From 1fd05608fb6eb330227b598394ad6fbd77acebdf Mon Sep 17 00:00:00 2001
From: Dzmitry Malyshau <kvark@mozilla.com>
Date: Tue, 2 Oct 2018 11:43:26 -0400
Subject: [PATCH 1/4] native: compute pass boilerplate

---
 wgpu-native/src/command/compute.rs | 32 +++++++++++++++++++++++++++++-
 wgpu-native/src/command/mod.rs     | 13 ++++++++++--
 wgpu-native/src/command/render.rs  | 19 ++++++++----------
 wgpu-native/src/lib.rs             |  1 +
 wgpu-native/src/registry.rs        |  4 +++-
 5 files changed, 54 insertions(+), 15 deletions(-)

diff --git a/wgpu-native/src/command/compute.rs b/wgpu-native/src/command/compute.rs
index fdf5473d5..245921fce 100644
--- a/wgpu-native/src/command/compute.rs
+++ b/wgpu-native/src/command/compute.rs
@@ -1,7 +1,37 @@
 use hal;
 
-//use {CommandBuffer, CommandBufferId, ComputePassId};
+use registry::{self, Items, Registry};
+use {
+    Stored,
+    CommandBufferId, ComputePassId
+};
+
 
 pub struct ComputePass<B: hal::Backend> {
     raw: B::CommandBuffer,
+    cmb_id: Stored<CommandBufferId>,
+}
+
+impl<B: hal::Backend> ComputePass<B> {
+    pub fn new(raw: B::CommandBuffer, cmb_id: CommandBufferId) -> Self {
+        ComputePass {
+            raw,
+            cmb_id: Stored(cmb_id),
+        }
+    }
+}
+
+#[no_mangle]
+pub extern "C" fn wgpu_compute_pass_end_pass(
+    pass_id: ComputePassId,
+) -> CommandBufferId {
+    let pass = registry::COMPUTE_PASS_REGISTRY
+        .lock()
+        .take(pass_id);
+
+    registry::COMMAND_BUFFER_REGISTRY
+        .lock()
+        .get_mut(pass.cmb_id.0)
+        .raw = Some(pass.raw);
+    pass.cmb_id.0
 }
diff --git a/wgpu-native/src/command/mod.rs b/wgpu-native/src/command/mod.rs
index c90f9c34f..7b6eacb56 100644
--- a/wgpu-native/src/command/mod.rs
+++ b/wgpu-native/src/command/mod.rs
@@ -112,6 +112,15 @@ pub extern "C" fn wgpu_command_buffer_begin_render_pass(
 }
 
 #[no_mangle]
-pub extern "C" fn wgpu_command_buffer_begin_compute_pass() -> ComputePassId {
-    unimplemented!()
+pub extern "C" fn wgpu_command_buffer_begin_compute_pass(
+    command_buffer_id: CommandBufferId,
+) -> ComputePassId {
+    let mut cmb_guard = registry::COMMAND_BUFFER_REGISTRY.lock();
+    let mut cmb = cmb_guard.get_mut(command_buffer_id);
+
+    let raw = cmb.raw.take().unwrap();
+
+    registry::COMPUTE_PASS_REGISTRY
+        .lock()
+        .register(ComputePass::new(raw, command_buffer_id))
 }
diff --git a/wgpu-native/src/command/render.rs b/wgpu-native/src/command/render.rs
index 94df08394..b807d4214 100644
--- a/wgpu-native/src/command/render.rs
+++ b/wgpu-native/src/command/render.rs
@@ -7,15 +7,12 @@ use {
     CommandBufferId, RenderPassId,
 };
 
+
 pub struct RenderPass<B: hal::Backend> {
     raw: B::CommandBuffer,
     cmb_id: Stored<CommandBufferId>,
 }
 
-// This is needed for `cmb_id` - would be great to remove.
-#[cfg(not(feature = "remote"))]
-unsafe impl<B: hal::Backend> Sync for RenderPass<B> {}
-
 impl<B: hal::Backend> RenderPass<B> {
     pub fn new(raw: B::CommandBuffer, cmb_id: CommandBufferId) -> Self {
         RenderPass {
@@ -27,16 +24,16 @@ impl<B: hal::Backend> RenderPass<B> {
 
 #[no_mangle]
 pub extern "C" fn wgpu_render_pass_end_pass(
-    render_pass_id: RenderPassId,
+    pass_id: RenderPassId,
 ) -> CommandBufferId {
-    let mut rp = registry::RENDER_PASS_REGISTRY
+    let mut pass = registry::RENDER_PASS_REGISTRY
         .lock()
-        .take(render_pass_id);
-    rp.raw.end_render_pass();
+        .take(pass_id);
+    pass.raw.end_render_pass();
 
     registry::COMMAND_BUFFER_REGISTRY
         .lock()
-        .get_mut(rp.cmb_id.0)
-        .raw = Some(rp.raw);
-    rp.cmb_id.0
+        .get_mut(pass.cmb_id.0)
+        .raw = Some(pass.raw);
+    pass.cmb_id.0
 }
diff --git a/wgpu-native/src/lib.rs b/wgpu-native/src/lib.rs
index 69e695f82..30a4b905b 100644
--- a/wgpu-native/src/lib.rs
+++ b/wgpu-native/src/lib.rs
@@ -126,3 +126,4 @@ type CommandBufferHandle = CommandBuffer<B>;
 pub type RenderPassId = Id;
 type RenderPassHandle = RenderPass<B>;
 pub type ComputePassId = Id;
+type ComputePassHandle = ComputePass<B>;
diff --git a/wgpu-native/src/registry.rs b/wgpu-native/src/registry.rs
index c55ca6ecc..89c6030b8 100644
--- a/wgpu-native/src/registry.rs
+++ b/wgpu-native/src/registry.rs
@@ -12,7 +12,7 @@ use std::sync::Arc;
 use {
     AdapterHandle, AttachmentStateHandle, BindGroupLayoutHandle, BlendStateHandle,
     CommandBufferHandle, DepthStencilStateHandle, DeviceHandle, InstanceHandle,
-    RenderPassHandle,
+    RenderPassHandle, ComputePassHandle,
     PipelineLayoutHandle, RenderPipelineHandle, ShaderModuleHandle,
 };
 
@@ -178,4 +178,6 @@ lazy_static! {
         ConcreteRegistry::new();
     pub(crate) static ref RENDER_PASS_REGISTRY: ConcreteRegistry<RenderPassHandle> =
         ConcreteRegistry::new();
+    pub(crate) static ref COMPUTE_PASS_REGISTRY: ConcreteRegistry<ComputePassHandle> =
+        ConcreteRegistry::new();
 }

From d713f3e3804ecc5811de407665408e538e4cb7fd Mon Sep 17 00:00:00 2001
From: Dzmitry Malyshau <kvark@mozilla.com>
Date: Tue, 2 Oct 2018 12:18:54 -0400
Subject: [PATCH 2/4] Consolidate all the registries together

---
 wgpu-native/src/command/compute.rs |   6 +-
 wgpu-native/src/command/mod.rs     |  18 ++---
 wgpu-native/src/command/render.rs  |   6 +-
 wgpu-native/src/device.rs          | 112 +++++++++++++++++------------
 wgpu-native/src/instance.rs        |  12 ++--
 wgpu-native/src/registry.rs        |  80 ++++++++++-----------
 6 files changed, 125 insertions(+), 109 deletions(-)

diff --git a/wgpu-native/src/command/compute.rs b/wgpu-native/src/command/compute.rs
index 245921fce..2ab7d2926 100644
--- a/wgpu-native/src/command/compute.rs
+++ b/wgpu-native/src/command/compute.rs
@@ -1,6 +1,6 @@
 use hal;
 
-use registry::{self, Items, Registry};
+use registry::{HUB, Items, Registry};
 use {
     Stored,
     CommandBufferId, ComputePassId
@@ -25,11 +25,11 @@ impl<B: hal::Backend> ComputePass<B> {
 pub extern "C" fn wgpu_compute_pass_end_pass(
     pass_id: ComputePassId,
 ) -> CommandBufferId {
-    let pass = registry::COMPUTE_PASS_REGISTRY
+    let pass = HUB.compute_passes
         .lock()
         .take(pass_id);
 
-    registry::COMMAND_BUFFER_REGISTRY
+    HUB.command_buffers
         .lock()
         .get_mut(pass.cmb_id.0)
         .raw = Some(pass.raw);
diff --git a/wgpu-native/src/command/mod.rs b/wgpu-native/src/command/mod.rs
index 7b6eacb56..25c44a745 100644
--- a/wgpu-native/src/command/mod.rs
+++ b/wgpu-native/src/command/mod.rs
@@ -12,7 +12,7 @@ use {
     Color, Origin3d, Stored,
     BufferId, CommandBufferId, ComputePassId, DeviceId, RenderPassId, TextureId, TextureViewId,
 };
-use registry::{self, Items, Registry};
+use registry::{HUB, Items, Registry};
 
 use std::thread::ThreadId;
 
@@ -86,13 +86,13 @@ pub extern "C" fn wgpu_command_buffer_begin_render_pass(
     command_buffer_id: CommandBufferId,
     _descriptor: RenderPassDescriptor<TextureViewId>,
 ) -> RenderPassId {
-    let mut cmb_guard = registry::COMMAND_BUFFER_REGISTRY.lock();
-    let mut cmb = cmb_guard.get_mut(command_buffer_id);
+    let mut cmb_guard = HUB.command_buffers.lock();
+    let cmb = cmb_guard.get_mut(command_buffer_id);
 
     let raw = cmb.raw.take().unwrap();
 
-    let mut device_guard = registry::DEVICE_REGISTRY.lock();
-    let device = &device_guard.get(cmb.device_id.0).raw;
+    let device_guard = HUB.devices.lock();
+    let _device = &device_guard.get(cmb.device_id.0).raw;
 
     //let render_pass = device.create_render_pass();
     //let framebuffer = device.create_framebuffer();
@@ -106,7 +106,7 @@ pub extern "C" fn wgpu_command_buffer_begin_render_pass(
         hal::SubpassContents::Inline,
     );*/
 
-    registry::RENDER_PASS_REGISTRY
+    HUB.render_passes
         .lock()
         .register(RenderPass::new(raw, command_buffer_id))
 }
@@ -115,12 +115,12 @@ pub extern "C" fn wgpu_command_buffer_begin_render_pass(
 pub extern "C" fn wgpu_command_buffer_begin_compute_pass(
     command_buffer_id: CommandBufferId,
 ) -> ComputePassId {
-    let mut cmb_guard = registry::COMMAND_BUFFER_REGISTRY.lock();
-    let mut cmb = cmb_guard.get_mut(command_buffer_id);
+    let mut cmb_guard = HUB.command_buffers.lock();
+    let cmb = cmb_guard.get_mut(command_buffer_id);
 
     let raw = cmb.raw.take().unwrap();
 
-    registry::COMPUTE_PASS_REGISTRY
+    HUB.compute_passes
         .lock()
         .register(ComputePass::new(raw, command_buffer_id))
 }
diff --git a/wgpu-native/src/command/render.rs b/wgpu-native/src/command/render.rs
index b807d4214..d382bbf8d 100644
--- a/wgpu-native/src/command/render.rs
+++ b/wgpu-native/src/command/render.rs
@@ -1,7 +1,7 @@
 use hal;
 use hal::command::RawCommandBuffer;
 
-use registry::{self, Items, Registry};
+use registry::{HUB, Items, Registry};
 use {
     Stored,
     CommandBufferId, RenderPassId,
@@ -26,12 +26,12 @@ impl<B: hal::Backend> RenderPass<B> {
 pub extern "C" fn wgpu_render_pass_end_pass(
     pass_id: RenderPassId,
 ) -> CommandBufferId {
-    let mut pass = registry::RENDER_PASS_REGISTRY
+    let mut pass = HUB.render_passes
         .lock()
         .take(pass_id);
     pass.raw.end_render_pass();
 
-    registry::COMMAND_BUFFER_REGISTRY
+    HUB.command_buffers
         .lock()
         .get_mut(pass.cmb_id.0)
         .raw = Some(pass.raw);
diff --git a/wgpu-native/src/device.rs b/wgpu-native/src/device.rs
index cc4f230b0..28c4c8694 100644
--- a/wgpu-native/src/device.rs
+++ b/wgpu-native/src/device.rs
@@ -1,15 +1,17 @@
-use hal::command::RawCommandBuffer;
-use hal::queue::RawCommandQueue;
-use hal::{self, Device as _Device};
 use {back, binding_model, command, conv, memory, pipeline};
-
-use registry::{self, Items, Registry};
-use std::{ffi, iter, slice};
+use registry::{HUB, Items, Registry};
 use {
     AttachmentStateId, BindGroupLayoutId, BlendStateId, CommandBufferId, DepthStencilStateId,
     DeviceId, PipelineLayoutId, QueueId, RenderPipelineId, ShaderModuleId,
 };
 
+use hal::command::RawCommandBuffer;
+use hal::queue::RawCommandQueue;
+use hal::{self, Device as _Device};
+
+use std::{ffi, slice};
+
+
 pub struct Device<B: hal::Backend> {
     pub(crate) raw: B::Device,
     queue_group: hal::QueueGroup<B, hal::General>,
@@ -42,7 +44,7 @@ pub extern "C" fn wgpu_device_create_bind_group_layout(
     desc: binding_model::BindGroupLayoutDescriptor,
 ) -> BindGroupLayoutId {
     let bindings = unsafe { slice::from_raw_parts(desc.bindings, desc.bindings_length) };
-    let device_guard = registry::DEVICE_REGISTRY.lock();
+    let device_guard = HUB.devices.lock();
     let device = device_guard.get(device_id);
     let descriptor_set_layout = device.raw.create_descriptor_set_layout(
         bindings.iter().map(|binding| {
@@ -56,7 +58,7 @@ pub extern "C" fn wgpu_device_create_bind_group_layout(
         }),
         &[],
     );
-    registry::BIND_GROUP_LAYOUT_REGISTRY
+    HUB.bind_group_layouts
         .lock()
         .register(binding_model::BindGroupLayout {
             raw: descriptor_set_layout,
@@ -68,17 +70,17 @@ pub extern "C" fn wgpu_device_create_pipeline_layout(
     device_id: DeviceId,
     desc: binding_model::PipelineLayoutDescriptor,
 ) -> PipelineLayoutId {
-    let bind_group_layout_guard = registry::BIND_GROUP_LAYOUT_REGISTRY.lock();
+    let bind_group_layout_guard = HUB.bind_group_layouts.lock();
     let descriptor_set_layouts =
         unsafe { slice::from_raw_parts(desc.bind_group_layouts, desc.bind_group_layouts_length) }
             .iter()
             .map(|id| bind_group_layout_guard.get(id.clone()))
             .collect::<Vec<_>>();
-    let device_guard = registry::DEVICE_REGISTRY.lock();
+    let device_guard = HUB.devices.lock();
     let device = &device_guard.get(device_id).raw;
     let pipeline_layout =
         device.create_pipeline_layout(descriptor_set_layouts.iter().map(|d| &d.raw), &[]); // TODO: push constants
-    registry::PIPELINE_LAYOUT_REGISTRY
+    HUB.pipeline_layouts
         .lock()
         .register(binding_model::PipelineLayout {
             raw: pipeline_layout,
@@ -90,7 +92,7 @@ pub extern "C" fn wgpu_device_create_blend_state(
     _device_id: DeviceId,
     desc: pipeline::BlendStateDescriptor,
 ) -> BlendStateId {
-    registry::BLEND_STATE_REGISTRY
+    HUB.blend_states
         .lock()
         .register(pipeline::BlendState {
             raw: conv::map_blend_state_descriptor(desc),
@@ -102,7 +104,7 @@ pub extern "C" fn wgpu_device_create_depth_stencil_state(
     _device_id: DeviceId,
     desc: pipeline::DepthStencilStateDescriptor,
 ) -> DepthStencilStateId {
-    registry::DEPTH_STENCIL_STATE_REGISTRY
+    HUB.depth_stencil_states
         .lock()
         .register(pipeline::DepthStencilState {
             raw: conv::map_depth_stencil_state(desc),
@@ -114,12 +116,12 @@ pub extern "C" fn wgpu_device_create_shader_module(
     device_id: DeviceId,
     desc: pipeline::ShaderModuleDescriptor,
 ) -> ShaderModuleId {
-    let device_guard = registry::DEVICE_REGISTRY.lock();
+    let device_guard = HUB.devices.lock();
     let device = &device_guard.get(device_id).raw;
     let shader = device
         .create_shader_module(unsafe { slice::from_raw_parts(desc.code.bytes, desc.code.length) })
         .unwrap();
-    registry::SHADER_MODULE_REGISTRY
+    HUB.shader_modules
         .lock()
         .register(ShaderModule { raw: shader })
 }
@@ -129,14 +131,14 @@ pub extern "C" fn wgpu_device_create_command_buffer(
     device_id: DeviceId,
     _desc: command::CommandBufferDescriptor,
 ) -> CommandBufferId {
-    let mut device_guard = registry::DEVICE_REGISTRY.lock();
+    let mut device_guard = HUB.devices.lock();
     let device = device_guard.get_mut(device_id);
     let mut cmd_buf = device.com_allocator.allocate(device_id, &device.raw);
     cmd_buf.raw.as_mut().unwrap().begin(
         hal::command::CommandBufferFlags::ONE_TIME_SUBMIT,
         hal::command::CommandBufferInheritanceInfo::default(),
     );
-    registry::COMMAND_BUFFER_REGISTRY.lock().register(cmd_buf)
+    HUB.command_buffers.lock().register(cmd_buf)
 }
 
 #[no_mangle]
@@ -150,28 +152,48 @@ pub extern "C" fn wgpu_queue_submit(
     command_buffer_ptr: *const CommandBufferId,
     command_buffer_count: usize,
 ) {
-    let mut device_guard = registry::DEVICE_REGISTRY.lock();
+    let mut device_guard = HUB.devices.lock();
     let device = device_guard.get_mut(queue_id);
-    let command_buffer_ids =
-        unsafe { slice::from_raw_parts(command_buffer_ptr, command_buffer_count) };
-    //TODO: submit at once, requires `get_all()`
-    let mut command_buffer_guard = registry::COMMAND_BUFFER_REGISTRY.lock();
+    let mut command_buffer_guard = HUB.command_buffers.lock();
+    let command_buffer_ids = unsafe {
+        slice::from_raw_parts(command_buffer_ptr, command_buffer_count)
+    };
+
+    // finish all the command buffers first
     for &cmb_id in command_buffer_ids {
-        let mut cmd_buf = command_buffer_guard.take(cmb_id);
-        {
-            let mut raw = cmd_buf.raw.as_mut().unwrap();
-            raw.finish();
-            let submission = hal::queue::RawSubmission {
-                cmd_buffers: iter::once(raw),
-                wait_semaphores: &[],
-                signal_semaphores: &[],
-            };
-            unsafe {
-                device.queue_group.queues[0]
-                    .as_raw_mut()
-                    .submit_raw(submission, None);
-            }
+        command_buffer_guard
+            .get_mut(cmb_id)
+            .raw
+            .as_mut()
+            .unwrap()
+            .finish();
+    }
+
+    // now prepare the GPU submission
+    {
+        let submission = hal::queue::RawSubmission {
+            cmd_buffers: command_buffer_ids
+                .iter()
+                .map(|&cmb_id| {
+                    command_buffer_guard
+                        .get(cmb_id)
+                        .raw
+                        .as_ref()
+                        .unwrap()
+                }),
+            wait_semaphores: &[],
+            signal_semaphores: &[],
+        };
+        unsafe {
+            device.queue_group.queues[0]
+                .as_raw_mut()
+                .submit_raw(submission, None);
         }
+    }
+
+    // finally, return the command buffers to the allocator
+    for &cmb_id in command_buffer_ids {
+        let cmd_buf = command_buffer_guard.take(cmb_id);
         device.com_allocator.submit(cmd_buf);
     }
 }
@@ -181,7 +203,7 @@ pub extern "C" fn wgpu_device_create_attachment_state(
     device_id: DeviceId,
     desc: pipeline::AttachmentStateDescriptor,
 ) -> AttachmentStateId {
-    let device_guard = registry::DEVICE_REGISTRY.lock();
+    let device_guard = HUB.devices.lock();
     let device = &device_guard.get(device_id).raw;
 
     let color_formats = unsafe {
@@ -224,7 +246,7 @@ pub extern "C" fn wgpu_device_create_attachment_state(
         depth_stencil_format,
     };
 
-    registry::ATTACHMENT_STATE_REGISTRY
+    HUB.attachment_states
         .lock()
         .register(at_state)
 }
@@ -240,12 +262,12 @@ pub extern "C" fn wgpu_device_create_render_pipeline(
         height: 100,
     };
 
-    let device_guard = registry::DEVICE_REGISTRY.lock();
+    let device_guard = HUB.devices.lock();
     let device = &device_guard.get(device_id).raw;
-    let pipeline_layout_guard = registry::PIPELINE_LAYOUT_REGISTRY.lock();
+    let pipeline_layout_guard = HUB.pipeline_layouts.lock();
     let layout = &pipeline_layout_guard.get(desc.layout).raw;
     let pipeline_stages = unsafe { slice::from_raw_parts(desc.stages, desc.stages_length) };
-    let shader_module_guard = registry::SHADER_MODULE_REGISTRY.lock();
+    let shader_module_guard = HUB.shader_modules.lock();
     let shaders = {
         let mut vertex = None;
         let mut fragment = None;
@@ -303,7 +325,7 @@ pub extern "C" fn wgpu_device_create_render_pipeline(
         primitive_restart: hal::pso::PrimitiveRestart::Disabled, // TODO
     };
 
-    let blend_state_guard = registry::BLEND_STATE_REGISTRY.lock();
+    let blend_state_guard = HUB.blend_states.lock();
     let blend_state = unsafe { slice::from_raw_parts(desc.blend_state, desc.blend_state_length) }
         .iter()
         .map(|id| blend_state_guard.get(id.clone()).raw)
@@ -314,7 +336,7 @@ pub extern "C" fn wgpu_device_create_render_pipeline(
         targets: blend_state,
     };
 
-    let depth_stencil_state_guard = registry::DEPTH_STENCIL_STATE_REGISTRY.lock();
+    let depth_stencil_state_guard = HUB.depth_stencil_states.lock();
     let depth_stencil = depth_stencil_state_guard.get(desc.depth_stencil_state).raw;
 
     // TODO
@@ -341,7 +363,7 @@ pub extern "C" fn wgpu_device_create_render_pipeline(
         depth_bounds: None,
     };
 
-    let attachment_state_guard = registry::ATTACHMENT_STATE_REGISTRY.lock();
+    let attachment_state_guard = HUB.attachment_states.lock();
     let attachment_state = attachment_state_guard.get(desc.attachment_state);
 
     // TODO
@@ -377,7 +399,7 @@ pub extern "C" fn wgpu_device_create_render_pipeline(
         .create_graphics_pipeline(&pipeline_desc, None)
         .unwrap();
 
-    registry::RENDER_PIPELINE_REGISTRY
+    HUB.render_pipelines
         .lock()
         .register(pipeline::RenderPipeline { raw: pipeline })
 }
diff --git a/wgpu-native/src/instance.rs b/wgpu-native/src/instance.rs
index e534e4785..5217bae11 100644
--- a/wgpu-native/src/instance.rs
+++ b/wgpu-native/src/instance.rs
@@ -1,6 +1,6 @@
 use hal::{self, Instance as _Instance, PhysicalDevice as _PhysicalDevice};
 
-use registry::{self, Items, Registry};
+use registry::{HUB, Items, Registry};
 use {AdapterId, Device, DeviceId, InstanceId};
 
 #[repr(C)]
@@ -35,7 +35,7 @@ pub extern "C" fn wgpu_create_instance() -> InstanceId {
     ))]
     {
         let inst = ::back::Instance::create("wgpu", 1);
-        registry::INSTANCE_REGISTRY.lock().register(inst)
+        HUB.instances.lock().register(inst)
     }
     #[cfg(not(any(
         feature = "gfx-backend-vulkan",
@@ -52,7 +52,7 @@ pub extern "C" fn wgpu_instance_get_adapter(
     instance_id: InstanceId,
     desc: AdapterDescriptor,
 ) -> AdapterId {
-    let instance_guard = registry::INSTANCE_REGISTRY.lock();
+    let instance_guard = HUB.instances.lock();
     let instance = instance_guard.get(instance_id);
     let (mut low, mut high, mut other) = (None, None, None);
     for adapter in instance.enumerate_adapters() {
@@ -67,7 +67,7 @@ pub extern "C" fn wgpu_instance_get_adapter(
         PowerPreference::LowPower => low.or(high),
         PowerPreference::HighPerformance | PowerPreference::Default => high.or(low),
     };
-    registry::ADAPTER_REGISTRY
+    HUB.adapters
         .lock()
         .register(some.or(other).unwrap())
 }
@@ -77,11 +77,11 @@ pub extern "C" fn wgpu_adapter_create_device(
     adapter_id: AdapterId,
     _desc: DeviceDescriptor,
 ) -> DeviceId {
-    let mut adapter_guard = registry::ADAPTER_REGISTRY.lock();
+    let mut adapter_guard = HUB.adapters.lock();
     let adapter = adapter_guard.get_mut(adapter_id);
     let (device, queue_group) = adapter.open_with::<_, hal::General>(1, |_qf| true).unwrap();
     let mem_props = adapter.physical_device.memory_properties();
-    registry::DEVICE_REGISTRY
+    HUB.devices
         .lock()
         .register(Device::new(device, queue_group, mem_props))
 }
diff --git a/wgpu-native/src/registry.rs b/wgpu-native/src/registry.rs
index 89c6030b8..b5b450146 100644
--- a/wgpu-native/src/registry.rs
+++ b/wgpu-native/src/registry.rs
@@ -17,9 +17,9 @@ use {
 };
 
 #[cfg(not(feature = "remote"))]
-pub(crate) type Id = *mut c_void;
+pub type Id = *mut c_void;
 #[cfg(feature = "remote")]
-pub(crate) type Id = u32;
+pub type Id = u32;
 
 type Item<'a, T> = &'a T;
 type ItemMut<'a, T> = &'a mut T;
@@ -29,12 +29,11 @@ type ItemsGuard<'a, T> = LocalItems<T>;
 #[cfg(feature = "remote")]
 type ItemsGuard<'a, T> = MutexGuard<'a, RemoteItems<T>>;
 
-pub(crate) trait Registry<T> {
-    fn new() -> Self;
+pub trait Registry<T>: Default {
     fn lock(&self) -> ItemsGuard<T>;
 }
 
-pub(crate) trait Items<T> {
+pub trait Items<T> {
     fn register(&mut self, handle: T) -> Id;
     fn get(&self, id: Id) -> Item<T>;
     fn get_mut(&mut self, id: Id) -> ItemMut<T>;
@@ -42,7 +41,7 @@ pub(crate) trait Items<T> {
 }
 
 #[cfg(not(feature = "remote"))]
-pub(crate) struct LocalItems<T> {
+pub struct LocalItems<T> {
     marker: PhantomData<T>,
 }
 
@@ -66,18 +65,19 @@ impl<T> Items<T> for LocalItems<T> {
 }
 
 #[cfg(not(feature = "remote"))]
-pub(crate) struct LocalRegistry<T> {
+pub struct LocalRegistry<T> {
     marker: PhantomData<T>,
 }
-
 #[cfg(not(feature = "remote"))]
-impl<T> Registry<T> for LocalRegistry<T> {
-    fn new() -> Self {
+impl<T> Default for LocalRegistry<T> {
+    fn default() -> Self {
         LocalRegistry {
             marker: PhantomData,
         }
     }
-
+}
+#[cfg(not(feature = "remote"))]
+impl<T> Registry<T> for LocalRegistry<T> {
     fn lock(&self) -> ItemsGuard<T> {
         LocalItems {
             marker: PhantomData,
@@ -86,7 +86,7 @@ impl<T> Registry<T> for LocalRegistry<T> {
 }
 
 #[cfg(feature = "remote")]
-pub(crate) struct RemoteItems<T> {
+pub struct RemoteItems<T> {
     next_id: Id,
     tracked: FastHashMap<Id, T>,
     free: Vec<Id>,
@@ -132,18 +132,19 @@ impl<T> Items<T> for RemoteItems<T> {
 }
 
 #[cfg(feature = "remote")]
-pub(crate) struct RemoteRegistry<T> {
+pub struct RemoteRegistry<T> {
     items: Arc<Mutex<RemoteItems<T>>>,
 }
-
 #[cfg(feature = "remote")]
-impl<T> Registry<T> for RemoteRegistry<T> {
-    fn new() -> Self {
+impl<T> Default for RemoteRegistry<T> {
+    fn default() -> Self {
         RemoteRegistry {
             items: Arc::new(Mutex::new(RemoteItems::new())),
         }
     }
-
+}
+#[cfg(feature = "remote")]
+impl<T> Registry<T> for RemoteRegistry<T> {
     fn lock(&self) -> ItemsGuard<T> {
         self.items.lock()
     }
@@ -154,30 +155,23 @@ type ConcreteRegistry<T> = LocalRegistry<T>;
 #[cfg(feature = "remote")]
 type ConcreteRegistry<T> = RemoteRegistry<T>;
 
-lazy_static! {
-    pub(crate) static ref ADAPTER_REGISTRY: ConcreteRegistry<AdapterHandle> =
-        ConcreteRegistry::new();
-    pub(crate) static ref ATTACHMENT_STATE_REGISTRY: ConcreteRegistry<AttachmentStateHandle> =
-        ConcreteRegistry::new();
-    pub(crate) static ref BIND_GROUP_LAYOUT_REGISTRY: ConcreteRegistry<BindGroupLayoutHandle> =
-        ConcreteRegistry::new();
-    pub(crate) static ref BLEND_STATE_REGISTRY: ConcreteRegistry<BlendStateHandle> =
-        ConcreteRegistry::new();
-    pub(crate) static ref DEPTH_STENCIL_STATE_REGISTRY: ConcreteRegistry<DepthStencilStateHandle> =
-        ConcreteRegistry::new();
-    pub(crate) static ref DEVICE_REGISTRY: ConcreteRegistry<DeviceHandle> = ConcreteRegistry::new();
-    pub(crate) static ref COMMAND_BUFFER_REGISTRY: ConcreteRegistry<CommandBufferHandle> =
-        ConcreteRegistry::new();
-    pub(crate) static ref INSTANCE_REGISTRY: ConcreteRegistry<InstanceHandle> =
-        ConcreteRegistry::new();
-    pub(crate) static ref PIPELINE_LAYOUT_REGISTRY: ConcreteRegistry<PipelineLayoutHandle> =
-        ConcreteRegistry::new();
-    pub(crate) static ref RENDER_PIPELINE_REGISTRY: ConcreteRegistry<RenderPipelineHandle> =
-        ConcreteRegistry::new();
-    pub(crate) static ref SHADER_MODULE_REGISTRY: ConcreteRegistry<ShaderModuleHandle> =
-        ConcreteRegistry::new();
-    pub(crate) static ref RENDER_PASS_REGISTRY: ConcreteRegistry<RenderPassHandle> =
-        ConcreteRegistry::new();
-    pub(crate) static ref COMPUTE_PASS_REGISTRY: ConcreteRegistry<ComputePassHandle> =
-        ConcreteRegistry::new();
+#[derive(Default)]
+pub struct Hub {
+    pub(crate) instances: ConcreteRegistry<InstanceHandle>,
+    pub(crate) adapters: ConcreteRegistry<AdapterHandle>,
+    pub(crate) devices: ConcreteRegistry<DeviceHandle>,
+    pub(crate) pipeline_layouts: ConcreteRegistry<PipelineLayoutHandle>,
+    pub(crate) bind_group_layouts: ConcreteRegistry<BindGroupLayoutHandle>,
+    pub(crate) attachment_states: ConcreteRegistry<AttachmentStateHandle>,
+    pub(crate) blend_states: ConcreteRegistry<BlendStateHandle>,
+    pub(crate) depth_stencil_states: ConcreteRegistry<DepthStencilStateHandle>,
+    pub(crate) shader_modules: ConcreteRegistry<ShaderModuleHandle>,
+    pub(crate) command_buffers: ConcreteRegistry<CommandBufferHandle>,
+    pub(crate) render_pipelines: ConcreteRegistry<RenderPipelineHandle>,
+    pub(crate) render_passes: ConcreteRegistry<RenderPassHandle>,
+    pub(crate) compute_passes: ConcreteRegistry<ComputePassHandle>,
+}
+
+lazy_static! {
+    pub static ref HUB: Hub = Hub::default();
 }

From 17527544891e5ccdef9f384694cb6095cc2b80a4 Mon Sep 17 00:00:00 2001
From: Dzmitry Malyshau <kvark@mozilla.com>
Date: Tue, 2 Oct 2018 12:31:02 -0400
Subject: [PATCH 3/4] Modularize the registry

---
 wgpu-native/src/registry.rs        | 177 -----------------------------
 wgpu-native/src/registry/local.rs  |  48 ++++++++
 wgpu-native/src/registry/mod.rs    |  52 +++++++++
 wgpu-native/src/registry/remote.rs |  68 +++++++++++
 4 files changed, 168 insertions(+), 177 deletions(-)
 delete mode 100644 wgpu-native/src/registry.rs
 create mode 100644 wgpu-native/src/registry/local.rs
 create mode 100644 wgpu-native/src/registry/mod.rs
 create mode 100644 wgpu-native/src/registry/remote.rs

diff --git a/wgpu-native/src/registry.rs b/wgpu-native/src/registry.rs
deleted file mode 100644
index b5b450146..000000000
--- a/wgpu-native/src/registry.rs
+++ /dev/null
@@ -1,177 +0,0 @@
-#[cfg(feature = "remote")]
-use hal::backend::FastHashMap;
-#[cfg(feature = "remote")]
-use parking_lot::{Mutex, MutexGuard};
-#[cfg(not(feature = "remote"))]
-use std::marker::PhantomData;
-#[cfg(not(feature = "remote"))]
-use std::os::raw::c_void;
-#[cfg(feature = "remote")]
-use std::sync::Arc;
-
-use {
-    AdapterHandle, AttachmentStateHandle, BindGroupLayoutHandle, BlendStateHandle,
-    CommandBufferHandle, DepthStencilStateHandle, DeviceHandle, InstanceHandle,
-    RenderPassHandle, ComputePassHandle,
-    PipelineLayoutHandle, RenderPipelineHandle, ShaderModuleHandle,
-};
-
-#[cfg(not(feature = "remote"))]
-pub type Id = *mut c_void;
-#[cfg(feature = "remote")]
-pub type Id = u32;
-
-type Item<'a, T> = &'a T;
-type ItemMut<'a, T> = &'a mut T;
-
-#[cfg(not(feature = "remote"))]
-type ItemsGuard<'a, T> = LocalItems<T>;
-#[cfg(feature = "remote")]
-type ItemsGuard<'a, T> = MutexGuard<'a, RemoteItems<T>>;
-
-pub trait Registry<T>: Default {
-    fn lock(&self) -> ItemsGuard<T>;
-}
-
-pub trait Items<T> {
-    fn register(&mut self, handle: T) -> Id;
-    fn get(&self, id: Id) -> Item<T>;
-    fn get_mut(&mut self, id: Id) -> ItemMut<T>;
-    fn take(&mut self, id: Id) -> T;
-}
-
-#[cfg(not(feature = "remote"))]
-pub struct LocalItems<T> {
-    marker: PhantomData<T>,
-}
-
-#[cfg(not(feature = "remote"))]
-impl<T> Items<T> for LocalItems<T> {
-    fn register(&mut self, handle: T) -> Id {
-        Box::into_raw(Box::new(handle)) as *mut _ as *mut c_void
-    }
-
-    fn get(&self, id: Id) -> Item<T> {
-        unsafe { (id as *mut T).as_ref() }.unwrap()
-    }
-
-    fn get_mut(&mut self, id: Id) -> ItemMut<T> {
-        unsafe { (id as *mut T).as_mut() }.unwrap()
-    }
-
-    fn take(&mut self, id: Id) -> T {
-        unsafe { *Box::from_raw(id as *mut T) }
-    }
-}
-
-#[cfg(not(feature = "remote"))]
-pub struct LocalRegistry<T> {
-    marker: PhantomData<T>,
-}
-#[cfg(not(feature = "remote"))]
-impl<T> Default for LocalRegistry<T> {
-    fn default() -> Self {
-        LocalRegistry {
-            marker: PhantomData,
-        }
-    }
-}
-#[cfg(not(feature = "remote"))]
-impl<T> Registry<T> for LocalRegistry<T> {
-    fn lock(&self) -> ItemsGuard<T> {
-        LocalItems {
-            marker: PhantomData,
-        }
-    }
-}
-
-#[cfg(feature = "remote")]
-pub struct RemoteItems<T> {
-    next_id: Id,
-    tracked: FastHashMap<Id, T>,
-    free: Vec<Id>,
-}
-
-#[cfg(feature = "remote")]
-impl<T> RemoteItems<T> {
-    fn new() -> Self {
-        RemoteItems {
-            next_id: 0,
-            tracked: FastHashMap::default(),
-            free: Vec::new(),
-        }
-    }
-}
-
-#[cfg(feature = "remote")]
-impl<T> Items<T> for RemoteItems<T> {
-    fn register(&mut self, handle: T) -> Id {
-        let id = match self.free.pop() {
-            Some(id) => id,
-            None => {
-                self.next_id += 1;
-                self.next_id - 1
-            }
-        };
-        self.tracked.insert(id, handle);
-        id
-    }
-
-    fn get(&self, id: Id) -> Item<T> {
-        self.tracked.get(&id).unwrap()
-    }
-
-    fn get_mut(&mut self, id: Id) -> ItemMut<T> {
-        self.tracked.get_mut(&id).unwrap()
-    }
-
-    fn take(&mut self, id: Id) -> T {
-        self.free.push(id);
-        self.tracked.remove(&id).unwrap()
-    }
-}
-
-#[cfg(feature = "remote")]
-pub struct RemoteRegistry<T> {
-    items: Arc<Mutex<RemoteItems<T>>>,
-}
-#[cfg(feature = "remote")]
-impl<T> Default for RemoteRegistry<T> {
-    fn default() -> Self {
-        RemoteRegistry {
-            items: Arc::new(Mutex::new(RemoteItems::new())),
-        }
-    }
-}
-#[cfg(feature = "remote")]
-impl<T> Registry<T> for RemoteRegistry<T> {
-    fn lock(&self) -> ItemsGuard<T> {
-        self.items.lock()
-    }
-}
-
-#[cfg(not(feature = "remote"))]
-type ConcreteRegistry<T> = LocalRegistry<T>;
-#[cfg(feature = "remote")]
-type ConcreteRegistry<T> = RemoteRegistry<T>;
-
-#[derive(Default)]
-pub struct Hub {
-    pub(crate) instances: ConcreteRegistry<InstanceHandle>,
-    pub(crate) adapters: ConcreteRegistry<AdapterHandle>,
-    pub(crate) devices: ConcreteRegistry<DeviceHandle>,
-    pub(crate) pipeline_layouts: ConcreteRegistry<PipelineLayoutHandle>,
-    pub(crate) bind_group_layouts: ConcreteRegistry<BindGroupLayoutHandle>,
-    pub(crate) attachment_states: ConcreteRegistry<AttachmentStateHandle>,
-    pub(crate) blend_states: ConcreteRegistry<BlendStateHandle>,
-    pub(crate) depth_stencil_states: ConcreteRegistry<DepthStencilStateHandle>,
-    pub(crate) shader_modules: ConcreteRegistry<ShaderModuleHandle>,
-    pub(crate) command_buffers: ConcreteRegistry<CommandBufferHandle>,
-    pub(crate) render_pipelines: ConcreteRegistry<RenderPipelineHandle>,
-    pub(crate) render_passes: ConcreteRegistry<RenderPassHandle>,
-    pub(crate) compute_passes: ConcreteRegistry<ComputePassHandle>,
-}
-
-lazy_static! {
-    pub static ref HUB: Hub = Hub::default();
-}
diff --git a/wgpu-native/src/registry/local.rs b/wgpu-native/src/registry/local.rs
new file mode 100644
index 000000000..e4e4bae6d
--- /dev/null
+++ b/wgpu-native/src/registry/local.rs
@@ -0,0 +1,48 @@
+use std::marker::PhantomData;
+use std::os::raw::c_void;
+
+
+pub type Id = *mut c_void;
+pub type ItemsGuard<'a, T> = Items<T>;
+
+pub struct Items<T> {
+    marker: PhantomData<T>,
+}
+
+impl<T> super::Items<T> for Items<T> {
+    fn register(&mut self, handle: T) -> Id {
+        Box::into_raw(Box::new(handle)) as *mut _ as *mut c_void
+    }
+
+    fn get(&self, id: Id) -> super::Item<T> {
+        unsafe { (id as *mut T).as_ref() }.unwrap()
+    }
+
+    fn get_mut(&mut self, id: Id) -> super::ItemMut<T> {
+        unsafe { (id as *mut T).as_mut() }.unwrap()
+    }
+
+    fn take(&mut self, id: Id) -> T {
+        unsafe { *Box::from_raw(id as *mut T) }
+    }
+}
+
+pub struct Registry<T> {
+    marker: PhantomData<T>,
+}
+
+impl<T> Default for Registry<T> {
+    fn default() -> Self {
+        Registry {
+            marker: PhantomData,
+        }
+    }
+}
+
+impl<T> super::Registry<T> for Registry<T> {
+    fn lock(&self) -> ItemsGuard<T> {
+        Items {
+            marker: PhantomData,
+        }
+    }
+}
diff --git a/wgpu-native/src/registry/mod.rs b/wgpu-native/src/registry/mod.rs
new file mode 100644
index 000000000..fecdf2c2b
--- /dev/null
+++ b/wgpu-native/src/registry/mod.rs
@@ -0,0 +1,52 @@
+#[cfg(not(feature = "remote"))]
+mod local;
+#[cfg(feature = "remote")]
+mod remote;
+
+#[cfg(not(feature = "remote"))]
+pub use self::local::{Id, ItemsGuard, Registry as ConcreteRegistry};
+#[cfg(feature = "remote")]
+pub use self::remote::{Id, ItemsGuard, Registry as ConcreteRegistry};
+
+use {
+    AdapterHandle, AttachmentStateHandle, BindGroupLayoutHandle, BlendStateHandle,
+    CommandBufferHandle, DepthStencilStateHandle, DeviceHandle, InstanceHandle,
+    RenderPassHandle, ComputePassHandle,
+    PipelineLayoutHandle, RenderPipelineHandle, ShaderModuleHandle,
+};
+
+
+type Item<'a, T> = &'a T;
+type ItemMut<'a, T> = &'a mut T;
+
+pub trait Registry<T>: Default {
+    fn lock(&self) -> ItemsGuard<T>;
+}
+
+pub trait Items<T> {
+    fn register(&mut self, handle: T) -> Id;
+    fn get(&self, id: Id) -> Item<T>;
+    fn get_mut(&mut self, id: Id) -> ItemMut<T>;
+    fn take(&mut self, id: Id) -> T;
+}
+
+#[derive(Default)]
+pub struct Hub {
+    pub(crate) instances: ConcreteRegistry<InstanceHandle>,
+    pub(crate) adapters: ConcreteRegistry<AdapterHandle>,
+    pub(crate) devices: ConcreteRegistry<DeviceHandle>,
+    pub(crate) pipeline_layouts: ConcreteRegistry<PipelineLayoutHandle>,
+    pub(crate) bind_group_layouts: ConcreteRegistry<BindGroupLayoutHandle>,
+    pub(crate) attachment_states: ConcreteRegistry<AttachmentStateHandle>,
+    pub(crate) blend_states: ConcreteRegistry<BlendStateHandle>,
+    pub(crate) depth_stencil_states: ConcreteRegistry<DepthStencilStateHandle>,
+    pub(crate) shader_modules: ConcreteRegistry<ShaderModuleHandle>,
+    pub(crate) command_buffers: ConcreteRegistry<CommandBufferHandle>,
+    pub(crate) render_pipelines: ConcreteRegistry<RenderPipelineHandle>,
+    pub(crate) render_passes: ConcreteRegistry<RenderPassHandle>,
+    pub(crate) compute_passes: ConcreteRegistry<ComputePassHandle>,
+}
+
+lazy_static! {
+    pub static ref HUB: Hub = Hub::default();
+}
diff --git a/wgpu-native/src/registry/remote.rs b/wgpu-native/src/registry/remote.rs
new file mode 100644
index 000000000..1253ef05e
--- /dev/null
+++ b/wgpu-native/src/registry/remote.rs
@@ -0,0 +1,68 @@
+use hal::backend::FastHashMap;
+use parking_lot::{Mutex, MutexGuard};
+use std::sync::Arc;
+
+
+pub type Id = u32;
+pub type ItemsGuard<'a, T> = MutexGuard<'a, Items<T>>;
+
+pub struct Items<T> {
+    next_id: Id,
+    tracked: FastHashMap<Id, T>,
+    free: Vec<Id>,
+}
+
+impl<T> Items<T> {
+    fn new() -> Self {
+        Items {
+            next_id: 0,
+            tracked: FastHashMap::default(),
+            free: Vec::new(),
+        }
+    }
+}
+
+impl<T> super::Items<T> for Items<T> {
+    fn register(&mut self, handle: T) -> Id {
+        let id = match self.free.pop() {
+            Some(id) => id,
+            None => {
+                self.next_id += 1;
+                self.next_id - 1
+            }
+        };
+        self.tracked.insert(id, handle);
+        id
+    }
+
+    fn get(&self, id: Id) -> super::Item<T> {
+        self.tracked.get(&id).unwrap()
+    }
+
+    fn get_mut(&mut self, id: Id) -> super::ItemMut<T> {
+        self.tracked.get_mut(&id).unwrap()
+    }
+
+    fn take(&mut self, id: Id) -> T {
+        self.free.push(id);
+        self.tracked.remove(&id).unwrap()
+    }
+}
+
+pub struct Registry<T> {
+    items: Arc<Mutex<Items<T>>>,
+}
+
+impl<T> Default for Registry<T> {
+    fn default() -> Self {
+        Registry {
+            items: Arc::new(Mutex::new(Items::new())),
+        }
+    }
+}
+
+impl<T> super::Registry<T> for Registry<T> {
+    fn lock(&self) -> ItemsGuard<T> {
+        self.items.lock()
+    }
+}

From a00c23e350bd7b65380abaab8aa9be6bcd882be8 Mon Sep 17 00:00:00 2001
From: Dzmitry Malyshau <kvark@mozilla.com>
Date: Tue, 2 Oct 2018 12:32:39 -0400
Subject: [PATCH 4/4] Basic compute passes on Rust side

---
 wgpu-rs/src/lib.rs | 19 +++++++++++++++++++
 1 file changed, 19 insertions(+)

diff --git a/wgpu-rs/src/lib.rs b/wgpu-rs/src/lib.rs
index 630074918..3b859990c 100644
--- a/wgpu-rs/src/lib.rs
+++ b/wgpu-rs/src/lib.rs
@@ -68,6 +68,11 @@ pub struct RenderPass<'a> {
     parent: &'a mut CommandBuffer,
 }
 
+pub struct ComputePass<'a> {
+    id: wgn::ComputePassId,
+    parent: &'a mut CommandBuffer,
+}
+
 pub struct Queue {
     id: wgn::QueueId,
 }
@@ -247,6 +252,13 @@ impl CommandBuffer {
             parent: self,
         }
     }
+
+    pub fn begin_compute_pass(&mut self) -> ComputePass {
+        ComputePass {
+            id: wgn::wgpu_command_buffer_begin_compute_pass(self.id),
+            parent: self,
+        }
+    }
 }
 
 impl<'a> RenderPass<'a> {
@@ -256,6 +268,13 @@ impl<'a> RenderPass<'a> {
     }
 }
 
+impl<'a> ComputePass<'a> {
+    pub fn end_pass(self) -> &'a mut CommandBuffer {
+        wgn::wgpu_compute_pass_end_pass(self.id);
+        self.parent
+    }
+}
+
 impl Queue {
     pub fn submit(&self, command_buffers: &[CommandBuffer]) {
         wgn::wgpu_queue_submit(