1 files changed, 475 insertions, 0 deletions
diff --git a/dom/webgpu/Queue.cpp b/dom/webgpu/Queue.cpp
new file mode 100644
index 0000000000..c675c42cae
--- /dev/null
+++ b/dom/webgpu/Queue.cpp
@@ -0,0 +1,475 @@
+/* -*- Mode: C++; tab-width: 4; indent-tabs-mode: nil; c-basic-offset: 2 -*- */
+/* This Source Code Form is subject to the terms of the Mozilla Public
+ * License, v. 2.0. If a copy of the MPL was not distributed with this
+ * file, You can obtain one at http://mozilla.org/MPL/2.0/. */
+
+#include "mozilla/dom/WebGPUBinding.h"
+#include "mozilla/dom/UnionTypes.h"
+#include "Queue.h"
+
+#include <algorithm>
+
+#include "CommandBuffer.h"
+#include "CommandEncoder.h"
+#include "ipc/WebGPUChild.h"
+#include "mozilla/Casting.h"
+#include "mozilla/ErrorResult.h"
+#include "mozilla/dom/HTMLCanvasElement.h"
+#include "mozilla/dom/ImageBitmap.h"
+#include "mozilla/dom/OffscreenCanvas.h"
+#include "mozilla/dom/WebGLTexelConversions.h"
+#include "mozilla/dom/WebGLTypes.h"
+#include "nsLayoutUtils.h"
+#include "Utility.h"
+
+namespace mozilla::webgpu {
+
+GPU_IMPL_CYCLE_COLLECTION(Queue, mParent, mBridge)
+GPU_IMPL_JS_WRAP(Queue)
+
+Queue::Queue(Device* const aParent, WebGPUChild* aBridge, RawId aId)
+    : ChildOf(aParent), mBridge(aBridge), mId(aId) {}
+
+Queue::~Queue() { Cleanup(); }
+
+void Queue::Submit(
+    const dom::Sequence<OwningNonNull<CommandBuffer>>& aCommandBuffers) {
+  nsTArray<RawId> list(aCommandBuffers.Length());
+  for (uint32_t i = 0; i < aCommandBuffers.Length(); ++i) {
+    auto idMaybe = aCommandBuffers[i]->Commit();
+    if (idMaybe) {
+      list.AppendElement(*idMaybe);
+    }
+  }
+
+  mBridge->SendQueueSubmit(mId, mParent->mId, list);
+}
+
+// Get the base address and length of part of a `BufferSource`.
+//
+// Given `aBufferSource` and an offset `aDataOffset` and optional length
+// `aSizeOrRemainder` describing the range of its contents we want to see, check
+// all arguments and set `aDataContents` and `aContentsSize` to a pointer to the
+// bytes and a length. Report errors in `aRv`.
+//
+// If `ASizeOrRemainder` was not passed, return a view from the starting offset
+// to the end of `aBufferSource`.
+//
+// On success, the returned `aDataContents` is never `nullptr`. If the
+// `ArrayBuffer` is detached, return a pointer to a dummy buffer and set
+// `aContentsSize` to zero.
+//
+// The `aBufferSource` argument is a WebIDL `BufferSource`, which WebGPU methods
+// use anywhere they accept a block of raw bytes. WebIDL defines `BufferSource`
+// as:
+//
+//     typedef (ArrayBufferView or ArrayBuffer) BufferSource;
+//
+// This appears in Gecko code as `dom::ArrayBufferViewOrArrayBuffer`.
+static void GetBufferSourceDataAndSize(
+    const dom::ArrayBufferViewOrArrayBuffer& aBufferSource,
+    uint64_t aDataOffset, const dom::Optional<uint64_t>& aSizeOrRemainder,
+    uint8_t*& aDataContents, uint64_t& aContentsSize, const char* aOffsetName,
+    ErrorResult& aRv) {
+  uint64_t dataSize = 0;
+  uint8_t* dataContents = nullptr;
+  if (aBufferSource.IsArrayBufferView()) {
+    const auto& view = aBufferSource.GetAsArrayBufferView();
+    view.ComputeState();
+    dataSize = view.Length();
+    dataContents = view.Data();
+  }
+  if (aBufferSource.IsArrayBuffer()) {
+    const auto& ab = aBufferSource.GetAsArrayBuffer();
+    ab.ComputeState();
+    dataSize = ab.Length();
+    dataContents = ab.Data();
+  }
+
+  if (aDataOffset > dataSize) {
+    aRv.ThrowOperationError(
+        nsPrintfCString("%s is greater than data length", aOffsetName));
+    return;
+  }
+
+  uint64_t contentsSize = 0;
+  if (aSizeOrRemainder.WasPassed()) {
+    contentsSize = aSizeOrRemainder.Value();
+  } else {
+    // We already know that aDataOffset <= length, so this cannot underflow.
+    contentsSize = dataSize - aDataOffset;
+  }
+
+  // This could be folded into the if above, but it's nice to make it
+  // obvious that the check always occurs.
+  // We already know that aDataOffset <= length, so this cannot underflow.
+  if (contentsSize > dataSize - aDataOffset) {
+    aRv.ThrowOperationError(
+        nsPrintfCString("%s + size is greater than data length", aOffsetName));
+    return;
+  }
+
+  if (!dataContents) {
+    // Passing `nullptr` as either the source or destination to
+    // `memcpy` is undefined behavior, even when the count is zero:
+    //
+    // https://en.cppreference.com/w/cpp/string/byte/memcpy
+    //
+    // We can either make callers responsible for checking the pointer
+    // before calling `memcpy`, or we can have it point to a
+    // permanently-live `static` dummy byte, so that the copies are
+    // harmless. The latter seems less error-prone.
+    static uint8_t dummy;
+    dataContents = &dummy;
+    MOZ_RELEASE_ASSERT(contentsSize == 0);
+  }
+  aDataContents = dataContents;
+  aContentsSize = contentsSize;
+}
+
+void Queue::WriteBuffer(const Buffer& aBuffer, uint64_t aBufferOffset,
+                        const dom::ArrayBufferViewOrArrayBuffer& aData,
+                        uint64_t aDataOffset,
+                        const dom::Optional<uint64_t>& aSize,
+                        ErrorResult& aRv) {
+  uint8_t* dataContents = nullptr;
+  uint64_t contentsSize = 0;
+  GetBufferSourceDataAndSize(aData, aDataOffset, aSize, dataContents,
+                             contentsSize, "dataOffset", aRv);
+  if (aRv.Failed()) {
+    return;
+  }
+
+  if (contentsSize % 4 != 0) {
+    aRv.ThrowAbortError("Byte size must be a multiple of 4");
+    return;
+  }
+
+  auto alloc =
+      mozilla::ipc::UnsafeSharedMemoryHandle::CreateAndMap(contentsSize);
+  if (alloc.isNothing()) {
+    aRv.Throw(NS_ERROR_OUT_OF_MEMORY);
+    return;
+  }
+
+  auto handle = std::move(alloc.ref().first);
+  auto mapping = std::move(alloc.ref().second);
+
+  memcpy(mapping.Bytes().data(), dataContents + aDataOffset, contentsSize);
+  ipc::ByteBuf bb;
+  ffi::wgpu_queue_write_buffer(aBuffer.mId, aBufferOffset, ToFFI(&bb));
+  if (!mBridge->SendQueueWriteAction(mId, mParent->mId, std::move(bb),
+                                     std::move(handle))) {
+    MOZ_CRASH("IPC failure");
+  }
+}
+
+void Queue::WriteTexture(const dom::GPUImageCopyTexture& aDestination,
+                         const dom::ArrayBufferViewOrArrayBuffer& aData,
+                         const dom::GPUImageDataLayout& aDataLayout,
+                         const dom::GPUExtent3D& aSize, ErrorResult& aRv) {
+  ffi::WGPUImageCopyTexture copyView = {};
+  CommandEncoder::ConvertTextureCopyViewToFFI(aDestination, &copyView);
+  ffi::WGPUImageDataLayout dataLayout = {};
+  CommandEncoder::ConvertTextureDataLayoutToFFI(aDataLayout, &dataLayout);
+  dataLayout.offset = 0;  // our Shmem has the contents starting from 0.
+  ffi::WGPUExtent3d extent = {};
+  ConvertExtent3DToFFI(aSize, &extent);
+
+  uint8_t* dataContents = nullptr;
+  uint64_t contentsSize = 0;
+  GetBufferSourceDataAndSize(aData, aDataLayout.mOffset,
+                             dom::Optional<uint64_t>(), dataContents,
+                             contentsSize, "dataLayout.offset", aRv);
+  if (aRv.Failed()) {
+    return;
+  }
+
+  if (!contentsSize) {
+    aRv.ThrowAbortError("Input size cannot be zero.");
+    return;
+  }
+  MOZ_ASSERT(dataContents != nullptr);
+
+  auto alloc =
+      mozilla::ipc::UnsafeSharedMemoryHandle::CreateAndMap(contentsSize);
+  if (alloc.isNothing()) {
+    aRv.Throw(NS_ERROR_OUT_OF_MEMORY);
+    return;
+  }
+
+  auto handle = std::move(alloc.ref().first);
+  auto mapping = std::move(alloc.ref().second);
+
+  memcpy(mapping.Bytes().data(), dataContents + aDataLayout.mOffset,
+         contentsSize);
+
+  ipc::ByteBuf bb;
+  ffi::wgpu_queue_write_texture(copyView, dataLayout, extent, ToFFI(&bb));
+  if (!mBridge->SendQueueWriteAction(mId, mParent->mId, std::move(bb),
+                                     std::move(handle))) {
+    MOZ_CRASH("IPC failure");
+  }
+}
+
+static WebGLTexelFormat ToWebGLTexelFormat(gfx::SurfaceFormat aFormat) {
+  switch (aFormat) {
+    case gfx::SurfaceFormat::B8G8R8A8:
+    case gfx::SurfaceFormat::B8G8R8X8:
+      return WebGLTexelFormat::BGRA8;
+    case gfx::SurfaceFormat::R8G8B8A8:
+    case gfx::SurfaceFormat::R8G8B8X8:
+      return WebGLTexelFormat::RGBA8;
+    default:
+      return WebGLTexelFormat::FormatNotSupportingAnyConversion;
+  }
+}
+
+static WebGLTexelFormat ToWebGLTexelFormat(dom::GPUTextureFormat aFormat) {
+  // TODO: We need support for Rbg10a2unorm as well.
+  switch (aFormat) {
+    case dom::GPUTextureFormat::R8unorm:
+      return WebGLTexelFormat::R8;
+    case dom::GPUTextureFormat::R16float:
+      return WebGLTexelFormat::R16F;
+    case dom::GPUTextureFormat::R32float:
+      return WebGLTexelFormat::R32F;
+    case dom::GPUTextureFormat::Rg8unorm:
+      return WebGLTexelFormat::RG8;
+    case dom::GPUTextureFormat::Rg16float:
+      return WebGLTexelFormat::RG16F;
+    case dom::GPUTextureFormat::Rg32float:
+      return WebGLTexelFormat::RG32F;
+    case dom::GPUTextureFormat::Rgba8unorm:
+    case dom::GPUTextureFormat::Rgba8unorm_srgb:
+      return WebGLTexelFormat::RGBA8;
+    case dom::GPUTextureFormat::Bgra8unorm:
+    case dom::GPUTextureFormat::Bgra8unorm_srgb:
+      return WebGLTexelFormat::BGRA8;
+    case dom::GPUTextureFormat::Rgba16float:
+      return WebGLTexelFormat::RGBA16F;
+    case dom::GPUTextureFormat::Rgba32float:
+      return WebGLTexelFormat::RGBA32F;
+    default:
+      return WebGLTexelFormat::FormatNotSupportingAnyConversion;
+  }
+}
+
+void Queue::CopyExternalImageToTexture(
+    const dom::GPUImageCopyExternalImage& aSource,
+    const dom::GPUImageCopyTextureTagged& aDestination,
+    const dom::GPUExtent3D& aCopySize, ErrorResult& aRv) {
+  const auto dstFormat = ToWebGLTexelFormat(aDestination.mTexture->Format());
+  if (dstFormat == WebGLTexelFormat::FormatNotSupportingAnyConversion) {
+    aRv.ThrowInvalidStateError("Unsupported destination format");
+    return;
+  }
+
+  const uint32_t surfaceFlags = nsLayoutUtils::SFE_ALLOW_NON_PREMULT;
+  SurfaceFromElementResult sfeResult;
+  switch (aSource.mSource.GetType()) {
+    case decltype(aSource.mSource)::Type::eImageBitmap: {
+      const auto& bitmap = aSource.mSource.GetAsImageBitmap();
+      if (bitmap->IsClosed()) {
+        aRv.ThrowInvalidStateError("Detached ImageBitmap");
+        return;
+      }
+
+      sfeResult = nsLayoutUtils::SurfaceFromImageBitmap(bitmap, surfaceFlags);
+      break;
+    }
+    case decltype(aSource.mSource)::Type::eHTMLCanvasElement: {
+      MOZ_ASSERT(NS_IsMainThread());
+
+      const auto& canvas = aSource.mSource.GetAsHTMLCanvasElement();
+      if (canvas->Width() == 0 || canvas->Height() == 0) {
+        aRv.ThrowInvalidStateError("Zero-sized HTMLCanvasElement");
+        return;
+      }
+
+      sfeResult = nsLayoutUtils::SurfaceFromElement(canvas, surfaceFlags);
+      break;
+    }
+    case decltype(aSource.mSource)::Type::eOffscreenCanvas: {
+      const auto& canvas = aSource.mSource.GetAsOffscreenCanvas();
+      if (canvas->Width() == 0 || canvas->Height() == 0) {
+        aRv.ThrowInvalidStateError("Zero-sized OffscreenCanvas");
+        return;
+      }
+
+      sfeResult =
+          nsLayoutUtils::SurfaceFromOffscreenCanvas(canvas, surfaceFlags);
+      break;
+    }
+  }
+
+  if (!sfeResult.mCORSUsed) {
+    nsIGlobalObject* global = mParent->GetOwnerGlobal();
+    nsIPrincipal* dstPrincipal = global ? global->PrincipalOrNull() : nullptr;
+    if (!sfeResult.mPrincipal || !dstPrincipal ||
+        !dstPrincipal->Subsumes(sfeResult.mPrincipal)) {
+      aRv.ThrowSecurityError("Cross-origin elements require CORS!");
+      return;
+    }
+  }
+
+  if (sfeResult.mIsWriteOnly) {
+    aRv.ThrowSecurityError("Write only source data not supported!");
+    return;
+  }
+
+  RefPtr<gfx::SourceSurface> surface = sfeResult.GetSourceSurface();
+  if (!surface) {
+    aRv.ThrowInvalidStateError("No surface available from source");
+    return;
+  }
+
+  RefPtr<gfx::DataSourceSurface> dataSurface = surface->GetDataSurface();
+  if (!dataSurface) {
+    aRv.Throw(NS_ERROR_OUT_OF_MEMORY);
+    return;
+  }
+
+  bool srcPremultiplied;
+  switch (sfeResult.mAlphaType) {
+    case gfxAlphaType::Premult:
+      srcPremultiplied = true;
+      break;
+    case gfxAlphaType::NonPremult:
+      srcPremultiplied = false;
+      break;
+    case gfxAlphaType::Opaque:
+      // No (un)premultiplication necessary so match the output.
+      srcPremultiplied = aDestination.mPremultipliedAlpha;
+      break;
+  }
+
+  const auto surfaceFormat = dataSurface->GetFormat();
+  const auto srcFormat = ToWebGLTexelFormat(surfaceFormat);
+  if (srcFormat == WebGLTexelFormat::FormatNotSupportingAnyConversion) {
+    gfxCriticalError() << "Unsupported surface format from source "
+                       << surfaceFormat;
+    MOZ_CRASH();
+  }
+
+  gfx::DataSourceSurface::ScopedMap map(dataSurface,
+                                        gfx::DataSourceSurface::READ);
+  if (!map.IsMapped()) {
+    aRv.ThrowInvalidStateError("Cannot map surface from source");
+    return;
+  }
+
+  if (!aSource.mOrigin.IsGPUOrigin2DDict()) {
+    aRv.ThrowInvalidStateError("Cannot get origin from source");
+    return;
+  }
+
+  ffi::WGPUExtent3d extent = {};
+  ConvertExtent3DToFFI(aCopySize, &extent);
+  if (extent.depth_or_array_layers > 1) {
+    aRv.ThrowOperationError("Depth is greater than 1");
+    return;
+  }
+
+  uint32_t srcOriginX;
+  uint32_t srcOriginY;
+  if (aSource.mOrigin.IsRangeEnforcedUnsignedLongSequence()) {
+    const auto& seq = aSource.mOrigin.GetAsRangeEnforcedUnsignedLongSequence();
+    srcOriginX = seq.Length() > 0 ? seq[0] : 0;
+    srcOriginY = seq.Length() > 1 ? seq[1] : 0;
+  } else if (aSource.mOrigin.IsGPUOrigin2DDict()) {
+    const auto& dict = aSource.mOrigin.GetAsGPUOrigin2DDict();
+    srcOriginX = dict.mX;
+    srcOriginY = dict.mY;
+  } else {
+    MOZ_CRASH("Unexpected origin type!");
+  }
+
+  const auto checkedMaxWidth = CheckedInt<uint32_t>(srcOriginX) + extent.width;
+  const auto checkedMaxHeight =
+      CheckedInt<uint32_t>(srcOriginY) + extent.height;
+  if (!checkedMaxWidth.isValid() || !checkedMaxHeight.isValid()) {
+    aRv.ThrowOperationError("Offset and copy size exceed integer bounds");
+    return;
+  }
+
+  const gfx::IntSize surfaceSize = dataSurface->GetSize();
+  const auto surfaceWidth = AssertedCast<uint32_t>(surfaceSize.width);
+  const auto surfaceHeight = AssertedCast<uint32_t>(surfaceSize.height);
+  if (surfaceWidth < checkedMaxWidth.value() ||
+      surfaceHeight < checkedMaxHeight.value()) {
+    aRv.ThrowOperationError("Offset and copy size exceed surface bounds");
+    return;
+  }
+
+  const auto dstWidth = extent.width;
+  const auto dstHeight = extent.height;
+  if (dstWidth == 0 || dstHeight == 0) {
+    aRv.ThrowOperationError("Destination size is empty");
+    return;
+  }
+
+  if (!aDestination.mTexture->mBytesPerBlock) {
+    // TODO(bug 1781071) This should emmit a GPUValidationError on the device
+    // timeline.
+    aRv.ThrowInvalidStateError("Invalid destination format");
+    return;
+  }
+
+  // Note: This assumes bytes per block == bytes per pixel which is the case
+  // here because the spec only allows non-compressed texture formats for the
+  // destination.
+  const auto dstStride = CheckedInt<uint32_t>(extent.width) *
+                         aDestination.mTexture->mBytesPerBlock.value();
+  const auto dstByteLength = dstStride * extent.height;
+  if (!dstStride.isValid() || !dstByteLength.isValid()) {
+    aRv.Throw(NS_ERROR_OUT_OF_MEMORY);
+    return;
+  }
+
+  auto alloc = mozilla::ipc::UnsafeSharedMemoryHandle::CreateAndMap(
+      dstByteLength.value());
+  if (alloc.isNothing()) {
+    aRv.Throw(NS_ERROR_OUT_OF_MEMORY);
+    return;
+  }
+
+  auto handle = std::move(alloc.ref().first);
+  auto mapping = std::move(alloc.ref().second);
+
+  const int32_t pixelSize = gfx::BytesPerPixel(surfaceFormat);
+  auto* dstBegin = mapping.Bytes().data();
+  const auto* srcBegin =
+      map.GetData() + srcOriginX * pixelSize + srcOriginY * map.GetStride();
+  const auto srcOriginPos = gl::OriginPos::TopLeft;
+  const auto srcStride = AssertedCast<uint32_t>(map.GetStride());
+  const auto dstOriginPos =
+      aSource.mFlipY ? gl::OriginPos::BottomLeft : gl::OriginPos::TopLeft;
+  bool wasTrivial;
+
+  auto dstStrideVal = dstStride.value();
+
+  if (!ConvertImage(dstWidth, dstHeight, srcBegin, srcStride, srcOriginPos,
+                    srcFormat, srcPremultiplied, dstBegin, dstStrideVal,
+                    dstOriginPos, dstFormat, aDestination.mPremultipliedAlpha,
+                    &wasTrivial)) {
+    MOZ_ASSERT_UNREACHABLE("ConvertImage failed!");
+    aRv.ThrowInvalidStateError(
+        nsPrintfCString("Failed to convert source to destination format "
+                        "(%i/%i), please file a bug!",
+                        (int)srcFormat, (int)dstFormat));
+    return;
+  }
+
+  ffi::WGPUImageDataLayout dataLayout = {0, &dstStrideVal, &dstHeight};
+  ffi::WGPUImageCopyTexture copyView = {};
+  CommandEncoder::ConvertTextureCopyViewToFFI(aDestination, &copyView);
+  ipc::ByteBuf bb;
+  ffi::wgpu_queue_write_texture(copyView, dataLayout, extent, ToFFI(&bb));
+  if (!mBridge->SendQueueWriteAction(mId, mParent->mId, std::move(bb),
+                                     std::move(handle))) {
+    MOZ_CRASH("IPC failure");
+  }
+}
+
+}  // namespace mozilla::webgpu