gl_shader_gen: Implement dual vertex shader mode.

- When VertexA shader stage is enabled, we combine with VertexB program to make a single Vertex Shader stage.

src/video_core/renderer_opengl/gl_rasterizer.cpp

@@ -181,6 +181,19 @@ std::pair<u8*, GLintptr> RasterizerOpenGL::SetupVertexArrays(u8* array_ptr,
     return {array_ptr, buffer_offset};
 }
 
+static GLShader::ProgramCode GetShaderProgramCode(Maxwell::ShaderProgram program) {
+    auto& gpu = Core::System().GetInstance().GPU().Maxwell3D();
+
+    // Fetch program code from memory
+    GLShader::ProgramCode program_code;
+    auto& shader_config = gpu.regs.shader_config[static_cast<size_t>(program)];
+    const u64 gpu_address{gpu.regs.code_address.CodeAddress() + shader_config.offset};
+    const boost::optional<VAddr> cpu_address{gpu.memory_manager.GpuToCpuAddress(gpu_address)};
+    Memory::ReadBlock(*cpu_address, program_code.data(), program_code.size() * sizeof(u64));
+
+    return program_code;
+}
+
 void RasterizerOpenGL::SetupShaders(u8* buffer_ptr, GLintptr buffer_offset) {
     // Helper function for uploading uniform data
     const auto copy_buffer = [&](GLuint handle, GLintptr offset, GLsizeiptr size) {
@@ -193,18 +206,17 @@ void RasterizerOpenGL::SetupShaders(u8* buffer_ptr, GLintptr buffer_offset) {
     };
 
     auto& gpu = Core::System().GetInstance().GPU().Maxwell3D();
-    ASSERT_MSG(!gpu.regs.shader_config[0].enable, "VertexA is unsupported!");
 
     // Next available bindpoints to use when uploading the const buffers and textures to the GLSL
     // shaders. The constbuffer bindpoint starts after the shader stage configuration bind points.
     u32 current_constbuffer_bindpoint = uniform_buffers.size();
     u32 current_texture_bindpoint = 0;
 
-    for (unsigned index = 1; index < Maxwell::MaxShaderProgram; ++index) {
+    for (size_t index = 0; index < Maxwell::MaxShaderProgram; ++index) {
         auto& shader_config = gpu.regs.shader_config[index];
         const Maxwell::ShaderProgram program{static_cast<Maxwell::ShaderProgram>(index)};
 
-        const auto& stage = index - 1; // Stage indices are 0 - 5
+        const size_t stage{index == 0 ? 0 : index - 1}; // Stage indices are 0 - 5
 
         const bool is_enabled = gpu.IsShaderStageEnabled(static_cast<Maxwell::ShaderStage>(stage));
 
@@ -228,16 +240,21 @@ void RasterizerOpenGL::SetupShaders(u8* buffer_ptr, GLintptr buffer_offset) {
         buffer_ptr += sizeof(GLShader::MaxwellUniformData);
         buffer_offset += sizeof(GLShader::MaxwellUniformData);
 
-        // Fetch program code from memory
-        GLShader::ProgramCode program_code;
-        const u64 gpu_address{gpu.regs.code_address.CodeAddress() + shader_config.offset};
-        const boost::optional<VAddr> cpu_address{gpu.memory_manager.GpuToCpuAddress(gpu_address)};
-        Memory::ReadBlock(*cpu_address, program_code.data(), program_code.size() * sizeof(u64));
-        GLShader::ShaderSetup setup{std::move(program_code)};
-
+        GLShader::ShaderSetup setup{GetShaderProgramCode(program)};
         GLShader::ShaderEntries shader_resources;
 
         switch (program) {
+        case Maxwell::ShaderProgram::VertexA: {
+            // VertexB is always enabled, so when VertexA is enabled, we have two vertex shaders.
+            // Conventional HW does not support this, so we combine VertexA and VertexB into one
+            // stage here.
+            setup.SetProgramB(GetShaderProgramCode(Maxwell::ShaderProgram::VertexB));
+            GLShader::MaxwellVSConfig vs_config{setup};
+            shader_resources =
+                shader_program_manager->UseProgrammableVertexShader(vs_config, setup);
+            break;
+        }
+
         case Maxwell::ShaderProgram::VertexB: {
             GLShader::MaxwellVSConfig vs_config{setup};
             shader_resources =
@@ -268,6 +285,12 @@ void RasterizerOpenGL::SetupShaders(u8* buffer_ptr, GLintptr buffer_offset) {
         current_texture_bindpoint =
             SetupTextures(static_cast<Maxwell::ShaderStage>(stage), gl_stage_program,
                           current_texture_bindpoint, shader_resources.texture_samplers);
+
+        // When VertexA is enabled, we have dual vertex shaders
+        if (program == Maxwell::ShaderProgram::VertexA) {
+            // VertexB was combined with VertexA, so we skip the VertexB iteration
+            index++;
+        }
     }
 
     shader_program_manager->UseTrivialGeometryShader();

src/video_core/renderer_opengl/gl_shader_decompiler.cpp

@@ -42,11 +42,12 @@ enum class ExitMethod {
 struct Subroutine {
     /// Generates a name suitable for GLSL source code.
     std::string GetName() const {
-        return "sub_" + std::to_string(begin) + '_' + std::to_string(end);
+        return "sub_" + std::to_string(begin) + '_' + std::to_string(end) + '_' + suffix;
     }
 
     u32 begin;                 ///< Entry point of the subroutine.
     u32 end;                   ///< Return point of the subroutine.
+    const std::string& suffix; ///< Suffix of the shader, used to make a unique subroutine name
     ExitMethod exit_method;    ///< Exit method of the subroutine.
     std::set<u32> labels;      ///< Addresses refereced by JMP instructions.
 
@@ -58,11 +59,11 @@ struct Subroutine {
 /// Analyzes shader code and produces a set of subroutines.
 class ControlFlowAnalyzer {
 public:
-    ControlFlowAnalyzer(const ProgramCode& program_code, u32 main_offset)
+    ControlFlowAnalyzer(const ProgramCode& program_code, u32 main_offset, const std::string& suffix)
         : program_code(program_code) {
 
         // Recursively finds all subroutines.
-        const Subroutine& program_main = AddSubroutine(main_offset, PROGRAM_END);
+        const Subroutine& program_main = AddSubroutine(main_offset, PROGRAM_END, suffix);
         if (program_main.exit_method != ExitMethod::AlwaysEnd)
             throw DecompileFail("Program does not always end");
     }
@@ -77,12 +78,12 @@ private:
     std::map<std::pair<u32, u32>, ExitMethod> exit_method_map;
 
     /// Adds and analyzes a new subroutine if it is not added yet.
-    const Subroutine& AddSubroutine(u32 begin, u32 end) {
-        auto iter = subroutines.find(Subroutine{begin, end});
+    const Subroutine& AddSubroutine(u32 begin, u32 end, const std::string& suffix) {
+        auto iter = subroutines.find(Subroutine{begin, end, suffix});
         if (iter != subroutines.end())
             return *iter;
 
-        Subroutine subroutine{begin, end};
+        Subroutine subroutine{begin, end, suffix};
         subroutine.exit_method = Scan(begin, end, subroutine.labels);
         if (subroutine.exit_method == ExitMethod::Undetermined)
             throw DecompileFail("Recursive function detected");
@@ -191,7 +192,8 @@ public:
         UnsignedInteger,
     };
 
-    GLSLRegister(size_t index, ShaderWriter& shader) : index{index}, shader{shader} {}
+    GLSLRegister(size_t index, ShaderWriter& shader, const std::string& suffix)
+        : index{index}, shader{shader}, suffix{suffix} {}
 
     /// Gets the GLSL type string for a register
     static std::string GetTypeString(Type type) {
@@ -216,7 +218,7 @@ public:
     /// Returns a GLSL string representing the current state of the register
     const std::string GetActiveString() {
         declr_type.insert(active_type);
-        return GetPrefixString(active_type) + std::to_string(index);
+        return GetPrefixString(active_type) + std::to_string(index) + '_' + suffix;
     }
 
     /// Returns true if the active type is a float
@@ -251,6 +253,7 @@ private:
     ShaderWriter& shader;
     Type active_type{Type::Float};
     std::set<Type> declr_type;
+    const std::string& suffix;
 };
 
 /**
@@ -262,8 +265,8 @@ private:
 class GLSLRegisterManager {
 public:
     GLSLRegisterManager(ShaderWriter& shader, ShaderWriter& declarations,
-                        const Maxwell3D::Regs::ShaderStage& stage)
-        : shader{shader}, declarations{declarations}, stage{stage} {
+                        const Maxwell3D::Regs::ShaderStage& stage, const std::string& suffix)
+        : shader{shader}, declarations{declarations}, stage{stage}, suffix{suffix} {
         BuildRegisterList();
     }
 
@@ -430,12 +433,12 @@ public:
     }
 
     /// Add declarations for registers
-    void GenerateDeclarations() {
+    void GenerateDeclarations(const std::string& suffix) {
         for (const auto& reg : regs) {
             for (const auto& type : reg.DeclaredTypes()) {
                 declarations.AddLine(GLSLRegister::GetTypeString(type) + ' ' +
-                                     GLSLRegister::GetPrefixString(type) +
-                                     std::to_string(reg.GetIndex()) + " = 0;");
+                                     reg.GetPrefixString(type) + std::to_string(reg.GetIndex()) +
+                                     '_' + suffix + " = 0;");
             }
         }
         declarations.AddNewLine();
@@ -558,7 +561,7 @@ private:
     /// Build the GLSL register list.
     void BuildRegisterList() {
         for (size_t index = 0; index < Register::NumRegisters; ++index) {
-            regs.emplace_back(index, shader);
+            regs.emplace_back(index, shader, suffix);
         }
     }
 
@@ -620,16 +623,17 @@ private:
     std::array<ConstBufferEntry, Maxwell3D::Regs::MaxConstBuffers> declr_const_buffers;
     std::vector<SamplerEntry> used_samplers;
     const Maxwell3D::Regs::ShaderStage& stage;
+    const std::string& suffix;
 };
 
 class GLSLGenerator {
 public:
     GLSLGenerator(const std::set<Subroutine>& subroutines, const ProgramCode& program_code,
-                  u32 main_offset, Maxwell3D::Regs::ShaderStage stage)
+                  u32 main_offset, Maxwell3D::Regs::ShaderStage stage, const std::string& suffix)
         : subroutines(subroutines), program_code(program_code), main_offset(main_offset),
-          stage(stage) {
+          stage(stage), suffix(suffix) {
 
-        Generate();
+        Generate(suffix);
     }
 
     std::string GetShaderCode() {
@@ -644,7 +648,7 @@ public:
 private:
     /// Gets the Subroutine object corresponding to the specified address.
     const Subroutine& GetSubroutine(u32 begin, u32 end) const {
-        auto iter = subroutines.find(Subroutine{begin, end});
+        auto iter = subroutines.find(Subroutine{begin, end, suffix});
         ASSERT(iter != subroutines.end());
         return *iter;
     }
@@ -689,7 +693,7 @@ private:
         // Can't assign to the constant predicate.
         ASSERT(pred != static_cast<u64>(Pred::UnusedIndex));
 
-        std::string variable = 'p' + std::to_string(pred);
+        std::string variable = 'p' + std::to_string(pred) + '_' + suffix;
         shader.AddLine(variable + " = " + value + ';');
         declr_predicates.insert(std::move(variable));
     }
@@ -707,7 +711,7 @@ private:
         if (index == static_cast<u64>(Pred::UnusedIndex))
             variable = "true";
         else
-            variable = 'p' + std::to_string(index);
+            variable = 'p' + std::to_string(index) + '_' + suffix;
 
         if (negate) {
             return "!(" + variable + ')';
@@ -1728,7 +1732,7 @@ private:
         return program_counter;
     }
 
-    void Generate() {
+    void Generate(const std::string& suffix) {
         // Add declarations for all subroutines
         for (const auto& subroutine : subroutines) {
             shader.AddLine("bool " + subroutine.GetName() + "();");
@@ -1736,7 +1740,7 @@ private:
         shader.AddNewLine();
 
         // Add the main entry point
-        shader.AddLine("bool exec_shader() {");
+        shader.AddLine("bool exec_" + suffix + "() {");
         ++shader.scope;
         CallSubroutine(GetSubroutine(main_offset, PROGRAM_END));
         --shader.scope;
@@ -1799,7 +1803,7 @@ private:
 
     /// Add declarations for registers
     void GenerateDeclarations() {
-        regs.GenerateDeclarations();
+        regs.GenerateDeclarations(suffix);
 
         for (const auto& pred : declr_predicates) {
             declarations.AddLine("bool " + pred + " = false;");
@@ -1812,27 +1816,30 @@ private:
     const ProgramCode& program_code;
     const u32 main_offset;
     Maxwell3D::Regs::ShaderStage stage;
+    const std::string& suffix;
 
     ShaderWriter shader;
     ShaderWriter declarations;
-    GLSLRegisterManager regs{shader, declarations, stage};
+    GLSLRegisterManager regs{shader, declarations, stage, suffix};
 
     // Declarations
     std::set<std::string> declr_predicates;
 }; // namespace Decompiler
 
 std::string GetCommonDeclarations() {
-    std::string declarations = "bool exec_shader();\n";
+    std::string declarations;
     declarations += "#define MAX_CONSTBUFFER_ELEMENTS " +
                     std::to_string(RasterizerOpenGL::MaxConstbufferSize / (sizeof(GLvec4)));
+    declarations += '\n';
     return declarations;
 }
 
 boost::optional<ProgramResult> DecompileProgram(const ProgramCode& program_code, u32 main_offset,
-                                                Maxwell3D::Regs::ShaderStage stage) {
+                                                Maxwell3D::Regs::ShaderStage stage,
+                                                const std::string& suffix) {
     try {
-        auto subroutines = ControlFlowAnalyzer(program_code, main_offset).GetSubroutines();
-        GLSLGenerator generator(subroutines, program_code, main_offset, stage);
+        auto subroutines = ControlFlowAnalyzer(program_code, main_offset, suffix).GetSubroutines();
+        GLSLGenerator generator(subroutines, program_code, main_offset, stage, suffix);
         return ProgramResult{generator.GetShaderCode(), generator.GetEntries()};
     } catch (const DecompileFail& exception) {
         LOG_ERROR(HW_GPU, "Shader decompilation failed: {}", exception.what());

src/video_core/renderer_opengl/gl_shader_decompiler.h

@@ -20,7 +20,8 @@ using Tegra::Engines::Maxwell3D;
 std::string GetCommonDeclarations();
 
 boost::optional<ProgramResult> DecompileProgram(const ProgramCode& program_code, u32 main_offset,
-                                                Maxwell3D::Regs::ShaderStage stage);
+                                                Maxwell3D::Regs::ShaderStage stage,
+                                                const std::string& suffix);
 
 } // namespace Decompiler
 } // namespace GLShader

src/video_core/renderer_opengl/gl_shader_gen.cpp

@@ -17,10 +17,17 @@ ProgramResult GenerateVertexShader(const ShaderSetup& setup, const MaxwellVSConf
     std::string out = "#version 430 core\n";
     out += "#extension GL_ARB_separate_shader_objects : enable\n\n";
     out += Decompiler::GetCommonDeclarations();
+    out += "bool exec_vertex();\n";
 
-    ProgramResult program = Decompiler::DecompileProgram(setup.program_code, PROGRAM_OFFSET,
-                                                         Maxwell3D::Regs::ShaderStage::Vertex)
+    if (setup.IsDualProgram()) {
+        out += "bool exec_vertex_b();\n";
+    }
+
+    ProgramResult program =
+        Decompiler::DecompileProgram(setup.program.code, PROGRAM_OFFSET,
+                                     Maxwell3D::Regs::ShaderStage::Vertex, "vertex")
             .get_value_or({});
+
     out += R"(
 
 out gl_PerVertex {
@@ -34,7 +41,14 @@ layout (std140) uniform vs_config {
 };
 
 void main() {
-    exec_shader();
+    exec_vertex();
+)";
+
+    if (setup.IsDualProgram()) {
+        out += "    exec_vertex_b();";
+    }
+
+    out += R"(
 
     // Viewport can be flipped, which is unsupported by glViewport
     position.xy *= viewport_flip.xy;
@@ -44,8 +58,19 @@ void main() {
     // For now, this is here to bring order in lieu of proper emulation
     position.w = 1.0;
 }
+
 )";
+
     out += program.first;
+
+    if (setup.IsDualProgram()) {
+        ProgramResult program_b =
+            Decompiler::DecompileProgram(setup.program.code_b, PROGRAM_OFFSET,
+                                         Maxwell3D::Regs::ShaderStage::Vertex, "vertex_b")
+                .get_value_or({});
+        out += program_b.first;
+    }
+
     return {out, program.second};
 }
 
@@ -53,12 +78,13 @@ ProgramResult GenerateFragmentShader(const ShaderSetup& setup, const MaxwellFSCo
     std::string out = "#version 430 core\n";
     out += "#extension GL_ARB_separate_shader_objects : enable\n\n";
     out += Decompiler::GetCommonDeclarations();
+    out += "bool exec_fragment();\n";
 
-    ProgramResult program = Decompiler::DecompileProgram(setup.program_code, PROGRAM_OFFSET,
-                                                         Maxwell3D::Regs::ShaderStage::Fragment)
+    ProgramResult program =
+        Decompiler::DecompileProgram(setup.program.code, PROGRAM_OFFSET,
+                                     Maxwell3D::Regs::ShaderStage::Fragment, "fragment")
             .get_value_or({});
     out += R"(
-
 in vec4 position;
 out vec4 color;
 
@@ -67,7 +93,7 @@ layout (std140) uniform fs_config {
 };
 
 void main() {
-    exec_shader();
+    exec_fragment();
 }
 
 )";

src/video_core/renderer_opengl/gl_shader_gen.h

@@ -115,21 +115,48 @@ struct ShaderEntries {
 using ProgramResult = std::pair<std::string, ShaderEntries>;
 
 struct ShaderSetup {
-    ShaderSetup(ProgramCode&& program_code) : program_code(std::move(program_code)) {}
+    ShaderSetup(const ProgramCode& program_code) {
+        program.code = program_code;
+    }
+
+    struct {
+        ProgramCode code;
+        ProgramCode code_b; // Used for dual vertex shaders
+    } program;
 
-    ProgramCode program_code;
     bool program_code_hash_dirty = true;
 
     u64 GetProgramCodeHash() {
         if (program_code_hash_dirty) {
-            program_code_hash = Common::ComputeHash64(&program_code, sizeof(program_code));
+            program_code_hash = GetNewHash();
             program_code_hash_dirty = false;
         }
         return program_code_hash;
     }
 
+    /// Used in scenarios where we have a dual vertex shaders
+    void SetProgramB(const ProgramCode& program_b) {
+        program.code_b = program_b;
+        has_program_b = true;
+    }
+
+    bool IsDualProgram() const {
+        return has_program_b;
+    }
+
 private:
+    u64 GetNewHash() const {
+        if (has_program_b) {
+            // Compute hash over dual shader programs
+            return Common::ComputeHash64(&program, sizeof(program));
+        } else {
+            // Compute hash over a single shader program
+            return Common::ComputeHash64(&program.code, program.code.size());
+        }
+    }
+
     u64 program_code_hash{};
+    bool has_program_b{};
 };
 
 struct MaxwellShaderConfigCommon {