Merge pull request #4894 from wwylele/remove-gs

opengl: remove hw geometry shader related stuff

src/video_core/renderer_opengl/gl_rasterizer.cpp

@@ -104,8 +104,6 @@ RasterizerOpenGL::RasterizerOpenGL(Frontend::EmuWindow& window)
     glGetIntegerv(GL_UNIFORM_BUFFER_OFFSET_ALIGNMENT, &uniform_buffer_alignment);
     uniform_size_aligned_vs =
         Common::AlignUp<std::size_t>(sizeof(VSUniformData), uniform_buffer_alignment);
-    uniform_size_aligned_gs =
-        Common::AlignUp<std::size_t>(sizeof(GSUniformData), uniform_buffer_alignment);
     uniform_size_aligned_fs =
         Common::AlignUp<std::size_t>(sizeof(UniformData), uniform_buffer_alignment);
 
@@ -392,8 +390,7 @@ bool RasterizerOpenGL::SetupGeometryShader() {
         shader_program_manager->UseFixedGeometryShader(gs_config);
         return true;
     } else {
-        PicaGSConfig gs_config(regs, Pica::g_state.gs);
+        LOG_ERROR(Render_OpenGL, "Accelerate draw doesn't support geometry shader");
-        return shader_program_manager->UseProgrammableGeometryShader(gs_config, Pica::g_state.gs);
     }
 }
 
@@ -417,42 +414,24 @@ bool RasterizerOpenGL::AccelerateDrawBatch(bool is_indexed) {
     return Draw(true, is_indexed);
 }
 
-static GLenum GetCurrentPrimitiveMode(bool use_gs) {
+static GLenum GetCurrentPrimitiveMode() {
     const auto& regs = Pica::g_state.regs;
-    if (use_gs) {
+    switch (regs.pipeline.triangle_topology) {
-        switch ((regs.gs.max_input_attribute_index + 1) /
+    case Pica::PipelineRegs::TriangleTopology::Shader:
-                (regs.pipeline.vs_outmap_total_minus_1_a + 1)) {
+    case Pica::PipelineRegs::TriangleTopology::List:
-        case 1:
+        return GL_TRIANGLES;
-            return GL_POINTS;
+    case Pica::PipelineRegs::TriangleTopology::Fan:
-        case 2:
+        return GL_TRIANGLE_FAN;
-            return GL_LINES;
+    case Pica::PipelineRegs::TriangleTopology::Strip:
-        case 4:
+        return GL_TRIANGLE_STRIP;
-            return GL_LINES_ADJACENCY;
+    default:
-        case 3:
+        UNREACHABLE();
-            return GL_TRIANGLES;
-        case 6:
-            return GL_TRIANGLES_ADJACENCY;
-        default:
-            UNREACHABLE();
-        }
-    } else {
-        switch (regs.pipeline.triangle_topology) {
-        case Pica::PipelineRegs::TriangleTopology::Shader:
-        case Pica::PipelineRegs::TriangleTopology::List:
-            return GL_TRIANGLES;
-        case Pica::PipelineRegs::TriangleTopology::Fan:
-            return GL_TRIANGLE_FAN;
-        case Pica::PipelineRegs::TriangleTopology::Strip:
-            return GL_TRIANGLE_STRIP;
-        default:
-            UNREACHABLE();
-        }
     }
 }
 
-bool RasterizerOpenGL::AccelerateDrawBatchInternal(bool is_indexed, bool use_gs) {
+bool RasterizerOpenGL::AccelerateDrawBatchInternal(bool is_indexed) {
     const auto& regs = Pica::g_state.regs;
-    GLenum primitive_mode = GetCurrentPrimitiveMode(use_gs);
+    GLenum primitive_mode = GetCurrentPrimitiveMode();
 
     auto [vs_input_index_min, vs_input_index_max, vs_input_size] = AnalyzeVertexArray(is_indexed);
 
@@ -787,8 +766,7 @@ bool RasterizerOpenGL::Draw(bool accelerate, bool is_indexed) {
     SyncAndUploadLUTs();
 
     // Sync the uniform data
-    const bool use_gs = regs.pipeline.use_gs == Pica::PipelineRegs::UseGS::Yes;
+    UploadUniforms(accelerate);
-    UploadUniforms(accelerate, use_gs);
 
     // Viewport can have negative offsets or larger
     // dimensions than our framebuffer sub-rect.
@@ -804,7 +782,7 @@ bool RasterizerOpenGL::Draw(bool accelerate, bool is_indexed) {
     // Draw the vertex batch
     bool succeeded = true;
     if (accelerate) {
-        succeeded = AccelerateDrawBatchInternal(is_indexed, use_gs);
+        succeeded = AccelerateDrawBatchInternal(is_indexed);
     } else {
         state.draw.vertex_array = sw_vao.handle;
         state.draw.vertex_buffer = vertex_buffer.GetHandle();
@@ -2119,21 +2097,19 @@ void RasterizerOpenGL::SyncAndUploadLUTs() {
     texture_buffer.Unmap(bytes_used);
 }
 
-void RasterizerOpenGL::UploadUniforms(bool accelerate_draw, bool use_gs) {
+void RasterizerOpenGL::UploadUniforms(bool accelerate_draw) {
     // glBindBufferRange below also changes the generic buffer binding point, so we sync the state
     // first
     state.draw.uniform_buffer = uniform_buffer.GetHandle();
     state.Apply();
 
     bool sync_vs = accelerate_draw;
-    bool sync_gs = accelerate_draw && use_gs;
     bool sync_fs = uniform_block_data.dirty;
 
-    if (!sync_vs && !sync_gs && !sync_fs)
+    if (!sync_vs && !sync_fs)
         return;
 
-    std::size_t uniform_size =
+    std::size_t uniform_size = uniform_size_aligned_vs + uniform_size_aligned_fs;
-        uniform_size_aligned_vs + uniform_size_aligned_gs + uniform_size_aligned_fs;
     std::size_t used_bytes = 0;
     u8* uniforms;
     GLintptr offset;
@@ -2150,15 +2126,6 @@ void RasterizerOpenGL::UploadUniforms(bool accelerate_draw, bool use_gs) {
         used_bytes += uniform_size_aligned_vs;
     }
 
-    if (sync_gs) {
-        GSUniformData gs_uniforms;
-        gs_uniforms.uniforms.SetFromRegs(Pica::g_state.regs.gs, Pica::g_state.gs);
-        std::memcpy(uniforms + used_bytes, &gs_uniforms, sizeof(gs_uniforms));
-        glBindBufferRange(GL_UNIFORM_BUFFER, static_cast<GLuint>(UniformBindings::GS),
-                          uniform_buffer.GetHandle(), offset + used_bytes, sizeof(GSUniformData));
-        used_bytes += uniform_size_aligned_gs;
-    }
-
     if (sync_fs || invalidate) {
         std::memcpy(uniforms + used_bytes, &uniform_block_data.data, sizeof(UniformData));
         glBindBufferRange(GL_UNIFORM_BUFFER, static_cast<GLuint>(UniformBindings::Common),

src/video_core/renderer_opengl/gl_rasterizer.h

@@ -231,13 +231,13 @@ private:
     void SyncAndUploadLUTs();
 
     /// Upload the uniform blocks to the uniform buffer object
-    void UploadUniforms(bool accelerate_draw, bool use_gs);
+    void UploadUniforms(bool accelerate_draw);
 
     /// Generic draw function for DrawTriangles and AccelerateDrawBatch
     bool Draw(bool accelerate, bool is_indexed);
 
     /// Internal implementation for AccelerateDrawBatch
-    bool AccelerateDrawBatchInternal(bool is_indexed, bool use_gs);
+    bool AccelerateDrawBatchInternal(bool is_indexed);
 
     struct VertexArrayInfo {
         u32 vs_input_index_min;
@@ -304,7 +304,6 @@ private:
     OGLFramebuffer framebuffer;
     GLint uniform_buffer_alignment;
     std::size_t uniform_size_aligned_vs;
-    std::size_t uniform_size_aligned_gs;
     std::size_t uniform_size_aligned_fs;
 
     SamplerInfo texture_cube_sampler;

src/video_core/renderer_opengl/gl_shader_decompiler.cpp

@@ -249,10 +249,10 @@ public:
     GLSLGenerator(const std::set<Subroutine>& subroutines, const ProgramCode& program_code,
                   const SwizzleData& swizzle_data, u32 main_offset,
                   const RegGetter& inputreg_getter, const RegGetter& outputreg_getter,
-                  bool sanitize_mul, bool is_gs)
+                  bool sanitize_mul)
         : subroutines(subroutines), program_code(program_code), swizzle_data(swizzle_data),
           main_offset(main_offset), inputreg_getter(inputreg_getter),
-          outputreg_getter(outputreg_getter), sanitize_mul(sanitize_mul), is_gs(is_gs) {
+          outputreg_getter(outputreg_getter), sanitize_mul(sanitize_mul) {
 
         Generate();
     }
@@ -342,13 +342,6 @@ private:
 
     /// Generates code representing a bool uniform
     std::string GetUniformBool(u32 index) const {
-        if (is_gs && index == 15) {
-            // In PICA geometry shader, b15 is set to true after every geometry shader invocation.
-            // Accessing b15 usually indicates that the program relies on register value
-            // preservation across invocation (and therefore it uses b15 to determine whether to
-            // initialize the registers), which cannot be implemented in GL shaders.
-            throw DecompileFail("b15 access in geometry shader");
-        }
         return "uniforms.b[" + std::to_string(index) + "]";
     }
 
@@ -751,22 +744,10 @@ private:
                 break;
             }
 
-            case OpCode::Id::EMIT: {
+            case OpCode::Id::EMIT:
-                if (is_gs) {
+            case OpCode::Id::SETEMIT:
-                    shader.AddLine("emit();");
+                LOG_ERROR(HW_GPU, "Geometry shader operation detected in vertex shader");
-                }
                 break;
-            }
-
-            case OpCode::Id::SETEMIT: {
-                if (is_gs) {
-                    ASSERT(instr.setemit.vertex_id < 3);
-                    shader.AddLine("setemit(" + std::to_string(instr.setemit.vertex_id) + "u, " +
-                                   ((instr.setemit.prim_emit != 0) ? "true" : "false") + ", " +
-                                   ((instr.setemit.winding != 0) ? "true" : "false") + ");");
-                }
-                break;
-            }
 
             default: {
                 LOG_ERROR(HW_GPU, "Unhandled instruction: 0x{:02x} ({}): 0x{:08x}",
@@ -890,7 +871,6 @@ private:
     const RegGetter& inputreg_getter;
     const RegGetter& outputreg_getter;
     const bool sanitize_mul;
-    const bool is_gs;
 
     ShaderWriter shader;
 };
@@ -911,13 +891,12 @@ bool exec_shader();
 std::optional<std::string> DecompileProgram(const ProgramCode& program_code,
                                             const SwizzleData& swizzle_data, u32 main_offset,
                                             const RegGetter& inputreg_getter,
-                                            const RegGetter& outputreg_getter, bool sanitize_mul,
+                                            const RegGetter& outputreg_getter, bool sanitize_mul) {
-                                            bool is_gs) {
 
     try {
         auto subroutines = ControlFlowAnalyzer(program_code, main_offset).MoveSubroutines();
         GLSLGenerator generator(subroutines, program_code, swizzle_data, main_offset,
-                                inputreg_getter, outputreg_getter, sanitize_mul, is_gs);
+                                inputreg_getter, outputreg_getter, sanitize_mul);
         return generator.MoveShaderCode();
     } catch (const DecompileFail& exception) {
         LOG_INFO(HW_GPU, "Shader decompilation failed: {}", exception.what());

src/video_core/renderer_opengl/gl_shader_decompiler.h

@@ -20,7 +20,6 @@ std::string GetCommonDeclarations();
 std::optional<std::string> DecompileProgram(const ProgramCode& program_code,
                                             const SwizzleData& swizzle_data, u32 main_offset,
                                             const RegGetter& inputreg_getter,
-                                            const RegGetter& outputreg_getter, bool sanitize_mul,
+                                            const RegGetter& outputreg_getter, bool sanitize_mul);
-                                            bool is_gs);
 
 } // namespace OpenGL::ShaderDecompiler

src/video_core/renderer_opengl/gl_shader_gen.cpp

@@ -283,22 +283,6 @@ void PicaGSConfigCommonRaw::Init(const Pica::Regs& regs) {
     }
 }
 
-void PicaGSConfigRaw::Init(const Pica::Regs& regs, Pica::Shader::ShaderSetup& setup) {
-    PicaShaderConfigCommon::Init(regs.gs, setup);
-    PicaGSConfigCommonRaw::Init(regs);
-
-    num_inputs = regs.gs.max_input_attribute_index + 1;
-    input_map.fill(16);
-
-    for (u32 attr = 0; attr < num_inputs; ++attr) {
-        input_map[regs.gs.GetRegisterForAttribute(attr)] = attr;
-    }
-
-    attributes_per_vertex = regs.pipeline.vs_outmap_total_minus_1_a + 1;
-
-    gs_output_attributes = num_outputs;
-}
-
 /// Detects if a TEV stage is configured to be skipped (to avoid generating unnecessary code)
 static bool IsPassThroughTevStage(const TevStageConfig& stage) {
     return (stage.color_op == TevStageConfig::Operation::Replace &&
@@ -1675,7 +1659,7 @@ std::optional<std::string> GenerateVertexShader(const Pica::Shader::ShaderSetup&
 
     auto program_source_opt = ShaderDecompiler::DecompileProgram(
         setup.program_code, setup.swizzle_data, config.state.main_offset, get_input_reg,
-        get_output_reg, config.state.sanitize_mul, false);
+        get_output_reg, config.state.sanitize_mul);
 
     if (!program_source_opt)
         return {};
@@ -1727,11 +1711,6 @@ static std::string GetGSCommonSource(const PicaGSConfigCommonRaw& config, bool s
     }
 
     out += R"(
-#define uniforms gs_uniforms
-layout (std140) uniform gs_config {
-    pica_uniforms uniforms;
-};
-
 struct Vertex {
 )";
     out += "    vec4 attributes[" + std::to_string(config.gs_output_attributes) + "];\n";
@@ -1838,116 +1817,4 @@ void main() {
 
     return out;
 }
-
-std::optional<std::string> GenerateGeometryShader(const Pica::Shader::ShaderSetup& setup,
-                                                  const PicaGSConfig& config,
-                                                  bool separable_shader) {
-    std::string out = "";
-    if (separable_shader) {
-        out += "#extension GL_ARB_separate_shader_objects : enable\n";
-    }
-
-    if (config.state.num_inputs % config.state.attributes_per_vertex != 0)
-        return {};
-
-    switch (config.state.num_inputs / config.state.attributes_per_vertex) {
-    case 1:
-        out += "layout(points) in;\n";
-        break;
-    case 2:
-        out += "layout(lines) in;\n";
-        break;
-    case 4:
-        out += "layout(lines_adjacency) in;\n";
-        break;
-    case 3:
-        out += "layout(triangles) in;\n";
-        break;
-    case 6:
-        out += "layout(triangles_adjacency) in;\n";
-        break;
-    default:
-        return {};
-    }
-    out += "layout(triangle_strip, max_vertices = 30) out;\n\n";
-
-    out += GetGSCommonSource(config.state, separable_shader);
-
-    auto get_input_reg = [&](u32 reg) -> std::string {
-        ASSERT(reg < 16);
-        u32 attr = config.state.input_map[reg];
-        if (attr < config.state.num_inputs) {
-            return "vs_out_attr" + std::to_string(attr % config.state.attributes_per_vertex) + "[" +
-                   std::to_string(attr / config.state.attributes_per_vertex) + "]";
-        }
-        return "vec4(0.0, 0.0, 0.0, 1.0)";
-    };
-
-    auto get_output_reg = [&](u32 reg) -> std::string {
-        ASSERT(reg < 16);
-        if (config.state.output_map[reg] < config.state.num_outputs) {
-            return "output_buffer.attributes[" + std::to_string(config.state.output_map[reg]) + "]";
-        }
-        return "";
-    };
-
-    auto program_source_opt = ShaderDecompiler::DecompileProgram(
-        setup.program_code, setup.swizzle_data, config.state.main_offset, get_input_reg,
-        get_output_reg, config.state.sanitize_mul, true);
-
-    if (!program_source_opt)
-        return {};
-
-    std::string& program_source = *program_source_opt;
-
-    out += R"(
-Vertex output_buffer;
-Vertex prim_buffer[3];
-uint vertex_id = 0u;
-bool prim_emit = false;
-bool winding = false;
-
-void setemit(uint vertex_id_, bool prim_emit_, bool winding_);
-void emit();
-
-void main() {
-)";
-    for (u32 i = 0; i < config.state.num_outputs; ++i) {
-        out +=
-            "    output_buffer.attributes[" + std::to_string(i) + "] = vec4(0.0, 0.0, 0.0, 1.0);\n";
-    }
-
-    // execute shader
-    out += "\n    exec_shader();\n\n";
-
-    out += "}\n\n";
-
-    // Put the definition of setemit and emit after main to avoid spurious warning about
-    // uninitialized output_buffer in some drivers
-    out += R"(
-void setemit(uint vertex_id_, bool prim_emit_, bool winding_) {
-    vertex_id = vertex_id_;
-    prim_emit = prim_emit_;
-    winding = winding_;
-}
-
-void emit() {
-    prim_buffer[vertex_id] = output_buffer;
-
-    if (prim_emit) {
-        if (winding) {
-            EmitPrim(prim_buffer[1], prim_buffer[0], prim_buffer[2]);
-            winding = false;
-        } else {
-            EmitPrim(prim_buffer[0], prim_buffer[1], prim_buffer[2]);
-        }
-    }
-}
-)";
-
-    out += program_source;
-
-    return out;
-}
-
 } // namespace OpenGL

src/video_core/renderer_opengl/gl_shader_gen.h

@@ -195,26 +195,6 @@ struct PicaFixedGSConfig : Common::HashableStruct<PicaGSConfigCommonRaw> {
     }
 };
 
-struct PicaGSConfigRaw : PicaShaderConfigCommon, PicaGSConfigCommonRaw {
-    void Init(const Pica::Regs& regs, Pica::Shader::ShaderSetup& setup);
-
-    u32 num_inputs;
-    u32 attributes_per_vertex;
-
-    // input_map[input register index] -> input attribute index
-    std::array<u32, 16> input_map;
-};
-
-/**
- * This struct contains information to identify a GL geometry shader generated from PICA geometry
- * shader.
- */
-struct PicaGSConfig : Common::HashableStruct<PicaGSConfigRaw> {
-    explicit PicaGSConfig(const Pica::Regs& regs, Pica::Shader::ShaderSetup& setups) {
-        state.Init(regs, setups);
-    }
-};
-
 /**
  * Generates the GLSL vertex shader program source code that accepts vertices from software shader
  * and directly passes them to the fragment shader.
@@ -236,15 +216,6 @@ std::optional<std::string> GenerateVertexShader(const Pica::Shader::ShaderSetup&
  */
 std::string GenerateFixedGeometryShader(const PicaFixedGSConfig& config, bool separable_shader);
 
-/**
- * Generates the GLSL geometry shader program source code for the given GS program and its
- * configuration
- * @returns String of the shader source code; boost::none on failure
- */
-std::optional<std::string> GenerateGeometryShader(const Pica::Shader::ShaderSetup& setup,
-                                                  const PicaGSConfig& config,
-                                                  bool separable_shader);
-
 /**
  * Generates the GLSL fragment shader program source code for the current Pica state
  * @param config ShaderCacheKey object generated for the current Pica state, used for the shader
@@ -277,11 +248,4 @@ struct hash<OpenGL::PicaFixedGSConfig> {
         return k.Hash();
     }
 };
-
-template <>
-struct hash<OpenGL::PicaGSConfig> {
-    std::size_t operator()(const OpenGL::PicaGSConfig& k) const {
-        return k.Hash();
-    }
-};
 } // namespace std

src/video_core/renderer_opengl/gl_shader_manager.cpp

@@ -27,7 +27,6 @@ static void SetShaderUniformBlockBindings(GLuint shader) {
     SetShaderUniformBlockBinding(shader, "shader_data", UniformBindings::Common,
                                  sizeof(UniformData));
     SetShaderUniformBlockBinding(shader, "vs_config", UniformBindings::VS, sizeof(VSUniformData));
-    SetShaderUniformBlockBinding(shader, "gs_config", UniformBindings::GS, sizeof(GSUniformData));
 }
 
 static void SetShaderSamplerBinding(GLuint shader, const char* name,
@@ -205,9 +204,6 @@ private:
 using ProgrammableVertexShaders =
     ShaderDoubleCache<PicaVSConfig, &GenerateVertexShader, GL_VERTEX_SHADER>;
 
-using ProgrammableGeometryShaders =
-    ShaderDoubleCache<PicaGSConfig, &GenerateGeometryShader, GL_GEOMETRY_SHADER>;
-
 using FixedGeometryShaders =
     ShaderCache<PicaFixedGSConfig, &GenerateFixedGeometryShader, GL_GEOMETRY_SHADER>;
 
@@ -217,8 +213,8 @@ class ShaderProgramManager::Impl {
 public:
     explicit Impl(bool separable, bool is_amd)
         : is_amd(is_amd), separable(separable), programmable_vertex_shaders(separable),
-          trivial_vertex_shader(separable), programmable_geometry_shaders(separable),
+          trivial_vertex_shader(separable), fixed_geometry_shaders(separable),
-          fixed_geometry_shaders(separable), fragment_shaders(separable) {
+          fragment_shaders(separable) {
         if (separable)
             pipeline.Create();
     }
@@ -254,7 +250,6 @@ public:
     ProgrammableVertexShaders programmable_vertex_shaders;
     TrivialVertexShader trivial_vertex_shader;
 
-    ProgrammableGeometryShaders programmable_geometry_shaders;
     FixedGeometryShaders fixed_geometry_shaders;
 
     FragmentShaders fragment_shaders;
@@ -282,15 +277,6 @@ void ShaderProgramManager::UseTrivialVertexShader() {
     impl->current.vs = impl->trivial_vertex_shader.Get();
 }
 
-bool ShaderProgramManager::UseProgrammableGeometryShader(const PicaGSConfig& config,
-                                                         const Pica::Shader::ShaderSetup setup) {
-    GLuint handle = impl->programmable_geometry_shaders.Get(config, setup);
-    if (handle == 0)
-        return false;
-    impl->current.gs = handle;
-    return true;
-}
-
 void ShaderProgramManager::UseFixedGeometryShader(const PicaFixedGSConfig& config) {
     impl->current.gs = impl->fixed_geometry_shaders.Get(config);
 }

src/video_core/renderer_opengl/gl_shader_manager.h

@@ -91,15 +91,6 @@ static_assert(
 static_assert(sizeof(VSUniformData) < 16384,
               "VSUniformData structure must be less than 16kb as per the OpenGL spec");
 
-struct GSUniformData {
-    PicaUniformsData uniforms;
-};
-static_assert(
-    sizeof(GSUniformData) == 1856,
-    "The size of the GSUniformData structure has changed, update the structure in the shader");
-static_assert(sizeof(GSUniformData) < 16384,
-              "GSUniformData structure must be less than 16kb as per the OpenGL spec");
-
 /// A class that manage different shader stages and configures them with given config data.
 class ShaderProgramManager {
 public:
@@ -111,9 +102,6 @@ public:
 
     void UseTrivialVertexShader();
 
-    bool UseProgrammableGeometryShader(const PicaGSConfig& config,
-                                       const Pica::Shader::ShaderSetup setup);
-
     void UseFixedGeometryShader(const PicaFixedGSConfig& config);
 
     void UseTrivialGeometryShader();