diff options
Diffstat (limited to 'src/video_core/renderer_opengl')
5 files changed, 277 insertions, 48 deletions
diff --git a/src/video_core/renderer_opengl/gl_rasterizer.cpp b/src/video_core/renderer_opengl/gl_rasterizer.cpp index 6a17bed72..a85f730a8 100644 --- a/src/video_core/renderer_opengl/gl_rasterizer.cpp +++ b/src/video_core/renderer_opengl/gl_rasterizer.cpp @@ -1340,7 +1340,9 @@ void RasterizerOpenGL::SyncPolygonOffset() { state.polygon_offset.fill_enable = regs.polygon_offset_fill_enable != 0; state.polygon_offset.line_enable = regs.polygon_offset_line_enable != 0; state.polygon_offset.point_enable = regs.polygon_offset_point_enable != 0; - state.polygon_offset.units = regs.polygon_offset_units; + + // Hardware divides polygon offset units by two + state.polygon_offset.units = regs.polygon_offset_units / 2.0f; state.polygon_offset.factor = regs.polygon_offset_factor; state.polygon_offset.clamp = regs.polygon_offset_clamp; diff --git a/src/video_core/renderer_opengl/gl_shader_decompiler.cpp b/src/video_core/renderer_opengl/gl_shader_decompiler.cpp index 8fa9e6534..6a610a3bc 100644 --- a/src/video_core/renderer_opengl/gl_shader_decompiler.cpp +++ b/src/video_core/renderer_opengl/gl_shader_decompiler.cpp @@ -19,6 +19,7 @@ #include "video_core/renderer_opengl/gl_device.h" #include "video_core/renderer_opengl/gl_rasterizer.h" #include "video_core/renderer_opengl/gl_shader_decompiler.h" +#include "video_core/shader/ast.h" #include "video_core/shader/node.h" #include "video_core/shader/shader_ir.h" @@ -334,39 +335,24 @@ constexpr bool IsVertexShader(ProgramType stage) { return stage == ProgramType::VertexA || stage == ProgramType::VertexB; } +class ASTDecompiler; +class ExprDecompiler; + class GLSLDecompiler final { public: explicit GLSLDecompiler(const Device& device, const ShaderIR& ir, ProgramType stage, std::string suffix) : device{device}, ir{ir}, stage{stage}, suffix{suffix}, header{ir.GetHeader()} {} - void Decompile() { - DeclareVertex(); - DeclareGeometry(); - DeclareRegisters(); - DeclarePredicates(); - DeclareLocalMemory(); - DeclareSharedMemory(); - DeclareInternalFlags(); - DeclareInputAttributes(); - DeclareOutputAttributes(); - DeclareConstantBuffers(); - DeclareGlobalMemory(); - DeclareSamplers(); - DeclarePhysicalAttributeReader(); - DeclareImages(); - - code.AddLine("void execute_{}() {{", suffix); - ++code.scope; - + void DecompileBranchMode() { // VM's program counter const auto first_address = ir.GetBasicBlocks().begin()->first; code.AddLine("uint jmp_to = {}U;", first_address); // TODO(Subv): Figure out the actual depth of the flow stack, for now it seems // unlikely that shaders will use 20 nested SSYs and PBKs. + constexpr u32 FLOW_STACK_SIZE = 20; if (!ir.IsFlowStackDisabled()) { - constexpr u32 FLOW_STACK_SIZE = 20; for (const auto stack : std::array{MetaStackClass::Ssy, MetaStackClass::Pbk}) { code.AddLine("uint {}[{}];", FlowStackName(stack), FLOW_STACK_SIZE); code.AddLine("uint {} = 0U;", FlowStackTopName(stack)); @@ -392,10 +378,37 @@ public: code.AddLine("default: return;"); code.AddLine("}}"); - for (std::size_t i = 0; i < 2; ++i) { - --code.scope; - code.AddLine("}}"); + --code.scope; + code.AddLine("}}"); + } + + void DecompileAST(); + + void Decompile() { + DeclareVertex(); + DeclareGeometry(); + DeclareRegisters(); + DeclarePredicates(); + DeclareLocalMemory(); + DeclareInternalFlags(); + DeclareInputAttributes(); + DeclareOutputAttributes(); + DeclareConstantBuffers(); + DeclareGlobalMemory(); + DeclareSamplers(); + DeclarePhysicalAttributeReader(); + + code.AddLine("void execute_{}() {{", suffix); + ++code.scope; + + if (ir.IsDecompiled()) { + DecompileAST(); + } else { + DecompileBranchMode(); } + + --code.scope; + code.AddLine("}}"); } std::string GetResult() { @@ -424,6 +437,9 @@ public: } private: + friend class ASTDecompiler; + friend class ExprDecompiler; + void DeclareVertex() { if (!IsVertexShader(stage)) return; @@ -1821,10 +1837,9 @@ private: return {}; } - Expression Exit(Operation operation) { + void PreExit() { if (stage != ProgramType::Fragment) { - code.AddLine("return;"); - return {}; + return; } const auto& used_registers = ir.GetRegisters(); const auto SafeGetRegister = [&](u32 reg) -> Expression { @@ -1856,7 +1871,10 @@ private: // already contains one past the last color register. code.AddLine("gl_FragDepth = {};", SafeGetRegister(current_reg + 1).AsFloat()); } + } + Expression Exit(Operation operation) { + PreExit(); code.AddLine("return;"); return {}; } @@ -2253,6 +2271,208 @@ private: ShaderWriter code; }; +static constexpr std::string_view flow_var = "flow_var_"; + +std::string GetFlowVariable(u32 i) { + return fmt::format("{}{}", flow_var, i); +} + +class ExprDecompiler { +public: + explicit ExprDecompiler(GLSLDecompiler& decomp) : decomp{decomp} {} + + void operator()(VideoCommon::Shader::ExprAnd& expr) { + inner += "( "; + std::visit(*this, *expr.operand1); + inner += " && "; + std::visit(*this, *expr.operand2); + inner += ')'; + } + + void operator()(VideoCommon::Shader::ExprOr& expr) { + inner += "( "; + std::visit(*this, *expr.operand1); + inner += " || "; + std::visit(*this, *expr.operand2); + inner += ')'; + } + + void operator()(VideoCommon::Shader::ExprNot& expr) { + inner += '!'; + std::visit(*this, *expr.operand1); + } + + void operator()(VideoCommon::Shader::ExprPredicate& expr) { + const auto pred = static_cast<Tegra::Shader::Pred>(expr.predicate); + inner += decomp.GetPredicate(pred); + } + + void operator()(VideoCommon::Shader::ExprCondCode& expr) { + const Node cc = decomp.ir.GetConditionCode(expr.cc); + std::string target; + + if (const auto pred = std::get_if<PredicateNode>(&*cc)) { + const auto index = pred->GetIndex(); + switch (index) { + case Tegra::Shader::Pred::NeverExecute: + target = "false"; + case Tegra::Shader::Pred::UnusedIndex: + target = "true"; + default: + target = decomp.GetPredicate(index); + } + } else if (const auto flag = std::get_if<InternalFlagNode>(&*cc)) { + target = decomp.GetInternalFlag(flag->GetFlag()); + } else { + UNREACHABLE(); + } + inner += target; + } + + void operator()(VideoCommon::Shader::ExprVar& expr) { + inner += GetFlowVariable(expr.var_index); + } + + void operator()(VideoCommon::Shader::ExprBoolean& expr) { + inner += expr.value ? "true" : "false"; + } + + std::string& GetResult() { + return inner; + } + +private: + std::string inner; + GLSLDecompiler& decomp; +}; + +class ASTDecompiler { +public: + explicit ASTDecompiler(GLSLDecompiler& decomp) : decomp{decomp} {} + + void operator()(VideoCommon::Shader::ASTProgram& ast) { + ASTNode current = ast.nodes.GetFirst(); + while (current) { + Visit(current); + current = current->GetNext(); + } + } + + void operator()(VideoCommon::Shader::ASTIfThen& ast) { + ExprDecompiler expr_parser{decomp}; + std::visit(expr_parser, *ast.condition); + decomp.code.AddLine("if ({}) {{", expr_parser.GetResult()); + decomp.code.scope++; + ASTNode current = ast.nodes.GetFirst(); + while (current) { + Visit(current); + current = current->GetNext(); + } + decomp.code.scope--; + decomp.code.AddLine("}}"); + } + + void operator()(VideoCommon::Shader::ASTIfElse& ast) { + decomp.code.AddLine("else {{"); + decomp.code.scope++; + ASTNode current = ast.nodes.GetFirst(); + while (current) { + Visit(current); + current = current->GetNext(); + } + decomp.code.scope--; + decomp.code.AddLine("}}"); + } + + void operator()(VideoCommon::Shader::ASTBlockEncoded& ast) { + UNREACHABLE(); + } + + void operator()(VideoCommon::Shader::ASTBlockDecoded& ast) { + decomp.VisitBlock(ast.nodes); + } + + void operator()(VideoCommon::Shader::ASTVarSet& ast) { + ExprDecompiler expr_parser{decomp}; + std::visit(expr_parser, *ast.condition); + decomp.code.AddLine("{} = {};", GetFlowVariable(ast.index), expr_parser.GetResult()); + } + + void operator()(VideoCommon::Shader::ASTLabel& ast) { + decomp.code.AddLine("// Label_{}:", ast.index); + } + + void operator()(VideoCommon::Shader::ASTGoto& ast) { + UNREACHABLE(); + } + + void operator()(VideoCommon::Shader::ASTDoWhile& ast) { + ExprDecompiler expr_parser{decomp}; + std::visit(expr_parser, *ast.condition); + decomp.code.AddLine("do {{"); + decomp.code.scope++; + ASTNode current = ast.nodes.GetFirst(); + while (current) { + Visit(current); + current = current->GetNext(); + } + decomp.code.scope--; + decomp.code.AddLine("}} while({});", expr_parser.GetResult()); + } + + void operator()(VideoCommon::Shader::ASTReturn& ast) { + const bool is_true = VideoCommon::Shader::ExprIsTrue(ast.condition); + if (!is_true) { + ExprDecompiler expr_parser{decomp}; + std::visit(expr_parser, *ast.condition); + decomp.code.AddLine("if ({}) {{", expr_parser.GetResult()); + decomp.code.scope++; + } + if (ast.kills) { + decomp.code.AddLine("discard;"); + } else { + decomp.PreExit(); + decomp.code.AddLine("return;"); + } + if (!is_true) { + decomp.code.scope--; + decomp.code.AddLine("}}"); + } + } + + void operator()(VideoCommon::Shader::ASTBreak& ast) { + const bool is_true = VideoCommon::Shader::ExprIsTrue(ast.condition); + if (!is_true) { + ExprDecompiler expr_parser{decomp}; + std::visit(expr_parser, *ast.condition); + decomp.code.AddLine("if ({}) {{", expr_parser.GetResult()); + decomp.code.scope++; + } + decomp.code.AddLine("break;"); + if (!is_true) { + decomp.code.scope--; + decomp.code.AddLine("}}"); + } + } + + void Visit(VideoCommon::Shader::ASTNode& node) { + std::visit(*this, *node->GetInnerData()); + } + +private: + GLSLDecompiler& decomp; +}; + +void GLSLDecompiler::DecompileAST() { + const u32 num_flow_variables = ir.GetASTNumVariables(); + for (u32 i = 0; i < num_flow_variables; i++) { + code.AddLine("bool {} = false;", GetFlowVariable(i)); + } + ASTDecompiler decompiler{*this}; + VideoCommon::Shader::ASTNode program = ir.GetASTProgram(); + decompiler.Visit(program); +} + } // Anonymous namespace std::string GetCommonDeclarations() { 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 6a7012b54..74cc33476 100644 --- a/src/video_core/renderer_opengl/gl_shader_disk_cache.cpp +++ b/src/video_core/renderer_opengl/gl_shader_disk_cache.cpp @@ -112,14 +112,15 @@ std::optional<std::pair<std::vector<ShaderDiskCacheRaw>, std::vector<ShaderDiskC ShaderDiskCacheOpenGL::LoadTransferable() { // Skip games without title id const bool has_title_id = system.CurrentProcess()->GetTitleID() != 0; - if (!Settings::values.use_disk_shader_cache || !has_title_id) + if (!Settings::values.use_disk_shader_cache || !has_title_id) { return {}; - tried_to_load = true; + } FileUtil::IOFile file(GetTransferablePath(), "rb"); if (!file.IsOpen()) { LOG_INFO(Render_OpenGL, "No transferable shader cache found for game with title id={}", GetTitleID()); + is_usable = true; return {}; } @@ -135,6 +136,7 @@ ShaderDiskCacheOpenGL::LoadTransferable() { LOG_INFO(Render_OpenGL, "Transferable shader cache is old - removing"); file.Close(); InvalidateTransferable(); + is_usable = true; return {}; } if (version > NativeVersion) { @@ -180,13 +182,15 @@ ShaderDiskCacheOpenGL::LoadTransferable() { } } - return {{raws, usages}}; + is_usable = true; + return {{std::move(raws), std::move(usages)}}; } std::pair<std::unordered_map<u64, ShaderDiskCacheDecompiled>, ShaderDumpsMap> ShaderDiskCacheOpenGL::LoadPrecompiled() { - if (!IsUsable()) + if (!is_usable) { return {}; + } FileUtil::IOFile file(GetPrecompiledPath(), "rb"); if (!file.IsOpen()) { @@ -479,8 +483,9 @@ void ShaderDiskCacheOpenGL::InvalidatePrecompiled() { } void ShaderDiskCacheOpenGL::SaveRaw(const ShaderDiskCacheRaw& entry) { - if (!IsUsable()) + if (!is_usable) { return; + } const u64 id = entry.GetUniqueIdentifier(); if (transferable.find(id) != transferable.end()) { @@ -501,8 +506,9 @@ void ShaderDiskCacheOpenGL::SaveRaw(const ShaderDiskCacheRaw& entry) { } void ShaderDiskCacheOpenGL::SaveUsage(const ShaderDiskCacheUsage& usage) { - if (!IsUsable()) + if (!is_usable) { return; + } const auto it = transferable.find(usage.unique_identifier); ASSERT_MSG(it != transferable.end(), "Saving shader usage without storing raw previously"); @@ -528,8 +534,9 @@ void ShaderDiskCacheOpenGL::SaveUsage(const ShaderDiskCacheUsage& usage) { void ShaderDiskCacheOpenGL::SaveDecompiled(u64 unique_identifier, const std::string& code, const GLShader::ShaderEntries& entries) { - if (!IsUsable()) + if (!is_usable) { return; + } if (precompiled_cache_virtual_file.GetSize() == 0) { SavePrecompiledHeaderToVirtualPrecompiledCache(); @@ -543,8 +550,9 @@ void ShaderDiskCacheOpenGL::SaveDecompiled(u64 unique_identifier, const std::str } void ShaderDiskCacheOpenGL::SaveDump(const ShaderDiskCacheUsage& usage, GLuint program) { - if (!IsUsable()) + if (!is_usable) { return; + } GLint binary_length{}; glGetProgramiv(program, GL_PROGRAM_BINARY_LENGTH, &binary_length); @@ -565,10 +573,6 @@ void ShaderDiskCacheOpenGL::SaveDump(const ShaderDiskCacheUsage& usage, GLuint p } } -bool ShaderDiskCacheOpenGL::IsUsable() const { - return tried_to_load && Settings::values.use_disk_shader_cache; -} - FileUtil::IOFile ShaderDiskCacheOpenGL::AppendTransferableFile() const { if (!EnsureDirectories()) return {}; diff --git a/src/video_core/renderer_opengl/gl_shader_disk_cache.h b/src/video_core/renderer_opengl/gl_shader_disk_cache.h index cc8bbd61e..9595bd71b 100644 --- a/src/video_core/renderer_opengl/gl_shader_disk_cache.h +++ b/src/video_core/renderer_opengl/gl_shader_disk_cache.h @@ -224,9 +224,6 @@ private: bool SaveDecompiledFile(u64 unique_identifier, const std::string& code, const GLShader::ShaderEntries& entries); - /// Returns if the cache can be used - bool IsUsable() const; - /// Opens current game's transferable file and write it's header if it doesn't exist FileUtil::IOFile AppendTransferableFile() const; @@ -297,7 +294,7 @@ private: std::unordered_map<u64, std::unordered_set<ShaderDiskCacheUsage>> transferable; // The cache has been loaded at boot - bool tried_to_load{}; + bool is_usable{}; }; } // namespace OpenGL diff --git a/src/video_core/renderer_opengl/gl_shader_gen.cpp b/src/video_core/renderer_opengl/gl_shader_gen.cpp index 3a8d9e1da..b5a43e79e 100644 --- a/src/video_core/renderer_opengl/gl_shader_gen.cpp +++ b/src/video_core/renderer_opengl/gl_shader_gen.cpp @@ -11,12 +11,16 @@ namespace OpenGL::GLShader { using Tegra::Engines::Maxwell3D; +using VideoCommon::Shader::CompileDepth; +using VideoCommon::Shader::CompilerSettings; using VideoCommon::Shader::ProgramCode; using VideoCommon::Shader::ShaderIR; static constexpr u32 PROGRAM_OFFSET = 10; static constexpr u32 COMPUTE_OFFSET = 0; +static constexpr CompilerSettings settings{CompileDepth::NoFlowStack, true}; + ProgramResult GenerateVertexShader(const Device& device, const ShaderSetup& setup) { const std::string id = fmt::format("{:016x}", setup.program.unique_identifier); @@ -31,13 +35,14 @@ layout (std140, binding = EMULATION_UBO_BINDING) uniform vs_config { )"; - const ShaderIR program_ir(setup.program.code, PROGRAM_OFFSET, setup.program.size_a); + const ShaderIR program_ir(setup.program.code, PROGRAM_OFFSET, setup.program.size_a, settings); const auto stage = setup.IsDualProgram() ? ProgramType::VertexA : ProgramType::VertexB; ProgramResult program = Decompile(device, program_ir, stage, "vertex"); out += program.first; if (setup.IsDualProgram()) { - const ShaderIR program_ir_b(setup.program.code_b, PROGRAM_OFFSET, setup.program.size_b); + const ShaderIR program_ir_b(setup.program.code_b, PROGRAM_OFFSET, setup.program.size_b, + settings); ProgramResult program_b = Decompile(device, program_ir_b, ProgramType::VertexB, "vertex_b"); out += program_b.first; } @@ -80,7 +85,7 @@ layout (std140, binding = EMULATION_UBO_BINDING) uniform gs_config { )"; - const ShaderIR program_ir(setup.program.code, PROGRAM_OFFSET, setup.program.size_a); + const ShaderIR program_ir(setup.program.code, PROGRAM_OFFSET, setup.program.size_a, settings); ProgramResult program = Decompile(device, program_ir, ProgramType::Geometry, "geometry"); out += program.first; @@ -114,7 +119,8 @@ layout (std140, binding = EMULATION_UBO_BINDING) uniform fs_config { }; )"; - const ShaderIR program_ir(setup.program.code, PROGRAM_OFFSET, setup.program.size_a); + + const ShaderIR program_ir(setup.program.code, PROGRAM_OFFSET, setup.program.size_a, settings); ProgramResult program = Decompile(device, program_ir, ProgramType::Fragment, "fragment"); out += program.first; @@ -133,7 +139,7 @@ ProgramResult GenerateComputeShader(const Device& device, const ShaderSetup& set std::string out = "// Shader Unique Id: CS" + id + "\n\n"; out += GetCommonDeclarations(); - const ShaderIR program_ir(setup.program.code, COMPUTE_OFFSET, setup.program.size_a); + const ShaderIR program_ir(setup.program.code, COMPUTE_OFFSET, setup.program.size_a, settings); ProgramResult program = Decompile(device, program_ir, ProgramType::Compute, "compute"); out += program.first; |