Mirrors/overte
3
0
Fork 0
mirror of https://github.com/overte-org/overte.git synced 2025-08-09 11:07:07 +02:00
Code Issues Projects Releases Packages Wiki Activity Actions 2

Merge pull request #13182 from jherico/feature/batch_pooling

Browse source 
Use pooled gpu batches
This commit is contained in:
Sam Gateau 2018-05-21 15:47:55 -07:00 committed by GitHub
commit 87989268bf
No known key found for this signature in database
GPG key ID: 4AEE18F83AFDEB23
9 changed files with 197 additions and 192 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

112
libraries/gpu/src/gpu/Batch.cpp
View file

@ -19,8 +19,7 @@
#if defined(NSIGHT_FOUND) #if defined(NSIGHT_FOUND)
#include "nvToolsExt.h" #include "nvToolsExt.h"
ProfileRangeBatch::ProfileRangeBatch(gpu::Batch& batch, const char* name) : _batch(batch) {
ProfileRangeBatch::ProfileRangeBatch(gpu::Batch& batch, const char *name) : _batch(batch) {
_batch.pushProfileRange(name); _batch.pushProfileRange(name);
} }
@ -38,19 +37,15 @@ static const int MAX_NUM_UNIFORM_BUFFERS = 14;
static const int MAX_NUM_RESOURCE_BUFFERS = 16; static const int MAX_NUM_RESOURCE_BUFFERS = 16;
static const int MAX_NUM_RESOURCE_TEXTURES = 16; static const int MAX_NUM_RESOURCE_TEXTURES = 16;
size_t Batch::_commandsMax { BATCH_PREALLOCATE_MIN }; size_t Batch::_commandsMax{ BATCH_PREALLOCATE_MIN };
size_t Batch::_commandOffsetsMax { BATCH_PREALLOCATE_MIN }; size_t Batch::_commandOffsetsMax{ BATCH_PREALLOCATE_MIN };
size_t Batch::_paramsMax { BATCH_PREALLOCATE_MIN }; size_t Batch::_paramsMax{ BATCH_PREALLOCATE_MIN };
size_t Batch::_dataMax { BATCH_PREALLOCATE_MIN }; size_t Batch::_dataMax{ BATCH_PREALLOCATE_MIN };
size_t Batch::_objectsMax { BATCH_PREALLOCATE_MIN }; size_t Batch::_objectsMax{ BATCH_PREALLOCATE_MIN };
size_t Batch::_drawCallInfosMax { BATCH_PREALLOCATE_MIN }; size_t Batch::_drawCallInfosMax{ BATCH_PREALLOCATE_MIN };
Batch::Batch(const char* name) { Batch::Batch(const char* name) {
#ifdef DEBUG _name = name;
if (name) {
_name = name;
}
#endif
_commands.reserve(_commandsMax); _commands.reserve(_commandsMax);
_commandOffsets.reserve(_commandOffsetsMax); _commandOffsets.reserve(_commandOffsetsMax);
_params.reserve(_paramsMax); _params.reserve(_paramsMax);
@ -59,38 +54,6 @@ Batch::Batch(const char* name) {
_drawCallInfos.reserve(_drawCallInfosMax); _drawCallInfos.reserve(_drawCallInfosMax);
} }
Batch::Batch(const Batch& batch_) {
Batch& batch = *const_cast<Batch*>(&batch_);
#ifdef DEBUG
_name = batch_._name;
#endif
_commands.swap(batch._commands);
_commandOffsets.swap(batch._commandOffsets);
_params.swap(batch._params);
_data.swap(batch._data);
_invalidModel = batch._invalidModel;
_currentModel = batch._currentModel;
_objects.swap(batch._objects);
_currentNamedCall = batch._currentNamedCall;
_buffers._items.swap(batch._buffers._items);
_textures._items.swap(batch._textures._items);
_textureTables._items.swap(batch._textureTables._items);
_streamFormats._items.swap(batch._streamFormats._items);
_transforms._items.swap(batch._transforms._items);
_pipelines._items.swap(batch._pipelines._items);
_framebuffers._items.swap(batch._framebuffers._items);
_swapChains._items.swap(batch._swapChains._items);
_drawCallInfos.swap(batch._drawCallInfos);
_queries._items.swap(batch._queries._items);
_lambdas._items.swap(batch._lambdas._items);
_profileRanges._items.swap(batch._profileRanges._items);
_names._items.swap(batch._names._items);
_namedData.swap(batch._namedData);
_enableStereo = batch._enableStereo;
_enableSkybox = batch._enableSkybox;
}
Batch::~Batch() { Batch::~Batch() {
_commandsMax = std::max(_commands.size(), _commandsMax); _commandsMax = std::max(_commands.size(), _commandsMax);
_commandOffsetsMax = std::max(_commandOffsets.size(), _commandOffsetsMax); _commandOffsetsMax = std::max(_commandOffsets.size(), _commandOffsetsMax);
@ -100,6 +63,10 @@ Batch::~Batch() {
_drawCallInfosMax = std::max(_drawCallInfos.size(), _drawCallInfosMax); _drawCallInfosMax = std::max(_drawCallInfos.size(), _drawCallInfosMax);
} }
void Batch::setName(const char* name) {
_name = name;
}
void Batch::clear() { void Batch::clear() {
_commandsMax = std::max(_commands.size(), _commandsMax); _commandsMax = std::max(_commands.size(), _commandsMax);
_commandOffsetsMax = std::max(_commandOffsets.size(), _commandOffsetsMax); _commandOffsetsMax = std::max(_commandOffsets.size(), _commandOffsetsMax);
@ -110,18 +77,30 @@ void Batch::clear() {
_commands.clear(); _commands.clear();
_commandOffsets.clear(); _commandOffsets.clear();
_params.clear();
_data.clear(); _data.clear();
_drawCallInfos.clear();
_buffers.clear(); _buffers.clear();
_framebuffers.clear();
_lambdas.clear();
_names.clear();
_namedData.clear();
_objects.clear();
_params.clear();
_pipelines.clear();
_profileRanges.clear();
_queries.clear();
_swapChains.clear();
_streamFormats.clear();
_textures.clear(); _textures.clear();
_textureTables.clear(); _textureTables.clear();
_streamFormats.clear();
_transforms.clear(); _transforms.clear();
_pipelines.clear();
_framebuffers.clear(); _name = nullptr;
_swapChains.clear(); _invalidModel = true;
_objects.clear(); _currentModel = Transform();
_drawCallInfos.clear(); _projectionJitter = glm::vec2(0.0f);
_enableStereo = true;
_enableSkybox = false;
} }
size_t Batch::cacheData(size_t size, const void* data) { size_t Batch::cacheData(size_t size, const void* data) {
@ -177,7 +156,6 @@ void Batch::drawIndexedInstanced(uint32 numInstances, Primitive primitiveType, u
captureDrawCallInfo(); captureDrawCallInfo();
} }
void Batch::multiDrawIndirect(uint32 numCommands, Primitive primitiveType) { void Batch::multiDrawIndirect(uint32 numCommands, Primitive primitiveType) {
ADD_COMMAND(multiDrawIndirect); ADD_COMMAND(multiDrawIndirect);
_params.emplace_back(numCommands); _params.emplace_back(numCommands);
@ -244,7 +222,6 @@ void Batch::setIndirectBuffer(const BufferPointer& buffer, Offset offset, Offset
_params.emplace_back(stride); _params.emplace_back(stride);
} }
void Batch::setModelTransform(const Transform& model) { void Batch::setModelTransform(const Transform& model) {
ADD_COMMAND(setModelTransform); ADD_COMMAND(setModelTransform);
@ -266,19 +243,19 @@ void Batch::setProjectionTransform(const Mat4& proj) {
} }
void Batch::setProjectionJitter(float jx, float jy) { void Batch::setProjectionJitter(float jx, float jy) {
_projectionJitter.x = jx; _projectionJitter.x = jx;
_projectionJitter.y = jy; _projectionJitter.y = jy;
pushProjectionJitter(jx, jy); pushProjectionJitter(jx, jy);
} }
void Batch::pushProjectionJitter(float jx, float jy) { void Batch::pushProjectionJitter(float jx, float jy) {
ADD_COMMAND(setProjectionJitter); ADD_COMMAND(setProjectionJitter);
_params.emplace_back(jx); _params.emplace_back(jx);
_params.emplace_back(jy); _params.emplace_back(jy);
} }
void Batch::popProjectionJitter() { void Batch::popProjectionJitter() {
pushProjectionJitter(_projectionJitter.x, _projectionJitter.y); pushProjectionJitter(_projectionJitter.x, _projectionJitter.y);
} }
void Batch::setViewportTransform(const Vec4i& viewport) { void Batch::setViewportTransform(const Vec4i& viewport) {
@ -374,7 +351,6 @@ void Batch::setFramebuffer(const FramebufferPointer& framebuffer) {
ADD_COMMAND(setFramebuffer); ADD_COMMAND(setFramebuffer);
_params.emplace_back(_framebuffers.cache(framebuffer)); _params.emplace_back(_framebuffers.cache(framebuffer));
} }
void Batch::setFramebufferSwapChain(const FramebufferSwapChainPointer& framebuffer, unsigned int swapChainIndex) { void Batch::setFramebufferSwapChain(const FramebufferSwapChainPointer& framebuffer, unsigned int swapChainIndex) {
@ -487,7 +463,7 @@ void Batch::runLambda(std::function<void()> f) {
void Batch::startNamedCall(const std::string& name) { void Batch::startNamedCall(const std::string& name) {
ADD_COMMAND(startNamedCall); ADD_COMMAND(startNamedCall);
_params.emplace_back(_names.cache(name)); _params.emplace_back(_names.cache(name));
_currentNamedCall = name; _currentNamedCall = name;
} }
@ -556,7 +532,7 @@ void Batch::captureDrawCallInfoImpl() {
TransformObject object; TransformObject object;
_currentModel.getMatrix(object._model); _currentModel.getMatrix(object._model);
// FIXME - we don't want to be using glm::inverse() here but it fixes the flickering issue we are // FIXME - we don't want to be using glm::inverse() here but it fixes the flickering issue we are
// seeing with planky blocks in toybox. Our implementation of getInverseMatrix() is buggy in cases // seeing with planky blocks in toybox. Our implementation of getInverseMatrix() is buggy in cases
// of non-uniform scale. We need to fix that. In the mean time, glm::inverse() works. // of non-uniform scale. We need to fix that. In the mean time, glm::inverse() works.
//_model.getInverseMatrix(_object._modelInverse); //_model.getInverseMatrix(_object._modelInverse);
@ -582,9 +558,9 @@ void Batch::captureDrawCallInfo() {
} }
void Batch::captureNamedDrawCallInfo(std::string name) { void Batch::captureNamedDrawCallInfo(std::string name) {
std::swap(_currentNamedCall, name); // Set and save _currentNamedCall std::swap(_currentNamedCall, name); // Set and save _currentNamedCall
captureDrawCallInfoImpl(); captureDrawCallInfoImpl();
std::swap(_currentNamedCall, name); // Restore _currentNamedCall std::swap(_currentNamedCall, name); // Restore _currentNamedCall
} }
// Debugging // Debugging

10
libraries/gpu/src/gpu/Batch.h
View file

@ -92,9 +92,12 @@ public:
void captureNamedDrawCallInfo(std::string name); void captureNamedDrawCallInfo(std::string name);
Batch(const char* name = nullptr); Batch(const char* name = nullptr);
Batch(const Batch& batch); // Disallow copy construction and assignement of batches
Batch(const Batch& batch) = delete;
Batch& operator=(const Batch& batch) = delete;
~Batch(); ~Batch();
void setName(const char* name);
void clear(); void clear();
// Batches may need to override the context level stereo settings // Batches may need to override the context level stereo settings
@ -506,10 +509,7 @@ public:
bool _enableSkybox { false }; bool _enableSkybox { false };
protected: protected:
const char* _name;
#ifdef DEBUG
std::string _name;
#endif
friend class Context; friend class Context;
friend class Frame; friend class Frame;

40
libraries/gpu/src/gpu/Context.cpp
View file

@ -47,6 +47,10 @@ Context::Context(const Context& context) {
} }
Context::~Context() { Context::~Context() {
for (auto batch : _batchPool) {
delete batch;
}
_batchPool.clear();
} }
const std::string& Context::getBackendVersion() const { const std::string& Context::getBackendVersion() const {
@ -65,7 +69,7 @@ void Context::beginFrame(const glm::mat4& renderView, const glm::mat4& renderPos
} }
} }
void Context::appendFrameBatch(Batch& batch) { void Context::appendFrameBatch(const BatchPointer& batch) {
if (!_frameActive) { if (!_frameActive) {
qWarning() << "Batch executed outside of frame boundaries"; qWarning() << "Batch executed outside of frame boundaries";
return; return;
@ -115,7 +119,7 @@ void Context::executeFrame(const FramePointer& frame) const {
// Execute the frame rendering commands // Execute the frame rendering commands
for (auto& batch : frame->batches) { for (auto& batch : frame->batches) {
_backend->render(batch); _backend->render(*batch);
} }
Batch endBatch("Context::executeFrame::end"); Batch endBatch("Context::executeFrame::end");
@ -323,6 +327,7 @@ Size Context::getTextureExternalGPUMemSize() {
uint32_t Context::getTexturePendingGPUTransferCount() { uint32_t Context::getTexturePendingGPUTransferCount() {
return Backend::texturePendingGPUTransferCount.getValue(); return Backend::texturePendingGPUTransferCount.getValue();
} }
Size Context::getTexturePendingGPUTransferMemSize() { Size Context::getTexturePendingGPUTransferMemSize() {
return Backend::texturePendingGPUTransferMemSize.getValue(); return Backend::texturePendingGPUTransferMemSize.getValue();
} }
@ -334,3 +339,34 @@ Size Context::getTextureResourcePopulatedGPUMemSize() {
Size Context::getTextureResourceIdealGPUMemSize() { Size Context::getTextureResourceIdealGPUMemSize() {
return Backend::textureResourceIdealGPUMemSize.getValue(); return Backend::textureResourceIdealGPUMemSize.getValue();
} }
BatchPointer Context::acquireBatch(const char* name) {
Batch* rawBatch = nullptr;
{
Lock lock(_batchPoolMutex);
if (!_batchPool.empty()) {
rawBatch = _batchPool.front();
_batchPool.pop_front();
}
}
if (!rawBatch) {
rawBatch = new Batch();
}
rawBatch->setName(name);
return BatchPointer(rawBatch, [this](Batch* batch) { releaseBatch(batch); });
}
void Context::releaseBatch(Batch* batch) {
batch->clear();
Lock lock(_batchPoolMutex);
_batchPool.push_back(batch);
}
void gpu::doInBatch(const char* name,
const std::shared_ptr<gpu::Context>& context,
const std::function<void(Batch& batch)>& f) {
auto batch = context->acquireBatch(name);
f(*batch);
context->appendFrameBatch(batch);
}

77
libraries/gpu/src/gpu/Context.h
View file

@ -43,17 +43,16 @@ public:
int _DSNumTriangles = 0; int _DSNumTriangles = 0;
int _PSNumSetPipelines = 0; int _PSNumSetPipelines = 0;
ContextStats() {} ContextStats() {}
ContextStats(const ContextStats& stats) = default; ContextStats(const ContextStats& stats) = default;
void evalDelta(const ContextStats& begin, const ContextStats& end); void evalDelta(const ContextStats& begin, const ContextStats& end);
}; };
class Backend { class Backend {
public: public:
virtual~ Backend() {}; virtual ~Backend(){};
virtual const std::string& getVersion() const = 0; virtual const std::string& getVersion() const = 0;
@ -78,12 +77,11 @@ public:
TransformCamera getEyeCamera(int eye, const StereoState& stereo, const Transform& xformView, Vec2 normalizedJitter) const; TransformCamera getEyeCamera(int eye, const StereoState& stereo, const Transform& xformView, Vec2 normalizedJitter) const;
}; };
template <typename T, typename U>
template<typename T, typename U>
static void setGPUObject(const U& object, T* gpuObject) { static void setGPUObject(const U& object, T* gpuObject) {
object.gpuObject.setGPUObject(gpuObject); object.gpuObject.setGPUObject(gpuObject);
} }
template<typename T, typename U> template <typename T, typename U>
static T* getGPUObject(const U& object) { static T* getGPUObject(const U& object) {
return reinterpret_cast<T*>(object.gpuObject.getGPUObject()); return reinterpret_cast<T*>(object.gpuObject.getGPUObject());
} }
@ -95,26 +93,25 @@ public:
// These should only be accessed by Backend implementation to report the buffer and texture allocations, // These should only be accessed by Backend implementation to report the buffer and texture allocations,
// they are NOT public objects // they are NOT public objects
static ContextMetricSize freeGPUMemSize; static ContextMetricSize freeGPUMemSize;
static ContextMetricCount bufferCount; static ContextMetricCount bufferCount;
static ContextMetricSize bufferGPUMemSize; static ContextMetricSize bufferGPUMemSize;
static ContextMetricCount textureResidentCount; static ContextMetricCount textureResidentCount;
static ContextMetricCount textureFramebufferCount; static ContextMetricCount textureFramebufferCount;
static ContextMetricCount textureResourceCount; static ContextMetricCount textureResourceCount;
static ContextMetricCount textureExternalCount; static ContextMetricCount textureExternalCount;
static ContextMetricSize textureResidentGPUMemSize; static ContextMetricSize textureResidentGPUMemSize;
static ContextMetricSize textureFramebufferGPUMemSize; static ContextMetricSize textureFramebufferGPUMemSize;
static ContextMetricSize textureResourceGPUMemSize; static ContextMetricSize textureResourceGPUMemSize;
static ContextMetricSize textureExternalGPUMemSize; static ContextMetricSize textureExternalGPUMemSize;
static ContextMetricCount texturePendingGPUTransferCount; static ContextMetricCount texturePendingGPUTransferCount;
static ContextMetricSize texturePendingGPUTransferMemSize; static ContextMetricSize texturePendingGPUTransferMemSize;
static ContextMetricSize textureResourcePopulatedGPUMemSize; static ContextMetricSize textureResourcePopulatedGPUMemSize;
static ContextMetricSize textureResourceIdealGPUMemSize; static ContextMetricSize textureResourceIdealGPUMemSize;
protected: protected:
virtual bool isStereo() { virtual bool isStereo() {
@ -144,7 +141,6 @@ public:
typedef BackendPointer (*CreateBackend)(); typedef BackendPointer (*CreateBackend)();
typedef bool (*MakeProgram)(Shader& shader, const Shader::BindingSet& bindings, const Shader::CompilationHandler& handler); typedef bool (*MakeProgram)(Shader& shader, const Shader::BindingSet& bindings, const Shader::CompilationHandler& handler);
// This one call must happen before any context is created or used (Shader::MakeProgram) in order to setup the Backend and any singleton data needed // This one call must happen before any context is created or used (Shader::MakeProgram) in order to setup the Backend and any singleton data needed
template <class T> template <class T>
static void init() { static void init() {
@ -161,40 +157,43 @@ public:
const std::string& getBackendVersion() const; const std::string& getBackendVersion() const;
void beginFrame(const glm::mat4& renderView = glm::mat4(), const glm::mat4& renderPose = glm::mat4()); void beginFrame(const glm::mat4& renderView = glm::mat4(), const glm::mat4& renderPose = glm::mat4());
void appendFrameBatch(Batch& batch); void appendFrameBatch(const BatchPointer& batch);
FramePointer endFrame(); FramePointer endFrame();
BatchPointer acquireBatch(const char* name = nullptr);
void releaseBatch(Batch* batch);
// MUST only be called on the rendering thread // MUST only be called on the rendering thread
// //
// Handle any pending operations to clean up (recycle / deallocate) resources no longer in use // Handle any pending operations to clean up (recycle / deallocate) resources no longer in use
void recycle() const; void recycle() const;
// MUST only be called on the rendering thread // MUST only be called on the rendering thread
// //
// Execute a batch immediately, rather than as part of a frame // Execute a batch immediately, rather than as part of a frame
void executeBatch(Batch& batch) const; void executeBatch(Batch& batch) const;
// MUST only be called on the rendering thread // MUST only be called on the rendering thread
// //
// Executes a frame, applying any updates contained in the frame batches to the rendering // Executes a frame, applying any updates contained in the frame batches to the rendering
// thread shadow copies. Either executeFrame or consumeFrameUpdates MUST be called on every frame // thread shadow copies. Either executeFrame or consumeFrameUpdates MUST be called on every frame
// generated, IN THE ORDER they were generated. // generated, IN THE ORDER they were generated.
void executeFrame(const FramePointer& frame) const; void executeFrame(const FramePointer& frame) const;
// MUST only be called on the rendering thread. // MUST only be called on the rendering thread.
// //
// Consuming a frame applies any updates queued from the recording thread and applies them to the // Consuming a frame applies any updates queued from the recording thread and applies them to the
// shadow copy used by the rendering thread. // shadow copy used by the rendering thread.
// //
// EVERY frame generated MUST be consumed, regardless of whether the frame is actually executed, // EVERY frame generated MUST be consumed, regardless of whether the frame is actually executed,
// or the buffer shadow copies can become unsynced from the recording thread copies. // or the buffer shadow copies can become unsynced from the recording thread copies.
// //
// Consuming a frame is idempotent, as the frame encapsulates the updates and clears them out as // Consuming a frame is idempotent, as the frame encapsulates the updates and clears them out as
// it applies them, so calling it more than once on a given frame will have no effect after the // it applies them, so calling it more than once on a given frame will have no effect after the
// first time // first time
// //
// //
// This is automatically called by executeFrame, so you only need to call it if you // This is automatically called by executeFrame, so you only need to call it if you
// have frames you aren't going to otherwise execute, for instance when a display plugin is // have frames you aren't going to otherwise execute, for instance when a display plugin is
// being disabled, or in the null display plugin where no rendering actually occurs // being disabled, or in the null display plugin where no rendering actually occurs
void consumeFrameUpdates(const FramePointer& frame) const; void consumeFrameUpdates(const FramePointer& frame) const;
@ -212,7 +211,7 @@ public:
// It s here for convenience to easily capture a snapshot // It s here for convenience to easily capture a snapshot
void downloadFramebuffer(const FramebufferPointer& srcFramebuffer, const Vec4i& region, QImage& destImage); void downloadFramebuffer(const FramebufferPointer& srcFramebuffer, const Vec4i& region, QImage& destImage);
// Repporting stats of the context // Repporting stats of the context
void resetStats() const; void resetStats() const;
void getStats(ContextStats& stats) const; void getStats(ContextStats& stats) const;
@ -237,7 +236,7 @@ public:
static Size getTextureGPUMemSize(); static Size getTextureGPUMemSize();
static Size getTextureResidentGPUMemSize(); static Size getTextureResidentGPUMemSize();
static Size getTextureFramebufferGPUMemSize(); static Size getTextureFramebufferGPUMemSize();
static Size getTextureResourceGPUMemSize(); static Size getTextureResourceGPUMemSize();
static Size getTextureExternalGPUMemSize(); static Size getTextureExternalGPUMemSize();
static uint32_t getTexturePendingGPUTransferCount(); static uint32_t getTexturePendingGPUTransferCount();
@ -250,10 +249,12 @@ protected:
Context(const Context& context); Context(const Context& context);
std::shared_ptr<Backend> _backend; std::shared_ptr<Backend> _backend;
bool _frameActive { false }; std::mutex _batchPoolMutex;
std::list<Batch*> _batchPool;
bool _frameActive{ false };
FramePointer _currentFrame; FramePointer _currentFrame;
RangeTimerPointer _frameRangeTimer; RangeTimerPointer _frameRangeTimer;
StereoState _stereo; StereoState _stereo;
// Sampled at the end of every frame, the stats of all the counters // Sampled at the end of every frame, the stats of all the counters
mutable ContextStats _frameStats; mutable ContextStats _frameStats;
@ -261,7 +262,7 @@ protected:
// This function can only be called by "static Shader::makeProgram()" // This function can only be called by "static Shader::makeProgram()"
// makeProgramShader(...) make a program shader ready to be used in a Batch. // makeProgramShader(...) make a program shader ready to be used in a Batch.
// It compiles the sub shaders, link them and defines the Slots and their bindings. // It compiles the sub shaders, link them and defines the Slots and their bindings.
// If the shader passed is not a program, nothing happens. // If the shader passed is not a program, nothing happens.
static bool makeProgram(Shader& shader, const Shader::BindingSet& bindings, const Shader::CompilationHandler& handler); static bool makeProgram(Shader& shader, const Shader::BindingSet& bindings, const Shader::CompilationHandler& handler);
static CreateBackend _createBackendCallback; static CreateBackend _createBackendCallback;
@ -273,14 +274,8 @@ protected:
}; };
typedef std::shared_ptr<Context> ContextPointer; typedef std::shared_ptr<Context> ContextPointer;
template<typename F> void doInBatch(const char* name, const std::shared_ptr<gpu::Context>& context, const std::function<void(Batch& batch)>& f);
void doInBatch(const char* name, std::shared_ptr<gpu::Context> context, F f) {
gpu::Batch batch(name);
f(batch);
context->appendFrameBatch(batch);
}
};
}; // namespace gpu
#endif #endif

1
libraries/gpu/src/gpu/Forward.h
View file

@ -21,6 +21,7 @@ namespace gpu {
using Lock = std::unique_lock<Mutex>; using Lock = std::unique_lock<Mutex>;
class Batch; class Batch;
using BatchPointer = std::shared_ptr<Batch>;
class Backend; class Backend;
using BackendPointer = std::shared_ptr<Backend>; using BackendPointer = std::shared_ptr<Backend>;
class Context; class Context;

4
libraries/gpu/src/gpu/Frame.cpp
View file

@ -28,8 +28,8 @@ Frame::~Frame() {
} }
void Frame::finish() { void Frame::finish() {
for (Batch& batch : batches) { for (const auto& batch : batches) {
batch.finishFrame(bufferUpdates); batch->finishFrame(bufferUpdates);
} }
} }

2
libraries/gpu/src/gpu/Frame.h
View file

@ -22,7 +22,7 @@ namespace gpu {
public: public:
Frame(); Frame();
virtual ~Frame(); virtual ~Frame();
using Batches = std::vector<Batch>; using Batches = std::vector<BatchPointer>;
using FramebufferRecycler = std::function<void(const FramebufferPointer&)>; using FramebufferRecycler = std::function<void(const FramebufferPointer&)>;
using OverlayRecycler = std::function<void(const TexturePointer&)>; using OverlayRecycler = std::function<void(const TexturePointer&)>;

17
libraries/render-utils/src/DeferredLightingEffect.cpp
View file

@ -738,21 +738,20 @@ void RenderDeferred::run(const RenderContextPointer& renderContext, const Inputs
} }
auto previousBatch = args->_batch; auto previousBatch = args->_batch;
gpu::Batch batch; gpu::doInBatch(nullptr, args->_context, [&](gpu::Batch& batch) {
args->_batch = &batch; args->_batch = &batch;
_gpuTimer->begin(batch); _gpuTimer->begin(batch);
setupJob.run(renderContext, deferredTransform, deferredFramebuffer, lightingModel, haze, surfaceGeometryFramebuffer, ssaoFramebuffer, subsurfaceScatteringResource); setupJob.run(renderContext, deferredTransform, deferredFramebuffer, lightingModel, haze, surfaceGeometryFramebuffer, ssaoFramebuffer, subsurfaceScatteringResource);
lightsJob.run(renderContext, deferredTransform, deferredFramebuffer, lightingModel, surfaceGeometryFramebuffer, lightClusters); lightsJob.run(renderContext, deferredTransform, deferredFramebuffer, lightingModel, surfaceGeometryFramebuffer, lightClusters);
cleanupJob.run(renderContext); cleanupJob.run(renderContext);
_gpuTimer->end(batch); _gpuTimer->end(batch);
args->_context->appendFrameBatch(batch); });
args->_batch = previousBatch; args->_batch = previousBatch;
auto config = std::static_pointer_cast<Config>(renderContext->jobConfig); auto config = std::static_pointer_cast<Config>(renderContext->jobConfig);
config->setGPUBatchRunTime(_gpuTimer->getGPUAverage(), _gpuTimer->getBatchAverage()); config->setGPUBatchRunTime(_gpuTimer->getGPUAverage(), _gpuTimer->getBatchAverage());
} }

126
libraries/render-utils/src/LightClusters.cpp
View file

@ -707,85 +707,83 @@ void DebugLightClusters::run(const render::RenderContextPointer& renderContext,
auto args = renderContext->args; auto args = renderContext->args;
gpu::Batch batch;
gpu::doInBatch(nullptr, args->_context, [&](gpu::Batch& batch) {
batch.enableStereo(false);
batch.enableStereo(false); // Assign the camera transform
batch.setViewportTransform(args->_viewport);
glm::mat4 projMat;
Transform viewMat;
args->getViewFrustum().evalProjectionMatrix(projMat);
args->getViewFrustum().evalViewTransform(viewMat);
batch.setProjectionTransform(projMat);
batch.setViewTransform(viewMat, true);
// Assign the camera transform // Then the actual ClusterGrid attributes
batch.setViewportTransform(args->_viewport); batch.setModelTransform(Transform());
glm::mat4 projMat;
Transform viewMat;
args->getViewFrustum().evalProjectionMatrix(projMat);
args->getViewFrustum().evalViewTransform(viewMat);
batch.setProjectionTransform(projMat);
batch.setViewTransform(viewMat, true);
// Bind the Light CLuster data strucutre
// Then the actual ClusterGrid attributes batch.setUniformBuffer(LIGHT_GPU_SLOT, lightClusters->_lightStage->getLightArrayBuffer());
batch.setModelTransform(Transform()); batch.setUniformBuffer(LIGHT_CLUSTER_GRID_FRUSTUM_GRID_SLOT, lightClusters->_frustumGridBuffer);
batch.setUniformBuffer(LIGHT_CLUSTER_GRID_CLUSTER_GRID_SLOT, lightClusters->_clusterGridBuffer);
// Bind the Light CLuster data strucutre batch.setUniformBuffer(LIGHT_CLUSTER_GRID_CLUSTER_CONTENT_SLOT, lightClusters->_clusterContentBuffer);
batch.setUniformBuffer(LIGHT_GPU_SLOT, lightClusters->_lightStage->getLightArrayBuffer());
batch.setUniformBuffer(LIGHT_CLUSTER_GRID_FRUSTUM_GRID_SLOT, lightClusters->_frustumGridBuffer);
batch.setUniformBuffer(LIGHT_CLUSTER_GRID_CLUSTER_GRID_SLOT, lightClusters->_clusterGridBuffer);
batch.setUniformBuffer(LIGHT_CLUSTER_GRID_CLUSTER_CONTENT_SLOT, lightClusters->_clusterContentBuffer);
if (doDrawClusterFromDepth) { if (doDrawClusterFromDepth) {
batch.setPipeline(getDrawClusterFromDepthPipeline()); batch.setPipeline(getDrawClusterFromDepthPipeline());
batch.setUniformBuffer(DEFERRED_FRAME_TRANSFORM_BUFFER_SLOT, deferredTransform->getFrameTransformBuffer()); batch.setUniformBuffer(DEFERRED_FRAME_TRANSFORM_BUFFER_SLOT, deferredTransform->getFrameTransformBuffer());
if (linearDepthTarget) { if (linearDepthTarget) {
batch.setResourceTexture(DEFERRED_BUFFER_LINEAR_DEPTH_UNIT, linearDepthTarget->getLinearDepthTexture()); batch.setResourceTexture(DEFERRED_BUFFER_LINEAR_DEPTH_UNIT, linearDepthTarget->getLinearDepthTexture());
}
batch.draw(gpu::TRIANGLE_STRIP, 4, 0);
batch.setResourceTexture(DEFERRED_BUFFER_LINEAR_DEPTH_UNIT, nullptr);
batch.setUniformBuffer(DEFERRED_FRAME_TRANSFORM_BUFFER_SLOT, nullptr);
} }
batch.draw(gpu::TRIANGLE_STRIP, 4, 0); if (doDrawContent) {
// bind the one gpu::Pipeline we need
batch.setPipeline(getDrawClusterContentPipeline());
batch.setUniformBuffer(DEFERRED_FRAME_TRANSFORM_BUFFER_SLOT, deferredTransform->getFrameTransformBuffer());
if (linearDepthTarget) {
batch.setResourceTexture(DEFERRED_BUFFER_LINEAR_DEPTH_UNIT, linearDepthTarget->getLinearDepthTexture());
}
batch.draw(gpu::TRIANGLE_STRIP, 4, 0);
batch.setResourceTexture(DEFERRED_BUFFER_LINEAR_DEPTH_UNIT, nullptr); batch.setResourceTexture(DEFERRED_BUFFER_LINEAR_DEPTH_UNIT, nullptr);
batch.setUniformBuffer(DEFERRED_FRAME_TRANSFORM_BUFFER_SLOT, nullptr); batch.setUniformBuffer(DEFERRED_FRAME_TRANSFORM_BUFFER_SLOT, nullptr);
} }
});
if (doDrawContent) {
// bind the one gpu::Pipeline we need
batch.setPipeline(getDrawClusterContentPipeline());
batch.setUniformBuffer(DEFERRED_FRAME_TRANSFORM_BUFFER_SLOT, deferredTransform->getFrameTransformBuffer());
if (linearDepthTarget) { gpu::doInBatch(nullptr, args->_context, [&](gpu::Batch& batch) {
batch.setResourceTexture(DEFERRED_BUFFER_LINEAR_DEPTH_UNIT, linearDepthTarget->getLinearDepthTexture()); auto& drawGridAndCleanBatch = batch;
if (doDrawGrid) {
// bind the one gpu::Pipeline we need
drawGridAndCleanBatch.setPipeline(getDrawClusterGridPipeline());
auto dims = lightClusters->_frustumGridBuffer->dims;
glm::ivec3 summedDims(dims.x * dims.y * dims.z, dims.x * dims.y, dims.x);
drawGridAndCleanBatch.drawInstanced(summedDims.x, gpu::LINES, 24, 0);
} }
batch.draw(gpu::TRIANGLE_STRIP, 4, 0); drawGridAndCleanBatch.setUniformBuffer(LIGHT_GPU_SLOT, nullptr);
drawGridAndCleanBatch.setUniformBuffer(LIGHT_CLUSTER_GRID_FRUSTUM_GRID_SLOT, nullptr);
batch.setResourceTexture(DEFERRED_BUFFER_LINEAR_DEPTH_UNIT, nullptr); drawGridAndCleanBatch.setUniformBuffer(LIGHT_CLUSTER_GRID_CLUSTER_GRID_SLOT, nullptr);
batch.setUniformBuffer(DEFERRED_FRAME_TRANSFORM_BUFFER_SLOT, nullptr); drawGridAndCleanBatch.setUniformBuffer(LIGHT_CLUSTER_GRID_CLUSTER_CONTENT_SLOT, nullptr);
}
drawGridAndCleanBatch.setResourceTexture(DEFERRED_BUFFER_COLOR_UNIT, nullptr);
drawGridAndCleanBatch.setResourceTexture(DEFERRED_BUFFER_NORMAL_UNIT, nullptr);
gpu::Batch drawGridAndCleanBatch; drawGridAndCleanBatch.setResourceTexture(DEFERRED_BUFFER_EMISSIVE_UNIT, nullptr);
drawGridAndCleanBatch.setResourceTexture(DEFERRED_BUFFER_DEPTH_UNIT, nullptr);
if (doDrawGrid) { });
// bind the one gpu::Pipeline we need
drawGridAndCleanBatch.setPipeline(getDrawClusterGridPipeline());
auto dims = lightClusters->_frustumGridBuffer->dims;
glm::ivec3 summedDims(dims.x*dims.y * dims.z, dims.x*dims.y, dims.x);
drawGridAndCleanBatch.drawInstanced(summedDims.x, gpu::LINES, 24, 0);
}
drawGridAndCleanBatch.setUniformBuffer(LIGHT_GPU_SLOT, nullptr);
drawGridAndCleanBatch.setUniformBuffer(LIGHT_CLUSTER_GRID_FRUSTUM_GRID_SLOT, nullptr);
drawGridAndCleanBatch.setUniformBuffer(LIGHT_CLUSTER_GRID_CLUSTER_GRID_SLOT, nullptr);
drawGridAndCleanBatch.setUniformBuffer(LIGHT_CLUSTER_GRID_CLUSTER_CONTENT_SLOT, nullptr);
drawGridAndCleanBatch.setResourceTexture(DEFERRED_BUFFER_COLOR_UNIT, nullptr);
drawGridAndCleanBatch.setResourceTexture(DEFERRED_BUFFER_NORMAL_UNIT, nullptr);
drawGridAndCleanBatch.setResourceTexture(DEFERRED_BUFFER_EMISSIVE_UNIT, nullptr);
drawGridAndCleanBatch.setResourceTexture(DEFERRED_BUFFER_DEPTH_UNIT, nullptr);
args->_context->appendFrameBatch(batch);
args->_context->appendFrameBatch(drawGridAndCleanBatch);
} }