gpu: Flush worker context every frame.

Adds support for flushing pending ordering barriers on a stream. This is
used to flush the worker context stream when operating in async mode.

R=piman
BUG=514813

Cq-Include-Trybots: master.tryserver.blink:linux_trusty_blink_rel;master.tryserver.chromium.android:android_optional_gpu_tests_rel;master.tryserver.chromium.linux:linux_optional_gpu_tests_rel;master.tryserver.chromium.mac:mac_optional_gpu_tests_rel;master.tryserver.chromium.win:win_optional_gpu_tests_rel
Change-Id: I2d21a9adf5d00f5417016b40dfaebbe0cc013fca
Reviewed-on: https://chromium-review.googlesource.com/505254
Reviewed-by: Antoine Labour <piman@chromium.org>
Commit-Queue: Sunny Sachanandani <sunnyps@chromium.org>
Cr-Commit-Position: refs/heads/master@{#472344}

cc/raster/bitmap_raster_buffer_provider.cc

@@ -98,6 +98,8 @@ void BitmapRasterBufferProvider::OrderingBarrier() {
   // No need to sync resources as this provider does not use GL context.
 }
 
+void BitmapRasterBufferProvider::Flush() {}
+
 ResourceFormat BitmapRasterBufferProvider::GetResourceFormat(
     bool must_support_alpha) const {
   return resource_provider_->best_texture_format();

cc/raster/bitmap_raster_buffer_provider.h

@@ -34,6 +34,7 @@ class CC_EXPORT BitmapRasterBufferProvider : public RasterBufferProvider {
       uint64_t previous_content_id) override;
   void ReleaseBufferForRaster(std::unique_ptr<RasterBuffer> buffer) override;
   void OrderingBarrier() override;
+  void Flush() override;
   ResourceFormat GetResourceFormat(bool must_support_alpha) const override;
   bool IsResourceSwizzleRequired(bool must_support_alpha) const override;
   bool CanPartialRasterIntoProvidedResource() const override;

cc/raster/gpu_raster_buffer_provider.cc

@@ -175,6 +175,16 @@ void GpuRasterBufferProvider::OrderingBarrier() {
   pending_raster_buffers_.clear();
 }
 
+void GpuRasterBufferProvider::Flush() {
+  if (async_worker_context_enabled_) {
+    int32_t worker_stream_id =
+        worker_context_provider_->ContextSupport()->GetStreamId();
+
+    compositor_context_provider_->ContextSupport()
+        ->FlushOrderingBarrierOnStream(worker_stream_id);
+  }
+}
+
 ResourceFormat GpuRasterBufferProvider::GetResourceFormat(
     bool must_support_alpha) const {
   if (resource_provider_->IsRenderBufferFormatSupported(

cc/raster/gpu_raster_buffer_provider.h

@@ -33,6 +33,7 @@ class CC_EXPORT GpuRasterBufferProvider : public RasterBufferProvider {
       uint64_t previous_content_id) override;
   void ReleaseBufferForRaster(std::unique_ptr<RasterBuffer> buffer) override;
   void OrderingBarrier() override;
+  void Flush() override;
   ResourceFormat GetResourceFormat(bool must_support_alpha) const override;
   bool IsResourceSwizzleRequired(bool must_support_alpha) const override;
   bool CanPartialRasterIntoProvidedResource() const override;

cc/raster/one_copy_raster_buffer_provider.cc

@@ -140,6 +140,16 @@ void OneCopyRasterBufferProvider::OrderingBarrier() {
   pending_raster_buffers_.clear();
 }
 
+void OneCopyRasterBufferProvider::Flush() {
+  if (async_worker_context_enabled_) {
+    int32_t worker_stream_id =
+        worker_context_provider_->ContextSupport()->GetStreamId();
+
+    compositor_context_provider_->ContextSupport()
+        ->FlushOrderingBarrierOnStream(worker_stream_id);
+  }
+}
+
 ResourceFormat OneCopyRasterBufferProvider::GetResourceFormat(
     bool must_support_alpha) const {
   if (resource_provider_->IsTextureFormatSupported(preferred_tile_format_) &&

cc/raster/one_copy_raster_buffer_provider.h

@@ -38,6 +38,7 @@ class CC_EXPORT OneCopyRasterBufferProvider : public RasterBufferProvider {
       uint64_t previous_content_id) override;
   void ReleaseBufferForRaster(std::unique_ptr<RasterBuffer> buffer) override;
   void OrderingBarrier() override;
+  void Flush() override;
   ResourceFormat GetResourceFormat(bool must_support_alpha) const override;
   bool IsResourceSwizzleRequired(bool must_support_alpha) const override;
   bool CanPartialRasterIntoProvidedResource() const override;

cc/raster/raster_buffer_provider.h

@@ -54,6 +54,10 @@ class CC_EXPORT RasterBufferProvider {
   // Used for syncing resources to the worker context.
   virtual void OrderingBarrier() = 0;
 
+  // In addition to above, also ensures that pending work is sent to the GPU
+  // process.
+  virtual void Flush() = 0;
+
   // Returns the format to use for the tiles.
   virtual ResourceFormat GetResourceFormat(bool must_support_alpha) const = 0;
 

cc/raster/zero_copy_raster_buffer_provider.cc

@@ -100,6 +100,8 @@ void ZeroCopyRasterBufferProvider::OrderingBarrier() {
   // No need to sync resources as this provider does not use GL context.
 }
 
+void ZeroCopyRasterBufferProvider::Flush() {}
+
 ResourceFormat ZeroCopyRasterBufferProvider::GetResourceFormat(
     bool must_support_alpha) const {
   if (resource_provider_->IsTextureFormatSupported(preferred_tile_format_) &&

cc/raster/zero_copy_raster_buffer_provider.h

@@ -36,6 +36,7 @@ class CC_EXPORT ZeroCopyRasterBufferProvider : public RasterBufferProvider {
       uint64_t previous_content_id) override;
   void ReleaseBufferForRaster(std::unique_ptr<RasterBuffer> buffer) override;
   void OrderingBarrier() override;
+  void Flush() override;
   ResourceFormat GetResourceFormat(bool must_support_alpha) const override;
   bool IsResourceSwizzleRequired(bool must_support_alpha) const override;
   bool CanPartialRasterIntoProvidedResource() const override;

cc/test/fake_raster_buffer_provider.cc

@@ -23,6 +23,8 @@ void FakeRasterBufferProviderImpl::ReleaseBufferForRaster(
 
 void FakeRasterBufferProviderImpl::OrderingBarrier() {}
 
+void FakeRasterBufferProviderImpl::Flush() {}
+
 ResourceFormat FakeRasterBufferProviderImpl::GetResourceFormat(
     bool must_support_alpha) const {
   return ResourceFormat::RGBA_8888;

cc/test/fake_raster_buffer_provider.h

@@ -22,6 +22,7 @@ class FakeRasterBufferProviderImpl : public RasterBufferProvider {
       uint64_t previous_content_id) override;
   void ReleaseBufferForRaster(std::unique_ptr<RasterBuffer> buffer) override;
   void OrderingBarrier() override;
+  void Flush() override;
   ResourceFormat GetResourceFormat(bool must_support_alpha) const override;
   bool IsResourceSwizzleRequired(bool must_support_alpha) const override;
   bool CanPartialRasterIntoProvidedResource() const override;

cc/test/test_context_support.cc

@@ -19,6 +19,12 @@ TestContextSupport::TestContextSupport()
 
 TestContextSupport::~TestContextSupport() {}
 
+int32_t TestContextSupport::GetStreamId() const {
+  return 0;
+}
+
+void TestContextSupport::FlushOrderingBarrierOnStream(int32_t stream_id) {}
+
 void TestContextSupport::SignalSyncToken(const gpu::SyncToken& sync_token,
                                          const base::Closure& callback) {
   sync_point_callbacks_.push_back(callback);

cc/test/test_context_support.h

@@ -27,6 +27,8 @@ class TestContextSupport : public gpu::ContextSupport {
   ~TestContextSupport() override;
 
   // gpu::ContextSupport implementation.
+  int32_t GetStreamId() const override;
+  void FlushOrderingBarrierOnStream(int32_t stream_id) override;
   void SignalSyncToken(const gpu::SyncToken& sync_token,
                        const base::Closure& callback) override;
   bool IsSyncTokenSignaled(const gpu::SyncToken& sync_token) override;

cc/tiles/tile_manager.cc

@@ -526,6 +526,10 @@ void TileManager::Flush() {
 
   tile_task_manager_->CheckForCompletedTasks();
   did_check_for_completed_tasks_since_last_schedule_tasks_ = true;
+
+  // Actually flush.
+  raster_buffer_provider_->Flush();
+
   CheckPendingGpuWorkTiles(true /* issue_signals */);
 
   TRACE_EVENT_INSTANT1("cc", "DidFlush", TRACE_EVENT_SCOPE_THREAD, "stats",
@@ -1431,6 +1435,11 @@ bool TileManager::UsePartialRaster() const {
 }
 
 void TileManager::CheckPendingGpuWorkTiles(bool issue_signals) {
+  TRACE_EVENT2("cc", "TileManager::CheckPendingGpuWorkTiles",
+               "pending_gpu_work_tiles", pending_gpu_work_tiles_.size(),
+               "tree_priority",
+               TreePriorityToString(global_state_.tree_priority));
+
   ResourceProvider::ResourceIdArray required_for_activation_ids;
   ResourceProvider::ResourceIdArray required_for_draw_ids;
 

gpu/command_buffer/client/client_test_helper.h

@@ -112,7 +112,8 @@ class MockClientGpuControl : public GpuControl {
   MOCK_METHOD0(EnsureWorkVisible, void());
   MOCK_CONST_METHOD0(GetNamespaceID, CommandBufferNamespace());
   MOCK_CONST_METHOD0(GetCommandBufferID, CommandBufferId());
-  MOCK_CONST_METHOD0(GetExtraCommandBufferData, int32_t());
+  MOCK_CONST_METHOD0(GetStreamId, int32_t());
+  MOCK_METHOD1(FlushOrderingBarrierOnStream, void(int32_t));
   MOCK_METHOD0(GenerateFenceSyncRelease, uint64_t());
   MOCK_METHOD1(IsFenceSyncRelease, bool(uint64_t release));
   MOCK_METHOD1(IsFenceSyncFlushed, bool(uint64_t release));

gpu/command_buffer/client/context_support.h

@@ -26,6 +26,12 @@ struct SyncToken;
 
 class ContextSupport {
  public:
+  // Returns the stream id for this context.
+  virtual int32_t GetStreamId() const = 0;
+
+  // Flush any outstanding ordering barriers on given stream.
+  virtual void FlushOrderingBarrierOnStream(int32_t stream_id) = 0;
+
   // Runs |callback| when the given sync token is signalled. The sync token may
   // belong to any context.
   virtual void SignalSyncToken(const SyncToken& sync_token,

gpu/command_buffer/client/gles2_implementation.cc

@@ -374,6 +374,14 @@ void GLES2Implementation::RunIfContextNotLost(const base::Closure& callback) {
     callback.Run();
 }
 
+int32_t GLES2Implementation::GetStreamId() const {
+  return gpu_control_->GetStreamId();
+}
+
+void GLES2Implementation::FlushOrderingBarrierOnStream(int32_t stream_id) {
+  gpu_control_->FlushOrderingBarrierOnStream(stream_id);
+}
+
 void GLES2Implementation::SignalSyncToken(const gpu::SyncToken& sync_token,
                                           const base::Closure& callback) {
   SyncToken verified_sync_token;
@@ -6121,7 +6129,7 @@ void GLES2Implementation::GenSyncTokenCHROMIUM(GLuint64 fence_sync,
 
   // Copy the data over after setting the data to ensure alignment.
   SyncToken sync_token_data(gpu_control_->GetNamespaceID(),
-                            gpu_control_->GetExtraCommandBufferData(),
+                            gpu_control_->GetStreamId(),
                             gpu_control_->GetCommandBufferID(), fence_sync);
   sync_token_data.SetVerifyFlush();
   memcpy(sync_token, &sync_token_data, sizeof(sync_token_data));
@@ -6145,7 +6153,7 @@ void GLES2Implementation::GenUnverifiedSyncTokenCHROMIUM(GLuint64 fence_sync,
 
   // Copy the data over after setting the data to ensure alignment.
   SyncToken sync_token_data(gpu_control_->GetNamespaceID(),
-                            gpu_control_->GetExtraCommandBufferData(),
+                            gpu_control_->GetStreamId(),
                             gpu_control_->GetCommandBufferID(), fence_sync);
   memcpy(sync_token, &sync_token_data, sizeof(sync_token_data));
 }

gpu/command_buffer/client/gles2_implementation.h

@@ -183,12 +183,22 @@ class GLES2_IMPL_EXPORT GLES2Implementation
   // Gets client side generated errors.
   GLenum GetClientSideGLError();
 
+  // GLES2Interface implementation
+  void FreeSharedMemory(void*) override;
+
   // Include the auto-generated part of this class. We split this because
   // it means we can easily edit the non-auto generated parts right here in
   // this file instead of having to edit some template or the code generator.
   #include "gpu/command_buffer/client/gles2_implementation_autogen.h"
 
   // ContextSupport implementation.
+  int32_t GetStreamId() const override;
+  void FlushOrderingBarrierOnStream(int32_t stream_id) override;
+  void SignalSyncToken(const gpu::SyncToken& sync_token,
+                       const base::Closure& callback) override;
+  bool IsSyncTokenSignaled(const gpu::SyncToken& sync_token) override;
+  void SignalQuery(uint32_t query, const base::Closure& callback) override;
+  void SetAggressivelyFreeResources(bool aggressively_free_resources) override;
   void Swap() override;
   void SwapWithBounds(const std::vector<gfx::Rect>& rects) override;
   void PartialSwapBuffers(const gfx::Rect& sub_buffer) override;
@@ -249,15 +259,6 @@ class GLES2_IMPL_EXPORT GLES2Implementation
   void FreeUnusedSharedMemory();
   void FreeEverything();
 
-  void FreeSharedMemory(void*) override;
-
-  // ContextSupport implementation.
-  void SignalSyncToken(const gpu::SyncToken& sync_token,
-                       const base::Closure& callback) override;
-  bool IsSyncTokenSignaled(const gpu::SyncToken& sync_token) override;
-  void SignalQuery(uint32_t query, const base::Closure& callback) override;
-  void SetAggressivelyFreeResources(bool aggressively_free_resources) override;
-
   // Helper to set verified bit on sync token if allowed by gpu control.
   bool GetVerifiedSyncTokenForIPC(const gpu::SyncToken& sync_token,
                                   gpu::SyncToken* verified_sync_token);

gpu/command_buffer/client/gles2_implementation_unittest.cc

@@ -3976,8 +3976,7 @@ TEST_F(GLES2ImplementationTest, GenSyncTokenCHROMIUM) {
       .WillRepeatedly(Return(kNamespaceId));
   EXPECT_CALL(*gpu_control_, GetCommandBufferID())
       .WillRepeatedly(Return(kCommandBufferId));
-  EXPECT_CALL(*gpu_control_, GetExtraCommandBufferData())
-      .WillRepeatedly(Return(0));
+  EXPECT_CALL(*gpu_control_, GetStreamId()).WillRepeatedly(Return(0));
 
   gl_->GenSyncTokenCHROMIUM(kFenceSync, nullptr);
   EXPECT_EQ(GL_INVALID_VALUE, CheckError());
@@ -4022,8 +4021,7 @@ TEST_F(GLES2ImplementationTest, GenUnverifiedSyncTokenCHROMIUM) {
       .WillRepeatedly(Return(kNamespaceId));
   EXPECT_CALL(*gpu_control_, GetCommandBufferID())
       .WillRepeatedly(Return(kCommandBufferId));
-  EXPECT_CALL(*gpu_control_, GetExtraCommandBufferData())
-      .WillRepeatedly(Return(0));
+  EXPECT_CALL(*gpu_control_, GetStreamId()).WillRepeatedly(Return(0));
 
   gl_->GenUnverifiedSyncTokenCHROMIUM(kFenceSync, nullptr);
   EXPECT_EQ(GL_INVALID_VALUE, CheckError());
@@ -4075,8 +4073,7 @@ TEST_F(GLES2ImplementationTest, VerifySyncTokensCHROMIUM) {
       .WillRepeatedly(Return(kNamespaceId));
   EXPECT_CALL(*gpu_control_, GetCommandBufferID())
       .WillRepeatedly(Return(kCommandBufferId));
-  EXPECT_CALL(*gpu_control_, GetExtraCommandBufferData())
-      .WillRepeatedly(Return(0));
+  EXPECT_CALL(*gpu_control_, GetStreamId()).WillRepeatedly(Return(0));
 
   EXPECT_CALL(*gpu_control_, IsFenceSyncRelease(kFenceSync))
       .WillOnce(Return(true));
@@ -4131,8 +4128,7 @@ TEST_F(GLES2ImplementationTest, VerifySyncTokensCHROMIUM_Sequence) {
       .WillRepeatedly(Return(kNamespaceId));
   EXPECT_CALL(*gpu_control_, GetCommandBufferID())
       .WillRepeatedly(Return(kCommandBufferId));
-  EXPECT_CALL(*gpu_control_, GetExtraCommandBufferData())
-      .WillRepeatedly(Return(0));
+  EXPECT_CALL(*gpu_control_, GetStreamId()).WillRepeatedly(Return(0));
 
   // Generate sync token 1.
   EXPECT_CALL(*gpu_control_, IsFenceSyncRelease(kFenceSync1))
@@ -4185,7 +4181,7 @@ TEST_F(GLES2ImplementationTest, WaitSyncTokenCHROMIUM) {
   EXPECT_CALL(*gpu_control_, GetNamespaceID()).WillOnce(Return(kNamespaceId));
   EXPECT_CALL(*gpu_control_, GetCommandBufferID())
       .WillOnce(Return(kCommandBufferId));
-  EXPECT_CALL(*gpu_control_, GetExtraCommandBufferData()).WillOnce(Return(0));
+  EXPECT_CALL(*gpu_control_, GetStreamId()).WillOnce(Return(0));
   gl_->GenSyncTokenCHROMIUM(kFenceSync, sync_token_data);
 
   struct Cmds {
@@ -4523,7 +4519,7 @@ TEST_F(GLES2ImplementationTest, SignalSyncToken) {
   EXPECT_CALL(*gpu_control_, IsFenceSyncFlushReceived(fence_sync))
       .WillOnce(Return(true));
   EXPECT_CALL(*gpu_control_, GetNamespaceID()).WillOnce(Return(GPU_IO));
-  EXPECT_CALL(*gpu_control_, GetExtraCommandBufferData()).WillOnce(Return(0));
+  EXPECT_CALL(*gpu_control_, GetStreamId()).WillOnce(Return(0));
   EXPECT_CALL(*gpu_control_, GetCommandBufferID())
       .WillOnce(Return(CommandBufferId::FromUnsafeValue(1)));
   gpu::SyncToken sync_token;
@@ -4555,7 +4551,7 @@ TEST_F(GLES2ImplementationTest, SignalSyncTokenAfterContextLoss) {
   EXPECT_CALL(*gpu_control_, IsFenceSyncFlushReceived(fence_sync))
       .WillOnce(Return(true));
   EXPECT_CALL(*gpu_control_, GetNamespaceID()).WillOnce(Return(GPU_IO));
-  EXPECT_CALL(*gpu_control_, GetExtraCommandBufferData()).WillOnce(Return(0));
+  EXPECT_CALL(*gpu_control_, GetStreamId()).WillOnce(Return(0));
   EXPECT_CALL(*gpu_control_, GetCommandBufferID())
       .WillOnce(Return(CommandBufferId::FromUnsafeValue(1)));
   gpu::SyncToken sync_token;

gpu/command_buffer/client/gpu_control.h

@@ -77,7 +77,13 @@ class GPU_EXPORT GpuControl {
   // the CanWaitUnverifiedSyncToken() function.
   virtual CommandBufferNamespace GetNamespaceID() const = 0;
   virtual CommandBufferId GetCommandBufferID() const = 0;
-  virtual int32_t GetExtraCommandBufferData() const = 0;
+
+  // Returns the stream id for this context. Only relevant for IPC command
+  // buffer proxy. Used as extra command buffer data in sync tokens.
+  virtual int32_t GetStreamId() const = 0;
+
+  // Flush any outstanding ordering barriers on given stream.
+  virtual void FlushOrderingBarrierOnStream(int32_t stream_id) = 0;
 
   // Generates a fence sync which should be inserted into the GL command stream.
   // When the service executes the fence sync it is released. Fence syncs are

gpu/command_buffer/tests/gl_manager.cc

@@ -584,10 +584,14 @@ CommandBufferId GLManager::GetCommandBufferID() const {
   return command_buffer_id_;
 }
 
-int32_t GLManager::GetExtraCommandBufferData() const {
+int32_t GLManager::GetStreamId() const {
   return 0;
 }
 
+void GLManager::FlushOrderingBarrierOnStream(int32_t stream_id) {
+  // This is only relevant for out-of-process command buffers.
+}
+
 uint64_t GLManager::GenerateFenceSyncRelease() {
   return next_fence_sync_release_++;
 }

gpu/command_buffer/tests/gl_manager.h

@@ -138,7 +138,8 @@ class GLManager : private GpuControl {
   void EnsureWorkVisible() override;
   gpu::CommandBufferNamespace GetNamespaceID() const override;
   CommandBufferId GetCommandBufferID() const override;
-  int32_t GetExtraCommandBufferData() const override;
+  int32_t GetStreamId() const override;
+  void FlushOrderingBarrierOnStream(int32_t stream_id) override;
   uint64_t GenerateFenceSyncRelease() override;
   bool IsFenceSyncRelease(uint64_t release) override;
   bool IsFenceSyncFlushed(uint64_t release) override;

gpu/gles2_conform_support/egl/context.cc

@@ -194,10 +194,14 @@ gpu::CommandBufferId Context::GetCommandBufferID() const {
   return gpu::CommandBufferId();
 }
 
-int32_t Context::GetExtraCommandBufferData() const {
+int32_t Context::GetStreamId() const {
   return 0;
 }
 
+void Context::FlushOrderingBarrierOnStream(int32_t stream_id) {
+  // This is only relevant for out-of-process command buffers.
+}
+
 uint64_t Context::GenerateFenceSyncRelease() {
   return display_->GenerateFenceSyncRelease();
 }

gpu/gles2_conform_support/egl/context.h

@@ -68,7 +68,8 @@ class Context : public base::RefCountedThreadSafe<Context>,
   void EnsureWorkVisible() override;
   gpu::CommandBufferNamespace GetNamespaceID() const override;
   gpu::CommandBufferId GetCommandBufferID() const override;
-  int32_t GetExtraCommandBufferData() const override;
+  int32_t GetStreamId() const override;
+  void FlushOrderingBarrierOnStream(int32_t stream_id) override;
   uint64_t GenerateFenceSyncRelease() override;
   bool IsFenceSyncRelease(uint64_t release) override;
   bool IsFenceSyncFlushed(uint64_t release) override;