diff options
Diffstat (limited to 'src/video_core/renderer_opengl')
36 files changed, 2009 insertions, 1480 deletions
diff --git a/src/video_core/renderer_opengl/gl_arb_decompiler.cpp b/src/video_core/renderer_opengl/gl_arb_decompiler.cpp index b7e9ed2e9..3e4d88c30 100644 --- a/src/video_core/renderer_opengl/gl_arb_decompiler.cpp +++ b/src/video_core/renderer_opengl/gl_arb_decompiler.cpp @@ -39,8 +39,8 @@ using Operation = const OperationNode&; constexpr std::array INTERNAL_FLAG_NAMES = {"ZERO", "SIGN", "CARRY", "OVERFLOW"}; char Swizzle(std::size_t component) { - ASSERT(component < 4); - return component["xyzw"]; + static constexpr std::string_view SWIZZLE{"xyzw"}; + return SWIZZLE.at(component); } constexpr bool IsGenericAttribute(Attribute::Index index) { @@ -71,7 +71,7 @@ std::string_view GetInputFlags(PixelImap attribute) { case PixelImap::Unused: break; } - UNIMPLEMENTED_MSG("Unknown attribute usage index={}", static_cast<int>(attribute)); + UNIMPLEMENTED_MSG("Unknown attribute usage index={}", attribute); return {}; } @@ -123,7 +123,7 @@ std::string_view PrimitiveDescription(Tegra::Engines::Maxwell3D::Regs::Primitive case Tegra::Engines::Maxwell3D::Regs::PrimitiveTopology::TriangleStripAdjacency: return "TRIANGLES_ADJACENCY"; default: - UNIMPLEMENTED_MSG("topology={}", static_cast<int>(topology)); + UNIMPLEMENTED_MSG("topology={}", topology); return "POINTS"; } } @@ -137,7 +137,7 @@ std::string_view TopologyName(Tegra::Shader::OutputTopology topology) { case Tegra::Shader::OutputTopology::TriangleStrip: return "TRIANGLE_STRIP"; default: - UNIMPLEMENTED_MSG("Unknown output topology: {}", static_cast<u32>(topology)); + UNIMPLEMENTED_MSG("Unknown output topology: {}", topology); return "points"; } } @@ -187,8 +187,8 @@ std::string TextureType(const MetaTexture& meta) { class ARBDecompiler final { public: - explicit ARBDecompiler(const Device& device, const ShaderIR& ir, const Registry& registry, - ShaderType stage, std::string_view identifier); + explicit ARBDecompiler(const Device& device_, const ShaderIR& ir_, const Registry& registry_, + ShaderType stage_, std::string_view identifier); std::string Code() const { return shader_source; @@ -224,7 +224,7 @@ private: std::string Visit(const Node& node); - std::pair<std::string, std::size_t> BuildCoords(Operation); + std::tuple<std::string, std::string, std::size_t> BuildCoords(Operation); std::string BuildAoffi(Operation); std::string GlobalMemoryPointer(const GmemNode& gmem); void Exit(); @@ -376,9 +376,11 @@ private: std::string temporary = AllocTemporary(); std::string address; std::string_view opname; + bool robust = false; if (const auto gmem = std::get_if<GmemNode>(&*operation[0])) { address = GlobalMemoryPointer(*gmem); opname = "ATOM"; + robust = true; } else if (const auto smem = std::get_if<SmemNode>(&*operation[0])) { address = fmt::format("shared_mem[{}]", Visit(smem->GetAddress())); opname = "ATOMS"; @@ -386,7 +388,15 @@ private: UNREACHABLE(); return "{0, 0, 0, 0}"; } + if (robust) { + AddLine("IF NE.x;"); + } AddLine("{}.{}.{} {}, {}, {};", opname, op, type, temporary, Visit(operation[1]), address); + if (robust) { + AddLine("ELSE;"); + AddLine("MOV.S {}, 0;", temporary); + AddLine("ENDIF;"); + } return temporary; } @@ -792,9 +802,9 @@ private: }; }; -ARBDecompiler::ARBDecompiler(const Device& device, const ShaderIR& ir, const Registry& registry, - ShaderType stage, std::string_view identifier) - : device{device}, ir{ir}, registry{registry}, stage{stage} { +ARBDecompiler::ARBDecompiler(const Device& device_, const ShaderIR& ir_, const Registry& registry_, + ShaderType stage_, std::string_view identifier) + : device{device_}, ir{ir_}, registry{registry_}, stage{stage_} { DefineGlobalMemory(); AddLine("TEMP RC;"); @@ -980,10 +990,9 @@ void ARBDecompiler::DeclareLocalMemory() { } void ARBDecompiler::DeclareGlobalMemory() { - const std::size_t num_entries = ir.GetGlobalMemory().size(); + const size_t num_entries = ir.GetGlobalMemory().size(); if (num_entries > 0) { - const std::size_t num_vectors = Common::AlignUp(num_entries, 2) / 2; - AddLine("PARAM c[{}] = {{ program.local[0..{}] }};", num_vectors, num_vectors - 1); + AddLine("PARAM c[{}] = {{ program.local[0..{}] }};", num_entries, num_entries - 1); } } @@ -1125,44 +1134,44 @@ void ARBDecompiler::VisitAST(const ASTNode& node) { for (ASTNode current = ast->nodes.GetFirst(); current; current = current->GetNext()) { VisitAST(current); } - } else if (const auto ast = std::get_if<ASTIfThen>(&*node->GetInnerData())) { - const std::string condition = VisitExpression(ast->condition); + } else if (const auto if_then = std::get_if<ASTIfThen>(&*node->GetInnerData())) { + const std::string condition = VisitExpression(if_then->condition); ResetTemporaries(); AddLine("MOVC.U RC.x, {};", condition); AddLine("IF NE.x;"); - for (ASTNode current = ast->nodes.GetFirst(); current; current = current->GetNext()) { + for (ASTNode current = if_then->nodes.GetFirst(); current; current = current->GetNext()) { VisitAST(current); } AddLine("ENDIF;"); - } else if (const auto ast = std::get_if<ASTIfElse>(&*node->GetInnerData())) { + } else if (const auto if_else = std::get_if<ASTIfElse>(&*node->GetInnerData())) { AddLine("ELSE;"); - for (ASTNode current = ast->nodes.GetFirst(); current; current = current->GetNext()) { + for (ASTNode current = if_else->nodes.GetFirst(); current; current = current->GetNext()) { VisitAST(current); } - } else if (const auto ast = std::get_if<ASTBlockDecoded>(&*node->GetInnerData())) { - VisitBlock(ast->nodes); - } else if (const auto ast = std::get_if<ASTVarSet>(&*node->GetInnerData())) { - AddLine("MOV.U F{}, {};", ast->index, VisitExpression(ast->condition)); + } else if (const auto decoded = std::get_if<ASTBlockDecoded>(&*node->GetInnerData())) { + VisitBlock(decoded->nodes); + } else if (const auto var_set = std::get_if<ASTVarSet>(&*node->GetInnerData())) { + AddLine("MOV.U F{}, {};", var_set->index, VisitExpression(var_set->condition)); ResetTemporaries(); - } else if (const auto ast = std::get_if<ASTDoWhile>(&*node->GetInnerData())) { - const std::string condition = VisitExpression(ast->condition); + } else if (const auto do_while = std::get_if<ASTDoWhile>(&*node->GetInnerData())) { + const std::string condition = VisitExpression(do_while->condition); ResetTemporaries(); AddLine("REP;"); - for (ASTNode current = ast->nodes.GetFirst(); current; current = current->GetNext()) { + for (ASTNode current = do_while->nodes.GetFirst(); current; current = current->GetNext()) { VisitAST(current); } AddLine("MOVC.U RC.x, {};", condition); AddLine("BRK (NE.x);"); AddLine("ENDREP;"); - } else if (const auto ast = std::get_if<ASTReturn>(&*node->GetInnerData())) { - const bool is_true = ExprIsTrue(ast->condition); + } else if (const auto ast_return = std::get_if<ASTReturn>(&*node->GetInnerData())) { + const bool is_true = ExprIsTrue(ast_return->condition); if (!is_true) { - AddLine("MOVC.U RC.x, {};", VisitExpression(ast->condition)); + AddLine("MOVC.U RC.x, {};", VisitExpression(ast_return->condition)); AddLine("IF NE.x;"); ResetTemporaries(); } - if (ast->kills) { + if (ast_return->kills) { AddLine("KIL TR;"); } else { Exit(); @@ -1170,11 +1179,11 @@ void ARBDecompiler::VisitAST(const ASTNode& node) { if (!is_true) { AddLine("ENDIF;"); } - } else if (const auto ast = std::get_if<ASTBreak>(&*node->GetInnerData())) { - if (ExprIsTrue(ast->condition)) { + } else if (const auto ast_break = std::get_if<ASTBreak>(&*node->GetInnerData())) { + if (ExprIsTrue(ast_break->condition)) { AddLine("BRK;"); } else { - AddLine("MOVC.U RC.x, {};", VisitExpression(ast->condition)); + AddLine("MOVC.U RC.x, {};", VisitExpression(ast_break->condition)); AddLine("BRK (NE.x);"); ResetTemporaries(); } @@ -1342,7 +1351,7 @@ std::string ARBDecompiler::Visit(const Node& node) { GetGenericAttributeIndex(index), swizzle); } } - UNIMPLEMENTED_MSG("Unimplemented input attribute={}", static_cast<int>(index)); + UNIMPLEMENTED_MSG("Unimplemented input attribute={}", index); break; } return "{0, 0, 0, 0}.x"; @@ -1363,7 +1372,8 @@ std::string ARBDecompiler::Visit(const Node& node) { if (const auto gmem = std::get_if<GmemNode>(&*node)) { std::string temporary = AllocTemporary(); - AddLine("LOAD.U32 {}, {};", temporary, GlobalMemoryPointer(*gmem)); + AddLine("MOV {}, 0;", temporary); + AddLine("LOAD.U32 {} (NE.x), {};", temporary, GlobalMemoryPointer(*gmem)); return temporary; } @@ -1406,12 +1416,12 @@ std::string ARBDecompiler::Visit(const Node& node) { return {}; } -std::pair<std::string, std::size_t> ARBDecompiler::BuildCoords(Operation operation) { +std::tuple<std::string, std::string, std::size_t> ARBDecompiler::BuildCoords(Operation operation) { const auto& meta = std::get<MetaTexture>(operation.GetMeta()); UNIMPLEMENTED_IF(meta.sampler.is_indexed); - UNIMPLEMENTED_IF(meta.sampler.is_shadow && meta.sampler.is_array && - meta.sampler.type == Tegra::Shader::TextureType::TextureCube); + const bool is_extended = meta.sampler.is_shadow && meta.sampler.is_array && + meta.sampler.type == Tegra::Shader::TextureType::TextureCube; const std::size_t count = operation.GetOperandsCount(); std::string temporary = AllocVectorTemporary(); std::size_t i = 0; @@ -1419,12 +1429,21 @@ std::pair<std::string, std::size_t> ARBDecompiler::BuildCoords(Operation operati AddLine("MOV.F {}.{}, {};", temporary, Swizzle(i), Visit(operation[i])); } if (meta.sampler.is_array) { - AddLine("I2F.S {}.{}, {};", temporary, Swizzle(i++), Visit(meta.array)); + AddLine("I2F.S {}.{}, {};", temporary, Swizzle(i), Visit(meta.array)); + ++i; } if (meta.sampler.is_shadow) { - AddLine("MOV.F {}.{}, {};", temporary, Swizzle(i++), Visit(meta.depth_compare)); + std::string compare = Visit(meta.depth_compare); + if (is_extended) { + ASSERT(i == 4); + std::string extra_coord = AllocVectorTemporary(); + AddLine("MOV.F {}.x, {};", extra_coord, compare); + return {fmt::format("{}, {}", temporary, extra_coord), extra_coord, 0}; + } + AddLine("MOV.F {}.{}, {};", temporary, Swizzle(i), compare); + ++i; } - return {std::move(temporary), i}; + return {temporary, temporary, i}; } std::string ARBDecompiler::BuildAoffi(Operation operation) { @@ -1441,18 +1460,21 @@ std::string ARBDecompiler::BuildAoffi(Operation operation) { } std::string ARBDecompiler::GlobalMemoryPointer(const GmemNode& gmem) { + // Read a bindless SSBO, return its address and set CC accordingly + // address = c[binding].xy + // length = c[binding].z const u32 binding = global_memory_names.at(gmem.GetDescriptor()); - const char result_swizzle = binding % 2 == 0 ? 'x' : 'y'; const std::string pointer = AllocLongVectorTemporary(); std::string temporary = AllocTemporary(); - const u32 local_index = binding / 2; - AddLine("PK64.U {}, c[{}];", pointer, local_index); + AddLine("PK64.U {}, c[{}];", pointer, binding); AddLine("SUB.U {}, {}, {};", temporary, Visit(gmem.GetRealAddress()), Visit(gmem.GetBaseAddress())); AddLine("CVT.U64.U32 {}.z, {};", pointer, temporary); - AddLine("ADD.U64 {}.x, {}.{}, {}.z;", pointer, pointer, result_swizzle, pointer); + AddLine("ADD.U64 {}.x, {}.x, {}.z;", pointer, pointer, pointer); + // Compare offset to length and set CC + AddLine("SLT.U.CC RC.x, {}, c[{}].z;", temporary, binding); return fmt::format("{}.x", pointer); } @@ -1463,9 +1485,7 @@ void ARBDecompiler::Exit() { } const auto safe_get_register = [this](u32 reg) -> std::string { - // TODO(Rodrigo): Replace with contains once C++20 releases - const auto& used_registers = ir.GetRegisters(); - if (used_registers.find(reg) != used_registers.end()) { + if (ir.GetRegisters().contains(reg)) { return fmt::format("R{}.x", reg); } return "{0, 0, 0, 0}.x"; @@ -1552,7 +1572,9 @@ std::string ARBDecompiler::Assign(Operation operation) { ResetTemporaries(); return {}; } else if (const auto gmem = std::get_if<GmemNode>(&*dest)) { + AddLine("IF NE.x;"); AddLine("STORE.U32 {}, {};", Visit(src), GlobalMemoryPointer(*gmem)); + AddLine("ENDIF;"); ResetTemporaries(); return {}; } else { @@ -1844,7 +1866,7 @@ std::string ARBDecompiler::LogicalAddCarry(Operation operation) { std::string ARBDecompiler::Texture(Operation operation) { const auto& meta = std::get<MetaTexture>(operation.GetMeta()); const u32 sampler_id = device.GetBaseBindings(stage).sampler + meta.sampler.index; - const auto [temporary, swizzle] = BuildCoords(operation); + const auto [coords, temporary, swizzle] = BuildCoords(operation); std::string_view opcode = "TEX"; std::string extra; @@ -1873,7 +1895,7 @@ std::string ARBDecompiler::Texture(Operation operation) { } } - AddLine("{}.F {}, {},{} texture[{}], {}{};", opcode, temporary, temporary, extra, sampler_id, + AddLine("{}.F {}, {},{} texture[{}], {}{};", opcode, temporary, coords, extra, sampler_id, TextureType(meta), BuildAoffi(operation)); AddLine("MOV.U {}.x, {}.{};", temporary, temporary, Swizzle(meta.element)); return fmt::format("{}.x", temporary); @@ -1882,7 +1904,7 @@ std::string ARBDecompiler::Texture(Operation operation) { std::string ARBDecompiler::TextureGather(Operation operation) { const auto& meta = std::get<MetaTexture>(operation.GetMeta()); const u32 sampler_id = device.GetBaseBindings(stage).sampler + meta.sampler.index; - const auto [temporary, swizzle] = BuildCoords(operation); + const auto [coords, temporary, swizzle] = BuildCoords(operation); std::string comp; if (!meta.sampler.is_shadow) { @@ -1892,7 +1914,7 @@ std::string ARBDecompiler::TextureGather(Operation operation) { AddLine("TXG.F {}, {}, texture[{}]{}, {}{};", temporary, temporary, sampler_id, comp, TextureType(meta), BuildAoffi(operation)); - AddLine("MOV.U {}.x, {}.{};", temporary, temporary, Swizzle(meta.element)); + AddLine("MOV.U {}.x, {}.{};", temporary, coords, Swizzle(meta.element)); return fmt::format("{}.x", temporary); } @@ -1930,13 +1952,13 @@ std::string ARBDecompiler::TextureQueryLod(Operation operation) { std::string ARBDecompiler::TexelFetch(Operation operation) { const auto& meta = std::get<MetaTexture>(operation.GetMeta()); const u32 sampler_id = device.GetBaseBindings(stage).sampler + meta.sampler.index; - const auto [temporary, swizzle] = BuildCoords(operation); + const auto [coords, temporary, swizzle] = BuildCoords(operation); if (!meta.sampler.is_buffer) { ASSERT(swizzle < 4); AddLine("MOV.F {}.w, {};", temporary, Visit(meta.lod)); } - AddLine("TXF.F {}, {}, texture[{}], {}{};", temporary, temporary, sampler_id, TextureType(meta), + AddLine("TXF.F {}, {}, texture[{}], {}{};", temporary, coords, sampler_id, TextureType(meta), BuildAoffi(operation)); AddLine("MOV.U {}.x, {}.{};", temporary, temporary, Swizzle(meta.element)); return fmt::format("{}.x", temporary); @@ -1947,7 +1969,7 @@ std::string ARBDecompiler::TextureGradient(Operation operation) { const u32 sampler_id = device.GetBaseBindings(stage).sampler + meta.sampler.index; const std::string ddx = AllocVectorTemporary(); const std::string ddy = AllocVectorTemporary(); - const std::string coord = BuildCoords(operation).first; + const std::string coord = std::get<1>(BuildCoords(operation)); const std::size_t num_components = meta.derivates.size() / 2; for (std::size_t index = 0; index < num_components; ++index) { diff --git a/src/video_core/renderer_opengl/gl_buffer_cache.cpp b/src/video_core/renderer_opengl/gl_buffer_cache.cpp index b1c4cd62f..5772cad87 100644 --- a/src/video_core/renderer_opengl/gl_buffer_cache.cpp +++ b/src/video_core/renderer_opengl/gl_buffer_cache.cpp @@ -22,11 +22,11 @@ using Maxwell = Tegra::Engines::Maxwell3D::Regs; MICROPROFILE_DEFINE(OpenGL_Buffer_Download, "OpenGL", "Buffer Download", MP_RGB(192, 192, 128)); -Buffer::Buffer(const Device& device, VAddr cpu_addr, std::size_t size) - : VideoCommon::BufferBlock{cpu_addr, size} { +Buffer::Buffer(const Device& device_, VAddr cpu_addr_, std::size_t size_) + : BufferBlock{cpu_addr_, size_} { gl_buffer.Create(); - glNamedBufferData(gl_buffer.handle, static_cast<GLsizeiptr>(size), nullptr, GL_DYNAMIC_DRAW); - if (device.UseAssemblyShaders() || device.HasVertexBufferUnifiedMemory()) { + glNamedBufferData(gl_buffer.handle, static_cast<GLsizeiptr>(size_), nullptr, GL_DYNAMIC_DRAW); + if (device_.UseAssemblyShaders() || device_.HasVertexBufferUnifiedMemory()) { glMakeNamedBufferResidentNV(gl_buffer.handle, GL_READ_WRITE); glGetNamedBufferParameterui64vNV(gl_buffer.handle, GL_BUFFER_GPU_ADDRESS_NV, &gpu_address); } @@ -34,14 +34,14 @@ Buffer::Buffer(const Device& device, VAddr cpu_addr, std::size_t size) Buffer::~Buffer() = default; -void Buffer::Upload(std::size_t offset, std::size_t size, const u8* data) { - glNamedBufferSubData(Handle(), static_cast<GLintptr>(offset), static_cast<GLsizeiptr>(size), - data); +void Buffer::Upload(std::size_t offset, std::size_t data_size, const u8* data) { + glNamedBufferSubData(Handle(), static_cast<GLintptr>(offset), + static_cast<GLsizeiptr>(data_size), data); } -void Buffer::Download(std::size_t offset, std::size_t size, u8* data) { +void Buffer::Download(std::size_t offset, std::size_t data_size, u8* data) { MICROPROFILE_SCOPE(OpenGL_Buffer_Download); - const GLsizeiptr gl_size = static_cast<GLsizeiptr>(size); + const GLsizeiptr gl_size = static_cast<GLsizeiptr>(data_size); const GLintptr gl_offset = static_cast<GLintptr>(offset); if (read_buffer.handle == 0) { read_buffer.Create(); @@ -54,17 +54,16 @@ void Buffer::Download(std::size_t offset, std::size_t size, u8* data) { } void Buffer::CopyFrom(const Buffer& src, std::size_t src_offset, std::size_t dst_offset, - std::size_t size) { + std::size_t copy_size) { glCopyNamedBufferSubData(src.Handle(), Handle(), static_cast<GLintptr>(src_offset), - static_cast<GLintptr>(dst_offset), static_cast<GLsizeiptr>(size)); + static_cast<GLintptr>(dst_offset), static_cast<GLsizeiptr>(copy_size)); } -OGLBufferCache::OGLBufferCache(VideoCore::RasterizerInterface& rasterizer, - Tegra::MemoryManager& gpu_memory, Core::Memory::Memory& cpu_memory, - const Device& device_, std::size_t stream_size) - : GenericBufferCache{rasterizer, gpu_memory, cpu_memory, - std::make_unique<OGLStreamBuffer>(device_, stream_size, true)}, - device{device_} { +OGLBufferCache::OGLBufferCache(VideoCore::RasterizerInterface& rasterizer_, + Tegra::MemoryManager& gpu_memory_, Core::Memory::Memory& cpu_memory_, + const Device& device_, OGLStreamBuffer& stream_buffer_, + StateTracker& state_tracker) + : GenericBufferCache{rasterizer_, gpu_memory_, cpu_memory_, stream_buffer_}, device{device_} { if (!device.HasFastBufferSubData()) { return; } diff --git a/src/video_core/renderer_opengl/gl_buffer_cache.h b/src/video_core/renderer_opengl/gl_buffer_cache.h index f75b32e31..17ee90316 100644 --- a/src/video_core/renderer_opengl/gl_buffer_cache.h +++ b/src/video_core/renderer_opengl/gl_buffer_cache.h @@ -22,18 +22,19 @@ namespace OpenGL { class Device; class OGLStreamBuffer; class RasterizerOpenGL; +class StateTracker; class Buffer : public VideoCommon::BufferBlock { public: - explicit Buffer(const Device& device, VAddr cpu_addr, std::size_t size); + explicit Buffer(const Device& device_, VAddr cpu_addr_, std::size_t size_); ~Buffer(); - void Upload(std::size_t offset, std::size_t size, const u8* data); + void Upload(std::size_t offset, std::size_t data_size, const u8* data); - void Download(std::size_t offset, std::size_t size, u8* data); + void Download(std::size_t offset, std::size_t data_size, u8* data); void CopyFrom(const Buffer& src, std::size_t src_offset, std::size_t dst_offset, - std::size_t size); + std::size_t copy_size); GLuint Handle() const noexcept { return gl_buffer.handle; @@ -54,7 +55,8 @@ class OGLBufferCache final : public GenericBufferCache { public: explicit OGLBufferCache(VideoCore::RasterizerInterface& rasterizer, Tegra::MemoryManager& gpu_memory, Core::Memory::Memory& cpu_memory, - const Device& device, std::size_t stream_size); + const Device& device, OGLStreamBuffer& stream_buffer, + StateTracker& state_tracker); ~OGLBufferCache(); BufferInfo GetEmptyBuffer(std::size_t) override; diff --git a/src/video_core/renderer_opengl/gl_device.cpp b/src/video_core/renderer_opengl/gl_device.cpp index e7d95149f..81b71edfb 100644 --- a/src/video_core/renderer_opengl/gl_device.cpp +++ b/src/video_core/renderer_opengl/gl_device.cpp @@ -5,9 +5,11 @@ #include <algorithm> #include <array> #include <cstddef> +#include <cstdlib> #include <cstring> #include <limits> #include <optional> +#include <span> #include <vector> #include <glad/glad.h> @@ -27,27 +29,29 @@ constexpr u32 ReservedUniformBlocks = 1; constexpr u32 NumStages = 5; -constexpr std::array LimitUBOs = { +constexpr std::array LIMIT_UBOS = { GL_MAX_VERTEX_UNIFORM_BLOCKS, GL_MAX_TESS_CONTROL_UNIFORM_BLOCKS, GL_MAX_TESS_EVALUATION_UNIFORM_BLOCKS, GL_MAX_GEOMETRY_UNIFORM_BLOCKS, - GL_MAX_FRAGMENT_UNIFORM_BLOCKS, GL_MAX_COMPUTE_UNIFORM_BLOCKS}; - -constexpr std::array LimitSSBOs = { + GL_MAX_FRAGMENT_UNIFORM_BLOCKS, GL_MAX_COMPUTE_UNIFORM_BLOCKS, +}; +constexpr std::array LIMIT_SSBOS = { GL_MAX_VERTEX_SHADER_STORAGE_BLOCKS, GL_MAX_TESS_CONTROL_SHADER_STORAGE_BLOCKS, GL_MAX_TESS_EVALUATION_SHADER_STORAGE_BLOCKS, GL_MAX_GEOMETRY_SHADER_STORAGE_BLOCKS, - GL_MAX_FRAGMENT_SHADER_STORAGE_BLOCKS, GL_MAX_COMPUTE_SHADER_STORAGE_BLOCKS}; - -constexpr std::array LimitSamplers = {GL_MAX_VERTEX_TEXTURE_IMAGE_UNITS, - GL_MAX_TESS_CONTROL_TEXTURE_IMAGE_UNITS, - GL_MAX_TESS_EVALUATION_TEXTURE_IMAGE_UNITS, - GL_MAX_GEOMETRY_TEXTURE_IMAGE_UNITS, - GL_MAX_TEXTURE_IMAGE_UNITS, - GL_MAX_COMPUTE_TEXTURE_IMAGE_UNITS}; - -constexpr std::array LimitImages = { + GL_MAX_FRAGMENT_SHADER_STORAGE_BLOCKS, GL_MAX_COMPUTE_SHADER_STORAGE_BLOCKS, +}; +constexpr std::array LIMIT_SAMPLERS = { + GL_MAX_VERTEX_TEXTURE_IMAGE_UNITS, + GL_MAX_TESS_CONTROL_TEXTURE_IMAGE_UNITS, + GL_MAX_TESS_EVALUATION_TEXTURE_IMAGE_UNITS, + GL_MAX_GEOMETRY_TEXTURE_IMAGE_UNITS, + GL_MAX_TEXTURE_IMAGE_UNITS, + GL_MAX_COMPUTE_TEXTURE_IMAGE_UNITS, +}; +constexpr std::array LIMIT_IMAGES = { GL_MAX_VERTEX_IMAGE_UNIFORMS, GL_MAX_TESS_CONTROL_IMAGE_UNIFORMS, GL_MAX_TESS_EVALUATION_IMAGE_UNIFORMS, GL_MAX_GEOMETRY_IMAGE_UNIFORMS, - GL_MAX_FRAGMENT_IMAGE_UNIFORMS, GL_MAX_COMPUTE_IMAGE_UNIFORMS}; + GL_MAX_FRAGMENT_IMAGE_UNIFORMS, GL_MAX_COMPUTE_IMAGE_UNIFORMS, +}; template <typename T> T GetInteger(GLenum pname) { @@ -76,8 +80,8 @@ std::vector<std::string_view> GetExtensions() { return extensions; } -bool HasExtension(const std::vector<std::string_view>& images, std::string_view extension) { - return std::find(images.begin(), images.end(), extension) != images.end(); +bool HasExtension(std::span<const std::string_view> extensions, std::string_view extension) { + return std::ranges::find(extensions, extension) != extensions.end(); } u32 Extract(u32& base, u32& num, u32 amount, std::optional<GLenum> limit = {}) { @@ -91,8 +95,8 @@ u32 Extract(u32& base, u32& num, u32 amount, std::optional<GLenum> limit = {}) { std::array<u32, Tegra::Engines::MaxShaderTypes> BuildMaxUniformBuffers() noexcept { std::array<u32, Tegra::Engines::MaxShaderTypes> max; - std::transform(LimitUBOs.begin(), LimitUBOs.end(), max.begin(), - [](GLenum pname) { return GetInteger<u32>(pname); }); + std::ranges::transform(LIMIT_UBOS, max.begin(), + [](GLenum pname) { return GetInteger<u32>(pname); }); return max; } @@ -115,9 +119,10 @@ std::array<Device::BaseBindings, Tegra::Engines::MaxShaderTypes> BuildBaseBindin for (std::size_t i = 0; i < NumStages; ++i) { const std::size_t stage = stage_swizzle[i]; bindings[stage] = { - Extract(base_ubo, num_ubos, total_ubos / NumStages, LimitUBOs[stage]), - Extract(base_ssbo, num_ssbos, total_ssbos / NumStages, LimitSSBOs[stage]), - Extract(base_samplers, num_samplers, total_samplers / NumStages, LimitSamplers[stage])}; + Extract(base_ubo, num_ubos, total_ubos / NumStages, LIMIT_UBOS[stage]), + Extract(base_ssbo, num_ssbos, total_ssbos / NumStages, LIMIT_SSBOS[stage]), + Extract(base_samplers, num_samplers, total_samplers / NumStages, + LIMIT_SAMPLERS[stage])}; } u32 num_images = GetInteger<u32>(GL_MAX_IMAGE_UNITS); @@ -130,7 +135,7 @@ std::array<Device::BaseBindings, Tegra::Engines::MaxShaderTypes> BuildBaseBindin // Reserve at least 4 image bindings on the fragment stage. bindings[4].image = - Extract(base_images, num_images, std::max(4U, num_images / NumStages), LimitImages[4]); + Extract(base_images, num_images, std::max(4U, num_images / NumStages), LIMIT_IMAGES[4]); // This is guaranteed to be at least 1. const u32 total_extracted_images = num_images / (NumStages - 1); @@ -142,7 +147,7 @@ std::array<Device::BaseBindings, Tegra::Engines::MaxShaderTypes> BuildBaseBindin continue; } bindings[stage].image = - Extract(base_images, num_images, total_extracted_images, LimitImages[stage]); + Extract(base_images, num_images, total_extracted_images, LIMIT_IMAGES[stage]); } // Compute doesn't care about any of this. @@ -188,17 +193,22 @@ bool IsASTCSupported() { return true; } +[[nodiscard]] bool IsDebugToolAttached(std::span<const std::string_view> extensions) { + const bool nsight = std::getenv("NVTX_INJECTION64_PATH") || std::getenv("NSIGHT_LAUNCHED"); + return nsight || HasExtension(extensions, "GL_EXT_debug_tool"); +} + } // Anonymous namespace Device::Device() : max_uniform_buffers{BuildMaxUniformBuffers()}, base_bindings{BuildBaseBindings()} { const std::string_view vendor = reinterpret_cast<const char*>(glGetString(GL_VENDOR)); - const std::string_view renderer = reinterpret_cast<const char*>(glGetString(GL_RENDERER)); const std::string_view version = reinterpret_cast<const char*>(glGetString(GL_VERSION)); const std::vector extensions = GetExtensions(); const bool is_nvidia = vendor == "NVIDIA Corporation"; const bool is_amd = vendor == "ATI Technologies Inc."; + const bool is_intel = vendor == "Intel"; bool disable_fast_buffer_sub_data = false; if (is_nvidia && version == "4.6.0 NVIDIA 443.24") { @@ -207,9 +217,8 @@ Device::Device() "Beta driver 443.24 is known to have issues. There might be performance issues."); disable_fast_buffer_sub_data = true; } - - uniform_buffer_alignment = GetInteger<std::size_t>(GL_UNIFORM_BUFFER_OFFSET_ALIGNMENT); - shader_storage_alignment = GetInteger<std::size_t>(GL_SHADER_STORAGE_BUFFER_OFFSET_ALIGNMENT); + uniform_buffer_alignment = GetInteger<size_t>(GL_UNIFORM_BUFFER_OFFSET_ALIGNMENT); + shader_storage_alignment = GetInteger<size_t>(GL_SHADER_STORAGE_BUFFER_OFFSET_ALIGNMENT); max_vertex_attributes = GetInteger<u32>(GL_MAX_VERTEX_ATTRIBS); max_varyings = GetInteger<u32>(GL_MAX_VARYING_VECTORS); max_compute_shared_memory_size = GetInteger<u32>(GL_MAX_COMPUTE_SHARED_MEMORY_SIZE); @@ -223,8 +232,10 @@ Device::Device() has_variable_aoffi = TestVariableAoffi(); has_component_indexing_bug = is_amd; has_precise_bug = TestPreciseBug(); + has_broken_texture_view_formats = is_amd || is_intel; has_nv_viewport_array2 = GLAD_GL_NV_viewport_array2; has_vertex_buffer_unified_memory = GLAD_GL_NV_vertex_buffer_unified_memory; + has_debugging_tool_attached = IsDebugToolAttached(extensions); // At the moment of writing this, only Nvidia's driver optimizes BufferSubData on exclusive // uniform buffers as "push constants" @@ -239,6 +250,8 @@ Device::Device() LOG_INFO(Render_OpenGL, "Renderer_VariableAOFFI: {}", has_variable_aoffi); LOG_INFO(Render_OpenGL, "Renderer_ComponentIndexingBug: {}", has_component_indexing_bug); LOG_INFO(Render_OpenGL, "Renderer_PreciseBug: {}", has_precise_bug); + LOG_INFO(Render_OpenGL, "Renderer_BrokenTextureViewFormats: {}", + has_broken_texture_view_formats); if (Settings::values.use_assembly_shaders.GetValue() && !use_assembly_shaders) { LOG_ERROR(Render_OpenGL, "Assembly shaders enabled but not supported"); diff --git a/src/video_core/renderer_opengl/gl_device.h b/src/video_core/renderer_opengl/gl_device.h index 8a4b6b9fc..3e79d1e37 100644 --- a/src/video_core/renderer_opengl/gl_device.h +++ b/src/video_core/renderer_opengl/gl_device.h @@ -36,11 +36,11 @@ public: return GetBaseBindings(static_cast<std::size_t>(shader_type)); } - std::size_t GetUniformBufferAlignment() const { + size_t GetUniformBufferAlignment() const { return uniform_buffer_alignment; } - std::size_t GetShaderStorageBufferAlignment() const { + size_t GetShaderStorageBufferAlignment() const { return shader_storage_alignment; } @@ -96,6 +96,10 @@ public: return has_precise_bug; } + bool HasBrokenTextureViewFormats() const { + return has_broken_texture_view_formats; + } + bool HasFastBufferSubData() const { return has_fast_buffer_sub_data; } @@ -104,6 +108,10 @@ public: return has_nv_viewport_array2; } + bool HasDebuggingToolAttached() const { + return has_debugging_tool_attached; + } + bool UseAssemblyShaders() const { return use_assembly_shaders; } @@ -118,8 +126,8 @@ private: std::array<u32, Tegra::Engines::MaxShaderTypes> max_uniform_buffers{}; std::array<BaseBindings, Tegra::Engines::MaxShaderTypes> base_bindings{}; - std::size_t uniform_buffer_alignment{}; - std::size_t shader_storage_alignment{}; + size_t uniform_buffer_alignment{}; + size_t shader_storage_alignment{}; u32 max_vertex_attributes{}; u32 max_varyings{}; u32 max_compute_shared_memory_size{}; @@ -133,8 +141,10 @@ private: bool has_variable_aoffi{}; bool has_component_indexing_bug{}; bool has_precise_bug{}; + bool has_broken_texture_view_formats{}; bool has_fast_buffer_sub_data{}; bool has_nv_viewport_array2{}; + bool has_debugging_tool_attached{}; bool use_assembly_shaders{}; bool use_asynchronous_shaders{}; }; diff --git a/src/video_core/renderer_opengl/gl_fence_manager.cpp b/src/video_core/renderer_opengl/gl_fence_manager.cpp index b532fdcc2..3e9c922f5 100644 --- a/src/video_core/renderer_opengl/gl_fence_manager.cpp +++ b/src/video_core/renderer_opengl/gl_fence_manager.cpp @@ -11,10 +11,10 @@ namespace OpenGL { -GLInnerFence::GLInnerFence(u32 payload, bool is_stubbed) : FenceBase(payload, is_stubbed) {} +GLInnerFence::GLInnerFence(u32 payload_, bool is_stubbed_) : FenceBase{payload_, is_stubbed_} {} -GLInnerFence::GLInnerFence(GPUVAddr address, u32 payload, bool is_stubbed) - : FenceBase(address, payload, is_stubbed) {} +GLInnerFence::GLInnerFence(GPUVAddr address_, u32 payload_, bool is_stubbed_) + : FenceBase{address_, payload_, is_stubbed_} {} GLInnerFence::~GLInnerFence() = default; @@ -45,10 +45,10 @@ void GLInnerFence::Wait() { glClientWaitSync(sync_object.handle, 0, GL_TIMEOUT_IGNORED); } -FenceManagerOpenGL::FenceManagerOpenGL(VideoCore::RasterizerInterface& rasterizer, Tegra::GPU& gpu, - TextureCacheOpenGL& texture_cache, - OGLBufferCache& buffer_cache, QueryCache& query_cache) - : GenericFenceManager{rasterizer, gpu, texture_cache, buffer_cache, query_cache} {} +FenceManagerOpenGL::FenceManagerOpenGL(VideoCore::RasterizerInterface& rasterizer_, + Tegra::GPU& gpu_, TextureCache& texture_cache_, + OGLBufferCache& buffer_cache_, QueryCache& query_cache_) + : GenericFenceManager{rasterizer_, gpu_, texture_cache_, buffer_cache_, query_cache_} {} Fence FenceManagerOpenGL::CreateFence(u32 value, bool is_stubbed) { return std::make_shared<GLInnerFence>(value, is_stubbed); diff --git a/src/video_core/renderer_opengl/gl_fence_manager.h b/src/video_core/renderer_opengl/gl_fence_manager.h index da1dcdace..30dbee613 100644 --- a/src/video_core/renderer_opengl/gl_fence_manager.h +++ b/src/video_core/renderer_opengl/gl_fence_manager.h @@ -17,8 +17,8 @@ namespace OpenGL { class GLInnerFence : public VideoCommon::FenceBase { public: - GLInnerFence(u32 payload, bool is_stubbed); - GLInnerFence(GPUVAddr address, u32 payload, bool is_stubbed); + explicit GLInnerFence(u32 payload_, bool is_stubbed_); + explicit GLInnerFence(GPUVAddr address_, u32 payload_, bool is_stubbed_); ~GLInnerFence(); void Queue(); @@ -33,13 +33,13 @@ private: using Fence = std::shared_ptr<GLInnerFence>; using GenericFenceManager = - VideoCommon::FenceManager<Fence, TextureCacheOpenGL, OGLBufferCache, QueryCache>; + VideoCommon::FenceManager<Fence, TextureCache, OGLBufferCache, QueryCache>; class FenceManagerOpenGL final : public GenericFenceManager { public: - explicit FenceManagerOpenGL(VideoCore::RasterizerInterface& rasterizer, Tegra::GPU& gpu, - TextureCacheOpenGL& texture_cache, OGLBufferCache& buffer_cache, - QueryCache& query_cache); + explicit FenceManagerOpenGL(VideoCore::RasterizerInterface& rasterizer_, Tegra::GPU& gpu_, + TextureCache& texture_cache_, OGLBufferCache& buffer_cache_, + QueryCache& query_cache_); protected: Fence CreateFence(u32 value, bool is_stubbed) override; diff --git a/src/video_core/renderer_opengl/gl_framebuffer_cache.cpp b/src/video_core/renderer_opengl/gl_framebuffer_cache.cpp deleted file mode 100644 index b8a512cb6..000000000 --- a/src/video_core/renderer_opengl/gl_framebuffer_cache.cpp +++ /dev/null @@ -1,85 +0,0 @@ -// Copyright 2019 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. - -#include <tuple> -#include <unordered_map> -#include <utility> - -#include <glad/glad.h> - -#include "common/common_types.h" -#include "video_core/engines/maxwell_3d.h" -#include "video_core/renderer_opengl/gl_framebuffer_cache.h" - -namespace OpenGL { - -using Maxwell = Tegra::Engines::Maxwell3D::Regs; -using VideoCore::Surface::SurfaceType; - -FramebufferCacheOpenGL::FramebufferCacheOpenGL() = default; - -FramebufferCacheOpenGL::~FramebufferCacheOpenGL() = default; - -GLuint FramebufferCacheOpenGL::GetFramebuffer(const FramebufferCacheKey& key) { - const auto [entry, is_cache_miss] = cache.try_emplace(key); - auto& framebuffer{entry->second}; - if (is_cache_miss) { - framebuffer = CreateFramebuffer(key); - } - return framebuffer.handle; -} - -OGLFramebuffer FramebufferCacheOpenGL::CreateFramebuffer(const FramebufferCacheKey& key) { - OGLFramebuffer framebuffer; - framebuffer.Create(); - - // TODO(Rodrigo): Use DSA here after Nvidia fixes their framebuffer DSA bugs. - glBindFramebuffer(GL_DRAW_FRAMEBUFFER, framebuffer.handle); - - if (key.zeta) { - const bool stencil = key.zeta->GetSurfaceParams().type == SurfaceType::DepthStencil; - const GLenum attach_target = stencil ? GL_DEPTH_STENCIL_ATTACHMENT : GL_DEPTH_ATTACHMENT; - key.zeta->Attach(attach_target, GL_DRAW_FRAMEBUFFER); - } - - std::size_t num_buffers = 0; - std::array<GLenum, Maxwell::NumRenderTargets> targets; - - for (std::size_t index = 0; index < Maxwell::NumRenderTargets; ++index) { - if (!key.colors[index]) { - targets[index] = GL_NONE; - continue; - } - const GLenum attach_target = GL_COLOR_ATTACHMENT0 + static_cast<GLenum>(index); - key.colors[index]->Attach(attach_target, GL_DRAW_FRAMEBUFFER); - - const u32 attachment = (key.color_attachments >> (BitsPerAttachment * index)) & 0b1111; - targets[index] = GL_COLOR_ATTACHMENT0 + attachment; - num_buffers = index + 1; - } - - if (num_buffers > 0) { - glDrawBuffers(static_cast<GLsizei>(num_buffers), std::data(targets)); - } else { - glDrawBuffer(GL_NONE); - } - - return framebuffer; -} - -std::size_t FramebufferCacheKey::Hash() const noexcept { - std::size_t hash = std::hash<View>{}(zeta); - for (const auto& color : colors) { - hash ^= std::hash<View>{}(color); - } - hash ^= static_cast<std::size_t>(color_attachments) << 16; - return hash; -} - -bool FramebufferCacheKey::operator==(const FramebufferCacheKey& rhs) const noexcept { - return std::tie(colors, zeta, color_attachments) == - std::tie(rhs.colors, rhs.zeta, rhs.color_attachments); -} - -} // namespace OpenGL diff --git a/src/video_core/renderer_opengl/gl_framebuffer_cache.h b/src/video_core/renderer_opengl/gl_framebuffer_cache.h deleted file mode 100644 index 8f698fee0..000000000 --- a/src/video_core/renderer_opengl/gl_framebuffer_cache.h +++ /dev/null @@ -1,68 +0,0 @@ -// Copyright 2019 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. - -#pragma once - -#include <array> -#include <cstddef> -#include <unordered_map> - -#include <glad/glad.h> - -#include "common/common_types.h" -#include "video_core/engines/maxwell_3d.h" -#include "video_core/renderer_opengl/gl_resource_manager.h" -#include "video_core/renderer_opengl/gl_texture_cache.h" - -namespace OpenGL { - -constexpr std::size_t BitsPerAttachment = 4; - -struct FramebufferCacheKey { - View zeta; - std::array<View, Tegra::Engines::Maxwell3D::Regs::NumRenderTargets> colors; - u32 color_attachments = 0; - - std::size_t Hash() const noexcept; - - bool operator==(const FramebufferCacheKey& rhs) const noexcept; - - bool operator!=(const FramebufferCacheKey& rhs) const noexcept { - return !operator==(rhs); - } - - void SetAttachment(std::size_t index, u32 attachment) { - color_attachments |= attachment << (BitsPerAttachment * index); - } -}; - -} // namespace OpenGL - -namespace std { - -template <> -struct hash<OpenGL::FramebufferCacheKey> { - std::size_t operator()(const OpenGL::FramebufferCacheKey& k) const noexcept { - return k.Hash(); - } -}; - -} // namespace std - -namespace OpenGL { - -class FramebufferCacheOpenGL { -public: - FramebufferCacheOpenGL(); - ~FramebufferCacheOpenGL(); - - GLuint GetFramebuffer(const FramebufferCacheKey& key); - -private: - OGLFramebuffer CreateFramebuffer(const FramebufferCacheKey& key); - - std::unordered_map<FramebufferCacheKey, OGLFramebuffer> cache; -}; - -} // namespace OpenGL diff --git a/src/video_core/renderer_opengl/gl_query_cache.cpp b/src/video_core/renderer_opengl/gl_query_cache.cpp index 1a3d9720e..acebbf5f4 100644 --- a/src/video_core/renderer_opengl/gl_query_cache.cpp +++ b/src/video_core/renderer_opengl/gl_query_cache.cpp @@ -30,11 +30,9 @@ constexpr GLenum GetTarget(VideoCore::QueryType type) { } // Anonymous namespace -QueryCache::QueryCache(RasterizerOpenGL& rasterizer, Tegra::Engines::Maxwell3D& maxwell3d, - Tegra::MemoryManager& gpu_memory) - : VideoCommon::QueryCacheBase<QueryCache, CachedQuery, CounterStream, HostCounter>( - rasterizer, maxwell3d, gpu_memory), - gl_rasterizer{rasterizer} {} +QueryCache::QueryCache(RasterizerOpenGL& rasterizer_, Tegra::Engines::Maxwell3D& maxwell3d_, + Tegra::MemoryManager& gpu_memory_) + : QueryCacheBase(rasterizer_, maxwell3d_, gpu_memory_), gl_rasterizer{rasterizer_} {} QueryCache::~QueryCache() = default; @@ -59,10 +57,11 @@ bool QueryCache::AnyCommandQueued() const noexcept { return gl_rasterizer.AnyCommandQueued(); } -HostCounter::HostCounter(QueryCache& cache, std::shared_ptr<HostCounter> dependency, - VideoCore::QueryType type) - : VideoCommon::HostCounterBase<QueryCache, HostCounter>{std::move(dependency)}, cache{cache}, - type{type}, query{cache.AllocateQuery(type)} { +HostCounter::HostCounter(QueryCache& cache_, std::shared_ptr<HostCounter> dependency_, + VideoCore::QueryType type_) + : HostCounterBase{std::move(dependency_)}, cache{cache_}, type{type_}, query{ + cache.AllocateQuery( + type)} { glBeginQuery(GetTarget(type), query.handle); } @@ -86,13 +85,14 @@ u64 HostCounter::BlockingQuery() const { return static_cast<u64>(value); } -CachedQuery::CachedQuery(QueryCache& cache, VideoCore::QueryType type, VAddr cpu_addr, u8* host_ptr) - : VideoCommon::CachedQueryBase<HostCounter>{cpu_addr, host_ptr}, cache{&cache}, type{type} {} +CachedQuery::CachedQuery(QueryCache& cache_, VideoCore::QueryType type_, VAddr cpu_addr_, + u8* host_ptr_) + : CachedQueryBase{cpu_addr_, host_ptr_}, cache{&cache_}, type{type_} {} CachedQuery::~CachedQuery() = default; CachedQuery::CachedQuery(CachedQuery&& rhs) noexcept - : VideoCommon::CachedQueryBase<HostCounter>(std::move(rhs)), cache{rhs.cache}, type{rhs.type} {} + : CachedQueryBase(std::move(rhs)), cache{rhs.cache}, type{rhs.type} {} CachedQuery& CachedQuery::operator=(CachedQuery&& rhs) noexcept { cache = rhs.cache; diff --git a/src/video_core/renderer_opengl/gl_query_cache.h b/src/video_core/renderer_opengl/gl_query_cache.h index 82cac51ee..7bbe5cfe9 100644 --- a/src/video_core/renderer_opengl/gl_query_cache.h +++ b/src/video_core/renderer_opengl/gl_query_cache.h @@ -29,8 +29,8 @@ using CounterStream = VideoCommon::CounterStreamBase<QueryCache, HostCounter>; class QueryCache final : public VideoCommon::QueryCacheBase<QueryCache, CachedQuery, CounterStream, HostCounter> { public: - explicit QueryCache(RasterizerOpenGL& rasterizer, Tegra::Engines::Maxwell3D& maxwell3d, - Tegra::MemoryManager& gpu_memory); + explicit QueryCache(RasterizerOpenGL& rasterizer_, Tegra::Engines::Maxwell3D& maxwell3d_, + Tegra::MemoryManager& gpu_memory_); ~QueryCache(); OGLQuery AllocateQuery(VideoCore::QueryType type); @@ -46,8 +46,8 @@ private: class HostCounter final : public VideoCommon::HostCounterBase<QueryCache, HostCounter> { public: - explicit HostCounter(QueryCache& cache, std::shared_ptr<HostCounter> dependency, - VideoCore::QueryType type); + explicit HostCounter(QueryCache& cache_, std::shared_ptr<HostCounter> dependency_, + VideoCore::QueryType type_); ~HostCounter(); void EndQuery(); @@ -62,8 +62,8 @@ private: class CachedQuery final : public VideoCommon::CachedQueryBase<HostCounter> { public: - explicit CachedQuery(QueryCache& cache, VideoCore::QueryType type, VAddr cpu_addr, - u8* host_ptr); + explicit CachedQuery(QueryCache& cache_, VideoCore::QueryType type_, VAddr cpu_addr_, + u8* host_ptr_); ~CachedQuery() override; CachedQuery(CachedQuery&& rhs) noexcept; diff --git a/src/video_core/renderer_opengl/gl_rasterizer.cpp b/src/video_core/renderer_opengl/gl_rasterizer.cpp index bbb2eb17c..8aa63d329 100644 --- a/src/video_core/renderer_opengl/gl_rasterizer.cpp +++ b/src/video_core/renderer_opengl/gl_rasterizer.cpp @@ -25,12 +25,15 @@ #include "video_core/engines/maxwell_3d.h" #include "video_core/engines/shader_type.h" #include "video_core/memory_manager.h" +#include "video_core/renderer_opengl/gl_device.h" #include "video_core/renderer_opengl/gl_query_cache.h" #include "video_core/renderer_opengl/gl_rasterizer.h" #include "video_core/renderer_opengl/gl_shader_cache.h" +#include "video_core/renderer_opengl/gl_texture_cache.h" #include "video_core/renderer_opengl/maxwell_to_gl.h" #include "video_core/renderer_opengl/renderer_opengl.h" #include "video_core/shader_cache.h" +#include "video_core/texture_cache/texture_cache.h" namespace OpenGL { @@ -55,18 +58,32 @@ MICROPROFILE_DEFINE(OpenGL_PrimitiveAssembly, "OpenGL", "Prim Asmbl", MP_RGB(255 namespace { -constexpr std::size_t NUM_CONST_BUFFERS_PER_STAGE = 18; -constexpr std::size_t NUM_CONST_BUFFERS_BYTES_PER_STAGE = +constexpr size_t NUM_CONST_BUFFERS_PER_STAGE = 18; +constexpr size_t NUM_CONST_BUFFERS_BYTES_PER_STAGE = NUM_CONST_BUFFERS_PER_STAGE * Maxwell::MaxConstBufferSize; -constexpr std::size_t TOTAL_CONST_BUFFER_BYTES = +constexpr size_t TOTAL_CONST_BUFFER_BYTES = NUM_CONST_BUFFERS_BYTES_PER_STAGE * Maxwell::MaxShaderStage; -constexpr std::size_t NUM_SUPPORTED_VERTEX_ATTRIBUTES = 16; -constexpr std::size_t NUM_SUPPORTED_VERTEX_BINDINGS = 16; +constexpr size_t NUM_SUPPORTED_VERTEX_ATTRIBUTES = 16; +constexpr size_t NUM_SUPPORTED_VERTEX_BINDINGS = 16; + +constexpr size_t MAX_TEXTURES = 192; +constexpr size_t MAX_IMAGES = 48; + +struct TextureHandle { + constexpr TextureHandle(u32 data, bool via_header_index) { + const Tegra::Texture::TextureHandle handle{data}; + image = handle.tic_id; + sampler = via_header_index ? image : handle.tsc_id.Value(); + } + + u32 image; + u32 sampler; +}; template <typename Engine, typename Entry> -Tegra::Texture::FullTextureInfo GetTextureInfo(const Engine& engine, const Entry& entry, - ShaderType shader_type, std::size_t index = 0) { +TextureHandle GetTextureInfo(const Engine& engine, bool via_header_index, const Entry& entry, + ShaderType shader_type, size_t index = 0) { if constexpr (std::is_same_v<Entry, SamplerEntry>) { if (entry.is_separated) { const u32 buffer_1 = entry.buffer; @@ -75,21 +92,16 @@ Tegra::Texture::FullTextureInfo GetTextureInfo(const Engine& engine, const Entry const u32 offset_2 = entry.secondary_offset; const u32 handle_1 = engine.AccessConstBuffer32(shader_type, buffer_1, offset_1); const u32 handle_2 = engine.AccessConstBuffer32(shader_type, buffer_2, offset_2); - return engine.GetTextureInfo(handle_1 | handle_2); + return TextureHandle(handle_1 | handle_2, via_header_index); } } if (entry.is_bindless) { - const u32 handle = engine.AccessConstBuffer32(shader_type, entry.buffer, entry.offset); - return engine.GetTextureInfo(handle); - } - - const auto& gpu_profile = engine.AccessGuestDriverProfile(); - const u32 offset = entry.offset + static_cast<u32>(index * gpu_profile.GetTextureHandlerSize()); - if constexpr (std::is_same_v<Engine, Tegra::Engines::Maxwell3D>) { - return engine.GetStageTexture(shader_type, offset); - } else { - return engine.GetTexture(offset); + const u32 raw = engine.AccessConstBuffer32(shader_type, entry.buffer, entry.offset); + return TextureHandle(raw, via_header_index); } + const u32 buffer = engine.GetBoundBuffer(); + const u64 offset = (entry.offset + index) * sizeof(u32); + return TextureHandle(engine.AccessConstBuffer32(shader_type, buffer, offset), via_header_index); } std::size_t GetConstBufferSize(const Tegra::Engines::ConstBufferInfo& buffer, @@ -97,7 +109,6 @@ std::size_t GetConstBufferSize(const Tegra::Engines::ConstBufferInfo& buffer, if (!entry.IsIndirect()) { return entry.GetSize(); } - if (buffer.size > Maxwell::MaxConstBufferSize) { LOG_WARNING(Render_OpenGL, "Indirect constbuffer size {} exceeds maximum {}", buffer.size, Maxwell::MaxConstBufferSize); @@ -131,7 +142,7 @@ std::pair<GLint, GLint> TransformFeedbackEnum(u8 location) { case 43: return {GL_BACK_SECONDARY_COLOR_NV, 0}; } - UNIMPLEMENTED_MSG("index={}", static_cast<int>(index)); + UNIMPLEMENTED_MSG("index={}", index); return {GL_POSITION, 0}; } @@ -139,35 +150,68 @@ void oglEnable(GLenum cap, bool state) { (state ? glEnable : glDisable)(cap); } -void UpdateBindlessPointers(GLenum target, GLuint64EXT* pointers, std::size_t num_entries) { - if (num_entries == 0) { +void UpdateBindlessSSBOs(GLenum target, const BindlessSSBO* ssbos, size_t num_ssbos) { + if (num_ssbos == 0) { return; } - if (num_entries % 2 == 1) { - pointers[num_entries] = 0; + glProgramLocalParametersI4uivNV(target, 0, static_cast<GLsizei>(num_ssbos), + reinterpret_cast<const GLuint*>(ssbos)); +} + +ImageViewType ImageViewTypeFromEntry(const SamplerEntry& entry) { + if (entry.is_buffer) { + return ImageViewType::Buffer; + } + switch (entry.type) { + case Tegra::Shader::TextureType::Texture1D: + return entry.is_array ? ImageViewType::e1DArray : ImageViewType::e1D; + case Tegra::Shader::TextureType::Texture2D: + return entry.is_array ? ImageViewType::e2DArray : ImageViewType::e2D; + case Tegra::Shader::TextureType::Texture3D: + return ImageViewType::e3D; + case Tegra::Shader::TextureType::TextureCube: + return entry.is_array ? ImageViewType::CubeArray : ImageViewType::Cube; + } + UNREACHABLE(); + return ImageViewType::e2D; +} + +ImageViewType ImageViewTypeFromEntry(const ImageEntry& entry) { + switch (entry.type) { + case Tegra::Shader::ImageType::Texture1D: + return ImageViewType::e1D; + case Tegra::Shader::ImageType::Texture1DArray: + return ImageViewType::e1DArray; + case Tegra::Shader::ImageType::Texture2D: + return ImageViewType::e2D; + case Tegra::Shader::ImageType::Texture2DArray: + return ImageViewType::e2DArray; + case Tegra::Shader::ImageType::Texture3D: + return ImageViewType::e3D; + case Tegra::Shader::ImageType::TextureBuffer: + return ImageViewType::Buffer; } - const GLsizei num_vectors = static_cast<GLsizei>((num_entries + 1) / 2); - glProgramLocalParametersI4uivNV(target, 0, num_vectors, - reinterpret_cast<const GLuint*>(pointers)); + UNREACHABLE(); + return ImageViewType::e2D; } } // Anonymous namespace -RasterizerOpenGL::RasterizerOpenGL(Core::Frontend::EmuWindow& emu_window, Tegra::GPU& gpu_, - Core::Memory::Memory& cpu_memory, const Device& device_, +RasterizerOpenGL::RasterizerOpenGL(Core::Frontend::EmuWindow& emu_window_, Tegra::GPU& gpu_, + Core::Memory::Memory& cpu_memory_, const Device& device_, ScreenInfo& screen_info_, ProgramManager& program_manager_, StateTracker& state_tracker_) - : RasterizerAccelerated{cpu_memory}, gpu(gpu_), maxwell3d(gpu.Maxwell3D()), + : RasterizerAccelerated(cpu_memory_), gpu(gpu_), maxwell3d(gpu.Maxwell3D()), kepler_compute(gpu.KeplerCompute()), gpu_memory(gpu.MemoryManager()), device(device_), screen_info(screen_info_), program_manager(program_manager_), state_tracker(state_tracker_), - texture_cache(*this, maxwell3d, gpu_memory, device, state_tracker), - shader_cache(*this, emu_window, gpu, maxwell3d, kepler_compute, gpu_memory, device), + stream_buffer(device, state_tracker), + texture_cache_runtime(device, program_manager, state_tracker), + texture_cache(texture_cache_runtime, *this, maxwell3d, kepler_compute, gpu_memory), + shader_cache(*this, emu_window_, gpu, maxwell3d, kepler_compute, gpu_memory, device), query_cache(*this, maxwell3d, gpu_memory), - buffer_cache(*this, gpu_memory, cpu_memory, device, STREAM_BUFFER_SIZE), + buffer_cache(*this, gpu_memory, cpu_memory_, device, stream_buffer, state_tracker), fence_manager(*this, gpu, texture_cache, buffer_cache, query_cache), - async_shaders(emu_window) { - CheckExtensions(); - + async_shaders(emu_window_) { unified_uniform_buffer.Create(); glNamedBufferStorage(unified_uniform_buffer.handle, TOTAL_CONST_BUFFER_BYTES, nullptr, 0); @@ -178,7 +222,6 @@ RasterizerOpenGL::RasterizerOpenGL(Core::Frontend::EmuWindow& emu_window, Tegra: nullptr, 0); } } - if (device.UseAsynchronousShaders()) { async_shaders.AllocateWorkers(); } @@ -190,14 +233,6 @@ RasterizerOpenGL::~RasterizerOpenGL() { } } -void RasterizerOpenGL::CheckExtensions() { - if (!GLAD_GL_ARB_texture_filter_anisotropic && !GLAD_GL_EXT_texture_filter_anisotropic) { - LOG_WARNING( - Render_OpenGL, - "Anisotropic filter is not supported! This can cause graphical issues in some games."); - } -} - void RasterizerOpenGL::SetupVertexFormat() { auto& flags = maxwell3d.dirty.flags; if (!flags[Dirty::VertexFormats]) { @@ -320,10 +355,16 @@ GLintptr RasterizerOpenGL::SetupIndexBuffer() { return info.offset; } -void RasterizerOpenGL::SetupShaders(GLenum primitive_mode) { +void RasterizerOpenGL::SetupShaders() { MICROPROFILE_SCOPE(OpenGL_Shader); u32 clip_distances = 0; + std::array<Shader*, Maxwell::MaxShaderStage> shaders{}; + image_view_indices.clear(); + sampler_handles.clear(); + + texture_cache.SynchronizeGraphicsDescriptors(); + for (std::size_t index = 0; index < Maxwell::MaxShaderProgram; ++index) { const auto& shader_config = maxwell3d.regs.shader_config[index]; const auto program{static_cast<Maxwell::ShaderProgram>(index)}; @@ -342,7 +383,6 @@ void RasterizerOpenGL::SetupShaders(GLenum primitive_mode) { } continue; } - // Currently this stages are not supported in the OpenGL backend. // TODO(Blinkhawk): Port tesselation shaders from Vulkan to OpenGL if (program == Maxwell::ShaderProgram::TesselationControl || @@ -351,7 +391,6 @@ void RasterizerOpenGL::SetupShaders(GLenum primitive_mode) { } Shader* const shader = shader_cache.GetStageProgram(program, async_shaders); - const GLuint program_handle = shader->IsBuilt() ? shader->GetHandle() : 0; switch (program) { case Maxwell::ShaderProgram::VertexA: @@ -367,14 +406,17 @@ void RasterizerOpenGL::SetupShaders(GLenum primitive_mode) { default: UNIMPLEMENTED_MSG("Unimplemented shader index={}, enable={}, offset=0x{:08X}", index, shader_config.enable.Value(), shader_config.offset); + break; } // Stage indices are 0 - 5 - const std::size_t stage = index == 0 ? 0 : index - 1; + const size_t stage = index == 0 ? 0 : index - 1; + shaders[stage] = shader; + SetupDrawConstBuffers(stage, shader); SetupDrawGlobalMemory(stage, shader); - SetupDrawTextures(stage, shader); - SetupDrawImages(stage, shader); + SetupDrawTextures(shader, stage); + SetupDrawImages(shader, stage); // Workaround for Intel drivers. // When a clip distance is enabled but not set in the shader it crops parts of the screen @@ -388,9 +430,23 @@ void RasterizerOpenGL::SetupShaders(GLenum primitive_mode) { ++index; } } - SyncClipEnabled(clip_distances); maxwell3d.dirty.flags[Dirty::Shaders] = false; + + const std::span indices_span(image_view_indices.data(), image_view_indices.size()); + texture_cache.FillGraphicsImageViews(indices_span, image_view_ids); + + size_t image_view_index = 0; + size_t texture_index = 0; + size_t image_index = 0; + for (size_t stage = 0; stage < Maxwell::MaxShaderStage; ++stage) { + const Shader* const shader = shaders[stage]; + if (shader) { + const auto base = device.GetBaseBindings(stage); + BindTextures(shader->GetEntries(), base.sampler, base.image, image_view_index, + texture_index, image_index); + } + } } std::size_t RasterizerOpenGL::CalculateVertexArraysSize() const { @@ -421,98 +477,6 @@ void RasterizerOpenGL::LoadDiskResources(u64 title_id, const std::atomic_bool& s shader_cache.LoadDiskCache(title_id, stop_loading, callback); } -void RasterizerOpenGL::ConfigureFramebuffers() { - MICROPROFILE_SCOPE(OpenGL_Framebuffer); - if (!maxwell3d.dirty.flags[VideoCommon::Dirty::RenderTargets]) { - return; - } - maxwell3d.dirty.flags[VideoCommon::Dirty::RenderTargets] = false; - - texture_cache.GuardRenderTargets(true); - - View depth_surface = texture_cache.GetDepthBufferSurface(true); - - const auto& regs = maxwell3d.regs; - UNIMPLEMENTED_IF(regs.rt_separate_frag_data == 0); - - // Bind the framebuffer surfaces - FramebufferCacheKey key; - const auto colors_count = static_cast<std::size_t>(regs.rt_control.count); - for (std::size_t index = 0; index < colors_count; ++index) { - View color_surface{texture_cache.GetColorBufferSurface(index, true)}; - if (!color_surface) { - continue; - } - // Assume that a surface will be written to if it is used as a framebuffer, even - // if the shader doesn't actually write to it. - texture_cache.MarkColorBufferInUse(index); - - key.SetAttachment(index, regs.rt_control.GetMap(index)); - key.colors[index] = std::move(color_surface); - } - - if (depth_surface) { - // Assume that a surface will be written to if it is used as a framebuffer, even if - // the shader doesn't actually write to it. - texture_cache.MarkDepthBufferInUse(); - key.zeta = std::move(depth_surface); - } - - texture_cache.GuardRenderTargets(false); - - glBindFramebuffer(GL_DRAW_FRAMEBUFFER, framebuffer_cache.GetFramebuffer(key)); -} - -void RasterizerOpenGL::ConfigureClearFramebuffer(bool using_color, bool using_depth_stencil) { - const auto& regs = maxwell3d.regs; - - texture_cache.GuardRenderTargets(true); - View color_surface; - - if (using_color) { - // Determine if we have to preserve the contents. - // First we have to make sure all clear masks are enabled. - bool preserve_contents = !regs.clear_buffers.R || !regs.clear_buffers.G || - !regs.clear_buffers.B || !regs.clear_buffers.A; - const std::size_t index = regs.clear_buffers.RT; - if (regs.clear_flags.scissor) { - // Then we have to confirm scissor testing clears the whole image. - const auto& scissor = regs.scissor_test[0]; - preserve_contents |= scissor.min_x > 0; - preserve_contents |= scissor.min_y > 0; - preserve_contents |= scissor.max_x < regs.rt[index].width; - preserve_contents |= scissor.max_y < regs.rt[index].height; - } - - color_surface = texture_cache.GetColorBufferSurface(index, preserve_contents); - texture_cache.MarkColorBufferInUse(index); - } - - View depth_surface; - if (using_depth_stencil) { - bool preserve_contents = false; - if (regs.clear_flags.scissor) { - // For depth stencil clears we only have to confirm scissor test covers the whole image. - const auto& scissor = regs.scissor_test[0]; - preserve_contents |= scissor.min_x > 0; - preserve_contents |= scissor.min_y > 0; - preserve_contents |= scissor.max_x < regs.zeta_width; - preserve_contents |= scissor.max_y < regs.zeta_height; - } - - depth_surface = texture_cache.GetDepthBufferSurface(preserve_contents); - texture_cache.MarkDepthBufferInUse(); - } - texture_cache.GuardRenderTargets(false); - - FramebufferCacheKey key; - key.colors[0] = std::move(color_surface); - key.zeta = std::move(depth_surface); - - state_tracker.NotifyFramebuffer(); - glBindFramebuffer(GL_DRAW_FRAMEBUFFER, framebuffer_cache.GetFramebuffer(key)); -} - void RasterizerOpenGL::Clear() { if (!maxwell3d.ShouldExecute()) { return; @@ -527,8 +491,9 @@ void RasterizerOpenGL::Clear() { regs.clear_buffers.A) { use_color = true; - state_tracker.NotifyColorMask0(); - glColorMaski(0, regs.clear_buffers.R != 0, regs.clear_buffers.G != 0, + const GLuint index = regs.clear_buffers.RT; + state_tracker.NotifyColorMask(index); + glColorMaski(index, regs.clear_buffers.R != 0, regs.clear_buffers.G != 0, regs.clear_buffers.B != 0, regs.clear_buffers.A != 0); // TODO(Rodrigo): Determine if clamping is used on clears @@ -561,15 +526,17 @@ void RasterizerOpenGL::Clear() { state_tracker.NotifyScissor0(); glDisablei(GL_SCISSOR_TEST, 0); } - UNIMPLEMENTED_IF(regs.clear_flags.viewport); - ConfigureClearFramebuffer(use_color, use_depth || use_stencil); + { + auto lock = texture_cache.AcquireLock(); + texture_cache.UpdateRenderTargets(true); + state_tracker.BindFramebuffer(texture_cache.GetFramebuffer()->Handle()); + } if (use_color) { - glClearBufferfv(GL_COLOR, 0, regs.clear_color); + glClearBufferfv(GL_COLOR, regs.clear_buffers.RT, regs.clear_color); } - if (use_depth && use_stencil) { glClearBufferfi(GL_DEPTH_STENCIL, 0, regs.clear_depth, regs.clear_stencil); } else if (use_depth) { @@ -626,16 +593,7 @@ void RasterizerOpenGL::Draw(bool is_indexed, bool is_instanced) { (Maxwell::MaxConstBufferSize + device.GetUniformBufferAlignment()); // Prepare the vertex array. - const bool invalidated = buffer_cache.Map(buffer_size); - - if (invalidated) { - // When the stream buffer has been invalidated, we have to consider vertex buffers as dirty - auto& dirty = maxwell3d.dirty.flags; - dirty[Dirty::VertexBuffers] = true; - for (int index = Dirty::VertexBuffer0; index <= Dirty::VertexBuffer31; ++index) { - dirty[index] = true; - } - } + buffer_cache.Map(buffer_size); // Prepare vertex array format. SetupVertexFormat(); @@ -659,22 +617,16 @@ void RasterizerOpenGL::Draw(bool is_indexed, bool is_instanced) { } // Setup shaders and their used resources. - texture_cache.GuardSamplers(true); - const GLenum primitive_mode = MaxwellToGL::PrimitiveTopology(maxwell3d.regs.draw.topology); - SetupShaders(primitive_mode); - texture_cache.GuardSamplers(false); - - ConfigureFramebuffers(); + auto lock = texture_cache.AcquireLock(); + SetupShaders(); // Signal the buffer cache that we are not going to upload more things. buffer_cache.Unmap(); - + texture_cache.UpdateRenderTargets(false); + state_tracker.BindFramebuffer(texture_cache.GetFramebuffer()->Handle()); program_manager.BindGraphicsPipeline(); - if (texture_cache.TextureBarrier()) { - glTextureBarrier(); - } - + const GLenum primitive_mode = MaxwellToGL::PrimitiveTopology(maxwell3d.regs.draw.topology); BeginTransformFeedback(primitive_mode); const GLuint base_instance = static_cast<GLuint>(maxwell3d.regs.vb_base_instance); @@ -726,15 +678,13 @@ void RasterizerOpenGL::DispatchCompute(GPUVAddr code_addr) { buffer_cache.Acquire(); current_cbuf = 0; - auto kernel = shader_cache.GetComputeKernel(code_addr); - program_manager.BindCompute(kernel->GetHandle()); + Shader* const kernel = shader_cache.GetComputeKernel(code_addr); - SetupComputeTextures(kernel); - SetupComputeImages(kernel); + auto lock = texture_cache.AcquireLock(); + BindComputeTextures(kernel); - const std::size_t buffer_size = - Tegra::Engines::KeplerCompute::NumConstBuffers * - (Maxwell::MaxConstBufferSize + device.GetUniformBufferAlignment()); + const size_t buffer_size = Tegra::Engines::KeplerCompute::NumConstBuffers * + (Maxwell::MaxConstBufferSize + device.GetUniformBufferAlignment()); buffer_cache.Map(buffer_size); SetupComputeConstBuffers(kernel); @@ -743,7 +693,6 @@ void RasterizerOpenGL::DispatchCompute(GPUVAddr code_addr) { buffer_cache.Unmap(); const auto& launch_desc = kepler_compute.launch_description; - program_manager.BindCompute(kernel->GetHandle()); glDispatchCompute(launch_desc.grid_dim_x, launch_desc.grid_dim_y, launch_desc.grid_dim_z); ++num_queued_commands; } @@ -764,7 +713,10 @@ void RasterizerOpenGL::FlushRegion(VAddr addr, u64 size) { if (addr == 0 || size == 0) { return; } - texture_cache.FlushRegion(addr, size); + { + auto lock = texture_cache.AcquireLock(); + texture_cache.DownloadMemory(addr, size); + } buffer_cache.FlushRegion(addr, size); query_cache.FlushRegion(addr, size); } @@ -773,7 +725,8 @@ bool RasterizerOpenGL::MustFlushRegion(VAddr addr, u64 size) { if (!Settings::IsGPULevelHigh()) { return buffer_cache.MustFlushRegion(addr, size); } - return texture_cache.MustFlushRegion(addr, size) || buffer_cache.MustFlushRegion(addr, size); + return texture_cache.IsRegionGpuModified(addr, size) || + buffer_cache.MustFlushRegion(addr, size); } void RasterizerOpenGL::InvalidateRegion(VAddr addr, u64 size) { @@ -781,7 +734,10 @@ void RasterizerOpenGL::InvalidateRegion(VAddr addr, u64 size) { if (addr == 0 || size == 0) { return; } - texture_cache.InvalidateRegion(addr, size); + { + auto lock = texture_cache.AcquireLock(); + texture_cache.WriteMemory(addr, size); + } shader_cache.InvalidateRegion(addr, size); buffer_cache.InvalidateRegion(addr, size); query_cache.InvalidateRegion(addr, size); @@ -792,18 +748,29 @@ void RasterizerOpenGL::OnCPUWrite(VAddr addr, u64 size) { if (addr == 0 || size == 0) { return; } - texture_cache.OnCPUWrite(addr, size); + { + auto lock = texture_cache.AcquireLock(); + texture_cache.WriteMemory(addr, size); + } shader_cache.OnCPUWrite(addr, size); buffer_cache.OnCPUWrite(addr, size); } void RasterizerOpenGL::SyncGuestHost() { MICROPROFILE_SCOPE(OpenGL_CacheManagement); - texture_cache.SyncGuestHost(); buffer_cache.SyncGuestHost(); shader_cache.SyncGuestHost(); } +void RasterizerOpenGL::UnmapMemory(VAddr addr, u64 size) { + { + auto lock = texture_cache.AcquireLock(); + texture_cache.UnmapMemory(addr, size); + } + buffer_cache.OnCPUWrite(addr, size); + shader_cache.OnCPUWrite(addr, size); +} + void RasterizerOpenGL::SignalSemaphore(GPUVAddr addr, u32 value) { if (!gpu.IsAsync()) { gpu_memory.Write<u32>(addr, value); @@ -845,6 +812,14 @@ void RasterizerOpenGL::WaitForIdle() { GL_SHADER_STORAGE_BARRIER_BIT | GL_QUERY_BUFFER_BARRIER_BIT); } +void RasterizerOpenGL::FragmentBarrier() { + glMemoryBarrier(GL_FRAMEBUFFER_BARRIER_BIT); +} + +void RasterizerOpenGL::TiledCacheBarrier() { + glTextureBarrier(); +} + void RasterizerOpenGL::FlushCommands() { // Only flush when we have commands queued to OpenGL. if (num_queued_commands == 0) { @@ -858,53 +833,103 @@ void RasterizerOpenGL::TickFrame() { // Ticking a frame means that buffers will be swapped, calling glFlush implicitly. num_queued_commands = 0; + fence_manager.TickFrame(); buffer_cache.TickFrame(); + { + auto lock = texture_cache.AcquireLock(); + texture_cache.TickFrame(); + } } -bool RasterizerOpenGL::AccelerateSurfaceCopy(const Tegra::Engines::Fermi2D::Regs::Surface& src, - const Tegra::Engines::Fermi2D::Regs::Surface& dst, +bool RasterizerOpenGL::AccelerateSurfaceCopy(const Tegra::Engines::Fermi2D::Surface& src, + const Tegra::Engines::Fermi2D::Surface& dst, const Tegra::Engines::Fermi2D::Config& copy_config) { MICROPROFILE_SCOPE(OpenGL_Blits); - texture_cache.DoFermiCopy(src, dst, copy_config); + auto lock = texture_cache.AcquireLock(); + texture_cache.BlitImage(dst, src, copy_config); return true; } bool RasterizerOpenGL::AccelerateDisplay(const Tegra::FramebufferConfig& config, VAddr framebuffer_addr, u32 pixel_stride) { - if (!framebuffer_addr) { - return {}; + if (framebuffer_addr == 0) { + return false; } - MICROPROFILE_SCOPE(OpenGL_CacheManagement); - const auto surface{texture_cache.TryFindFramebufferSurface(framebuffer_addr)}; - if (!surface) { - return {}; + auto lock = texture_cache.AcquireLock(); + ImageView* const image_view{texture_cache.TryFindFramebufferImageView(framebuffer_addr)}; + if (!image_view) { + return false; } - // Verify that the cached surface is the same size and format as the requested framebuffer - const auto& params{surface->GetSurfaceParams()}; - const auto& pixel_format{ - VideoCore::Surface::PixelFormatFromGPUPixelFormat(config.pixel_format)}; - ASSERT_MSG(params.width == config.width, "Framebuffer width is different"); - ASSERT_MSG(params.height == config.height, "Framebuffer height is different"); + // ASSERT_MSG(image_view->size.width == config.width, "Framebuffer width is different"); + // ASSERT_MSG(image_view->size.height == config.height, "Framebuffer height is different"); - if (params.pixel_format != pixel_format) { - LOG_DEBUG(Render_OpenGL, "Framebuffer pixel_format is different"); - } + screen_info.display_texture = image_view->Handle(ImageViewType::e2D); + screen_info.display_srgb = VideoCore::Surface::IsPixelFormatSRGB(image_view->format); + return true; +} - screen_info.display_texture = surface->GetTexture(); - screen_info.display_srgb = surface->GetSurfaceParams().srgb_conversion; +void RasterizerOpenGL::BindComputeTextures(Shader* kernel) { + image_view_indices.clear(); + sampler_handles.clear(); - return true; + texture_cache.SynchronizeComputeDescriptors(); + + SetupComputeTextures(kernel); + SetupComputeImages(kernel); + + const std::span indices_span(image_view_indices.data(), image_view_indices.size()); + texture_cache.FillComputeImageViews(indices_span, image_view_ids); + + program_manager.BindCompute(kernel->GetHandle()); + size_t image_view_index = 0; + size_t texture_index = 0; + size_t image_index = 0; + BindTextures(kernel->GetEntries(), 0, 0, image_view_index, texture_index, image_index); +} + +void RasterizerOpenGL::BindTextures(const ShaderEntries& entries, GLuint base_texture, + GLuint base_image, size_t& image_view_index, + size_t& texture_index, size_t& image_index) { + const GLuint* const samplers = sampler_handles.data() + texture_index; + const GLuint* const textures = texture_handles.data() + texture_index; + const GLuint* const images = image_handles.data() + image_index; + + const size_t num_samplers = entries.samplers.size(); + for (const auto& sampler : entries.samplers) { + for (size_t i = 0; i < sampler.size; ++i) { + const ImageViewId image_view_id = image_view_ids[image_view_index++]; + const ImageView& image_view = texture_cache.GetImageView(image_view_id); + const GLuint handle = image_view.Handle(ImageViewTypeFromEntry(sampler)); + texture_handles[texture_index++] = handle; + } + } + const size_t num_images = entries.images.size(); + for (size_t unit = 0; unit < num_images; ++unit) { + // TODO: Mark as modified + const ImageViewId image_view_id = image_view_ids[image_view_index++]; + const ImageView& image_view = texture_cache.GetImageView(image_view_id); + const GLuint handle = image_view.Handle(ImageViewTypeFromEntry(entries.images[unit])); + image_handles[image_index] = handle; + ++image_index; + } + if (num_samplers > 0) { + glBindSamplers(base_texture, static_cast<GLsizei>(num_samplers), samplers); + glBindTextures(base_texture, static_cast<GLsizei>(num_samplers), textures); + } + if (num_images > 0) { + glBindImageTextures(base_image, static_cast<GLsizei>(num_images), images); + } } void RasterizerOpenGL::SetupDrawConstBuffers(std::size_t stage_index, Shader* shader) { - static constexpr std::array PARAMETER_LUT = { - GL_VERTEX_PROGRAM_PARAMETER_BUFFER_NV, GL_TESS_CONTROL_PROGRAM_PARAMETER_BUFFER_NV, + static constexpr std::array PARAMETER_LUT{ + GL_VERTEX_PROGRAM_PARAMETER_BUFFER_NV, GL_TESS_CONTROL_PROGRAM_PARAMETER_BUFFER_NV, GL_TESS_EVALUATION_PROGRAM_PARAMETER_BUFFER_NV, GL_GEOMETRY_PROGRAM_PARAMETER_BUFFER_NV, - GL_FRAGMENT_PROGRAM_PARAMETER_BUFFER_NV}; - + GL_FRAGMENT_PROGRAM_PARAMETER_BUFFER_NV, + }; MICROPROFILE_SCOPE(OpenGL_UBO); const auto& stages = maxwell3d.state.shader_stages; const auto& shader_stage = stages[stage_index]; @@ -1003,12 +1028,11 @@ void RasterizerOpenGL::SetupDrawGlobalMemory(std::size_t stage_index, Shader* sh GL_VERTEX_PROGRAM_NV, GL_TESS_CONTROL_PROGRAM_NV, GL_TESS_EVALUATION_PROGRAM_NV, GL_GEOMETRY_PROGRAM_NV, GL_FRAGMENT_PROGRAM_NV, }; - const auto& cbufs{maxwell3d.state.shader_stages[stage_index]}; const auto& entries{shader->GetEntries().global_memory_entries}; - std::array<GLuint64EXT, 32> pointers; - ASSERT(entries.size() < pointers.size()); + std::array<BindlessSSBO, 32> ssbos; + ASSERT(entries.size() < ssbos.size()); const bool assembly_shaders = device.UseAssemblyShaders(); u32 binding = assembly_shaders ? 0 : device.GetBaseBindings(stage_index).shader_storage_buffer; @@ -1016,11 +1040,11 @@ void RasterizerOpenGL::SetupDrawGlobalMemory(std::size_t stage_index, Shader* sh const GPUVAddr addr{cbufs.const_buffers[entry.cbuf_index].address + entry.cbuf_offset}; const GPUVAddr gpu_addr{gpu_memory.Read<u64>(addr)}; const u32 size{gpu_memory.Read<u32>(addr + 8)}; - SetupGlobalMemory(binding, entry, gpu_addr, size, &pointers[binding]); + SetupGlobalMemory(binding, entry, gpu_addr, size, &ssbos[binding]); ++binding; } if (assembly_shaders) { - UpdateBindlessPointers(TARGET_LUT[stage_index], pointers.data(), entries.size()); + UpdateBindlessSSBOs(TARGET_LUT[stage_index], ssbos.data(), entries.size()); } } @@ -1028,106 +1052,85 @@ void RasterizerOpenGL::SetupComputeGlobalMemory(Shader* kernel) { const auto& cbufs{kepler_compute.launch_description.const_buffer_config}; const auto& entries{kernel->GetEntries().global_memory_entries}; - std::array<GLuint64EXT, 32> pointers; - ASSERT(entries.size() < pointers.size()); + std::array<BindlessSSBO, 32> ssbos; + ASSERT(entries.size() < ssbos.size()); u32 binding = 0; for (const auto& entry : entries) { const GPUVAddr addr{cbufs[entry.cbuf_index].Address() + entry.cbuf_offset}; const GPUVAddr gpu_addr{gpu_memory.Read<u64>(addr)}; const u32 size{gpu_memory.Read<u32>(addr + 8)}; - SetupGlobalMemory(binding, entry, gpu_addr, size, &pointers[binding]); + SetupGlobalMemory(binding, entry, gpu_addr, size, &ssbos[binding]); ++binding; } if (device.UseAssemblyShaders()) { - UpdateBindlessPointers(GL_COMPUTE_PROGRAM_NV, pointers.data(), entries.size()); + UpdateBindlessSSBOs(GL_COMPUTE_PROGRAM_NV, ssbos.data(), ssbos.size()); } } void RasterizerOpenGL::SetupGlobalMemory(u32 binding, const GlobalMemoryEntry& entry, - GPUVAddr gpu_addr, std::size_t size, - GLuint64EXT* pointer) { - const std::size_t alignment{device.GetShaderStorageBufferAlignment()}; + GPUVAddr gpu_addr, size_t size, BindlessSSBO* ssbo) { + const size_t alignment{device.GetShaderStorageBufferAlignment()}; const auto info = buffer_cache.UploadMemory(gpu_addr, size, alignment, entry.is_written); if (device.UseAssemblyShaders()) { - *pointer = info.address + info.offset; + *ssbo = BindlessSSBO{ + .address = static_cast<GLuint64EXT>(info.address + info.offset), + .length = static_cast<GLsizei>(size), + .padding = 0, + }; } else { glBindBufferRange(GL_SHADER_STORAGE_BUFFER, binding, info.handle, info.offset, static_cast<GLsizeiptr>(size)); } } -void RasterizerOpenGL::SetupDrawTextures(std::size_t stage_index, Shader* shader) { - MICROPROFILE_SCOPE(OpenGL_Texture); - u32 binding = device.GetBaseBindings(stage_index).sampler; +void RasterizerOpenGL::SetupDrawTextures(const Shader* shader, size_t stage_index) { + const bool via_header_index = + maxwell3d.regs.sampler_index == Maxwell::SamplerIndex::ViaHeaderIndex; for (const auto& entry : shader->GetEntries().samplers) { const auto shader_type = static_cast<ShaderType>(stage_index); - for (std::size_t i = 0; i < entry.size; ++i) { - const auto texture = GetTextureInfo(maxwell3d, entry, shader_type, i); - SetupTexture(binding++, texture, entry); + for (size_t index = 0; index < entry.size; ++index) { + const auto handle = + GetTextureInfo(maxwell3d, via_header_index, entry, shader_type, index); + const Sampler* const sampler = texture_cache.GetGraphicsSampler(handle.sampler); + sampler_handles.push_back(sampler->Handle()); + image_view_indices.push_back(handle.image); } } } -void RasterizerOpenGL::SetupComputeTextures(Shader* kernel) { - MICROPROFILE_SCOPE(OpenGL_Texture); - u32 binding = 0; +void RasterizerOpenGL::SetupComputeTextures(const Shader* kernel) { + const bool via_header_index = kepler_compute.launch_description.linked_tsc; for (const auto& entry : kernel->GetEntries().samplers) { - for (std::size_t i = 0; i < entry.size; ++i) { - const auto texture = GetTextureInfo(kepler_compute, entry, ShaderType::Compute, i); - SetupTexture(binding++, texture, entry); + for (size_t i = 0; i < entry.size; ++i) { + const auto handle = + GetTextureInfo(kepler_compute, via_header_index, entry, ShaderType::Compute, i); + const Sampler* const sampler = texture_cache.GetComputeSampler(handle.sampler); + sampler_handles.push_back(sampler->Handle()); + image_view_indices.push_back(handle.image); } } } -void RasterizerOpenGL::SetupTexture(u32 binding, const Tegra::Texture::FullTextureInfo& texture, - const SamplerEntry& entry) { - const auto view = texture_cache.GetTextureSurface(texture.tic, entry); - if (!view) { - // Can occur when texture addr is null or its memory is unmapped/invalid - glBindSampler(binding, 0); - glBindTextureUnit(binding, 0); - return; - } - const GLuint handle = view->GetTexture(texture.tic.x_source, texture.tic.y_source, - texture.tic.z_source, texture.tic.w_source); - glBindTextureUnit(binding, handle); - if (!view->GetSurfaceParams().IsBuffer()) { - glBindSampler(binding, sampler_cache.GetSampler(texture.tsc)); - } -} - -void RasterizerOpenGL::SetupDrawImages(std::size_t stage_index, Shader* shader) { - u32 binding = device.GetBaseBindings(stage_index).image; +void RasterizerOpenGL::SetupDrawImages(const Shader* shader, size_t stage_index) { + const bool via_header_index = + maxwell3d.regs.sampler_index == Maxwell::SamplerIndex::ViaHeaderIndex; for (const auto& entry : shader->GetEntries().images) { const auto shader_type = static_cast<ShaderType>(stage_index); - const auto tic = GetTextureInfo(maxwell3d, entry, shader_type).tic; - SetupImage(binding++, tic, entry); + const auto handle = GetTextureInfo(maxwell3d, via_header_index, entry, shader_type); + image_view_indices.push_back(handle.image); } } -void RasterizerOpenGL::SetupComputeImages(Shader* shader) { - u32 binding = 0; +void RasterizerOpenGL::SetupComputeImages(const Shader* shader) { + const bool via_header_index = kepler_compute.launch_description.linked_tsc; for (const auto& entry : shader->GetEntries().images) { - const auto tic = GetTextureInfo(kepler_compute, entry, ShaderType::Compute).tic; - SetupImage(binding++, tic, entry); + const auto handle = + GetTextureInfo(kepler_compute, via_header_index, entry, ShaderType::Compute); + image_view_indices.push_back(handle.image); } } -void RasterizerOpenGL::SetupImage(u32 binding, const Tegra::Texture::TICEntry& tic, - const ImageEntry& entry) { - const auto view = texture_cache.GetImageSurface(tic, entry); - if (!view) { - glBindImageTexture(binding, 0, 0, GL_FALSE, 0, GL_READ_ONLY, GL_R8); - return; - } - if (entry.is_written) { - view->MarkAsModified(texture_cache.Tick()); - } - const GLuint handle = view->GetTexture(tic.x_source, tic.y_source, tic.z_source, tic.w_source); - glBindImageTexture(binding, handle, 0, GL_TRUE, 0, GL_READ_WRITE, view->GetFormat()); -} - void RasterizerOpenGL::SyncViewport() { auto& flags = maxwell3d.dirty.flags; const auto& regs = maxwell3d.regs; @@ -1157,7 +1160,7 @@ void RasterizerOpenGL::SyncViewport() { flags[Dirty::ClipControl] = false; bool flip_y = false; - if (regs.viewport_transform[0].scale_y < 0.0) { + if (regs.viewport_transform[0].scale_y < 0.0f) { flip_y = !flip_y; } if (regs.screen_y_control.y_negate != 0) { @@ -1527,17 +1530,9 @@ void RasterizerOpenGL::SyncPointState() { flags[Dirty::PointSize] = false; oglEnable(GL_POINT_SPRITE, maxwell3d.regs.point_sprite_enable); + oglEnable(GL_PROGRAM_POINT_SIZE, maxwell3d.regs.vp_point_size.enable); - if (maxwell3d.regs.vp_point_size.enable) { - // By definition of GL_POINT_SIZE, it only matters if GL_PROGRAM_POINT_SIZE is disabled. - glEnable(GL_PROGRAM_POINT_SIZE); - return; - } - - // Limit the point size to 1 since nouveau sometimes sets a point size of 0 (and that's invalid - // in OpenGL). glPointSize(std::max(1.0f, maxwell3d.regs.point_size)); - glDisable(GL_PROGRAM_POINT_SIZE); } void RasterizerOpenGL::SyncLineState() { @@ -1580,10 +1575,6 @@ void RasterizerOpenGL::SyncAlphaTest() { flags[Dirty::AlphaTest] = false; const auto& regs = maxwell3d.regs; - if (regs.alpha_test_enabled && regs.rt_control.count > 1) { - LOG_WARNING(Render_OpenGL, "Alpha testing with more than one render target is not tested"); - } - if (regs.alpha_test_enabled) { glEnable(GL_ALPHA_TEST); glAlphaFunc(MaxwellToGL::ComparisonOp(regs.alpha_test_func), regs.alpha_test_ref); diff --git a/src/video_core/renderer_opengl/gl_rasterizer.h b/src/video_core/renderer_opengl/gl_rasterizer.h index f451404b2..82e03e677 100644 --- a/src/video_core/renderer_opengl/gl_rasterizer.h +++ b/src/video_core/renderer_opengl/gl_rasterizer.h @@ -7,12 +7,13 @@ #include <array> #include <atomic> #include <cstddef> -#include <map> #include <memory> #include <optional> #include <tuple> #include <utility> +#include <boost/container/static_vector.hpp> + #include <glad/glad.h> #include "common/common_types.h" @@ -23,16 +24,14 @@ #include "video_core/renderer_opengl/gl_buffer_cache.h" #include "video_core/renderer_opengl/gl_device.h" #include "video_core/renderer_opengl/gl_fence_manager.h" -#include "video_core/renderer_opengl/gl_framebuffer_cache.h" #include "video_core/renderer_opengl/gl_query_cache.h" #include "video_core/renderer_opengl/gl_resource_manager.h" -#include "video_core/renderer_opengl/gl_sampler_cache.h" #include "video_core/renderer_opengl/gl_shader_cache.h" #include "video_core/renderer_opengl/gl_shader_decompiler.h" #include "video_core/renderer_opengl/gl_shader_manager.h" #include "video_core/renderer_opengl/gl_state_tracker.h" +#include "video_core/renderer_opengl/gl_stream_buffer.h" #include "video_core/renderer_opengl/gl_texture_cache.h" -#include "video_core/renderer_opengl/utils.h" #include "video_core/shader/async_shaders.h" #include "video_core/textures/texture.h" @@ -51,14 +50,21 @@ class MemoryManager; namespace OpenGL { struct ScreenInfo; -struct DrawParameters; +struct ShaderEntries; + +struct BindlessSSBO { + GLuint64EXT address; + GLsizei length; + GLsizei padding; +}; +static_assert(sizeof(BindlessSSBO) * CHAR_BIT == 128); class RasterizerOpenGL : public VideoCore::RasterizerAccelerated { public: - explicit RasterizerOpenGL(Core::Frontend::EmuWindow& emu_window, Tegra::GPU& gpu, - Core::Memory::Memory& cpu_memory, const Device& device, - ScreenInfo& screen_info, ProgramManager& program_manager, - StateTracker& state_tracker); + explicit RasterizerOpenGL(Core::Frontend::EmuWindow& emu_window_, Tegra::GPU& gpu_, + Core::Memory::Memory& cpu_memory_, const Device& device_, + ScreenInfo& screen_info_, ProgramManager& program_manager_, + StateTracker& state_tracker_); ~RasterizerOpenGL() override; void Draw(bool is_indexed, bool is_instanced) override; @@ -72,15 +78,18 @@ public: void InvalidateRegion(VAddr addr, u64 size) override; void OnCPUWrite(VAddr addr, u64 size) override; void SyncGuestHost() override; + void UnmapMemory(VAddr addr, u64 size) override; void SignalSemaphore(GPUVAddr addr, u32 value) override; void SignalSyncPoint(u32 value) override; void ReleaseFences() override; void FlushAndInvalidateRegion(VAddr addr, u64 size) override; void WaitForIdle() override; + void FragmentBarrier() override; + void TiledCacheBarrier() override; void FlushCommands() override; void TickFrame() override; - bool AccelerateSurfaceCopy(const Tegra::Engines::Fermi2D::Regs::Surface& src, - const Tegra::Engines::Fermi2D::Regs::Surface& dst, + bool AccelerateSurfaceCopy(const Tegra::Engines::Fermi2D::Surface& src, + const Tegra::Engines::Fermi2D::Surface& dst, const Tegra::Engines::Fermi2D::Config& copy_config) override; bool AccelerateDisplay(const Tegra::FramebufferConfig& config, VAddr framebuffer_addr, u32 pixel_stride) override; @@ -101,11 +110,14 @@ public: } private: - /// Configures the color and depth framebuffer states. - void ConfigureFramebuffers(); + static constexpr size_t MAX_TEXTURES = 192; + static constexpr size_t MAX_IMAGES = 48; + static constexpr size_t MAX_IMAGE_VIEWS = MAX_TEXTURES + MAX_IMAGES; + + void BindComputeTextures(Shader* kernel); - /// Configures the color and depth framebuffer for clearing. - void ConfigureClearFramebuffer(bool using_color, bool using_depth_stencil); + void BindTextures(const ShaderEntries& entries, GLuint base_texture, GLuint base_image, + size_t& image_view_index, size_t& texture_index, size_t& image_index); /// Configures the current constbuffers to use for the draw command. void SetupDrawConstBuffers(std::size_t stage_index, Shader* shader); @@ -126,26 +138,19 @@ private: /// Configures a global memory buffer. void SetupGlobalMemory(u32 binding, const GlobalMemoryEntry& entry, GPUVAddr gpu_addr, - std::size_t size, GLuint64EXT* pointer); + size_t size, BindlessSSBO* ssbo); /// Configures the current textures to use for the draw command. - void SetupDrawTextures(std::size_t stage_index, Shader* shader); + void SetupDrawTextures(const Shader* shader, size_t stage_index); /// Configures the textures used in a compute shader. - void SetupComputeTextures(Shader* kernel); - - /// Configures a texture. - void SetupTexture(u32 binding, const Tegra::Texture::FullTextureInfo& texture, - const SamplerEntry& entry); + void SetupComputeTextures(const Shader* kernel); /// Configures images in a graphics shader. - void SetupDrawImages(std::size_t stage_index, Shader* shader); + void SetupDrawImages(const Shader* shader, size_t stage_index); /// Configures images in a compute shader. - void SetupComputeImages(Shader* shader); - - /// Configures an image. - void SetupImage(u32 binding, const Tegra::Texture::TICEntry& tic, const ImageEntry& entry); + void SetupComputeImages(const Shader* shader); /// Syncs the viewport and depth range to match the guest state void SyncViewport(); @@ -220,9 +225,6 @@ private: /// End a transform feedback void EndTransformFeedback(); - /// Check for extension that are not strictly required but are needed for correct emulation - void CheckExtensions(); - std::size_t CalculateVertexArraysSize() const; std::size_t CalculateIndexBufferSize() const; @@ -235,7 +237,7 @@ private: GLintptr SetupIndexBuffer(); - void SetupShaders(GLenum primitive_mode); + void SetupShaders(); Tegra::GPU& gpu; Tegra::Engines::Maxwell3D& maxwell3d; @@ -247,19 +249,21 @@ private: ProgramManager& program_manager; StateTracker& state_tracker; - TextureCacheOpenGL texture_cache; + OGLStreamBuffer stream_buffer; + TextureCacheRuntime texture_cache_runtime; + TextureCache texture_cache; ShaderCacheOpenGL shader_cache; - SamplerCacheOpenGL sampler_cache; - FramebufferCacheOpenGL framebuffer_cache; QueryCache query_cache; OGLBufferCache buffer_cache; FenceManagerOpenGL fence_manager; VideoCommon::Shader::AsyncShaders async_shaders; - static constexpr std::size_t STREAM_BUFFER_SIZE = 128 * 1024 * 1024; - - GLint vertex_binding = 0; + boost::container::static_vector<u32, MAX_IMAGE_VIEWS> image_view_indices; + std::array<ImageViewId, MAX_IMAGE_VIEWS> image_view_ids; + boost::container::static_vector<GLuint, MAX_TEXTURES> sampler_handles; + std::array<GLuint, MAX_TEXTURES> texture_handles; + std::array<GLuint, MAX_IMAGES> image_handles; std::array<OGLBuffer, Tegra::Engines::Maxwell3D::Regs::NumTransformFeedbackBuffers> transform_feedback_buffers; @@ -273,7 +277,7 @@ private: std::size_t current_cbuf = 0; OGLBuffer unified_uniform_buffer; - /// Number of commands queued to the OpenGL driver. Reseted on flush. + /// Number of commands queued to the OpenGL driver. Resetted on flush. std::size_t num_queued_commands = 0; u32 last_clip_distance_mask = 0; diff --git a/src/video_core/renderer_opengl/gl_resource_manager.cpp b/src/video_core/renderer_opengl/gl_resource_manager.cpp index 0ebcec427..0e34a0f20 100644 --- a/src/video_core/renderer_opengl/gl_resource_manager.cpp +++ b/src/video_core/renderer_opengl/gl_resource_manager.cpp @@ -71,7 +71,7 @@ void OGLSampler::Create() { return; MICROPROFILE_SCOPE(OpenGL_ResourceCreation); - glGenSamplers(1, &handle); + glCreateSamplers(1, &handle); } void OGLSampler::Release() { diff --git a/src/video_core/renderer_opengl/gl_sampler_cache.cpp b/src/video_core/renderer_opengl/gl_sampler_cache.cpp deleted file mode 100644 index 5c174879a..000000000 --- a/src/video_core/renderer_opengl/gl_sampler_cache.cpp +++ /dev/null @@ -1,52 +0,0 @@ -// Copyright 2019 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. - -#include "common/logging/log.h" -#include "video_core/renderer_opengl/gl_resource_manager.h" -#include "video_core/renderer_opengl/gl_sampler_cache.h" -#include "video_core/renderer_opengl/maxwell_to_gl.h" - -namespace OpenGL { - -SamplerCacheOpenGL::SamplerCacheOpenGL() = default; - -SamplerCacheOpenGL::~SamplerCacheOpenGL() = default; - -OGLSampler SamplerCacheOpenGL::CreateSampler(const Tegra::Texture::TSCEntry& tsc) const { - OGLSampler sampler; - sampler.Create(); - - const GLuint sampler_id{sampler.handle}; - glSamplerParameteri( - sampler_id, GL_TEXTURE_MAG_FILTER, - MaxwellToGL::TextureFilterMode(tsc.mag_filter, Tegra::Texture::TextureMipmapFilter::None)); - glSamplerParameteri(sampler_id, GL_TEXTURE_MIN_FILTER, - MaxwellToGL::TextureFilterMode(tsc.min_filter, tsc.mipmap_filter)); - glSamplerParameteri(sampler_id, GL_TEXTURE_WRAP_S, MaxwellToGL::WrapMode(tsc.wrap_u)); - glSamplerParameteri(sampler_id, GL_TEXTURE_WRAP_T, MaxwellToGL::WrapMode(tsc.wrap_v)); - glSamplerParameteri(sampler_id, GL_TEXTURE_WRAP_R, MaxwellToGL::WrapMode(tsc.wrap_p)); - glSamplerParameteri(sampler_id, GL_TEXTURE_COMPARE_MODE, - tsc.depth_compare_enabled == 1 ? GL_COMPARE_REF_TO_TEXTURE : GL_NONE); - glSamplerParameteri(sampler_id, GL_TEXTURE_COMPARE_FUNC, - MaxwellToGL::DepthCompareFunc(tsc.depth_compare_func)); - glSamplerParameterfv(sampler_id, GL_TEXTURE_BORDER_COLOR, tsc.GetBorderColor().data()); - glSamplerParameterf(sampler_id, GL_TEXTURE_MIN_LOD, tsc.GetMinLod()); - glSamplerParameterf(sampler_id, GL_TEXTURE_MAX_LOD, tsc.GetMaxLod()); - glSamplerParameterf(sampler_id, GL_TEXTURE_LOD_BIAS, tsc.GetLodBias()); - if (GLAD_GL_ARB_texture_filter_anisotropic) { - glSamplerParameterf(sampler_id, GL_TEXTURE_MAX_ANISOTROPY, tsc.GetMaxAnisotropy()); - } else if (GLAD_GL_EXT_texture_filter_anisotropic) { - glSamplerParameterf(sampler_id, GL_TEXTURE_MAX_ANISOTROPY_EXT, tsc.GetMaxAnisotropy()); - } else { - LOG_WARNING(Render_OpenGL, "Anisotropy not supported by host GPU driver"); - } - - return sampler; -} - -GLuint SamplerCacheOpenGL::ToSamplerType(const OGLSampler& sampler) const { - return sampler.handle; -} - -} // namespace OpenGL diff --git a/src/video_core/renderer_opengl/gl_sampler_cache.h b/src/video_core/renderer_opengl/gl_sampler_cache.h deleted file mode 100644 index 34ee37f00..000000000 --- a/src/video_core/renderer_opengl/gl_sampler_cache.h +++ /dev/null @@ -1,25 +0,0 @@ -// Copyright 2019 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. - -#pragma once - -#include <glad/glad.h> - -#include "video_core/renderer_opengl/gl_resource_manager.h" -#include "video_core/sampler_cache.h" - -namespace OpenGL { - -class SamplerCacheOpenGL final : public VideoCommon::SamplerCache<GLuint, OGLSampler> { -public: - explicit SamplerCacheOpenGL(); - ~SamplerCacheOpenGL(); - -protected: - OGLSampler CreateSampler(const Tegra::Texture::TSCEntry& tsc) const override; - - GLuint ToSamplerType(const OGLSampler& sampler) const override; -}; - -} // namespace OpenGL diff --git a/src/video_core/renderer_opengl/gl_shader_cache.cpp b/src/video_core/renderer_opengl/gl_shader_cache.cpp index bd56bed0c..d4841fdb7 100644 --- a/src/video_core/renderer_opengl/gl_shader_cache.cpp +++ b/src/video_core/renderer_opengl/gl_shader_cache.cpp @@ -27,7 +27,6 @@ #include "video_core/renderer_opengl/gl_shader_decompiler.h" #include "video_core/renderer_opengl/gl_shader_disk_cache.h" #include "video_core/renderer_opengl/gl_state_tracker.h" -#include "video_core/renderer_opengl/utils.h" #include "video_core/shader/memory_util.h" #include "video_core/shader/registry.h" #include "video_core/shader/shader_ir.h" @@ -198,10 +197,10 @@ ProgramSharedPtr BuildShader(const Device& device, ShaderType shader_type, u64 u return program; } -Shader::Shader(std::shared_ptr<VideoCommon::Shader::Registry> registry_, ShaderEntries entries_, - ProgramSharedPtr program_, bool is_built) +Shader::Shader(std::shared_ptr<Registry> registry_, ShaderEntries entries_, + ProgramSharedPtr program_, bool is_built_) : registry{std::move(registry_)}, entries{std::move(entries_)}, program{std::move(program_)}, - is_built(is_built) { + is_built{is_built_} { handle = program->assembly_program.handle; if (handle == 0) { handle = program->source_program.handle; @@ -318,14 +317,13 @@ std::unique_ptr<Shader> Shader::CreateFromCache(const ShaderParameters& params, precompiled_shader.registry, precompiled_shader.entries, precompiled_shader.program)); } -ShaderCacheOpenGL::ShaderCacheOpenGL(RasterizerOpenGL& rasterizer, +ShaderCacheOpenGL::ShaderCacheOpenGL(RasterizerOpenGL& rasterizer_, Core::Frontend::EmuWindow& emu_window_, Tegra::GPU& gpu_, Tegra::Engines::Maxwell3D& maxwell3d_, Tegra::Engines::KeplerCompute& kepler_compute_, Tegra::MemoryManager& gpu_memory_, const Device& device_) - : VideoCommon::ShaderCache<Shader>{rasterizer}, emu_window{emu_window_}, gpu{gpu_}, - gpu_memory{gpu_memory_}, maxwell3d{maxwell3d_}, - kepler_compute{kepler_compute_}, device{device_} {} + : ShaderCache{rasterizer_}, emu_window{emu_window_}, gpu{gpu_}, gpu_memory{gpu_memory_}, + maxwell3d{maxwell3d_}, kepler_compute{kepler_compute_}, device{device_} {} ShaderCacheOpenGL::~ShaderCacheOpenGL() = default; @@ -460,7 +458,7 @@ void ShaderCacheOpenGL::LoadDiskCache(u64 title_id, const std::atomic_bool& stop ProgramSharedPtr ShaderCacheOpenGL::GeneratePrecompiledProgram( const ShaderDiskCacheEntry& entry, const ShaderDiskCachePrecompiled& precompiled_entry, const std::unordered_set<GLenum>& supported_formats) { - if (supported_formats.find(precompiled_entry.binary_format) == supported_formats.end()) { + if (!supported_formats.contains(precompiled_entry.binary_format)) { LOG_INFO(Render_OpenGL, "Precompiled cache entry with unsupported format, removing"); return {}; } diff --git a/src/video_core/renderer_opengl/gl_shader_cache.h b/src/video_core/renderer_opengl/gl_shader_cache.h index 1708af06a..2aed0697e 100644 --- a/src/video_core/renderer_opengl/gl_shader_cache.h +++ b/src/video_core/renderer_opengl/gl_shader_cache.h @@ -108,7 +108,7 @@ public: private: explicit Shader(std::shared_ptr<VideoCommon::Shader::Registry> registry, ShaderEntries entries, - ProgramSharedPtr program, bool is_built = true); + ProgramSharedPtr program, bool is_built_ = true); std::shared_ptr<VideoCommon::Shader::Registry> registry; ShaderEntries entries; @@ -119,10 +119,11 @@ private: class ShaderCacheOpenGL final : public VideoCommon::ShaderCache<Shader> { public: - explicit ShaderCacheOpenGL(RasterizerOpenGL& rasterizer, Core::Frontend::EmuWindow& emu_window, - Tegra::GPU& gpu, Tegra::Engines::Maxwell3D& maxwell3d, - Tegra::Engines::KeplerCompute& kepler_compute, - Tegra::MemoryManager& gpu_memory, const Device& device); + explicit ShaderCacheOpenGL(RasterizerOpenGL& rasterizer_, + Core::Frontend::EmuWindow& emu_window_, Tegra::GPU& gpu, + Tegra::Engines::Maxwell3D& maxwell3d_, + Tegra::Engines::KeplerCompute& kepler_compute_, + Tegra::MemoryManager& gpu_memory_, const Device& device_); ~ShaderCacheOpenGL() override; /// Loads disk cache for the current game diff --git a/src/video_core/renderer_opengl/gl_shader_decompiler.cpp b/src/video_core/renderer_opengl/gl_shader_decompiler.cpp index bbb8fb095..2e1fa252d 100644 --- a/src/video_core/renderer_opengl/gl_shader_decompiler.cpp +++ b/src/video_core/renderer_opengl/gl_shader_decompiler.cpp @@ -38,11 +38,9 @@ using Tegra::Shader::IpaSampleMode; using Tegra::Shader::PixelImap; using Tegra::Shader::Register; using Tegra::Shader::TextureType; -using VideoCommon::Shader::BuildTransformFeedback; -using VideoCommon::Shader::Registry; -using namespace std::string_literals; using namespace VideoCommon::Shader; +using namespace std::string_literals; using Maxwell = Tegra::Engines::Maxwell3D::Regs; using Operation = const OperationNode&; @@ -131,7 +129,7 @@ private: class Expression final { public: - Expression(std::string code, Type type) : code{std::move(code)}, type{type} { + Expression(std::string code_, Type type_) : code{std::move(code_)}, type{type_} { ASSERT(type != Type::Void); } Expression() : type{Type::Void} {} @@ -148,8 +146,8 @@ public: ASSERT(type == Type::Void); } - std::string As(Type type) const { - switch (type) { + std::string As(Type type_) const { + switch (type_) { case Type::Bool: return AsBool(); case Type::Bool2: @@ -316,7 +314,7 @@ std::pair<const char*, u32> GetPrimitiveDescription(Maxwell::PrimitiveTopology t case Maxwell::PrimitiveTopology::TriangleStripAdjacency: return {"triangles_adjacency", 6}; default: - UNIMPLEMENTED_MSG("topology={}", static_cast<int>(topology)); + UNIMPLEMENTED_MSG("topology={}", topology); return {"points", 1}; } } @@ -342,7 +340,7 @@ std::string GetTopologyName(Tegra::Shader::OutputTopology topology) { case Tegra::Shader::OutputTopology::TriangleStrip: return "triangle_strip"; default: - UNIMPLEMENTED_MSG("Unknown output topology: {}", static_cast<u32>(topology)); + UNIMPLEMENTED_MSG("Unknown output topology: {}", topology); return "points"; } } @@ -418,11 +416,12 @@ struct GenericVaryingDescription { class GLSLDecompiler final { public: - explicit GLSLDecompiler(const Device& device, const ShaderIR& ir, const Registry& registry, - ShaderType stage, std::string_view identifier, std::string_view suffix) - : device{device}, ir{ir}, registry{registry}, stage{stage}, identifier{identifier}, - suffix{suffix}, header{ir.GetHeader()}, use_unified_uniforms{ - UseUnifiedUniforms(device, ir, stage)} { + explicit GLSLDecompiler(const Device& device_, const ShaderIR& ir_, const Registry& registry_, + ShaderType stage_, std::string_view identifier_, + std::string_view suffix_) + : device{device_}, ir{ir_}, registry{registry_}, stage{stage_}, identifier{identifier_}, + suffix{suffix_}, header{ir.GetHeader()}, use_unified_uniforms{ + UseUnifiedUniforms(device_, ir_, stage_)} { if (stage != ShaderType::Compute) { transform_feedback = BuildTransformFeedback(registry.GetGraphicsInfo()); } @@ -744,7 +743,7 @@ private: case PixelImap::Unused: break; } - UNIMPLEMENTED_MSG("Unknown attribute usage index={}", static_cast<int>(attribute)); + UNIMPLEMENTED_MSG("Unknown attribute usage index={}", attribute); return {}; } @@ -777,16 +776,16 @@ private: name = "gs_" + name + "[]"; } - std::string suffix; + std::string suffix_; if (stage == ShaderType::Fragment) { const auto input_mode{header.ps.GetPixelImap(location)}; if (input_mode == PixelImap::Unused) { return; } - suffix = GetInputFlags(input_mode); + suffix_ = GetInputFlags(input_mode); } - code.AddLine("layout (location = {}) {} in vec4 {};", location, suffix, name); + code.AddLine("layout (location = {}) {} in vec4 {};", location, suffix_, name); } void DeclareOutputAttributes() { @@ -877,7 +876,7 @@ private: } u32 binding = device.GetBaseBindings(stage).uniform_buffer; - for (const auto [index, info] : ir.GetConstantBuffers()) { + for (const auto& [index, info] : ir.GetConstantBuffers()) { const u32 num_elements = Common::AlignUp(info.GetSize(), 4) / 4; const u32 size = info.IsIndirect() ? MAX_CONSTBUFFER_ELEMENTS : num_elements; code.AddLine("layout (std140, binding = {}) uniform {} {{", binding++, @@ -1251,7 +1250,7 @@ private: } break; } - UNIMPLEMENTED_MSG("Unhandled input attribute: {}", static_cast<u32>(attribute)); + UNIMPLEMENTED_MSG("Unhandled input attribute: {}", attribute); return {"0", Type::Int}; } @@ -1331,7 +1330,7 @@ private: GetSwizzle(element)), Type::Float}}; } - UNIMPLEMENTED_MSG("Unhandled output attribute: {}", static_cast<u32>(attribute)); + UNIMPLEMENTED_MSG("Unhandled output attribute: {}", attribute); return std::nullopt; } } @@ -2056,15 +2055,19 @@ private: } Expression Texture(Operation operation) { - const auto meta = std::get_if<MetaTexture>(&operation.GetMeta()); - ASSERT(meta); - - std::string expr = GenerateTexture( - operation, "", {TextureOffset{}, TextureArgument{Type::Float, meta->bias}}); - if (meta->sampler.is_shadow) { - expr = "vec4(" + expr + ')'; + const auto meta = std::get<MetaTexture>(operation.GetMeta()); + const bool separate_dc = meta.sampler.type == TextureType::TextureCube && + meta.sampler.is_array && meta.sampler.is_shadow; + // TODO: Replace this with an array and make GenerateTexture use C++20 std::span + const std::vector<TextureIR> extras{ + TextureOffset{}, + TextureArgument{Type::Float, meta.bias}, + }; + std::string expr = GenerateTexture(operation, "", extras, separate_dc); + if (meta.sampler.is_shadow) { + expr = fmt::format("vec4({})", expr); } - return {expr + GetSwizzle(meta->element), Type::Float}; + return {expr + GetSwizzle(meta.element), Type::Float}; } Expression TextureLod(Operation operation) { @@ -2096,13 +2099,13 @@ private: const auto type = meta.sampler.is_shadow ? Type::Float : Type::Int; const bool separate_dc = meta.sampler.is_shadow; - std::vector<TextureIR> ir; + std::vector<TextureIR> ir_; if (meta.sampler.is_shadow) { - ir = {TextureOffset{}}; + ir_ = {TextureOffset{}}; } else { - ir = {TextureOffset{}, TextureArgument{type, meta.component}}; + ir_ = {TextureOffset{}, TextureArgument{type, meta.component}}; } - return {GenerateTexture(operation, "Gather", ir, separate_dc) + GetSwizzle(meta.element), + return {GenerateTexture(operation, "Gather", ir_, separate_dc) + GetSwizzle(meta.element), Type::Float}; } @@ -2748,11 +2751,11 @@ private: } } - std::string GetSampler(const Sampler& sampler) const { + std::string GetSampler(const SamplerEntry& sampler) const { return AppendSuffix(sampler.index, "sampler"); } - std::string GetImage(const Image& image) const { + std::string GetImage(const ImageEntry& image) const { return AppendSuffix(image.index, "image"); } @@ -2797,7 +2800,7 @@ std::string GetFlowVariable(u32 index) { class ExprDecompiler { public: - explicit ExprDecompiler(GLSLDecompiler& decomp) : decomp{decomp} {} + explicit ExprDecompiler(GLSLDecompiler& decomp_) : decomp{decomp_} {} void operator()(const ExprAnd& expr) { inner += '('; @@ -2852,7 +2855,7 @@ private: class ASTDecompiler { public: - explicit ASTDecompiler(GLSLDecompiler& decomp) : decomp{decomp} {} + explicit ASTDecompiler(GLSLDecompiler& decomp_) : decomp{decomp_} {} void operator()(const ASTProgram& ast) { ASTNode current = ast.nodes.GetFirst(); diff --git a/src/video_core/renderer_opengl/gl_shader_decompiler.h b/src/video_core/renderer_opengl/gl_shader_decompiler.h index 451c9689a..be68994bb 100644 --- a/src/video_core/renderer_opengl/gl_shader_decompiler.h +++ b/src/video_core/renderer_opengl/gl_shader_decompiler.h @@ -20,13 +20,13 @@ namespace OpenGL { class Device; using Maxwell = Tegra::Engines::Maxwell3D::Regs; -using SamplerEntry = VideoCommon::Shader::Sampler; -using ImageEntry = VideoCommon::Shader::Image; +using SamplerEntry = VideoCommon::Shader::SamplerEntry; +using ImageEntry = VideoCommon::Shader::ImageEntry; class ConstBufferEntry : public VideoCommon::Shader::ConstBuffer { public: - explicit ConstBufferEntry(u32 max_offset, bool is_indirect, u32 index) - : VideoCommon::Shader::ConstBuffer{max_offset, is_indirect}, index{index} {} + explicit ConstBufferEntry(u32 max_offset_, bool is_indirect_, u32 index_) + : ConstBuffer{max_offset_, is_indirect_}, index{index_} {} u32 GetIndex() const { return index; @@ -37,10 +37,10 @@ private: }; struct GlobalMemoryEntry { - constexpr explicit GlobalMemoryEntry(u32 cbuf_index, u32 cbuf_offset, bool is_read, - bool is_written) - : cbuf_index{cbuf_index}, cbuf_offset{cbuf_offset}, is_read{is_read}, is_written{ - is_written} {} + constexpr explicit GlobalMemoryEntry(u32 cbuf_index_, u32 cbuf_offset_, bool is_read_, + bool is_written_) + : cbuf_index{cbuf_index_}, cbuf_offset{cbuf_offset_}, is_read{is_read_}, is_written{ + is_written_} {} u32 cbuf_index = 0; u32 cbuf_offset = 0; diff --git a/src/video_core/renderer_opengl/gl_shader_disk_cache.cpp b/src/video_core/renderer_opengl/gl_shader_disk_cache.cpp index 166ee34e1..955b2abc4 100644 --- a/src/video_core/renderer_opengl/gl_shader_disk_cache.cpp +++ b/src/video_core/renderer_opengl/gl_shader_disk_cache.cpp @@ -317,8 +317,7 @@ std::optional<std::vector<ShaderDiskCachePrecompiled>> ShaderDiskCacheOpenGL::Lo return std::nullopt; } } - - return std::move(entries); + return entries; } void ShaderDiskCacheOpenGL::InvalidateTransferable() { @@ -344,7 +343,7 @@ void ShaderDiskCacheOpenGL::SaveEntry(const ShaderDiskCacheEntry& entry) { } const u64 id = entry.unique_identifier; - if (stored_transferable.find(id) != stored_transferable.end()) { + if (stored_transferable.contains(id)) { // The shader already exists return; } diff --git a/src/video_core/renderer_opengl/gl_shader_manager.cpp b/src/video_core/renderer_opengl/gl_shader_manager.cpp index 691c6c79b..553e6e8d6 100644 --- a/src/video_core/renderer_opengl/gl_shader_manager.cpp +++ b/src/video_core/renderer_opengl/gl_shader_manager.cpp @@ -83,6 +83,21 @@ void ProgramManager::RestoreGuestPipeline() { } } +void ProgramManager::BindHostCompute(GLuint program) { + if (use_assembly_programs) { + glDisable(GL_COMPUTE_PROGRAM_NV); + } + glUseProgram(program); + is_graphics_bound = false; +} + +void ProgramManager::RestoreGuestCompute() { + if (use_assembly_programs) { + glEnable(GL_COMPUTE_PROGRAM_NV); + glUseProgram(0); + } +} + void ProgramManager::UseVertexShader(GLuint program) { if (use_assembly_programs) { BindProgram(GL_VERTEX_PROGRAM_NV, program, current_state.vertex, vertex_enabled); diff --git a/src/video_core/renderer_opengl/gl_shader_manager.h b/src/video_core/renderer_opengl/gl_shader_manager.h index 950e0dfcb..ad42cce74 100644 --- a/src/video_core/renderer_opengl/gl_shader_manager.h +++ b/src/video_core/renderer_opengl/gl_shader_manager.h @@ -45,6 +45,12 @@ public: /// Rewinds BindHostPipeline state changes. void RestoreGuestPipeline(); + /// Binds an OpenGL GLSL program object unsynchronized with the guest state. + void BindHostCompute(GLuint program); + + /// Rewinds BindHostCompute state changes. + void RestoreGuestCompute(); + void UseVertexShader(GLuint program); void UseGeometryShader(GLuint program); void UseFragmentShader(GLuint program); diff --git a/src/video_core/renderer_opengl/gl_state_tracker.cpp b/src/video_core/renderer_opengl/gl_state_tracker.cpp index 6bcf831f2..60e6fa39f 100644 --- a/src/video_core/renderer_opengl/gl_state_tracker.cpp +++ b/src/video_core/renderer_opengl/gl_state_tracker.cpp @@ -13,7 +13,7 @@ #include "video_core/renderer_opengl/gl_state_tracker.h" #define OFF(field_name) MAXWELL3D_REG_INDEX(field_name) -#define NUM(field_name) (sizeof(Maxwell3D::Regs::field_name) / sizeof(u32)) +#define NUM(field_name) (sizeof(Maxwell3D::Regs::field_name) / (sizeof(u32))) namespace OpenGL { @@ -249,4 +249,11 @@ StateTracker::StateTracker(Tegra::GPU& gpu) : flags{gpu.Maxwell3D().dirty.flags} } } +void StateTracker::InvalidateStreamBuffer() { + flags[Dirty::VertexBuffers] = true; + for (int index = Dirty::VertexBuffer0; index <= Dirty::VertexBuffer31; ++index) { + flags[index] = true; + } +} + } // namespace OpenGL diff --git a/src/video_core/renderer_opengl/gl_state_tracker.h b/src/video_core/renderer_opengl/gl_state_tracker.h index 9d127548f..574615d3c 100644 --- a/src/video_core/renderer_opengl/gl_state_tracker.h +++ b/src/video_core/renderer_opengl/gl_state_tracker.h @@ -92,6 +92,8 @@ class StateTracker { public: explicit StateTracker(Tegra::GPU& gpu); + void InvalidateStreamBuffer(); + void BindIndexBuffer(GLuint new_index_buffer) { if (index_buffer == new_index_buffer) { return; @@ -100,6 +102,14 @@ public: glBindBuffer(GL_ELEMENT_ARRAY_BUFFER, new_index_buffer); } + void BindFramebuffer(GLuint new_framebuffer) { + if (framebuffer == new_framebuffer) { + return; + } + framebuffer = new_framebuffer; + glBindFramebuffer(GL_DRAW_FRAMEBUFFER, framebuffer); + } + void NotifyScreenDrawVertexArray() { flags[OpenGL::Dirty::VertexFormats] = true; flags[OpenGL::Dirty::VertexFormat0 + 0] = true; @@ -129,9 +139,9 @@ public: flags[OpenGL::Dirty::Scissor0] = true; } - void NotifyColorMask0() { + void NotifyColorMask(size_t index) { flags[OpenGL::Dirty::ColorMasks] = true; - flags[OpenGL::Dirty::ColorMask0] = true; + flags[OpenGL::Dirty::ColorMask0 + index] = true; } void NotifyBlend0() { @@ -190,6 +200,7 @@ public: private: Tegra::Engines::Maxwell3D::DirtyState::Flags& flags; + GLuint framebuffer = 0; GLuint index_buffer = 0; }; diff --git a/src/video_core/renderer_opengl/gl_stream_buffer.cpp b/src/video_core/renderer_opengl/gl_stream_buffer.cpp index 887995cf4..e0819cdf2 100644 --- a/src/video_core/renderer_opengl/gl_stream_buffer.cpp +++ b/src/video_core/renderer_opengl/gl_stream_buffer.cpp @@ -9,6 +9,7 @@ #include "common/assert.h" #include "common/microprofile.h" #include "video_core/renderer_opengl/gl_device.h" +#include "video_core/renderer_opengl/gl_state_tracker.h" #include "video_core/renderer_opengl/gl_stream_buffer.h" MICROPROFILE_DEFINE(OpenGL_StreamBuffer, "OpenGL", "Stream Buffer Orphaning", @@ -16,24 +17,14 @@ MICROPROFILE_DEFINE(OpenGL_StreamBuffer, "OpenGL", "Stream Buffer Orphaning", namespace OpenGL { -OGLStreamBuffer::OGLStreamBuffer(const Device& device, GLsizeiptr size, bool vertex_data_usage) - : buffer_size(size) { +OGLStreamBuffer::OGLStreamBuffer(const Device& device, StateTracker& state_tracker_) + : state_tracker{state_tracker_} { gl_buffer.Create(); - GLsizeiptr allocate_size = size; - if (vertex_data_usage) { - // On AMD GPU there is a strange crash in indexed drawing. The crash happens when the buffer - // read position is near the end and is an out-of-bound access to the vertex buffer. This is - // probably a bug in the driver and is related to the usage of vec3<byte> attributes in the - // vertex array. Doubling the allocation size for the vertex buffer seems to avoid the - // crash. - allocate_size *= 2; - } - static constexpr GLbitfield flags = GL_MAP_WRITE_BIT | GL_MAP_PERSISTENT_BIT; - glNamedBufferStorage(gl_buffer.handle, allocate_size, nullptr, flags); + glNamedBufferStorage(gl_buffer.handle, BUFFER_SIZE, nullptr, flags); mapped_ptr = static_cast<u8*>( - glMapNamedBufferRange(gl_buffer.handle, 0, buffer_size, flags | GL_MAP_FLUSH_EXPLICIT_BIT)); + glMapNamedBufferRange(gl_buffer.handle, 0, BUFFER_SIZE, flags | GL_MAP_FLUSH_EXPLICIT_BIT)); if (device.UseAssemblyShaders() || device.HasVertexBufferUnifiedMemory()) { glMakeNamedBufferResidentNV(gl_buffer.handle, GL_READ_ONLY); @@ -46,25 +37,24 @@ OGLStreamBuffer::~OGLStreamBuffer() { gl_buffer.Release(); } -std::tuple<u8*, GLintptr, bool> OGLStreamBuffer::Map(GLsizeiptr size, GLintptr alignment) { - ASSERT(size <= buffer_size); - ASSERT(alignment <= buffer_size); +std::pair<u8*, GLintptr> OGLStreamBuffer::Map(GLsizeiptr size, GLintptr alignment) { + ASSERT(size <= BUFFER_SIZE); + ASSERT(alignment <= BUFFER_SIZE); mapped_size = size; if (alignment > 0) { buffer_pos = Common::AlignUp<std::size_t>(buffer_pos, alignment); } - bool invalidate = false; - if (buffer_pos + size > buffer_size) { + if (buffer_pos + size > BUFFER_SIZE) { MICROPROFILE_SCOPE(OpenGL_StreamBuffer); glInvalidateBufferData(gl_buffer.handle); + state_tracker.InvalidateStreamBuffer(); buffer_pos = 0; - invalidate = true; } - return std::make_tuple(mapped_ptr + buffer_pos, buffer_pos, invalidate); + return std::make_pair(mapped_ptr + buffer_pos, buffer_pos); } void OGLStreamBuffer::Unmap(GLsizeiptr size) { diff --git a/src/video_core/renderer_opengl/gl_stream_buffer.h b/src/video_core/renderer_opengl/gl_stream_buffer.h index 307a67113..dd9cf67eb 100644 --- a/src/video_core/renderer_opengl/gl_stream_buffer.h +++ b/src/video_core/renderer_opengl/gl_stream_buffer.h @@ -4,29 +4,31 @@ #pragma once -#include <tuple> +#include <utility> + #include <glad/glad.h> + #include "common/common_types.h" #include "video_core/renderer_opengl/gl_resource_manager.h" namespace OpenGL { class Device; +class StateTracker; class OGLStreamBuffer : private NonCopyable { public: - explicit OGLStreamBuffer(const Device& device, GLsizeiptr size, bool vertex_data_usage); + explicit OGLStreamBuffer(const Device& device, StateTracker& state_tracker_); ~OGLStreamBuffer(); /* * Allocates a linear chunk of memory in the GPU buffer with at least "size" bytes * and the optional alignment requirement. * If the buffer is full, the whole buffer is reallocated which invalidates old chunks. - * The return values are the pointer to the new chunk, the offset within the buffer, - * and the invalidation flag for previous chunks. + * The return values are the pointer to the new chunk, and the offset within the buffer. * The actual used size must be specified on unmapping the chunk. */ - std::tuple<u8*, GLintptr, bool> Map(GLsizeiptr size, GLintptr alignment = 0); + std::pair<u8*, GLintptr> Map(GLsizeiptr size, GLintptr alignment = 0); void Unmap(GLsizeiptr size); @@ -39,15 +41,18 @@ public: } GLsizeiptr Size() const noexcept { - return buffer_size; + return BUFFER_SIZE; } private: + static constexpr GLsizeiptr BUFFER_SIZE = 256 * 1024 * 1024; + + StateTracker& state_tracker; + OGLBuffer gl_buffer; GLuint64EXT gpu_address = 0; GLintptr buffer_pos = 0; - GLsizeiptr buffer_size = 0; GLsizeiptr mapped_size = 0; u8* mapped_ptr = nullptr; }; diff --git a/src/video_core/renderer_opengl/gl_texture_cache.cpp b/src/video_core/renderer_opengl/gl_texture_cache.cpp index a863ef218..546cb6d00 100644 --- a/src/video_core/renderer_opengl/gl_texture_cache.cpp +++ b/src/video_core/renderer_opengl/gl_texture_cache.cpp @@ -2,37 +2,60 @@ // Licensed under GPLv2 or any later version // Refer to the license.txt file included. -#include "common/assert.h" -#include "common/bit_util.h" -#include "common/common_types.h" -#include "common/microprofile.h" -#include "common/scope_exit.h" -#include "core/core.h" -#include "video_core/morton.h" -#include "video_core/renderer_opengl/gl_resource_manager.h" +#include <algorithm> +#include <array> +#include <bit> +#include <string> + +#include <glad/glad.h> + +#include "video_core/renderer_opengl/gl_device.h" +#include "video_core/renderer_opengl/gl_shader_manager.h" #include "video_core/renderer_opengl/gl_state_tracker.h" #include "video_core/renderer_opengl/gl_texture_cache.h" -#include "video_core/renderer_opengl/utils.h" -#include "video_core/texture_cache/surface_base.h" +#include "video_core/renderer_opengl/maxwell_to_gl.h" +#include "video_core/renderer_opengl/util_shaders.h" +#include "video_core/surface.h" +#include "video_core/texture_cache/format_lookup_table.h" +#include "video_core/texture_cache/samples_helper.h" #include "video_core/texture_cache/texture_cache.h" -#include "video_core/textures/convert.h" -#include "video_core/textures/texture.h" +#include "video_core/textures/decoders.h" namespace OpenGL { -using Tegra::Texture::SwizzleSource; -using VideoCore::MortonSwizzleMode; +namespace { +using Tegra::Texture::SwizzleSource; +using Tegra::Texture::TextureMipmapFilter; +using Tegra::Texture::TextureType; +using Tegra::Texture::TICEntry; +using Tegra::Texture::TSCEntry; +using VideoCommon::CalculateLevelStrideAlignment; +using VideoCommon::ImageCopy; +using VideoCommon::ImageFlagBits; +using VideoCommon::ImageType; +using VideoCommon::NUM_RT; +using VideoCommon::SamplesLog2; +using VideoCommon::SwizzleParameters; +using VideoCore::Surface::BytesPerBlock; +using VideoCore::Surface::IsPixelFormatASTC; +using VideoCore::Surface::IsPixelFormatSRGB; +using VideoCore::Surface::MaxPixelFormat; using VideoCore::Surface::PixelFormat; -using VideoCore::Surface::SurfaceTarget; using VideoCore::Surface::SurfaceType; -MICROPROFILE_DEFINE(OpenGL_Texture_Upload, "OpenGL", "Texture Upload", MP_RGB(128, 192, 128)); -MICROPROFILE_DEFINE(OpenGL_Texture_Download, "OpenGL", "Texture Download", MP_RGB(128, 192, 128)); -MICROPROFILE_DEFINE(OpenGL_Texture_Buffer_Copy, "OpenGL", "Texture Buffer Copy", - MP_RGB(128, 192, 128)); +struct CopyOrigin { + GLint level; + GLint x; + GLint y; + GLint z; +}; -namespace { +struct CopyRegion { + GLsizei width; + GLsizei height; + GLsizei depth; +}; struct FormatTuple { GLenum internal_format; @@ -40,7 +63,7 @@ struct FormatTuple { GLenum type = GL_NONE; }; -constexpr std::array<FormatTuple, VideoCore::Surface::MaxPixelFormat> tex_format_tuples = {{ +constexpr std::array<FormatTuple, MaxPixelFormat> FORMAT_TABLE = {{ {GL_RGBA8, GL_RGBA, GL_UNSIGNED_INT_8_8_8_8_REV}, // A8B8G8R8_UNORM {GL_RGBA8_SNORM, GL_RGBA, GL_BYTE}, // A8B8G8R8_SNORM {GL_RGBA8I, GL_RGBA_INTEGER, GL_BYTE}, // A8B8G8R8_SINT @@ -103,72 +126,113 @@ constexpr std::array<FormatTuple, VideoCore::Surface::MaxPixelFormat> tex_format {GL_COMPRESSED_RGBA_ASTC_8x5_KHR}, // ASTC_2D_8X5_UNORM {GL_COMPRESSED_RGBA_ASTC_5x4_KHR}, // ASTC_2D_5X4_UNORM {GL_SRGB8_ALPHA8, GL_BGRA, GL_UNSIGNED_BYTE}, // B8G8R8A8_UNORM - // Compressed sRGB formats - {GL_COMPRESSED_SRGB_ALPHA_S3TC_DXT1_EXT}, // BC1_RGBA_SRGB - {GL_COMPRESSED_SRGB_ALPHA_S3TC_DXT3_EXT}, // BC2_SRGB - {GL_COMPRESSED_SRGB_ALPHA_S3TC_DXT5_EXT}, // BC3_SRGB - {GL_COMPRESSED_SRGB_ALPHA_BPTC_UNORM}, // BC7_SRGB - {GL_RGBA4, GL_RGBA, GL_UNSIGNED_SHORT_4_4_4_4_REV}, // A4B4G4R4_UNORM - {GL_COMPRESSED_SRGB8_ALPHA8_ASTC_4x4_KHR}, // ASTC_2D_4X4_SRGB - {GL_COMPRESSED_SRGB8_ALPHA8_ASTC_8x8_KHR}, // ASTC_2D_8X8_SRGB - {GL_COMPRESSED_SRGB8_ALPHA8_ASTC_8x5_KHR}, // ASTC_2D_8X5_SRGB - {GL_COMPRESSED_SRGB8_ALPHA8_ASTC_5x4_KHR}, // ASTC_2D_5X4_SRGB - {GL_COMPRESSED_RGBA_ASTC_5x5_KHR}, // ASTC_2D_5X5_UNORM - {GL_COMPRESSED_SRGB8_ALPHA8_ASTC_5x5_KHR}, // ASTC_2D_5X5_SRGB - {GL_COMPRESSED_RGBA_ASTC_10x8_KHR}, // ASTC_2D_10X8_UNORM - {GL_COMPRESSED_SRGB8_ALPHA8_ASTC_10x8_KHR}, // ASTC_2D_10X8_SRGB - {GL_COMPRESSED_RGBA_ASTC_6x6_KHR}, // ASTC_2D_6X6_UNORM - {GL_COMPRESSED_SRGB8_ALPHA8_ASTC_6x6_KHR}, // ASTC_2D_6X6_SRGB - {GL_COMPRESSED_RGBA_ASTC_10x10_KHR}, // ASTC_2D_10X10_UNORM - {GL_COMPRESSED_SRGB8_ALPHA8_ASTC_10x10_KHR}, // ASTC_2D_10X10_SRGB - {GL_COMPRESSED_RGBA_ASTC_12x12_KHR}, // ASTC_2D_12X12_UNORM - {GL_COMPRESSED_SRGB8_ALPHA8_ASTC_12x12_KHR}, // ASTC_2D_12X12_SRGB - {GL_COMPRESSED_RGBA_ASTC_8x6_KHR}, // ASTC_2D_8X6_UNORM - {GL_COMPRESSED_SRGB8_ALPHA8_ASTC_8x6_KHR}, // ASTC_2D_8X6_SRGB - {GL_COMPRESSED_RGBA_ASTC_6x5_KHR}, // ASTC_2D_6X5_UNORM - {GL_COMPRESSED_SRGB8_ALPHA8_ASTC_6x5_KHR}, // ASTC_2D_6X5_SRGB - {GL_RGB9_E5, GL_RGB, GL_UNSIGNED_INT_5_9_9_9_REV}, // E5B9G9R9_FLOAT - - // Depth formats - {GL_DEPTH_COMPONENT32F, GL_DEPTH_COMPONENT, GL_FLOAT}, // D32_FLOAT - {GL_DEPTH_COMPONENT16, GL_DEPTH_COMPONENT, GL_UNSIGNED_SHORT}, // D16_UNORM - - // DepthStencil formats - {GL_DEPTH24_STENCIL8, GL_DEPTH_STENCIL, GL_UNSIGNED_INT_24_8}, // D24_UNORM_S8_UINT - {GL_DEPTH24_STENCIL8, GL_DEPTH_STENCIL, GL_UNSIGNED_INT_24_8}, // S8_UINT_D24_UNORM + {GL_COMPRESSED_SRGB_ALPHA_S3TC_DXT1_EXT}, // BC1_RGBA_SRGB + {GL_COMPRESSED_SRGB_ALPHA_S3TC_DXT3_EXT}, // BC2_SRGB + {GL_COMPRESSED_SRGB_ALPHA_S3TC_DXT5_EXT}, // BC3_SRGB + {GL_COMPRESSED_SRGB_ALPHA_BPTC_UNORM}, // BC7_SRGB + {GL_RGBA4, GL_RGBA, GL_UNSIGNED_SHORT_4_4_4_4_REV}, // A4B4G4R4_UNORM + {GL_COMPRESSED_SRGB8_ALPHA8_ASTC_4x4_KHR}, // ASTC_2D_4X4_SRGB + {GL_COMPRESSED_SRGB8_ALPHA8_ASTC_8x8_KHR}, // ASTC_2D_8X8_SRGB + {GL_COMPRESSED_SRGB8_ALPHA8_ASTC_8x5_KHR}, // ASTC_2D_8X5_SRGB + {GL_COMPRESSED_SRGB8_ALPHA8_ASTC_5x4_KHR}, // ASTC_2D_5X4_SRGB + {GL_COMPRESSED_RGBA_ASTC_5x5_KHR}, // ASTC_2D_5X5_UNORM + {GL_COMPRESSED_SRGB8_ALPHA8_ASTC_5x5_KHR}, // ASTC_2D_5X5_SRGB + {GL_COMPRESSED_RGBA_ASTC_10x8_KHR}, // ASTC_2D_10X8_UNORM + {GL_COMPRESSED_SRGB8_ALPHA8_ASTC_10x8_KHR}, // ASTC_2D_10X8_SRGB + {GL_COMPRESSED_RGBA_ASTC_6x6_KHR}, // ASTC_2D_6X6_UNORM + {GL_COMPRESSED_SRGB8_ALPHA8_ASTC_6x6_KHR}, // ASTC_2D_6X6_SRGB + {GL_COMPRESSED_RGBA_ASTC_10x10_KHR}, // ASTC_2D_10X10_UNORM + {GL_COMPRESSED_SRGB8_ALPHA8_ASTC_10x10_KHR}, // ASTC_2D_10X10_SRGB + {GL_COMPRESSED_RGBA_ASTC_12x12_KHR}, // ASTC_2D_12X12_UNORM + {GL_COMPRESSED_SRGB8_ALPHA8_ASTC_12x12_KHR}, // ASTC_2D_12X12_SRGB + {GL_COMPRESSED_RGBA_ASTC_8x6_KHR}, // ASTC_2D_8X6_UNORM + {GL_COMPRESSED_SRGB8_ALPHA8_ASTC_8x6_KHR}, // ASTC_2D_8X6_SRGB + {GL_COMPRESSED_RGBA_ASTC_6x5_KHR}, // ASTC_2D_6X5_UNORM + {GL_COMPRESSED_SRGB8_ALPHA8_ASTC_6x5_KHR}, // ASTC_2D_6X5_SRGB + {GL_RGB9_E5, GL_RGB, GL_UNSIGNED_INT_5_9_9_9_REV}, // E5B9G9R9_FLOAT + {GL_DEPTH_COMPONENT32F, GL_DEPTH_COMPONENT, GL_FLOAT}, // D32_FLOAT + {GL_DEPTH_COMPONENT16, GL_DEPTH_COMPONENT, GL_UNSIGNED_SHORT}, // D16_UNORM + {GL_DEPTH24_STENCIL8, GL_DEPTH_STENCIL, GL_UNSIGNED_INT_24_8}, // D24_UNORM_S8_UINT + {GL_DEPTH24_STENCIL8, GL_DEPTH_STENCIL, GL_UNSIGNED_INT_24_8}, // S8_UINT_D24_UNORM {GL_DEPTH32F_STENCIL8, GL_DEPTH_STENCIL, GL_FLOAT_32_UNSIGNED_INT_24_8_REV}, // D32_FLOAT_S8_UINT }}; +constexpr std::array ACCELERATED_FORMATS{ + GL_RGBA32F, GL_RGBA16F, GL_RG32F, GL_RG16F, GL_R11F_G11F_B10F, GL_R32F, + GL_R16F, GL_RGBA32UI, GL_RGBA16UI, GL_RGB10_A2UI, GL_RGBA8UI, GL_RG32UI, + GL_RG16UI, GL_RG8UI, GL_R32UI, GL_R16UI, GL_R8UI, GL_RGBA32I, + GL_RGBA16I, GL_RGBA8I, GL_RG32I, GL_RG16I, GL_RG8I, GL_R32I, + GL_R16I, GL_R8I, GL_RGBA16, GL_RGB10_A2, GL_RGBA8, GL_RG16, + GL_RG8, GL_R16, GL_R8, GL_RGBA16_SNORM, GL_RGBA8_SNORM, GL_RG16_SNORM, + GL_RG8_SNORM, GL_R16_SNORM, GL_R8_SNORM, +}; + const FormatTuple& GetFormatTuple(PixelFormat pixel_format) { - ASSERT(static_cast<std::size_t>(pixel_format) < tex_format_tuples.size()); - return tex_format_tuples[static_cast<std::size_t>(pixel_format)]; + ASSERT(static_cast<size_t>(pixel_format) < FORMAT_TABLE.size()); + return FORMAT_TABLE[static_cast<size_t>(pixel_format)]; } -GLenum GetTextureTarget(const SurfaceTarget& target) { - switch (target) { - case SurfaceTarget::TextureBuffer: +GLenum ImageTarget(const VideoCommon::ImageInfo& info) { + switch (info.type) { + case ImageType::e1D: + return GL_TEXTURE_1D_ARRAY; + case ImageType::e2D: + if (info.num_samples > 1) { + return GL_TEXTURE_2D_MULTISAMPLE_ARRAY; + } + return GL_TEXTURE_2D_ARRAY; + case ImageType::e3D: + return GL_TEXTURE_3D; + case ImageType::Linear: + return GL_TEXTURE_2D_ARRAY; + case ImageType::Buffer: return GL_TEXTURE_BUFFER; - case SurfaceTarget::Texture1D: + } + UNREACHABLE_MSG("Invalid image type={}", info.type); + return GL_NONE; +} + +GLenum ImageTarget(ImageViewType type, int num_samples = 1) { + const bool is_multisampled = num_samples > 1; + switch (type) { + case ImageViewType::e1D: return GL_TEXTURE_1D; - case SurfaceTarget::Texture2D: - return GL_TEXTURE_2D; - case SurfaceTarget::Texture3D: + case ImageViewType::e2D: + return is_multisampled ? GL_TEXTURE_2D_MULTISAMPLE : GL_TEXTURE_2D; + case ImageViewType::Cube: + return GL_TEXTURE_CUBE_MAP; + case ImageViewType::e3D: return GL_TEXTURE_3D; - case SurfaceTarget::Texture1DArray: + case ImageViewType::e1DArray: return GL_TEXTURE_1D_ARRAY; - case SurfaceTarget::Texture2DArray: - return GL_TEXTURE_2D_ARRAY; - case SurfaceTarget::TextureCubemap: - return GL_TEXTURE_CUBE_MAP; - case SurfaceTarget::TextureCubeArray: + case ImageViewType::e2DArray: + return is_multisampled ? GL_TEXTURE_2D_MULTISAMPLE_ARRAY : GL_TEXTURE_2D_ARRAY; + case ImageViewType::CubeArray: return GL_TEXTURE_CUBE_MAP_ARRAY; + case ImageViewType::Rect: + return GL_TEXTURE_RECTANGLE; + case ImageViewType::Buffer: + return GL_TEXTURE_BUFFER; } - UNREACHABLE(); - return {}; + UNREACHABLE_MSG("Invalid image view type={}", type); + return GL_NONE; } -GLint GetSwizzleSource(SwizzleSource source) { +GLenum TextureMode(PixelFormat format, bool is_first) { + switch (format) { + case PixelFormat::D24_UNORM_S8_UINT: + case PixelFormat::D32_FLOAT_S8_UINT: + return is_first ? GL_DEPTH_COMPONENT : GL_STENCIL_INDEX; + case PixelFormat::S8_UINT_D24_UNORM: + return is_first ? GL_STENCIL_INDEX : GL_DEPTH_COMPONENT; + default: + UNREACHABLE(); + return GL_DEPTH_COMPONENT; + } +} + +GLint Swizzle(SwizzleSource source) { switch (source) { case SwizzleSource::Zero: return GL_ZERO; @@ -184,531 +248,813 @@ GLint GetSwizzleSource(SwizzleSource source) { case SwizzleSource::OneFloat: return GL_ONE; } - UNREACHABLE(); + UNREACHABLE_MSG("Invalid swizzle source={}", source); return GL_NONE; } -GLenum GetComponent(PixelFormat format, bool is_first) { - switch (format) { - case PixelFormat::D24_UNORM_S8_UINT: - case PixelFormat::D32_FLOAT_S8_UINT: - return is_first ? GL_DEPTH_COMPONENT : GL_STENCIL_INDEX; - case PixelFormat::S8_UINT_D24_UNORM: - return is_first ? GL_STENCIL_INDEX : GL_DEPTH_COMPONENT; +GLenum AttachmentType(PixelFormat format) { + switch (const SurfaceType type = VideoCore::Surface::GetFormatType(format); type) { + case SurfaceType::Depth: + return GL_DEPTH_ATTACHMENT; + case SurfaceType::DepthStencil: + return GL_DEPTH_STENCIL_ATTACHMENT; default: - UNREACHABLE(); - return GL_DEPTH_COMPONENT; + UNIMPLEMENTED_MSG("Unimplemented type={}", type); + return GL_NONE; } } -void ApplyTextureDefaults(const SurfaceParams& params, GLuint texture) { - if (params.IsBuffer()) { - return; +[[nodiscard]] bool IsConverted(const Device& device, PixelFormat format, ImageType type) { + if (!device.HasASTC() && IsPixelFormatASTC(format)) { + return true; } - glTextureParameteri(texture, GL_TEXTURE_MIN_FILTER, GL_LINEAR); - glTextureParameteri(texture, GL_TEXTURE_MAG_FILTER, GL_LINEAR); - glTextureParameteri(texture, GL_TEXTURE_WRAP_S, GL_CLAMP_TO_EDGE); - glTextureParameteri(texture, GL_TEXTURE_WRAP_T, GL_CLAMP_TO_EDGE); - glTextureParameteri(texture, GL_TEXTURE_MAX_LEVEL, static_cast<GLint>(params.num_levels - 1)); - if (params.num_levels == 1) { - glTextureParameterf(texture, GL_TEXTURE_LOD_BIAS, 1000.0f); + switch (format) { + case PixelFormat::BC4_UNORM: + case PixelFormat::BC5_UNORM: + return type == ImageType::e3D; + default: + break; } + return false; } -OGLTexture CreateTexture(const SurfaceParams& params, GLenum target, GLenum internal_format, - OGLBuffer& texture_buffer) { - OGLTexture texture; - texture.Create(target); +[[nodiscard]] constexpr SwizzleSource ConvertGreenRed(SwizzleSource value) { + switch (value) { + case SwizzleSource::G: + return SwizzleSource::R; + default: + return value; + } +} - switch (params.target) { - case SurfaceTarget::Texture1D: - glTextureStorage1D(texture.handle, params.emulated_levels, internal_format, params.width); - break; - case SurfaceTarget::TextureBuffer: - texture_buffer.Create(); - glNamedBufferStorage(texture_buffer.handle, params.width * params.GetBytesPerPixel(), - nullptr, GL_DYNAMIC_STORAGE_BIT); - glTextureBuffer(texture.handle, internal_format, texture_buffer.handle); +void ApplySwizzle(GLuint handle, PixelFormat format, std::array<SwizzleSource, 4> swizzle) { + switch (format) { + case PixelFormat::D24_UNORM_S8_UINT: + case PixelFormat::D32_FLOAT_S8_UINT: + case PixelFormat::S8_UINT_D24_UNORM: + UNIMPLEMENTED_IF(swizzle[0] != SwizzleSource::R && swizzle[0] != SwizzleSource::G); + glTextureParameteri(handle, GL_DEPTH_STENCIL_TEXTURE_MODE, + TextureMode(format, swizzle[0] == SwizzleSource::R)); + std::ranges::transform(swizzle, swizzle.begin(), ConvertGreenRed); break; - case SurfaceTarget::Texture2D: - case SurfaceTarget::TextureCubemap: - glTextureStorage2D(texture.handle, params.emulated_levels, internal_format, params.width, - params.height); + default: break; - case SurfaceTarget::Texture3D: - case SurfaceTarget::Texture2DArray: - case SurfaceTarget::TextureCubeArray: - glTextureStorage3D(texture.handle, params.emulated_levels, internal_format, params.width, - params.height, params.depth); + } + std::array<GLint, 4> gl_swizzle; + std::ranges::transform(swizzle, gl_swizzle.begin(), Swizzle); + glTextureParameteriv(handle, GL_TEXTURE_SWIZZLE_RGBA, gl_swizzle.data()); +} + +[[nodiscard]] bool CanBeAccelerated(const TextureCacheRuntime& runtime, + const VideoCommon::ImageInfo& info) { + // Disable accelerated uploads for now as they don't implement swizzled uploads + return false; + switch (info.type) { + case ImageType::e2D: + case ImageType::e3D: + case ImageType::Linear: break; default: - UNREACHABLE(); + return false; + } + const GLenum internal_format = GetFormatTuple(info.format).internal_format; + const auto& format_info = runtime.FormatInfo(info.type, internal_format); + if (format_info.is_compressed) { + return false; + } + if (std::ranges::find(ACCELERATED_FORMATS, internal_format) == ACCELERATED_FORMATS.end()) { + return false; } + if (format_info.compatibility_by_size) { + return true; + } + const GLenum store_format = StoreFormat(BytesPerBlock(info.format)); + const GLenum store_class = runtime.FormatInfo(info.type, store_format).compatibility_class; + return format_info.compatibility_class == store_class; +} - ApplyTextureDefaults(params, texture.handle); +[[nodiscard]] CopyOrigin MakeCopyOrigin(VideoCommon::Offset3D offset, + VideoCommon::SubresourceLayers subresource, GLenum target) { + switch (target) { + case GL_TEXTURE_2D_ARRAY: + case GL_TEXTURE_2D_MULTISAMPLE_ARRAY: + return CopyOrigin{ + .level = static_cast<GLint>(subresource.base_level), + .x = static_cast<GLint>(offset.x), + .y = static_cast<GLint>(offset.y), + .z = static_cast<GLint>(subresource.base_layer), + }; + case GL_TEXTURE_3D: + return CopyOrigin{ + .level = static_cast<GLint>(subresource.base_level), + .x = static_cast<GLint>(offset.x), + .y = static_cast<GLint>(offset.y), + .z = static_cast<GLint>(offset.z), + }; + default: + UNIMPLEMENTED_MSG("Unimplemented copy target={}", target); + return CopyOrigin{.level = 0, .x = 0, .y = 0, .z = 0}; + } +} - return texture; +[[nodiscard]] CopyRegion MakeCopyRegion(VideoCommon::Extent3D extent, + VideoCommon::SubresourceLayers dst_subresource, + GLenum target) { + switch (target) { + case GL_TEXTURE_2D_ARRAY: + case GL_TEXTURE_2D_MULTISAMPLE_ARRAY: + return CopyRegion{ + .width = static_cast<GLsizei>(extent.width), + .height = static_cast<GLsizei>(extent.height), + .depth = static_cast<GLsizei>(dst_subresource.num_layers), + }; + case GL_TEXTURE_3D: + return CopyRegion{ + .width = static_cast<GLsizei>(extent.width), + .height = static_cast<GLsizei>(extent.height), + .depth = static_cast<GLsizei>(extent.depth), + }; + default: + UNIMPLEMENTED_MSG("Unimplemented copy target={}", target); + return CopyRegion{.width = 0, .height = 0, .depth = 0}; + } } -constexpr u32 EncodeSwizzle(SwizzleSource x_source, SwizzleSource y_source, SwizzleSource z_source, - SwizzleSource w_source) { - return (static_cast<u32>(x_source) << 24) | (static_cast<u32>(y_source) << 16) | - (static_cast<u32>(z_source) << 8) | static_cast<u32>(w_source); +void AttachTexture(GLuint fbo, GLenum attachment, const ImageView* image_view) { + if (False(image_view->flags & VideoCommon::ImageViewFlagBits::Slice)) { + const GLuint texture = image_view->DefaultHandle(); + glNamedFramebufferTexture(fbo, attachment, texture, 0); + return; + } + const GLuint texture = image_view->Handle(ImageViewType::e3D); + if (image_view->range.extent.layers > 1) { + // TODO: OpenGL doesn't support rendering to a fixed number of slices + glNamedFramebufferTexture(fbo, attachment, texture, 0); + } else { + const u32 slice = image_view->range.base.layer; + glNamedFramebufferTextureLayer(fbo, attachment, texture, 0, slice); + } } } // Anonymous namespace -CachedSurface::CachedSurface(const GPUVAddr gpu_addr, const SurfaceParams& params, - bool is_astc_supported) - : VideoCommon::SurfaceBase<View>(gpu_addr, params, is_astc_supported) { - if (is_converted) { - internal_format = params.srgb_conversion ? GL_SRGB8_ALPHA8 : GL_RGBA8; - format = GL_RGBA; - type = GL_UNSIGNED_BYTE; - } else { - const auto& tuple{GetFormatTuple(params.pixel_format)}; - internal_format = tuple.internal_format; - format = tuple.format; - type = tuple.type; - is_compressed = params.IsCompressed(); - } - target = GetTextureTarget(params.target); - texture = CreateTexture(params, target, internal_format, texture_buffer); - DecorateSurfaceName(); +ImageBufferMap::ImageBufferMap(GLuint handle_, u8* map, size_t size, OGLSync* sync_) + : span(map, size), sync{sync_}, handle{handle_} {} - u32 num_layers = 1; - if (params.is_layered || params.target == SurfaceTarget::Texture3D) { - num_layers = params.depth; +ImageBufferMap::~ImageBufferMap() { + if (sync) { + sync->Create(); } - - main_view = - CreateViewInner(ViewParams(params.target, 0, num_layers, 0, params.num_levels), true); } -CachedSurface::~CachedSurface() = default; +TextureCacheRuntime::TextureCacheRuntime(const Device& device_, ProgramManager& program_manager, + StateTracker& state_tracker_) + : device{device_}, state_tracker{state_tracker_}, util_shaders(program_manager) { + static constexpr std::array TARGETS{GL_TEXTURE_1D_ARRAY, GL_TEXTURE_2D_ARRAY, GL_TEXTURE_3D}; + for (size_t i = 0; i < TARGETS.size(); ++i) { + const GLenum target = TARGETS[i]; + for (const FormatTuple& tuple : FORMAT_TABLE) { + const GLenum format = tuple.internal_format; + GLint compat_class; + GLint compat_type; + GLint is_compressed; + glGetInternalformativ(target, format, GL_IMAGE_COMPATIBILITY_CLASS, 1, &compat_class); + glGetInternalformativ(target, format, GL_IMAGE_FORMAT_COMPATIBILITY_TYPE, 1, + &compat_type); + glGetInternalformativ(target, format, GL_TEXTURE_COMPRESSED, 1, &is_compressed); + const FormatProperties properties{ + .compatibility_class = static_cast<GLenum>(compat_class), + .compatibility_by_size = compat_type == GL_IMAGE_FORMAT_COMPATIBILITY_BY_SIZE, + .is_compressed = is_compressed == GL_TRUE, + }; + format_properties[i].emplace(format, properties); + } + } + has_broken_texture_view_formats = device.HasBrokenTextureViewFormats(); + + null_image_1d_array.Create(GL_TEXTURE_1D_ARRAY); + null_image_cube_array.Create(GL_TEXTURE_CUBE_MAP_ARRAY); + null_image_3d.Create(GL_TEXTURE_3D); + null_image_rect.Create(GL_TEXTURE_RECTANGLE); + glTextureStorage2D(null_image_1d_array.handle, 1, GL_R8, 1, 1); + glTextureStorage3D(null_image_cube_array.handle, 1, GL_R8, 1, 1, 6); + glTextureStorage3D(null_image_3d.handle, 1, GL_R8, 1, 1, 1); + glTextureStorage2D(null_image_rect.handle, 1, GL_R8, 1, 1); + + std::array<GLuint, 4> new_handles; + glGenTextures(static_cast<GLsizei>(new_handles.size()), new_handles.data()); + null_image_view_1d.handle = new_handles[0]; + null_image_view_2d.handle = new_handles[1]; + null_image_view_2d_array.handle = new_handles[2]; + null_image_view_cube.handle = new_handles[3]; + glTextureView(null_image_view_1d.handle, GL_TEXTURE_1D, null_image_1d_array.handle, GL_R8, 0, 1, + 0, 1); + glTextureView(null_image_view_2d.handle, GL_TEXTURE_2D, null_image_cube_array.handle, GL_R8, 0, + 1, 0, 1); + glTextureView(null_image_view_2d_array.handle, GL_TEXTURE_2D_ARRAY, + null_image_cube_array.handle, GL_R8, 0, 1, 0, 1); + glTextureView(null_image_view_cube.handle, GL_TEXTURE_CUBE_MAP, null_image_cube_array.handle, + GL_R8, 0, 1, 0, 6); + const std::array texture_handles{ + null_image_1d_array.handle, null_image_cube_array.handle, null_image_3d.handle, + null_image_rect.handle, null_image_view_1d.handle, null_image_view_2d.handle, + null_image_view_2d_array.handle, null_image_view_cube.handle, + }; + for (const GLuint handle : texture_handles) { + static constexpr std::array NULL_SWIZZLE{GL_ZERO, GL_ZERO, GL_ZERO, GL_ZERO}; + glTextureParameteriv(handle, GL_TEXTURE_SWIZZLE_RGBA, NULL_SWIZZLE.data()); + } + const auto set_view = [this](ImageViewType type, GLuint handle) { + if (device.HasDebuggingToolAttached()) { + const std::string name = fmt::format("NullImage {}", type); + glObjectLabel(GL_TEXTURE, handle, static_cast<GLsizei>(name.size()), name.data()); + } + null_image_views[static_cast<size_t>(type)] = handle; + }; + set_view(ImageViewType::e1D, null_image_view_1d.handle); + set_view(ImageViewType::e2D, null_image_view_2d.handle); + set_view(ImageViewType::Cube, null_image_view_cube.handle); + set_view(ImageViewType::e3D, null_image_3d.handle); + set_view(ImageViewType::e1DArray, null_image_1d_array.handle); + set_view(ImageViewType::e2DArray, null_image_view_2d_array.handle); + set_view(ImageViewType::CubeArray, null_image_cube_array.handle); + set_view(ImageViewType::Rect, null_image_rect.handle); +} -void CachedSurface::DownloadTexture(std::vector<u8>& staging_buffer) { - MICROPROFILE_SCOPE(OpenGL_Texture_Download); +TextureCacheRuntime::~TextureCacheRuntime() = default; - if (params.IsBuffer()) { - glGetNamedBufferSubData(texture_buffer.handle, 0, - static_cast<GLsizeiptr>(params.GetHostSizeInBytes(false)), - staging_buffer.data()); - return; - } +void TextureCacheRuntime::Finish() { + glFinish(); +} - SCOPE_EXIT({ glPixelStorei(GL_PACK_ROW_LENGTH, 0); }); +ImageBufferMap TextureCacheRuntime::MapUploadBuffer(size_t size) { + return upload_buffers.RequestMap(size, true); +} - for (u32 level = 0; level < params.emulated_levels; ++level) { - glPixelStorei(GL_PACK_ALIGNMENT, std::min(8U, params.GetRowAlignment(level, is_converted))); - glPixelStorei(GL_PACK_ROW_LENGTH, static_cast<GLint>(params.GetMipWidth(level))); - const std::size_t mip_offset = params.GetHostMipmapLevelOffset(level, is_converted); +ImageBufferMap TextureCacheRuntime::MapDownloadBuffer(size_t size) { + return download_buffers.RequestMap(size, false); +} - u8* const mip_data = staging_buffer.data() + mip_offset; - const GLsizei size = static_cast<GLsizei>(params.GetHostMipmapSize(level)); - if (is_compressed) { - glGetCompressedTextureImage(texture.handle, level, size, mip_data); - } else { - glGetTextureImage(texture.handle, level, format, type, size, mip_data); - } +void TextureCacheRuntime::CopyImage(Image& dst_image, Image& src_image, + std::span<const ImageCopy> copies) { + const GLuint dst_name = dst_image.Handle(); + const GLuint src_name = src_image.Handle(); + const GLenum dst_target = ImageTarget(dst_image.info); + const GLenum src_target = ImageTarget(src_image.info); + for (const ImageCopy& copy : copies) { + const auto src_origin = MakeCopyOrigin(copy.src_offset, copy.src_subresource, src_target); + const auto dst_origin = MakeCopyOrigin(copy.dst_offset, copy.dst_subresource, dst_target); + const auto region = MakeCopyRegion(copy.extent, copy.dst_subresource, dst_target); + glCopyImageSubData(src_name, src_target, src_origin.level, src_origin.x, src_origin.y, + src_origin.z, dst_name, dst_target, dst_origin.level, dst_origin.x, + dst_origin.y, dst_origin.z, region.width, region.height, region.depth); } } -void CachedSurface::UploadTexture(const std::vector<u8>& staging_buffer) { - MICROPROFILE_SCOPE(OpenGL_Texture_Upload); - SCOPE_EXIT({ glPixelStorei(GL_UNPACK_ROW_LENGTH, 0); }); - for (u32 level = 0; level < params.emulated_levels; ++level) { - UploadTextureMipmap(level, staging_buffer); +bool TextureCacheRuntime::CanImageBeCopied(const Image& dst, const Image& src) { + if (dst.info.type == ImageType::e3D && dst.info.format == PixelFormat::BC4_UNORM) { + return false; } + return true; } -void CachedSurface::UploadTextureMipmap(u32 level, const std::vector<u8>& staging_buffer) { - glPixelStorei(GL_UNPACK_ALIGNMENT, std::min(8U, params.GetRowAlignment(level, is_converted))); - glPixelStorei(GL_UNPACK_ROW_LENGTH, static_cast<GLint>(params.GetMipWidth(level))); - - const std::size_t mip_offset = params.GetHostMipmapLevelOffset(level, is_converted); - const u8* buffer{staging_buffer.data() + mip_offset}; - if (is_compressed) { - const auto image_size{static_cast<GLsizei>(params.GetHostMipmapSize(level))}; - switch (params.target) { - case SurfaceTarget::Texture2D: - glCompressedTextureSubImage2D(texture.handle, level, 0, 0, - static_cast<GLsizei>(params.GetMipWidth(level)), - static_cast<GLsizei>(params.GetMipHeight(level)), - internal_format, image_size, buffer); - break; - case SurfaceTarget::Texture3D: - case SurfaceTarget::Texture2DArray: - case SurfaceTarget::TextureCubeArray: - glCompressedTextureSubImage3D(texture.handle, level, 0, 0, 0, - static_cast<GLsizei>(params.GetMipWidth(level)), - static_cast<GLsizei>(params.GetMipHeight(level)), - static_cast<GLsizei>(params.GetMipDepth(level)), - internal_format, image_size, buffer); - break; - case SurfaceTarget::TextureCubemap: { - const std::size_t layer_size{params.GetHostLayerSize(level)}; - for (std::size_t face = 0; face < params.depth; ++face) { - glCompressedTextureSubImage3D(texture.handle, level, 0, 0, static_cast<GLint>(face), - static_cast<GLsizei>(params.GetMipWidth(level)), - static_cast<GLsizei>(params.GetMipHeight(level)), 1, - internal_format, static_cast<GLsizei>(layer_size), - buffer); - buffer += layer_size; - } - break; - } - default: - UNREACHABLE(); - } +void TextureCacheRuntime::EmulateCopyImage(Image& dst, Image& src, + std::span<const ImageCopy> copies) { + if (dst.info.type == ImageType::e3D && dst.info.format == PixelFormat::BC4_UNORM) { + ASSERT(src.info.type == ImageType::e3D); + util_shaders.CopyBC4(dst, src, copies); } else { - switch (params.target) { - case SurfaceTarget::Texture1D: - glTextureSubImage1D(texture.handle, level, 0, params.GetMipWidth(level), format, type, - buffer); - break; - case SurfaceTarget::TextureBuffer: - ASSERT(level == 0); - glNamedBufferSubData(texture_buffer.handle, 0, - params.GetMipWidth(level) * params.GetBytesPerPixel(), buffer); - break; - case SurfaceTarget::Texture1DArray: - case SurfaceTarget::Texture2D: - glTextureSubImage2D(texture.handle, level, 0, 0, params.GetMipWidth(level), - params.GetMipHeight(level), format, type, buffer); - break; - case SurfaceTarget::Texture3D: - case SurfaceTarget::Texture2DArray: - case SurfaceTarget::TextureCubeArray: - glTextureSubImage3D( - texture.handle, level, 0, 0, 0, static_cast<GLsizei>(params.GetMipWidth(level)), - static_cast<GLsizei>(params.GetMipHeight(level)), - static_cast<GLsizei>(params.GetMipDepth(level)), format, type, buffer); - break; - case SurfaceTarget::TextureCubemap: - for (std::size_t face = 0; face < params.depth; ++face) { - glTextureSubImage3D(texture.handle, level, 0, 0, static_cast<GLint>(face), - params.GetMipWidth(level), params.GetMipHeight(level), 1, - format, type, buffer); - buffer += params.GetHostLayerSize(level); - } - break; - default: - UNREACHABLE(); - } + UNREACHABLE(); } } -void CachedSurface::DecorateSurfaceName() { - LabelGLObject(GL_TEXTURE, texture.handle, GetGpuAddr(), params.TargetName()); -} +void TextureCacheRuntime::BlitFramebuffer(Framebuffer* dst, Framebuffer* src, + const std::array<Offset2D, 2>& dst_region, + const std::array<Offset2D, 2>& src_region, + Tegra::Engines::Fermi2D::Filter filter, + Tegra::Engines::Fermi2D::Operation operation) { + state_tracker.NotifyScissor0(); + state_tracker.NotifyRasterizeEnable(); + state_tracker.NotifyFramebufferSRGB(); -void CachedSurfaceView::DecorateViewName(GPUVAddr gpu_addr, const std::string& prefix) { - LabelGLObject(GL_TEXTURE, main_view.handle, gpu_addr, prefix); + ASSERT(dst->BufferBits() == src->BufferBits()); + + glEnable(GL_FRAMEBUFFER_SRGB); + glDisable(GL_RASTERIZER_DISCARD); + glDisablei(GL_SCISSOR_TEST, 0); + + const GLbitfield buffer_bits = dst->BufferBits(); + const bool has_depth = (buffer_bits & ~GL_COLOR_BUFFER_BIT) != 0; + const bool is_linear = !has_depth && filter == Tegra::Engines::Fermi2D::Filter::Bilinear; + glBlitNamedFramebuffer(src->Handle(), dst->Handle(), src_region[0].x, src_region[0].y, + src_region[1].x, src_region[1].y, dst_region[0].x, dst_region[0].y, + dst_region[1].x, dst_region[1].y, buffer_bits, + is_linear ? GL_LINEAR : GL_NEAREST); } -View CachedSurface::CreateView(const ViewParams& view_key) { - return CreateViewInner(view_key, false); +void TextureCacheRuntime::AccelerateImageUpload(Image& image, const ImageBufferMap& map, + size_t buffer_offset, + std::span<const SwizzleParameters> swizzles) { + switch (image.info.type) { + case ImageType::e2D: + return util_shaders.BlockLinearUpload2D(image, map, buffer_offset, swizzles); + case ImageType::e3D: + return util_shaders.BlockLinearUpload3D(image, map, buffer_offset, swizzles); + case ImageType::Linear: + return util_shaders.PitchUpload(image, map, buffer_offset, swizzles); + default: + UNREACHABLE(); + break; + } } -View CachedSurface::CreateViewInner(const ViewParams& view_key, const bool is_proxy) { - auto view = std::make_shared<CachedSurfaceView>(*this, view_key, is_proxy); - views[view_key] = view; - if (!is_proxy) - view->DecorateViewName(gpu_addr, params.TargetName() + "V:" + std::to_string(view_count++)); - return view; +void TextureCacheRuntime::InsertUploadMemoryBarrier() { + glMemoryBarrier(GL_TEXTURE_FETCH_BARRIER_BIT | GL_SHADER_IMAGE_ACCESS_BARRIER_BIT); } -CachedSurfaceView::CachedSurfaceView(CachedSurface& surface, const ViewParams& params, - bool is_proxy) - : VideoCommon::ViewBase(params), surface{surface}, format{surface.internal_format}, - target{GetTextureTarget(params.target)}, is_proxy{is_proxy} { - if (!is_proxy) { - main_view = CreateTextureView(); +FormatProperties TextureCacheRuntime::FormatInfo(ImageType type, GLenum internal_format) const { + switch (type) { + case ImageType::e1D: + return format_properties[0].at(internal_format); + case ImageType::e2D: + case ImageType::Linear: + return format_properties[1].at(internal_format); + case ImageType::e3D: + return format_properties[2].at(internal_format); + default: + UNREACHABLE(); + return FormatProperties{}; } } -CachedSurfaceView::~CachedSurfaceView() = default; +TextureCacheRuntime::StagingBuffers::StagingBuffers(GLenum storage_flags_, GLenum map_flags_) + : storage_flags{storage_flags_}, map_flags{map_flags_} {} -void CachedSurfaceView::Attach(GLenum attachment, GLenum fb_target) const { - ASSERT(params.num_levels == 1); +TextureCacheRuntime::StagingBuffers::~StagingBuffers() = default; - if (params.target == SurfaceTarget::Texture3D) { - if (params.num_layers > 1) { - ASSERT(params.base_layer == 0); - glFramebufferTexture(fb_target, attachment, surface.texture.handle, params.base_level); - } else { - glFramebufferTexture3D(fb_target, attachment, target, surface.texture.handle, - params.base_level, params.base_layer); - } - return; +ImageBufferMap TextureCacheRuntime::StagingBuffers::RequestMap(size_t requested_size, + bool insert_fence) { + const size_t index = RequestBuffer(requested_size); + OGLSync* const sync = insert_fence ? &syncs[index] : nullptr; + return ImageBufferMap(buffers[index].handle, maps[index], requested_size, sync); +} + +size_t TextureCacheRuntime::StagingBuffers::RequestBuffer(size_t requested_size) { + if (const std::optional<size_t> index = FindBuffer(requested_size); index) { + return *index; } - if (params.num_layers > 1) { - UNIMPLEMENTED_IF(params.base_layer != 0); - glFramebufferTexture(fb_target, attachment, GetTexture(), 0); - return; + OGLBuffer& buffer = buffers.emplace_back(); + buffer.Create(); + glNamedBufferStorage(buffer.handle, requested_size, nullptr, + storage_flags | GL_MAP_PERSISTENT_BIT); + maps.push_back(static_cast<u8*>(glMapNamedBufferRange(buffer.handle, 0, requested_size, + map_flags | GL_MAP_PERSISTENT_BIT))); + + syncs.emplace_back(); + sizes.push_back(requested_size); + + ASSERT(syncs.size() == buffers.size() && buffers.size() == maps.size() && + maps.size() == sizes.size()); + + return buffers.size() - 1; +} + +std::optional<size_t> TextureCacheRuntime::StagingBuffers::FindBuffer(size_t requested_size) { + size_t smallest_buffer = std::numeric_limits<size_t>::max(); + std::optional<size_t> found; + const size_t num_buffers = sizes.size(); + for (size_t index = 0; index < num_buffers; ++index) { + const size_t buffer_size = sizes[index]; + if (buffer_size < requested_size || buffer_size >= smallest_buffer) { + continue; + } + if (syncs[index].handle != 0) { + GLint status; + glGetSynciv(syncs[index].handle, GL_SYNC_STATUS, 1, nullptr, &status); + if (status != GL_SIGNALED) { + continue; + } + syncs[index].Release(); + } + smallest_buffer = buffer_size; + found = index; } + return found; +} - const GLenum view_target = surface.GetTarget(); - const GLuint texture = surface.GetTexture(); - switch (surface.GetSurfaceParams().target) { - case SurfaceTarget::Texture1D: - glFramebufferTexture1D(fb_target, attachment, view_target, texture, params.base_level); +Image::Image(TextureCacheRuntime& runtime, const VideoCommon::ImageInfo& info_, GPUVAddr gpu_addr_, + VAddr cpu_addr_) + : VideoCommon::ImageBase(info_, gpu_addr_, cpu_addr_) { + if (CanBeAccelerated(runtime, info)) { + flags |= ImageFlagBits::AcceleratedUpload; + } + if (IsConverted(runtime.device, info.format, info.type)) { + flags |= ImageFlagBits::Converted; + gl_internal_format = IsPixelFormatSRGB(info.format) ? GL_SRGB8_ALPHA8 : GL_RGBA8; + gl_format = GL_RGBA; + gl_type = GL_UNSIGNED_INT_8_8_8_8_REV; + } else { + const auto& tuple = GetFormatTuple(info.format); + gl_internal_format = tuple.internal_format; + gl_format = tuple.format; + gl_type = tuple.type; + } + const GLenum target = ImageTarget(info); + const GLsizei width = info.size.width; + const GLsizei height = info.size.height; + const GLsizei depth = info.size.depth; + const int max_host_mip_levels = std::bit_width(info.size.width); + const GLsizei num_levels = std::min(info.resources.levels, max_host_mip_levels); + const GLsizei num_layers = info.resources.layers; + const GLsizei num_samples = info.num_samples; + + GLuint handle = 0; + if (target != GL_TEXTURE_BUFFER) { + texture.Create(target); + handle = texture.handle; + } + switch (target) { + case GL_TEXTURE_1D_ARRAY: + glTextureStorage2D(handle, num_levels, gl_internal_format, width, num_layers); break; - case SurfaceTarget::Texture2D: - glFramebufferTexture2D(fb_target, attachment, view_target, texture, params.base_level); + case GL_TEXTURE_2D_ARRAY: + glTextureStorage3D(handle, num_levels, gl_internal_format, width, height, num_layers); break; - case SurfaceTarget::Texture1DArray: - case SurfaceTarget::Texture2DArray: - case SurfaceTarget::TextureCubemap: - case SurfaceTarget::TextureCubeArray: - glFramebufferTextureLayer(fb_target, attachment, texture, params.base_level, - params.base_layer); + case GL_TEXTURE_2D_MULTISAMPLE_ARRAY: { + // TODO: Where should 'fixedsamplelocations' come from? + const auto [samples_x, samples_y] = SamplesLog2(info.num_samples); + glTextureStorage3DMultisample(handle, num_samples, gl_internal_format, width >> samples_x, + height >> samples_y, num_layers, GL_FALSE); + break; + } + case GL_TEXTURE_RECTANGLE: + glTextureStorage2D(handle, num_levels, gl_internal_format, width, height); + break; + case GL_TEXTURE_3D: + glTextureStorage3D(handle, num_levels, gl_internal_format, width, height, depth); + break; + case GL_TEXTURE_BUFFER: + buffer.Create(); + glNamedBufferStorage(buffer.handle, guest_size_bytes, nullptr, 0); break; default: - UNIMPLEMENTED(); + UNREACHABLE_MSG("Invalid target=0x{:x}", target); + break; + } + if (runtime.device.HasDebuggingToolAttached()) { + const std::string name = VideoCommon::Name(*this); + glObjectLabel(target == GL_TEXTURE_BUFFER ? GL_BUFFER : GL_TEXTURE, handle, + static_cast<GLsizei>(name.size()), name.data()); } } -GLuint CachedSurfaceView::GetTexture(SwizzleSource x_source, SwizzleSource y_source, - SwizzleSource z_source, SwizzleSource w_source) { - if (GetSurfaceParams().IsBuffer()) { - return GetTexture(); - } - const u32 new_swizzle = EncodeSwizzle(x_source, y_source, z_source, w_source); - if (current_swizzle == new_swizzle) { - return current_view; - } - current_swizzle = new_swizzle; +void Image::UploadMemory(const ImageBufferMap& map, size_t buffer_offset, + std::span<const VideoCommon::BufferImageCopy> copies) { + glBindBuffer(GL_PIXEL_UNPACK_BUFFER, map.Handle()); + glFlushMappedBufferRange(GL_PIXEL_UNPACK_BUFFER, buffer_offset, unswizzled_size_bytes); - const auto [entry, is_cache_miss] = view_cache.try_emplace(new_swizzle); - OGLTextureView& view = entry->second; - if (!is_cache_miss) { - current_view = view.handle; - return view.handle; - } - view = CreateTextureView(); - current_view = view.handle; + glPixelStorei(GL_UNPACK_ALIGNMENT, 1); - std::array swizzle{x_source, y_source, z_source, w_source}; + u32 current_row_length = std::numeric_limits<u32>::max(); + u32 current_image_height = std::numeric_limits<u32>::max(); - switch (const PixelFormat format = GetSurfaceParams().pixel_format) { - case PixelFormat::D24_UNORM_S8_UINT: - case PixelFormat::D32_FLOAT_S8_UINT: - case PixelFormat::S8_UINT_D24_UNORM: - UNIMPLEMENTED_IF(x_source != SwizzleSource::R && x_source != SwizzleSource::G); - glTextureParameteri(view.handle, GL_DEPTH_STENCIL_TEXTURE_MODE, - GetComponent(format, x_source == SwizzleSource::R)); - - // Make sure we sample the first component - std::transform(swizzle.begin(), swizzle.end(), swizzle.begin(), [](SwizzleSource value) { - return value == SwizzleSource::G ? SwizzleSource::R : value; - }); - [[fallthrough]]; - default: { - const std::array gl_swizzle = {GetSwizzleSource(swizzle[0]), GetSwizzleSource(swizzle[1]), - GetSwizzleSource(swizzle[2]), GetSwizzleSource(swizzle[3])}; - glTextureParameteriv(view.handle, GL_TEXTURE_SWIZZLE_RGBA, gl_swizzle.data()); - break; - } + for (const VideoCommon::BufferImageCopy& copy : copies) { + if (current_row_length != copy.buffer_row_length) { + current_row_length = copy.buffer_row_length; + glPixelStorei(GL_UNPACK_ROW_LENGTH, current_row_length); + } + if (current_image_height != copy.buffer_image_height) { + current_image_height = copy.buffer_image_height; + glPixelStorei(GL_UNPACK_IMAGE_HEIGHT, current_image_height); + } + CopyBufferToImage(copy, buffer_offset); } - return view.handle; } -OGLTextureView CachedSurfaceView::CreateTextureView() const { - OGLTextureView texture_view; - texture_view.Create(); - - if (target == GL_TEXTURE_3D) { - glTextureView(texture_view.handle, target, surface.texture.handle, format, - params.base_level, params.num_levels, 0, 1); - } else { - glTextureView(texture_view.handle, target, surface.texture.handle, format, - params.base_level, params.num_levels, params.base_layer, params.num_layers); +void Image::UploadMemory(const ImageBufferMap& map, size_t buffer_offset, + std::span<const VideoCommon::BufferCopy> copies) { + for (const VideoCommon::BufferCopy& copy : copies) { + glCopyNamedBufferSubData(map.Handle(), buffer.handle, copy.src_offset + buffer_offset, + copy.dst_offset, copy.size); } - ApplyTextureDefaults(surface.GetSurfaceParams(), texture_view.handle); - - return texture_view; } -TextureCacheOpenGL::TextureCacheOpenGL(VideoCore::RasterizerInterface& rasterizer, - Tegra::Engines::Maxwell3D& maxwell3d, - Tegra::MemoryManager& gpu_memory, const Device& device, - StateTracker& state_tracker_) - : TextureCacheBase{rasterizer, maxwell3d, gpu_memory, device.HasASTC()}, state_tracker{ - state_tracker_} { - src_framebuffer.Create(); - dst_framebuffer.Create(); -} +void Image::DownloadMemory(ImageBufferMap& map, size_t buffer_offset, + std::span<const VideoCommon::BufferImageCopy> copies) { + glMemoryBarrier(GL_PIXEL_BUFFER_BARRIER_BIT); // TODO: Move this to its own API -TextureCacheOpenGL::~TextureCacheOpenGL() = default; + glBindBuffer(GL_PIXEL_PACK_BUFFER, map.Handle()); + glPixelStorei(GL_PACK_ALIGNMENT, 1); -Surface TextureCacheOpenGL::CreateSurface(GPUVAddr gpu_addr, const SurfaceParams& params) { - return std::make_shared<CachedSurface>(gpu_addr, params, is_astc_supported); -} + u32 current_row_length = std::numeric_limits<u32>::max(); + u32 current_image_height = std::numeric_limits<u32>::max(); -void TextureCacheOpenGL::ImageCopy(Surface& src_surface, Surface& dst_surface, - const VideoCommon::CopyParams& copy_params) { - const auto& src_params = src_surface->GetSurfaceParams(); - const auto& dst_params = dst_surface->GetSurfaceParams(); - if (src_params.type != dst_params.type) { - // A fallback is needed - return; + for (const VideoCommon::BufferImageCopy& copy : copies) { + if (current_row_length != copy.buffer_row_length) { + current_row_length = copy.buffer_row_length; + glPixelStorei(GL_PACK_ROW_LENGTH, current_row_length); + } + if (current_image_height != copy.buffer_image_height) { + current_image_height = copy.buffer_image_height; + glPixelStorei(GL_PACK_IMAGE_HEIGHT, current_image_height); + } + CopyImageToBuffer(copy, buffer_offset); } - const auto src_handle = src_surface->GetTexture(); - const auto src_target = src_surface->GetTarget(); - const auto dst_handle = dst_surface->GetTexture(); - const auto dst_target = dst_surface->GetTarget(); - glCopyImageSubData(src_handle, src_target, copy_params.source_level, copy_params.source_x, - copy_params.source_y, copy_params.source_z, dst_handle, dst_target, - copy_params.dest_level, copy_params.dest_x, copy_params.dest_y, - copy_params.dest_z, copy_params.width, copy_params.height, - copy_params.depth); } -void TextureCacheOpenGL::ImageBlit(View& src_view, View& dst_view, - const Tegra::Engines::Fermi2D::Config& copy_config) { - const auto& src_params{src_view->GetSurfaceParams()}; - const auto& dst_params{dst_view->GetSurfaceParams()}; - UNIMPLEMENTED_IF(src_params.depth != 1); - UNIMPLEMENTED_IF(dst_params.depth != 1); - - state_tracker.NotifyScissor0(); - state_tracker.NotifyFramebuffer(); - state_tracker.NotifyRasterizeEnable(); - state_tracker.NotifyFramebufferSRGB(); +void Image::CopyBufferToImage(const VideoCommon::BufferImageCopy& copy, size_t buffer_offset) { + // Compressed formats don't have a pixel format or type + const bool is_compressed = gl_format == GL_NONE; + const void* const offset = reinterpret_cast<const void*>(copy.buffer_offset + buffer_offset); - if (dst_params.srgb_conversion) { - glEnable(GL_FRAMEBUFFER_SRGB); - } else { - glDisable(GL_FRAMEBUFFER_SRGB); + switch (info.type) { + case ImageType::e1D: + if (is_compressed) { + glCompressedTextureSubImage2D(texture.handle, copy.image_subresource.base_level, + copy.image_offset.x, copy.image_subresource.base_layer, + copy.image_extent.width, + copy.image_subresource.num_layers, gl_internal_format, + static_cast<GLsizei>(copy.buffer_size), offset); + } else { + glTextureSubImage2D(texture.handle, copy.image_subresource.base_level, + copy.image_offset.x, copy.image_subresource.base_layer, + copy.image_extent.width, copy.image_subresource.num_layers, + gl_format, gl_type, offset); + } + break; + case ImageType::e2D: + case ImageType::Linear: + if (is_compressed) { + glCompressedTextureSubImage3D( + texture.handle, copy.image_subresource.base_level, copy.image_offset.x, + copy.image_offset.y, copy.image_subresource.base_layer, copy.image_extent.width, + copy.image_extent.height, copy.image_subresource.num_layers, gl_internal_format, + static_cast<GLsizei>(copy.buffer_size), offset); + } else { + glTextureSubImage3D(texture.handle, copy.image_subresource.base_level, + copy.image_offset.x, copy.image_offset.y, + copy.image_subresource.base_layer, copy.image_extent.width, + copy.image_extent.height, copy.image_subresource.num_layers, + gl_format, gl_type, offset); + } + break; + case ImageType::e3D: + if (is_compressed) { + glCompressedTextureSubImage3D( + texture.handle, copy.image_subresource.base_level, copy.image_offset.x, + copy.image_offset.y, copy.image_offset.z, copy.image_extent.width, + copy.image_extent.height, copy.image_extent.depth, gl_internal_format, + static_cast<GLsizei>(copy.buffer_size), offset); + } else { + glTextureSubImage3D(texture.handle, copy.image_subresource.base_level, + copy.image_offset.x, copy.image_offset.y, copy.image_offset.z, + copy.image_extent.width, copy.image_extent.height, + copy.image_extent.depth, gl_format, gl_type, offset); + } + break; + default: + UNREACHABLE(); } - glDisable(GL_RASTERIZER_DISCARD); - glDisablei(GL_SCISSOR_TEST, 0); - - glBindFramebuffer(GL_READ_FRAMEBUFFER, src_framebuffer.handle); - glBindFramebuffer(GL_DRAW_FRAMEBUFFER, dst_framebuffer.handle); - - GLenum buffers = 0; - if (src_params.type == SurfaceType::ColorTexture) { - src_view->Attach(GL_COLOR_ATTACHMENT0, GL_READ_FRAMEBUFFER); - glFramebufferTexture2D(GL_READ_FRAMEBUFFER, GL_DEPTH_STENCIL_ATTACHMENT, GL_TEXTURE_2D, 0, - 0); - - dst_view->Attach(GL_COLOR_ATTACHMENT0, GL_DRAW_FRAMEBUFFER); - glFramebufferTexture2D(GL_DRAW_FRAMEBUFFER, GL_DEPTH_STENCIL_ATTACHMENT, GL_TEXTURE_2D, 0, - 0); - - buffers = GL_COLOR_BUFFER_BIT; - } else if (src_params.type == SurfaceType::Depth) { - glFramebufferTexture2D(GL_READ_FRAMEBUFFER, GL_COLOR_ATTACHMENT0, GL_TEXTURE_2D, 0, 0); - src_view->Attach(GL_DEPTH_ATTACHMENT, GL_READ_FRAMEBUFFER); - glFramebufferTexture2D(GL_READ_FRAMEBUFFER, GL_STENCIL_ATTACHMENT, GL_TEXTURE_2D, 0, 0); +} - glFramebufferTexture2D(GL_DRAW_FRAMEBUFFER, GL_COLOR_ATTACHMENT0, GL_TEXTURE_2D, 0, 0); - dst_view->Attach(GL_DEPTH_ATTACHMENT, GL_DRAW_FRAMEBUFFER); - glFramebufferTexture2D(GL_DRAW_FRAMEBUFFER, GL_STENCIL_ATTACHMENT, GL_TEXTURE_2D, 0, 0); +void Image::CopyImageToBuffer(const VideoCommon::BufferImageCopy& copy, size_t buffer_offset) { + const GLint x_offset = copy.image_offset.x; + const GLsizei width = copy.image_extent.width; - buffers = GL_DEPTH_BUFFER_BIT; - } else if (src_params.type == SurfaceType::DepthStencil) { - glFramebufferTexture2D(GL_READ_FRAMEBUFFER, GL_COLOR_ATTACHMENT0, GL_TEXTURE_2D, 0, 0); - src_view->Attach(GL_DEPTH_STENCIL_ATTACHMENT, GL_READ_FRAMEBUFFER); + const GLint level = copy.image_subresource.base_level; + const GLsizei buffer_size = static_cast<GLsizei>(copy.buffer_size); + void* const offset = reinterpret_cast<void*>(copy.buffer_offset + buffer_offset); - glFramebufferTexture2D(GL_DRAW_FRAMEBUFFER, GL_COLOR_ATTACHMENT0, GL_TEXTURE_2D, 0, 0); - dst_view->Attach(GL_DEPTH_STENCIL_ATTACHMENT, GL_DRAW_FRAMEBUFFER); + GLint y_offset = 0; + GLint z_offset = 0; + GLsizei height = 1; + GLsizei depth = 1; - buffers = GL_DEPTH_BUFFER_BIT | GL_STENCIL_BUFFER_BIT; + switch (info.type) { + case ImageType::e1D: + y_offset = copy.image_subresource.base_layer; + height = copy.image_subresource.num_layers; + break; + case ImageType::e2D: + case ImageType::Linear: + y_offset = copy.image_offset.y; + z_offset = copy.image_subresource.base_layer; + height = copy.image_extent.height; + depth = copy.image_subresource.num_layers; + break; + case ImageType::e3D: + y_offset = copy.image_offset.y; + z_offset = copy.image_offset.z; + height = copy.image_extent.height; + depth = copy.image_extent.depth; + break; + default: + UNREACHABLE(); + } + // Compressed formats don't have a pixel format or type + const bool is_compressed = gl_format == GL_NONE; + if (is_compressed) { + glGetCompressedTextureSubImage(texture.handle, level, x_offset, y_offset, z_offset, width, + height, depth, buffer_size, offset); + } else { + glGetTextureSubImage(texture.handle, level, x_offset, y_offset, z_offset, width, height, + depth, gl_format, gl_type, buffer_size, offset); } - - const Common::Rectangle<u32>& src_rect = copy_config.src_rect; - const Common::Rectangle<u32>& dst_rect = copy_config.dst_rect; - const bool is_linear = copy_config.filter == Tegra::Engines::Fermi2D::Filter::Linear; - - glBlitFramebuffer(static_cast<GLint>(src_rect.left), static_cast<GLint>(src_rect.top), - static_cast<GLint>(src_rect.right), static_cast<GLint>(src_rect.bottom), - static_cast<GLint>(dst_rect.left), static_cast<GLint>(dst_rect.top), - static_cast<GLint>(dst_rect.right), static_cast<GLint>(dst_rect.bottom), - buffers, - is_linear && (buffers == GL_COLOR_BUFFER_BIT) ? GL_LINEAR : GL_NEAREST); } -void TextureCacheOpenGL::BufferCopy(Surface& src_surface, Surface& dst_surface) { - MICROPROFILE_SCOPE(OpenGL_Texture_Buffer_Copy); - const auto& src_params = src_surface->GetSurfaceParams(); - const auto& dst_params = dst_surface->GetSurfaceParams(); - UNIMPLEMENTED_IF(src_params.num_levels > 1 || dst_params.num_levels > 1); +ImageView::ImageView(TextureCacheRuntime& runtime, const VideoCommon::ImageViewInfo& info, + ImageId image_id_, Image& image) + : VideoCommon::ImageViewBase{info, image.info, image_id_}, views{runtime.null_image_views} { + const Device& device = runtime.device; + if (True(image.flags & ImageFlagBits::Converted)) { + internal_format = IsPixelFormatSRGB(info.format) ? GL_SRGB8_ALPHA8 : GL_RGBA8; + } else { + internal_format = GetFormatTuple(format).internal_format; + } + VideoCommon::SubresourceRange flatten_range = info.range; + std::array<GLuint, 2> handles; + stored_views.reserve(2); - const auto source_format = GetFormatTuple(src_params.pixel_format); - const auto dest_format = GetFormatTuple(dst_params.pixel_format); + switch (info.type) { + case ImageViewType::e1DArray: + flatten_range.extent.layers = 1; + [[fallthrough]]; + case ImageViewType::e1D: + glGenTextures(2, handles.data()); + SetupView(device, image, ImageViewType::e1D, handles[0], info, flatten_range); + SetupView(device, image, ImageViewType::e1DArray, handles[1], info, info.range); + break; + case ImageViewType::e2DArray: + flatten_range.extent.layers = 1; + [[fallthrough]]; + case ImageViewType::e2D: + if (True(flags & VideoCommon::ImageViewFlagBits::Slice)) { + // 2D and 2D array views on a 3D textures are used exclusively for render targets + ASSERT(info.range.extent.levels == 1); + const VideoCommon::SubresourceRange slice_range{ + .base = {.level = info.range.base.level, .layer = 0}, + .extent = {.levels = 1, .layers = 1}, + }; + glGenTextures(1, handles.data()); + SetupView(device, image, ImageViewType::e3D, handles[0], info, slice_range); + break; + } + glGenTextures(2, handles.data()); + SetupView(device, image, ImageViewType::e2D, handles[0], info, flatten_range); + SetupView(device, image, ImageViewType::e2DArray, handles[1], info, info.range); + break; + case ImageViewType::e3D: + glGenTextures(1, handles.data()); + SetupView(device, image, ImageViewType::e3D, handles[0], info, info.range); + break; + case ImageViewType::CubeArray: + flatten_range.extent.layers = 6; + [[fallthrough]]; + case ImageViewType::Cube: + glGenTextures(2, handles.data()); + SetupView(device, image, ImageViewType::Cube, handles[0], info, flatten_range); + SetupView(device, image, ImageViewType::CubeArray, handles[1], info, info.range); + break; + case ImageViewType::Rect: + glGenTextures(1, handles.data()); + SetupView(device, image, ImageViewType::Rect, handles[0], info, info.range); + break; + case ImageViewType::Buffer: + glCreateTextures(GL_TEXTURE_BUFFER, 1, handles.data()); + SetupView(device, image, ImageViewType::Buffer, handles[0], info, info.range); + break; + } + default_handle = Handle(info.type); +} - const std::size_t source_size = src_surface->GetHostSizeInBytes(); - const std::size_t dest_size = dst_surface->GetHostSizeInBytes(); +ImageView::ImageView(TextureCacheRuntime& runtime, const VideoCommon::NullImageParams& params) + : VideoCommon::ImageViewBase{params}, views{runtime.null_image_views} {} - const std::size_t buffer_size = std::max(source_size, dest_size); +void ImageView::SetupView(const Device& device, Image& image, ImageViewType view_type, + GLuint handle, const VideoCommon::ImageViewInfo& info, + VideoCommon::SubresourceRange view_range) { + if (info.type == ImageViewType::Buffer) { + // TODO: Take offset from buffer cache + glTextureBufferRange(handle, internal_format, image.buffer.handle, 0, + image.guest_size_bytes); + } else { + const GLuint parent = image.texture.handle; + const GLenum target = ImageTarget(view_type, image.info.num_samples); + glTextureView(handle, target, parent, internal_format, view_range.base.level, + view_range.extent.levels, view_range.base.layer, view_range.extent.layers); + if (!info.IsRenderTarget()) { + ApplySwizzle(handle, format, info.Swizzle()); + } + } + if (device.HasDebuggingToolAttached()) { + const std::string name = VideoCommon::Name(*this, view_type); + glObjectLabel(GL_TEXTURE, handle, static_cast<GLsizei>(name.size()), name.data()); + } + stored_views.emplace_back().handle = handle; + views[static_cast<size_t>(view_type)] = handle; +} - GLuint copy_pbo_handle = FetchPBO(buffer_size); +Sampler::Sampler(TextureCacheRuntime& runtime, const TSCEntry& config) { + const GLenum compare_mode = config.depth_compare_enabled ? GL_COMPARE_REF_TO_TEXTURE : GL_NONE; + const GLenum compare_func = MaxwellToGL::DepthCompareFunc(config.depth_compare_func); + const GLenum mag = MaxwellToGL::TextureFilterMode(config.mag_filter, TextureMipmapFilter::None); + const GLenum min = MaxwellToGL::TextureFilterMode(config.min_filter, config.mipmap_filter); + const GLenum reduction_filter = MaxwellToGL::ReductionFilter(config.reduction_filter); + const GLint seamless = config.cubemap_interface_filtering ? GL_TRUE : GL_FALSE; + + UNIMPLEMENTED_IF(config.cubemap_anisotropy != 1); + UNIMPLEMENTED_IF(config.float_coord_normalization != 0); + + sampler.Create(); + const GLuint handle = sampler.handle; + glSamplerParameteri(handle, GL_TEXTURE_WRAP_S, MaxwellToGL::WrapMode(config.wrap_u)); + glSamplerParameteri(handle, GL_TEXTURE_WRAP_T, MaxwellToGL::WrapMode(config.wrap_v)); + glSamplerParameteri(handle, GL_TEXTURE_WRAP_R, MaxwellToGL::WrapMode(config.wrap_p)); + glSamplerParameteri(handle, GL_TEXTURE_COMPARE_MODE, compare_mode); + glSamplerParameteri(handle, GL_TEXTURE_COMPARE_FUNC, compare_func); + glSamplerParameteri(handle, GL_TEXTURE_MAG_FILTER, mag); + glSamplerParameteri(handle, GL_TEXTURE_MIN_FILTER, min); + glSamplerParameterf(handle, GL_TEXTURE_LOD_BIAS, config.LodBias()); + glSamplerParameterf(handle, GL_TEXTURE_MIN_LOD, config.MinLod()); + glSamplerParameterf(handle, GL_TEXTURE_MAX_LOD, config.MaxLod()); + glSamplerParameterfv(handle, GL_TEXTURE_BORDER_COLOR, config.BorderColor().data()); + + if (GLAD_GL_ARB_texture_filter_anisotropic || GLAD_GL_EXT_texture_filter_anisotropic) { + glSamplerParameterf(handle, GL_TEXTURE_MAX_ANISOTROPY, config.MaxAnisotropy()); + } else { + LOG_WARNING(Render_OpenGL, "GL_ARB_texture_filter_anisotropic is required"); + } + if (GLAD_GL_ARB_texture_filter_minmax || GLAD_GL_EXT_texture_filter_minmax) { + glSamplerParameteri(handle, GL_TEXTURE_REDUCTION_MODE_ARB, reduction_filter); + } else if (reduction_filter != GL_WEIGHTED_AVERAGE_ARB) { + LOG_WARNING(Render_OpenGL, "GL_ARB_texture_filter_minmax is required"); + } + if (GLAD_GL_ARB_seamless_cubemap_per_texture || GLAD_GL_AMD_seamless_cubemap_per_texture) { + glSamplerParameteri(handle, GL_TEXTURE_CUBE_MAP_SEAMLESS, seamless); + } else if (seamless == GL_FALSE) { + // We default to false because it's more common + LOG_WARNING(Render_OpenGL, "GL_ARB_seamless_cubemap_per_texture is required"); + } +} - glBindBuffer(GL_PIXEL_PACK_BUFFER, copy_pbo_handle); +Framebuffer::Framebuffer(TextureCacheRuntime& runtime, std::span<ImageView*, NUM_RT> color_buffers, + ImageView* depth_buffer, const VideoCommon::RenderTargets& key) { + // Bind to READ_FRAMEBUFFER to stop Nvidia's driver from creating an EXT_framebuffer instead of + // a core framebuffer. EXT framebuffer attachments have to match in size and can be shared + // across contexts. yuzu doesn't share framebuffers across contexts and we need attachments with + // mismatching size, this is why core framebuffers are preferred. + GLuint handle; + glGenFramebuffers(1, &handle); + glBindFramebuffer(GL_READ_FRAMEBUFFER, handle); + + GLsizei num_buffers = 0; + std::array<GLenum, NUM_RT> gl_draw_buffers; + gl_draw_buffers.fill(GL_NONE); + + for (size_t index = 0; index < color_buffers.size(); ++index) { + const ImageView* const image_view = color_buffers[index]; + if (!image_view) { + continue; + } + buffer_bits |= GL_COLOR_BUFFER_BIT; + gl_draw_buffers[index] = GL_COLOR_ATTACHMENT0 + key.draw_buffers[index]; + num_buffers = static_cast<GLsizei>(index + 1); - if (src_surface->IsCompressed()) { - glGetCompressedTextureImage(src_surface->GetTexture(), 0, static_cast<GLsizei>(source_size), - nullptr); - } else { - glGetTextureImage(src_surface->GetTexture(), 0, source_format.format, source_format.type, - static_cast<GLsizei>(source_size), nullptr); + const GLenum attachment = static_cast<GLenum>(GL_COLOR_ATTACHMENT0 + index); + AttachTexture(handle, attachment, image_view); } - glBindBuffer(GL_PIXEL_PACK_BUFFER, 0); - glBindBuffer(GL_PIXEL_UNPACK_BUFFER, copy_pbo_handle); + if (const ImageView* const image_view = depth_buffer; image_view) { + if (GetFormatType(image_view->format) == SurfaceType::DepthStencil) { + buffer_bits |= GL_DEPTH_BUFFER_BIT | GL_STENCIL_BUFFER_BIT; + } else { + buffer_bits |= GL_DEPTH_BUFFER_BIT; + } + const GLenum attachment = AttachmentType(image_view->format); + AttachTexture(handle, attachment, image_view); + } - const GLsizei width = static_cast<GLsizei>(dst_params.width); - const GLsizei height = static_cast<GLsizei>(dst_params.height); - const GLsizei depth = static_cast<GLsizei>(dst_params.depth); - if (dst_surface->IsCompressed()) { - LOG_CRITICAL(HW_GPU, "Compressed buffer copy is unimplemented!"); - UNREACHABLE(); + if (num_buffers > 1) { + glNamedFramebufferDrawBuffers(handle, num_buffers, gl_draw_buffers.data()); + } else if (num_buffers > 0) { + glNamedFramebufferDrawBuffer(handle, gl_draw_buffers[0]); } else { - switch (dst_params.target) { - case SurfaceTarget::Texture1D: - glTextureSubImage1D(dst_surface->GetTexture(), 0, 0, width, dest_format.format, - dest_format.type, nullptr); - break; - case SurfaceTarget::Texture2D: - glTextureSubImage2D(dst_surface->GetTexture(), 0, 0, 0, width, height, - dest_format.format, dest_format.type, nullptr); - break; - case SurfaceTarget::Texture3D: - case SurfaceTarget::Texture2DArray: - case SurfaceTarget::TextureCubeArray: - glTextureSubImage3D(dst_surface->GetTexture(), 0, 0, 0, 0, width, height, depth, - dest_format.format, dest_format.type, nullptr); - break; - case SurfaceTarget::TextureCubemap: - glTextureSubImage3D(dst_surface->GetTexture(), 0, 0, 0, 0, width, height, depth, - dest_format.format, dest_format.type, nullptr); - break; - default: - LOG_CRITICAL(Render_OpenGL, "Unimplemented surface target={}", - static_cast<u32>(dst_params.target)); - UNREACHABLE(); - } + glNamedFramebufferDrawBuffer(handle, GL_NONE); } - glBindBuffer(GL_PIXEL_UNPACK_BUFFER, 0); - glTextureBarrier(); -} + glNamedFramebufferParameteri(handle, GL_FRAMEBUFFER_DEFAULT_WIDTH, key.size.width); + glNamedFramebufferParameteri(handle, GL_FRAMEBUFFER_DEFAULT_HEIGHT, key.size.height); + // TODO + // glNamedFramebufferParameteri(handle, GL_FRAMEBUFFER_DEFAULT_LAYERS, ...); + // glNamedFramebufferParameteri(handle, GL_FRAMEBUFFER_DEFAULT_SAMPLES, ...); + // glNamedFramebufferParameteri(handle, GL_FRAMEBUFFER_DEFAULT_FIXED_SAMPLE_LOCATIONS, ...); -GLuint TextureCacheOpenGL::FetchPBO(std::size_t buffer_size) { - ASSERT_OR_EXECUTE(buffer_size > 0, { return 0; }); - const u32 l2 = Common::Log2Ceil64(static_cast<u64>(buffer_size)); - OGLBuffer& cp = copy_pbo_cache[l2]; - if (cp.handle == 0) { - const std::size_t ceil_size = 1ULL << l2; - cp.Create(); - cp.MakeStreamCopy(ceil_size); + if (runtime.device.HasDebuggingToolAttached()) { + const std::string name = VideoCommon::Name(key); + glObjectLabel(GL_FRAMEBUFFER, handle, static_cast<GLsizei>(name.size()), name.data()); } - return cp.handle; + framebuffer.handle = handle; } } // namespace OpenGL diff --git a/src/video_core/renderer_opengl/gl_texture_cache.h b/src/video_core/renderer_opengl/gl_texture_cache.h index 7787134fc..15b7c3676 100644 --- a/src/video_core/renderer_opengl/gl_texture_cache.h +++ b/src/video_core/renderer_opengl/gl_texture_cache.h @@ -4,156 +4,251 @@ #pragma once -#include <array> -#include <functional> #include <memory> -#include <unordered_map> -#include <utility> -#include <vector> +#include <span> #include <glad/glad.h> -#include "common/common_types.h" -#include "video_core/engines/shader_bytecode.h" -#include "video_core/renderer_opengl/gl_device.h" #include "video_core/renderer_opengl/gl_resource_manager.h" +#include "video_core/renderer_opengl/util_shaders.h" #include "video_core/texture_cache/texture_cache.h" namespace OpenGL { -using VideoCommon::SurfaceParams; -using VideoCommon::ViewParams; - -class CachedSurfaceView; -class CachedSurface; -class TextureCacheOpenGL; +class Device; +class ProgramManager; class StateTracker; -using Surface = std::shared_ptr<CachedSurface>; -using View = std::shared_ptr<CachedSurfaceView>; -using TextureCacheBase = VideoCommon::TextureCache<Surface, View>; +class Framebuffer; +class Image; +class ImageView; +class Sampler; -class CachedSurface final : public VideoCommon::SurfaceBase<View> { - friend CachedSurfaceView; +using VideoCommon::ImageId; +using VideoCommon::ImageViewId; +using VideoCommon::ImageViewType; +using VideoCommon::NUM_RT; +using VideoCommon::Offset2D; +using VideoCommon::RenderTargets; +class ImageBufferMap { public: - explicit CachedSurface(GPUVAddr gpu_addr, const SurfaceParams& params, bool is_astc_supported); - ~CachedSurface(); - - void UploadTexture(const std::vector<u8>& staging_buffer) override; - void DownloadTexture(std::vector<u8>& staging_buffer) override; + explicit ImageBufferMap(GLuint handle, u8* map, size_t size, OGLSync* sync); + ~ImageBufferMap(); - GLenum GetTarget() const { - return target; + GLuint Handle() const noexcept { + return handle; } - GLuint GetTexture() const { - return texture.handle; + std::span<u8> Span() const noexcept { + return span; } - bool IsCompressed() const { - return is_compressed; +private: + std::span<u8> span; + OGLSync* sync; + GLuint handle; +}; + +struct FormatProperties { + GLenum compatibility_class; + bool compatibility_by_size; + bool is_compressed; +}; + +class TextureCacheRuntime { + friend Framebuffer; + friend Image; + friend ImageView; + friend Sampler; + +public: + explicit TextureCacheRuntime(const Device& device, ProgramManager& program_manager, + StateTracker& state_tracker); + ~TextureCacheRuntime(); + + void Finish(); + + ImageBufferMap MapUploadBuffer(size_t size); + + ImageBufferMap MapDownloadBuffer(size_t size); + + void CopyImage(Image& dst, Image& src, std::span<const VideoCommon::ImageCopy> copies); + + void ConvertImage(Framebuffer* dst, ImageView& dst_view, ImageView& src_view) { + UNIMPLEMENTED(); } -protected: - void DecorateSurfaceName() override; + bool CanImageBeCopied(const Image& dst, const Image& src); + + void EmulateCopyImage(Image& dst, Image& src, std::span<const VideoCommon::ImageCopy> copies); + + void BlitFramebuffer(Framebuffer* dst, Framebuffer* src, + const std::array<Offset2D, 2>& dst_region, + const std::array<Offset2D, 2>& src_region, + Tegra::Engines::Fermi2D::Filter filter, + Tegra::Engines::Fermi2D::Operation operation); + + void AccelerateImageUpload(Image& image, const ImageBufferMap& map, size_t buffer_offset, + std::span<const VideoCommon::SwizzleParameters> swizzles); - View CreateView(const ViewParams& view_key) override; - View CreateViewInner(const ViewParams& view_key, bool is_proxy); + void InsertUploadMemoryBarrier(); + + FormatProperties FormatInfo(VideoCommon::ImageType type, GLenum internal_format) const; + + bool HasBrokenTextureViewFormats() const noexcept { + return has_broken_texture_view_formats; + } private: - void UploadTextureMipmap(u32 level, const std::vector<u8>& staging_buffer); + struct StagingBuffers { + explicit StagingBuffers(GLenum storage_flags_, GLenum map_flags_); + ~StagingBuffers(); - GLenum internal_format{}; - GLenum format{}; - GLenum type{}; - bool is_compressed{}; - GLenum target{}; - u32 view_count{}; + ImageBufferMap RequestMap(size_t requested_size, bool insert_fence); - OGLTexture texture; - OGLBuffer texture_buffer; + size_t RequestBuffer(size_t requested_size); + + std::optional<size_t> FindBuffer(size_t requested_size); + + std::vector<OGLSync> syncs; + std::vector<OGLBuffer> buffers; + std::vector<u8*> maps; + std::vector<size_t> sizes; + GLenum storage_flags; + GLenum map_flags; + }; + + const Device& device; + StateTracker& state_tracker; + UtilShaders util_shaders; + + std::array<std::unordered_map<GLenum, FormatProperties>, 3> format_properties; + bool has_broken_texture_view_formats = false; + + StagingBuffers upload_buffers{GL_MAP_WRITE_BIT, GL_MAP_WRITE_BIT | GL_MAP_FLUSH_EXPLICIT_BIT}; + StagingBuffers download_buffers{GL_MAP_READ_BIT, GL_MAP_READ_BIT}; + + OGLTexture null_image_1d_array; + OGLTexture null_image_cube_array; + OGLTexture null_image_3d; + OGLTexture null_image_rect; + OGLTextureView null_image_view_1d; + OGLTextureView null_image_view_2d; + OGLTextureView null_image_view_2d_array; + OGLTextureView null_image_view_cube; + + std::array<GLuint, VideoCommon::NUM_IMAGE_VIEW_TYPES> null_image_views; }; -class CachedSurfaceView final : public VideoCommon::ViewBase { +class Image : public VideoCommon::ImageBase { + friend ImageView; + public: - explicit CachedSurfaceView(CachedSurface& surface, const ViewParams& params, bool is_proxy); - ~CachedSurfaceView(); + explicit Image(TextureCacheRuntime&, const VideoCommon::ImageInfo& info, GPUVAddr gpu_addr, + VAddr cpu_addr); - /// @brief Attaches this texture view to the currently bound fb_target framebuffer - /// @param attachment Attachment to bind textures to - /// @param fb_target Framebuffer target to attach to (e.g. DRAW_FRAMEBUFFER) - void Attach(GLenum attachment, GLenum fb_target) const; + void UploadMemory(const ImageBufferMap& map, size_t buffer_offset, + std::span<const VideoCommon::BufferImageCopy> copies); - GLuint GetTexture(Tegra::Texture::SwizzleSource x_source, - Tegra::Texture::SwizzleSource y_source, - Tegra::Texture::SwizzleSource z_source, - Tegra::Texture::SwizzleSource w_source); + void UploadMemory(const ImageBufferMap& map, size_t buffer_offset, + std::span<const VideoCommon::BufferCopy> copies); - void DecorateViewName(GPUVAddr gpu_addr, const std::string& prefix); + void DownloadMemory(ImageBufferMap& map, size_t buffer_offset, + std::span<const VideoCommon::BufferImageCopy> copies); - void MarkAsModified(u64 tick) { - surface.MarkAsModified(true, tick); + GLuint Handle() const noexcept { + return texture.handle; } - GLuint GetTexture() const { - if (is_proxy) { - return surface.GetTexture(); - } - return main_view.handle; +private: + void CopyBufferToImage(const VideoCommon::BufferImageCopy& copy, size_t buffer_offset); + + void CopyImageToBuffer(const VideoCommon::BufferImageCopy& copy, size_t buffer_offset); + + OGLTexture texture; + OGLTextureView store_view; + OGLBuffer buffer; + GLenum gl_internal_format = GL_NONE; + GLenum gl_format = GL_NONE; + GLenum gl_type = GL_NONE; +}; + +class ImageView : public VideoCommon::ImageViewBase { + friend Image; + +public: + explicit ImageView(TextureCacheRuntime&, const VideoCommon::ImageViewInfo&, ImageId, Image&); + explicit ImageView(TextureCacheRuntime&, const VideoCommon::NullImageParams&); + + [[nodiscard]] GLuint Handle(ImageViewType query_type) const noexcept { + return views[static_cast<size_t>(query_type)]; } - GLenum GetFormat() const { - return format; + [[nodiscard]] GLuint DefaultHandle() const noexcept { + return default_handle; } - const SurfaceParams& GetSurfaceParams() const { - return surface.GetSurfaceParams(); + [[nodiscard]] GLenum Format() const noexcept { + return internal_format; } private: - OGLTextureView CreateTextureView() const; + void SetupView(const Device& device, Image& image, ImageViewType view_type, GLuint handle, + const VideoCommon::ImageViewInfo& info, + VideoCommon::SubresourceRange view_range); + + std::array<GLuint, VideoCommon::NUM_IMAGE_VIEW_TYPES> views{}; + std::vector<OGLTextureView> stored_views; + GLuint default_handle = 0; + GLenum internal_format = GL_NONE; +}; + +class ImageAlloc : public VideoCommon::ImageAllocBase {}; - CachedSurface& surface; - const GLenum format; - const GLenum target; - const bool is_proxy; +class Sampler { +public: + explicit Sampler(TextureCacheRuntime&, const Tegra::Texture::TSCEntry&); - std::unordered_map<u32, OGLTextureView> view_cache; - OGLTextureView main_view; + GLuint Handle() const noexcept { + return sampler.handle; + } - // Use an invalid default so it always fails the comparison test - u32 current_swizzle = 0xffffffff; - GLuint current_view = 0; +private: + OGLSampler sampler; }; -class TextureCacheOpenGL final : public TextureCacheBase { +class Framebuffer { public: - explicit TextureCacheOpenGL(VideoCore::RasterizerInterface& rasterizer, - Tegra::Engines::Maxwell3D& maxwell3d, - Tegra::MemoryManager& gpu_memory, const Device& device, - StateTracker& state_tracker); - ~TextureCacheOpenGL(); - -protected: - Surface CreateSurface(GPUVAddr gpu_addr, const SurfaceParams& params) override; - - void ImageCopy(Surface& src_surface, Surface& dst_surface, - const VideoCommon::CopyParams& copy_params) override; + explicit Framebuffer(TextureCacheRuntime&, std::span<ImageView*, NUM_RT> color_buffers, + ImageView* depth_buffer, const VideoCommon::RenderTargets& key); - void ImageBlit(View& src_view, View& dst_view, - const Tegra::Engines::Fermi2D::Config& copy_config) override; + [[nodiscard]] GLuint Handle() const noexcept { + return framebuffer.handle; + } - void BufferCopy(Surface& src_surface, Surface& dst_surface) override; + [[nodiscard]] GLbitfield BufferBits() const noexcept { + return buffer_bits; + } private: - GLuint FetchPBO(std::size_t buffer_size); - - StateTracker& state_tracker; + OGLFramebuffer framebuffer; + GLbitfield buffer_bits = GL_NONE; +}; - OGLFramebuffer src_framebuffer; - OGLFramebuffer dst_framebuffer; - std::unordered_map<u32, OGLBuffer> copy_pbo_cache; +struct TextureCacheParams { + static constexpr bool ENABLE_VALIDATION = true; + static constexpr bool FRAMEBUFFER_BLITS = true; + static constexpr bool HAS_EMULATED_COPIES = true; + + using Runtime = OpenGL::TextureCacheRuntime; + using Image = OpenGL::Image; + using ImageAlloc = OpenGL::ImageAlloc; + using ImageView = OpenGL::ImageView; + using Sampler = OpenGL::Sampler; + using Framebuffer = OpenGL::Framebuffer; }; +using TextureCache = VideoCommon::TextureCache<TextureCacheParams>; + } // namespace OpenGL diff --git a/src/video_core/renderer_opengl/maxwell_to_gl.h b/src/video_core/renderer_opengl/maxwell_to_gl.h index a8be2aa37..cbccfdeb4 100644 --- a/src/video_core/renderer_opengl/maxwell_to_gl.h +++ b/src/video_core/renderer_opengl/maxwell_to_gl.h @@ -107,7 +107,7 @@ inline GLenum IndexFormat(Maxwell::IndexFormat index_format) { case Maxwell::IndexFormat::UnsignedInt: return GL_UNSIGNED_INT; } - UNREACHABLE_MSG("Invalid index_format={}", static_cast<u32>(index_format)); + UNREACHABLE_MSG("Invalid index_format={}", index_format); return {}; } @@ -144,7 +144,7 @@ inline GLenum PrimitiveTopology(Maxwell::PrimitiveTopology topology) { case Maxwell::PrimitiveTopology::Patches: return GL_PATCHES; } - UNREACHABLE_MSG("Invalid topology={}", static_cast<int>(topology)); + UNREACHABLE_MSG("Invalid topology={}", topology); return GL_POINTS; } @@ -172,8 +172,8 @@ inline GLenum TextureFilterMode(Tegra::Texture::TextureFilter filter_mode, } break; } - UNREACHABLE_MSG("Invalid texture filter mode={} and mipmap filter mode={}", - static_cast<u32>(filter_mode), static_cast<u32>(mipmap_filter_mode)); + UNREACHABLE_MSG("Invalid texture filter mode={} and mipmap filter mode={}", filter_mode, + mipmap_filter_mode); return GL_NEAREST; } @@ -204,7 +204,7 @@ inline GLenum WrapMode(Tegra::Texture::WrapMode wrap_mode) { return GL_MIRROR_CLAMP_TO_EDGE; } } - UNIMPLEMENTED_MSG("Unimplemented texture wrap mode={}", static_cast<u32>(wrap_mode)); + UNIMPLEMENTED_MSG("Unimplemented texture wrap mode={}", wrap_mode); return GL_REPEAT; } @@ -227,7 +227,7 @@ inline GLenum DepthCompareFunc(Tegra::Texture::DepthCompareFunc func) { case Tegra::Texture::DepthCompareFunc::Always: return GL_ALWAYS; } - UNIMPLEMENTED_MSG("Unimplemented texture depth compare function={}", static_cast<u32>(func)); + UNIMPLEMENTED_MSG("Unimplemented texture depth compare function={}", func); return GL_GREATER; } @@ -249,7 +249,7 @@ inline GLenum BlendEquation(Maxwell::Blend::Equation equation) { case Maxwell::Blend::Equation::MaxGL: return GL_MAX; } - UNIMPLEMENTED_MSG("Unimplemented blend equation={}", static_cast<u32>(equation)); + UNIMPLEMENTED_MSG("Unimplemented blend equation={}", equation); return GL_FUNC_ADD; } @@ -313,7 +313,7 @@ inline GLenum BlendFunc(Maxwell::Blend::Factor factor) { case Maxwell::Blend::Factor::OneMinusConstantAlphaGL: return GL_ONE_MINUS_CONSTANT_ALPHA; } - UNIMPLEMENTED_MSG("Unimplemented blend factor={}", static_cast<u32>(factor)); + UNIMPLEMENTED_MSG("Unimplemented blend factor={}", factor); return GL_ZERO; } @@ -333,7 +333,7 @@ inline GLenum SwizzleSource(Tegra::Texture::SwizzleSource source) { case Tegra::Texture::SwizzleSource::OneFloat: return GL_ONE; } - UNIMPLEMENTED_MSG("Unimplemented swizzle source={}", static_cast<u32>(source)); + UNIMPLEMENTED_MSG("Unimplemented swizzle source={}", source); return GL_ZERO; } @@ -364,7 +364,7 @@ inline GLenum ComparisonOp(Maxwell::ComparisonOp comparison) { case Maxwell::ComparisonOp::AlwaysOld: return GL_ALWAYS; } - UNIMPLEMENTED_MSG("Unimplemented comparison op={}", static_cast<u32>(comparison)); + UNIMPLEMENTED_MSG("Unimplemented comparison op={}", comparison); return GL_ALWAYS; } @@ -395,7 +395,7 @@ inline GLenum StencilOp(Maxwell::StencilOp stencil) { case Maxwell::StencilOp::DecrWrapOGL: return GL_DECR_WRAP; } - UNIMPLEMENTED_MSG("Unimplemented stencil op={}", static_cast<u32>(stencil)); + UNIMPLEMENTED_MSG("Unimplemented stencil op={}", stencil); return GL_KEEP; } @@ -406,7 +406,7 @@ inline GLenum FrontFace(Maxwell::FrontFace front_face) { case Maxwell::FrontFace::CounterClockWise: return GL_CCW; } - UNIMPLEMENTED_MSG("Unimplemented front face cull={}", static_cast<u32>(front_face)); + UNIMPLEMENTED_MSG("Unimplemented front face cull={}", front_face); return GL_CCW; } @@ -419,7 +419,7 @@ inline GLenum CullFace(Maxwell::CullFace cull_face) { case Maxwell::CullFace::FrontAndBack: return GL_FRONT_AND_BACK; } - UNIMPLEMENTED_MSG("Unimplemented cull face={}", static_cast<u32>(cull_face)); + UNIMPLEMENTED_MSG("Unimplemented cull face={}", cull_face); return GL_BACK; } @@ -458,7 +458,7 @@ inline GLenum LogicOp(Maxwell::LogicOperation operation) { case Maxwell::LogicOperation::Set: return GL_SET; } - UNIMPLEMENTED_MSG("Unimplemented logic operation={}", static_cast<u32>(operation)); + UNIMPLEMENTED_MSG("Unimplemented logic operation={}", operation); return GL_COPY; } @@ -471,10 +471,23 @@ inline GLenum PolygonMode(Maxwell::PolygonMode polygon_mode) { case Maxwell::PolygonMode::Fill: return GL_FILL; } - UNREACHABLE_MSG("Invalid polygon mode={}", static_cast<int>(polygon_mode)); + UNREACHABLE_MSG("Invalid polygon mode={}", polygon_mode); return GL_FILL; } +inline GLenum ReductionFilter(Tegra::Texture::SamplerReduction filter) { + switch (filter) { + case Tegra::Texture::SamplerReduction::WeightedAverage: + return GL_WEIGHTED_AVERAGE_ARB; + case Tegra::Texture::SamplerReduction::Min: + return GL_MIN; + case Tegra::Texture::SamplerReduction::Max: + return GL_MAX; + } + UNREACHABLE_MSG("Invalid reduction filter={}", static_cast<int>(filter)); + return GL_WEIGHTED_AVERAGE_ARB; +} + inline GLenum ViewportSwizzle(Maxwell::ViewportSwizzle swizzle) { // Enumeration order matches register order. We can convert it arithmetically. return GL_VIEWPORT_SWIZZLE_POSITIVE_X_NV + static_cast<GLenum>(swizzle); diff --git a/src/video_core/renderer_opengl/renderer_opengl.cpp b/src/video_core/renderer_opengl/renderer_opengl.cpp index 2ccca1993..dd77a543c 100644 --- a/src/video_core/renderer_opengl/renderer_opengl.cpp +++ b/src/video_core/renderer_opengl/renderer_opengl.cpp @@ -23,10 +23,10 @@ #include "core/telemetry_session.h" #include "video_core/host_shaders/opengl_present_frag.h" #include "video_core/host_shaders/opengl_present_vert.h" -#include "video_core/morton.h" #include "video_core/renderer_opengl/gl_rasterizer.h" #include "video_core/renderer_opengl/gl_shader_manager.h" #include "video_core/renderer_opengl/renderer_opengl.h" +#include "video_core/textures/decoders.h" namespace OpenGL { @@ -130,8 +130,8 @@ void APIENTRY DebugHandler(GLenum source, GLenum type, GLuint id, GLenum severit RendererOpenGL::RendererOpenGL(Core::TelemetrySession& telemetry_session_, Core::Frontend::EmuWindow& emu_window_, Core::Memory::Memory& cpu_memory_, Tegra::GPU& gpu_, - std::unique_ptr<Core::Frontend::GraphicsContext> context) - : RendererBase{emu_window_, std::move(context)}, telemetry_session{telemetry_session_}, + std::unique_ptr<Core::Frontend::GraphicsContext> context_) + : RendererBase{emu_window_, std::move(context_)}, telemetry_session{telemetry_session_}, emu_window{emu_window_}, cpu_memory{cpu_memory_}, gpu{gpu_}, program_manager{device} {} RendererOpenGL::~RendererOpenGL() = default; @@ -140,19 +140,18 @@ void RendererOpenGL::SwapBuffers(const Tegra::FramebufferConfig* framebuffer) { if (!framebuffer) { return; } - PrepareRendertarget(framebuffer); RenderScreenshot(); - glBindFramebuffer(GL_DRAW_FRAMEBUFFER, 0); + state_tracker.BindFramebuffer(0); DrawScreen(emu_window.GetFramebufferLayout()); ++m_current_frame; rasterizer->TickFrame(); - render_window.PollEvents(); context->SwapBuffers(); + render_window.OnFrameDisplayed(); } void RendererOpenGL::PrepareRendertarget(const Tegra::FramebufferConfig* framebuffer) { @@ -187,19 +186,20 @@ void RendererOpenGL::LoadFBToScreenInfo(const Tegra::FramebufferConfig& framebuf // Reset the screen info's display texture to its own permanent texture screen_info.display_texture = screen_info.texture.resource.handle; - const auto pixel_format{ - VideoCore::Surface::PixelFormatFromGPUPixelFormat(framebuffer.pixel_format)}; - const u32 bytes_per_pixel{VideoCore::Surface::GetBytesPerPixel(pixel_format)}; - const u64 size_in_bytes{framebuffer.stride * framebuffer.height * bytes_per_pixel}; - u8* const host_ptr{cpu_memory.GetPointer(framebuffer_addr)}; - rasterizer->FlushRegion(ToCacheAddr(host_ptr), size_in_bytes); - // TODO(Rodrigo): Read this from HLE constexpr u32 block_height_log2 = 4; - VideoCore::MortonSwizzle(VideoCore::MortonSwizzleMode::MortonToLinear, pixel_format, - framebuffer.stride, block_height_log2, framebuffer.height, 0, 1, 1, - gl_framebuffer_data.data(), host_ptr); - + const auto pixel_format{ + VideoCore::Surface::PixelFormatFromGPUPixelFormat(framebuffer.pixel_format)}; + const u32 bytes_per_pixel{VideoCore::Surface::BytesPerBlock(pixel_format)}; + const u64 size_in_bytes{Tegra::Texture::CalculateSize( + true, bytes_per_pixel, framebuffer.stride, framebuffer.height, 1, block_height_log2, 0)}; + const u8* const host_ptr{cpu_memory.GetPointer(framebuffer_addr)}; + const std::span<const u8> input_data(host_ptr, size_in_bytes); + Tegra::Texture::UnswizzleTexture(gl_framebuffer_data, input_data, bytes_per_pixel, + framebuffer.width, framebuffer.height, 1, block_height_log2, + 0); + + glBindBuffer(GL_PIXEL_UNPACK_BUFFER, 0); glPixelStorei(GL_UNPACK_ROW_LENGTH, static_cast<GLint>(framebuffer.stride)); // Update existing texture @@ -238,6 +238,10 @@ void RendererOpenGL::InitOpenGLObjects() { glUseProgramStages(pipeline.handle, GL_VERTEX_SHADER_BIT, vertex_program.handle); glUseProgramStages(pipeline.handle, GL_FRAGMENT_SHADER_BIT, fragment_program.handle); + // Generate presentation sampler + present_sampler.Create(); + glSamplerParameteri(present_sampler.handle, GL_TEXTURE_MIN_FILTER, GL_LINEAR); + // Generate VBO handle for drawing vertex_buffer.Create(); @@ -255,6 +259,11 @@ void RendererOpenGL::InitOpenGLObjects() { // Clear screen to black LoadColorToActiveGLTexture(0, 0, 0, 0, screen_info.texture); + // Enable seamless cubemaps when per texture parameters are not available + if (!GLAD_GL_ARB_seamless_cubemap_per_texture && !GLAD_GL_AMD_seamless_cubemap_per_texture) { + glEnable(GL_TEXTURE_CUBE_MAP_SEAMLESS); + } + // Enable unified vertex attributes and query vertex buffer address when the driver supports it if (device.HasVertexBufferUnifiedMemory()) { glEnableClientState(GL_VERTEX_ATTRIB_ARRAY_UNIFIED_NV); @@ -275,9 +284,9 @@ void RendererOpenGL::AddTelemetryFields() { LOG_INFO(Render_OpenGL, "GL_RENDERER: {}", gpu_model); constexpr auto user_system = Common::Telemetry::FieldType::UserSystem; - telemetry_session.AddField(user_system, "GPU_Vendor", gpu_vendor); - telemetry_session.AddField(user_system, "GPU_Model", gpu_model); - telemetry_session.AddField(user_system, "GPU_OpenGL_Version", gl_version); + telemetry_session.AddField(user_system, "GPU_Vendor", std::string(gpu_vendor)); + telemetry_session.AddField(user_system, "GPU_Model", std::string(gpu_model)); + telemetry_session.AddField(user_system, "GPU_OpenGL_Version", std::string(gl_version)); } void RendererOpenGL::CreateRasterizer() { @@ -296,7 +305,7 @@ void RendererOpenGL::ConfigureFramebufferTexture(TextureInfo& texture, const auto pixel_format{ VideoCore::Surface::PixelFormatFromGPUPixelFormat(framebuffer.pixel_format)}; - const u32 bytes_per_pixel{VideoCore::Surface::GetBytesPerPixel(pixel_format)}; + const u32 bytes_per_pixel{VideoCore::Surface::BytesPerBlock(pixel_format)}; gl_framebuffer_data.resize(texture.width * texture.height * bytes_per_pixel); GLint internal_format; @@ -315,8 +324,8 @@ void RendererOpenGL::ConfigureFramebufferTexture(TextureInfo& texture, internal_format = GL_RGBA8; texture.gl_format = GL_RGBA; texture.gl_type = GL_UNSIGNED_INT_8_8_8_8_REV; - UNIMPLEMENTED_MSG("Unknown framebuffer pixel format: {}", - static_cast<u32>(framebuffer.pixel_format)); + // UNIMPLEMENTED_MSG("Unknown framebuffer pixel format: {}", + // static_cast<u32>(framebuffer.pixel_format)); } texture.resource.Release(); @@ -348,7 +357,7 @@ void RendererOpenGL::DrawScreen(const Layout::FramebufferLayout& layout) { } else { // Other transformations are unsupported LOG_CRITICAL(Render_OpenGL, "Unsupported framebuffer_transform_flags={}", - static_cast<u32>(framebuffer_transform_flags)); + framebuffer_transform_flags); UNIMPLEMENTED(); } } @@ -382,7 +391,7 @@ void RendererOpenGL::DrawScreen(const Layout::FramebufferLayout& layout) { state_tracker.NotifyPolygonModes(); state_tracker.NotifyViewport0(); state_tracker.NotifyScissor0(); - state_tracker.NotifyColorMask0(); + state_tracker.NotifyColorMask(0); state_tracker.NotifyBlend0(); state_tracker.NotifyFramebuffer(); state_tracker.NotifyFrontFace(); @@ -440,7 +449,7 @@ void RendererOpenGL::DrawScreen(const Layout::FramebufferLayout& layout) { } glBindTextureUnit(0, screen_info.display_texture); - glBindSampler(0, 0); + glBindSampler(0, present_sampler.handle); glClear(GL_COLOR_BUFFER_BIT); glDrawArrays(GL_TRIANGLE_STRIP, 0, 4); @@ -473,6 +482,8 @@ void RendererOpenGL::RenderScreenshot() { DrawScreen(layout); + glBindBuffer(GL_PIXEL_PACK_BUFFER, 0); + glPixelStorei(GL_PACK_ROW_LENGTH, 0); glReadPixels(0, 0, layout.width, layout.height, GL_BGRA, GL_UNSIGNED_INT_8_8_8_8_REV, renderer_settings.screenshot_bits); diff --git a/src/video_core/renderer_opengl/renderer_opengl.h b/src/video_core/renderer_opengl/renderer_opengl.h index 9ef181f95..44e109794 100644 --- a/src/video_core/renderer_opengl/renderer_opengl.h +++ b/src/video_core/renderer_opengl/renderer_opengl.h @@ -57,10 +57,10 @@ struct ScreenInfo { class RendererOpenGL final : public VideoCore::RendererBase { public: - explicit RendererOpenGL(Core::TelemetrySession& telemetry_session, - Core::Frontend::EmuWindow& emu_window, Core::Memory::Memory& cpu_memory, - Tegra::GPU& gpu, - std::unique_ptr<Core::Frontend::GraphicsContext> context); + explicit RendererOpenGL(Core::TelemetrySession& telemetry_session_, + Core::Frontend::EmuWindow& emu_window_, + Core::Memory::Memory& cpu_memory_, Tegra::GPU& gpu_, + std::unique_ptr<Core::Frontend::GraphicsContext> context_); ~RendererOpenGL() override; bool Init() override; @@ -102,6 +102,7 @@ private: StateTracker state_tracker{gpu}; // OpenGL object IDs + OGLSampler present_sampler; OGLBuffer vertex_buffer; OGLProgram vertex_program; OGLProgram fragment_program; diff --git a/src/video_core/renderer_opengl/util_shaders.cpp b/src/video_core/renderer_opengl/util_shaders.cpp new file mode 100644 index 000000000..eb849cbf2 --- /dev/null +++ b/src/video_core/renderer_opengl/util_shaders.cpp @@ -0,0 +1,224 @@ +// Copyright 2020 yuzu Emulator Project +// Licensed under GPLv2 or any later version +// Refer to the license.txt file included. + +#include <bit> +#include <span> +#include <string_view> + +#include <glad/glad.h> + +#include "common/assert.h" +#include "common/common_types.h" +#include "common/div_ceil.h" +#include "video_core/host_shaders/block_linear_unswizzle_2d_comp.h" +#include "video_core/host_shaders/block_linear_unswizzle_3d_comp.h" +#include "video_core/host_shaders/opengl_copy_bc4_comp.h" +#include "video_core/host_shaders/pitch_unswizzle_comp.h" +#include "video_core/renderer_opengl/gl_resource_manager.h" +#include "video_core/renderer_opengl/gl_shader_manager.h" +#include "video_core/renderer_opengl/gl_texture_cache.h" +#include "video_core/renderer_opengl/util_shaders.h" +#include "video_core/surface.h" +#include "video_core/texture_cache/accelerated_swizzle.h" +#include "video_core/texture_cache/types.h" +#include "video_core/texture_cache/util.h" +#include "video_core/textures/decoders.h" + +namespace OpenGL { + +using namespace HostShaders; + +using VideoCommon::Extent3D; +using VideoCommon::ImageCopy; +using VideoCommon::ImageType; +using VideoCommon::SwizzleParameters; +using VideoCommon::Accelerated::MakeBlockLinearSwizzle2DParams; +using VideoCommon::Accelerated::MakeBlockLinearSwizzle3DParams; +using VideoCore::Surface::BytesPerBlock; + +namespace { + +OGLProgram MakeProgram(std::string_view source) { + OGLShader shader; + shader.Create(source, GL_COMPUTE_SHADER); + + OGLProgram program; + program.Create(true, false, shader.handle); + return program; +} + +} // Anonymous namespace + +UtilShaders::UtilShaders(ProgramManager& program_manager_) + : program_manager{program_manager_}, + block_linear_unswizzle_2d_program(MakeProgram(BLOCK_LINEAR_UNSWIZZLE_2D_COMP)), + block_linear_unswizzle_3d_program(MakeProgram(BLOCK_LINEAR_UNSWIZZLE_3D_COMP)), + pitch_unswizzle_program(MakeProgram(PITCH_UNSWIZZLE_COMP)), + copy_bc4_program(MakeProgram(OPENGL_COPY_BC4_COMP)) { + const auto swizzle_table = Tegra::Texture::MakeSwizzleTable(); + swizzle_table_buffer.Create(); + glNamedBufferStorage(swizzle_table_buffer.handle, sizeof(swizzle_table), &swizzle_table, 0); +} + +UtilShaders::~UtilShaders() = default; + +void UtilShaders::BlockLinearUpload2D(Image& image, const ImageBufferMap& map, size_t buffer_offset, + std::span<const SwizzleParameters> swizzles) { + static constexpr Extent3D WORKGROUP_SIZE{32, 32, 1}; + static constexpr GLuint BINDING_SWIZZLE_BUFFER = 0; + static constexpr GLuint BINDING_INPUT_BUFFER = 1; + static constexpr GLuint BINDING_OUTPUT_IMAGE = 0; + + program_manager.BindHostCompute(block_linear_unswizzle_2d_program.handle); + glFlushMappedNamedBufferRange(map.Handle(), buffer_offset, image.guest_size_bytes); + glBindBufferBase(GL_SHADER_STORAGE_BUFFER, BINDING_SWIZZLE_BUFFER, swizzle_table_buffer.handle); + + const GLenum store_format = StoreFormat(BytesPerBlock(image.info.format)); + for (const SwizzleParameters& swizzle : swizzles) { + const Extent3D num_tiles = swizzle.num_tiles; + const size_t input_offset = swizzle.buffer_offset + buffer_offset; + + const u32 num_dispatches_x = Common::DivCeil(num_tiles.width, WORKGROUP_SIZE.width); + const u32 num_dispatches_y = Common::DivCeil(num_tiles.height, WORKGROUP_SIZE.height); + + const auto params = MakeBlockLinearSwizzle2DParams(swizzle, image.info); + glUniform3uiv(0, 1, params.origin.data()); + glUniform3iv(1, 1, params.destination.data()); + glUniform1ui(2, params.bytes_per_block_log2); + glUniform1ui(3, params.layer_stride); + glUniform1ui(4, params.block_size); + glUniform1ui(5, params.x_shift); + glUniform1ui(6, params.block_height); + glUniform1ui(7, params.block_height_mask); + glBindBufferRange(GL_SHADER_STORAGE_BUFFER, BINDING_INPUT_BUFFER, map.Handle(), + input_offset, image.guest_size_bytes - swizzle.buffer_offset); + glBindImageTexture(BINDING_OUTPUT_IMAGE, image.Handle(), swizzle.level, GL_TRUE, 0, + GL_WRITE_ONLY, store_format); + glDispatchCompute(num_dispatches_x, num_dispatches_y, image.info.resources.layers); + } + program_manager.RestoreGuestCompute(); +} + +void UtilShaders::BlockLinearUpload3D(Image& image, const ImageBufferMap& map, size_t buffer_offset, + std::span<const SwizzleParameters> swizzles) { + static constexpr Extent3D WORKGROUP_SIZE{16, 8, 8}; + + static constexpr GLuint BINDING_SWIZZLE_BUFFER = 0; + static constexpr GLuint BINDING_INPUT_BUFFER = 1; + static constexpr GLuint BINDING_OUTPUT_IMAGE = 0; + + glFlushMappedNamedBufferRange(map.Handle(), buffer_offset, image.guest_size_bytes); + program_manager.BindHostCompute(block_linear_unswizzle_3d_program.handle); + glBindBufferBase(GL_SHADER_STORAGE_BUFFER, BINDING_SWIZZLE_BUFFER, swizzle_table_buffer.handle); + + const GLenum store_format = StoreFormat(BytesPerBlock(image.info.format)); + for (const SwizzleParameters& swizzle : swizzles) { + const Extent3D num_tiles = swizzle.num_tiles; + const size_t input_offset = swizzle.buffer_offset + buffer_offset; + + const u32 num_dispatches_x = Common::DivCeil(num_tiles.width, WORKGROUP_SIZE.width); + const u32 num_dispatches_y = Common::DivCeil(num_tiles.height, WORKGROUP_SIZE.height); + const u32 num_dispatches_z = Common::DivCeil(num_tiles.depth, WORKGROUP_SIZE.depth); + + const auto params = MakeBlockLinearSwizzle3DParams(swizzle, image.info); + glUniform3uiv(0, 1, params.origin.data()); + glUniform3iv(1, 1, params.destination.data()); + glUniform1ui(2, params.bytes_per_block_log2); + glUniform1ui(3, params.slice_size); + glUniform1ui(4, params.block_size); + glUniform1ui(5, params.x_shift); + glUniform1ui(6, params.block_height); + glUniform1ui(7, params.block_height_mask); + glUniform1ui(8, params.block_depth); + glUniform1ui(9, params.block_depth_mask); + glBindBufferRange(GL_SHADER_STORAGE_BUFFER, BINDING_INPUT_BUFFER, map.Handle(), + input_offset, image.guest_size_bytes - swizzle.buffer_offset); + glBindImageTexture(BINDING_OUTPUT_IMAGE, image.Handle(), swizzle.level, GL_TRUE, 0, + GL_WRITE_ONLY, store_format); + glDispatchCompute(num_dispatches_x, num_dispatches_y, num_dispatches_z); + } + program_manager.RestoreGuestCompute(); +} + +void UtilShaders::PitchUpload(Image& image, const ImageBufferMap& map, size_t buffer_offset, + std::span<const SwizzleParameters> swizzles) { + static constexpr Extent3D WORKGROUP_SIZE{32, 32, 1}; + static constexpr GLuint BINDING_INPUT_BUFFER = 0; + static constexpr GLuint BINDING_OUTPUT_IMAGE = 0; + static constexpr GLuint LOC_ORIGIN = 0; + static constexpr GLuint LOC_DESTINATION = 1; + static constexpr GLuint LOC_BYTES_PER_BLOCK = 2; + static constexpr GLuint LOC_PITCH = 3; + + const u32 bytes_per_block = BytesPerBlock(image.info.format); + const GLenum format = StoreFormat(bytes_per_block); + const u32 pitch = image.info.pitch; + + UNIMPLEMENTED_IF_MSG(!std::has_single_bit(bytes_per_block), + "Non-power of two images are not implemented"); + + program_manager.BindHostCompute(pitch_unswizzle_program.handle); + glFlushMappedNamedBufferRange(map.Handle(), buffer_offset, image.guest_size_bytes); + glUniform2ui(LOC_ORIGIN, 0, 0); + glUniform2i(LOC_DESTINATION, 0, 0); + glUniform1ui(LOC_BYTES_PER_BLOCK, bytes_per_block); + glUniform1ui(LOC_PITCH, pitch); + glBindImageTexture(BINDING_OUTPUT_IMAGE, image.Handle(), 0, GL_FALSE, 0, GL_WRITE_ONLY, format); + for (const SwizzleParameters& swizzle : swizzles) { + const Extent3D num_tiles = swizzle.num_tiles; + const size_t input_offset = swizzle.buffer_offset + buffer_offset; + + const u32 num_dispatches_x = Common::DivCeil(num_tiles.width, WORKGROUP_SIZE.width); + const u32 num_dispatches_y = Common::DivCeil(num_tiles.height, WORKGROUP_SIZE.height); + + glBindBufferRange(GL_SHADER_STORAGE_BUFFER, BINDING_INPUT_BUFFER, map.Handle(), + input_offset, image.guest_size_bytes - swizzle.buffer_offset); + glDispatchCompute(num_dispatches_x, num_dispatches_y, 1); + } + program_manager.RestoreGuestCompute(); +} + +void UtilShaders::CopyBC4(Image& dst_image, Image& src_image, std::span<const ImageCopy> copies) { + static constexpr GLuint BINDING_INPUT_IMAGE = 0; + static constexpr GLuint BINDING_OUTPUT_IMAGE = 1; + static constexpr GLuint LOC_SRC_OFFSET = 0; + static constexpr GLuint LOC_DST_OFFSET = 1; + + program_manager.BindHostCompute(copy_bc4_program.handle); + + for (const ImageCopy& copy : copies) { + ASSERT(copy.src_subresource.base_layer == 0); + ASSERT(copy.src_subresource.num_layers == 1); + ASSERT(copy.dst_subresource.base_layer == 0); + ASSERT(copy.dst_subresource.num_layers == 1); + + glUniform3ui(LOC_SRC_OFFSET, copy.src_offset.x, copy.src_offset.y, copy.src_offset.z); + glUniform3ui(LOC_DST_OFFSET, copy.dst_offset.x, copy.dst_offset.y, copy.dst_offset.z); + glBindImageTexture(BINDING_INPUT_IMAGE, src_image.Handle(), copy.src_subresource.base_level, + GL_FALSE, 0, GL_READ_ONLY, GL_RG32UI); + glBindImageTexture(BINDING_OUTPUT_IMAGE, dst_image.Handle(), + copy.dst_subresource.base_level, GL_FALSE, 0, GL_WRITE_ONLY, GL_RGBA8UI); + glDispatchCompute(copy.extent.width, copy.extent.height, copy.extent.depth); + } + program_manager.RestoreGuestCompute(); +} + +GLenum StoreFormat(u32 bytes_per_block) { + switch (bytes_per_block) { + case 1: + return GL_R8UI; + case 2: + return GL_R16UI; + case 4: + return GL_R32UI; + case 8: + return GL_RG32UI; + case 16: + return GL_RGBA32UI; + } + UNREACHABLE(); + return GL_R8UI; +} + +} // namespace OpenGL diff --git a/src/video_core/renderer_opengl/util_shaders.h b/src/video_core/renderer_opengl/util_shaders.h new file mode 100644 index 000000000..359997255 --- /dev/null +++ b/src/video_core/renderer_opengl/util_shaders.h @@ -0,0 +1,51 @@ +// Copyright 2020 yuzu Emulator Project +// Licensed under GPLv2 or any later version +// Refer to the license.txt file included. + +#pragma once + +#include <span> + +#include <glad/glad.h> + +#include "common/common_types.h" +#include "video_core/renderer_opengl/gl_resource_manager.h" +#include "video_core/texture_cache/types.h" + +namespace OpenGL { + +class Image; +class ImageBufferMap; +class ProgramManager; + +class UtilShaders { +public: + explicit UtilShaders(ProgramManager& program_manager); + ~UtilShaders(); + + void BlockLinearUpload2D(Image& image, const ImageBufferMap& map, size_t buffer_offset, + std::span<const VideoCommon::SwizzleParameters> swizzles); + + void BlockLinearUpload3D(Image& image, const ImageBufferMap& map, size_t buffer_offset, + std::span<const VideoCommon::SwizzleParameters> swizzles); + + void PitchUpload(Image& image, const ImageBufferMap& map, size_t buffer_offset, + std::span<const VideoCommon::SwizzleParameters> swizzles); + + void CopyBC4(Image& dst_image, Image& src_image, + std::span<const VideoCommon::ImageCopy> copies); + +private: + ProgramManager& program_manager; + + OGLBuffer swizzle_table_buffer; + + OGLProgram block_linear_unswizzle_2d_program; + OGLProgram block_linear_unswizzle_3d_program; + OGLProgram pitch_unswizzle_program; + OGLProgram copy_bc4_program; +}; + +GLenum StoreFormat(u32 bytes_per_block); + +} // namespace OpenGL diff --git a/src/video_core/renderer_opengl/utils.cpp b/src/video_core/renderer_opengl/utils.cpp deleted file mode 100644 index 6d7bb16b2..000000000 --- a/src/video_core/renderer_opengl/utils.cpp +++ /dev/null @@ -1,42 +0,0 @@ -// Copyright 2014 Citra Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. - -#include <string> -#include <vector> - -#include <fmt/format.h> -#include <glad/glad.h> - -#include "common/common_types.h" -#include "video_core/renderer_opengl/gl_state_tracker.h" -#include "video_core/renderer_opengl/utils.h" - -namespace OpenGL { - -void LabelGLObject(GLenum identifier, GLuint handle, VAddr addr, std::string_view extra_info) { - if (!GLAD_GL_KHR_debug) { - // We don't need to throw an error as this is just for debugging - return; - } - - std::string object_label; - if (extra_info.empty()) { - switch (identifier) { - case GL_TEXTURE: - object_label = fmt::format("Texture@0x{:016X}", addr); - break; - case GL_PROGRAM: - object_label = fmt::format("Shader@0x{:016X}", addr); - break; - default: - object_label = fmt::format("Object(0x{:X})@0x{:016X}", identifier, addr); - break; - } - } else { - object_label = fmt::format("{}@0x{:016X}", extra_info, addr); - } - glObjectLabel(identifier, handle, -1, static_cast<const GLchar*>(object_label.c_str())); -} - -} // namespace OpenGL diff --git a/src/video_core/renderer_opengl/utils.h b/src/video_core/renderer_opengl/utils.h deleted file mode 100644 index 9c09ee12c..000000000 --- a/src/video_core/renderer_opengl/utils.h +++ /dev/null @@ -1,16 +0,0 @@ -// Copyright 2014 Citra Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. - -#pragma once - -#include <string_view> -#include <vector> -#include <glad/glad.h> -#include "common/common_types.h" - -namespace OpenGL { - -void LabelGLObject(GLenum identifier, GLuint handle, VAddr addr, std::string_view extra_info = {}); - -} // namespace OpenGL |