VKBuffer.hpp

Go to the documentation of this file.

#pragma once
 
#include "Buffer.hpp"
 
#include "MayaFlux/Buffers/BufferProcessor.hpp"
 
#include "MayaFlux/Portal/Graphics/ShaderUtils.hpp"
 
namespace MayaFlux::Registry::Service {
struct BufferService;
struct ComputeService;
}
 
namespace MayaFlux::Core {
class Window;
}
 
namespace MayaFlux::Buffers {
 
class RenderProcessor;
struct ShaderConfig;
 
/**
 * @struct VKBufferResources
 * @brief Raw Vulkan handles owned by a VKBuffer instance.
 *
 * The index_buffer / index_memory / index_size_bytes fields are populated
 * by GeometryBindingsProcessor after index data upload and consumed by
 * RenderProcessor to select the indexed draw path. They are null/zero for
 * all non-indexed geometry; no separate allocation object is required.
 */
struct VKBufferResources {
    vk::Buffer buffer;
    vk::DeviceMemory memory;
    void* mapped_ptr { nullptr };
 
    vk::Buffer index_buffer;
    vk::DeviceMemory index_memory;
    size_t index_size_bytes { 0 };
};
 
using RenderPipelineID = uint64_t;
using CommandBufferID = uint64_t;
 
/**
 * @class VKBuffer
 * @brief Vulkan-backed buffer wrapper used in processing chains
 *
 * VKBuffer is a lightweight, high-level representation of a GPU buffer used by
 * the MayaFlux processing pipeline. It carries semantic metadata (Kakshya
 * modalities and dimensions), integrates with the Buffer processing chain and
 * BufferManager, and exposes Vulkan handles once the backend registers the
 * buffer. Prior to registration the object contains no GPU resources and can
 * be manipulated cheaply (like AudioBuffer).
 *
 * Responsibilities:
 * - Store buffer size, usage intent and semantic modality
 * - Provide inferred data dimensions for processors and pipeline inspection
 * - Hold Vulkan handles (VkBuffer, VkDeviceMemory) assigned by graphics backend
 * - Provide convenience helpers for common Vulkan creation flags and memory props
 * - Integrate with BufferProcessor / BufferProcessingChain for default processing
 *
 * Note: Actual allocation, mapping and command-based transfers are performed by
 * the graphics backend / BufferManager. VKBuffer only stores handles and
 * metadata and provides helpers for processors to operate on it.
 */
class MAYAFLUX_API VKBuffer : public Buffer {
public:
    using RenderConfig = Portal::Graphics::RenderConfig;
 
    enum class Usage : uint8_t {
        STAGING, ///< Host-visible staging buffer (CPU-writable, eTransferSrc|Dst)
        DEVICE, ///< Device-local GPU-only buffer
        COMPUTE, ///< Storage buffer for compute shaders (device-local)
        VERTEX, ///< Vertex buffer
        INDEX, ///< Index buffer
        UNIFORM, ///< Uniform buffer (host-visible)
        UNIFORM_BDA, ///< Uniform buffer with device address query support
        STORAGE_BDA, ///< Storage buffer with device address query support
        HOST_STORAGE, ///< Host-visible storage buffer (eStorageBuffer + eHostVisible|eHostCoherent)
    };
 
    /**
     * @brief Context shared with BufferProcessors during pipeline execution
     *
     * Processors can use this struct to share data (e.g., push constant staging)
     * and metadata during processing of this buffer in a chain.
     */
    struct PipelineContext {
        std::vector<uint8_t> push_constant_staging;
 
        std::vector<Portal::Graphics::DescriptorBindingInfo> descriptor_buffer_bindings;
 
        std::unordered_map<std::string, std::any> metadata;
    };
 
    /**
     * @brief Engine-internal per-frame binding state
     *
     * Carries descriptor bindings written by engine processors (e.g.
     * MeshNetworkProcessor SSBOs) that must reach RenderProcessor::execute_shader
     * without passing through the user-facing PipelineContext. Access is
     * unrestricted by convention only: engine processors write here, user code
     * should not.
     */
    struct EngineContext {
        std::vector<Portal::Graphics::DescriptorBindingInfo> ssbo_bindings;
    };
 
    /**
     * @brief Construct an unregistered VKBuffer
     *
     * Creates a VKBuffer object with the requested capacity, usage intent and
     * semantic modality. No Vulkan resources are created by this constructor —
     * registration with the BufferManager / backend is required to allocate
     * VkBuffer and VkDeviceMemory.
     *
     * @param size_bytes Buffer capacity in bytes.
     * @param usage Intended usage pattern (affects Vulkan flags and memory).
     * @param modality Semantic interpretation of the buffer contents.
     */
    VKBuffer(
        size_t size_bytes,
        Usage usage,
        Kakshya::DataModality modality = Kakshya::DataModality::VERTICES_3D);
 
    VKBuffer() = default;
 
    /**
     * @brief Virtual destructor
     *
     * VKBuffer does not own Vulkan resources directly; cleanup is handled by
     * the backend during unregistration. The destructor ensures derived class
     * cleanup and safe destruction semantics.
     */
    ~VKBuffer() override;
 
    /**
     * @brief Get the Vulkan device address for this buffer (if applicable)
     * @return 64-bit device address, or 0 if not BDA-capable or not initialized
     *
     * Only buffers created with Usage::UNIFORM_BDA or Usage::STORAGE_BDA and
     * registered with the backend will return a valid device address. This is
     * used by processors that need to bind buffers via device address (e.g.,
     * for ray tracing or bindless access).
     */
    [[nodiscard]] uint64_t get_device_address() const;
 
    /**
     * @brief Clear buffer contents
     *
     * If the buffer is host-visible and mapped the memory is zeroed. For
     * device-local buffers the backend must provide a ClearBufferProcessor
     * that performs the appropriate GPU-side clear.
     */
    void clear() override;
 
    /**
     * @brief Read buffer contents as Kakshya DataVariant
     *
     * For host-visible buffers this returns a single DataVariant containing the
     * raw bytes. For device-local buffers this warns and returns empty — a
     * BufferDownloadProcessor should be used to read GPU-only memory.
     *
     * @return Vector of DataVariant objects representing buffer contents.
     */
    std::vector<Kakshya::DataVariant> get_data();
 
    /**
     * @brief Write data into the buffer
     *
     * If the buffer is host-visible and mapped the provided data is copied into
     * the mapped memory. For device-local buffers, a BufferUploadProcessor must
     * be present in the processing chain to perform the staging/upload.
     *
     * @param data Vector of Kakshya::DataVariant containing the payload to copy.
     */
    void set_data(const std::vector<Kakshya::DataVariant>& data);
 
    /**
     * @brief Resize buffer and recreate GPU resources if needed
     * @param new_size New size in bytes
     * @param preserve_data If true, copy existing data to new buffer
     *
     * If buffer is already initialized (has GPU resources), this will:
     * 1. Create new GPU buffer with new size
     * 2. Optionally copy old data
     * 3. Destroy old GPU buffer
     * 4. Update buffer resources
     *
     * If buffer is not initialized, just updates logical size.
     */
    void resize(size_t new_size, bool preserve_data = false);
 
    /**
     * @brief Get current logical size in bytes
     * @return Buffer size in bytes.
     */
    size_t get_size() const { return m_size_bytes; }
 
    /**
     * @brief Run the buffer's default processor (if set and enabled)
     *
     * Invokes the attached default BufferProcessor. Processors should be
     * prepared to handle mapped/unmapped memory according to buffer usage.
     */
    void process_default() override;
 
    /**
     * @brief Set the buffer's default processor
     *
     * Attaches a processor that will be invoked by process_default(). The
     * previous default processor (if any) is detached first.
     *
     * @param processor Shared pointer to a BufferProcessor or nullptr to clear.
     */
    void set_default_processor(const std::shared_ptr<BufferProcessor>& processor) override;
 
    /**
     * @brief Get the currently attached default processor
     * @return Shared pointer to the default BufferProcessor or nullptr.
     */
    std::shared_ptr<Buffers::BufferProcessor> get_default_processor() const override;
 
    /**
     * @brief Access the buffer's processing chain
     * @return Shared pointer to the BufferProcessingChain used for this buffer.
     */
    std::shared_ptr<Buffers::BufferProcessingChain> get_processing_chain() override;
 
    /**
     * @brief Replace the buffer's processing chain
     * @param chain New processing chain to assign.
     * @param force If true, replaces existing chain even if one is set.
     */
    void set_processing_chain(const std::shared_ptr<BufferProcessingChain>& chain, bool force = false) override;
 
    bool has_data_for_cycle() const override { return m_has_data; }
    bool needs_removal() const override { return m_needs_removal; }
    void mark_for_processing(bool has_data) override { m_has_data = has_data; }
    void mark_for_removal() override { m_needs_removal = true; }
    void enforce_default_processing(bool should_process) override { m_process_default = should_process; }
    bool needs_default_processing() override { return m_process_default; }
 
    /**
     * @brief Try to acquire processing lock for this buffer
     * @return True if lock acquired, false if already processing.
     *
     * Uses an atomic flag to guard concurrent processing attempts.
     */
    inline bool try_acquire_processing() override
    {
        bool expected = false;
        return m_is_processing.compare_exchange_strong(expected, true,
            std::memory_order_acquire, std::memory_order_relaxed);
    }
 
    /**
     * @brief Release previously acquired processing lock
     */
    inline void release_processing() override
    {
        m_is_processing.store(false, std::memory_order_release);
    }
 
    /**
     * @brief Query whether the buffer is currently being processed
     * @return True if a processing operation holds the lock.
     */
    inline bool is_processing() const override
    {
        return m_is_processing.load(std::memory_order_acquire);
    }
 
    /** Get VkBuffer handle (VK_NULL_HANDLE if not registered) */
    vk::Buffer& get_buffer() { return m_resources.buffer; }
 
    /* Get logical buffer size as VkDeviceSize */
    vk::DeviceSize get_size_bytes() const { return m_size_bytes; }
 
    /** Check whether Vulkan handles are present (buffer registered) */
    bool is_initialized() const { return m_resources.buffer != VK_NULL_HANDLE; }
 
    /**
     * @brief Setup processors with a processing token
     * @param token ProcessingToken to assign.
     *
     * For VKBuffer this is a no-op as processors get the token.
     * This is meant for derived classes that need to setup default processors
     */
    virtual void setup_processors(ProcessingToken token) { }
 
    /** Get the buffer's semantic modality */
    Kakshya::DataModality get_modality() const { return m_modality; }
 
    /** Get the inferred data dimensions for the buffer contents */
    const std::vector<Kakshya::DataDimension>& get_dimensions() const { return m_dimensions; }
 
    /**
     * @brief Update the semantic modality and re-infer dimensions
     * @param modality New Kakshya::DataModality to apply.
     */
    void set_modality(Kakshya::DataModality modality);
 
    /** Retrieve the declared usage intent */
    Usage get_usage() const { return m_usage; }
 
    /** Set VkBuffer handle after backend allocation */
    void set_buffer(vk::Buffer buffer) { m_resources.buffer = buffer; }
 
    /** Set device memory handle after backend allocation */
    void set_memory(vk::DeviceMemory memory) { m_resources.memory = memory; }
 
    /** Set mapped host pointer (for host-visible allocations) */
    void set_mapped_ptr(void* ptr) { m_resources.mapped_ptr = ptr; }
 
    /** Set all buffer resources at once */
    inline void set_buffer_resources(const VKBufferResources& resources)
    {
        m_resources = resources;
    }
 
    /**
     * @brief Store raw index buffer handles produced by the geometry upload path.
     *
     * Called by GeometryBindingsProcessor after allocating and uploading index
     * data. Overwrites any previously stored handles. Pass null handles and
     * zero size to clear (non-indexed geometry).
     *
     * @param buf   Allocated vk::Buffer with INDEX usage flags.
     * @param mem   Backing vk::DeviceMemory.
     * @param size  Byte size of the index buffer.
     */
    void set_index_resources(vk::Buffer buf, vk::DeviceMemory mem, size_t size)
    {
        m_resources.index_buffer = buf;
        m_resources.index_memory = mem;
        m_resources.index_size_bytes = size;
    }
 
    /** Get all buffer resources at once */
    inline const VKBufferResources& get_buffer_resources() { return m_resources; }
 
    /**
     * @brief Return the raw index buffer handle.
     * @return vk::Buffer; operator bool() returns false when non-indexed.
     */
    [[nodiscard]] vk::Buffer get_index_buffer() const { return m_resources.index_buffer; }
 
    /**
     * @brief Number of bytes in the index buffer.
     * @return Byte count; divide by sizeof(uint32_t) for index count.
     *         Zero when non-indexed.
     */
    [[nodiscard]] size_t get_index_buffer_size() const { return m_resources.index_size_bytes; }
 
    /**
     * @brief True when an index buffer has been associated with this buffer.
     */
    [[nodiscard]] bool has_index_buffer() const
    {
        return static_cast<bool>(m_resources.index_buffer);
    }
 
    /**
     * @brief Whether this VKBuffer should be host-visible
     * @return True for staging or uniform buffers, false for device-local types.
     */
    bool is_host_visible() const
    {
        return m_usage == Usage::STAGING
            || m_usage == Usage::UNIFORM
            || m_usage == Usage::UNIFORM_BDA
            || m_usage == Usage::STORAGE_BDA
            || m_usage == Usage::HOST_STORAGE;
    }
 
    /**
     * @brief Get appropriate VkBufferUsageFlags for creation based on Usage
     * @return VkBufferUsageFlags to be used when creating VkBuffer.
     */
    vk::BufferUsageFlags get_usage_flags() const;
 
    /**
     * @brief Get appropriate VkMemoryPropertyFlags for allocation based on Usage
     * @return VkMemoryPropertyFlags to request during memory allocation.
     */
    vk::MemoryPropertyFlags get_memory_properties() const;
 
    /** Get mapped host pointer (nullptr if not host-visible or unmapped) */
    void* get_mapped_ptr() const { return m_resources.mapped_ptr; }
 
    /** Get device memory handle */
    void mark_dirty_range(size_t offset, size_t size);
 
    /** Mark a range as invalid (needs download) */
    void mark_invalid_range(size_t offset, size_t size);
 
    /** Retrieve and clear all dirty ranges */
    std::vector<std::pair<size_t, size_t>> get_and_clear_dirty_ranges();
 
    /** Retrieve and clear all invalid ranges */
    std::vector<std::pair<size_t, size_t>> get_and_clear_invalid_ranges();
 
    /**
     * @brief Associate this buffer with a window for rendering
     * @param window Target window for rendering this buffer's content
     *
     * When this buffer is processed, its content will be rendered to the associated window.
     * Currently supports one window per buffer (will be extended to multiple windows).
     */
    void set_pipeline_window(RenderPipelineID id, const std::shared_ptr<Core::Window>& window)
    {
        m_window_pipelines[id] = window;
    }
 
    /**
     * @brief Get the window associated with this buffer
     * @return Target window, or nullptr if not set
     */
    std::shared_ptr<Core::Window> get_pipeline_window(RenderPipelineID id) const
    {
        auto it = m_window_pipelines.find(id);
        if (it != m_window_pipelines.end()) {
            return it->second;
        }
        return nullptr;
    }
 
    /**
     * @brief Check if this buffer has a rendering pipeline configured
     */
    bool has_render_pipeline() const
    {
        return !m_window_pipelines.empty();
    }
 
    /**
     * @brief Get all render pipelines associated with this buffer
     * @return Map of RenderPipelineID to associated windows
     */
    std::unordered_map<RenderPipelineID, std::shared_ptr<Core::Window>> get_render_pipelines() const
    {
        return m_window_pipelines;
    }
 
    /**
     * @brief Store recorded command buffer for a pipeline
     */
    void set_pipeline_command(RenderPipelineID pipeline_id,
        CommandBufferID cmd_id)
    {
        m_pipeline_commands[pipeline_id] = cmd_id;
    }
 
    /**
     * @brief Get recorded command buffer for a pipeline
     */
    CommandBufferID get_pipeline_command(RenderPipelineID pipeline_id) const
    {
        auto it = m_pipeline_commands.find(pipeline_id);
        return it != m_pipeline_commands.end() ? it->second : 0;
    }
 
    /**
     * @brief Clear all recorded commands (called after presentation)
     */
    void clear_pipeline_commands()
    {
        m_pipeline_commands.clear();
    }
 
    /**
     * @brief Set vertex layout for this buffer
     *
     * Required before using buffer with graphics rendering.
     * Describes how to interpret buffer data as vertices.
     *
     * @param layout VertexLayout describing vertex structure
     */
    void set_vertex_layout(const Kakshya::VertexLayout& layout);
 
    /**
     * @brief Get vertex layout if set
     * @return Optional containing layout, or empty if not set
     */
    std::optional<Kakshya::VertexLayout> get_vertex_layout() const { return m_vertex_layout; }
 
    /**
     * @brief Check if this buffer has vertex layout configured
     */
    bool has_vertex_layout() const { return m_vertex_layout.has_value(); }
 
    /**
     * @brief Clear vertex layout
     */
    void clear_vertex_layout()
    {
        m_vertex_layout.reset();
    }
 
    std::shared_ptr<Buffer> clone_to(uint8_t dest_desc) override;
 
    /**
     * @brief Create a clone of this buffer with the same data and properties
     * @param usage Usage enum for the cloned buffer
     * @return Shared pointer to the cloned VKBuffer
     *
     * The cloned buffer will have the same size, modality, dimensions,
     * processing chain and default processor as the original. Changes to
     * one buffer after cloning do not affect the other.
     */
    std::shared_ptr<VKBuffer> clone_to(Usage usage);
 
    /** Set whether this buffer is for internal engine usage */
    void force_internal_usage(bool internal) override { m_internal_usage = internal; }
 
    /** Check whether this buffer is for internal engine usage */
    bool is_internal_only() const override { return m_internal_usage; }
 
    /** Access the pipeline context for custom metadata (non-const) */
    PipelineContext& get_pipeline_context() { return m_pipeline_context; }
 
    /** Access the pipeline context for custom metadata (const) */
    const PipelineContext& get_pipeline_context() const { return m_pipeline_context; }
 
    [[nodiscard]] EngineContext& get_engine_context() { return m_engine_context; }
    [[nodiscard]] const EngineContext& get_engine_context() const { return m_engine_context; }
 
    /**
     * @brief Mark config as changed (processors will detect and react)
     * @param is_dirty Whether the config is now dirty (default: true)
     * NOTE: Child classes override this to call their internal setup_rendering() with the new config, which may have additional side effects.
     */
    virtual void mark_render_config_dirty(bool is_dirty = true) { m_render_config_dirty = is_dirty; }
 
    /**
     * @brief Check if config has changed since last frame
     */
    [[nodiscard]] bool is_render_config_dirty() const { return m_render_config_dirty; }
 
    /**
     * @brief Get the current render configuration
     * @return RenderConfig struct with current settings
     */
    RenderConfig get_render_config() const { return m_render_config; }
 
    /**
     * @brief Update the render configuration and mark as dirty
     * @param config New RenderConfig to apply
     *
     * This will update the buffer's render configuration and set the dirty flag,
     * signaling to any RenderProcessor that it needs to reconfigure rendering for this buffer.
     * NOTE: Child classes override this to call their internal setup_rendering() with the new config, which may have additional side effects.
     */
    virtual void set_render_config(const RenderConfig& config)
    {
        m_render_config = config;
        m_render_config_dirty = true;
    }
 
    /**
     * @brief Mark this buffer as requiring depth testing when rendered
     */
    void set_needs_depth_attachment(bool needs) { m_needs_depth = needs; }
 
    /**
     * @brief Check if this buffer requires depth attachment for rendering
     */
    [[nodiscard]] bool needs_depth_attachment() const { return m_needs_depth; }
 
    /**
     * @brief Get a RenderProcessor suitable for rendering this buffer
     * @return Shared pointer to a RenderProcessor, or nullptr if not renderable
     *
     * By default returns nullptr. Derived classes that support rendering should
     * override this to return an appropriate RenderProcessor instance.
     */
    virtual std::shared_ptr<RenderProcessor> get_render_processor() const { return m_render_processor; }
 
    inline void set_render_processor(std::shared_ptr<RenderProcessor> rp) { m_render_processor = std::move(rp); }
 
protected:
    /**
     * @brief Called by derived classes to set their context-specific defaults
     * @param config RenderConfig with default values for this buffer type
     *
     * Example (GeometryBuffer calls this in constructor):
     *   RenderConfig defaults;
     *   defaults.vertex_shader = "point.vert.spv";
     *   defaults.fragment_shader = "point.frag.spv";
     *   defaults.topology = PrimitiveTopology::POINT_LIST;
     *   set_default_render_config(defaults);
     */
    void set_default_render_config(const RenderConfig& config)
    {
        m_render_config = config;
        m_render_config_dirty = false;
    }
 
    /**
     * @brief Configure the internal m_render_processor from a RenderConfig.
     */
    void apply_render_config(const RenderConfig& config, const ShaderConfig& shader_config);
 
    /**
     * @brief Configure a RenderProcessor, creating one if null.
     * @param render_processor Existing processor to configure, or nullptr to create one.
     * @param config RenderConfig with settings to apply
     */
    void apply_render_config(
        std::shared_ptr<RenderProcessor>& render_processor,
        const RenderConfig& config,
        const ShaderConfig& shader_config);
 
    bool m_render_config_dirty {};
    RenderConfig m_render_config;
    std::shared_ptr<RenderProcessor> m_render_processor;
 
private:
    VKBufferResources m_resources;
 
    // Buffer parameters
    size_t m_size_bytes {};
    Usage m_usage {};
    bool m_needs_depth {};
 
    // Semantic metadata
    Kakshya::DataModality m_modality;
    std::vector<Kakshya::DataDimension> m_dimensions;
 
    std::optional<Kakshya::VertexLayout> m_vertex_layout;
 
    // Buffer interface state
    bool m_has_data { true };
    bool m_needs_removal {};
    bool m_process_default { true };
    bool m_internal_usage {};
    std::atomic<bool> m_is_processing;
    std::shared_ptr<Buffers::BufferProcessor> m_default_processor;
    std::shared_ptr<Buffers::BufferProcessingChain> m_processing_chain;
    ProcessingToken m_processing_token;
    PipelineContext m_pipeline_context;
    EngineContext m_engine_context;
 
    std::unordered_map<RenderPipelineID, std::shared_ptr<Core::Window>> m_window_pipelines;
    std::unordered_map<RenderPipelineID, CommandBufferID> m_pipeline_commands;
 
    std::vector<std::pair<size_t, size_t>> m_dirty_ranges;
    std::vector<std::pair<size_t, size_t>> m_invalid_ranges;
 
    /**
     * @brief Infer Kakshya::DataDimension entries from a given byte count
     *
     * Uses the current modality and provided byte count to populate
     * m_dimensions so processors and UI code can reason about the buffer's layout.
     *
     * @param byte_count Number of bytes of data to infer dimensions from.
     */
    void infer_dimensions_from_data(size_t byte_count);
};
 
class VKBufferProcessor : public BufferProcessor {
protected:
    Registry::Service::BufferService* m_buffer_service = nullptr;
    Registry::Service::ComputeService* m_compute_service = nullptr;
 
    void initialize_buffer_service();
    void initialize_compute_service();
    void ensure_initialized(const std::shared_ptr<VKBuffer>& buffer);
};
 
} // namespace MayaFlux::Buffers