blob: 2f5b0a9d7917ea2fd00472cea2992a0c268b67a2 [file] [log] [blame]
// Copyright 2012 The Chromium Authors. All rights reserved.
// Use of this source code is governed by a BSD-style license that can be
// found in the LICENSE file.
#include "cc/resources/resource_provider.h"
#include <algorithm>
#include <limits>
#include "base/atomic_sequence_num.h"
#include "base/containers/hash_tables.h"
#include "base/metrics/histogram.h"
#include "base/numerics/safe_math.h"
#include "base/stl_util.h"
#include "base/strings/string_split.h"
#include "base/strings/string_util.h"
#include "base/strings/stringprintf.h"
#include "base/thread_task_runner_handle.h"
#include "base/trace_event/memory_dump_manager.h"
#include "base/trace_event/trace_event.h"
#include "cc/resources/platform_color.h"
#include "cc/resources/resource_util.h"
#include "cc/resources/returned_resource.h"
#include "cc/resources/shared_bitmap_manager.h"
#include "cc/resources/transferable_resource.h"
#include "gpu/GLES2/gl2extchromium.h"
#include "gpu/command_buffer/client/context_support.h"
#include "gpu/command_buffer/client/gles2_interface.h"
#include "gpu/command_buffer/client/gpu_memory_buffer_manager.h"
#include "third_party/khronos/GLES2/gl2.h"
#include "third_party/khronos/GLES2/gl2ext.h"
#include "third_party/skia/include/core/SkSurface.h"
#include "third_party/skia/include/gpu/GrContext.h"
#include "third_party/skia/include/gpu/GrTextureProvider.h"
#include "ui/gfx/geometry/rect.h"
#include "ui/gfx/geometry/vector2d.h"
#include "ui/gl/trace_util.h"
using gpu::gles2::GLES2Interface;
namespace cc {
class IdAllocator {
public:
virtual ~IdAllocator() {}
virtual GLuint NextId() = 0;
protected:
IdAllocator(GLES2Interface* gl, size_t id_allocation_chunk_size)
: gl_(gl),
id_allocation_chunk_size_(id_allocation_chunk_size),
ids_(new GLuint[id_allocation_chunk_size]),
next_id_index_(id_allocation_chunk_size) {
DCHECK(id_allocation_chunk_size_);
DCHECK_LE(id_allocation_chunk_size_,
static_cast<size_t>(std::numeric_limits<int>::max()));
}
GLES2Interface* gl_;
const size_t id_allocation_chunk_size_;
scoped_ptr<GLuint[]> ids_;
size_t next_id_index_;
};
namespace {
GLenum TextureToStorageFormat(ResourceFormat format) {
GLenum storage_format = GL_RGBA8_OES;
switch (format) {
case RGBA_8888:
break;
case BGRA_8888:
storage_format = GL_BGRA8_EXT;
break;
case RGBA_4444:
case ALPHA_8:
case LUMINANCE_8:
case RGB_565:
case ETC1:
case RED_8:
NOTREACHED();
break;
}
return storage_format;
}
bool IsFormatSupportedForStorage(ResourceFormat format, bool use_bgra) {
switch (format) {
case RGBA_8888:
return true;
case BGRA_8888:
return use_bgra;
case RGBA_4444:
case ALPHA_8:
case LUMINANCE_8:
case RGB_565:
case ETC1:
case RED_8:
return false;
}
return false;
}
GrPixelConfig ToGrPixelConfig(ResourceFormat format) {
switch (format) {
case RGBA_8888:
return kRGBA_8888_GrPixelConfig;
case BGRA_8888:
return kBGRA_8888_GrPixelConfig;
case RGBA_4444:
return kRGBA_4444_GrPixelConfig;
default:
break;
}
DCHECK(false) << "Unsupported resource format.";
return kSkia8888_GrPixelConfig;
}
class ScopedSetActiveTexture {
public:
ScopedSetActiveTexture(GLES2Interface* gl, GLenum unit)
: gl_(gl), unit_(unit) {
DCHECK_EQ(GL_TEXTURE0, ResourceProvider::GetActiveTextureUnit(gl_));
if (unit_ != GL_TEXTURE0)
gl_->ActiveTexture(unit_);
}
~ScopedSetActiveTexture() {
// Active unit being GL_TEXTURE0 is effectively the ground state.
if (unit_ != GL_TEXTURE0)
gl_->ActiveTexture(GL_TEXTURE0);
}
private:
GLES2Interface* gl_;
GLenum unit_;
};
class TextureIdAllocator : public IdAllocator {
public:
TextureIdAllocator(GLES2Interface* gl,
size_t texture_id_allocation_chunk_size)
: IdAllocator(gl, texture_id_allocation_chunk_size) {}
~TextureIdAllocator() override {
gl_->DeleteTextures(
static_cast<int>(id_allocation_chunk_size_ - next_id_index_),
ids_.get() + next_id_index_);
}
// Overridden from IdAllocator:
GLuint NextId() override {
if (next_id_index_ == id_allocation_chunk_size_) {
gl_->GenTextures(static_cast<int>(id_allocation_chunk_size_), ids_.get());
next_id_index_ = 0;
}
return ids_[next_id_index_++];
}
private:
DISALLOW_COPY_AND_ASSIGN(TextureIdAllocator);
};
class BufferIdAllocator : public IdAllocator {
public:
BufferIdAllocator(GLES2Interface* gl, size_t buffer_id_allocation_chunk_size)
: IdAllocator(gl, buffer_id_allocation_chunk_size) {}
~BufferIdAllocator() override {
gl_->DeleteBuffers(
static_cast<int>(id_allocation_chunk_size_ - next_id_index_),
ids_.get() + next_id_index_);
}
// Overridden from IdAllocator:
GLuint NextId() override {
if (next_id_index_ == id_allocation_chunk_size_) {
gl_->GenBuffers(static_cast<int>(id_allocation_chunk_size_), ids_.get());
next_id_index_ = 0;
}
return ids_[next_id_index_++];
}
private:
DISALLOW_COPY_AND_ASSIGN(BufferIdAllocator);
};
// Generates process-unique IDs to use for tracing a ResourceProvider's
// resources.
base::StaticAtomicSequenceNumber g_next_resource_provider_tracing_id;
} // namespace
ResourceProvider::Resource::~Resource() {}
ResourceProvider::Resource::Resource(GLuint texture_id,
const gfx::Size& size,
Origin origin,
GLenum target,
GLenum filter,
TextureHint hint,
ResourceFormat format)
: child_id(0),
gl_id(texture_id),
gl_pixel_buffer_id(0),
gl_upload_query_id(0),
gl_read_lock_query_id(0),
pixels(NULL),
lock_for_read_count(0),
imported_count(0),
exported_count(0),
dirty_image(false),
locked_for_write(false),
lost(false),
marked_for_deletion(false),
allocated(false),
read_lock_fences_enabled(false),
has_shared_bitmap_id(false),
is_overlay_candidate(false),
read_lock_fence(NULL),
size(size),
origin(origin),
target(target),
original_filter(filter),
filter(filter),
image_id(0),
bound_image_id(0),
hint(hint),
type(RESOURCE_TYPE_GL_TEXTURE),
format(format),
shared_bitmap(NULL),
gpu_memory_buffer(NULL) {
}
ResourceProvider::Resource::Resource(uint8_t* pixels,
SharedBitmap* bitmap,
const gfx::Size& size,
Origin origin,
GLenum filter)
: child_id(0),
gl_id(0),
gl_pixel_buffer_id(0),
gl_upload_query_id(0),
gl_read_lock_query_id(0),
pixels(pixels),
lock_for_read_count(0),
imported_count(0),
exported_count(0),
dirty_image(false),
locked_for_write(false),
lost(false),
marked_for_deletion(false),
allocated(false),
read_lock_fences_enabled(false),
has_shared_bitmap_id(!!bitmap),
is_overlay_candidate(false),
read_lock_fence(NULL),
size(size),
origin(origin),
target(0),
original_filter(filter),
filter(filter),
image_id(0),
bound_image_id(0),
hint(TEXTURE_HINT_IMMUTABLE),
type(RESOURCE_TYPE_BITMAP),
format(RGBA_8888),
shared_bitmap(bitmap),
gpu_memory_buffer(NULL) {
DCHECK(origin == DELEGATED || pixels);
if (bitmap)
shared_bitmap_id = bitmap->id();
}
ResourceProvider::Resource::Resource(const SharedBitmapId& bitmap_id,
const gfx::Size& size,
Origin origin,
GLenum filter)
: child_id(0),
gl_id(0),
gl_pixel_buffer_id(0),
gl_upload_query_id(0),
gl_read_lock_query_id(0),
pixels(NULL),
lock_for_read_count(0),
imported_count(0),
exported_count(0),
dirty_image(false),
locked_for_write(false),
lost(false),
marked_for_deletion(false),
allocated(false),
read_lock_fences_enabled(false),
has_shared_bitmap_id(true),
is_overlay_candidate(false),
read_lock_fence(NULL),
size(size),
origin(origin),
target(0),
original_filter(filter),
filter(filter),
image_id(0),
bound_image_id(0),
hint(TEXTURE_HINT_IMMUTABLE),
type(RESOURCE_TYPE_BITMAP),
format(RGBA_8888),
shared_bitmap_id(bitmap_id),
shared_bitmap(NULL),
gpu_memory_buffer(NULL) {}
ResourceProvider::Child::Child()
: marked_for_deletion(false), needs_sync_points(true) {
}
ResourceProvider::Child::~Child() {}
scoped_ptr<ResourceProvider> ResourceProvider::Create(
OutputSurface* output_surface,
SharedBitmapManager* shared_bitmap_manager,
gpu::GpuMemoryBufferManager* gpu_memory_buffer_manager,
BlockingTaskRunner* blocking_main_thread_task_runner,
int highp_threshold_min,
size_t id_allocation_chunk_size,
const std::vector<unsigned>& use_image_texture_targets) {
scoped_ptr<ResourceProvider> resource_provider(new ResourceProvider(
output_surface, shared_bitmap_manager, gpu_memory_buffer_manager,
blocking_main_thread_task_runner, highp_threshold_min,
id_allocation_chunk_size, use_image_texture_targets));
resource_provider->Initialize();
return resource_provider;
}
ResourceProvider::~ResourceProvider() {
base::trace_event::MemoryDumpManager::GetInstance()->UnregisterDumpProvider(
this);
while (!children_.empty())
DestroyChildInternal(children_.begin(), FOR_SHUTDOWN);
while (!resources_.empty())
DeleteResourceInternal(resources_.begin(), FOR_SHUTDOWN);
GLES2Interface* gl = ContextGL();
if (default_resource_type_ != RESOURCE_TYPE_GL_TEXTURE) {
// We are not in GL mode, but double check before returning.
DCHECK(!gl);
return;
}
DCHECK(gl);
#if DCHECK_IS_ON()
// Check that all GL resources has been deleted.
for (ResourceMap::const_iterator itr = resources_.begin();
itr != resources_.end(); ++itr) {
DCHECK_NE(RESOURCE_TYPE_GL_TEXTURE, itr->second.type);
}
#endif // DCHECK_IS_ON()
texture_id_allocator_ = nullptr;
buffer_id_allocator_ = nullptr;
gl->Finish();
}
bool ResourceProvider::InUseByConsumer(ResourceId id) {
Resource* resource = GetResource(id);
return resource->lock_for_read_count > 0 || resource->exported_count > 0 ||
resource->lost;
}
bool ResourceProvider::IsLost(ResourceId id) {
Resource* resource = GetResource(id);
return resource->lost;
}
void ResourceProvider::LoseResourceForTesting(ResourceId id) {
Resource* resource = GetResource(id);
DCHECK(resource);
resource->lost = true;
}
ResourceId ResourceProvider::CreateResource(const gfx::Size& size,
TextureHint hint,
ResourceFormat format) {
DCHECK(!size.IsEmpty());
switch (default_resource_type_) {
case RESOURCE_TYPE_GL_TEXTURE:
return CreateGLTexture(size,
GL_TEXTURE_2D,
hint,
format);
case RESOURCE_TYPE_BITMAP:
DCHECK_EQ(RGBA_8888, format);
return CreateBitmap(size);
}
LOG(FATAL) << "Invalid default resource type.";
return 0;
}
ResourceId ResourceProvider::CreateResourceWithImageTextureTarget(
const gfx::Size& size,
TextureHint hint,
ResourceFormat format) {
DCHECK(!size.IsEmpty());
switch (default_resource_type_) {
case RESOURCE_TYPE_GL_TEXTURE: {
return CreateGLTexture(size, GetImageTextureTarget(format), hint, format);
}
case RESOURCE_TYPE_BITMAP:
DCHECK_EQ(RGBA_8888, format);
return CreateBitmap(size);
}
LOG(FATAL) << "Invalid default resource type.";
return 0;
}
ResourceId ResourceProvider::CreateGLTexture(const gfx::Size& size,
GLenum target,
TextureHint hint,
ResourceFormat format) {
DCHECK_LE(size.width(), max_texture_size_);
DCHECK_LE(size.height(), max_texture_size_);
DCHECK(thread_checker_.CalledOnValidThread());
ResourceId id = next_id_++;
Resource* resource = InsertResource(
id,
Resource(0, size, Resource::INTERNAL, target, GL_LINEAR, hint, format));
resource->allocated = false;
return id;
}
ResourceId ResourceProvider::CreateBitmap(const gfx::Size& size) {
DCHECK(thread_checker_.CalledOnValidThread());
scoped_ptr<SharedBitmap> bitmap =
shared_bitmap_manager_->AllocateSharedBitmap(size);
uint8_t* pixels = bitmap->pixels();
DCHECK(pixels);
ResourceId id = next_id_++;
Resource* resource = InsertResource(
id,
Resource(pixels, bitmap.release(), size, Resource::INTERNAL, GL_LINEAR));
resource->allocated = true;
return id;
}
ResourceId ResourceProvider::CreateResourceFromIOSurface(
const gfx::Size& size,
unsigned io_surface_id) {
DCHECK(thread_checker_.CalledOnValidThread());
ResourceId id = next_id_++;
Resource* resource = InsertResource(
id, Resource(0, gfx::Size(), Resource::INTERNAL, GL_TEXTURE_RECTANGLE_ARB,
GL_LINEAR, TEXTURE_HINT_IMMUTABLE, RGBA_8888));
LazyCreate(resource);
GLES2Interface* gl = ContextGL();
DCHECK(gl);
gl->BindTexture(GL_TEXTURE_RECTANGLE_ARB, resource->gl_id);
gl->TexImageIOSurface2DCHROMIUM(
GL_TEXTURE_RECTANGLE_ARB, size.width(), size.height(), io_surface_id, 0);
resource->allocated = true;
return id;
}
ResourceId ResourceProvider::CreateResourceFromTextureMailbox(
const TextureMailbox& mailbox,
scoped_ptr<SingleReleaseCallbackImpl> release_callback_impl,
bool read_lock_fences_enabled) {
DCHECK(thread_checker_.CalledOnValidThread());
// Just store the information. Mailbox will be consumed in LockForRead().
ResourceId id = next_id_++;
DCHECK(mailbox.IsValid());
Resource* resource = nullptr;
if (mailbox.IsTexture()) {
resource = InsertResource(
id, Resource(0, gfx::Size(), Resource::EXTERNAL, mailbox.target(),
mailbox.nearest_neighbor() ? GL_NEAREST : GL_LINEAR,
TEXTURE_HINT_IMMUTABLE, RGBA_8888));
} else {
DCHECK(mailbox.IsSharedMemory());
SharedBitmap* shared_bitmap = mailbox.shared_bitmap();
uint8_t* pixels = shared_bitmap->pixels();
DCHECK(pixels);
resource = InsertResource(
id, Resource(pixels, shared_bitmap, mailbox.size_in_pixels(),
Resource::EXTERNAL, GL_LINEAR));
}
resource->allocated = true;
resource->mailbox = mailbox;
resource->release_callback_impl =
base::Bind(&SingleReleaseCallbackImpl::Run,
base::Owned(release_callback_impl.release()));
resource->read_lock_fences_enabled = read_lock_fences_enabled;
resource->is_overlay_candidate = mailbox.is_overlay_candidate();
return id;
}
ResourceId ResourceProvider::CreateResourceFromTextureMailbox(
const TextureMailbox& mailbox,
scoped_ptr<SingleReleaseCallbackImpl> release_callback_impl) {
return CreateResourceFromTextureMailbox(mailbox, release_callback_impl.Pass(),
false);
}
void ResourceProvider::DeleteResource(ResourceId id) {
DCHECK(thread_checker_.CalledOnValidThread());
ResourceMap::iterator it = resources_.find(id);
CHECK(it != resources_.end());
Resource* resource = &it->second;
DCHECK(!resource->marked_for_deletion);
DCHECK_EQ(resource->imported_count, 0);
DCHECK(!resource->locked_for_write);
if (resource->exported_count > 0 || resource->lock_for_read_count > 0 ||
!ReadLockFenceHasPassed(resource)) {
resource->marked_for_deletion = true;
return;
} else {
DeleteResourceInternal(it, NORMAL);
}
}
void ResourceProvider::DeleteResourceInternal(ResourceMap::iterator it,
DeleteStyle style) {
TRACE_EVENT0("cc", "ResourceProvider::DeleteResourceInternal");
Resource* resource = &it->second;
bool lost_resource = resource->lost;
DCHECK(resource->exported_count == 0 || style != NORMAL);
if (style == FOR_SHUTDOWN && resource->exported_count > 0)
lost_resource = true;
if (resource->image_id) {
DCHECK(resource->origin == Resource::INTERNAL);
GLES2Interface* gl = ContextGL();
DCHECK(gl);
gl->DestroyImageCHROMIUM(resource->image_id);
}
if (resource->gl_upload_query_id) {
DCHECK(resource->origin == Resource::INTERNAL);
GLES2Interface* gl = ContextGL();
DCHECK(gl);
gl->DeleteQueriesEXT(1, &resource->gl_upload_query_id);
}
if (resource->gl_read_lock_query_id) {
DCHECK(resource->origin == Resource::INTERNAL);
GLES2Interface* gl = ContextGL();
DCHECK(gl);
gl->DeleteQueriesEXT(1, &resource->gl_read_lock_query_id);
}
if (resource->gl_pixel_buffer_id) {
DCHECK(resource->origin == Resource::INTERNAL);
GLES2Interface* gl = ContextGL();
DCHECK(gl);
gl->DeleteBuffers(1, &resource->gl_pixel_buffer_id);
}
if (resource->origin == Resource::EXTERNAL) {
DCHECK(resource->mailbox.IsValid());
GLuint sync_point = resource->mailbox.sync_point();
if (resource->type == RESOURCE_TYPE_GL_TEXTURE) {
DCHECK(resource->mailbox.IsTexture());
lost_resource |= lost_output_surface_;
GLES2Interface* gl = ContextGL();
DCHECK(gl);
if (resource->gl_id) {
gl->DeleteTextures(1, &resource->gl_id);
resource->gl_id = 0;
if (!lost_resource)
sync_point = gl->InsertSyncPointCHROMIUM();
}
} else {
DCHECK(resource->mailbox.IsSharedMemory());
resource->shared_bitmap = nullptr;
resource->pixels = nullptr;
}
resource->release_callback_impl.Run(
sync_point, lost_resource, blocking_main_thread_task_runner_);
}
if (resource->gl_id) {
GLES2Interface* gl = ContextGL();
DCHECK(gl);
gl->DeleteTextures(1, &resource->gl_id);
resource->gl_id = 0;
}
if (resource->shared_bitmap) {
DCHECK(resource->origin != Resource::EXTERNAL);
DCHECK_EQ(RESOURCE_TYPE_BITMAP, resource->type);
delete resource->shared_bitmap;
resource->pixels = NULL;
}
if (resource->pixels) {
DCHECK(resource->origin == Resource::INTERNAL);
delete[] resource->pixels;
resource->pixels = NULL;
}
if (resource->gpu_memory_buffer) {
DCHECK(resource->origin == Resource::INTERNAL);
delete resource->gpu_memory_buffer;
resource->gpu_memory_buffer = NULL;
}
resources_.erase(it);
}
ResourceProvider::ResourceType ResourceProvider::GetResourceType(
ResourceId id) {
return GetResource(id)->type;
}
void ResourceProvider::CopyToResource(ResourceId id,
const uint8_t* image,
const gfx::Size& image_size) {
Resource* resource = GetResource(id);
DCHECK(!resource->locked_for_write);
DCHECK(!resource->lock_for_read_count);
DCHECK(resource->origin == Resource::INTERNAL);
DCHECK_EQ(resource->exported_count, 0);
DCHECK(ReadLockFenceHasPassed(resource));
DCHECK_EQ(image_size.width(), resource->size.width());
DCHECK_EQ(image_size.height(), resource->size.height());
if (resource->type == RESOURCE_TYPE_BITMAP) {
DCHECK_EQ(RESOURCE_TYPE_BITMAP, resource->type);
DCHECK(resource->allocated);
DCHECK_EQ(RGBA_8888, resource->format);
SkImageInfo source_info =
SkImageInfo::MakeN32Premul(image_size.width(), image_size.height());
size_t image_stride = image_size.width() * 4;
ScopedWriteLockSoftware lock(this, id);
SkCanvas dest(lock.sk_bitmap());
dest.writePixels(source_info, image, image_stride, 0, 0);
} else {
ScopedWriteLockGL lock(this, id);
DCHECK(lock.texture_id());
GLES2Interface* gl = ContextGL();
DCHECK(gl);
gl->BindTexture(resource->target, lock.texture_id());
if (resource->format == ETC1) {
DCHECK_EQ(resource->target, static_cast<GLenum>(GL_TEXTURE_2D));
int image_bytes = ResourceUtil::CheckedSizeInBytes<int>(image_size, ETC1);
gl->CompressedTexImage2D(resource->target, 0, GLInternalFormat(ETC1),
image_size.width(), image_size.height(), 0,
image_bytes, image);
} else {
gl->TexSubImage2D(resource->target, 0, 0, 0, image_size.width(),
image_size.height(), GLDataFormat(resource->format),
GLDataType(resource->format), image);
}
}
}
ResourceProvider::Resource* ResourceProvider::InsertResource(
ResourceId id,
const Resource& resource) {
std::pair<ResourceMap::iterator, bool> result =
resources_.insert(ResourceMap::value_type(id, resource));
DCHECK(result.second);
return &result.first->second;
}
ResourceProvider::Resource* ResourceProvider::GetResource(ResourceId id) {
DCHECK(thread_checker_.CalledOnValidThread());
DCHECK(id);
ResourceMap::iterator it = resources_.find(id);
DCHECK(it != resources_.end());
return &it->second;
}
const ResourceProvider::Resource* ResourceProvider::LockForRead(ResourceId id) {
Resource* resource = GetResource(id);
DCHECK(!resource->locked_for_write) << "locked for write: "
<< resource->locked_for_write;
DCHECK_EQ(resource->exported_count, 0);
// Uninitialized! Call SetPixels or LockForWrite first.
DCHECK(resource->allocated);
LazyCreate(resource);
if (resource->type == RESOURCE_TYPE_GL_TEXTURE && !resource->gl_id) {
DCHECK(resource->origin != Resource::INTERNAL);
DCHECK(resource->mailbox.IsTexture());
// Mailbox sync_points must be processed by a call to
// WaitSyncPointIfNeeded() prior to calling LockForRead().
DCHECK(!resource->mailbox.sync_point());
GLES2Interface* gl = ContextGL();
DCHECK(gl);
resource->gl_id = gl->CreateAndConsumeTextureCHROMIUM(
resource->mailbox.target(), resource->mailbox.name());
}
if (!resource->pixels && resource->has_shared_bitmap_id &&
shared_bitmap_manager_) {
scoped_ptr<SharedBitmap> bitmap =
shared_bitmap_manager_->GetSharedBitmapFromId(
resource->size, resource->shared_bitmap_id);
if (bitmap) {
resource->shared_bitmap = bitmap.release();
resource->pixels = resource->shared_bitmap->pixels();
}
}
resource->lock_for_read_count++;
if (resource->read_lock_fences_enabled) {
if (current_read_lock_fence_.get())
current_read_lock_fence_->Set();
resource->read_lock_fence = current_read_lock_fence_;
}
return resource;
}
void ResourceProvider::UnlockForRead(ResourceId id) {
DCHECK(thread_checker_.CalledOnValidThread());
ResourceMap::iterator it = resources_.find(id);
CHECK(it != resources_.end());
Resource* resource = &it->second;
DCHECK_GT(resource->lock_for_read_count, 0);
DCHECK_EQ(resource->exported_count, 0);
resource->lock_for_read_count--;
if (resource->marked_for_deletion && !resource->lock_for_read_count) {
if (!resource->child_id) {
// The resource belongs to this ResourceProvider, so it can be destroyed.
DeleteResourceInternal(it, NORMAL);
} else {
ChildMap::iterator child_it = children_.find(resource->child_id);
ResourceIdArray unused;
unused.push_back(id);
DeleteAndReturnUnusedResourcesToChild(child_it, NORMAL, unused);
}
}
}
ResourceProvider::Resource* ResourceProvider::LockForWrite(ResourceId id) {
Resource* resource = GetResource(id);
DCHECK(CanLockForWrite(id));
resource->locked_for_write = true;
return resource;
}
bool ResourceProvider::CanLockForWrite(ResourceId id) {
Resource* resource = GetResource(id);
return !resource->locked_for_write && !resource->lock_for_read_count &&
!resource->exported_count && resource->origin == Resource::INTERNAL &&
!resource->lost && ReadLockFenceHasPassed(resource);
}
bool ResourceProvider::IsOverlayCandidate(ResourceId id) {
Resource* resource = GetResource(id);
return resource->is_overlay_candidate;
}
void ResourceProvider::UnlockForWrite(ResourceProvider::Resource* resource) {
DCHECK(resource->locked_for_write);
DCHECK_EQ(resource->exported_count, 0);
DCHECK(resource->origin == Resource::INTERNAL);
resource->locked_for_write = false;
}
void ResourceProvider::EnableReadLockFencesForTesting(ResourceId id) {
Resource* resource = GetResource(id);
DCHECK(resource);
resource->read_lock_fences_enabled = true;
}
ResourceProvider::ScopedReadLockGL::ScopedReadLockGL(
ResourceProvider* resource_provider,
ResourceId resource_id)
: resource_provider_(resource_provider),
resource_id_(resource_id),
resource_(resource_provider->LockForRead(resource_id)) {
DCHECK(resource_);
}
ResourceProvider::ScopedReadLockGL::~ScopedReadLockGL() {
resource_provider_->UnlockForRead(resource_id_);
}
ResourceProvider::ScopedSamplerGL::ScopedSamplerGL(
ResourceProvider* resource_provider,
ResourceId resource_id,
GLenum filter)
: ScopedReadLockGL(resource_provider, resource_id),
unit_(GL_TEXTURE0),
target_(resource_provider->BindForSampling(resource_id, unit_, filter)) {
}
ResourceProvider::ScopedSamplerGL::ScopedSamplerGL(
ResourceProvider* resource_provider,
ResourceId resource_id,
GLenum unit,
GLenum filter)
: ScopedReadLockGL(resource_provider, resource_id),
unit_(unit),
target_(resource_provider->BindForSampling(resource_id, unit_, filter)) {
}
ResourceProvider::ScopedSamplerGL::~ScopedSamplerGL() {
}
ResourceProvider::ScopedWriteLockGL::ScopedWriteLockGL(
ResourceProvider* resource_provider,
ResourceId resource_id)
: resource_provider_(resource_provider),
resource_(resource_provider->LockForWrite(resource_id)) {
resource_provider_->LazyAllocate(resource_);
texture_id_ = resource_->gl_id;
DCHECK(texture_id_);
}
ResourceProvider::ScopedWriteLockGL::~ScopedWriteLockGL() {
resource_provider_->UnlockForWrite(resource_);
}
void ResourceProvider::PopulateSkBitmapWithResource(
SkBitmap* sk_bitmap, const Resource* resource) {
DCHECK_EQ(RGBA_8888, resource->format);
SkImageInfo info = SkImageInfo::MakeN32Premul(resource->size.width(),
resource->size.height());
sk_bitmap->installPixels(info, resource->pixels, info.minRowBytes());
}
ResourceProvider::ScopedReadLockSoftware::ScopedReadLockSoftware(
ResourceProvider* resource_provider,
ResourceId resource_id)
: resource_provider_(resource_provider), resource_id_(resource_id) {
const Resource* resource = resource_provider->LockForRead(resource_id);
ResourceProvider::PopulateSkBitmapWithResource(&sk_bitmap_, resource);
}
ResourceProvider::ScopedReadLockSoftware::~ScopedReadLockSoftware() {
resource_provider_->UnlockForRead(resource_id_);
}
ResourceProvider::ScopedWriteLockSoftware::ScopedWriteLockSoftware(
ResourceProvider* resource_provider,
ResourceId resource_id)
: resource_provider_(resource_provider),
resource_(resource_provider->LockForWrite(resource_id)) {
ResourceProvider::PopulateSkBitmapWithResource(&sk_bitmap_, resource_);
DCHECK(valid());
}
ResourceProvider::ScopedWriteLockSoftware::~ScopedWriteLockSoftware() {
DCHECK(thread_checker_.CalledOnValidThread());
resource_provider_->UnlockForWrite(resource_);
}
ResourceProvider::ScopedWriteLockGpuMemoryBuffer::
ScopedWriteLockGpuMemoryBuffer(ResourceProvider* resource_provider,
ResourceId resource_id)
: resource_provider_(resource_provider),
resource_(resource_provider->LockForWrite(resource_id)),
gpu_memory_buffer_manager_(resource_provider->gpu_memory_buffer_manager_),
gpu_memory_buffer_(nullptr),
size_(resource_->size),
format_(resource_->format) {
DCHECK_EQ(RESOURCE_TYPE_GL_TEXTURE, resource_->type);
std::swap(gpu_memory_buffer_, resource_->gpu_memory_buffer);
}
ResourceProvider::ScopedWriteLockGpuMemoryBuffer::
~ScopedWriteLockGpuMemoryBuffer() {
DCHECK(thread_checker_.CalledOnValidThread());
resource_provider_->UnlockForWrite(resource_);
if (!gpu_memory_buffer_)
return;
resource_provider_->LazyCreate(resource_);
if (!resource_->image_id) {
GLES2Interface* gl = resource_provider_->ContextGL();
DCHECK(gl);
#if defined(OS_CHROMEOS)
// TODO(reveman): GL_COMMANDS_ISSUED_CHROMIUM is used for synchronization
// on ChromeOS to avoid some performance issues. This only works with
// shared memory backed buffers. crbug.com/436314
DCHECK_EQ(gpu_memory_buffer_->GetHandle().type, gfx::SHARED_MEMORY_BUFFER);
#endif
resource_->image_id = gl->CreateImageCHROMIUM(
gpu_memory_buffer_->AsClientBuffer(), size_.width(), size_.height(),
GLInternalFormat(resource_->format));
}
std::swap(resource_->gpu_memory_buffer, gpu_memory_buffer_);
resource_->allocated = true;
resource_->dirty_image = true;
// GpuMemoryBuffer provides direct access to the memory used by the GPU.
// Read lock fences are required to ensure that we're not trying to map a
// buffer that is currently in-use by the GPU.
resource_->read_lock_fences_enabled = true;
}
gfx::GpuMemoryBuffer*
ResourceProvider::ScopedWriteLockGpuMemoryBuffer::GetGpuMemoryBuffer() {
if (gpu_memory_buffer_)
return gpu_memory_buffer_;
scoped_ptr<gfx::GpuMemoryBuffer> gpu_memory_buffer =
gpu_memory_buffer_manager_->AllocateGpuMemoryBuffer(
size_, BufferFormat(format_),
gfx::BufferUsage::GPU_READ_CPU_READ_WRITE);
gpu_memory_buffer_ = gpu_memory_buffer.release();
return gpu_memory_buffer_;
}
ResourceProvider::ScopedWriteLockGr::ScopedWriteLockGr(
ResourceProvider* resource_provider,
ResourceId resource_id)
: resource_provider_(resource_provider),
resource_(resource_provider->LockForWrite(resource_id)) {
DCHECK(thread_checker_.CalledOnValidThread());
resource_provider_->LazyAllocate(resource_);
}
ResourceProvider::ScopedWriteLockGr::~ScopedWriteLockGr() {
DCHECK(thread_checker_.CalledOnValidThread());
DCHECK(resource_->locked_for_write);
resource_provider_->UnlockForWrite(resource_);
}
void ResourceProvider::ScopedWriteLockGr::InitSkSurface(
bool use_distance_field_text,
bool can_use_lcd_text,
int msaa_sample_count) {
DCHECK(resource_->locked_for_write);
GrBackendTextureDesc desc;
desc.fFlags = kRenderTarget_GrBackendTextureFlag;
desc.fWidth = resource_->size.width();
desc.fHeight = resource_->size.height();
desc.fConfig = ToGrPixelConfig(resource_->format);
desc.fOrigin = kTopLeft_GrSurfaceOrigin;
desc.fTextureHandle = resource_->gl_id;
desc.fSampleCnt = msaa_sample_count;
bool use_worker_context = true;
class GrContext* gr_context =
resource_provider_->GrContext(use_worker_context);
uint32_t flags =
use_distance_field_text ? SkSurfaceProps::kUseDistanceFieldFonts_Flag : 0;
// Use unknown pixel geometry to disable LCD text.
SkSurfaceProps surface_props(flags, kUnknown_SkPixelGeometry);
if (can_use_lcd_text) {
// LegacyFontHost will get LCD text and skia figures out what type to use.
surface_props =
SkSurfaceProps(flags, SkSurfaceProps::kLegacyFontHost_InitType);
}
sk_surface_ = skia::AdoptRef(
SkSurface::NewWrappedRenderTarget(gr_context, desc, &surface_props));
}
void ResourceProvider::ScopedWriteLockGr::ReleaseSkSurface() {
sk_surface_.clear();
}
ResourceProvider::SynchronousFence::SynchronousFence(
gpu::gles2::GLES2Interface* gl)
: gl_(gl), has_synchronized_(true) {
}
ResourceProvider::SynchronousFence::~SynchronousFence() {
}
void ResourceProvider::SynchronousFence::Set() {
has_synchronized_ = false;
}
bool ResourceProvider::SynchronousFence::HasPassed() {
if (!has_synchronized_) {
has_synchronized_ = true;
Synchronize();
}
return true;
}
void ResourceProvider::SynchronousFence::Wait() {
HasPassed();
}
void ResourceProvider::SynchronousFence::Synchronize() {
TRACE_EVENT0("cc", "ResourceProvider::SynchronousFence::Synchronize");
gl_->Finish();
}
ResourceProvider::ResourceProvider(
OutputSurface* output_surface,
SharedBitmapManager* shared_bitmap_manager,
gpu::GpuMemoryBufferManager* gpu_memory_buffer_manager,
BlockingTaskRunner* blocking_main_thread_task_runner,
int highp_threshold_min,
size_t id_allocation_chunk_size,
const std::vector<unsigned>& use_image_texture_targets)
: output_surface_(output_surface),
shared_bitmap_manager_(shared_bitmap_manager),
gpu_memory_buffer_manager_(gpu_memory_buffer_manager),
blocking_main_thread_task_runner_(blocking_main_thread_task_runner),
lost_output_surface_(false),
highp_threshold_min_(highp_threshold_min),
next_id_(1),
next_child_(1),
default_resource_type_(RESOURCE_TYPE_BITMAP),
use_texture_storage_ext_(false),
use_texture_format_bgra_(false),
use_texture_usage_hint_(false),
use_compressed_texture_etc1_(false),
yuv_resource_format_(LUMINANCE_8),
max_texture_size_(0),
best_texture_format_(RGBA_8888),
best_render_buffer_format_(RGBA_8888),
id_allocation_chunk_size_(id_allocation_chunk_size),
use_sync_query_(false),
use_image_texture_targets_(use_image_texture_targets),
tracing_id_(g_next_resource_provider_tracing_id.GetNext()) {
DCHECK(output_surface_->HasClient());
DCHECK(id_allocation_chunk_size_);
}
void ResourceProvider::Initialize() {
DCHECK(thread_checker_.CalledOnValidThread());
// In certain cases, ThreadTaskRunnerHandle isn't set (Android Webview).
// Don't register a dump provider in these cases.
// TODO(ericrk): Get this working in Android Webview. crbug.com/517156
if (base::ThreadTaskRunnerHandle::IsSet()) {
base::trace_event::MemoryDumpManager::GetInstance()->RegisterDumpProvider(
this, "cc::ResourceProvider", base::ThreadTaskRunnerHandle::Get());
}
GLES2Interface* gl = ContextGL();
if (!gl) {
default_resource_type_ = RESOURCE_TYPE_BITMAP;
// Pick an arbitrary limit here similar to what hardware might.
max_texture_size_ = 16 * 1024;
best_texture_format_ = RGBA_8888;
return;
}
DCHECK(!texture_id_allocator_);
DCHECK(!buffer_id_allocator_);
const ContextProvider::Capabilities& caps =
output_surface_->context_provider()->ContextCapabilities();
default_resource_type_ = RESOURCE_TYPE_GL_TEXTURE;
use_texture_storage_ext_ = caps.gpu.texture_storage;
use_texture_format_bgra_ = caps.gpu.texture_format_bgra8888;
use_texture_usage_hint_ = caps.gpu.texture_usage;
use_compressed_texture_etc1_ = caps.gpu.texture_format_etc1;
yuv_resource_format_ = caps.gpu.texture_rg ? RED_8 : LUMINANCE_8;
use_sync_query_ = caps.gpu.sync_query;
max_texture_size_ = 0; // Context expects cleared value.
gl->GetIntegerv(GL_MAX_TEXTURE_SIZE, &max_texture_size_);
best_texture_format_ =
PlatformColor::BestTextureFormat(use_texture_format_bgra_);
best_render_buffer_format_ =
PlatformColor::BestTextureFormat(caps.gpu.render_buffer_format_bgra8888);
texture_id_allocator_.reset(
new TextureIdAllocator(gl, id_allocation_chunk_size_));
buffer_id_allocator_.reset(
new BufferIdAllocator(gl, id_allocation_chunk_size_));
}
int ResourceProvider::CreateChild(const ReturnCallback& return_callback) {
DCHECK(thread_checker_.CalledOnValidThread());
Child child_info;
child_info.return_callback = return_callback;
int child = next_child_++;
children_[child] = child_info;
return child;
}
void ResourceProvider::SetChildNeedsSyncPoints(int child_id, bool needs) {
ChildMap::iterator it = children_.find(child_id);
DCHECK(it != children_.end());
it->second.needs_sync_points = needs;
}
void ResourceProvider::DestroyChild(int child_id) {
ChildMap::iterator it = children_.find(child_id);
DCHECK(it != children_.end());
DestroyChildInternal(it, NORMAL);
}
void ResourceProvider::DestroyChildInternal(ChildMap::iterator it,
DeleteStyle style) {
DCHECK(thread_checker_.CalledOnValidThread());
Child& child = it->second;
DCHECK(style == FOR_SHUTDOWN || !child.marked_for_deletion);
ResourceIdArray resources_for_child;
for (ResourceIdMap::iterator child_it = child.child_to_parent_map.begin();
child_it != child.child_to_parent_map.end();
++child_it) {
ResourceId id = child_it->second;
resources_for_child.push_back(id);
}
child.marked_for_deletion = true;
DeleteAndReturnUnusedResourcesToChild(it, style, resources_for_child);
}
const ResourceProvider::ResourceIdMap& ResourceProvider::GetChildToParentMap(
int child) const {
DCHECK(thread_checker_.CalledOnValidThread());
ChildMap::const_iterator it = children_.find(child);
DCHECK(it != children_.end());
DCHECK(!it->second.marked_for_deletion);
return it->second.child_to_parent_map;
}
void ResourceProvider::PrepareSendToParent(const ResourceIdArray& resources,
TransferableResourceArray* list) {
DCHECK(thread_checker_.CalledOnValidThread());
GLES2Interface* gl = ContextGL();
bool need_sync_point = false;
for (ResourceIdArray::const_iterator it = resources.begin();
it != resources.end();
++it) {
TransferableResource resource;
TransferResource(gl, *it, &resource);
if (!resource.mailbox_holder.sync_point && !resource.is_software)
need_sync_point = true;
++resources_.find(*it)->second.exported_count;
list->push_back(resource);
}
if (need_sync_point &&
output_surface_->capabilities().delegated_sync_points_required) {
GLuint sync_point = gl->InsertSyncPointCHROMIUM();
for (TransferableResourceArray::iterator it = list->begin();
it != list->end();
++it) {
if (!it->mailbox_holder.sync_point)
it->mailbox_holder.sync_point = sync_point;
}
}
}
void ResourceProvider::ReceiveFromChild(
int child, const TransferableResourceArray& resources) {
DCHECK(thread_checker_.CalledOnValidThread());
GLES2Interface* gl = ContextGL();
Child& child_info = children_.find(child)->second;
DCHECK(!child_info.marked_for_deletion);
for (TransferableResourceArray::const_iterator it = resources.begin();
it != resources.end();
++it) {
ResourceIdMap::iterator resource_in_map_it =
child_info.child_to_parent_map.find(it->id);
if (resource_in_map_it != child_info.child_to_parent_map.end()) {
Resource* resource = GetResource(resource_in_map_it->second);
resource->marked_for_deletion = false;
resource->imported_count++;
continue;
}
if ((!it->is_software && !gl) ||
(it->is_software && !shared_bitmap_manager_)) {
TRACE_EVENT0("cc", "ResourceProvider::ReceiveFromChild dropping invalid");
ReturnedResourceArray to_return;
to_return.push_back(it->ToReturnedResource());
child_info.return_callback.Run(to_return,
blocking_main_thread_task_runner_);
continue;
}
ResourceId local_id = next_id_++;
Resource* resource = nullptr;
if (it->is_software) {
resource = InsertResource(local_id,
Resource(it->mailbox_holder.mailbox, it->size,
Resource::DELEGATED, GL_LINEAR));
} else {
resource = InsertResource(
local_id, Resource(0, it->size, Resource::DELEGATED,
it->mailbox_holder.texture_target, it->filter,
TEXTURE_HINT_IMMUTABLE, it->format));
resource->mailbox = TextureMailbox(it->mailbox_holder.mailbox,
it->mailbox_holder.texture_target,
it->mailbox_holder.sync_point);
resource->read_lock_fences_enabled = it->read_lock_fences_enabled;
resource->is_overlay_candidate = it->is_overlay_candidate;
}
resource->child_id = child;
// Don't allocate a texture for a child.
resource->allocated = true;
resource->imported_count = 1;
child_info.parent_to_child_map[local_id] = it->id;
child_info.child_to_parent_map[it->id] = local_id;
}
}
void ResourceProvider::DeclareUsedResourcesFromChild(
int child,
const ResourceIdSet& resources_from_child) {
DCHECK(thread_checker_.CalledOnValidThread());
ChildMap::iterator child_it = children_.find(child);
DCHECK(child_it != children_.end());
Child& child_info = child_it->second;
DCHECK(!child_info.marked_for_deletion);
ResourceIdArray unused;
for (ResourceIdMap::iterator it = child_info.child_to_parent_map.begin();
it != child_info.child_to_parent_map.end();
++it) {
ResourceId local_id = it->second;
bool resource_is_in_use = resources_from_child.count(it->first) > 0;
if (!resource_is_in_use)
unused.push_back(local_id);
}
DeleteAndReturnUnusedResourcesToChild(child_it, NORMAL, unused);
}
void ResourceProvider::ReceiveReturnsFromParent(
const ReturnedResourceArray& resources) {
DCHECK(thread_checker_.CalledOnValidThread());
GLES2Interface* gl = ContextGL();
base::hash_map<int, ResourceIdArray> resources_for_child;
for (const ReturnedResource& returned : resources) {
ResourceId local_id = returned.id;
ResourceMap::iterator map_iterator = resources_.find(local_id);
// Resource was already lost (e.g. it belonged to a child that was
// destroyed).
if (map_iterator == resources_.end())
continue;
Resource* resource = &map_iterator->second;
CHECK_GE(resource->exported_count, returned.count);
resource->exported_count -= returned.count;
resource->lost |= returned.lost;
if (resource->exported_count)
continue;
if (returned.sync_point) {
DCHECK(!resource->has_shared_bitmap_id);
if (resource->origin == Resource::INTERNAL) {
DCHECK(resource->gl_id);
gl->WaitSyncPointCHROMIUM(returned.sync_point);
} else {
DCHECK(!resource->gl_id);
resource->mailbox.set_sync_point(returned.sync_point);
}
}
if (!resource->marked_for_deletion)
continue;
if (!resource->child_id) {
// The resource belongs to this ResourceProvider, so it can be destroyed.
DeleteResourceInternal(map_iterator, NORMAL);
continue;
}
DCHECK(resource->origin == Resource::DELEGATED);
resources_for_child[resource->child_id].push_back(local_id);
}
for (const auto& children : resources_for_child) {
ChildMap::iterator child_it = children_.find(children.first);
DCHECK(child_it != children_.end());
DeleteAndReturnUnusedResourcesToChild(child_it, NORMAL, children.second);
}
}
void ResourceProvider::TransferResource(GLES2Interface* gl,
ResourceId id,
TransferableResource* resource) {
Resource* source = GetResource(id);
DCHECK(!source->locked_for_write);
DCHECK(!source->lock_for_read_count);
DCHECK(source->origin != Resource::EXTERNAL || source->mailbox.IsValid());
DCHECK(source->allocated);
resource->id = id;
resource->format = source->format;
resource->mailbox_holder.texture_target = source->target;
resource->filter = source->filter;
resource->size = source->size;
resource->read_lock_fences_enabled = source->read_lock_fences_enabled;
resource->is_overlay_candidate = source->is_overlay_candidate;
if (source->type == RESOURCE_TYPE_BITMAP) {
resource->mailbox_holder.mailbox = source->shared_bitmap_id;
resource->is_software = true;
} else if (!source->mailbox.IsValid()) {
LazyCreate(source);
DCHECK(source->gl_id);
DCHECK(source->origin == Resource::INTERNAL);
if (source->image_id) {
DCHECK(source->dirty_image);
gl->BindTexture(resource->mailbox_holder.texture_target, source->gl_id);
BindImageForSampling(source);
}
// This is a resource allocated by the compositor, we need to produce it.
// Don't set a sync point, the caller will do it.
gl->GenMailboxCHROMIUM(resource->mailbox_holder.mailbox.name);
gl->ProduceTextureDirectCHROMIUM(source->gl_id,
resource->mailbox_holder.texture_target,
resource->mailbox_holder.mailbox.name);
source->mailbox = TextureMailbox(resource->mailbox_holder);
} else {
DCHECK(source->mailbox.IsTexture());
if (source->image_id && source->dirty_image) {
DCHECK(source->gl_id);
DCHECK(source->origin == Resource::INTERNAL);
gl->BindTexture(resource->mailbox_holder.texture_target, source->gl_id);
BindImageForSampling(source);
}
// This is either an external resource, or a compositor resource that we
// already exported. Make sure to forward the sync point that we were given.
resource->mailbox_holder.mailbox = source->mailbox.mailbox();
resource->mailbox_holder.texture_target = source->mailbox.target();
resource->mailbox_holder.sync_point = source->mailbox.sync_point();
source->mailbox.set_sync_point(0);
}
}
void ResourceProvider::DeleteAndReturnUnusedResourcesToChild(
ChildMap::iterator child_it,
DeleteStyle style,
const ResourceIdArray& unused) {
DCHECK(thread_checker_.CalledOnValidThread());
DCHECK(child_it != children_.end());
Child* child_info = &child_it->second;
if (unused.empty() && !child_info->marked_for_deletion)
return;
ReturnedResourceArray to_return;
GLES2Interface* gl = ContextGL();
bool need_sync_point = false;
for (size_t i = 0; i < unused.size(); ++i) {
ResourceId local_id = unused[i];
ResourceMap::iterator it = resources_.find(local_id);
CHECK(it != resources_.end());
Resource& resource = it->second;
DCHECK(!resource.locked_for_write);
DCHECK(child_info->parent_to_child_map.count(local_id));
ResourceId child_id = child_info->parent_to_child_map[local_id];
DCHECK(child_info->child_to_parent_map.count(child_id));
bool is_lost =
resource.lost ||
(resource.type == RESOURCE_TYPE_GL_TEXTURE && lost_output_surface_);
if (resource.exported_count > 0 || resource.lock_for_read_count > 0) {
if (style != FOR_SHUTDOWN) {
// Defer this resource deletion.
resource.marked_for_deletion = true;
continue;
}
// We can't postpone the deletion, so we'll have to lose it.
is_lost = true;
} else if (!ReadLockFenceHasPassed(&resource)) {
// TODO(dcastagna): see if it's possible to use this logic for
// the branch above too, where the resource is locked or still exported.
if (style != FOR_SHUTDOWN && !child_info->marked_for_deletion) {
// Defer this resource deletion.
resource.marked_for_deletion = true;
continue;
}
// We can't postpone the deletion, so we'll have to lose it.
is_lost = true;
}
if (gl && resource.filter != resource.original_filter) {
DCHECK(resource.target);
DCHECK(resource.gl_id);
gl->BindTexture(resource.target, resource.gl_id);
gl->TexParameteri(resource.target, GL_TEXTURE_MIN_FILTER,
resource.original_filter);
gl->TexParameteri(resource.target, GL_TEXTURE_MAG_FILTER,
resource.original_filter);
}
ReturnedResource returned;
returned.id = child_id;
returned.sync_point = resource.mailbox.sync_point();
if (!returned.sync_point && resource.type == RESOURCE_TYPE_GL_TEXTURE)
need_sync_point = true;
returned.count = resource.imported_count;
returned.lost = is_lost;
to_return.push_back(returned);
child_info->parent_to_child_map.erase(local_id);
child_info->child_to_parent_map.erase(child_id);
resource.imported_count = 0;
DeleteResourceInternal(it, style);
}
if (need_sync_point && child_info->needs_sync_points) {
DCHECK(gl);
GLuint sync_point = gl->InsertSyncPointCHROMIUM();
for (size_t i = 0; i < to_return.size(); ++i) {
if (!to_return[i].sync_point)
to_return[i].sync_point = sync_point;
}
}
if (!to_return.empty())
child_info->return_callback.Run(to_return,
blocking_main_thread_task_runner_);
if (child_info->marked_for_deletion &&
child_info->parent_to_child_map.empty()) {
DCHECK(child_info->child_to_parent_map.empty());
children_.erase(child_it);
}
}
GLenum ResourceProvider::BindForSampling(ResourceId resource_id,
GLenum unit,
GLenum filter) {
DCHECK(thread_checker_.CalledOnValidThread());
GLES2Interface* gl = ContextGL();
ResourceMap::iterator it = resources_.find(resource_id);
DCHECK(it != resources_.end());
Resource* resource = &it->second;
DCHECK(resource->lock_for_read_count);
DCHECK(!resource->locked_for_write);
ScopedSetActiveTexture scoped_active_tex(gl, unit);
GLenum target = resource->target;
gl->BindTexture(target, resource->gl_id);
if (filter != resource->filter) {
gl->TexParameteri(target, GL_TEXTURE_MIN_FILTER, filter);
gl->TexParameteri(target, GL_TEXTURE_MAG_FILTER, filter);
resource->filter = filter;
}
if (resource->image_id && resource->dirty_image)
BindImageForSampling(resource);
return target;
}
void ResourceProvider::CreateForTesting(ResourceId id) {
LazyCreate(GetResource(id));
}
GLenum ResourceProvider::TargetForTesting(ResourceId id) {
Resource* resource = GetResource(id);
return resource->target;
}
void ResourceProvider::LazyCreate(Resource* resource) {
if (resource->type != RESOURCE_TYPE_GL_TEXTURE ||
resource->origin != Resource::INTERNAL)
return;
if (resource->gl_id)
return;
DCHECK(resource->origin == Resource::INTERNAL);
DCHECK(!resource->mailbox.IsValid());
resource->gl_id = texture_id_allocator_->NextId();
GLES2Interface* gl = ContextGL();
DCHECK(gl);
// Create and set texture properties. Allocation is delayed until needed.
gl->BindTexture(resource->target, resource->gl_id);
gl->TexParameteri(resource->target, GL_TEXTURE_MIN_FILTER,
resource->original_filter);
gl->TexParameteri(resource->target, GL_TEXTURE_MAG_FILTER,
resource->original_filter);
gl->TexParameteri(resource->target, GL_TEXTURE_WRAP_S, GL_CLAMP_TO_EDGE);
gl->TexParameteri(resource->target, GL_TEXTURE_WRAP_T, GL_CLAMP_TO_EDGE);
if (use_texture_usage_hint_ && (resource->hint & TEXTURE_HINT_FRAMEBUFFER)) {
gl->TexParameteri(resource->target, GL_TEXTURE_USAGE_ANGLE,
GL_FRAMEBUFFER_ATTACHMENT_ANGLE);
}
}
void ResourceProvider::AllocateForTesting(ResourceId id) {
LazyAllocate(GetResource(id));
}
void ResourceProvider::LazyAllocate(Resource* resource) {
DCHECK(resource);
if (resource->allocated)
return;
LazyCreate(resource);
if (!resource->gl_id)
return;
resource->allocated = true;
GLES2Interface* gl = ContextGL();
gfx::Size& size = resource->size;
ResourceFormat format = resource->format;
gl->BindTexture(resource->target, resource->gl_id);
if (use_texture_storage_ext_ &&
IsFormatSupportedForStorage(format, use_texture_format_bgra_) &&
(resource->hint & TEXTURE_HINT_IMMUTABLE)) {
GLenum storage_format = TextureToStorageFormat(format);
gl->TexStorage2DEXT(resource->target, 1, storage_format, size.width(),
size.height());
} else {
// ETC1 does not support preallocation.
if (format != ETC1) {
gl->TexImage2D(resource->target, 0, GLInternalFormat(format),
size.width(), size.height(), 0, GLDataFormat(format),
GLDataType(format), NULL);
}
}
}
void ResourceProvider::BindImageForSampling(Resource* resource) {
GLES2Interface* gl = ContextGL();
DCHECK(resource->gl_id);
DCHECK(resource->image_id);
// Release image currently bound to texture.
if (resource->bound_image_id)
gl->ReleaseTexImage2DCHROMIUM(resource->target, resource->bound_image_id);
gl->BindTexImage2DCHROMIUM(resource->target, resource->image_id);
resource->bound_image_id = resource->image_id;
resource->dirty_image = false;
}
void ResourceProvider::WaitSyncPointIfNeeded(ResourceId id) {
Resource* resource = GetResource(id);
DCHECK_EQ(resource->exported_count, 0);
DCHECK(resource->allocated);
if (resource->type != RESOURCE_TYPE_GL_TEXTURE || resource->gl_id)
return;
if (!resource->mailbox.sync_point())
return;
DCHECK(resource->mailbox.IsValid());
GLES2Interface* gl = ContextGL();
DCHECK(gl);
gl->WaitSyncPointCHROMIUM(resource->mailbox.sync_point());
resource->mailbox.set_sync_point(0);
}
GLint ResourceProvider::GetActiveTextureUnit(GLES2Interface* gl) {
GLint active_unit = 0;
gl->GetIntegerv(GL_ACTIVE_TEXTURE, &active_unit);
return active_unit;
}
GLenum ResourceProvider::GetImageTextureTarget(ResourceFormat format) {
gfx::BufferFormat buffer_format = BufferFormat(format);
DCHECK_GT(use_image_texture_targets_.size(),
static_cast<size_t>(buffer_format));
return use_image_texture_targets_[static_cast<size_t>(buffer_format)];
}
void ResourceProvider::ValidateResource(ResourceId id) const {
DCHECK(thread_checker_.CalledOnValidThread());
DCHECK(id);
DCHECK(resources_.find(id) != resources_.end());
}
GLES2Interface* ResourceProvider::ContextGL() const {
ContextProvider* context_provider = output_surface_->context_provider();
return context_provider ? context_provider->ContextGL() : NULL;
}
class GrContext* ResourceProvider::GrContext(bool worker_context) const {
ContextProvider* context_provider =
worker_context ? output_surface_->worker_context_provider()
: output_surface_->context_provider();
return context_provider ? context_provider->GrContext() : NULL;
}
bool ResourceProvider::OnMemoryDump(
const base::trace_event::MemoryDumpArgs& args,
base::trace_event::ProcessMemoryDump* pmd) {
DCHECK(thread_checker_.CalledOnValidThread());
const uint64 tracing_process_id =
base::trace_event::MemoryDumpManager::GetInstance()
->GetTracingProcessId();
for (const auto& resource_entry : resources_) {
const auto& resource = resource_entry.second;
// Resource IDs are not process-unique, so log with the ResourceProvider's
// unique id.
std::string dump_name =
base::StringPrintf("cc/resource_memory/provider_%d/resource_%d",
tracing_id_, resource_entry.first);
base::trace_event::MemoryAllocatorDump* dump =
pmd->CreateAllocatorDump(dump_name);
uint64_t total_bytes = ResourceUtil::UncheckedSizeInBytesAligned<size_t>(
resource.size, resource.format);
dump->AddScalar(base::trace_event::MemoryAllocatorDump::kNameSize,
base::trace_event::MemoryAllocatorDump::kUnitsBytes,
static_cast<uint64_t>(total_bytes));
// Resources which are shared across processes require a shared GUID to
// prevent double counting the memory. We currently support shared GUIDs for
// GpuMemoryBuffer, SharedBitmap, and GL backed resources.
base::trace_event::MemoryAllocatorDumpGuid guid;
if (resource.gpu_memory_buffer) {
guid = gfx::GetGpuMemoryBufferGUIDForTracing(
tracing_process_id, resource.gpu_memory_buffer->GetHandle().id);
} else if (resource.shared_bitmap) {
guid = GetSharedBitmapGUIDForTracing(resource.shared_bitmap->id());
} else if (resource.gl_id && resource.allocated) {
guid = gfx::GetGLTextureClientGUIDForTracing(
output_surface_->context_provider()
->ContextSupport()
->ShareGroupTracingGUID(),
resource.gl_id);
}
if (!guid.empty()) {
const int kImportance = 2;
pmd->CreateSharedGlobalAllocatorDump(guid);
pmd->AddOwnershipEdge(dump->guid(), guid, kImportance);
}
}
return true;
}
} // namespace cc