global: &Global,
+    self_id: id::CommandEncoderId,
+    byte_buf: &ByteBuf,
+    error_buf: ErrorBuffer,
+) {
+    let action = bincode::deserialize(byte_buf.as_slice()).unwrap();
+    gfx_select!(self_id => global.command_encoder_action(self_id, action, error_buf));
+}
+
+#[no_mangle]
+pub extern "C" fn wgpu_server_device_create_encoder(
+    global: &Global,
+    self_id: id::DeviceId,
+    desc: &wgt::CommandEncoderDescriptor<Option<&nsACString>>,
+    new_id: id::CommandEncoderId,
+    mut error_buf: ErrorBuffer,
+) {
+    let utf8_label = desc.label.map(|utf16| utf16.to_string());
+    let label = utf8_label.as_ref().map(|s| Cow::from(&s[..]));
+
+    let desc = desc.map_label(|_| label);
+    let (_, error) =
+        gfx_select!(self_id => global.device_create_command_encoder(self_id, &desc, new_id));
+    if let Some(err) = error {
+        error_buf.init(err);
+    }
+}
+
+#[no_mangle]
+pub extern "C" fn wgpu_server_encoder_finish(
+    global: &Global,
+    self_id: id::CommandEncoderId,
+    desc: &wgt::CommandBufferDescriptor<Option<&nsACString>>,
+    mut error_buf: ErrorBuffer,
+) {
+    let label = wgpu_string(desc.label);
+    let desc = desc.map_label(|_| label);
+    let (_, error) = gfx_select!(self_id => global.command_encoder_finish(self_id, &desc));
+    if let Some(err) = error {
+        error_buf.init(err);
+    }
+}
+
+#[no_mangle]
+pub extern "C" fn wgpu_server_encoder_drop(global: &Global, self_id: id::CommandEncoderId) {
+    gfx_select!(self_id => global.command_encoder_drop(self_id));
+}
+
+#[no_mangle]
+pub extern "C" fn wgpu_server_command_buffer_drop(global: &Global, self_id: id::CommandBufferId) {
+    gfx_select!(self_id => global.command_buffer_drop(self_id));
+}
+
+#[no_mangle]
+pub extern "C" fn wgpu_server_render_bundle_drop(global: &Global, self_id: id::RenderBundleId) {
+    gfx_select!(self_id => global.render_bundle_drop(self_id));
+}
+
+#[no_mangle]
+pub unsafe extern "C" fn wgpu_server_encoder_copy_texture_to_buffer(
+    global: &Global,
+    self_id: id::CommandEncoderId,
+    source: &wgc::command::ImageCopyTexture,
+    dst_buffer: wgc::id::BufferId,
+    dst_layout: &crate::ImageDataLayout,
+    size: &wgt::Extent3d,
+) {
+    let destination = wgc::command::ImageCopyBuffer {
+        buffer: dst_buffer,
+        layout: dst_layout.into_wgt(),
+    };
+    gfx_select!(self_id => global.command_encoder_copy_texture_to_buffer(self_id, source, &destination, size)).unwrap();
+}
+
+/// # Safety
+///
+/// This function is unsafe as there is no guarantee that the given pointer is
+/// valid for `command_buffer_id_length` elements.
+#[no_mangle]
+pub unsafe extern "C" fn wgpu_server_queue_submit(
+    global: &Global,
+    self_id: id::QueueId,
+    command_buffer_ids: *const id::CommandBufferId,
+    command_buffer_id_length: usize,
+    mut error_buf: ErrorBuffer,
+) {
+    let command_buffers = slice::from_raw_parts(command_buffer_ids, command_buffer_id_length);
+    let result = gfx_select!(self_id => global.queue_submit(self_id, command_buffers));
+    if let Err(err) = result {
+        error_buf.init(err);
+    }
+}
+
+/// # Safety
+///
+/// This function is unsafe as there is no guarantee that the given pointer is
+/// valid for `data_length` elements.
+#[no_mangle]
+pub unsafe extern "C" fn wgpu_server_queue_write_action(
+    global: &Global,
+    self_id: id::QueueId,
+    byte_buf: &ByteBuf,
+    data: *const u8,
+    data_length: usize,
+    mut error_buf: ErrorBuffer,
+) {
+    let action: QueueWriteAction = bincode::deserialize(byte_buf.as_slice()).unwrap();
+    let data = slice::from_raw_parts(data, data_length);
+    let result = match action {
+        QueueWriteAction::Buffer { dst, offset } => {
+            gfx_select!(self_id => global.queue_write_buffer(self_id, dst, offset, data))
+        }
+        QueueWriteAction::Texture { dst, layout, size } => {
+            gfx_select!(self_id => global.queue_write_texture(self_id, &dst, data, &layout, &size))
+        }
+    };
+    if let Err(err) = result {
+        error_buf.init(err);
+    }
+}
+
+#[no_mangle]
+pub extern "C" fn wgpu_server_bind_group_layout_drop(
+    global: &Global,
+    self_id: id::BindGroupLayoutId,
+) {
+    gfx_select!(self_id => global.bind_group_layout_drop(self_id));
+}
+
+#[no_mangle]
+pub extern "C" fn wgpu_server_pipeline_layout_drop(global: &Global, self_id: id::PipelineLayoutId) {
+    gfx_select!(self_id => global.pipeline_layout_drop(self_id));
+}
+
+#[no_mangle]
+pub extern "C" fn wgpu_server_bind_group_drop(global: &Global, self_id: id::BindGroupId) {
+    gfx_select!(self_id => global.bind_group_drop(self_id));
+}
+
+#[no_mangle]
+pub extern "C" fn wgpu_server_shader_module_drop(global: &Global, self_id: id::ShaderModuleId) {
+    gfx_select!(self_id => global.shader_module_drop(self_id));
+}
+
+#[no_mangle]
+pub extern "C" fn wgpu_server_compute_pipeline_drop(
+    global: &Global,
+    self_id: id::ComputePipelineId,
+) {
+    gfx_select!(self_id => global.compute_pipeline_drop(self_id));
+}
+
+#[no_mangle]
+pub extern "C" fn wgpu_server_render_pipeline_drop(global: &Global, self_id: id::RenderPipelineId) {
+    gfx_select!(self_id => global.render_pipeline_drop(self_id));
+}
+
+#[no_mangle]
+pub extern "C" fn wgpu_server_texture_drop(global: &Global, self_id: id::TextureId) {
+    gfx_select!(self_id => global.texture_drop(self_id, false));
+}
+
+#[no_mangle]
+pub extern "C" fn wgpu_server_texture_view_drop(global: &Global, self_id: id::TextureViewId) {
+    gfx_select!(self_id => global.texture_view_drop(self_id, false)).unwrap();
+}
+
+#[no_mangle]
+pub extern "C" fn wgpu_server_sampler_drop(global: &Global, self_id: id::SamplerId) {
+    gfx_select!(self_id => global.sampler_drop(self_id));
+}
+
+#[no_mangle]
+pub extern "C" fn wgpu_server_compute_pipeline_get_bind_group_layout(
+    global: &Global,
+    self_id: id::ComputePipelineId,
+    index: u32,
+    assign_id: id::BindGroupLayoutId,
+    mut error_buf: ErrorBuffer,
+) {
+    let (_, error) = gfx_select!(self_id => global.compute_pipeline_get_bind_group_layout(self_id, index, assign_id));
+    if let Some(err) = error {
+        error_buf.init(err);
+    }
+}
+
+#[no_mangle]
+pub extern "C" fn wgpu_server_render_pipeline_get_bind_group_layout(
+    global: &Global,
+    self_id: id::RenderPipelineId,
+    index: u32,
+    assign_id: id::BindGroupLayoutId,
+    mut error_buf: ErrorBuffer,
+) {
+    let (_, error) = gfx_select!(self_id => global.render_pipeline_get_bind_group_layout(self_id, index, assign_id));
+    if let Some(err) = error {
+        error_buf.init(err);
+    }
+}
+
+/// Encode the freeing of the selected ID into a byte buf.
+#[no_mangle]
+pub extern "C" fn wgpu_server_adapter_free(id: id::AdapterId, drop_byte_buf: &mut ByteBuf) {
+    *drop_byte_buf = DropAction::Adapter(id).to_byte_buf();
+}
+#[no_mangle]
+pub extern "C" fn wgpu_server_device_free(id: id::DeviceId, drop_byte_buf: &mut ByteBuf) {
+    *drop_byte_buf = DropAction::Device(id).to_byte_buf();
+}
+#[no_mangle]
+pub extern "C" fn wgpu_server_shader_module_free(
+    id: id::ShaderModuleId,
+    drop_byte_buf: &mut ByteBuf,
+) {
+    *drop_byte_buf = DropAction::ShaderModule(id).to_byte_buf();
+}
+#[no_mangle]
+pub extern "C" fn wgpu_server_pipeline_layout_free(
+    id: id::PipelineLayoutId,
+    drop_byte_buf: &mut ByteBuf,
+) {
+    *drop_byte_buf = DropAction::PipelineLayout(id).to_byte_buf();
+}
+#[no_mangle]
+pub extern "C" fn wgpu_server_bind_group_layout_free(
+    id: id::BindGroupLayoutId,
+    drop_byte_buf: &mut ByteBuf,
+) {
+    *drop_byte_buf = DropAction::BindGroupLayout(id).to_byte_buf();
+}
+#[no_mangle]
+pub extern "C" fn wgpu_server_bind_group_free(id: id::BindGroupId, drop_byte_buf: &mut ByteBuf) {
+    *drop_byte_buf = DropAction::BindGroup(id).to_byte_buf();
+}
+#[no_mangle]
+pub extern "C" fn wgpu_server_command_buffer_free(
+    id: id::CommandBufferId,
+    drop_byte_buf: &mut ByteBuf,
+) {
+    *drop_byte_buf = DropAction::CommandBuffer(id).to_byte_buf();
+}
+#[no_mangle]
+pub extern "C" fn wgpu_server_render_bundle_free(
+    id: id::RenderBundleId,
+    drop_byte_buf: &mut ByteBuf,
+) {
+    *drop_byte_buf = DropAction::RenderBundle(id).to_byte_buf();
+}
+#[no_mangle]
+pub extern "C" fn wgpu_server_render_pipeline_free(
+    id: id::RenderPipelineId,
+    drop_byte_buf: &mut ByteBuf,
+) {
+    *drop_byte_buf = DropAction::RenderPipeline(id).to_byte_buf();
+}
+#[no_mangle]
+pub extern "C" fn wgpu_server_compute_pipeline_free(
+    id: id::ComputePipelineId,
+    drop_byte_buf: &mut ByteBuf,
+) {
+    *drop_byte_buf = DropAction::ComputePipeline(id).to_byte_buf();
+}
+#[no_mangle]
+pub extern "C" fn wgpu_server_buffer_free(id: id::BufferId, drop_byte_buf: &mut ByteBuf) {
+    *drop_byte_buf = DropAction::Buffer(id).to_byte_buf();
+}
+#[no_mangle]
+pub extern "C" fn wgpu_server_texture_free(id: id::TextureId, drop_byte_buf: &mut ByteBuf) {
+    *drop_byte_buf = DropAction::Texture(id).to_byte_buf();
+}
+#[no_mangle]
+pub extern "C" fn wgpu_server_texture_view_free(
+    id: id::TextureViewId,
+    drop_byte_buf: &mut ByteBuf,
+) {
+    *drop_byte_buf = DropAction::TextureView(id).to_byte_buf();
+}
+#[no_mangle]
+pub extern "C" fn wgpu_server_sampler_free(id: id::SamplerId, drop_byte_buf: &mut ByteBuf) {
+    *drop_byte_buf = DropAction::Sampler(id).to_byte_buf();
+}