Rework Maxwell3D Shader/Pipeline Stages Compilation with UBO support

Shader compilation is now broken into shader program parsing and pipeline shader compilation which will allow for supporting dual vertex shaders and more atomic invalidation depending on runtime state limiting the amount of work that is redone. 

Bindings are now properly handled allowing for bound UBOs to be converted to the appropriate host UBO as designated by the shader compiler by creating Vulkan Descriptor Sets that match it.

app/src/main/cpp/skyline/gpu/interconnect/graphics_context.h

@@ -556,61 +556,156 @@ namespace skyline::gpu::interconnect {
       private:
         struct Shader {
             bool enabled{false};
+            ShaderCompiler::Stage stage;
+
             bool invalidated{true}; //!< If the shader that existed earlier has been invalidated
             bool shouldCheckSame{false}; //!< If we should do a check for the shader being the same as before
-            ShaderCompiler::Stage stage;
-            vk::ShaderStageFlagBits vkStage;
             u32 offset{}; //!< Offset of the shader from the base IOVA
             std::vector<u8> data; //!< The shader bytecode in a vector
+            std::optional<ShaderCompiler::IR::Program> program;
+
+            Shader(ShaderCompiler::Stage stage) : stage(stage) {}
+
+            maxwell3d::PipelineStage ToPipelineStage() {
+                using ShaderStage = ShaderCompiler::Stage;
+                using PipelineStage = maxwell3d::PipelineStage;
+
+                switch (stage) {
+                    case ShaderStage::VertexA:
+                    case ShaderStage::VertexB:
+                        return PipelineStage::Vertex;
+
+                    case ShaderStage::TessellationControl:
+                        return PipelineStage::TessellationControl;
+                    case ShaderStage::TessellationEval:
+                        return PipelineStage::TessellationEvaluation;
+
+                    case ShaderStage::Geometry:
+                        return PipelineStage::Geometry;
+
+                    case ShaderStage::Fragment:
+                        return PipelineStage::Fragment;
+
+                    case ShaderStage::Compute:
+                        throw exception("Unexpected compute shader in Maxwell3D");
+                }
+            }
+        };
+
+        struct Shaders : public std::array<Shader, maxwell3d::ShaderStageCount> {
+            Shaders() : array({
+                                  Shader{ShaderCompiler::Stage::VertexA},
+                                  Shader{ShaderCompiler::Stage::VertexB},
+                                  Shader{ShaderCompiler::Stage::TessellationControl},
+                                  Shader{ShaderCompiler::Stage::TessellationEval},
+                                  Shader{ShaderCompiler::Stage::Geometry},
+                                  Shader{ShaderCompiler::Stage::Fragment},
+                              }) {}
+
+            Shader &at(maxwell3d::ShaderStage stage) {
+                return array::at(static_cast<size_t>(stage));
+            }
+
+            Shader &operator[](maxwell3d::ShaderStage stage) {
+                return array::operator[](static_cast<size_t>(stage));
+            }
+        };
+
+        struct PipelineStage {
+            bool enabled{false};
+            vk::ShaderStageFlagBits vkStage;
+
+            std::variant<ShaderCompiler::IR::Program, std::reference_wrapper<ShaderCompiler::IR::Program>> program; //!< The shader program by value or by reference (VertexA and VertexB shaders when combined will store by value, otherwise only a reference is stored)
+
+            bool needsRecompile{}; //!< If the shader needs to be recompiled as runtime information has changed
+            ShaderCompiler::VaryingState previousStageStores{};
+            u32 bindingBase{}, bindingLast{}; //!< The base and last binding for descriptors bound to this stage
             std::optional<vk::raii::ShaderModule> vkModule;
 
-            Shader(ShaderCompiler::Stage stage, vk::ShaderStageFlagBits vkStage) : stage(stage), vkStage(vkStage) {}
+            std::array<std::shared_ptr<BufferView>, maxwell3d::PipelineStageConstantBufferCount> constantBuffers{};
+
+            PipelineStage(vk::ShaderStageFlagBits vkStage) : vkStage(vkStage) {}
+        };
+
+        struct PipelineStages : public std::array<PipelineStage, maxwell3d::PipelineStageCount> {
+            PipelineStages() : array({
+                                         PipelineStage{vk::ShaderStageFlagBits::eVertex},
+                                         PipelineStage{vk::ShaderStageFlagBits::eTessellationControl},
+                                         PipelineStage{vk::ShaderStageFlagBits::eTessellationEvaluation},
+                                         PipelineStage{vk::ShaderStageFlagBits::eGeometry},
+                                         PipelineStage{vk::ShaderStageFlagBits::eFragment},
+                                     }) {}
+
+            PipelineStage &at(maxwell3d::PipelineStage stage) {
+                return array::at(static_cast<size_t>(stage));
+            }
+
+            PipelineStage &operator[](maxwell3d::PipelineStage stage) {
+                return array::operator[](static_cast<size_t>(stage));
+            }
         };
 
         IOVA shaderBaseIova{}; //!< The base IOVA that shaders are located at an offset from
-        std::array<Shader, maxwell3d::ShaderStageCount> shaders{
+        Shaders shaders;
-            Shader{ShaderCompiler::Stage::VertexA, vk::ShaderStageFlagBits::eVertex},
+        PipelineStages pipelineStages;
-            Shader{ShaderCompiler::Stage::VertexB, vk::ShaderStageFlagBits::eVertex},
-            Shader{ShaderCompiler::Stage::TessellationControl, vk::ShaderStageFlagBits::eTessellationControl},
-            Shader{ShaderCompiler::Stage::TessellationEval, vk::ShaderStageFlagBits::eTessellationEvaluation},
-            Shader{ShaderCompiler::Stage::Geometry, vk::ShaderStageFlagBits::eGeometry},
-            Shader{ShaderCompiler::Stage::Fragment, vk::ShaderStageFlagBits::eFragment},
-        };
 
-        std::array<vk::PipelineShaderStageCreateInfo, maxwell3d::StageCount - 1> shaderStagesInfo{}; //!< Storage backing for the pipeline shader stage information for all shaders aside from 'VertexA' which uses the same stage as 'VertexB'
+        std::array<vk::PipelineShaderStageCreateInfo, maxwell3d::PipelineStageCount> shaderStagesInfo{}; //!< Storage backing for the pipeline shader stage information for all shaders aside from 'VertexA' which uses the same stage as 'VertexB'
-        size_t activeShaderStagesInfoCount{}; //!< The amount of active shader stages with valid entries in 'shaderStagesInfo'
+        std::optional<vk::raii::DescriptorSetLayout> descriptorSetLayout{}; //!< The descriptor set layout for the pipeline (Only valid when `activeShaderStagesInfoCount` is non-zero)
 
         ShaderCompiler::RuntimeInfo runtimeInfo{};
 
         constexpr static size_t MaxShaderBytecodeSize{1 * 1024 * 1024}; //!< The largest shader binary that we support (1 MiB)
 
-        span<vk::PipelineShaderStageCreateInfo> GetShaderStages() {
+        constexpr static size_t PipelineUniqueDescriptorTypeCount{1}; //!< The amount of unique descriptor types that may be bound to a pipeline
-            if (!activeShaderStagesInfoCount) {
+        constexpr static size_t MaxPipelineDescriptorWriteCount{maxwell3d::PipelineStageCount * PipelineUniqueDescriptorTypeCount}; //!< The maxium amount of descriptors writes that are used to bind a pipeline
-                runtimeInfo.previous_stage_stores.mask.set(); // First stage should always have all bits set
+        constexpr static size_t MaxPipelineDescriptorCount{100}; //!< The maxium amount of descriptors we support being bound to a pipeline
-                ShaderCompiler::Backend::Bindings bindings;
 
-                size_t count{};
+        boost::container::static_vector<vk::WriteDescriptorSet, MaxPipelineDescriptorWriteCount> descriptorSetWrites;
-                for (auto &shader : shaders) {
+        boost::container::static_vector<vk::DescriptorSetLayoutBinding, MaxPipelineDescriptorCount> layoutBindings;
-                    if (shader.enabled) {
+        boost::container::static_vector<vk::DescriptorBufferInfo, MaxPipelineDescriptorCount> bufferInfo;
-                        // We only want to add the stage if it is enabled on the guest
-                        if (shader.invalidated) {
-                            // If a shader is invalidated, we need to ensure the corresponding VkShaderModule is accurate
-                            if (!shader.data.empty() && shader.shouldCheckSame) {
-                                auto newIovaRanges{channelCtx.asCtx->gmmu.TranslateRange(shaderBaseIova + shader.offset, shader.data.size())};
-                                auto originalShader{shader.data.data()};
 
-                                // A fast path to check if the shader is the same as before to avoid work
+        /**
-                                for (auto &range : newIovaRanges) {
+         * @brief All state concerning the shader programs and their bindings
-                                    if (range.data() && std::memcmp(range.data(), originalShader, range.size()) == 0) {
+         * @note The `descriptorSetWrite` will have a null `dstSet` which needs to be assigned prior to usage
-                                        originalShader += range.size();
+         */
-                                    } else {
+        struct ShaderProgramState {
-                                        break;
+            span<vk::PipelineShaderStageCreateInfo> shaders;
-                                    }
+            vk::DescriptorSetLayout descriptorSetLayout;
+            span<vk::WriteDescriptorSet> descriptorSetWrites; //!< The writes to the descriptor set that need to be done prior to executing a pipeline
+        };
+
+        /**
+         * @note The return value of previous calls will be invalidated on a call to this as values are provided by reference
+         * @note Any bound resources will automatically be attached to the CommandExecutor, there's no need to manually attach them
+         */
+        ShaderProgramState CompileShaderProgramState() {
+            for (auto &shader : shaders) {
+                auto &pipelineStage{pipelineStages[shader.ToPipelineStage()]};
+                if (shader.enabled) {
+                    // We only want to include the shader if it is enabled on the guest
+                    if (shader.invalidated) {
+                        // If a shader is invalidated, we need to reparse the program (given that it has changed)
+
+                        bool shouldParseShader{true};
+                        if (!shader.data.empty() && shader.shouldCheckSame) {
+                            // A fast path to check if the shader is the same as before to avoid reparsing the shader
+                            auto newIovaRanges{channelCtx.asCtx->gmmu.TranslateRange(shaderBaseIova + shader.offset, shader.data.size())};
+                            auto originalShader{shader.data.data()};
+
+                            shouldParseShader = false;
+                            for (auto &range : newIovaRanges) {
+                                if (range.data() && std::memcmp(range.data(), originalShader, range.size()) == 0) {
+                                    originalShader += range.size();
+                                } else {
+                                    shouldParseShader = true;
+                                    break;
                                 }
-
-                                shader.shouldCheckSame = true;
                             }
 
+                            shader.shouldCheckSame = true;
+                        }
+
+                        if (shouldParseShader) {
                             // A pass to check if the shader has a BRA infloop opcode ending (On most commercial games)
                             shader.data.resize(MaxShaderBytecodeSize);
                             auto foundEnd{channelCtx.asCtx->gmmu.ReadTill(shader.data, shaderBaseIova + shader.offset, [](span<u8> data) -> std::optional<size_t> {
@@ -628,21 +723,93 @@ namespace skyline::gpu::interconnect {
                                 return std::nullopt;
                             })};
 
-                            shader.vkModule = gpu.shader.CompileGraphicsShader(shader.data, shader.stage, shader.offset, runtimeInfo, bindings);
+                            shader.program = gpu.shader.ParseGraphicsShader(shader.data, shader.stage, shader.offset);
+
+                            if (shader.stage != ShaderCompiler::Stage::VertexA) {
+                                pipelineStage.program.emplace<std::reference_wrapper<ShaderCompiler::IR::Program>>(*shader.program);
+                            }
+
+                            pipelineStage.enabled = true;
+                            pipelineStage.needsRecompile = true;
                         }
 
-                        shaderStagesInfo[count++] = vk::PipelineShaderStageCreateInfo{
+                        shader.invalidated = false;
-                            .stage = shader.vkStage,
+                    }
-                            .module = **shader.vkModule,
+                } else if (shader.stage != ShaderCompiler::Stage::VertexA) {
-                            .pName = "main",
+                    pipelineStage.enabled = false;
-                        };
+                }
+            }
+
+            runtimeInfo.previous_stage_stores.mask.set(); // First stage should always have all bits set
+            ShaderCompiler::Backend::Bindings bindings{};
+
+            size_t count{};
+            for (auto &pipelineStage : pipelineStages) {
+                if (!pipelineStage.enabled)
+                    continue;
+
+                auto &program{std::visit(VariantVisitor{
+                    [](ShaderCompiler::IR::Program &program) -> ShaderCompiler::IR::Program & { return program; },
+                    [](std::reference_wrapper<ShaderCompiler::IR::Program> program) -> ShaderCompiler::IR::Program & { return program.get(); },
+                }, pipelineStage.program)};
+
+                if (pipelineStage.needsRecompile || bindings.unified != pipelineStage.bindingBase || pipelineStage.previousStageStores.mask != runtimeInfo.previous_stage_stores.mask) {
+                    pipelineStage.previousStageStores = runtimeInfo.previous_stage_stores;
+                    pipelineStage.bindingBase = bindings.unified;
+                    pipelineStage.vkModule = gpu.shader.CompileShader(runtimeInfo, program, bindings);
+                    pipelineStage.bindingLast = bindings.unified;
+                }
+
+                runtimeInfo.previous_stage_stores = program.info.stores;
+                if (program.is_geometry_passthrough)
+                    runtimeInfo.previous_stage_stores.mask |= program.info.passthrough.mask;
+                bindings.unified = pipelineStage.bindingLast;
+
+                u32 bindingIndex{pipelineStage.bindingBase};
+                if (!program.info.constant_buffer_descriptors.empty()) {
+                    descriptorSetWrites.push_back(vk::WriteDescriptorSet{
+                        .dstBinding = bindingIndex,
+                        .descriptorCount = static_cast<u32>(program.info.constant_buffer_descriptors.size()),
+                        .descriptorType = vk::DescriptorType::eUniformBuffer,
+                        .pBufferInfo = bufferInfo.data() + bufferInfo.size(),
+                    });
+
+                    for (auto &constantBuffer : program.info.constant_buffer_descriptors) {
+                        layoutBindings.push_back(vk::DescriptorSetLayoutBinding{
+                            .binding = bindingIndex++,
+                            .descriptorType = vk::DescriptorType::eUniformBuffer,
+                            .descriptorCount = 1,
+                            .stageFlags = pipelineStage.vkStage,
+                        });
+
+                        auto view{pipelineStage.constantBuffers[constantBuffer.index]};
+                        std::scoped_lock lock(*view);
+                        bufferInfo.push_back(vk::DescriptorBufferInfo{
+                            .buffer = view->buffer->GetBacking(),
+                            .offset = view->offset,
+                            .range = view->range,
+                        });
+                        executor.AttachBuffer(view.get());
                     }
                 }
 
-                activeShaderStagesInfoCount = count;
+                shaderStagesInfo[count++] = vk::PipelineShaderStageCreateInfo{
+                    .stage = pipelineStage.vkStage,
+                    .module = **pipelineStage.vkModule,
+                    .pName = "main",
+                };
             }
 
-            return span(shaderStagesInfo.data(), activeShaderStagesInfoCount);
+            descriptorSetLayout.emplace(gpu.vkDevice, vk::DescriptorSetLayoutCreateInfo{
+                .pBindings = layoutBindings.data(),
+                .bindingCount = static_cast<u32>(layoutBindings.size()),
+            });
+
+            return {
+                span(shaderStagesInfo.data(), count),
+                **descriptorSetLayout,
+                descriptorSetWrites,
+            };
         }
 
       public:
@@ -663,17 +830,27 @@ namespace skyline::gpu::interconnect {
         }
 
         void SetShaderEnabled(maxwell3d::ShaderStage stage, bool enabled) {
-            auto &shader{shaders[static_cast<size_t>(stage)]};
+            auto &shader{shaders[stage]};
             shader.enabled = enabled;
             shader.invalidated = true;
         }
 
         void SetShaderOffset(maxwell3d::ShaderStage stage, u32 offset) {
-            auto &shader{shaders[static_cast<size_t>(stage)]};
+            auto &shader{shaders[stage]};
             shader.offset = offset;
             shader.invalidated = true;
         }
 
+        void BindPipelineConstantBuffer(maxwell3d::PipelineStage stage, bool enable, u32 index) {
+            auto &constantBufferView{pipelineStages[stage].constantBuffers[index]};
+
+            if (enable) {
+                constantBufferView = GetConstantBufferSelectorView()->shared_from_this();
+            } else {
+                constantBufferView = nullptr;
+            }
+        }
+
         /* Rasterizer State */
       private:
         vk::StructureChain<vk::PipelineRasterizationStateCreateInfo, vk::PipelineRasterizationProvokingVertexStateCreateInfoEXT> rasterizerState;
@@ -1388,16 +1565,35 @@ namespace skyline::gpu::interconnect {
 
             vertexState.get<vk::PipelineVertexInputStateCreateInfo>().vertexAttributeDescriptionCount = static_cast<u32>(vertexAttributesDescriptions.size());
 
-            // Shader Setup
+            // Shader + Binding Setup
-            auto shaderStages{GetShaderStages()};
+            auto programState{CompileShaderProgramState()};
-            pipelineState.pStages = shaderStages.data();
+            pipelineState.pStages = programState.shaders.data();
-            pipelineState.stageCount = static_cast<u32>(shaderStages.size());
+            pipelineState.stageCount = static_cast<u32>(programState.shaders.size());
+
+            auto descriptorSet{gpu.descriptor.AllocateSet(programState.descriptorSetLayout)};
+            for (auto &descriptorSetWrite : programState.descriptorSetWrites)
+                descriptorSetWrite.dstSet = descriptorSet;
+            gpu.vkDevice.updateDescriptorSets(programState.descriptorSetWrites, nullptr);
+
+            vk::raii::PipelineLayout pipelineLayout(gpu.vkDevice, vk::PipelineLayoutCreateInfo{
+                .pSetLayouts = &programState.descriptorSetLayout,
+                .setLayoutCount = 1,
+            });
+
+            // Draw Persistent Storage
+            struct Storage : FenceCycleDependency {
+                vk::raii::PipelineLayout pipelineLayout;
+                std::optional<vk::raii::Pipeline> pipeline;
+                DescriptorAllocator::ActiveDescriptorSet descriptorSet;
+
+                Storage(vk::raii::PipelineLayout &&pipelineLayout, DescriptorAllocator::ActiveDescriptorSet &&descriptorSet) : pipelineLayout(std::move(pipelineLayout)), descriptorSet(std::move(descriptorSet)) {}
+            };
+
+            auto storage{std::make_shared<Storage>(std::move(pipelineLayout), std::move(descriptorSet))};
 
             // Submit Draw
-            executor.AddSubpass([vertexCount, firstVertex, &vkDevice = gpu.vkDevice, pipelineCreateInfo = pipelineState, vertexBufferHandles = std::move(vertexBufferHandles), vertexBufferOffsets = std::move(vertexBufferOffsets), pipelineCache = *pipelineCache](vk::raii::CommandBuffer &commandBuffer, const std::shared_ptr<FenceCycle> &cycle, GPU &, vk::RenderPass renderPass, u32 subpassIndex) mutable {
+            executor.AddSubpass([vertexCount, firstVertex, &vkDevice = gpu.vkDevice, pipelineCreateInfo = pipelineState, storage = std::move(storage), vertexBufferHandles = std::move(vertexBufferHandles), vertexBufferOffsets = std::move(vertexBufferOffsets), pipelineCache = *pipelineCache](vk::raii::CommandBuffer &commandBuffer, const std::shared_ptr<FenceCycle> &cycle, GPU &, vk::RenderPass renderPass, u32 subpassIndex) mutable {
-                vk::PipelineLayoutCreateInfo layoutCreateInfo{};
+                pipelineCreateInfo.layout = *storage->pipelineLayout;
-                vk::raii::PipelineLayout pipelineLayout(vkDevice, layoutCreateInfo);
-                pipelineCreateInfo.layout = *pipelineLayout;
 
                 pipelineCreateInfo.renderPass = renderPass;
                 pipelineCreateInfo.subpass = subpassIndex;
@@ -1420,16 +1616,13 @@ namespace skyline::gpu::interconnect {
                     }
                 }
 
+                commandBuffer.bindDescriptorSets(vk::PipelineBindPoint::eGraphics, *storage->pipelineLayout, 0, storage->descriptorSet, nullptr);
+
                 commandBuffer.draw(vertexCount, 1, firstVertex, 0);
 
-                struct Storage : FenceCycleDependency {
+                storage->pipeline = vk::raii::Pipeline(vkDevice, pipeline.value);
-                    vk::raii::PipelineLayout pipelineLayout;
-                    vk::raii::Pipeline pipeline;
 
-                    Storage(vk::raii::PipelineLayout &&pipelineLayout, vk::raii::Pipeline &&pipeline) : pipelineLayout(std::move(pipelineLayout)), pipeline(std::move(pipeline)) {}
+                cycle->AttachObject(storage);
-                };
-
-                cycle->AttachObject(std::make_shared<Storage>(std::move(pipelineLayout), vk::raii::Pipeline(vkDevice, pipeline.value)));
             }, vk::Rect2D{
                 .extent = activeColorRenderTargets[0]->texture->dimensions,
             }, {}, activeColorRenderTargets, depthRenderTargetView);

app/src/main/cpp/skyline/gpu/shader_manager.cpp

@@ -83,11 +83,11 @@ namespace skyline::gpu {
      */
     class GraphicsEnvironment : public Shader::Environment {
       private:
-        std::vector<u8> binary;
+        span<u8> binary;
         u32 baseOffset;
 
       public:
-        explicit GraphicsEnvironment(std::vector<u8> pBinary, u32 baseOffset, Shader::Stage pStage) : binary(std::move(pBinary)), baseOffset(baseOffset) {
+        GraphicsEnvironment(span<u8> pBinary, u32 baseOffset, Shader::Stage pStage) : binary(pBinary), baseOffset(baseOffset) {
             sph = *reinterpret_cast<Shader::ProgramHeader *>(binary.data());
             start_address = baseOffset;
             stage = pStage;
@@ -125,24 +125,21 @@ namespace skyline::gpu {
         }
     };
 
-    vk::raii::ShaderModule ShaderManager::CompileGraphicsShader(const std::vector<u8> &binary, Shader::Stage stage, u32 baseOffset, Shader::RuntimeInfo &runtimeInfo, Shader::Backend::Bindings &bindings) {
+    Shader::IR::Program ShaderManager::ParseGraphicsShader(span<u8> binary, Shader::Stage stage, u32 baseOffset) {
         GraphicsEnvironment environment{binary, baseOffset, stage};
         Shader::Maxwell::Flow::CFG cfg(environment, flowBlockPool, Shader::Maxwell::Location{static_cast<u32>(baseOffset + sizeof(Shader::ProgramHeader))});
 
-        auto program{Shader::Maxwell::TranslateProgram(instPool, blockPool, environment, cfg, hostTranslateInfo)};
+        return Shader::Maxwell::TranslateProgram(instPool, blockPool, environment, cfg, hostTranslateInfo);
+    }
 
+    vk::raii::ShaderModule ShaderManager::CompileShader(Shader::RuntimeInfo &runtimeInfo, Shader::IR::Program &program, Shader::Backend::Bindings &bindings) {
         auto spirv{Shader::Backend::SPIRV::EmitSPIRV(profile, runtimeInfo, program, bindings)};
 
-        runtimeInfo.previous_stage_stores = program.info.stores;
-        if (program.is_geometry_passthrough)
-            runtimeInfo.previous_stage_stores.mask |= program.info.passthrough.mask;
-
         vk::ShaderModuleCreateInfo createInfo{
             .pCode = spirv.data(),
             .codeSize = spirv.size() * sizeof(u32),
         };
         vk::raii::ShaderModule shaderModule(gpu.vkDevice, createInfo);
-
         return shaderModule;
     }
 }

app/src/main/cpp/skyline/gpu/shader_manager.h

@@ -8,6 +8,7 @@
 #include <shader_compiler/frontend/maxwell/control_flow.h>
 #include <shader_compiler/frontend/ir/value.h>
 #include <shader_compiler/frontend/ir/basic_block.h>
+#include <shader_compiler/frontend/ir/program.h>
 #include <shader_compiler/host_translate_info.h>
 #include <shader_compiler/profile.h>
 #include <shader_compiler/runtime_info.h>
@@ -30,9 +31,8 @@ namespace skyline::gpu {
       public:
         ShaderManager(const DeviceState &state, GPU &gpu);
 
-        /**
+        Shader::IR::Program ParseGraphicsShader(span<u8> binary, Shader::Stage stage, u32 baseOffset);
-         * @note `runtimeInfo::previous_stage_stores` will automatically be updated for the next stage
+
-         */
+        vk::raii::ShaderModule CompileShader(Shader::RuntimeInfo &runtimeInfo, Shader::IR::Program &program, Shader::Backend::Bindings &bindings);
-        vk::raii::ShaderModule CompileGraphicsShader(const std::vector<u8> &binary, Shader::Stage stage, u32 baseOffset, Shader::RuntimeInfo &runtimeInfo, Shader::Backend::Bindings &bindings);
     };
 }

app/src/main/cpp/skyline/soc/gm20b/engines/maxwell_3d.cpp

@@ -518,6 +518,15 @@ namespace skyline::soc::gm20b::engine::maxwell3d {
             static_assert(type::ShaderStageCount == 6 && type::ShaderStageCount < BOOST_PP_LIMIT_REPEAT);
             #undef SHADER_CALLBACKS
 
+            #define PIPELINE_CALLBACKS(z, idx, data)                                                                                       \
+                MAXWELL3D_ARRAY_STRUCT_CASE(bind, idx, constantBuffer, {                                                                   \
+                    context.BindPipelineConstantBuffer(static_cast<type::PipelineStage>(idx), constantBuffer.valid, constantBuffer.index); \
+                })
+
+            BOOST_PP_REPEAT(5, PIPELINE_CALLBACKS, 0)
+            static_assert(type::PipelineStageCount == 5 && type::PipelineStageCount < BOOST_PP_LIMIT_REPEAT);
+            #undef PIPELINE_CALLBACKS
+
             MAXWELL3D_ARRAY_CASE(firmwareCall, 4, {
                 registers.raw[0xD00] = 1;
             })

app/src/main/cpp/skyline/soc/gm20b/engines/maxwell_3d.h

@@ -287,6 +287,8 @@ namespace skyline::soc::gm20b::engine::maxwell3d {
                 type::Address address;
             };
             Register<0x8E0, ConstantBufferSelector> constantBufferSelector;
+
+            Register<0x900, std::array<type::Bind, type::PipelineStageCount>> bind; //!< Binds constant buffers to pipeline stages
         };
         static_assert(sizeof(Registers) == (RegisterCount * sizeof(u32)));
         #pragma pack(pop)