forked from mirrors/gecko-dev
This is a bigger patch but it was hard to split. Some of the changes look non-trivial; the guiding principle in these changes was "look at what the code was like before document splitting landed". Differential Revision: https://phabricator.services.mozilla.com/D73879
685 lines
24 KiB
C++
685 lines
24 KiB
C++
/* -*- Mode: C++; tab-width: 8; indent-tabs-mode: nil; c-basic-offset: 2 -*- */
|
|
/* vim: set ts=8 sts=2 et sw=2 tw=80: */
|
|
/* 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 "AsyncImagePipelineManager.h"
|
|
|
|
#include <algorithm>
|
|
#include <iterator>
|
|
|
|
#include "CompositableHost.h"
|
|
#include "gfxEnv.h"
|
|
#include "mozilla/gfx/gfxVars.h"
|
|
#include "mozilla/layers/CompositorThread.h"
|
|
#include "mozilla/layers/SharedSurfacesParent.h"
|
|
#include "mozilla/layers/WebRenderImageHost.h"
|
|
#include "mozilla/layers/WebRenderTextureHost.h"
|
|
#include "mozilla/webrender/RenderThread.h"
|
|
#include "mozilla/webrender/WebRenderAPI.h"
|
|
#include "mozilla/webrender/WebRenderTypes.h"
|
|
|
|
namespace mozilla {
|
|
namespace layers {
|
|
|
|
AsyncImagePipelineManager::ForwardingExternalImage::~ForwardingExternalImage() {
|
|
DebugOnly<bool> released = SharedSurfacesParent::Release(mImageId);
|
|
MOZ_ASSERT(released);
|
|
}
|
|
|
|
AsyncImagePipelineManager::AsyncImagePipeline::AsyncImagePipeline()
|
|
: mInitialised(false),
|
|
mRenderRoot(wr::RenderRoot::Default),
|
|
mIsChanged(false),
|
|
mUseExternalImage(false),
|
|
mFilter(wr::ImageRendering::Auto),
|
|
mMixBlendMode(wr::MixBlendMode::Normal) {}
|
|
|
|
AsyncImagePipelineManager::AsyncImagePipelineManager(
|
|
RefPtr<wr::WebRenderAPI>&& aApi, bool aUseCompositorWnd)
|
|
: mApi(aApi),
|
|
mUseCompositorWnd(aUseCompositorWnd),
|
|
mIdNamespace(mApi->GetNamespace()),
|
|
mUseTripleBuffering(mApi->GetUseTripleBuffering()),
|
|
mResourceId(0),
|
|
mAsyncImageEpoch{0},
|
|
mWillGenerateFrame{},
|
|
mDestroyed(false),
|
|
mRenderSubmittedUpdatesLock("SubmittedUpdatesLock"),
|
|
mLastCompletedFrameId(0) {
|
|
MOZ_COUNT_CTOR(AsyncImagePipelineManager);
|
|
}
|
|
|
|
AsyncImagePipelineManager::~AsyncImagePipelineManager() {
|
|
MOZ_COUNT_DTOR(AsyncImagePipelineManager);
|
|
}
|
|
|
|
void AsyncImagePipelineManager::Destroy() {
|
|
MOZ_ASSERT(!mDestroyed);
|
|
mApi = nullptr;
|
|
mPipelineTexturesHolders.Clear();
|
|
mDestroyed = true;
|
|
}
|
|
|
|
void AsyncImagePipelineManager::SetWillGenerateFrameAllRenderRoots() {
|
|
for (auto renderRoot : wr::kRenderRoots) {
|
|
SetWillGenerateFrame(renderRoot);
|
|
}
|
|
}
|
|
|
|
void AsyncImagePipelineManager::SetWillGenerateFrame(
|
|
wr::RenderRoot aRenderRoot) {
|
|
MOZ_ASSERT(CompositorThreadHolder::IsInCompositorThread());
|
|
|
|
mWillGenerateFrame[aRenderRoot] = true;
|
|
}
|
|
|
|
bool AsyncImagePipelineManager::GetAndResetWillGenerateFrame(
|
|
wr::RenderRoot aRenderRoot) {
|
|
MOZ_ASSERT(CompositorThreadHolder::IsInCompositorThread());
|
|
|
|
bool ret = mWillGenerateFrame[aRenderRoot];
|
|
mWillGenerateFrame[aRenderRoot] = false;
|
|
return ret;
|
|
}
|
|
|
|
void AsyncImagePipelineManager::AddPipeline(const wr::PipelineId& aPipelineId,
|
|
WebRenderBridgeParent* aWrBridge) {
|
|
if (mDestroyed) {
|
|
return;
|
|
}
|
|
uint64_t id = wr::AsUint64(aPipelineId);
|
|
|
|
PipelineTexturesHolder* holder =
|
|
mPipelineTexturesHolders.Get(wr::AsUint64(aPipelineId));
|
|
if (holder) {
|
|
// This could happen during tab move between different windows.
|
|
// Previously removed holder could be still alive for waiting destroyed.
|
|
MOZ_ASSERT(holder->mDestroyedEpoch.isSome());
|
|
holder->mDestroyedEpoch = Nothing(); // Revive holder
|
|
holder->mWrBridge = aWrBridge;
|
|
return;
|
|
}
|
|
holder = new PipelineTexturesHolder();
|
|
holder->mWrBridge = aWrBridge;
|
|
mPipelineTexturesHolders.Put(id, holder);
|
|
}
|
|
|
|
void AsyncImagePipelineManager::RemovePipeline(
|
|
const wr::PipelineId& aPipelineId, const wr::Epoch& aEpoch) {
|
|
if (mDestroyed) {
|
|
return;
|
|
}
|
|
|
|
PipelineTexturesHolder* holder =
|
|
mPipelineTexturesHolders.Get(wr::AsUint64(aPipelineId));
|
|
MOZ_ASSERT(holder);
|
|
if (!holder) {
|
|
return;
|
|
}
|
|
holder->mWrBridge = nullptr;
|
|
holder->mDestroyedEpoch = Some(aEpoch);
|
|
}
|
|
|
|
WebRenderBridgeParent* AsyncImagePipelineManager::GetWrBridge(
|
|
const wr::PipelineId& aPipelineId) {
|
|
if (mDestroyed) {
|
|
return nullptr;
|
|
}
|
|
|
|
PipelineTexturesHolder* holder =
|
|
mPipelineTexturesHolders.Get(wr::AsUint64(aPipelineId));
|
|
if (!holder) {
|
|
return nullptr;
|
|
}
|
|
if (holder->mWrBridge) {
|
|
MOZ_ASSERT(holder->mDestroyedEpoch.isNothing());
|
|
return holder->mWrBridge;
|
|
}
|
|
|
|
return nullptr;
|
|
}
|
|
|
|
void AsyncImagePipelineManager::AddAsyncImagePipeline(
|
|
const wr::PipelineId& aPipelineId, WebRenderImageHost* aImageHost,
|
|
wr::RenderRoot aRenderRoot) {
|
|
if (mDestroyed) {
|
|
return;
|
|
}
|
|
MOZ_ASSERT(aImageHost);
|
|
uint64_t id = wr::AsUint64(aPipelineId);
|
|
|
|
MOZ_ASSERT(!mAsyncImagePipelines.Get(id));
|
|
AsyncImagePipeline* holder = new AsyncImagePipeline();
|
|
holder->mRenderRoot = aRenderRoot;
|
|
holder->mImageHost = aImageHost;
|
|
mAsyncImagePipelines.Put(id, holder);
|
|
AddPipeline(aPipelineId, /* aWrBridge */ nullptr);
|
|
}
|
|
|
|
void AsyncImagePipelineManager::RemoveAsyncImagePipeline(
|
|
const wr::PipelineId& aPipelineId, wr::TransactionBuilder& aTxn) {
|
|
if (mDestroyed) {
|
|
return;
|
|
}
|
|
|
|
uint64_t id = wr::AsUint64(aPipelineId);
|
|
if (auto entry = mAsyncImagePipelines.Lookup(id)) {
|
|
const auto& holder = entry.Data();
|
|
wr::Epoch epoch = GetNextImageEpoch();
|
|
aTxn.ClearDisplayList(epoch, aPipelineId);
|
|
for (wr::ImageKey key : holder->mKeys) {
|
|
aTxn.DeleteImage(key);
|
|
}
|
|
entry.Remove();
|
|
RemovePipeline(aPipelineId, epoch);
|
|
}
|
|
}
|
|
|
|
void AsyncImagePipelineManager::UpdateAsyncImagePipeline(
|
|
const wr::PipelineId& aPipelineId, const LayoutDeviceRect& aScBounds,
|
|
const gfx::Matrix4x4& aScTransform, const gfx::MaybeIntSize& aScaleToSize,
|
|
const wr::ImageRendering& aFilter, const wr::MixBlendMode& aMixBlendMode) {
|
|
if (mDestroyed) {
|
|
return;
|
|
}
|
|
AsyncImagePipeline* pipeline =
|
|
mAsyncImagePipelines.Get(wr::AsUint64(aPipelineId));
|
|
if (!pipeline) {
|
|
return;
|
|
}
|
|
pipeline->mInitialised = true;
|
|
pipeline->Update(aScBounds, aScTransform, aScaleToSize, aFilter,
|
|
aMixBlendMode);
|
|
}
|
|
|
|
Maybe<TextureHost::ResourceUpdateOp> AsyncImagePipelineManager::UpdateImageKeys(
|
|
const wr::Epoch& aEpoch, const wr::PipelineId& aPipelineId,
|
|
AsyncImagePipeline* aPipeline, nsTArray<wr::ImageKey>& aKeys,
|
|
wr::TransactionBuilder& aSceneBuilderTxn,
|
|
wr::TransactionBuilder& aMaybeFastTxn) {
|
|
MOZ_ASSERT(aKeys.IsEmpty());
|
|
MOZ_ASSERT(aPipeline);
|
|
|
|
TextureHost* texture =
|
|
aPipeline->mImageHost->GetAsTextureHostForComposite(this);
|
|
TextureHost* previousTexture = aPipeline->mCurrentTexture.get();
|
|
|
|
if (texture == previousTexture) {
|
|
// The texture has not changed, just reuse previous ImageKeys.
|
|
aKeys = aPipeline->mKeys.Clone();
|
|
return Nothing();
|
|
}
|
|
|
|
if (!texture || texture->NumSubTextures() == 0) {
|
|
// We don't have a new texture or texture does not have SubTextures, there
|
|
// isn't much we can do.
|
|
aKeys = aPipeline->mKeys.Clone();
|
|
return Nothing();
|
|
}
|
|
|
|
aPipeline->mCurrentTexture = texture;
|
|
|
|
WebRenderTextureHost* wrTexture = texture->AsWebRenderTextureHost();
|
|
MOZ_ASSERT(wrTexture);
|
|
if (!wrTexture) {
|
|
gfxCriticalNote << "WebRenderTextureHost is not used";
|
|
}
|
|
|
|
bool useExternalImage = !!wrTexture;
|
|
aPipeline->mUseExternalImage = useExternalImage;
|
|
|
|
// The non-external image code path falls back to converting the texture into
|
|
// an rgb image.
|
|
auto numKeys = useExternalImage ? texture->NumSubTextures() : 1;
|
|
MOZ_ASSERT(numKeys > 0);
|
|
|
|
// If we already had a texture and the format hasn't changed, better to reuse
|
|
// the image keys than create new ones.
|
|
bool canUpdate = !!previousTexture &&
|
|
previousTexture->GetSize() == texture->GetSize() &&
|
|
previousTexture->GetFormat() == texture->GetFormat() &&
|
|
previousTexture->NeedsYFlip() == texture->NeedsYFlip() &&
|
|
aPipeline->mKeys.Length() == numKeys;
|
|
|
|
if (!canUpdate) {
|
|
for (auto key : aPipeline->mKeys) {
|
|
// Destroy ImageKeys on transaction of scene builder thread, since
|
|
// DisplayList is updated on SceneBuilder thread. It prevents too early
|
|
// ImageKey deletion.
|
|
aSceneBuilderTxn.DeleteImage(key);
|
|
}
|
|
aPipeline->mKeys.Clear();
|
|
for (uint32_t i = 0; i < numKeys; ++i) {
|
|
aPipeline->mKeys.AppendElement(GenerateImageKey());
|
|
}
|
|
}
|
|
|
|
aKeys = aPipeline->mKeys.Clone();
|
|
|
|
auto op = canUpdate ? TextureHost::UPDATE_IMAGE : TextureHost::ADD_IMAGE;
|
|
|
|
if (!useExternalImage) {
|
|
return UpdateWithoutExternalImage(texture, aKeys[0], op, aMaybeFastTxn);
|
|
}
|
|
|
|
Range<wr::ImageKey> keys(&aKeys[0], aKeys.Length());
|
|
auto externalImageKey = wrTexture->GetExternalImageKey();
|
|
wrTexture->PushResourceUpdates(aMaybeFastTxn, op, keys, externalImageKey);
|
|
|
|
return Some(op);
|
|
}
|
|
|
|
Maybe<TextureHost::ResourceUpdateOp>
|
|
AsyncImagePipelineManager::UpdateWithoutExternalImage(
|
|
TextureHost* aTexture, wr::ImageKey aKey, TextureHost::ResourceUpdateOp aOp,
|
|
wr::TransactionBuilder& aTxn) {
|
|
MOZ_ASSERT(aTexture);
|
|
|
|
RefPtr<gfx::DataSourceSurface> dSurf = aTexture->GetAsSurface();
|
|
if (!dSurf) {
|
|
NS_ERROR("TextureHost does not return DataSourceSurface");
|
|
return Nothing();
|
|
}
|
|
gfx::DataSourceSurface::MappedSurface map;
|
|
if (!dSurf->Map(gfx::DataSourceSurface::MapType::READ, &map)) {
|
|
NS_ERROR("DataSourceSurface failed to map");
|
|
return Nothing();
|
|
}
|
|
|
|
gfx::IntSize size = dSurf->GetSize();
|
|
wr::ImageDescriptor descriptor(size, map.mStride, dSurf->GetFormat());
|
|
|
|
// Costly copy right here...
|
|
wr::Vec<uint8_t> bytes;
|
|
bytes.PushBytes(Range<uint8_t>(map.mData, size.height * map.mStride));
|
|
|
|
if (aOp == TextureHost::UPDATE_IMAGE) {
|
|
aTxn.UpdateImageBuffer(aKey, descriptor, bytes);
|
|
} else {
|
|
aTxn.AddImage(aKey, descriptor, bytes);
|
|
}
|
|
|
|
dSurf->Unmap();
|
|
|
|
return Some(aOp);
|
|
}
|
|
|
|
void AsyncImagePipelineManager::ApplyAsyncImagesOfImageBridge(
|
|
wr::TransactionBuilder& aSceneBuilderTxn,
|
|
wr::TransactionBuilder& aFastTxn) {
|
|
if (mDestroyed || mAsyncImagePipelines.Count() == 0) {
|
|
return;
|
|
}
|
|
|
|
wr::Epoch epoch = GetNextImageEpoch();
|
|
|
|
// We use a pipeline with a very small display list for each video element.
|
|
// Update each of them if needed.
|
|
for (auto iter = mAsyncImagePipelines.Iter(); !iter.Done(); iter.Next()) {
|
|
wr::PipelineId pipelineId = wr::AsPipelineId(iter.Key());
|
|
AsyncImagePipeline* pipeline = iter.UserData();
|
|
// If aync image pipeline does not use ImageBridge, do not need to apply.
|
|
if (!pipeline->mImageHost->GetAsyncRef()) {
|
|
continue;
|
|
}
|
|
ApplyAsyncImageForPipeline(epoch, pipelineId, pipeline, aSceneBuilderTxn,
|
|
aFastTxn);
|
|
}
|
|
}
|
|
|
|
void AsyncImagePipelineManager::ApplyAsyncImageForPipeline(
|
|
const wr::Epoch& aEpoch, const wr::PipelineId& aPipelineId,
|
|
AsyncImagePipeline* aPipeline, wr::TransactionBuilder& aSceneBuilderTxn,
|
|
wr::TransactionBuilder& aMaybeFastTxn) {
|
|
nsTArray<wr::ImageKey> keys;
|
|
auto op = UpdateImageKeys(aEpoch, aPipelineId, aPipeline, keys,
|
|
aSceneBuilderTxn, aMaybeFastTxn);
|
|
|
|
bool updateDisplayList =
|
|
aPipeline->mInitialised &&
|
|
(aPipeline->mIsChanged || op == Some(TextureHost::ADD_IMAGE)) &&
|
|
!!aPipeline->mCurrentTexture;
|
|
|
|
if (!updateDisplayList) {
|
|
// We don't need to update the display list, either because we can't or
|
|
// because the previous one is still up to date. We may, however, have
|
|
// updated some resources.
|
|
|
|
// Use transaction of scene builder thread to notify epoch.
|
|
// It is for making epoch update consistent.
|
|
aSceneBuilderTxn.UpdateEpoch(aPipelineId, aEpoch);
|
|
if (aPipeline->mCurrentTexture) {
|
|
HoldExternalImage(aPipelineId, aEpoch, aPipeline->mCurrentTexture);
|
|
}
|
|
return;
|
|
}
|
|
|
|
aPipeline->mIsChanged = false;
|
|
|
|
gfx::Matrix4x4 scTransform = aPipeline->mScTransform;
|
|
if (aPipeline->mCurrentTexture && aPipeline->mCurrentTexture->NeedsYFlip()) {
|
|
scTransform.PreTranslate(0, aPipeline->mCurrentTexture->GetSize().height, 0)
|
|
.PreScale(1, -1, 1);
|
|
}
|
|
|
|
wr::LayoutSize contentSize{aPipeline->mScBounds.Width(),
|
|
aPipeline->mScBounds.Height()};
|
|
wr::DisplayListBuilder builder(aPipelineId, contentSize);
|
|
|
|
float opacity = 1.0f;
|
|
wr::StackingContextParams params;
|
|
params.opacity = &opacity;
|
|
params.mTransformPtr = scTransform.IsIdentity() ? nullptr : &scTransform;
|
|
params.mix_blend_mode = aPipeline->mMixBlendMode;
|
|
|
|
Maybe<wr::WrSpatialId> referenceFrameId = builder.PushStackingContext(
|
|
params, wr::ToLayoutRect(aPipeline->mScBounds),
|
|
// This is fine to do unconditionally because we only push images here.
|
|
wr::RasterSpace::Screen());
|
|
|
|
Maybe<wr::SpaceAndClipChainHelper> spaceAndClipChainHelper;
|
|
if (referenceFrameId) {
|
|
spaceAndClipChainHelper.emplace(builder, referenceFrameId.ref());
|
|
}
|
|
|
|
if (aPipeline->mCurrentTexture && !keys.IsEmpty()) {
|
|
LayoutDeviceRect rect(0, 0, aPipeline->mCurrentTexture->GetSize().width,
|
|
aPipeline->mCurrentTexture->GetSize().height);
|
|
if (aPipeline->mScaleToSize.isSome()) {
|
|
rect = LayoutDeviceRect(0, 0, aPipeline->mScaleToSize.value().width,
|
|
aPipeline->mScaleToSize.value().height);
|
|
}
|
|
|
|
if (aPipeline->mUseExternalImage) {
|
|
MOZ_ASSERT(aPipeline->mCurrentTexture->AsWebRenderTextureHost());
|
|
Range<wr::ImageKey> range_keys(&keys[0], keys.Length());
|
|
bool prefer_compositor_surface =
|
|
IsOpaque(aPipeline->mCurrentTexture->GetFormat()) ||
|
|
bool(aPipeline->mCurrentTexture->GetFlags() &
|
|
TextureFlags::IS_OPAQUE);
|
|
aPipeline->mCurrentTexture->PushDisplayItems(
|
|
builder, wr::ToLayoutRect(rect), wr::ToLayoutRect(rect),
|
|
aPipeline->mFilter, range_keys,
|
|
/* aPreferCompositorSurface */ prefer_compositor_surface);
|
|
HoldExternalImage(aPipelineId, aEpoch, aPipeline->mCurrentTexture);
|
|
} else {
|
|
MOZ_ASSERT(keys.Length() == 1);
|
|
builder.PushImage(wr::ToLayoutRect(rect), wr::ToLayoutRect(rect), true,
|
|
aPipeline->mFilter, keys[0]);
|
|
}
|
|
}
|
|
|
|
spaceAndClipChainHelper.reset();
|
|
builder.PopStackingContext(referenceFrameId.isSome());
|
|
|
|
wr::BuiltDisplayList dl;
|
|
wr::LayoutSize builderContentSize;
|
|
builder.Finalize(builderContentSize, dl);
|
|
aSceneBuilderTxn.SetDisplayList(gfx::DeviceColor(0.f, 0.f, 0.f, 0.f), aEpoch,
|
|
wr::ToLayoutSize(aPipeline->mScBounds.Size()),
|
|
aPipelineId, builderContentSize, dl.dl_desc,
|
|
dl.dl);
|
|
}
|
|
|
|
void AsyncImagePipelineManager::ApplyAsyncImageForPipeline(
|
|
const wr::PipelineId& aPipelineId, wr::TransactionBuilder& aTxn,
|
|
wr::TransactionBuilder& aTxnForImageBridge, wr::RenderRoot aRenderRoot) {
|
|
AsyncImagePipeline* pipeline =
|
|
mAsyncImagePipelines.Get(wr::AsUint64(aPipelineId));
|
|
if (!pipeline) {
|
|
return;
|
|
}
|
|
wr::TransactionBuilder fastTxn(/* aUseSceneBuilderThread */ false);
|
|
wr::AutoTransactionSender sender(mApi, &fastTxn);
|
|
|
|
// Transaction for async image pipeline that uses ImageBridge always need to
|
|
// be non low priority.
|
|
auto& sceneBuilderTxn =
|
|
pipeline->mImageHost->GetAsyncRef() ? aTxnForImageBridge : aTxn;
|
|
|
|
// Use transaction of using non scene builder thread when ImageHost uses
|
|
// ImageBridge. ApplyAsyncImagesOfImageBridge() handles transaction of adding
|
|
// and updating wr::ImageKeys of ImageHosts that uses ImageBridge. Then
|
|
// AsyncImagePipelineManager always needs to use non scene builder thread
|
|
// transaction for adding and updating wr::ImageKeys of ImageHosts that uses
|
|
// ImageBridge. Otherwise, ordering of wr::ImageKeys updating in webrender
|
|
// becomes inconsistent.
|
|
auto& maybeFastTxn = pipeline->mImageHost->GetAsyncRef() ? fastTxn : aTxn;
|
|
|
|
wr::Epoch epoch = GetNextImageEpoch();
|
|
|
|
ApplyAsyncImageForPipeline(epoch, aPipelineId, pipeline, sceneBuilderTxn,
|
|
maybeFastTxn);
|
|
}
|
|
|
|
void AsyncImagePipelineManager::SetEmptyDisplayList(
|
|
const wr::PipelineId& aPipelineId, wr::TransactionBuilder& aTxn,
|
|
wr::TransactionBuilder& aTxnForImageBridge) {
|
|
AsyncImagePipeline* pipeline =
|
|
mAsyncImagePipelines.Get(wr::AsUint64(aPipelineId));
|
|
if (!pipeline) {
|
|
return;
|
|
}
|
|
|
|
// Transaction for async image pipeline that uses ImageBridge always need to
|
|
// be non low priority.
|
|
auto& txn = pipeline->mImageHost->GetAsyncRef() ? aTxnForImageBridge : aTxn;
|
|
|
|
wr::Epoch epoch = GetNextImageEpoch();
|
|
wr::LayoutSize contentSize{pipeline->mScBounds.Width(),
|
|
pipeline->mScBounds.Height()};
|
|
wr::DisplayListBuilder builder(aPipelineId, contentSize);
|
|
|
|
wr::BuiltDisplayList dl;
|
|
wr::LayoutSize builderContentSize;
|
|
builder.Finalize(builderContentSize, dl);
|
|
txn.SetDisplayList(gfx::DeviceColor(0.f, 0.f, 0.f, 0.f), epoch,
|
|
wr::ToLayoutSize(pipeline->mScBounds.Size()), aPipelineId,
|
|
builderContentSize, dl.dl_desc, dl.dl);
|
|
}
|
|
|
|
void AsyncImagePipelineManager::HoldExternalImage(
|
|
const wr::PipelineId& aPipelineId, const wr::Epoch& aEpoch,
|
|
TextureHost* aTexture) {
|
|
if (mDestroyed) {
|
|
return;
|
|
}
|
|
MOZ_ASSERT(aTexture);
|
|
|
|
PipelineTexturesHolder* holder =
|
|
mPipelineTexturesHolders.Get(wr::AsUint64(aPipelineId));
|
|
MOZ_ASSERT(holder);
|
|
if (!holder) {
|
|
return;
|
|
}
|
|
if (aTexture->HasIntermediateBuffer()) {
|
|
// Hold WebRenderTextureHost until submitted for rendering if it has an
|
|
// intermediate buffer.
|
|
holder->mTextureHostsUntilRenderSubmitted.emplace_back(aEpoch, aTexture);
|
|
} else {
|
|
// Hold WebRenderTextureHost until rendering completed if not.
|
|
holder->mTextureHostsUntilRenderCompleted.emplace_back(
|
|
MakeUnique<ForwardingTextureHost>(aEpoch, aTexture));
|
|
}
|
|
}
|
|
|
|
void AsyncImagePipelineManager::HoldExternalImage(
|
|
const wr::PipelineId& aPipelineId, const wr::Epoch& aEpoch,
|
|
const wr::ExternalImageId& aImageId) {
|
|
if (mDestroyed) {
|
|
SharedSurfacesParent::Release(aImageId);
|
|
return;
|
|
}
|
|
|
|
PipelineTexturesHolder* holder =
|
|
mPipelineTexturesHolders.Get(wr::AsUint64(aPipelineId));
|
|
MOZ_ASSERT(holder);
|
|
if (!holder) {
|
|
SharedSurfacesParent::Release(aImageId);
|
|
return;
|
|
}
|
|
|
|
holder->mExternalImages.emplace_back(
|
|
MakeUnique<ForwardingExternalImage>(aEpoch, aImageId));
|
|
}
|
|
|
|
void AsyncImagePipelineManager::NotifyPipelinesUpdated(
|
|
RefPtr<const wr::WebRenderPipelineInfo> aInfo,
|
|
wr::RenderedFrameId aLatestFrameId,
|
|
wr::RenderedFrameId aLastCompletedFrameId) {
|
|
MOZ_ASSERT(wr::RenderThread::IsInRenderThread());
|
|
MOZ_ASSERT(mLastCompletedFrameId <= aLastCompletedFrameId.mId);
|
|
MOZ_ASSERT(aLatestFrameId.IsValid());
|
|
|
|
// This is called on the render thread, so we just stash the data into
|
|
// mPendingUpdates and process it later on the compositor thread.
|
|
mPendingUpdates.push_back(std::move(aInfo));
|
|
mLastCompletedFrameId = aLastCompletedFrameId.mId;
|
|
|
|
{
|
|
// We need to lock for mRenderSubmittedUpdates because it can be accessed
|
|
// on the compositor thread.
|
|
MutexAutoLock lock(mRenderSubmittedUpdatesLock);
|
|
|
|
// Move the pending updates into the submitted ones. Note that this clears
|
|
// mPendingUpdates.
|
|
mRenderSubmittedUpdates.emplace_back(aLatestFrameId,
|
|
std::move(mPendingUpdates));
|
|
}
|
|
|
|
// Queue a runnable on the compositor thread to process the updates.
|
|
// This will also call CheckForTextureHostsNotUsedByGPU.
|
|
layers::CompositorThreadHolder::Loop()->PostTask(
|
|
NewRunnableMethod("ProcessPipelineUpdates", this,
|
|
&AsyncImagePipelineManager::ProcessPipelineUpdates));
|
|
}
|
|
|
|
void AsyncImagePipelineManager::ProcessPipelineUpdates() {
|
|
MOZ_ASSERT(CompositorThreadHolder::IsInCompositorThread());
|
|
|
|
if (mDestroyed) {
|
|
return;
|
|
}
|
|
|
|
std::vector<std::pair<wr::RenderedFrameId, PipelineInfoVector>>
|
|
submittedUpdates;
|
|
{
|
|
// We need to lock for mRenderSubmittedUpdates because it can be accessed on
|
|
// the compositor thread.
|
|
MutexAutoLock lock(mRenderSubmittedUpdatesLock);
|
|
mRenderSubmittedUpdates.swap(submittedUpdates);
|
|
}
|
|
|
|
// submittedUpdates is a vector of RenderedFrameIds paired with vectors of
|
|
// WebRenderPipelineInfo.
|
|
for (auto update : submittedUpdates) {
|
|
for (auto pipelineInfo : update.second) {
|
|
auto& info = pipelineInfo->Raw();
|
|
|
|
for (auto& epoch : info.epochs) {
|
|
ProcessPipelineRendered(epoch.pipeline_id, epoch.epoch, update.first);
|
|
}
|
|
for (auto& removedPipeline : info.removed_pipelines) {
|
|
ProcessPipelineRemoved(removedPipeline, update.first);
|
|
}
|
|
}
|
|
}
|
|
CheckForTextureHostsNotUsedByGPU();
|
|
}
|
|
|
|
void AsyncImagePipelineManager::ProcessPipelineRendered(
|
|
const wr::PipelineId& aPipelineId, const wr::Epoch& aEpoch,
|
|
wr::RenderedFrameId aRenderedFrameId) {
|
|
if (auto entry = mPipelineTexturesHolders.Lookup(wr::AsUint64(aPipelineId))) {
|
|
const auto& holder = entry.Data();
|
|
// For TextureHosts that can be released on render submission, using aEpoch
|
|
// find the first that we can't release and then release all prior to that.
|
|
auto firstSubmittedHostToKeep = std::find_if(
|
|
holder->mTextureHostsUntilRenderSubmitted.begin(),
|
|
holder->mTextureHostsUntilRenderSubmitted.end(),
|
|
[&aEpoch](const auto& entry) { return aEpoch <= entry.mEpoch; });
|
|
holder->mTextureHostsUntilRenderSubmitted.erase(
|
|
holder->mTextureHostsUntilRenderSubmitted.begin(),
|
|
firstSubmittedHostToKeep);
|
|
|
|
// For TextureHosts that need to wait until render completed, find the first
|
|
// that is later than aEpoch and then move all prior to that to
|
|
// mTexturesInUseByGPU paired with aRenderedFrameId. These will be released
|
|
// once rendering has completed for aRenderedFrameId.
|
|
auto firstCompletedHostToKeep = std::find_if(
|
|
holder->mTextureHostsUntilRenderCompleted.begin(),
|
|
holder->mTextureHostsUntilRenderCompleted.end(),
|
|
[&aEpoch](const auto& entry) { return aEpoch <= entry->mEpoch; });
|
|
if (firstCompletedHostToKeep !=
|
|
holder->mTextureHostsUntilRenderCompleted.begin()) {
|
|
std::vector<UniquePtr<ForwardingTextureHost>> hostsUntilCompleted(
|
|
std::make_move_iterator(
|
|
holder->mTextureHostsUntilRenderCompleted.begin()),
|
|
std::make_move_iterator(firstCompletedHostToKeep));
|
|
mTexturesInUseByGPU.emplace_back(aRenderedFrameId,
|
|
std::move(hostsUntilCompleted));
|
|
holder->mTextureHostsUntilRenderCompleted.erase(
|
|
holder->mTextureHostsUntilRenderCompleted.begin(),
|
|
firstCompletedHostToKeep);
|
|
}
|
|
|
|
// Using aEpoch, find the first external image that we can't release and
|
|
// then release all prior to that.
|
|
auto firstImageToKeep = std::find_if(
|
|
holder->mExternalImages.begin(), holder->mExternalImages.end(),
|
|
[&aEpoch](const auto& entry) { return aEpoch <= entry->mEpoch; });
|
|
holder->mExternalImages.erase(holder->mExternalImages.begin(),
|
|
firstImageToKeep);
|
|
}
|
|
}
|
|
|
|
void AsyncImagePipelineManager::ProcessPipelineRemoved(
|
|
const wr::RemovedPipeline& aRemovedPipeline,
|
|
wr::RenderedFrameId aRenderedFrameId) {
|
|
if (mDestroyed) {
|
|
return;
|
|
}
|
|
if (auto entry = mPipelineTexturesHolders.Lookup(
|
|
wr::AsUint64(aRemovedPipeline.pipeline_id))) {
|
|
const auto& holder = entry.Data();
|
|
if (holder->mDestroyedEpoch.isSome()) {
|
|
if (!holder->mTextureHostsUntilRenderCompleted.empty()) {
|
|
// Move all TextureHosts that must be held until render completed to
|
|
// mTexturesInUseByGPU paired with aRenderedFrameId.
|
|
mTexturesInUseByGPU.emplace_back(
|
|
aRenderedFrameId,
|
|
std::move(holder->mTextureHostsUntilRenderCompleted));
|
|
}
|
|
|
|
// Remove Pipeline releasing all remaining TextureHosts and external
|
|
// images.
|
|
entry.Remove();
|
|
}
|
|
|
|
// If mDestroyedEpoch contains nothing it means we reused the same pipeline
|
|
// id (probably because we moved the tab to another window). In this case we
|
|
// need to keep the holder.
|
|
}
|
|
}
|
|
|
|
void AsyncImagePipelineManager::CheckForTextureHostsNotUsedByGPU() {
|
|
uint64_t lastCompletedFrameId = mLastCompletedFrameId;
|
|
|
|
// Find first entry after mLastCompletedFrameId and release all prior ones.
|
|
auto firstTexturesToKeep =
|
|
std::find_if(mTexturesInUseByGPU.begin(), mTexturesInUseByGPU.end(),
|
|
[lastCompletedFrameId](const auto& entry) {
|
|
return lastCompletedFrameId < entry.first.mId;
|
|
});
|
|
mTexturesInUseByGPU.erase(mTexturesInUseByGPU.begin(), firstTexturesToKeep);
|
|
}
|
|
|
|
wr::Epoch AsyncImagePipelineManager::GetNextImageEpoch() {
|
|
mAsyncImageEpoch.mHandle++;
|
|
return mAsyncImageEpoch;
|
|
}
|
|
|
|
} // namespace layers
|
|
} // namespace mozilla
|