Bug 1602131 - WebGPU render passes r=jgilbert,webidl,smaug

Adds support for recording render passes in WebGPU. The `wgpu` logo is also added to the repository ("gfx/wgpu/logo.png") under [CC-BY-SA](https://creativecommons.org/licenses/by-sa/3.0/) license. Differential Revision: https://phabricator.services.mozilla.com/D62461 --HG-- extra : moz-landing-system : lando
2020-02-19 19:25:30 +00:00 · 2020-02-19 19:25:30 +00:00 · 199a362906
--- a/dom/webgpu/CommandEncoder.cpp
+++ b/dom/webgpu/CommandEncoder.cpp
@ -10,6 +10,7 @@
 #include "Buffer.h"
 #include "ComputePassEncoder.h"
 #include "Device.h"
+#include "RenderPassEncoder.h"

 namespace mozilla {
 namespace webgpu {
@ -51,6 +52,12 @@ already_AddRefed<ComputePassEncoder> CommandEncoder::BeginComputePass(
  return pass.forget();
 }

+already_AddRefed<RenderPassEncoder> CommandEncoder::BeginRenderPass(
+    const dom::GPURenderPassDescriptor& aDesc) {
+  RefPtr<RenderPassEncoder> pass = new RenderPassEncoder(this, aDesc);
+  return pass.forget();
+}
+
 void CommandEncoder::EndComputePass(Span<const uint8_t> aData,
                                    ErrorResult& aRv) {
  if (!mValid) {
@ -67,6 +74,22 @@ void CommandEncoder::EndComputePass(Span<const uint8_t> aData,
  mBridge->SendCommandEncoderRunComputePass(mId, std::move(shmem));
 }

+void CommandEncoder::EndRenderPass(Span<const uint8_t> aData,
+                                   ErrorResult& aRv) {
+  if (!mValid) {
+    return aRv.ThrowInvalidStateError("Command encoder is not valid");
+  }
+  ipc::Shmem shmem;
+  if (!mBridge->AllocShmem(aData.Length(), ipc::Shmem::SharedMemory::TYPE_BASIC,
+                           &shmem)) {
+    return aRv.ThrowAbortError(nsPrintfCString(
+        "Unable to allocate shmem of size %zu", aData.Length()));
+  }
+
+  memcpy(shmem.get<uint8_t>(), aData.data(), aData.Length());
+  mBridge->SendCommandEncoderRunRenderPass(mId, std::move(shmem));
+}
+
 already_AddRefed<CommandBuffer> CommandEncoder::Finish(
    const dom::GPUCommandBufferDescriptor& aDesc) {
  RawId id = 0;
--- a/dom/webgpu/CommandEncoder.h
+++ b/dom/webgpu/CommandEncoder.h
@ -49,6 +49,7 @@ class CommandEncoder final : public ObjectBase, public ChildOf<Device> {

 public:
  void EndComputePass(Span<const uint8_t> aData, ErrorResult& aRv);
+  void EndRenderPass(Span<const uint8_t> aData, ErrorResult& aRv);

  void CopyBufferToBuffer(const Buffer& aSource, BufferAddress aSourceOffset,
                          const Buffer& aDestination,
@ -56,6 +57,8 @@ class CommandEncoder final : public ObjectBase, public ChildOf<Device> {
                          BufferAddress aSize);
  already_AddRefed<ComputePassEncoder> BeginComputePass(
      const dom::GPUComputePassDescriptor& aDesc);
+  already_AddRefed<RenderPassEncoder> BeginRenderPass(
+      const dom::GPURenderPassDescriptor& aDesc);
  already_AddRefed<CommandBuffer> Finish(
      const dom::GPUCommandBufferDescriptor& aDesc);
 };
--- a/dom/webgpu/RenderPassEncoder.cpp
+++ b/dom/webgpu/RenderPassEncoder.cpp
@ -5,7 +5,9 @@

 #include "mozilla/dom/WebGPUBinding.h"
 #include "RenderPassEncoder.h"
+#include "BindGroup.h"
 #include "CommandEncoder.h"
+#include "RenderPipeline.h"

 namespace mozilla {
 namespace webgpu {
@ -13,13 +15,174 @@ namespace webgpu {
 GPU_IMPL_CYCLE_COLLECTION(RenderPassEncoder, mParent)
 GPU_IMPL_JS_WRAP(RenderPassEncoder)

-RenderPassEncoder::~RenderPassEncoder() = default;
+ffi::WGPULoadOp ConvertLoadOp(const dom::GPULoadOp& aOp) {
+  switch (aOp) {
+    case dom::GPULoadOp::Load:
+      return ffi::WGPULoadOp_Load;
+    default:
+      MOZ_CRASH("Unexpected load op");
+  }
+}
+
+ffi::WGPUStoreOp ConvertStoreOp(const dom::GPUStoreOp& aOp) {
+  switch (aOp) {
+    case dom::GPUStoreOp::Store:
+      return ffi::WGPUStoreOp_Store;
+    case dom::GPUStoreOp::Clear:
+      return ffi::WGPUStoreOp_Clear;
+    default:
+      MOZ_CRASH("Unexpected load op");
+  }
+}
+
+ffi::WGPUColor ConvertColor(const dom::GPUColorDict& aColor) {
+  ffi::WGPUColor color = {aColor.mR, aColor.mG, aColor.mB, aColor.mA};
+  return color;
+}
+
+ffi::WGPURawPass BeginRenderPass(RawId aEncoderId,
+                                 const dom::GPURenderPassDescriptor& aDesc) {
+  ffi::WGPURenderPassDescriptor desc = {};
+
+  ffi::WGPURenderPassDepthStencilAttachmentDescriptor dsDesc = {};
+  if (aDesc.mDepthStencilAttachment.WasPassed()) {
+    const auto& dsa = aDesc.mDepthStencilAttachment.Value();
+    dsDesc.attachment = dsa.mAttachment->mId;
+
+    if (dsa.mDepthLoadValue.IsFloat()) {
+      dsDesc.depth_load_op = ffi::WGPULoadOp_Clear;
+      dsDesc.clear_depth = dsa.mDepthLoadValue.GetAsFloat();
+    }
+    if (dsa.mDepthLoadValue.IsGPULoadOp()) {
+      dsDesc.depth_load_op =
+          ConvertLoadOp(dsa.mDepthLoadValue.GetAsGPULoadOp());
+    }
+    dsDesc.depth_store_op = ConvertStoreOp(dsa.mDepthStoreOp);
+
+    if (dsa.mStencilLoadValue.IsUnsignedLong()) {
+      dsDesc.stencil_load_op = ffi::WGPULoadOp_Clear;
+      dsDesc.clear_stencil = dsa.mStencilLoadValue.GetAsUnsignedLong();
+    }
+    if (dsa.mStencilLoadValue.IsGPULoadOp()) {
+      dsDesc.stencil_load_op =
+          ConvertLoadOp(dsa.mStencilLoadValue.GetAsGPULoadOp());
+    }
+    dsDesc.stencil_store_op = ConvertStoreOp(dsa.mStencilStoreOp);
+
+    desc.depth_stencil_attachment = &dsDesc;
+  }
+
+  std::array<ffi::WGPURenderPassColorAttachmentDescriptor,
+             WGPUMAX_COLOR_TARGETS>
+      colorDescs = {};
+  desc.color_attachments = colorDescs.data();
+  for (size_t i = 0; i < aDesc.mColorAttachments.Length(); ++i) {
+    const auto& ca = aDesc.mColorAttachments[i];
+    ffi::WGPURenderPassColorAttachmentDescriptor& cd = colorDescs[i];
+    cd.attachment = ca.mAttachment->mId;
+    cd.store_op = ConvertStoreOp(ca.mStoreOp);
+
+    if (ca.mResolveTarget.WasPassed()) {
+      cd.resolve_target = &ca.mResolveTarget.Value().mId;
+    }
+    if (ca.mLoadValue.IsGPULoadOp()) {
+      cd.load_op = ConvertLoadOp(ca.mLoadValue.GetAsGPULoadOp());
+    } else {
+      cd.load_op = ffi::WGPULoadOp_Clear;
+      if (ca.mLoadValue.IsDoubleSequence()) {
+        const auto& seq = ca.mLoadValue.GetAsDoubleSequence();
+        if (seq.Length() >= 1) {
+          cd.clear_color.r = seq[0];
+        }
+        if (seq.Length() >= 2) {
+          cd.clear_color.g = seq[1];
+        }
+        if (seq.Length() >= 3) {
+          cd.clear_color.b = seq[2];
+        }
+        if (seq.Length() >= 4) {
+          cd.clear_color.a = seq[3];
+        }
+      }
+      if (ca.mLoadValue.IsGPUColorDict()) {
+        cd.clear_color =
+            ConvertColor(ca.mLoadValue.GetAsGPUColorDict());
+      }
+    }
+  }
+
+  return ffi::wgpu_command_encoder_begin_render_pass(aEncoderId, &desc);
+}
+
+RenderPassEncoder::RenderPassEncoder(CommandEncoder* const aParent,
+                                     const dom::GPURenderPassDescriptor& aDesc)
+    : ChildOf(aParent), mRaw(BeginRenderPass(aParent->mId, aDesc)) {}
+
+RenderPassEncoder::~RenderPassEncoder() {
+  if (mValid) {
+    mValid = false;
+    ffi::wgpu_render_pass_destroy(mRaw);
+  }
+}

 void RenderPassEncoder::SetBindGroup(
    uint32_t aSlot, const BindGroup& aBindGroup,
    const dom::Sequence<uint32_t>& aDynamicOffsets) {
  if (mValid) {
-    MOZ_CRASH("TODO");
+    ffi::wgpu_render_pass_set_bind_group(&mRaw, aSlot, aBindGroup.mId,
+                                         aDynamicOffsets.Elements(),
+                                         aDynamicOffsets.Length());
+  }
+}
+
+void RenderPassEncoder::SetPipeline(const RenderPipeline& aPipeline) {
+  if (mValid) {
+    ffi::wgpu_render_pass_set_pipeline(&mRaw, aPipeline.mId);
+  }
+}
+
+void RenderPassEncoder::SetIndexBuffer(const Buffer& aBuffer,
+                                       uint64_t aOffset) {
+  if (mValid) {
+    ffi::wgpu_render_pass_set_index_buffer(&mRaw, aBuffer.mId, aOffset);
+  }
+}
+
+void RenderPassEncoder::SetVertexBuffer(uint32_t aSlot, const Buffer& aBuffer,
+                                        uint64_t aOffset) {
+  if (mValid) {
+    // TODO: change the Rust API to use a single vertex buffer?
+    ffi::wgpu_render_pass_set_vertex_buffers(&mRaw, aSlot, &aBuffer.mId,
+                                             &aOffset, 1);
+  }
+}
+
+void RenderPassEncoder::Draw(uint32_t aVertexCount, uint32_t aInstanceCount,
+                             uint32_t aFirstVertex, uint32_t aFirstInstance) {
+  if (mValid) {
+    ffi::wgpu_render_pass_draw(&mRaw, aVertexCount, aInstanceCount,
+                               aFirstVertex, aFirstInstance);
+  }
+}
+
+void RenderPassEncoder::DrawIndexed(uint32_t aIndexCount,
+                                    uint32_t aInstanceCount,
+                                    uint32_t aFirstIndex, int32_t aBaseVertex,
+                                    uint32_t aFirstInstance) {
+  if (mValid) {
+    ffi::wgpu_render_pass_draw_indexed(&mRaw, aIndexCount, aInstanceCount,
+                                       aFirstIndex, aBaseVertex,
+                                       aFirstInstance);
+  }
+}
+
+void RenderPassEncoder::EndPass(ErrorResult& aRv) {
+  if (mValid) {
+    mValid = false;
+    uintptr_t length = 0;
+    const uint8_t* pass_data = ffi::wgpu_render_pass_finish(&mRaw, &length);
+    mParent->EndRenderPass(Span(pass_data, length), aRv);
+    ffi::wgpu_render_pass_destroy(mRaw);
  }
 }

--- a/dom/webgpu/RenderPassEncoder.h
+++ b/dom/webgpu/RenderPassEncoder.h
@ -22,6 +22,7 @@ namespace webgpu {

 class CommandEncoder;
 class RenderBundle;
+class RenderPipeline;

 class RenderPassEncoder final : public ObjectBase,
                                public ChildOf<CommandEncoder> {
@ -29,15 +30,27 @@ class RenderPassEncoder final : public ObjectBase,
  GPU_DECL_CYCLE_COLLECTION(RenderPassEncoder)
  GPU_DECL_JS_WRAP(RenderPassEncoder)

-  RenderPassEncoder() = delete;
+  RenderPassEncoder(CommandEncoder* const aParent,
+                    const dom::GPURenderPassDescriptor& aDesc);

 protected:
  virtual ~RenderPassEncoder();
  void Cleanup() {}

+  ffi::WGPURawPass mRaw;
+
 public:
  void SetBindGroup(uint32_t aSlot, const BindGroup& aBindGroup,
                    const dom::Sequence<uint32_t>& aDynamicOffsets);
+  void SetPipeline(const RenderPipeline& aPipeline);
+  void SetIndexBuffer(const Buffer& aBuffer, uint64_t aOffset);
+  void SetVertexBuffer(uint32_t aSlot, const Buffer& aBuffer, uint64_t aOffset);
+  void Draw(uint32_t aVertexCount, uint32_t aInstanceCount,
+            uint32_t aFirstVertex, uint32_t aFirstInstance);
+  void DrawIndexed(uint32_t aIndexCount, uint32_t aInstanceCount,
+                   uint32_t aFirstIndex, int32_t aBaseVertex,
+                   uint32_t aFirstInstance);
+  void EndPass(ErrorResult& aRv);
 };

 }  // namespace webgpu
--- a/dom/webgpu/RenderPipeline.cpp
+++ b/dom/webgpu/RenderPipeline.cpp
@ -13,7 +13,20 @@ namespace webgpu {
 GPU_IMPL_CYCLE_COLLECTION(RenderPipeline, mParent)
 GPU_IMPL_JS_WRAP(RenderPipeline)

-RenderPipeline::~RenderPipeline() = default;
+RenderPipeline::RenderPipeline(Device* const aParent, RawId aId)
+    : ChildOf(aParent), mId(aId) {}
+
+RenderPipeline::~RenderPipeline() { Cleanup(); }
+
+void RenderPipeline::Cleanup() {
+  if (mValid && mParent) {
+    mValid = false;
+    WebGPUChild* bridge = mParent->mBridge;
+    if (bridge && bridge->IsOpen()) {
+      bridge->DestroyRenderPipeline(mId);
+    }
+  }
+}

 }  // namespace webgpu
 }  // namespace mozilla
--- a/dom/webgpu/RenderPipeline.h
+++ b/dom/webgpu/RenderPipeline.h
@ -19,10 +19,13 @@ class RenderPipeline final : public ObjectBase, public ChildOf<Device> {
  GPU_DECL_CYCLE_COLLECTION(RenderPipeline)
  GPU_DECL_JS_WRAP(RenderPipeline)

+  RenderPipeline(Device* const aParent, RawId aId);
+
+  const RawId mId;
+
 private:
-  RenderPipeline() = delete;
  virtual ~RenderPipeline();
-  void Cleanup() {}
+  void Cleanup();
 };

 }  // namespace webgpu
--- a/dom/webgpu/ipc/PWebGPU.ipdl
+++ b/dom/webgpu/ipc/PWebGPU.ipdl
@ -45,6 +45,7 @@ parent:
  async DeviceCreateCommandEncoder(RawId selfId, GPUCommandEncoderDescriptor desc, RawId newId);
  async CommandEncoderCopyBufferToBuffer(RawId selfId, RawId sourceId, BufferAddress sourceOffset, RawId destinationId, BufferAddress destinationOffset, BufferAddress size);
  async CommandEncoderRunComputePass(RawId selfId, Shmem shmem);
+  async CommandEncoderRunRenderPass(RawId selfId, Shmem shmem);
  async CommandEncoderFinish(RawId selfId, GPUCommandBufferDescriptor desc);
  async CommandEncoderDestroy(RawId selfId);
  async CommandBufferDestroy(RawId selfId);
@ -59,6 +60,7 @@ parent:
  async ShaderModuleDestroy(RawId selfId);
  async DeviceCreateComputePipeline(RawId selfId, SerialComputePipelineDescriptor desc, RawId newId);
  async ComputePipelineDestroy(RawId selfId);
+  async RenderPipelineDestroy(RawId selfId);
  async Shutdown();

 child:
--- a/dom/webgpu/ipc/WebGPUChild.cpp
+++ b/dom/webgpu/ipc/WebGPUChild.cpp
@ -248,6 +248,10 @@ void WebGPUChild::DestroyComputePipeline(RawId aId) {
  SendComputePipelineDestroy(aId);
  ffi::wgpu_client_kill_compute_pipeline_id(mClient, aId);
 }
+void WebGPUChild::DestroyRenderPipeline(RawId aId) {
+  SendRenderPipelineDestroy(aId);
+  ffi::wgpu_client_kill_render_pipeline_id(mClient, aId);
+}

 }  // namespace webgpu
 }  // namespace mozilla
--- a/dom/webgpu/ipc/WebGPUChild.h
+++ b/dom/webgpu/ipc/WebGPUChild.h
@ -66,6 +66,7 @@ class WebGPUChild final : public PWebGPUChild {
  void DestroyBindGroup(RawId aId);
  void DestroyShaderModule(RawId aId);
  void DestroyComputePipeline(RawId aId);
+  void DestroyRenderPipeline(RawId aId);

 private:
  virtual ~WebGPUChild();
--- a/dom/webgpu/ipc/WebGPUParent.cpp
+++ b/dom/webgpu/ipc/WebGPUParent.cpp
@ -146,6 +146,13 @@ ipc::IPCResult WebGPUParent::RecvCommandEncoderRunComputePass(RawId aSelfId,
  return IPC_OK();
 }

+ipc::IPCResult WebGPUParent::RecvCommandEncoderRunRenderPass(RawId aSelfId,
+                                                             Shmem&& shmem) {
+  ffi::wgpu_server_encode_render_pass(mContext, aSelfId, shmem.get<uint8_t>(),
+                                      shmem.Size<uint8_t>());
+  return IPC_OK();
+}
+
 ipc::IPCResult WebGPUParent::RecvCommandEncoderFinish(
    RawId aSelfId, const dom::GPUCommandBufferDescriptor& aDesc) {
  Unused << aDesc;
@ -272,6 +279,11 @@ ipc::IPCResult WebGPUParent::RecvComputePipelineDestroy(RawId aSelfId) {
  return IPC_OK();
 }

+ipc::IPCResult WebGPUParent::RecvRenderPipelineDestroy(RawId aSelfId) {
+  ffi::wgpu_server_render_pipeline_destroy(mContext, aSelfId);
+  return IPC_OK();
+}
+
 ipc::IPCResult WebGPUParent::RecvShutdown() {
  mTimer.Stop();
  ffi::wgpu_server_poll_all_devices(mContext, true);
--- a/dom/webgpu/ipc/WebGPUParent.h
+++ b/dom/webgpu/ipc/WebGPUParent.h
@ -47,6 +47,7 @@ class WebGPUParent final : public PWebGPUParent {
      RawId aDestinationId, BufferAddress aDestinationOffset,
      BufferAddress aSize);
  ipc::IPCResult RecvCommandEncoderRunComputePass(RawId aSelfId, Shmem&& shmem);
+  ipc::IPCResult RecvCommandEncoderRunRenderPass(RawId aSelfId, Shmem&& shmem);
  ipc::IPCResult RecvCommandEncoderFinish(
      RawId aSelfId, const dom::GPUCommandBufferDescriptor& aDesc);
  ipc::IPCResult RecvCommandEncoderDestroy(RawId aSelfId);
@ -71,6 +72,7 @@ class WebGPUParent final : public PWebGPUParent {
      RawId aSelfId, const SerialComputePipelineDescriptor& aDesc,
      RawId aNewId);
  ipc::IPCResult RecvComputePipelineDestroy(RawId aSelfId);
+  ipc::IPCResult RecvRenderPipelineDestroy(RawId aSelfId);
  ipc::IPCResult RecvShutdown();

 private:
--- a/dom/webgpu/mochitest/mochitest.ini
+++ b/dom/webgpu/mochitest/mochitest.ini
@ -7,3 +7,5 @@ prefs = dom.webgpu.enabled=true
 [test_buffer_mapping.html]
 [test_command_buffer_creation.html]
 [test_submit_compute_empty.html]
+[test_submit_render_empty.html]
+disabled=https://bugzilla.mozilla.org/show_bug.cgi?id=1614702
--- a/dom/webgpu/mochitest/test_submit_render_empty.html
+++ b/dom/webgpu/mochitest/test_submit_render_empty.html
@ -0,0 +1,32 @@
+<!DOCTYPE HTML>
+<html>
+<head>
+<meta charset='utf-8'>
+<script src="/tests/SimpleTest/SimpleTest.js"></script>
+<link rel="stylesheet" href="/tests/SimpleTest/test.css">
+</head>
+<body>
+<script>
+
+ok(SpecialPowers.getBoolPref('dom.webgpu.enabled'), 'Pref should be enabled.');
+
+const func = async function() {
+  const adapter = await navigator.gpu.requestAdapter();
+  const device = await adapter.requestDevice();
+  const encoder = device.createCommandEncoder();
+  const pass = encoder.beginRenderPass({
+    colorAttachments: [],
+  });
+  pass.endPass();
+  const command_buffer = encoder.finish();
+  device.defaultQueue.submit([command_buffer]);
+  ok(command_buffer !== undefined, 'command_buffer !== undefined');
+};
+
+SimpleTest.waitForExplicitFinish();
+func().finally(() => SimpleTest.finish());
+
+</script>
+</body>
+</html>
+
--- a/dom/webidl/WebGPU.webidl
+++ b/dom/webidl/WebGPU.webidl
@ -760,9 +760,10 @@ dictionary GPUCommandEncoderDescriptor : GPUObjectDescriptorBase {
 [Pref="dom.webgpu.enabled",
 Exposed=Window]
 interface GPUCommandEncoder {
-    //GPURenderPassEncoder beginRenderPass(GPURenderPassDescriptor descriptor);
    [NewObject]
    GPUComputePassEncoder beginComputePass(optional GPUComputePassDescriptor descriptor = {});
+    [NewObject]
+    GPURenderPassEncoder beginRenderPass(GPURenderPassDescriptor descriptor);

    void copyBufferToBuffer(
        GPUBuffer source,
@ -813,16 +814,15 @@ interface mixin GPUProgrammablePassEncoder {

 // Render Pass
 interface mixin GPURenderEncoderBase {
-    //void setPipeline(GPURenderPipeline pipeline);
+    void setPipeline(GPURenderPipeline pipeline);

-    //void setIndexBuffer(GPUBuffer buffer, u64 offset);
-    //void setVertexBuffers(u32 startSlot,
-    //                      sequence<GPUBuffer> buffers, sequence<u64> offsets);
+    void setIndexBuffer(GPUBuffer buffer, optional u64 offset = 0);
+    void setVertexBuffer(u32 slot, GPUBuffer buffer, optional u64 offset = 0);

-    //void draw(u32 vertexCount, u32 instanceCount,
-    //          u32 firstVertex, u32 firstInstance);
-    //void drawIndexed(u32 indexCount, u32 instanceCount,
-    //                 u32 firstIndex, i32 baseVertex, u32 firstInstance);
+    void draw(u32 vertexCount, u32 instanceCount,
+              u32 firstVertex, u32 firstInstance);
+    void drawIndexed(u32 indexCount, u32 instanceCount,
+                     u32 firstIndex, i32 baseVertex, u32 firstInstance);

    //void drawIndirect(GPUBuffer indirectBuffer, u64 indirectOffset);
    //void drawIndexedIndirect(GPUBuffer indirectBuffer, u64 indirectOffset);
@ -841,7 +841,8 @@ interface GPURenderPassEncoder {
    //void setStencilReference(u32 reference);

    //void executeBundles(sequence<GPURenderBundle> bundles);
-    //void endPass();
+    [Throws]
+    void endPass();
 };
 GPURenderPassEncoder includes GPUObjectBase;
 GPURenderPassEncoder includes GPUProgrammablePassEncoder;
@ -887,7 +888,7 @@ interface GPURenderBundleEncoder {
    //GPURenderBundle finish(optional GPURenderBundleDescriptor descriptor = {});
 };
 GPURenderBundleEncoder includes GPUObjectBase;
-GPURenderBundleEncoder includes GPURenderEncoderBase;
+//GPURenderBundleEncoder includes GPURenderEncoderBase;

 dictionary GPURenderBundleDescriptor : GPUObjectDescriptorBase {
 };
--- a/gfx/wgpu/README.md
+++ b/gfx/wgpu/README.md
@ -1,28 +1,30 @@
-This is an active GitHub mirror of the WebGPU native implementation in Rust, which now lives in [Mozilla-central](https://hg.mozilla.org/mozilla-central). Issues and pull requests are accepted, but we merge them in m-c manually and then sync to GitHub instead of landing directly here.
+<img align="right" width="25%" src="logo.png">
+
+This is an active GitHub mirror of the WebGPU implementation in Rust, which now lives in "gfx/wgpu" of [Mozilla-central](https://hg.mozilla.org/mozilla-central/file/tip/gfx/wgpu). Issues and pull requests are accepted, but some bidirectional synchronization may be involved.

---
 # WebGPU
-[![Build Status](https://travis-ci.org/gfx-rs/wgpu.svg)](https://travis-ci.org/gfx-rs/wgpu)
+[![Matrix](https://img.shields.io/badge/Matrix-%23wgpu%3Amatrix.org-blueviolet.svg)](https://matrix.to/#/#wgpu:matrix.org)
+[![Build Status](https://travis-ci.org/gfx-rs/wgpu.svg?branch=master)](https://travis-ci.org/gfx-rs/wgpu)
+[![Crates.io](https://img.shields.io/crates/v/wgpu-core.svg?label=wgpu-core)](https://crates.io/crates/wgpu-core)
 [![Crates.io](https://img.shields.io/crates/v/wgpu-native.svg?label=wgpu-native)](https://crates.io/crates/wgpu-native)
-[![Gitter](https://badges.gitter.im/gfx-rs/webgpu.svg)](https://gitter.im/gfx-rs/webgpu)

-This is an experimental [WebGPU](https://www.w3.org/community/gpu/) implementation as a native static library. It's written in Rust and is based on [gfx-hal](https://github.com/gfx-rs/gfx) and [Rendy](https://github.com/amethyst/rendy) libraries. The corresponding WebIDL specification can be found at [gpuweb project](https://github.com/gpuweb/gpuweb/blob/master/spec/index.bs).
+This is an experimental [WebGPU](https://www.w3.org/community/gpu/) implementation, exposing both Rust and C interfaces as a native static library. It's written in Rust and is based on [gfx-hal](https://github.com/gfx-rs/gfx) and a few [Rendy](https://github.com/amethyst/rendy) bits. See the upstream [WebGPU specification](https://gpuweb.github.io/gpuweb/) (work in progress).

 The implementation consists of the following parts:
  1. `wgpu-core` - internal Rust API for WebGPU implementations to use
  2. `wgpu-native` - the native implementation of WebGPU as a C API library
-  3. `wgpu-remote` - remoting layer to work with WebGPU across the process boundary
-  4. `ffi` - the C headers generated by [cbindgen](https://github.com/eqrion/cbindgen) for both of the libraries
+  3. `wgpu-remote` - remoting layer to work with WebGPU across the process boundary, as a C API library used by Gecko
+  4. `ffi` - the C headers generated by [cbindgen](https://github.com/eqrion/cbindgen) for the native headers

 ## Supported Platforms

   API   |       Windows      |       Linux        |    macOS & iOS     |
  -----  | ------------------ | ------------------ | ------------------ |
-  DX11   | :heavy_check_mark: |                    |                    |
+  DX11   | :white_check_mark: |                    |                    |
  DX12   | :heavy_check_mark: |                    |                    |
  Vulkan | :heavy_check_mark: | :heavy_check_mark: |                    |
  Metal  |                    |                    | :heavy_check_mark: |
-  OpenGL |                    |                    |                    |
+  OpenGL | :construction:     | :construction:     | :construction:     |

 ## Usage

--- a/gfx/wgpu/ffi/wgpu.h
+++ b/gfx/wgpu/ffi/wgpu.h
@ -2,7 +2,7 @@
 * 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/. */

-/* Generated with cbindgen:0.12.2 */
+/* Generated with cbindgen:0.13.1 */

 /* DO NOT MODIFY THIS MANUALLY! This file was generated using cbindgen.
 * To generate this file:
@ -311,6 +311,8 @@ typedef uint64_t WGPUId_TextureView_Dummy;

 typedef WGPUId_TextureView_Dummy WGPUTextureViewId;

+typedef const WGPUTextureViewId *WGPUOptionRef_TextureViewId;
+
 typedef struct {
  double r;
  double g;
@ -326,13 +328,13 @@ typedef struct {

 typedef struct {
  WGPUTextureViewId attachment;
-  WGPUTextureViewId resolve_target;
+  WGPUOptionRef_TextureViewId resolve_target;
  WGPULoadOp load_op;
  WGPUStoreOp store_op;
  WGPUColor clear_color;
-} WGPURenderPassColorAttachmentDescriptorBase_TextureViewId__TextureViewId;
+} WGPURenderPassColorAttachmentDescriptorBase_TextureViewId__OptionRef_TextureViewId;

-typedef WGPURenderPassColorAttachmentDescriptorBase_TextureViewId__TextureViewId WGPURawRenderPassColorAttachmentDescriptor;
+typedef WGPURenderPassColorAttachmentDescriptorBase_TextureViewId__OptionRef_TextureViewId WGPURenderPassColorAttachmentDescriptor;

 typedef struct {
  WGPUTextureViewId attachment;
@ -346,28 +348,6 @@ typedef struct {

 typedef WGPURenderPassDepthStencilAttachmentDescriptorBase_TextureViewId WGPURenderPassDepthStencilAttachmentDescriptor;

-typedef struct {
-  WGPURawRenderPassColorAttachmentDescriptor colors[WGPUMAX_COLOR_TARGETS];
-  WGPURenderPassDepthStencilAttachmentDescriptor depth_stencil;
-} WGPURawRenderTargets;
-
-typedef struct {
-  WGPURawPass raw;
-  WGPURawRenderTargets targets;
-} WGPURawRenderPass;
-
-typedef const WGPUTextureViewId *WGPUOptionRef_TextureViewId;
-
-typedef struct {
-  WGPUTextureViewId attachment;
-  WGPUOptionRef_TextureViewId resolve_target;
-  WGPULoadOp load_op;
-  WGPUStoreOp store_op;
-  WGPUColor clear_color;
-} WGPURenderPassColorAttachmentDescriptorBase_TextureViewId__OptionRef_TextureViewId;
-
-typedef WGPURenderPassColorAttachmentDescriptorBase_TextureViewId__OptionRef_TextureViewId WGPURenderPassColorAttachmentDescriptor;
-
 typedef struct {
  const WGPURenderPassColorAttachmentDescriptor *color_attachments;
  uintptr_t color_attachments_length;
@ -680,7 +660,7 @@ typedef struct {

 typedef WGPUDeviceId WGPUQueueId;

-typedef WGPURawRenderPass *WGPURenderPassId;
+typedef WGPURawPass *WGPURenderPassId;

 typedef uint64_t WGPUId_RenderBundle_Dummy;

@ -746,12 +726,12 @@ WGPURawPass *wgpu_command_encoder_begin_compute_pass(WGPUCommandEncoderId encode
 /**
 * # Safety
 *
- * This function is unsafe as there is no guarantee that the given pointer
- * (`RenderPassDescriptor::color_attachments`) is valid for
- * `RenderPassDescriptor::color_attachments_length` elements.
+ * 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.
 */
-WGPURawRenderPass *wgpu_command_encoder_begin_render_pass(WGPUCommandEncoderId encoder_id,
-                                                          const WGPURenderPassDescriptor *desc);
+WGPURawPass *wgpu_command_encoder_begin_render_pass(WGPUCommandEncoderId encoder_id,
+                                                    const WGPURenderPassDescriptor *desc);

 void wgpu_command_encoder_copy_buffer_to_buffer(WGPUCommandEncoderId command_encoder_id,
                                                WGPUBufferId source,
@ -882,26 +862,26 @@ void wgpu_queue_submit(WGPUQueueId queue_id,
                       const WGPUCommandBufferId *command_buffers,
                       uintptr_t command_buffers_length);

-void wgpu_render_pass_destroy(WGPURawRenderPass *pass);
+void wgpu_render_pass_destroy(WGPURawPass *pass);

-void wgpu_render_pass_draw(WGPURawRenderPass *pass,
+void wgpu_render_pass_draw(WGPURawPass *pass,
                           uint32_t vertex_count,
                           uint32_t instance_count,
                           uint32_t first_vertex,
                           uint32_t first_instance);

-void wgpu_render_pass_draw_indexed(WGPURawRenderPass *pass,
+void wgpu_render_pass_draw_indexed(WGPURawPass *pass,
                                   uint32_t index_count,
                                   uint32_t instance_count,
                                   uint32_t first_index,
                                   int32_t base_vertex,
                                   uint32_t first_instance);

-void wgpu_render_pass_draw_indexed_indirect(WGPURawRenderPass *pass,
+void wgpu_render_pass_draw_indexed_indirect(WGPURawPass *pass,
                                            WGPUBufferId buffer_id,
                                            WGPUBufferAddress offset);

-void wgpu_render_pass_draw_indirect(WGPURawRenderPass *pass,
+void wgpu_render_pass_draw_indirect(WGPURawPass *pass,
                                    WGPUBufferId buffer_id,
                                    WGPUBufferAddress offset);

@ -914,17 +894,17 @@ void wgpu_render_pass_draw_indirect(WGPURawRenderPass *pass,
 */
 void wgpu_render_pass_end_pass(WGPURenderPassId pass_id);

-void wgpu_render_pass_execute_bundles(WGPURawRenderPass *_pass,
+void wgpu_render_pass_execute_bundles(WGPURawPass *_pass,
                                      const WGPURenderBundleId *_bundles,
                                      uintptr_t _bundles_length);

-const uint8_t *wgpu_render_pass_finish(WGPURawRenderPass *pass, uintptr_t *length);
+const uint8_t *wgpu_render_pass_finish(WGPURawPass *pass, uintptr_t *length);

-void wgpu_render_pass_insert_debug_marker(WGPURawRenderPass *_pass, WGPURawString _label);
+void wgpu_render_pass_insert_debug_marker(WGPURawPass *_pass, WGPURawString _label);

-void wgpu_render_pass_pop_debug_group(WGPURawRenderPass *_pass);
+void wgpu_render_pass_pop_debug_group(WGPURawPass *_pass);

-void wgpu_render_pass_push_debug_group(WGPURawRenderPass *_pass, WGPURawString _label);
+void wgpu_render_pass_push_debug_group(WGPURawPass *_pass, WGPURawString _label);

 /**
 * # Safety
@ -932,27 +912,27 @@ void wgpu_render_pass_push_debug_group(WGPURawRenderPass *_pass, WGPURawString _
 * This function is unsafe as there is no guarantee that the given pointer is
 * valid for `offset_length` elements.
 */
-void wgpu_render_pass_set_bind_group(WGPURawRenderPass *pass,
+void wgpu_render_pass_set_bind_group(WGPURawPass *pass,
                                     uint32_t index,
                                     WGPUBindGroupId bind_group_id,
                                     const WGPUDynamicOffset *offsets,
                                     uintptr_t offset_length);

-void wgpu_render_pass_set_blend_color(WGPURawRenderPass *pass, const WGPUColor *color);
+void wgpu_render_pass_set_blend_color(WGPURawPass *pass, const WGPUColor *color);

-void wgpu_render_pass_set_index_buffer(WGPURawRenderPass *pass,
+void wgpu_render_pass_set_index_buffer(WGPURawPass *pass,
                                       WGPUBufferId buffer_id,
                                       WGPUBufferAddress offset);

-void wgpu_render_pass_set_pipeline(WGPURawRenderPass *pass, WGPURenderPipelineId pipeline_id);
+void wgpu_render_pass_set_pipeline(WGPURawPass *pass, WGPURenderPipelineId pipeline_id);

-void wgpu_render_pass_set_scissor_rect(WGPURawRenderPass *pass,
+void wgpu_render_pass_set_scissor_rect(WGPURawPass *pass,
                                       uint32_t x,
                                       uint32_t y,
                                       uint32_t w,
                                       uint32_t h);

-void wgpu_render_pass_set_stencil_reference(WGPURawRenderPass *pass, uint32_t value);
+void wgpu_render_pass_set_stencil_reference(WGPURawPass *pass, uint32_t value);

 /**
 * # Safety
@ -960,13 +940,13 @@ void wgpu_render_pass_set_stencil_reference(WGPURawRenderPass *pass, uint32_t va
 * This function is unsafe as there is no guarantee that the given pointers
 * (`buffer_ids` and `offsets`) are valid for `length` elements.
 */
-void wgpu_render_pass_set_vertex_buffers(WGPURawRenderPass *pass,
+void wgpu_render_pass_set_vertex_buffers(WGPURawPass *pass,
                                         uint32_t start_slot,
                                         const WGPUBufferId *buffer_ids,
                                         const WGPUBufferAddress *offsets,
                                         uintptr_t length);

-void wgpu_render_pass_set_viewport(WGPURawRenderPass *pass,
+void wgpu_render_pass_set_viewport(WGPURawPass *pass,
                                   float x,
                                   float y,
                                   float w,
--- a/gfx/wgpu/logo.png
+++ b/gfx/wgpu/logo.png
--- a/gfx/wgpu/wgpu-core/src/binding_model.rs
+++ b/gfx/wgpu/wgpu-core/src/binding_model.rs
@ -87,6 +87,7 @@ pub struct PipelineLayout<B: hal::Backend> {

 #[repr(C)]
 #[derive(Debug)]
+#[cfg_attr(feature = "serde", derive(Serialize, Deserialize))]
 pub struct BufferBinding {
    pub buffer: BufferId,
    pub offset: BufferAddress,
@ -95,6 +96,7 @@ pub struct BufferBinding {

 #[repr(C)]
 #[derive(Debug)]
+#[cfg_attr(feature = "serde", derive(Serialize, Deserialize))]
 pub enum BindingResource {
    Buffer(BufferBinding),
    Sampler(SamplerId),
@ -103,6 +105,7 @@ pub enum BindingResource {

 #[repr(C)]
 #[derive(Debug)]
+#[cfg_attr(feature = "serde", derive(Serialize, Deserialize))]
 pub struct BindGroupBinding {
    pub binding: u32,
    pub resource: BindingResource,
--- a/gfx/wgpu/wgpu-core/src/command/bind.rs
+++ b/gfx/wgpu/wgpu-core/src/command/bind.rs
@ -81,7 +81,7 @@ impl BindGroupEntry {
                }
                self.expected_layout_id == Some(layout_id)
            }
-            None => true,
+            None => false,
        };

        self.provided = Some(BindGroupPair {
--- a/gfx/wgpu/wgpu-core/src/command/compute.rs
+++ b/gfx/wgpu/wgpu-core/src/command/compute.rs
@ -40,7 +40,7 @@ enum ComputeCommand {
 }

 impl super::RawPass {
-    pub fn new_compute(parent: id::CommandEncoderId) -> Self {
+    pub unsafe fn new_compute(parent: id::CommandEncoderId) -> Self {
        Self::from_vec(Vec::<ComputeCommand>::with_capacity(1), parent)
    }

--- a/gfx/wgpu/wgpu-core/src/command/mod.rs
+++ b/gfx/wgpu/wgpu-core/src/command/mod.rs
@ -208,56 +208,16 @@ pub struct CommandBufferDescriptor {
    pub todo: u32,
 }

-type RawRenderPassColorAttachmentDescriptor =
+pub type RawRenderPassColorAttachmentDescriptor =
    RenderPassColorAttachmentDescriptorBase<id::TextureViewId, id::TextureViewId>;

 #[repr(C)]
+#[derive(peek_poke::PeekCopy, peek_poke::Poke)]
 pub struct RawRenderTargets {
    pub colors: [RawRenderPassColorAttachmentDescriptor; MAX_COLOR_TARGETS],
    pub depth_stencil: RenderPassDepthStencilAttachmentDescriptor,
 }

-#[repr(C)]
-pub struct RawRenderPass {
-    raw: RawPass,
-    targets: RawRenderTargets,
-}
-
-/// # Safety
-///
-/// This function is unsafe as there is no guarantee that the given pointer
-/// (`RenderPassDescriptor::color_attachments`) is valid for
-/// `RenderPassDescriptor::color_attachments_length` elements.
-#[no_mangle]
-pub unsafe extern "C" fn wgpu_command_encoder_begin_render_pass(
-    encoder_id: id::CommandEncoderId,
-    desc: &RenderPassDescriptor,
-) -> *mut RawRenderPass {
-    let mut colors: [RawRenderPassColorAttachmentDescriptor; MAX_COLOR_TARGETS] = mem::zeroed();
-    for (color, at) in colors
-        .iter_mut()
-        .zip(slice::from_raw_parts(desc.color_attachments, desc.color_attachments_length))
-    {
-        *color = RawRenderPassColorAttachmentDescriptor {
-            attachment: at.attachment,
-            resolve_target: at.resolve_target.map_or(id::TextureViewId::ERROR, |rt| *rt),
-            load_op: at.load_op,
-            store_op: at.store_op,
-            clear_color: at.clear_color,
-        };
-    }
-    let pass = RawRenderPass {
-        raw: RawPass::new_render(encoder_id),
-        targets: RawRenderTargets {
-            colors,
-            depth_stencil: desc.depth_stencil_attachment
-                .cloned()
-                .unwrap_or_else(|| mem::zeroed()),
-        },
-    };
-    Box::into_raw(Box::new(pass))
-}
-
 impl<F> Global<F> {
    pub fn command_encoder_finish<B: GfxBackend>(
        &self,
--- a/gfx/wgpu/wgpu-core/src/command/render.rs
+++ b/gfx/wgpu/wgpu-core/src/command/render.rs
@ -6,6 +6,7 @@ use crate::{
    command::{
        bind::{Binder, LayoutChange},
        PhantomSlice,
+        RawRenderPassColorAttachmentDescriptor,
        RawRenderTargets,
    },
    conv,
@ -37,26 +38,28 @@ use std::{
    collections::hash_map::Entry,
    iter,
    marker::PhantomData,
+    mem,
    ops::Range,
+    slice,
 };


 #[repr(C)]
-#[derive(Copy, Clone, Debug, Hash, Eq, PartialEq)]
+#[derive(Copy, Clone, Debug, Hash, Eq, PartialEq, PeekCopy, Poke)]
 pub enum LoadOp {
    Clear = 0,
    Load = 1,
 }

 #[repr(C)]
-#[derive(Copy, Clone, Debug, Hash, Eq, PartialEq)]
+#[derive(Copy, Clone, Debug, Hash, Eq, PartialEq, PeekCopy, Poke)]
 pub enum StoreOp {
    Clear = 0,
    Store = 1,
 }

 #[repr(C)]
-#[derive(Debug)]
+#[derive(Debug, PeekCopy, Poke)]
 pub struct RenderPassColorAttachmentDescriptorBase<T, R> {
    pub attachment: T,
    pub resolve_target: R,
@ -66,7 +69,7 @@ pub struct RenderPassColorAttachmentDescriptorBase<T, R> {
 }

 #[repr(C)]
-#[derive(Clone, Debug)]
+#[derive(Clone, Debug, PeekCopy, Poke)]
 pub struct RenderPassDepthStencilAttachmentDescriptorBase<T> {
    pub attachment: T,
    pub depth_load_op: LoadOp,
@ -154,16 +157,36 @@ enum RenderCommand {
 }

 impl super::RawPass {
-    pub fn new_render(parent_id: id::CommandEncoderId) -> Self {
-        Self::from_vec(Vec::<RenderCommand>::with_capacity(1), parent_id)
-    }
-}
+    pub unsafe fn new_render(parent_id: id::CommandEncoderId, desc: &RenderPassDescriptor) -> Self {
+        let mut pass = Self::from_vec(Vec::<RenderCommand>::with_capacity(1), parent_id);

-impl super::RawRenderPass {
-    pub unsafe fn finish_render(mut self) -> (Vec<u8>, id::CommandEncoderId, RawRenderTargets) {
-        self.raw.finish(RenderCommand::End);
-        let (vec, parent_id) = self.raw.into_vec();
-        (vec, parent_id, self.targets)
+        let mut targets = RawRenderTargets {
+            depth_stencil: desc.depth_stencil_attachment
+                .cloned()
+                .unwrap_or_else(|| mem::zeroed()),
+            colors: mem::zeroed(),
+        };
+        for (color, at) in targets.colors
+            .iter_mut()
+            .zip(slice::from_raw_parts(desc.color_attachments, desc.color_attachments_length))
+        {
+            *color = RawRenderPassColorAttachmentDescriptor {
+                attachment: at.attachment,
+                resolve_target: at.resolve_target.map_or(id::TextureViewId::ERROR, |rt| *rt),
+                load_op: at.load_op,
+                store_op: at.store_op,
+                clear_color: at.clear_color,
+            };
+        }
+
+        pass.encode(&targets);
+        pass
+    }
+
+    pub unsafe fn finish_render(mut self) -> (Vec<u8>, id::CommandEncoderId) {
+        self.finish(RenderCommand::End);
+        let (vec, parent_id) = self.into_vec();
+        (vec, parent_id)
    }
 }

@ -289,8 +312,6 @@ impl<F> Global<F> {
    pub fn command_encoder_run_render_pass<B: GfxBackend>(
        &self,
        encoder_id: id::CommandEncoderId,
-        color_attachments: &[RenderPassColorAttachmentDescriptor],
-        depth_stencil_attachment: Option<&RenderPassDepthStencilAttachmentDescriptor>,
        raw_data: &[u8],
    ) {
        let hub = B::hub(self);
@ -316,6 +337,38 @@ impl<F> Global<F> {
        let (texture_guard, mut token) = hub.textures.read(&mut token);
        let (view_guard, _) = hub.texture_views.read(&mut token);

+        let mut peeker = raw_data.as_ptr();
+        let raw_data_end = unsafe {
+            raw_data.as_ptr().add(raw_data.len())
+        };
+
+        let mut targets: RawRenderTargets = unsafe { mem::zeroed() };
+        assert!(unsafe { peeker.add(RawRenderTargets::max_size()) <= raw_data_end });
+        peeker = unsafe { targets.peek_from(peeker) };
+
+        let color_attachments = targets.colors
+            .iter()
+            .take_while(|at| at.attachment != id::TextureViewId::ERROR)
+            .map(|at| {
+                RenderPassColorAttachmentDescriptor {
+                    attachment: at.attachment,
+                    resolve_target: if at.resolve_target == id::TextureViewId::ERROR {
+                        None
+                    } else {
+                        Some(&at.resolve_target)
+                    },
+                    load_op: at.load_op,
+                    store_op: at.store_op,
+                    clear_color: at.clear_color,
+                }
+            })
+            .collect::<arrayvec::ArrayVec<[_; MAX_COLOR_TARGETS]>>();
+        let depth_stencil_attachment = if targets.depth_stencil.attachment == id::TextureViewId::ERROR {
+            None
+        } else {
+            Some(&targets.depth_stencil)
+        };
+
        let (context, sample_count) = {
            use hal::{adapter::PhysicalDevice as _, device::Device as _};

@ -401,7 +454,7 @@ impl<F> Global<F> {
                let mut colors = ArrayVec::new();
                let mut resolves = ArrayVec::new();

-                for at in color_attachments {
+                for at in &color_attachments {
                    let view = &view_guard[at.attachment];
                    if let Some(ex) = extent {
                        assert_eq!(ex, view.extent);
@ -770,10 +823,6 @@ impl<F> Global<F> {
            },
        };

-        let mut peeker = raw_data.as_ptr();
-        let raw_data_end = unsafe {
-            raw_data.as_ptr().add(raw_data.len())
-        };
        let mut command = RenderCommand::Draw {
            vertex_count: 0,
            instance_count: 0,
@ -1113,7 +1162,7 @@ impl<F> Global<F> {
 pub mod render_ffi {
    use super::{
        RenderCommand,
-        super::{PhantomSlice, RawRenderPass, Rect},
+        super::{PhantomSlice, RawPass, Rect},
    };
    use crate::{
        id,
@ -1132,38 +1181,38 @@ pub mod render_ffi {
    // `RawPass::encode` and `RawPass::encode_slice`.
    #[no_mangle]
    pub unsafe extern "C" fn wgpu_render_pass_set_bind_group(
-        pass: &mut RawRenderPass,
+        pass: &mut RawPass,
        index: u32,
        bind_group_id: id::BindGroupId,
        offsets: *const DynamicOffset,
        offset_length: usize,
    ) {
-        pass.raw.encode(&RenderCommand::SetBindGroup {
+        pass.encode(&RenderCommand::SetBindGroup {
            index: index.try_into().unwrap(),
            num_dynamic_offsets: offset_length.try_into().unwrap(),
            bind_group_id,
            phantom_offsets: PhantomSlice::new(),
        });
-        pass.raw.encode_slice(
+        pass.encode_slice(
            slice::from_raw_parts(offsets, offset_length),
        );
    }

    #[no_mangle]
    pub unsafe extern "C" fn wgpu_render_pass_set_pipeline(
-        pass: &mut RawRenderPass,
+        pass: &mut RawPass,
        pipeline_id: id::RenderPipelineId,
    ) {
-        pass.raw.encode(&RenderCommand::SetPipeline(pipeline_id));
+        pass.encode(&RenderCommand::SetPipeline(pipeline_id));
    }

    #[no_mangle]
    pub unsafe extern "C" fn wgpu_render_pass_set_index_buffer(
-        pass: &mut RawRenderPass,
+        pass: &mut RawPass,
        buffer_id: id::BufferId,
        offset: BufferAddress,
    ) {
-        pass.raw.encode(&RenderCommand::SetIndexBuffer {
+        pass.encode(&RenderCommand::SetIndexBuffer {
            buffer_id,
            offset,
        });
@ -1177,45 +1226,45 @@ pub mod render_ffi {
    // `RawPass::encode` and `RawPass::encode_slice`.
    #[no_mangle]
    pub unsafe extern "C" fn wgpu_render_pass_set_vertex_buffers(
-        pass: &mut RawRenderPass,
+        pass: &mut RawPass,
        start_slot: u32,
        buffer_ids: *const id::BufferId,
        offsets: *const BufferAddress,
        length: usize,
    ) {
-        pass.raw.encode(&RenderCommand::SetVertexBuffers {
+        pass.encode(&RenderCommand::SetVertexBuffers {
            start_index: start_slot.try_into().unwrap(),
            count: length.try_into().unwrap(),
            phantom_buffer_ids: PhantomSlice::new(),
            phantom_offsets: PhantomSlice::new(),
        });
-        pass.raw.encode_slice(
+        pass.encode_slice(
            slice::from_raw_parts(buffer_ids, length),
        );
-        pass.raw.encode_slice(
+        pass.encode_slice(
            slice::from_raw_parts(offsets, length),
        );
    }

    #[no_mangle]
    pub unsafe extern "C" fn wgpu_render_pass_set_blend_color(
-        pass: &mut RawRenderPass,
+        pass: &mut RawPass,
        color: &Color,
    ) {
-        pass.raw.encode(&RenderCommand::SetBlendColor(*color));
+        pass.encode(&RenderCommand::SetBlendColor(*color));
    }

    #[no_mangle]
    pub unsafe extern "C" fn wgpu_render_pass_set_stencil_reference(
-        pass: &mut RawRenderPass,
+        pass: &mut RawPass,
        value: u32,
    ) {
-        pass.raw.encode(&RenderCommand::SetStencilReference(value));
+        pass.encode(&RenderCommand::SetStencilReference(value));
    }

    #[no_mangle]
    pub unsafe extern "C" fn wgpu_render_pass_set_viewport(
-        pass: &mut RawRenderPass,
+        pass: &mut RawPass,
        x: f32,
        y: f32,
        w: f32,
@ -1223,7 +1272,7 @@ pub mod render_ffi {
        depth_min: f32,
        depth_max: f32,
    ) {
-        pass.raw.encode(&RenderCommand::SetViewport {
+        pass.encode(&RenderCommand::SetViewport {
            rect: Rect { x, y, w, h },
            depth_min,
            depth_max,
@ -1232,24 +1281,24 @@ pub mod render_ffi {

    #[no_mangle]
    pub unsafe extern "C" fn wgpu_render_pass_set_scissor_rect(
-        pass: &mut RawRenderPass,
+        pass: &mut RawPass,
        x: u32,
        y: u32,
        w: u32,
        h: u32,
    ) {
-        pass.raw.encode(&RenderCommand::SetScissor(Rect { x, y, w, h }));
+        pass.encode(&RenderCommand::SetScissor(Rect { x, y, w, h }));
    }

    #[no_mangle]
    pub unsafe extern "C" fn wgpu_render_pass_draw(
-        pass: &mut RawRenderPass,
+        pass: &mut RawPass,
        vertex_count: u32,
        instance_count: u32,
        first_vertex: u32,
        first_instance: u32,
    ) {
-        pass.raw.encode(&RenderCommand::Draw {
+        pass.encode(&RenderCommand::Draw {
            vertex_count,
            instance_count,
            first_vertex,
@ -1259,14 +1308,14 @@ pub mod render_ffi {

    #[no_mangle]
    pub unsafe extern "C" fn wgpu_render_pass_draw_indexed(
-        pass: &mut RawRenderPass,
+        pass: &mut RawPass,
        index_count: u32,
        instance_count: u32,
        first_index: u32,
        base_vertex: i32,
        first_instance: u32,
    ) {
-        pass.raw.encode(&RenderCommand::DrawIndexed {
+        pass.encode(&RenderCommand::DrawIndexed {
            index_count,
            instance_count,
            first_index,
@ -1277,11 +1326,11 @@ pub mod render_ffi {

    #[no_mangle]
    pub unsafe extern "C" fn wgpu_render_pass_draw_indirect(
-        pass: &mut RawRenderPass,
+        pass: &mut RawPass,
        buffer_id: id::BufferId,
        offset: BufferAddress,
    ) {
-        pass.raw.encode(&RenderCommand::DrawIndirect {
+        pass.encode(&RenderCommand::DrawIndirect {
            buffer_id,
            offset,
        });
@ -1289,11 +1338,11 @@ pub mod render_ffi {

    #[no_mangle]
    pub unsafe extern "C" fn wgpu_render_pass_draw_indexed_indirect(
-        pass: &mut RawRenderPass,
+        pass: &mut RawPass,
        buffer_id: id::BufferId,
        offset: BufferAddress,
    ) {
-        pass.raw.encode(&RenderCommand::DrawIndexedIndirect {
+        pass.encode(&RenderCommand::DrawIndexedIndirect {
            buffer_id,
            offset,
        });
@ -1301,7 +1350,7 @@ pub mod render_ffi {

    #[no_mangle]
    pub extern "C" fn wgpu_render_pass_execute_bundles(
-        _pass: &mut RawRenderPass,
+        _pass: &mut RawPass,
        _bundles: *const id::RenderBundleId,
        _bundles_length: usize,
    ) {
@ -1310,7 +1359,7 @@ pub mod render_ffi {

    #[no_mangle]
    pub extern "C" fn wgpu_render_pass_push_debug_group(
-        _pass: &mut RawRenderPass,
+        _pass: &mut RawPass,
        _label: RawString,
    ) {
        //TODO
@ -1318,14 +1367,14 @@ pub mod render_ffi {

    #[no_mangle]
    pub extern "C" fn wgpu_render_pass_pop_debug_group(
-        _pass: &mut RawRenderPass,
+        _pass: &mut RawPass,
    ) {
        //TODO
    }

    #[no_mangle]
    pub extern "C" fn wgpu_render_pass_insert_debug_marker(
-        _pass: &mut RawRenderPass,
+        _pass: &mut RawPass,
        _label: RawString,
    ) {
        //TODO
@ -1333,17 +1382,11 @@ pub mod render_ffi {

    #[no_mangle]
    pub unsafe extern "C" fn wgpu_render_pass_finish(
-        pass: &mut RawRenderPass,
+        pass: &mut RawPass,
        length: &mut usize,
    ) -> *const u8 {
-        //TODO: put target information into the byte stream
-        pass.raw.finish(RenderCommand::End);
-        *length = pass.raw.size();
-        pass.raw.base
-    }
-
-    #[no_mangle]
-    pub unsafe extern "C" fn wgpu_render_pass_destroy(pass: *mut RawRenderPass) {
-        let _ = Box::from_raw(pass).raw.into_vec();
+        pass.finish(RenderCommand::End);
+        *length = pass.size();
+        pass.base
    }
 }
--- a/gfx/wgpu/wgpu-core/src/device/mod.rs
+++ b/gfx/wgpu/wgpu-core/src/device/mod.rs
@ -1369,6 +1369,9 @@ impl<F: AllIdentityFilter + IdentityFilter<id::CommandBufferId>> Global<F> {

                if let Some((sc_id, fbo)) = comb.used_swap_chain.take() {
                    let sc = &mut swap_chain_guard[sc_id.value];
+                    assert!(sc.acquired_view_id.is_some(),
+                        "SwapChainOutput for {:?} was dropped before the respective command buffer {:?} got submitted!",
+                        sc_id.value, cmb_id);
                    if sc.acquired_framebuffers.is_empty() {
                        signal_swapchain_semaphores.push(sc_id.value);
                    }
--- a/gfx/wgpu/wgpu-core/src/id.rs
+++ b/gfx/wgpu/wgpu-core/src/id.rs
@ -115,7 +115,7 @@ pub type ComputePipelineId = Id<crate::pipeline::ComputePipeline<Dummy>>;
 // Command
 pub type CommandBufferId = Id<crate::command::CommandBuffer<Dummy>>;
 pub type CommandEncoderId = CommandBufferId;
-pub type RenderPassId = *mut crate::command::RawRenderPass;
+pub type RenderPassId = *mut crate::command::RawPass;
 pub type ComputePassId = *mut crate::command::RawPass;
 pub type RenderBundleId = Id<crate::command::RenderBundle<Dummy>>;
 // Swap chain
--- a/gfx/wgpu/wgpu-core/src/instance.rs
+++ b/gfx/wgpu/wgpu-core/src/instance.rs
@ -15,8 +15,12 @@ use crate::{
 #[cfg(feature = "serde")]
 use serde::{Deserialize, Serialize};

-pub use hal::adapter::{AdapterInfo, DeviceType};
-use hal::{self, adapter::PhysicalDevice as _, queue::QueueFamily as _, Instance as _};
+use hal::{
+    self,
+    adapter::{AdapterInfo as HalAdapterInfo, DeviceType as HalDeviceType, PhysicalDevice as _},
+    queue::QueueFamily as _,
+    Instance as _,
+};


 #[derive(Debug)]
@ -173,6 +177,69 @@ impl From<Backend> for BackendBit {
    }
 }

+/// Metadata about a backend adapter.
+#[derive(Clone, Debug, PartialEq)]
+#[cfg_attr(feature = "serde", derive(Serialize, Deserialize))]
+pub struct AdapterInfo {
+    /// Adapter name
+    pub name: String,
+    /// Vendor PCI id of the adapter
+    pub vendor: usize,
+    /// PCI id of the adapter
+    pub device: usize,
+    /// Type of device
+    pub device_type: DeviceType,
+    /// Backend used for device
+    pub backend: Backend,
+}
+
+impl AdapterInfo {
+    fn from_gfx(adapter_info: HalAdapterInfo, backend: Backend) -> Self {
+        let HalAdapterInfo {
+            name,
+            vendor,
+            device,
+            device_type,
+        } = adapter_info;
+
+        AdapterInfo {
+            name,
+            vendor,
+            device,
+            device_type: device_type.into(),
+            backend,
+        }
+    }
+}
+
+/// Supported physical device types
+#[derive(Clone, Debug, PartialEq)]
+#[cfg_attr(feature = "serde", derive(Serialize, Deserialize))]
+pub enum DeviceType {
+    /// Other
+    Other,
+    /// Integrated
+    IntegratedGpu,
+    /// Discrete
+    DiscreteGpu,
+    /// Virtual / Hosted
+    VirtualGpu,
+    /// Cpu / Software Rendering
+    Cpu,
+}
+
+impl From<HalDeviceType> for DeviceType {
+    fn from(device_type: HalDeviceType) -> Self {
+        match device_type {
+            HalDeviceType::Other => Self::Other,
+            HalDeviceType::IntegratedGpu => Self::IntegratedGpu,
+            HalDeviceType::DiscreteGpu => Self::DiscreteGpu,
+            HalDeviceType::VirtualGpu => Self::VirtualGpu,
+            HalDeviceType::Cpu => Self::Cpu,
+        }
+    }
+}
+
 pub enum AdapterInputs<'a, I> {
    IdSet(&'a [I], fn(&I) -> Backend),
    Mask(BackendBit, fn() -> I),
@ -194,6 +261,76 @@ impl<I: Clone> AdapterInputs<'_, I> {
 }

 impl<F: IdentityFilter<AdapterId>> Global<F> {
+    pub fn enumerate_adapters(&self, inputs: AdapterInputs<F::Input>) -> Vec<AdapterId> {
+        let instance = &self.instance;
+        let mut token = Token::root();
+        let mut adapters = Vec::new();
+
+        #[cfg(any(
+            not(any(target_os = "ios", target_os = "macos")),
+            feature = "gfx-backend-vulkan"
+        ))]
+        {
+            if let Some(ref inst) = instance.vulkan {
+                if let Some(id_vulkan) = inputs.find(Backend::Vulkan) {
+                    for raw in inst.enumerate_adapters() {
+                        let adapter = Adapter { raw };
+                        log::info!("Adapter Vulkan {:?}", adapter.raw.info);
+                        adapters.push(backend::Vulkan::hub(self).adapters.register_identity(
+                            id_vulkan.clone(),
+                            adapter,
+                            &mut token,
+                        ));
+                    }
+                }
+            }
+        }
+        #[cfg(any(target_os = "ios", target_os = "macos"))]
+        {
+            if let Some(id_metal) = inputs.find(Backend::Metal) {
+                for raw in instance.metal.enumerate_adapters() {
+                    let adapter = Adapter { raw };
+                    log::info!("Adapter Metal {:?}", adapter.raw.info);
+                    adapters.push(backend::Metal::hub(self).adapters.register_identity(
+                        id_metal.clone(),
+                        adapter,
+                        &mut token,
+                    ));
+                }
+            }
+        }
+        #[cfg(windows)]
+        {
+            if let Some(ref inst) = instance.dx12 {
+                if let Some(id_dx12) = inputs.find(Backend::Dx12) {
+                    for raw in inst.enumerate_adapters() {
+                        let adapter = Adapter { raw };
+                        log::info!("Adapter Dx12 {:?}", adapter.raw.info);
+                        adapters.push(backend::Dx12::hub(self).adapters.register_identity(
+                            id_dx12.clone(),
+                            adapter,
+                            &mut token,
+                        ));
+                    }
+                }
+            }
+
+            if let Some(id_dx11) = inputs.find(Backend::Dx11) {
+                for raw in instance.dx11.enumerate_adapters() {
+                    let adapter = Adapter { raw };
+                    log::info!("Adapter Dx11 {:?}", adapter.raw.info);
+                    adapters.push(backend::Dx11::hub(self).adapters.register_identity(
+                        id_dx11.clone(),
+                        adapter,
+                        &mut token,
+                    ));
+                }
+            }
+        }
+
+        adapters
+    }
+
    pub fn pick_adapter(
        &self,
        desc: &RequestAdapterOptions,
@ -361,7 +498,7 @@ impl<F: IdentityFilter<AdapterId>> Global<F> {
        let mut token = Token::root();
        let (adapter_guard, _) = hub.adapters.read(&mut token);
        let adapter = &adapter_guard[adapter_id];
-        adapter.raw.info.clone()
+        AdapterInfo::from_gfx(adapter.raw.info.clone(), adapter_id.backend())
    }

    pub fn adapter_destroy<B: GfxBackend>(&self, adapter_id: AdapterId) {
--- a/gfx/wgpu/wgpu-core/src/lib.rs
+++ b/gfx/wgpu/wgpu-core/src/lib.rs
@ -30,6 +30,9 @@ pub mod resource;
 pub mod swap_chain;
 pub mod track;

+#[cfg(feature = "serde")]
+use serde::{Deserialize, Serialize};
+
 pub use hal::pso::read_spirv;
 use peek_poke::{PeekCopy, Poke};

@ -45,6 +48,7 @@ type Epoch = u32;

 #[repr(u8)]
 #[derive(Clone, Copy, Debug, PartialEq)]
+#[cfg_attr(feature = "serde", derive(Serialize, Deserialize))]
 pub enum Backend {
    Empty = 0,
    Vulkan = 1,
--- a/gfx/wgpu/wgpu-native/src/command.rs
+++ b/gfx/wgpu/wgpu-native/src/command.rs
@ -5,7 +5,6 @@
 use crate::GLOBAL;

 pub use core::command::{
-    wgpu_command_encoder_begin_render_pass,
    compute_ffi::*,
    render_ffi::*,
 };
@ -82,6 +81,20 @@ pub extern "C" fn wgpu_command_encoder_copy_texture_to_texture(
 }


+/// # 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_command_encoder_begin_render_pass(
+    encoder_id: id::CommandEncoderId,
+    desc: &core::command::RenderPassDescriptor,
+) -> *mut core::command::RawPass {
+    let pass = core::command::RawPass::new_render(encoder_id, desc);
+    Box::into_raw(Box::new(pass))
+}
+
 /// # Safety
 ///
 /// This function is unsafe because improper use may lead to memory
@ -89,33 +102,13 @@ pub extern "C" fn wgpu_command_encoder_copy_texture_to_texture(
 /// twice on the same raw pointer.
 #[no_mangle]
 pub unsafe extern "C" fn wgpu_render_pass_end_pass(pass_id: id::RenderPassId) {
-    let (pass_data, encoder_id, targets) = Box::from_raw(pass_id).finish_render();
-    let color_attachments: arrayvec::ArrayVec<[_; core::device::MAX_COLOR_TARGETS]> = targets.colors
-        .iter()
-        .flat_map(|at| {
-            if at.attachment == id::TextureViewId::ERROR {
-                None
-            } else {
-                Some(core::command::RenderPassColorAttachmentDescriptor {
-                    attachment: at.attachment,
-                    resolve_target: if at.resolve_target == id::TextureViewId::ERROR {
-                        None
-                    } else {
-                        Some(&at.resolve_target)
-                    },
-                    load_op: at.load_op,
-                    store_op: at.store_op,
-                    clear_color: at.clear_color,
-                })
-            }
-        })
-        .collect();
-    let depth_stencil_attachment = if targets.depth_stencil.attachment == id::TextureViewId::ERROR {
-        None
-    } else {
-        Some(&targets.depth_stencil)
-    };
-    gfx_select!(encoder_id => GLOBAL.command_encoder_run_render_pass(encoder_id, &color_attachments, depth_stencil_attachment, &pass_data))
+    let (pass_data, encoder_id) = Box::from_raw(pass_id).finish_render();
+    gfx_select!(encoder_id => GLOBAL.command_encoder_run_render_pass(encoder_id, &pass_data))
+}
+
+#[no_mangle]
+pub unsafe extern "C" fn wgpu_render_pass_destroy(pass: *mut core::command::RawPass) {
+    let _ = Box::from_raw(pass).into_vec();
 }

 /// # Safety
--- a/gfx/wgpu/wgpu-remote/src/lib.rs
+++ b/gfx/wgpu/wgpu-remote/src/lib.rs
@ -9,7 +9,6 @@ use core::{
 };

 pub use core::command::{
-    wgpu_command_encoder_begin_render_pass,
    compute_ffi::*,
    render_ffi::*,
 };
@ -32,6 +31,7 @@ struct IdentityHub {
    bind_groups: IdentityManager,
    shader_modules: IdentityManager,
    compute_pipelines: IdentityManager,
+    render_pipelines: IdentityManager,
 }

 #[derive(Debug, Default)]
@ -219,6 +219,19 @@ pub unsafe extern "C" fn wgpu_compute_pass_destroy(pass: core::command::RawPass)
    let _ = pass.into_vec();
 }

+#[no_mangle]
+pub unsafe extern "C" fn wgpu_command_encoder_begin_render_pass(
+    encoder_id: id::CommandEncoderId,
+    desc: &core::command::RenderPassDescriptor,
+) -> core::command::RawPass {
+    core::command::RawPass::new_render(encoder_id, desc)
+}
+
+#[no_mangle]
+pub unsafe extern "C" fn wgpu_render_pass_destroy(pass: core::command::RawPass) {
+    let _ = pass.into_vec();
+}
+
 #[no_mangle]
 pub extern "C" fn wgpu_client_make_bind_group_layout_id(
    client: &Client,
@ -353,3 +366,16 @@ pub extern "C" fn wgpu_client_kill_compute_pipeline_id(
        .compute_pipelines
        .free(id)
 }
+
+#[no_mangle]
+pub extern "C" fn wgpu_client_kill_render_pipeline_id(
+    client: &Client,
+    id: id::RenderPipelineId,
+) {
+    client
+        .identities
+        .lock()
+        .select(id.backend())
+        .render_pipelines
+        .free(id)
+}
--- a/gfx/wgpu/wgpu-remote/src/server.rs
+++ b/gfx/wgpu/wgpu-remote/src/server.rs
@ -190,6 +190,11 @@ pub unsafe extern "C" fn wgpu_server_encoder_copy_buffer_to_buffer(
    gfx_select!(self_id => global.command_encoder_copy_buffer_to_buffer(self_id, source_id, source_offset, destination_id, destination_offset, size));
 }

+/// # Safety
+///
+/// This function is unsafe as there is no guarantee that the given pointers are
+/// valid for `color_attachments_length` and `command_length` elements,
+/// respectively.
 #[no_mangle]
 pub unsafe extern "C" fn wgpu_server_encode_compute_pass(
    global: &Global,
@ -210,15 +215,11 @@ pub unsafe extern "C" fn wgpu_server_encode_compute_pass(
 pub unsafe extern "C" fn wgpu_server_encode_render_pass(
    global: &Global,
    self_id: id::CommandEncoderId,
-    color_attachments: *const core::command::RenderPassColorAttachmentDescriptor,
-    color_attachment_length: usize,
-    depth_stencil_attachment: Option<&core::command::RenderPassDepthStencilAttachmentDescriptor>,
    commands: *const u8,
    command_length: usize,
 ) {
-    let color_attachments = slice::from_raw_parts(color_attachments, color_attachment_length);
    let raw_pass = slice::from_raw_parts(commands, command_length);
-    gfx_select!(self_id => global.command_encoder_run_render_pass(self_id, color_attachments, depth_stencil_attachment, raw_pass));
+    gfx_select!(self_id => global.command_encoder_run_render_pass(self_id, raw_pass));
 }

 /// # Safety
@ -325,3 +326,11 @@ pub extern "C" fn wgpu_server_compute_pipeline_destroy(
 ) {
    gfx_select!(self_id => global.compute_pipeline_destroy(self_id));
 }
+
+#[no_mangle]
+pub extern "C" fn wgpu_server_render_pipeline_destroy(
+    global: &Global,
+    self_id: id::RenderPipelineId,
+) {
+    gfx_select!(self_id => global.render_pipeline_destroy(self_id));
+}