#include "boo/graphicsdev/GL.hpp" #include "boo/graphicsdev/glew.h" #include "boo/IGraphicsContext.hpp" #include #include #include #include #include #include #undef min #undef max namespace boo { static LogVisor::LogModule Log("boo::GL"); ThreadLocalPtr GLDataFactory::m_deferredData; struct GLData : IGraphicsData { std::vector> m_SPs; std::vector> m_SBinds; std::vector> m_SBufs; std::vector> m_DBufs; std::vector> m_STexs; std::vector> m_SATexs; std::vector> m_DTexs; std::vector> m_RTexs; std::vector> m_VFmts; }; static const GLenum USE_TABLE[] = { GL_INVALID_ENUM, GL_ARRAY_BUFFER, GL_ELEMENT_ARRAY_BUFFER, GL_UNIFORM_BUFFER }; class GLGraphicsBufferS : public IGraphicsBufferS { friend class GLDataFactory; friend struct GLCommandQueue; GLuint m_buf; GLenum m_target; GLGraphicsBufferS(BufferUse use, const void* data, size_t sz) { m_target = USE_TABLE[int(use)]; glGenBuffers(1, &m_buf); glBindBuffer(m_target, m_buf); glBufferData(m_target, sz, data, GL_STATIC_DRAW); } public: ~GLGraphicsBufferS() {glDeleteBuffers(1, &m_buf);} void bindVertex() const {glBindBuffer(GL_ARRAY_BUFFER, m_buf);} void bindIndex() const {glBindBuffer(GL_ELEMENT_ARRAY_BUFFER, m_buf);} void bindUniform(size_t idx) const {glBindBufferBase(GL_UNIFORM_BUFFER, idx, m_buf);} }; class GLGraphicsBufferD : public IGraphicsBufferD { friend class GLDataFactory; friend struct GLCommandQueue; GLuint m_bufs[3]; GLenum m_target; std::unique_ptr m_cpuBuf; size_t m_cpuSz = 0; int m_validMask = 0; GLGraphicsBufferD(BufferUse use, size_t sz) : m_target(USE_TABLE[int(use)]), m_cpuBuf(new uint8_t[sz]), m_cpuSz(sz) { glGenBuffers(3, m_bufs); for (int i=0 ; i<3 ; ++i) { glBindBuffer(m_target, m_bufs[i]); glBufferData(m_target, m_cpuSz, nullptr, GL_STREAM_DRAW); } } void update(int b); public: ~GLGraphicsBufferD() {glDeleteBuffers(3, m_bufs);} void load(const void* data, size_t sz); void* map(size_t sz); void unmap(); void bindVertex(int b); void bindIndex(int b); void bindUniform(size_t idx, int b); }; IGraphicsBufferS* GLDataFactory::newStaticBuffer(BufferUse use, const void* data, size_t stride, size_t count) { GLGraphicsBufferS* retval = new GLGraphicsBufferS(use, data, stride * count); if (!m_deferredData.get()) m_deferredData.reset(new struct GLData()); m_deferredData->m_SBufs.emplace_back(retval); return retval; } class GLTextureS : public ITextureS { friend class GLDataFactory; GLuint m_tex; GLTextureS(size_t width, size_t height, size_t mips, TextureFormat fmt, const void* data, size_t sz) { const uint8_t* dataIt = static_cast(data); glGenTextures(1, &m_tex); glBindTexture(GL_TEXTURE_2D, m_tex); glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MAG_FILTER, GL_LINEAR); if (mips > 1) { glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, GL_LINEAR_MIPMAP_LINEAR); glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MAX_LEVEL, mips-1); } else glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, GL_LINEAR); GLenum intFormat, format; int pxPitch; bool compressed = false; switch (fmt) { case TextureFormat::RGBA8: intFormat = GL_RGBA; format = GL_RGBA; pxPitch = 4; break; case TextureFormat::I8: intFormat = GL_R8; format = GL_RED; pxPitch = 1; break; case TextureFormat::DXT1: intFormat = GL_COMPRESSED_RGBA_S3TC_DXT1_EXT; compressed = true; break; default: Log.report(LogVisor::FatalError, "unsupported tex format"); } if (compressed) { for (size_t i=0 ; i m_cpuBuf; size_t m_cpuSz = 0; GLenum m_intFormat, m_format; size_t m_width = 0; size_t m_height = 0; int m_validMask = 0; GLTextureD(size_t width, size_t height, TextureFormat fmt); void update(int b); public: ~GLTextureD(); void load(const void* data, size_t sz); void* map(size_t sz); void unmap(); void bind(size_t idx, int b); }; class GLTextureR : public ITextureR { friend class GLDataFactory; friend struct GLCommandQueue; struct GLCommandQueue* m_q; GLuint m_texs[2]; GLuint m_bindTexs[2] = {}; GLuint m_fbo = 0; size_t m_width = 0; size_t m_height = 0; size_t m_samples = 0; GLenum m_target; GLTextureR(GLCommandQueue* q, size_t width, size_t height, size_t samples, bool enableShaderColorBinding, bool enableShaderDepthBinding); public: ~GLTextureR(); void bind(size_t idx) const { glActiveTexture(GL_TEXTURE0 + idx); glBindTexture(m_target, m_bindTexs[0]); } void resize(size_t width, size_t height) { m_width = width; m_height = height; if (m_samples > 1) { glBindTexture(GL_TEXTURE_2D_MULTISAMPLE, m_texs[0]); glTexImage2DMultisample(GL_TEXTURE_2D_MULTISAMPLE, m_samples, GL_RGBA, width, height, GL_FALSE); glBindTexture(GL_TEXTURE_2D_MULTISAMPLE, m_texs[1]); glTexImage2DMultisample(GL_TEXTURE_2D_MULTISAMPLE, m_samples, GL_DEPTH_COMPONENT24, width, height, GL_FALSE); if (m_bindTexs[0]) { glBindTexture(GL_TEXTURE_2D_MULTISAMPLE, m_bindTexs[0]); glTexImage2DMultisample(GL_TEXTURE_2D_MULTISAMPLE, m_samples, GL_RGBA, width, height, GL_FALSE); } if (m_bindTexs[1]) { glBindTexture(GL_TEXTURE_2D_MULTISAMPLE, m_bindTexs[1]); glTexImage2DMultisample(GL_TEXTURE_2D_MULTISAMPLE, m_samples, GL_DEPTH_COMPONENT24, width, height, GL_FALSE); } } else { glBindTexture(GL_TEXTURE_2D, m_texs[0]); glTexImage2D(GL_TEXTURE_2D, 0, GL_RGBA, width, height, 0, GL_RGBA, GL_UNSIGNED_BYTE, nullptr); glBindTexture(GL_TEXTURE_2D, m_texs[1]); glTexImage2D(GL_TEXTURE_2D, 0, GL_DEPTH_COMPONENT24, width, height, 0, GL_DEPTH_COMPONENT, GL_UNSIGNED_INT, nullptr); if (m_bindTexs[0]) { glBindTexture(GL_TEXTURE_2D, m_bindTexs[0]); glTexImage2D(GL_TEXTURE_2D, 0, GL_RGBA, width, height, 0, GL_RGBA, GL_UNSIGNED_BYTE, nullptr); } if (m_bindTexs[1]) { glBindTexture(GL_TEXTURE_2D, m_bindTexs[1]); glTexImage2D(GL_TEXTURE_2D, 0, GL_DEPTH_COMPONENT24, width, height, 0, GL_DEPTH_COMPONENT, GL_UNSIGNED_INT, nullptr); } } } }; ITextureS* GLDataFactory::newStaticTexture(size_t width, size_t height, size_t mips, TextureFormat fmt, const void* data, size_t sz) { GLTextureS* retval = new GLTextureS(width, height, mips, fmt, data, sz); if (!m_deferredData.get()) m_deferredData.reset(new struct GLData()); m_deferredData->m_STexs.emplace_back(retval); return retval; } GraphicsDataToken GLDataFactory::newStaticTextureNoContext(size_t width, size_t height, size_t mips, TextureFormat fmt, const void* data, size_t sz, ITextureS*& texOut) { GLTextureS* retval = new GLTextureS(width, height, mips, fmt, data, sz); GLData* tokData = new struct GLData(); tokData->m_STexs.emplace_back(retval); texOut = retval; std::unique_lock lk(m_committedMutex); m_committedData.insert(tokData); lk.unlock(); /* Let's go ahead and flush to ensure our data gets to the GPU While this isn't strictly required, some drivers might behave differently */ glFlush(); return GraphicsDataToken(this, tokData); } ITextureSA* GLDataFactory::newStaticArrayTexture(size_t width, size_t height, size_t layers, TextureFormat fmt, const void *data, size_t sz) { GLTextureSA* retval = new GLTextureSA(width, height, layers, fmt, data, sz); if (!m_deferredData.get()) m_deferredData.reset(new struct GLData()); m_deferredData->m_SATexs.emplace_back(retval); return retval; } class GLShaderPipeline : public IShaderPipeline { friend class GLDataFactory; friend struct GLShaderDataBinding; GLuint m_vert = 0; GLuint m_frag = 0; GLuint m_prog = 0; GLenum m_sfactor = GL_ONE; GLenum m_dfactor = GL_ZERO; bool m_depthTest = true; bool m_depthWrite = true; bool m_backfaceCulling = true; std::vector m_uniLocs; bool initObjects() { m_vert = glCreateShader(GL_VERTEX_SHADER); m_frag = glCreateShader(GL_FRAGMENT_SHADER); m_prog = glCreateProgram(); if (!m_vert || !m_frag || !m_prog) { glDeleteShader(m_vert); m_vert = 0; glDeleteShader(m_frag); m_frag = 0; glDeleteProgram(m_prog); m_prog = 0; return false; } glAttachShader(m_prog, m_vert); glAttachShader(m_prog, m_frag); return true; } void clearObjects() { if (m_vert) glDeleteShader(m_vert); if (m_frag) glDeleteShader(m_frag); if (m_prog) glDeleteProgram(m_prog); } GLShaderPipeline() = default; public: operator bool() const {return m_prog != 0;} ~GLShaderPipeline() {clearObjects();} GLShaderPipeline& operator=(const GLShaderPipeline&) = delete; GLShaderPipeline(const GLShaderPipeline&) = delete; GLShaderPipeline& operator=(GLShaderPipeline&& other) { m_vert = other.m_vert; other.m_vert = 0; m_frag = other.m_frag; other.m_frag = 0; m_prog = other.m_prog; other.m_prog = 0; m_sfactor = other.m_sfactor; m_dfactor = other.m_dfactor; m_depthTest = other.m_depthTest; m_depthWrite = other.m_depthWrite; m_backfaceCulling = other.m_backfaceCulling; m_uniLocs = std::move(other.m_uniLocs); return *this; } GLShaderPipeline(GLShaderPipeline&& other) {*this = std::move(other);} GLuint bind() const { glUseProgram(m_prog); if (m_dfactor != GL_ZERO) { glEnable(GL_BLEND); glBlendFunc(m_sfactor, m_dfactor); } else glDisable(GL_BLEND); if (m_depthTest) glEnable(GL_DEPTH_TEST); else glDisable(GL_DEPTH_TEST); glDepthMask(m_depthWrite); if (m_backfaceCulling) glEnable(GL_CULL_FACE); else glDisable(GL_CULL_FACE); return m_prog; } }; static const GLenum BLEND_FACTOR_TABLE[] = { GL_ZERO, GL_ONE, GL_SRC_COLOR, GL_ONE_MINUS_SRC_COLOR, GL_DST_COLOR, GL_ONE_MINUS_DST_COLOR, GL_SRC_ALPHA, GL_ONE_MINUS_SRC_ALPHA, GL_DST_ALPHA, GL_ONE_MINUS_DST_ALPHA, GL_SRC1_COLOR, GL_ONE_MINUS_SRC1_COLOR }; IShaderPipeline* GLDataFactory::newShaderPipeline (const char* vertSource, const char* fragSource, size_t texCount, const char* texArrayName, size_t uniformBlockCount, const char** uniformBlockNames, BlendFactor srcFac, BlendFactor dstFac, bool depthTest, bool depthWrite, bool backfaceCulling) { GLShaderPipeline shader; if (!shader.initObjects()) { Log.report(LogVisor::Error, "unable to create shader objects\n"); return nullptr; } shader.m_sfactor = BLEND_FACTOR_TABLE[int(srcFac)]; shader.m_dfactor = BLEND_FACTOR_TABLE[int(dstFac)]; shader.m_depthTest = depthTest; shader.m_depthWrite = depthWrite; shader.m_backfaceCulling = backfaceCulling; glShaderSource(shader.m_vert, 1, &vertSource, nullptr); glCompileShader(shader.m_vert); GLint status; glGetShaderiv(shader.m_vert, GL_COMPILE_STATUS, &status); if (status != GL_TRUE) { GLint logLen; glGetShaderiv(shader.m_vert, GL_INFO_LOG_LENGTH, &logLen); char* log = (char*)malloc(logLen); glGetShaderInfoLog(shader.m_vert, logLen, nullptr, log); Log.report(LogVisor::Error, "unable to compile vert source\n%s\n%s\n", log, vertSource); free(log); return nullptr; } glShaderSource(shader.m_frag, 1, &fragSource, nullptr); glCompileShader(shader.m_frag); glGetShaderiv(shader.m_frag, GL_COMPILE_STATUS, &status); if (status != GL_TRUE) { GLint logLen; glGetShaderiv(shader.m_frag, GL_INFO_LOG_LENGTH, &logLen); char* log = (char*)malloc(logLen); glGetShaderInfoLog(shader.m_frag, logLen, nullptr, log); Log.report(LogVisor::Error, "unable to compile frag source\n%s\n%s\n", log, fragSource); free(log); return nullptr; } glLinkProgram(shader.m_prog); glGetProgramiv(shader.m_prog, GL_LINK_STATUS, &status); if (status != GL_TRUE) { GLint logLen; glGetProgramiv(shader.m_prog, GL_INFO_LOG_LENGTH, &logLen); char* log = (char*)malloc(logLen); glGetProgramInfoLog(shader.m_prog, logLen, nullptr, log); Log.report(LogVisor::Error, "unable to link shader program\n%s\n", log); free(log); return nullptr; } glUseProgram(shader.m_prog); if (uniformBlockCount) { shader.m_uniLocs.reserve(uniformBlockCount); for (size_t i=0 ; i m_texUnis.size()) for (size_t i=m_texUnis.size() ; im_SPs.emplace_back(retval); return retval; } struct GLVertexFormat : IVertexFormat { GLCommandQueue* m_q; GLuint m_vao[3] = {}; size_t m_elementCount; std::unique_ptr m_elements; GLVertexFormat(GLCommandQueue* q, size_t elementCount, const VertexElementDescriptor* elements); ~GLVertexFormat(); void bind(int idx) const {glBindVertexArray(m_vao[idx]);} }; struct GLShaderDataBinding : IShaderDataBinding { const GLShaderPipeline* m_pipeline; const GLVertexFormat* m_vtxFormat; size_t m_ubufCount; std::unique_ptr m_ubufs; size_t m_texCount; std::unique_ptr m_texs; #ifndef NDEBUG /* Debugging aids */ bool m_committed = false; #endif GLShaderDataBinding(IShaderPipeline* pipeline, IVertexFormat* vtxFormat, size_t ubufCount, IGraphicsBuffer** ubufs, size_t texCount, ITexture** texs) : m_pipeline(static_cast(pipeline)), m_vtxFormat(static_cast(vtxFormat)), m_ubufCount(ubufCount), m_ubufs(new IGraphicsBuffer*[ubufCount]), m_texCount(texCount), m_texs(new ITexture*[texCount]) { for (size_t i=0 ; ibind(); m_vtxFormat->bind(b); for (size_t i=0 ; idynamic()) static_cast(ubuf)->bindUniform(i, b); else static_cast(ubuf)->bindUniform(i); glUniformBlockBinding(prog, m_pipeline->m_uniLocs.at(i), i); } for (size_t i=0 ; itype()) { case TextureType::Dynamic: static_cast(tex)->bind(i, b); break; case TextureType::Static: static_cast(tex)->bind(i); break; case TextureType::StaticArray: static_cast(tex)->bind(i); break; case TextureType::Render: static_cast(tex)->bind(i); break; default: break; } } } }; IShaderDataBinding* GLDataFactory::newShaderDataBinding(IShaderPipeline* pipeline, IVertexFormat* vtxFormat, IGraphicsBuffer*, IGraphicsBuffer*, IGraphicsBuffer*, size_t ubufCount, IGraphicsBuffer** ubufs, size_t texCount, ITexture** texs) { GLShaderDataBinding* retval = new GLShaderDataBinding(pipeline, vtxFormat, ubufCount, ubufs, texCount, texs); if (!m_deferredData.get()) m_deferredData.reset(new struct GLData()); m_deferredData->m_SBinds.emplace_back(retval); return retval; } GLDataFactory::GLDataFactory(IGraphicsContext* parent, uint32_t drawSamples) : m_parent(parent), m_drawSamples(drawSamples) {} void GLDataFactory::reset() { delete m_deferredData.get(); m_deferredData.reset(); } GraphicsDataToken GLDataFactory::commit() { if (!m_deferredData.get()) return GraphicsDataToken(this, nullptr); std::unique_lock lk(m_committedMutex); GLData* retval = m_deferredData.get(); #ifndef NDEBUG for (std::unique_ptr& b : retval->m_SBinds) b->m_committed = true; #endif m_deferredData.reset(); m_committedData.insert(retval); lk.unlock(); /* Let's go ahead and flush to ensure our data gets to the GPU While this isn't strictly required, some drivers might behave differently */ glFlush(); return GraphicsDataToken(this, retval); } void GLDataFactory::destroyData(IGraphicsData* d) { std::unique_lock lk(m_committedMutex); GLData* data = static_cast(d); m_committedData.erase(data); delete data; } void GLDataFactory::destroyAllData() { std::unique_lock lk(m_committedMutex); for (IGraphicsData* data : m_committedData) delete static_cast(data); m_committedData.clear(); } static const GLint SEMANTIC_COUNT_TABLE[] = { 0, 3, 4, 3, 4, 4, 4, 2, 4, 4, 4 }; static const size_t SEMANTIC_SIZE_TABLE[] = { 0, 12, 16, 12, 16, 16, 4, 8, 16, 16, 16 }; static const GLenum SEMANTIC_TYPE_TABLE[] = { GL_INVALID_ENUM, GL_FLOAT, GL_FLOAT, GL_FLOAT, GL_FLOAT, GL_FLOAT, GL_UNSIGNED_BYTE, GL_FLOAT, GL_FLOAT, GL_FLOAT, GL_FLOAT }; struct GLCommandQueue : IGraphicsCommandQueue { Platform platform() const {return IGraphicsDataFactory::Platform::OGL;} const SystemChar* platformName() const {return _S("OGL");} IGraphicsContext* m_parent = nullptr; struct Command { enum class Op { SetShaderDataBinding, SetRenderTarget, SetViewport, SetScissor, SetClearColor, ClearTarget, Draw, DrawIndexed, DrawInstances, DrawInstancesIndexed, ResolveBindTexture, Present } m_op; union { const IShaderDataBinding* binding; const ITextureR* target; const ITextureR* source; SWindowRect rect; float rgba[4]; GLbitfield flags; struct { size_t start; size_t count; size_t instCount; }; }; const ITextureR* resolveTex; bool resolveColor : 1; bool resolveDepth : 1; Command(Op op) : m_op(op) {} }; std::vector m_cmdBufs[3]; size_t m_fillBuf = 0; size_t m_completeBuf = 0; size_t m_drawBuf = 0; bool m_running = true; std::mutex m_mt; std::condition_variable m_cv; std::mutex m_initmt; std::condition_variable m_initcv; std::unique_lock m_initlk; std::thread m_thr; struct RenderTextureResize { GLTextureR* tex; size_t width; size_t height; }; /* These members are locked for multithreaded access */ std::vector m_pendingResizes; std::vector> m_pendingPosts1; std::vector> m_pendingPosts2; std::vector m_pendingFmtAdds; std::vector> m_pendingFmtDels; std::vector m_pendingFboAdds; std::vector m_pendingFboDels; static void ConfigureVertexFormat(GLVertexFormat* fmt) { glGenVertexArrays(3, fmt->m_vao); size_t stride = 0; size_t instStride = 0; for (size_t i=0 ; im_elementCount ; ++i) { const VertexElementDescriptor* desc = &fmt->m_elements[i]; if ((desc->semantic & VertexSemantic::Instanced) != VertexSemantic::None) instStride += SEMANTIC_SIZE_TABLE[int(desc->semantic & VertexSemantic::SemanticMask)]; else stride += SEMANTIC_SIZE_TABLE[int(desc->semantic & VertexSemantic::SemanticMask)]; } for (int b=0 ; b<3 ; ++b) { size_t offset = 0; size_t instOffset = 0; glBindVertexArray(fmt->m_vao[b]); IGraphicsBuffer* lastVBO = nullptr; IGraphicsBuffer* lastEBO = nullptr; for (size_t i=0 ; im_elementCount ; ++i) { const VertexElementDescriptor* desc = &fmt->m_elements[i]; if (desc->vertBuffer != lastVBO) { lastVBO = desc->vertBuffer; if (lastVBO->dynamic()) static_cast(lastVBO)->bindVertex(b); else static_cast(lastVBO)->bindVertex(); } if (desc->indexBuffer != lastEBO) { lastEBO = desc->indexBuffer; if (lastEBO->dynamic()) static_cast(lastEBO)->bindIndex(b); else static_cast(lastEBO)->bindIndex(); } glEnableVertexAttribArray(i); int maskedSem = int(desc->semantic & VertexSemantic::SemanticMask); if ((desc->semantic & VertexSemantic::Instanced) != VertexSemantic::None) { glVertexAttribPointer(i, SEMANTIC_COUNT_TABLE[maskedSem], SEMANTIC_TYPE_TABLE[maskedSem], GL_TRUE, instStride, (void*)instOffset); glVertexAttribDivisor(i, 1); instOffset += SEMANTIC_SIZE_TABLE[maskedSem]; } else { glVertexAttribPointer(i, SEMANTIC_COUNT_TABLE[maskedSem], SEMANTIC_TYPE_TABLE[maskedSem], GL_TRUE, stride, (void*)offset); offset += SEMANTIC_SIZE_TABLE[maskedSem]; } } } } static void ConfigureFBO(GLTextureR* tex) { glGenFramebuffers(1, &tex->m_fbo); glBindFramebuffer(GL_DRAW_FRAMEBUFFER, tex->m_fbo); glFramebufferTexture2D(GL_DRAW_FRAMEBUFFER, GL_COLOR_ATTACHMENT0, GL_TEXTURE_2D, tex->m_texs[0], 0); glFramebufferTexture2D(GL_DRAW_FRAMEBUFFER, GL_DEPTH_ATTACHMENT, GL_TEXTURE_2D, tex->m_texs[1], 0); } static void RenderingWorker(GLCommandQueue* self) { { std::unique_lock lk(self->m_initmt); self->m_parent->makeCurrent(); if (glewInit() != GLEW_OK) Log.report(LogVisor::FatalError, "unable to init glew"); const GLubyte* version = glGetString(GL_VERSION); Log.report(LogVisor::Info, "OpenGL Version: %s", version); self->m_parent->postInit(); } self->m_initcv.notify_one(); while (self->m_running) { std::vector> posts; { std::unique_lock lk(self->m_mt); self->m_cv.wait(lk); if (!self->m_running) break; self->m_drawBuf = self->m_completeBuf; if (self->m_pendingResizes.size()) { for (const RenderTextureResize& resize : self->m_pendingResizes) resize.tex->resize(resize.width, resize.height); self->m_pendingResizes.clear(); } if (self->m_pendingFmtAdds.size()) { for (GLVertexFormat* fmt : self->m_pendingFmtAdds) if (fmt) ConfigureVertexFormat(fmt); self->m_pendingFmtAdds.clear(); } if (self->m_pendingFmtDels.size()) { for (const auto& fmt : self->m_pendingFmtDels) glDeleteVertexArrays(3, fmt.data()); self->m_pendingFmtDels.clear(); } if (self->m_pendingFboAdds.size()) { for (GLTextureR* tex : self->m_pendingFboAdds) ConfigureFBO(tex); self->m_pendingFboAdds.clear(); } if (self->m_pendingFboDels.size()) { for (GLuint fbo : self->m_pendingFboDels) glDeleteFramebuffers(1, &fbo); self->m_pendingFboDels.clear(); } if (self->m_pendingPosts2.size()) posts.swap(self->m_pendingPosts2); } std::vector& cmds = self->m_cmdBufs[self->m_drawBuf]; for (const Command& cmd : cmds) { switch (cmd.m_op) { case Command::Op::SetShaderDataBinding: static_cast(cmd.binding)->bind(self->m_drawBuf); break; case Command::Op::SetRenderTarget: { const GLTextureR* tex = static_cast(cmd.target); if (!tex) glBindFramebuffer(GL_DRAW_FRAMEBUFFER, 0); else glBindFramebuffer(GL_DRAW_FRAMEBUFFER, tex->m_fbo); break; } case Command::Op::SetViewport: glViewport(cmd.rect.location[0], cmd.rect.location[1], cmd.rect.size[0], cmd.rect.size[1]); break; case Command::Op::SetScissor: if (cmd.rect.size[0] == 0 && cmd.rect.size[1] == 0) glDisable(GL_SCISSOR_TEST); else { glEnable(GL_SCISSOR_TEST); glScissor(cmd.rect.location[0], cmd.rect.location[1], cmd.rect.size[0], cmd.rect.size[1]); } break; case Command::Op::SetClearColor: glClearColor(cmd.rgba[0], cmd.rgba[1], cmd.rgba[2], cmd.rgba[3]); break; case Command::Op::ClearTarget: glClear(cmd.flags); break; case Command::Op::Draw: glDrawArrays(GL_TRIANGLE_STRIP, cmd.start, cmd.count); break; case Command::Op::DrawIndexed: glDrawElements(GL_TRIANGLE_STRIP, cmd.count, GL_UNSIGNED_INT, (void*)cmd.start); break; case Command::Op::DrawInstances: glDrawArraysInstanced(GL_TRIANGLE_STRIP, cmd.start, cmd.count, cmd.instCount); break; case Command::Op::DrawInstancesIndexed: glDrawElementsInstanced(GL_TRIANGLE_STRIP, cmd.count, GL_UNSIGNED_INT, (void*)cmd.start, cmd.instCount); break; case Command::Op::ResolveBindTexture: { const GLTextureR* tex = static_cast(cmd.resolveTex); GLenum target = (tex->m_samples > 1) ? GL_TEXTURE_2D_MULTISAMPLE : GL_TEXTURE_2D; glBindFramebuffer(GL_READ_FRAMEBUFFER, tex->m_fbo); glActiveTexture(GL_TEXTURE9); if (cmd.resolveColor && tex->m_bindTexs[0]) { glBindTexture(target, tex->m_bindTexs[0]); glCopyTexSubImage2D(target, 0, cmd.rect.location[0], cmd.rect.location[1], cmd.rect.location[0], cmd.rect.location[1], cmd.rect.size[0], cmd.rect.size[1]); } if (cmd.resolveDepth && tex->m_bindTexs[1]) { glBindTexture(target, tex->m_bindTexs[1]); glCopyTexSubImage2D(target, 0, cmd.rect.location[0], cmd.rect.location[1], cmd.rect.location[0], cmd.rect.location[1], cmd.rect.size[0], cmd.rect.size[1]); } break; } case Command::Op::Present: { const GLTextureR* tex = static_cast(cmd.source); if (tex) { glBindFramebuffer(GL_READ_FRAMEBUFFER, tex->m_fbo); glBindFramebuffer(GL_DRAW_FRAMEBUFFER, 0); glBlitFramebuffer(0, 0, tex->m_width, tex->m_height, 0, 0, tex->m_width, tex->m_height, GL_COLOR_BUFFER_BIT, GL_NEAREST); } self->m_parent->present(); break; } default: break; } } cmds.clear(); for (auto& p : posts) p(); } } GLCommandQueue(IGraphicsContext* parent) : m_parent(parent), m_initlk(m_initmt), m_thr(RenderingWorker, this) { m_initcv.wait(m_initlk); m_initlk.unlock(); } void stopRenderer() { m_running = false; m_cv.notify_one(); m_thr.join(); } ~GLCommandQueue() { if (m_running) stopRenderer(); } void setShaderDataBinding(IShaderDataBinding* binding) { std::vector& cmds = m_cmdBufs[m_fillBuf]; cmds.emplace_back(Command::Op::SetShaderDataBinding); cmds.back().binding = binding; } void setRenderTarget(ITextureR* target) { std::vector& cmds = m_cmdBufs[m_fillBuf]; cmds.emplace_back(Command::Op::SetRenderTarget); cmds.back().target = target; } void setViewport(const SWindowRect& rect) { std::vector& cmds = m_cmdBufs[m_fillBuf]; cmds.emplace_back(Command::Op::SetViewport); cmds.back().rect = rect; } void setScissor(const SWindowRect& rect) { std::vector& cmds = m_cmdBufs[m_fillBuf]; cmds.emplace_back(Command::Op::SetScissor); cmds.back().rect = rect; } void resizeRenderTexture(ITextureR* tex, size_t width, size_t height) { std::unique_lock lk(m_mt); GLTextureR* texgl = static_cast(tex); m_pendingResizes.push_back({texgl, width, height}); } void schedulePostFrameHandler(std::function&& func) { m_pendingPosts1.push_back(std::move(func)); } void setClearColor(const float rgba[4]) { std::vector& cmds = m_cmdBufs[m_fillBuf]; cmds.emplace_back(Command::Op::SetClearColor); cmds.back().rgba[0] = rgba[0]; cmds.back().rgba[1] = rgba[1]; cmds.back().rgba[2] = rgba[2]; cmds.back().rgba[3] = rgba[3]; } void clearTarget(bool render=true, bool depth=true) { std::vector& cmds = m_cmdBufs[m_fillBuf]; cmds.emplace_back(Command::Op::ClearTarget); cmds.back().flags = 0; if (render) cmds.back().flags |= GL_COLOR_BUFFER_BIT; if (depth) cmds.back().flags |= GL_DEPTH_BUFFER_BIT; } void draw(size_t start, size_t count) { std::vector& cmds = m_cmdBufs[m_fillBuf]; cmds.emplace_back(Command::Op::Draw); cmds.back().start = start; cmds.back().count = count; } void drawIndexed(size_t start, size_t count) { std::vector& cmds = m_cmdBufs[m_fillBuf]; cmds.emplace_back(Command::Op::DrawIndexed); cmds.back().start = start; cmds.back().count = count; } void drawInstances(size_t start, size_t count, size_t instCount) { std::vector& cmds = m_cmdBufs[m_fillBuf]; cmds.emplace_back(Command::Op::DrawInstances); cmds.back().start = start; cmds.back().count = count; cmds.back().instCount = instCount; } void drawInstancesIndexed(size_t start, size_t count, size_t instCount) { std::vector& cmds = m_cmdBufs[m_fillBuf]; cmds.emplace_back(Command::Op::DrawInstancesIndexed); cmds.back().start = start; cmds.back().count = count; cmds.back().instCount = instCount; } void resolveBindTexture(ITextureR* texture, const SWindowRect& rect, bool tlOrigin, bool color, bool depth) { GLTextureR* tex = static_cast(texture); std::vector& cmds = m_cmdBufs[m_fillBuf]; cmds.emplace_back(Command::Op::ResolveBindTexture); cmds.back().resolveTex = texture; cmds.back().resolveColor = color; cmds.back().resolveDepth = depth; SWindowRect intersectRect = rect.intersect(SWindowRect(0, 0, tex->m_width, tex->m_height)); SWindowRect& targetRect = cmds.back().rect; targetRect.location[0] = intersectRect.location[0]; if (tlOrigin) targetRect.location[1] = tex->m_height - intersectRect.location[1] - intersectRect.size[1]; else targetRect.location[1] = intersectRect.location[1]; targetRect.size[0] = intersectRect.size[0]; targetRect.size[1] = intersectRect.size[1]; } void resolveDisplay(ITextureR* source) { std::vector& cmds = m_cmdBufs[m_fillBuf]; cmds.emplace_back(Command::Op::Present); cmds.back().source = source; } void addVertexFormat(GLVertexFormat* fmt) { std::unique_lock lk(m_mt); m_pendingFmtAdds.push_back(fmt); } void delVertexFormat(GLVertexFormat* fmt) { std::unique_lock lk(m_mt); bool foundAdd = false; for (GLVertexFormat*& afmt : m_pendingFmtAdds) if (afmt == fmt) { foundAdd = true; afmt = nullptr; break; } if (!foundAdd) m_pendingFmtDels.push_back({fmt->m_vao[0], fmt->m_vao[1], fmt->m_vao[2]}); } void addFBO(GLTextureR* tex) { std::unique_lock lk(m_mt); m_pendingFboAdds.push_back(tex); } void delFBO(GLTextureR* tex) { std::unique_lock lk(m_mt); m_pendingFboDels.push_back(tex->m_fbo); } void execute() { std::unique_lock lk(m_mt); m_completeBuf = m_fillBuf; for (size_t i=0 ; i<3 ; ++i) { if (i == m_completeBuf || i == m_drawBuf) continue; m_fillBuf = i; break; } /* Update dynamic data here */ GLDataFactory* gfxF = static_cast(m_parent->getDataFactory()); std::unique_lock datalk(gfxF->m_committedMutex); for (GLData* d : gfxF->m_committedData) { for (std::unique_ptr& b : d->m_DBufs) b->update(m_completeBuf); for (std::unique_ptr& t : d->m_DTexs) t->update(m_completeBuf); } datalk.unlock(); glFlush(); for (auto& p : m_pendingPosts1) m_pendingPosts2.push_back(std::move(p)); m_pendingPosts1.clear(); lk.unlock(); m_cv.notify_one(); m_cmdBufs[m_fillBuf].clear(); } }; void GLGraphicsBufferD::update(int b) { int slot = 1 << b; if ((slot & m_validMask) == 0) { glBindBuffer(m_target, m_bufs[b]); glBufferSubData(m_target, 0, m_cpuSz, m_cpuBuf.get()); m_validMask |= slot; } } void GLGraphicsBufferD::load(const void* data, size_t sz) { size_t bufSz = std::min(sz, m_cpuSz); memcpy(m_cpuBuf.get(), data, bufSz); m_validMask = 0; } void* GLGraphicsBufferD::map(size_t sz) { if (sz < m_cpuSz) return nullptr; return m_cpuBuf.get(); } void GLGraphicsBufferD::unmap() { m_validMask = 0; } void GLGraphicsBufferD::bindVertex(int b) {glBindBuffer(GL_ARRAY_BUFFER, m_bufs[b]);} void GLGraphicsBufferD::bindIndex(int b) {glBindBuffer(GL_ELEMENT_ARRAY_BUFFER, m_bufs[b]);} void GLGraphicsBufferD::bindUniform(size_t idx, int b) {glBindBufferBase(GL_UNIFORM_BUFFER, idx, m_bufs[b]);} IGraphicsBufferD* GLDataFactory::newDynamicBuffer(BufferUse use, size_t stride, size_t count) { GLGraphicsBufferD* retval = new GLGraphicsBufferD(use, stride * count); if (!m_deferredData.get()) m_deferredData.reset(new struct GLData()); m_deferredData->m_DBufs.emplace_back(retval); return retval; } GLTextureD::GLTextureD(size_t width, size_t height, TextureFormat fmt) : m_width(width), m_height(height) { int pxPitch = 4; switch (fmt) { case TextureFormat::RGBA8: m_intFormat = GL_RGBA; m_format = GL_RGBA; pxPitch = 4; break; case TextureFormat::I8: m_intFormat = GL_R8; m_format = GL_RED; pxPitch = 1; break; default: Log.report(LogVisor::FatalError, "unsupported tex format"); } m_cpuSz = width * height * pxPitch; m_cpuBuf.reset(new uint8_t[m_cpuSz]); glGenTextures(3, m_texs); glBindTexture(GL_TEXTURE_2D, m_texs[0]); glTexImage2D(GL_TEXTURE_2D, 0, m_intFormat, width, height, 0, m_format, GL_UNSIGNED_BYTE, nullptr); glBindTexture(GL_TEXTURE_2D, m_texs[1]); glTexImage2D(GL_TEXTURE_2D, 0, m_intFormat, width, height, 0, m_format, GL_UNSIGNED_BYTE, nullptr); glBindTexture(GL_TEXTURE_2D, m_texs[2]); glTexImage2D(GL_TEXTURE_2D, 0, m_intFormat, width, height, 0, m_format, GL_UNSIGNED_BYTE, nullptr); } GLTextureD::~GLTextureD() {glDeleteTextures(3, m_texs);} void GLTextureD::update(int b) { int slot = 1 << b; if ((slot & m_validMask) == 0) { glBindTexture(GL_TEXTURE_2D, m_texs[b]); glTexImage2D(GL_TEXTURE_2D, 0, m_intFormat, m_width, m_height, 0, m_format, GL_UNSIGNED_BYTE, m_cpuBuf.get()); m_validMask |= slot; } } void GLTextureD::load(const void* data, size_t sz) { size_t bufSz = std::min(sz, m_cpuSz); memcpy(m_cpuBuf.get(), data, bufSz); m_validMask = 0; } void* GLTextureD::map(size_t sz) { if (sz > m_cpuSz) return nullptr; return m_cpuBuf.get(); } void GLTextureD::unmap() { m_validMask = 0; } void GLTextureD::bind(size_t idx, int b) { glActiveTexture(GL_TEXTURE0 + idx); glBindTexture(GL_TEXTURE_2D, m_texs[b]); } ITextureD* GLDataFactory::newDynamicTexture(size_t width, size_t height, TextureFormat fmt) { GLTextureD* retval = new GLTextureD(width, height, fmt); if (!m_deferredData.get()) m_deferredData.reset(new struct GLData()); m_deferredData->m_DTexs.emplace_back(retval); return retval; } GLTextureR::GLTextureR(GLCommandQueue* q, size_t width, size_t height, size_t samples, bool enableShaderColorBinding, bool enableShaderDepthBinding) : m_q(q), m_width(width), m_height(height), m_samples(samples) { glGenTextures(2, m_texs); if (enableShaderColorBinding) glGenTextures(1, &m_bindTexs[0]); if (enableShaderDepthBinding) glGenTextures(1, &m_bindTexs[1]); if (samples > 1) { m_target = GL_TEXTURE_2D_MULTISAMPLE; glBindTexture(GL_TEXTURE_2D_MULTISAMPLE, m_texs[0]); glTexImage2DMultisample(GL_TEXTURE_2D_MULTISAMPLE, samples, GL_RGBA, width, height, GL_FALSE); glBindTexture(GL_TEXTURE_2D_MULTISAMPLE, m_texs[1]); glTexImage2DMultisample(GL_TEXTURE_2D_MULTISAMPLE, samples, GL_DEPTH_COMPONENT24, width, height, GL_FALSE); if (enableShaderColorBinding) { glBindTexture(GL_TEXTURE_2D_MULTISAMPLE, m_bindTexs[0]); glTexImage2DMultisample(GL_TEXTURE_2D_MULTISAMPLE, samples, GL_RGBA, width, height, GL_FALSE); } if (enableShaderDepthBinding) { glBindTexture(GL_TEXTURE_2D_MULTISAMPLE, m_bindTexs[1]); glTexImage2DMultisample(GL_TEXTURE_2D_MULTISAMPLE, samples, GL_DEPTH_COMPONENT24, width, height, GL_FALSE); } } else { m_target = GL_TEXTURE_2D; glBindTexture(GL_TEXTURE_2D, m_texs[0]); glTexImage2D(GL_TEXTURE_2D, 0, GL_RGBA, width, height, 0, GL_RGBA, GL_UNSIGNED_BYTE, nullptr); glBindTexture(GL_TEXTURE_2D, m_texs[1]); glTexImage2D(GL_TEXTURE_2D, 0, GL_DEPTH_COMPONENT24, width, height, 0, GL_DEPTH_COMPONENT, GL_UNSIGNED_INT, nullptr); if (enableShaderColorBinding) { glBindTexture(GL_TEXTURE_2D, m_bindTexs[0]); glTexImage2D(GL_TEXTURE_2D, 0, GL_RGBA, width, height, 0, GL_RGBA, GL_UNSIGNED_BYTE, nullptr); glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MAG_FILTER, GL_LINEAR); glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, GL_LINEAR); } if (enableShaderDepthBinding) { glBindTexture(GL_TEXTURE_2D, m_bindTexs[1]); glTexImage2D(GL_TEXTURE_2D, 0, GL_DEPTH_COMPONENT24, width, height, 0, GL_DEPTH_COMPONENT, GL_UNSIGNED_INT, nullptr); glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MAG_FILTER, GL_LINEAR); glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, GL_LINEAR); } } m_q->addFBO(this); } GLTextureR::~GLTextureR() { glDeleteTextures(2, m_texs); glDeleteTextures(2, m_bindTexs); m_q->delFBO(this); } ITextureR* GLDataFactory::newRenderTexture(size_t width, size_t height, bool enableShaderColorBinding, bool enableShaderDepthBinding) { GLCommandQueue* q = static_cast(m_parent->getCommandQueue()); GLTextureR* retval = new GLTextureR(q, width, height, m_drawSamples, enableShaderColorBinding, enableShaderDepthBinding); q->resizeRenderTexture(retval, width, height); if (!m_deferredData.get()) m_deferredData.reset(new struct GLData()); m_deferredData->m_RTexs.emplace_back(retval); return retval; } GLVertexFormat::GLVertexFormat(GLCommandQueue* q, size_t elementCount, const VertexElementDescriptor* elements) : m_q(q), m_elementCount(elementCount), m_elements(new VertexElementDescriptor[elementCount]) { for (size_t i=0 ; iaddVertexFormat(this); } GLVertexFormat::~GLVertexFormat() {m_q->delVertexFormat(this);} IVertexFormat* GLDataFactory::newVertexFormat (size_t elementCount, const VertexElementDescriptor* elements) { GLCommandQueue* q = static_cast(m_parent->getCommandQueue()); GLVertexFormat* retval = new struct GLVertexFormat(q, elementCount, elements); if (!m_deferredData.get()) m_deferredData.reset(new struct GLData()); m_deferredData->m_VFmts.emplace_back(retval); return retval; } IGraphicsCommandQueue* _NewGLCommandQueue(IGraphicsContext* parent) { return new struct GLCommandQueue(parent); } }