More shared pointer deduplication

This commit is contained in:
Brad Davis 2018-09-06 16:03:43 -07:00
parent 9a435cdca2
commit cfff28ad0a
18 changed files with 330 additions and 182 deletions

View file

@ -150,6 +150,7 @@ void GLBackend::init() {
});
}
GLBackend::GLBackend(bool syncCache) {
_pipeline._cameraCorrectionBuffer._buffer->flush();
initShaderBinaryCache();
@ -201,9 +202,10 @@ void GLBackend::renderPassTransfer(const Batch& batch) {
{
Vec2u outputSize{ 1,1 };
if (_output._framebuffer) {
outputSize.x = _output._framebuffer->getWidth();
outputSize.y = _output._framebuffer->getHeight();
auto framebuffer = acquire(_output._framebuffer);
if (framebuffer) {
outputSize.x = framebuffer->getWidth();
outputSize.y = framebuffer->getHeight();
} else if (glm::dot(_transform._projectionJitter, _transform._projectionJitter)>0.0f) {
qCWarning(gpugllogging) << "Jittering needs to have a frame buffer to be set";
}
@ -220,6 +222,7 @@ void GLBackend::renderPassTransfer(const Batch& batch) {
_stereo._contextDisable = false;
break;
case Batch::COMMAND_setFramebuffer:
case Batch::COMMAND_setViewportTransform:
case Batch::COMMAND_setViewTransform:
case Batch::COMMAND_setProjectionTransform:

View file

@ -30,7 +30,6 @@
#include "GLShared.h"
// Different versions for the stereo drawcall
// Current preferred is "instanced" which draw the shape twice but instanced and rely on clipping plane to draw left/right side only
#if defined(USE_GLES)
@ -40,7 +39,6 @@
#define GPU_STEREO_TECHNIQUE_INSTANCED
#endif
// Let these be configured by the one define picked above
#ifdef GPU_STEREO_TECHNIQUE_DOUBLED_SIMPLE
#define GPU_STEREO_DRAWCALL_DOUBLED
@ -56,8 +54,151 @@
#define GPU_STEREO_CAMERA_BUFFER
#endif
//
// GL Backend pointer storage mechanism
// One of the following three defines must be uncommented.
// Equivalent to current state of affairs in master,
// Works pretty well, but ends up creating a lot of needless smart pointer duplication
// which means there's a high aggregate cost of manipulating std::shared_ptr counters
//#define GPU_POINTER_STORAGE_SHARED
// The platonic ideal, use references to smart pointers.
// However, this produces artifacts because there are too many places in the code right now that
// create temporary values (undesirable smart pointer duplications) and then those temp variables
// get passed on and have their reference taken, and then invalidated
//#define GPU_POINTER_STORAGE_REF
// Raw pointer manipulation. Seems more dangerous than the reference wrappers,
// but in practice, the danger of grabbing a reference to a temporary variable
// is causing issues
#define GPU_POINTER_STORAGE_RAW
namespace gpu { namespace gl {
#if defined(GPU_POINTER_STORAGE_SHARED)
template <typename T>
static inline bool compare(const std::shared_ptr<T>& a, const std::shared_ptr<T>& b) {
return a == b;
}
template <typename T>
static inline T* acquire(const std::shared_ptr<T>& pointer) {
return pointer.get();
}
template <typename T>
static inline void reset(std::shared_ptr<T>& pointer) {
return pointer.reset();
}
template <typename T>
static inline bool valid(const std::shared_ptr<T>& pointer) {
return pointer.operator bool();
}
template <typename T>
static inline void assign(std::shared_ptr<T>&pointer, const std::shared_ptr<T>& source) {
pointer = source;
}
using BufferReference = BufferPointer;
using TextureReference = TexturePointer;
using FramebufferReference = FramebufferPointer;
using FormatReference = Stream::FormatPointer;
using PipelineReference = PipelinePointer;
#define GPU_REFERENCE_INIT_VALUE nullptr
#elif defined(GPU_POINTER_STORAGE_REF)
template <typename T>
class PointerReferenceWrapper : public std::reference_wrapper<const std::shared_ptr<T>> {
using Parent = std::reference_wrapper<const std::shared_ptr<T>>;
public:
using Pointer = std::shared_ptr<T>;
PointerReferenceWrapper() : Parent(EMPTY()) {}
PointerReferenceWrapper(const Pointer& pointer) : Parent(pointer) {}
void clear() { *this = EMPTY(); }
private:
static const Pointer& EMPTY() {
static const Pointer EMPTY_VALUE;
return EMPTY_VALUE;
};
};
template<typename T>
static bool compare(const PointerReferenceWrapper<T>& reference, const std::shared_ptr<T>& pointer) {
return reference.get() == pointer;
}
template <typename T>
static inline T* acquire(const PointerReferenceWrapper<T>& reference) {
return reference.get().get();
}
template<typename T>
static void assign(PointerReferenceWrapper<T>& reference, const std::shared_ptr<T>& pointer) {
reference = pointer;
}
template<typename T>
static bool valid(const PointerReferenceWrapper<T>& reference) {
return reference.get().operator bool();
}
template <typename T>
static inline void reset(PointerReferenceWrapper<T>& reference) {
return reference.clear();
}
using BufferReference = PointerReferenceWrapper<Buffer>;
using TextureReference = PointerReferenceWrapper<Texture>;
using FramebufferReference = PointerReferenceWrapper<Framebuffer>;
using FormatReference = PointerReferenceWrapper<Stream::Format>;
using PipelineReference = PointerReferenceWrapper<Pipeline>;
#define GPU_REFERENCE_INIT_VALUE
#elif defined(GPU_POINTER_STORAGE_RAW)
template<typename T>
static bool compare(const T*const& rawPointer, const std::shared_ptr<T>& pointer) {
return rawPointer == pointer.get();
}
template <typename T>
static inline T* acquire(T*& rawPointer) {
return rawPointer;
}
template <typename T>
static inline bool valid(const T*const& rawPointer) {
return rawPointer;
}
template <typename T>
static inline void reset(T*& rawPointer) {
rawPointer = nullptr;
}
template <typename T>
static inline void assign(T*& rawPointer, const std::shared_ptr<T>& pointer) {
rawPointer = pointer.get();
}
using BufferReference = Buffer*;
using TextureReference = Texture*;
using FramebufferReference = Framebuffer*;
using FormatReference = Stream::Format*;
using PipelineReference = Pipeline*;
#define GPU_REFERENCE_INIT_VALUE nullptr
#endif
class GLBackend : public Backend, public std::enable_shared_from_this<GLBackend> {
// Context Backend static interface required
friend class gpu::Context;
@ -67,8 +208,9 @@ class GLBackend : public Backend, public std::enable_shared_from_this<GLBackend>
protected:
explicit GLBackend(bool syncCache);
GLBackend();
public:
public:
#if defined(USE_GLES)
// https://www.khronos.org/registry/OpenGL-Refpages/es3/html/glGet.xhtml
static const GLint MIN_REQUIRED_TEXTURE_IMAGE_UNITS = 16;
@ -109,8 +251,8 @@ public:
// This is the ugly "download the pixels to sysmem for taking a snapshot"
// Just avoid using it, it's ugly and will break performances
virtual void downloadFramebuffer(const FramebufferPointer& srcFramebuffer,
const Vec4i& region, QImage& destImage) final override;
const Vec4i& region,
QImage& destImage) final override;
// this is the maximum numeber of available input buffers
size_t getNumInputBuffers() const { return _input._invalidBuffers.size(); }
@ -131,7 +273,6 @@ public:
static const int MAX_NUM_RESOURCE_TABLE_TEXTURES = 2;
size_t getMaxNumResourceTextureTables() const { return MAX_NUM_RESOURCE_TABLE_TEXTURES; }
// Draw Stage
virtual void do_draw(const Batch& batch, size_t paramOffset) = 0;
virtual void do_drawIndexed(const Batch& batch, size_t paramOffset) = 0;
@ -183,7 +324,6 @@ public:
// Reset stages
virtual void do_resetStages(const Batch& batch, size_t paramOffset) final;
virtual void do_disableContextViewCorrection(const Batch& batch, size_t paramOffset) final;
virtual void do_restoreContextViewCorrection(const Batch& batch, size_t paramOffset) final;
@ -203,7 +343,7 @@ public:
virtual void do_popProfileRange(const Batch& batch, size_t paramOffset) final;
// TODO: As long as we have gl calls explicitely issued from interface
// code, we need to be able to record and batch these calls. THe long
// code, we need to be able to record and batch these calls. THe long
// term strategy is to get rid of any GL calls in favor of the HIFI GPU API
virtual void do_glUniform1i(const Batch& batch, size_t paramOffset) final;
virtual void do_glUniform1f(const Batch& batch, size_t paramOffset) final;
@ -228,7 +368,9 @@ public:
virtual void do_setStateAntialiasedLineEnable(bool enable) final;
virtual void do_setStateDepthBias(Vec2 bias) final;
virtual void do_setStateDepthTest(State::DepthTest test) final;
virtual void do_setStateStencil(State::StencilActivation activation, State::StencilTest frontTest, State::StencilTest backTest) final;
virtual void do_setStateStencil(State::StencilActivation activation,
State::StencilTest frontTest,
State::StencilTest backTest) final;
virtual void do_setStateAlphaToCoverageEnable(bool enable) final;
virtual void do_setStateSampleMask(uint32 mask) final;
virtual void do_setStateBlend(State::BlendFunction blendFunction) final;
@ -257,7 +399,9 @@ public:
virtual void releaseQuery(GLuint id) const;
virtual void queueLambda(const std::function<void()> lambda) const;
bool isTextureManagementSparseEnabled() const override { return (_textureManagement._sparseCapable && Texture::getEnableSparseTextures()); }
bool isTextureManagementSparseEnabled() const override {
return (_textureManagement._sparseCapable && Texture::getEnableSparseTextures());
}
protected:
virtual GLint getRealUniformLocation(GLint location) const;
@ -266,11 +410,11 @@ protected:
// FIXME instead of a single flag, create a features struct similar to
// https://www.khronos.org/registry/vulkan/specs/1.0/man/html/VkPhysicalDeviceFeatures.html
virtual bool supportsBindless() const { return false; }
virtual bool supportsBindless() const { return false; }
static const size_t INVALID_OFFSET = (size_t)-1;
bool _inRenderTransferPass { false };
int _currentDraw { -1 };
bool _inRenderTransferPass{ false };
int _currentDraw{ -1 };
std::list<std::string> profileRanges;
mutable Mutex _trashMutex;
@ -299,46 +443,42 @@ protected:
virtual void updateInput() = 0;
struct InputStageState {
bool _invalidFormat { true };
bool _lastUpdateStereoState{ false };
bool _invalidFormat{ true };
bool _lastUpdateStereoState{ false };
bool _hadColorAttribute{ true };
Stream::FormatPointer _format;
FormatReference _format{ GPU_REFERENCE_INIT_VALUE };
std::string _formatKey;
typedef std::bitset<MAX_NUM_ATTRIBUTES> ActivationCache;
ActivationCache _attributeActivation { 0 };
ActivationCache _attributeActivation{ 0 };
typedef std::bitset<MAX_NUM_INPUT_BUFFERS> BuffersState;
BuffersState _invalidBuffers{ 0 };
BuffersState _attribBindingBuffers{ 0 };
Buffers _buffers;
Offsets _bufferOffsets;
Offsets _bufferStrides;
std::vector<GLuint> _bufferVBOs;
std::vector<BufferReference> _buffers{ MAX_NUM_INPUT_BUFFERS, GPU_REFERENCE_INIT_VALUE };
Offsets _bufferOffsets;
Offsets _bufferStrides;
std::vector<GLuint> _bufferVBOs;
glm::vec4 _colorAttribute{ 0.0f };
BufferPointer _indexBuffer;
Offset _indexBufferOffset { 0 };
Type _indexBufferType { UINT32 };
BufferPointer _indirectBuffer;
BufferReference _indexBuffer{ GPU_REFERENCE_INIT_VALUE };
Offset _indexBufferOffset{ 0 };
Type _indexBufferType{ UINT32 };
BufferReference _indirectBuffer{ GPU_REFERENCE_INIT_VALUE };
Offset _indirectBufferOffset{ 0 };
Offset _indirectBufferStride{ 0 };
GLuint _defaultVAO { 0 };
GLuint _defaultVAO{ 0 };
InputStageState() :
_invalidFormat(true),
_format(0),
_formatKey(),
_attributeActivation(0),
_buffers(_invalidBuffers.size(), BufferPointer(0)),
_bufferOffsets(_invalidBuffers.size(), 0),
_bufferStrides(_invalidBuffers.size(), 0),
_bufferVBOs(_invalidBuffers.size(), 0) {}
InputStageState() {
_bufferOffsets.resize(MAX_NUM_INPUT_BUFFERS, 0);
_bufferStrides.resize(MAX_NUM_INPUT_BUFFERS, 0);
_bufferVBOs.resize(MAX_NUM_INPUT_BUFFERS, 0);
}
} _input;
virtual void initTransform() = 0;
@ -349,7 +489,7 @@ protected:
virtual void resetTransformStage();
// Allows for correction of the camera pose to account for changes
// between the time when a was recorded and the time(s) when it is
// between the time when a was recorded and the time(s) when it is
// executed
// Prev is the previous correction used at previous frame
struct CameraCorrection {
@ -364,9 +504,12 @@ protected:
struct Cameras {
TransformCamera _cams[2];
Cameras() {};
Cameras(){};
Cameras(const TransformCamera& cam) { memcpy(_cams, &cam, sizeof(TransformCamera)); };
Cameras(const TransformCamera& camL, const TransformCamera& camR) { memcpy(_cams, &camL, sizeof(TransformCamera)); memcpy(_cams + 1, &camR, sizeof(TransformCamera)); };
Cameras(const TransformCamera& camL, const TransformCamera& camR) {
memcpy(_cams, &camL, sizeof(TransformCamera));
memcpy(_cams + 1, &camR, sizeof(TransformCamera));
};
};
using CameraBufferElement = Cameras;
@ -380,25 +523,24 @@ protected:
mutable std::map<std::string, GLvoid*> _drawCallInfoOffsets;
GLuint _objectBuffer { 0 };
GLuint _cameraBuffer { 0 };
GLuint _drawCallInfoBuffer { 0 };
GLuint _objectBufferTexture { 0 };
size_t _cameraUboSize { 0 };
GLuint _objectBuffer{ 0 };
GLuint _cameraBuffer{ 0 };
GLuint _drawCallInfoBuffer{ 0 };
GLuint _objectBufferTexture{ 0 };
size_t _cameraUboSize{ 0 };
bool _viewIsCamera{ false };
bool _skybox { false };
bool _skybox{ false };
Transform _view;
CameraCorrection _correction;
bool _viewCorrectionEnabled{ true };
Mat4 _projection;
Vec4i _viewport { 0, 0, 1, 1 };
Vec2 _depthRange { 0.0f, 1.0f };
Vec4i _viewport{ 0, 0, 1, 1 };
Vec2 _depthRange{ 0.0f, 1.0f };
Vec2 _projectionJitter{ 0.0f, 0.0f };
bool _invalidView { false };
bool _invalidProj { false };
bool _invalidViewport { false };
bool _invalidView{ false };
bool _invalidProj{ false };
bool _invalidViewport{ false };
bool _enabledDrawcallInfoBuffer{ false };
@ -417,44 +559,47 @@ protected:
struct UniformStageState {
struct BufferState {
BufferPointer buffer;
BufferReference buffer{ GPU_REFERENCE_INIT_VALUE };
GLintptr offset{ 0 };
GLsizeiptr size{ 0 };
BufferState(const BufferPointer& buffer = nullptr, GLintptr offset = 0, GLsizeiptr size = 0);
bool operator ==(BufferState& other) const {
return offset == other.offset && size == other.size && buffer == other.buffer;
//BufferState(const BufferPointer& buffer = nullptr, GLintptr offset = 0, GLsizeiptr size = 0);
BufferState& operator=(const BufferState& other) = delete;
void reset() { gpu::gl::reset(buffer); offset = 0; size = 0; }
bool compare(const BufferPointer& buffer, GLintptr offset, GLsizeiptr size) {
const auto& self = *this;
return (self.offset == offset && self.size == size && gpu::gl::compare(self.buffer, buffer));
}
};
// MAX_NUM_UNIFORM_BUFFERS-1 is the max uniform index BATCHES are allowed to set, but
// MIN_REQUIRED_UNIFORM_BUFFER_BINDINGS is used here because the backend sets some
// internal UBOs for things like camera correction
// MIN_REQUIRED_UNIFORM_BUFFER_BINDINGS is used here because the backend sets some
// internal UBOs for things like camera correction
std::array<BufferState, MIN_REQUIRED_UNIFORM_BUFFER_BINDINGS> _buffers;
} _uniform;
// Helper function that provides common code
// Helper function that provides common code
void bindUniformBuffer(uint32_t slot, const BufferPointer& buffer, GLintptr offset = 0, GLsizeiptr size = 0);
void releaseUniformBuffer(uint32_t slot);
void resetUniformStage();
// update resource cache and do the gl bind/unbind call with the current gpu::Buffer cached at slot s
// This is using different gl object depending on the gl version
virtual bool bindResourceBuffer(uint32_t slot, BufferPointer& buffer) = 0;
virtual bool bindResourceBuffer(uint32_t slot, const BufferPointer& buffer) = 0;
virtual void releaseResourceBuffer(uint32_t slot) = 0;
// Helper function that provides common code used by do_setResourceTexture and
// Helper function that provides common code used by do_setResourceTexture and
// do_setResourceTextureTable (in non-bindless mode)
void bindResourceTexture(uint32_t slot, const TexturePointer& texture);
// update resource cache and do the gl unbind call with the current gpu::Texture cached at slot s
void releaseResourceTexture(uint32_t slot);
void resetResourceStage();
struct ResourceStageState {
std::array<BufferPointer, MAX_NUM_RESOURCE_BUFFERS> _buffers;
std::array<TexturePointer, MAX_NUM_RESOURCE_TEXTURES> _textures;
std::vector<BufferReference> _buffers{ MAX_NUM_RESOURCE_BUFFERS, BufferReference() };
std::vector<TextureReference> _textures{ MAX_NUM_RESOURCE_TEXTURES, GPU_REFERENCE_INIT_VALUE };
//Textures _textures { { MAX_NUM_RESOURCE_TEXTURES } };
int findEmptyTextureSlot() const;
} _resource;
@ -470,21 +615,22 @@ protected:
void resetPipelineStage();
struct PipelineStageState {
PipelinePointer _pipeline;
PipelineReference _pipeline{ GPU_REFERENCE_INIT_VALUE };
GLuint _program { 0 };
bool _cameraCorrection { false };
GLShader* _programShader { nullptr };
bool _invalidProgram { false };
GLuint _program{ 0 };
bool _cameraCorrection{ false };
GLShader* _programShader{ nullptr };
bool _invalidProgram{ false };
BufferView _cameraCorrectionBuffer { gpu::BufferView(std::make_shared<gpu::Buffer>(sizeof(CameraCorrection), nullptr )) };
BufferView _cameraCorrectionBufferIdentity { gpu::BufferView(std::make_shared<gpu::Buffer>(sizeof(CameraCorrection), nullptr )) };
BufferView _cameraCorrectionBuffer{ gpu::BufferView(std::make_shared<gpu::Buffer>(sizeof(CameraCorrection), nullptr)) };
BufferView _cameraCorrectionBufferIdentity{ gpu::BufferView(
std::make_shared<gpu::Buffer>(sizeof(CameraCorrection), nullptr)) };
State::Data _stateCache{ State::DEFAULT };
State::Signature _stateSignatureCache { 0 };
State::Signature _stateSignatureCache{ 0 };
GLState* _state { nullptr };
bool _invalidState { false };
GLState* _state{ nullptr };
bool _invalidState{ false };
PipelineStageState() {
_cameraCorrectionBuffer.edit<CameraCorrection>() = CameraCorrection();
@ -498,9 +644,9 @@ protected:
virtual GLShader* compileBackendProgram(const Shader& program, const Shader::CompilationHandler& handler);
virtual GLShader* compileBackendShader(const Shader& shader, const Shader::CompilationHandler& handler);
virtual std::string getBackendShaderHeader() const = 0;
// For a program, this will return a string containing all the source files (without any
// backend headers or defines). For a vertex, fragment or geometry shader, this will
// return the fully customized shader with all the version and backend specific
// For a program, this will return a string containing all the source files (without any
// backend headers or defines). For a vertex, fragment or geometry shader, this will
// return the fully customized shader with all the version and backend specific
// preprocessor directives
// The program string returned can be used as a key for a cache of shader binaries
// The shader strings can be reliably sent to the low level `compileShader` functions
@ -516,22 +662,22 @@ protected:
// Synchronize the state cache of this Backend with the actual real state of the GL Context
void syncOutputStateCache();
void resetOutputStage();
struct OutputStageState {
FramebufferPointer _framebuffer { nullptr };
GLuint _drawFBO { 0 };
FramebufferReference _framebuffer{ GPU_REFERENCE_INIT_VALUE };
GLuint _drawFBO{ 0 };
} _output;
void resetQueryStage();
struct QueryStageState {
uint32_t _rangeQueryDepth { 0 };
uint32_t _rangeQueryDepth{ 0 };
} _queryStage;
void resetStages();
// Stores cached binary versions of the shaders for quicker startup on subsequent runs
// Note that shaders in the cache can still fail to load due to hardware or driver
// changes that invalidate the cached binary, in which case we fall back on compiling
// Note that shaders in the cache can still fail to load due to hardware or driver
// changes that invalidate the cached binary, in which case we fall back on compiling
// the source again
struct ShaderBinaryCache {
std::mutex _mutex;
@ -543,7 +689,7 @@ protected:
virtual void killShaderBinaryCache();
struct TextureManagementStageState {
bool _sparseCapable { false };
bool _sparseCapable{ false };
GLTextureTransferEnginePointer _transferEngine;
} _textureManagement;
virtual void initTextureManagementStage();
@ -556,6 +702,6 @@ protected:
friend class GLShader;
};
} }
}} // namespace gpu::gl
#endif

View file

@ -18,8 +18,8 @@ using namespace gpu::gl;
void GLBackend::do_setInputFormat(const Batch& batch, size_t paramOffset) {
const auto& format = batch._streamFormats.get(batch._params[paramOffset]._uint);
if (format != _input._format) {
_input._format = format;
if (!compare(_input._format, format)) {
assign(_input._format, format);
if (format) {
auto inputFormat = GLInputFormat::sync((*format));
assert(inputFormat);
@ -42,8 +42,8 @@ void GLBackend::do_setInputBuffer(const Batch& batch, size_t paramOffset) {
if (channel < getNumInputBuffers()) {
bool isModified = false;
if (_input._buffers[channel] != buffer) {
_input._buffers[channel] = buffer;
if (!compare(_input._buffers[channel], buffer)) {
assign(_input._buffers[channel], buffer);
_input._bufferVBOs[channel] = getBufferIDUnsynced((*buffer));
isModified = true;
}
@ -94,18 +94,18 @@ void GLBackend::resetInputStage() {
// Reset index buffer
_input._indexBufferType = UINT32;
_input._indexBufferOffset = 0;
_input._indexBuffer.reset();
reset(_input._indexBuffer);
glBindBuffer(GL_ELEMENT_ARRAY_BUFFER, 0);
(void) CHECK_GL_ERROR();
// Reset vertex buffer and format
_input._format.reset();
reset(_input._format);
_input._formatKey.clear();
_input._invalidFormat = false;
_input._attributeActivation.reset();
for (uint32_t i = 0; i < _input._buffers.size(); i++) {
_input._buffers[i].reset();
reset(_input._buffers[i]);
_input._bufferOffsets[i] = 0;
_input._bufferStrides[i] = 0;
_input._bufferVBOs[i] = 0;
@ -120,8 +120,8 @@ void GLBackend::do_setIndexBuffer(const Batch& batch, size_t paramOffset) {
_input._indexBufferOffset = batch._params[paramOffset + 0]._uint;
const auto& indexBuffer = batch._buffers.get(batch._params[paramOffset + 1]._uint);
if (indexBuffer != _input._indexBuffer) {
_input._indexBuffer = indexBuffer;
if (!compare(_input._indexBuffer, indexBuffer)) {
assign(_input._indexBuffer, indexBuffer);
if (indexBuffer) {
glBindBuffer(GL_ELEMENT_ARRAY_BUFFER, getBufferIDUnsynced(*indexBuffer));
} else {
@ -137,8 +137,8 @@ void GLBackend::do_setIndirectBuffer(const Batch& batch, size_t paramOffset) {
_input._indirectBufferStride = batch._params[paramOffset + 2]._uint;
const auto& buffer = batch._buffers.get(batch._params[paramOffset]._uint);
if (buffer != _input._indirectBuffer) {
_input._indirectBuffer = buffer;
if (!compare(_input._indirectBuffer, buffer)) {
assign(_input._indirectBuffer, buffer);
if (buffer) {
glBindBuffer(GL_DRAW_INDIRECT_BUFFER, getBufferIDUnsynced(*buffer));
} else {
@ -152,7 +152,7 @@ void GLBackend::do_setIndirectBuffer(const Batch& batch, size_t paramOffset) {
void GLBackend::updateInput() {
bool isStereoNow = isStereo();
// track stereo state change potentially happening wihtout changing the input format
// track stereo state change potentially happening without changing the input format
// this is a rare case requesting to invalid the format
#ifdef GPU_STEREO_DRAWCALL_INSTANCED
_input._invalidFormat |= (isStereoNow != _input._lastUpdateStereoState);
@ -163,13 +163,14 @@ void GLBackend::updateInput() {
InputStageState::ActivationCache newActivation;
// Assign the vertex format required
if (_input._format) {
auto format = acquire(_input._format);
if (format) {
bool hasColorAttribute{ false };
_input._attribBindingBuffers.reset();
const Stream::Format::AttributeMap& attributes = _input._format->getAttributes();
auto& inputChannels = _input._format->getChannels();
const auto& attributes = format->getAttributes();
const auto& inputChannels = format->getChannels();
for (auto& channelIt : inputChannels) {
auto bufferChannelNum = (channelIt).first;
const Stream::Format::ChannelMap::value_type::second_type& channel = (channelIt).second;

View file

@ -25,23 +25,19 @@ using namespace gpu::gl;
void GLBackend::syncOutputStateCache() {
GLint currentFBO;
glGetIntegerv(GL_DRAW_FRAMEBUFFER_BINDING, &currentFBO);
_output._drawFBO = currentFBO;
_output._framebuffer.reset();
reset(_output._framebuffer);
}
void GLBackend::resetOutputStage() {
if (_output._framebuffer) {
_output._framebuffer.reset();
_output._drawFBO = 0;
glBindFramebuffer(GL_DRAW_FRAMEBUFFER, 0);
}
_output._drawFBO = 0;
glBindFramebuffer(GL_DRAW_FRAMEBUFFER, 0);
glEnable(GL_FRAMEBUFFER_SRGB);
reset(_output._framebuffer);
}
void GLBackend::do_setFramebuffer(const Batch& batch, size_t paramOffset) {
auto framebuffer = batch._framebuffers.get(batch._params[paramOffset]._uint);
const auto& framebuffer = batch._framebuffers.get(batch._params[paramOffset]._uint);
setFramebuffer(framebuffer);
}
@ -55,13 +51,13 @@ void GLBackend::do_setFramebufferSwapChain(const Batch& batch, size_t paramOffse
}
void GLBackend::setFramebuffer(const FramebufferPointer& framebuffer) {
if (_output._framebuffer != framebuffer) {
if (!compare(_output._framebuffer, framebuffer)) {
auto newFBO = getFramebufferID(framebuffer);
if (_output._drawFBO != newFBO) {
_output._drawFBO = newFBO;
glBindFramebuffer(GL_DRAW_FRAMEBUFFER, newFBO);
}
_output._framebuffer = framebuffer;
assign(_output._framebuffer, framebuffer);
}
}
@ -114,8 +110,9 @@ void GLBackend::do_clearFramebuffer(const Batch& batch, size_t paramOffset) {
}
std::vector<GLenum> drawBuffers;
auto framebuffer = acquire(_output._framebuffer);
if (masks & Framebuffer::BUFFER_COLORS) {
if (_output._framebuffer) {
if (framebuffer) {
for (unsigned int i = 0; i < Framebuffer::MAX_NUM_RENDER_BUFFERS; i++) {
if (masks & (1 << i)) {
drawBuffers.push_back(GL_COLOR_ATTACHMENT0 + i);
@ -163,8 +160,8 @@ void GLBackend::do_clearFramebuffer(const Batch& batch, size_t paramOffset) {
}
// Restore the color draw buffers only if a frmaebuffer is bound
if (_output._framebuffer && !drawBuffers.empty()) {
auto glFramebuffer = syncGPUObject(*_output._framebuffer);
if (framebuffer && !drawBuffers.empty()) {
auto glFramebuffer = syncGPUObject(*framebuffer);
if (glFramebuffer) {
glDrawBuffers((GLsizei)glFramebuffer->_colorBuffers.size(), glFramebuffer->_colorBuffers.data());
}

View file

@ -25,7 +25,7 @@ using namespace gpu::gl;
void GLBackend::do_setPipeline(const Batch& batch, size_t paramOffset) {
const auto& pipeline = batch._pipelines.get(batch._params[paramOffset + 0]._uint);
if (_pipeline._pipeline == pipeline) {
if (compare(_pipeline._pipeline, pipeline)) {
return;
}
@ -34,7 +34,7 @@ void GLBackend::do_setPipeline(const Batch& batch, size_t paramOffset) {
// null pipeline == reset
if (!pipeline) {
_pipeline._pipeline.reset();
reset(_pipeline._pipeline);
_pipeline._program = 0;
_pipeline._cameraCorrection = false;
@ -73,7 +73,7 @@ void GLBackend::do_setPipeline(const Batch& batch, size_t paramOffset) {
}
// Remember the new pipeline
_pipeline._pipeline = pipeline;
assign(_pipeline._pipeline, pipeline);
}
// THis should be done on Pipeline::update...
@ -81,7 +81,7 @@ void GLBackend::do_setPipeline(const Batch& batch, size_t paramOffset) {
glUseProgram(_pipeline._program);
if (_pipeline._cameraCorrection) {
// Invalidate uniform buffer cache slot
_uniform._buffers[gpu::slot::buffer::CameraCorrection] = {};
_uniform._buffers[gpu::slot::buffer::CameraCorrection].reset();
auto& cameraCorrectionBuffer = _transform._viewCorrectionEnabled ?
_pipeline._cameraCorrectionBuffer._buffer :
_pipeline._cameraCorrectionBufferIdentity._buffer;
@ -112,7 +112,7 @@ void GLBackend::updatePipeline() {
_pipeline._stateSignatureCache |= _pipeline._state->_signature;
// And perform
for (auto command : _pipeline._state->_commands) {
for (const auto& command : _pipeline._state->_commands) {
command->run(this);
}
} else {
@ -134,23 +134,21 @@ void GLBackend::resetPipelineStage() {
_pipeline._invalidProgram = false;
_pipeline._program = 0;
_pipeline._programShader = nullptr;
_pipeline._pipeline.reset();
reset(_pipeline._pipeline);
glUseProgram(0);
}
GLBackend::UniformStageState::BufferState::BufferState(const BufferPointer& buffer, GLintptr offset, GLsizeiptr size)
: buffer(buffer), offset(offset), size(size) {}
void GLBackend::releaseUniformBuffer(uint32_t slot) {
auto& buf = _uniform._buffers[slot];
if (buf.buffer) {
auto* object = Backend::getGPUObject<GLBuffer>(*buf.buffer);
auto& bufferState = _uniform._buffers[slot];
auto buffer = acquire(bufferState.buffer);
if (buffer) {
auto* object = Backend::getGPUObject<GLBuffer>(*buffer);
if (object) {
glBindBufferBase(GL_UNIFORM_BUFFER, slot, 0); // RELEASE
(void)CHECK_GL_ERROR();
}
buf = UniformStageState::BufferState();
}
bufferState.reset();
}
void GLBackend::resetUniformStage() {
@ -165,18 +163,20 @@ void GLBackend::bindUniformBuffer(uint32_t slot, const BufferPointer& buffer, GL
return;
}
UniformStageState::BufferState bufferState{ buffer, offset, size };
auto& currentBufferState = _uniform._buffers[slot];
// check cache before thinking
if (_uniform._buffers[slot] == bufferState) {
if (currentBufferState.compare(buffer, offset, size)) {
return;
}
// Grab the true gl Buffer object
auto glBO = getBufferIDUnsynced(*buffer);
if (glBO) {
glBindBufferRange(GL_UNIFORM_BUFFER, slot, glBO, bufferState.offset, bufferState.size);
_uniform._buffers[slot] = bufferState;
glBindBufferRange(GL_UNIFORM_BUFFER, slot, glBO, offset, size);
assign(currentBufferState.buffer, buffer);
currentBufferState.offset = offset;
currentBufferState.size = size;
(void)CHECK_GL_ERROR();
} else {
releaseUniformBuffer(slot);
@ -201,7 +201,7 @@ void GLBackend::do_setUniformBuffer(const Batch& batch, size_t paramOffset) {
}
void GLBackend::releaseResourceTexture(uint32_t slot) {
auto& tex = _resource._textures[slot];
auto tex = acquire(_resource._textures[slot]);
if (tex) {
auto* object = Backend::getGPUObject<GLTexture>(*tex);
if (object) {
@ -210,8 +210,8 @@ void GLBackend::releaseResourceTexture(uint32_t slot) {
glBindTexture(target, 0); // RELEASE
(void)CHECK_GL_ERROR();
}
tex.reset();
}
reset(_resource._textures[slot]);
}
void GLBackend::resetResourceStage() {
@ -232,14 +232,14 @@ void GLBackend::do_setResourceBuffer(const Batch& batch, size_t paramOffset) {
return;
}
auto resourceBuffer = batch._buffers.get(batch._params[paramOffset + 0]._uint);
const auto& resourceBuffer = batch._buffers.get(batch._params[paramOffset + 0]._uint);
if (!resourceBuffer) {
releaseResourceBuffer(slot);
return;
}
// check cache before thinking
if (_resource._buffers[slot] == resourceBuffer) {
if (compare(_resource._buffers[slot], resourceBuffer)) {
return;
}
@ -248,7 +248,7 @@ void GLBackend::do_setResourceBuffer(const Batch& batch, size_t paramOffset) {
// If successful bind then cache it
if (bindResourceBuffer(slot, resourceBuffer)) {
_resource._buffers[slot] = resourceBuffer;
assign(_resource._buffers[slot], resourceBuffer);
} else { // else clear slot and cache
releaseResourceBuffer(slot);
return;
@ -293,14 +293,14 @@ void GLBackend::do_setResourceFramebufferSwapChainTexture(const Batch& batch, si
}
auto index = batch._params[paramOffset + 2]._uint;
auto renderBufferSlot = batch._params[paramOffset + 3]._uint;
auto resourceFramebuffer = swapChain->get(index);
auto resourceTexture = resourceFramebuffer->getRenderBuffer(renderBufferSlot);
const auto& resourceFramebuffer = swapChain->get(index);
const auto& resourceTexture = resourceFramebuffer->getRenderBuffer(renderBufferSlot);
setResourceTexture(slot, resourceTexture);
}
void GLBackend::setResourceTexture(unsigned int slot, const TexturePointer& resourceTexture) {
// check cache before thinking
if (_resource._textures[slot] == resourceTexture) {
if (compare(_resource._textures[slot], resourceTexture)) {
return;
}
@ -317,7 +317,7 @@ void GLBackend::setResourceTexture(unsigned int slot, const TexturePointer& reso
(void)CHECK_GL_ERROR();
_resource._textures[slot] = resourceTexture;
assign(_resource._textures[slot], resourceTexture);
_stats._RSAmountTextureMemoryBounded += (int)object->size();
@ -343,7 +343,7 @@ void GLBackend::do_setResourceTextureTable(const Batch& batch, size_t paramOffse
int GLBackend::ResourceStageState::findEmptyTextureSlot() const {
// start from the end of the slots, try to find an empty one that can be used
for (auto i = MAX_NUM_RESOURCE_TEXTURES - 1; i > 0; i--) {
if (!_textures[i]) {
if (!valid(_textures[i])) {
return i;
}
}

View file

@ -24,7 +24,7 @@ GLPipeline* GLPipeline::sync(GLBackend& backend, const Pipeline& pipeline) {
}
// No object allocated yet, let's see if it's worth it...
ShaderPointer shader = pipeline.getProgram();
const auto& shader = pipeline.getProgram();
// If this pipeline's shader has already failed to compile, don't try again
if (shader->compilationHasFailed()) {
@ -37,7 +37,7 @@ GLPipeline* GLPipeline::sync(GLBackend& backend, const Pipeline& pipeline) {
return nullptr;
}
StatePointer state = pipeline.getState();
const auto& state = pipeline.getState();
GLState* stateObject = GLState::sync(*state);
if (stateObject == nullptr) {
return nullptr;

View file

@ -161,7 +161,7 @@ protected:
void updateTransform(const Batch& batch) override;
// Resource Stage
bool bindResourceBuffer(uint32_t slot, BufferPointer& buffer) override;
bool bindResourceBuffer(uint32_t slot, const BufferPointer& buffer) override;
void releaseResourceBuffer(uint32_t slot) override;
// Output stage

View file

@ -100,7 +100,7 @@ GLBuffer* GL41Backend::syncGPUObject(const Buffer& buffer) {
return GL41Buffer::sync<GL41Buffer>(*this, buffer);
}
bool GL41Backend::bindResourceBuffer(uint32_t slot, BufferPointer& buffer) {
bool GL41Backend::bindResourceBuffer(uint32_t slot, const BufferPointer& buffer) {
GLuint texBuffer = GL41Backend::getResourceBufferID((*buffer));
if (texBuffer) {
glActiveTexture(GL_TEXTURE0 + GL41Backend::RESOURCE_BUFFER_SLOT0_TEX_UNIT + slot);
@ -108,7 +108,7 @@ bool GL41Backend::bindResourceBuffer(uint32_t slot, BufferPointer& buffer) {
(void)CHECK_GL_ERROR();
_resource._buffers[slot] = buffer;
assign(_resource._buffers[slot], buffer);
return true;
}
@ -117,10 +117,7 @@ bool GL41Backend::bindResourceBuffer(uint32_t slot, BufferPointer& buffer) {
}
void GL41Backend::releaseResourceBuffer(uint32_t slot) {
auto& buf = _resource._buffers[slot];
if (buf) {
glActiveTexture(GL_TEXTURE0 + GL41Backend::RESOURCE_BUFFER_SLOT0_TEX_UNIT + slot);
glBindTexture(GL_TEXTURE_BUFFER, 0);
buf.reset();
}
reset(_resource._buffers[slot]);
glActiveTexture(GL_TEXTURE0 + GL41Backend::RESOURCE_BUFFER_SLOT0_TEX_UNIT + slot);
glBindTexture(GL_TEXTURE_BUFFER, 0);
}

View file

@ -35,14 +35,15 @@ void GL41Backend::updateInput() {
if (_input._invalidFormat || _input._invalidBuffers.any()) {
auto format = acquire(_input._format);
if (_input._invalidFormat) {
InputStageState::ActivationCache newActivation;
_stats._ISNumFormatChanges++;
// Check expected activation
if (_input._format) {
for (auto& it : _input._format->getAttributes()) {
if (format) {
for (auto& it : format->getAttributes()) {
const Stream::Attribute& attrib = (it).second;
uint8_t locationCount = attrib._element.getLocationCount();
for (int i = 0; i < locationCount; ++i) {
@ -69,15 +70,15 @@ void GL41Backend::updateInput() {
}
// now we need to bind the buffers and assign the attrib pointers
if (_input._format) {
if (format) {
bool hasColorAttribute{ false };
const Buffers& buffers = _input._buffers;
const auto& buffers = _input._buffers;
const Offsets& offsets = _input._bufferOffsets;
const Offsets& strides = _input._bufferStrides;
const Stream::Format::AttributeMap& attributes = _input._format->getAttributes();
auto& inputChannels = _input._format->getChannels();
const auto& attributes = format->getAttributes();
const auto& inputChannels = format->getChannels();
int numInvalids = (int)_input._invalidBuffers.count();
_stats._ISNumInputBufferChanges += numInvalids;

View file

@ -262,7 +262,7 @@ protected:
void updateTransform(const Batch& batch) override;
// Resource Stage
bool bindResourceBuffer(uint32_t slot, BufferPointer& buffer) override;
bool bindResourceBuffer(uint32_t slot, const BufferPointer& buffer) override;
void releaseResourceBuffer(uint32_t slot) override;
// Output stage

View file

@ -60,14 +60,14 @@ GLBuffer* GL45Backend::syncGPUObject(const Buffer& buffer) {
}
bool GL45Backend::bindResourceBuffer(uint32_t slot, BufferPointer& buffer) {
bool GL45Backend::bindResourceBuffer(uint32_t slot, const BufferPointer& buffer) {
GLBuffer* object = syncGPUObject((*buffer));
if (object) {
glBindBufferBase(GL_SHADER_STORAGE_BUFFER, slot, object->_id);
(void)CHECK_GL_ERROR();
_resource._buffers[slot] = buffer;
assign(_resource._buffers[slot], buffer);
return true;
}
@ -76,10 +76,10 @@ bool GL45Backend::bindResourceBuffer(uint32_t slot, BufferPointer& buffer) {
}
void GL45Backend::releaseResourceBuffer(uint32_t slot) {
auto& buf = _resource._buffers[slot];
auto buf = acquire(_resource._buffers[slot]);
if (buf) {
glBindBufferBase(GL_SHADER_STORAGE_BUFFER, slot, 0);
buf.reset();
reset(_resource._buffers[slot]);
}
}

View file

@ -28,7 +28,7 @@ void GL45Backend::resetInputStage() {
void GL45Backend::updateInput() {
bool isStereoNow = isStereo();
// track stereo state change potentially happening wihtout changing the input format
// track stereo state change potentially happening without changing the input format
// this is a rare case requesting to invalid the format
#ifdef GPU_STEREO_DRAWCALL_INSTANCED
_input._invalidFormat |= (isStereoNow != _input._lastUpdateStereoState);
@ -39,13 +39,14 @@ void GL45Backend::updateInput() {
InputStageState::ActivationCache newActivation;
// Assign the vertex format required
if (_input._format) {
auto format = acquire(_input._format);
if (format) {
bool hasColorAttribute{ false };
_input._attribBindingBuffers.reset();
const Stream::Format::AttributeMap& attributes = _input._format->getAttributes();
auto& inputChannels = _input._format->getChannels();
const auto& attributes = format->getAttributes();
const auto& inputChannels = format->getChannels();
for (auto& channelIt : inputChannels) {
auto bufferChannelNum = (channelIt).first;
const Stream::Format::ChannelMap::value_type::second_type& channel = (channelIt).second;

View file

@ -157,7 +157,7 @@ protected:
void updateTransform(const Batch& batch) override;
// Resource Stage
bool bindResourceBuffer(uint32_t slot, BufferPointer& buffer) override;
bool bindResourceBuffer(uint32_t slot, const BufferPointer& buffer) override;
void releaseResourceBuffer(uint32_t slot) override;
// Output stage

View file

@ -72,14 +72,14 @@ GLBuffer* GLESBackend::syncGPUObject(const Buffer& buffer) {
return GLESBuffer::sync<GLESBuffer>(*this, buffer);
}
bool GLESBackend::bindResourceBuffer(uint32_t slot, BufferPointer& buffer) {
bool GLESBackend::bindResourceBuffer(uint32_t slot, const BufferPointer& buffer) {
GLBuffer* object = syncGPUObject((*buffer));
if (object) {
glBindBufferBase(GL_SHADER_STORAGE_BUFFER, slot, object->_id);
(void)CHECK_GL_ERROR();
_resource._buffers[slot] = buffer;
assign(_resource._buffers[slot], buffer);
return true;
}
@ -91,7 +91,7 @@ void GLESBackend::releaseResourceBuffer(uint32_t slot) {
auto& buf = _resource._buffers[slot];
if (buf) {
glBindBufferBase(GL_SHADER_STORAGE_BUFFER, slot, 0);
buf.reset();
reset(buf);
}
}

View file

@ -498,7 +498,7 @@ void Batch::setupNamedCalls(const std::string& instanceName, NamedBatchData::Fun
captureNamedDrawCallInfo(instanceName);
}
BufferPointer Batch::getNamedBuffer(const std::string& instanceName, uint8_t index) {
const BufferPointer& Batch::getNamedBuffer(const std::string& instanceName, uint8_t index) {
NamedBatchData& instance = _namedData[instanceName];
if (instance.buffers.size() <= index) {
instance.buffers.resize(index + 1);

View file

@ -119,7 +119,7 @@ public:
void multiDrawIndexedIndirect(uint32 numCommands, Primitive primitiveType);
void setupNamedCalls(const std::string& instanceName, NamedBatchData::Function function);
BufferPointer getNamedBuffer(const std::string& instanceName, uint8_t index = 0);
const BufferPointer& getNamedBuffer(const std::string& instanceName, uint8_t index = 0);
// Input Stage
// InputFormat

View file

@ -203,11 +203,12 @@ uint32 Framebuffer::getNumRenderBuffers() const {
return nb;
}
TexturePointer Framebuffer::getRenderBuffer(uint32 slot) const {
const TexturePointer& Framebuffer::getRenderBuffer(uint32 slot) const {
static const TexturePointer EMPTY;
if (!isSwapchain() && (slot < getMaxNumRenderBuffers())) {
return _renderBuffers[slot]._texture;
} else {
return TexturePointer();
return EMPTY;
}
}
@ -297,9 +298,10 @@ bool Framebuffer::setDepthStencilBuffer(const TexturePointer& texture, const For
return false;
}
TexturePointer Framebuffer::getDepthStencilBuffer() const {
const TexturePointer& Framebuffer::getDepthStencilBuffer() const {
static const TexturePointer EMPTY;
if (isSwapchain()) {
return TexturePointer();
return EMPTY;
} else {
return _depthStencilBuffer._texture;
}

View file

@ -95,7 +95,7 @@ public:
static Framebuffer* createShadowmap(uint16 width);
bool isSwapchain() const;
SwapchainPointer getSwapchain() const { return _swapchain; }
const SwapchainPointer& getSwapchain() const { return _swapchain; }
uint32 getFrameCount() const;
@ -105,13 +105,13 @@ public:
const TextureViews& getRenderBuffers() const { return _renderBuffers; }
int32 setRenderBuffer(uint32 slot, const TexturePointer& texture, uint32 subresource = 0);
TexturePointer getRenderBuffer(uint32 slot) const;
const TexturePointer& getRenderBuffer(uint32 slot) const;
uint32 getRenderBufferSubresource(uint32 slot) const;
bool setDepthBuffer(const TexturePointer& texture, const Format& format, uint32 subresource = 0);
bool setStencilBuffer(const TexturePointer& texture, const Format& format, uint32 subresource = 0);
bool setDepthStencilBuffer(const TexturePointer& texture, const Format& format, uint32 subresource = 0);
TexturePointer getDepthStencilBuffer() const;
const TexturePointer& getDepthStencilBuffer() const;
uint32 getDepthStencilBufferSubresource() const;
Format getDepthStencilBufferFormat() const;