BackendResoureManager.hpp

Go to the documentation of this file.

#pragma once
 
#include "vulkan/vulkan.hpp"
 
namespace MayaFlux::Registry::Service {
struct BufferService;
}
 
namespace MayaFlux::Buffers {
class VKBuffer;
}
 
namespace MayaFlux::Core {
 
class VKContext;
class VKImage;
class VKCommandManager;
 
/**
 * @class BackendResourceManager
 * @brief Manages Vulkan resources (buffers, images, samplers) for the graphics backend
 */
class MAYAFLUX_API BackendResourceManager {
public:
    BackendResourceManager(VKContext& context, VKCommandManager& command_manager);
    ~BackendResourceManager() = default;
 
    void setup_backend_service(const std::shared_ptr<Registry::Service::BufferService>& buffer_service);
 
    // ========================================================================
    // Buffer management
    // ========================================================================
 
    /**
     * @brief Initialize a buffer for use with the graphics backend
     * @param buffer Shared pointer to the buffer to initialize
     */
    void initialize_buffer(const std::shared_ptr<Buffers::VKBuffer>& buffer);
 
    /**
     * @brief Allocate a raw VkBuffer/VkDeviceMemory pair without an owning
     *        VKBuffer object.
     * @param size_bytes Buffer capacity in bytes.
     * @param usage Vulkan buffer usage flags.
     * @param memory_properties Vulkan memory property flags.
     * @param host_visible When true, the allocation is mapped and out_mapped_ptr
     *        is populated; when false, out_mapped_ptr is left null.
     * @param out_buffer Receives the created vk::Buffer.
     * @param out_memory Receives the bound vk::DeviceMemory.
     * @param out_mapped_ptr Receives the mapped host pointer, or nullptr if
     *        @p host_visible is false.
     *
     * Same allocation sequence as initialize_buffer() (createBuffer,
     * getBufferMemoryRequirements, allocateMemory, bindBufferMemory, optional
     * mapMemory), without requiring or populating a VKBuffer. Intended for
     * VKBuffer subclasses that own secondary raw handle pairs directly in
     * VKBufferResources::back_buffers rather than as separate VKBuffer objects.
     */
    void allocate_raw_buffer(
        size_t size_bytes,
        vk::BufferUsageFlags usage,
        vk::MemoryPropertyFlags memory_properties,
        bool host_visible,
        vk::Buffer& out_buffer,
        vk::DeviceMemory& out_memory,
        void*& out_mapped_ptr);
 
    /**
     * @brief Cleanup a buffer and release associated resources
     * @param buffer Shared pointer to the buffer to cleanup
     */
    void cleanup_buffer(const std::shared_ptr<Buffers::VKBuffer>& buffer);
 
    /**
     * @brief Query the Vulkan device address of an initialized BDA-capable buffer
     * @param buffer Buffer with Usage::UNIFORM_BDA or Usage::STORAGE_BDA
     * @return vk::DeviceAddress, or 0 if the buffer was not created with eShaderDeviceAddress
     */
    [[nodiscard]] vk::DeviceAddress get_buffer_device_address(
        const std::shared_ptr<Buffers::VKBuffer>& buffer) const;
 
    /**
     * @brief Flush any pending buffer operations (e.g., uploads/downloads)
     */
    void flush_pending_buffer_operations();
 
    // ========================================================================
    // Buffer management
    // ========================================================================
 
    /**
     * @brief Initialize a VKImage (allocate VkImage, memory, and create image view)
     * @param image VKImage to initialize
     *
     * Follows the same pattern as initialize_buffer:
     * 1. Create VkImage
     * 2. Allocate VkDeviceMemory
     * 3. Bind memory to image
     * 4. Create VkImageView
     * 5. Store handles in VKImage
     */
    void initialize_image(const std::shared_ptr<VKImage>& image);
 
    /**
     * @brief Cleanup a VKImage (destroy view, image, and free memory)
     * @param image VKImage to cleanup
     */
    void cleanup_image(const std::shared_ptr<VKImage>& image);
 
    /**
     * @brief Transition image layout using a pipeline barrier
     * @param image VkImage handle
     * @param old_layout Current layout
     * @param new_layout Target layout
     * @param mip_levels Number of mip levels to transition
     * @param array_layers Number of array layers to transition
     *
     * Executes immediately on graphics queue. Use for initial setup and
     * one-off transitions. For rendering, prefer manual barriers.
     */
    void transition_image_layout(
        vk::Image image,
        vk::ImageLayout old_layout,
        vk::ImageLayout new_layout,
        uint32_t mip_levels = 1,
        uint32_t array_layers = 1,
        vk::ImageAspectFlags aspect_flags = vk::ImageAspectFlagBits::eColor);
 
    /**
     * @brief Upload data to an image (creates staging buffer internally)
     * @param image Target VKImage
     * @param data Pixel data pointer
     * @param size Data size in bytes
     */
    void upload_image_data(
        std::shared_ptr<VKImage> image,
        const void* data,
        size_t size);
 
    /**
     * @brief Upload image data using a caller-supplied persistent staging buffer.
     *        Identical to upload_image_data() but skips the per-call VkBuffer
     *        allocation. The staging buffer must be host-visible and at least
     *        @p size bytes. Intended for high-frequency streaming uploads.
     * @param image   Target VKImage.
     * @param data    Source pixel data pointer.
     * @param size    Byte count.
     * @param staging Pre-allocated host-visible staging buffer.
     * @param deferred If true, command recording will be deferred and must be flushed
     */
    void upload_image_data(
        std::shared_ptr<VKImage> image,
        const void* data,
        size_t size,
        const std::shared_ptr<Buffers::VKBuffer>& staging, bool deferred = false);
 
    /**
     * @brief Download image data to a host pointer.
     *
     * When @p staging is null, allocates and destroys a per-call staging
     * buffer internally. When @p staging is supplied, it is used as-is and
     * not destroyed after — the caller owns its lifetime.
     *
     * When @p deferred is false (default), submits under a fence and blocks
     * the calling thread on vkWaitForFences. This is non-blocking relative
     * to the graphics queue (waitIdle is never called).
     * When @p deferred is true and staging is supplied, records commands for
     * deferred submission; the caller is responsible for flushing.
     *
     * @param image          Source image.
     * @param data           Destination host pointer, at least @p size bytes.
     * @param size           Byte count to read.
     * @param staging        Persistent host-visible buffer, or nullptr for
     *                       per-call allocation.
     * @param deferred       Record for deferred submission (requires staging).
     * @param restore_layout Layout to restore after the copy.
     * @param restore_stage  Pipeline stage that consumes the image after restore.
     */
    void download_image_data(
        std::shared_ptr<VKImage> image,
        void* data,
        size_t size,
        const std::shared_ptr<Buffers::VKBuffer>& staging = nullptr,
        bool deferred = false,
        vk::ImageLayout restore_layout = vk::ImageLayout::eShaderReadOnlyOptimal,
        vk::PipelineStageFlags restore_stage = vk::PipelineStageFlagBits::eFragmentShader);
 
    // ========================================================================
    // Sampler management
    // ========================================================================
 
    /**
     * @brief Create sampler
     * @param filter Mag/min filter
     * @param address_mode Texture address mode (wrap, clamp, etc.)
     * @param anisotropy Max anisotropy (0 = disabled)
     * @return Sampler handle
     */
    vk::Sampler create_sampler(
        vk::Filter filter = vk::Filter::eLinear,
        vk::SamplerAddressMode address_mode = vk::SamplerAddressMode::eRepeat,
        float max_anisotropy = 0.0F);
 
    /**
     * @brief Destroy sampler
     */
    void destroy_sampler(vk::Sampler sampler);
 
    // ========================================================================
    // Memory management
    // ========================================================================
 
    /**
     * @brief Find a suitable memory type for Vulkan buffer allocation
     * @param type_filter Memory type bits filter
     * @param properties Desired memory property flags
     * @return Index of the suitable memory type
     */
    uint32_t find_memory_type(uint32_t type_filter, vk::MemoryPropertyFlags properties) const;
 
    // ========================================================================
    // Command management
    // ========================================================================
 
    /**
     * @brief Execute immediate command recording for buffer operations
     * @param recorder Command recording function
     */
    void execute_immediate_commands(const std::function<void(vk::CommandBuffer)>& recorder);
 
    /**
     * @brief Record deferred command recording for buffer operations
     * @param recorder Command recording function
     */
    void record_deferred_commands(const std::function<void(vk::CommandBuffer)>& recorder);
 
    /**
     * @brief Flush deferred commands and return a semaphore that signals when they are complete
     * @return Semaphore that will be signaled when deferred commands are finished
     */
    vk::Semaphore flush_deferred_commands();
 
    // ========================================================================
    // Cleanup
    // ========================================================================
 
    void cleanup();
 
private:
    VKContext& m_context;
    VKCommandManager& m_command_manager;
 
    std::vector<std::shared_ptr<Buffers::VKBuffer>> m_managed_buffers;
    std::unordered_map<size_t, vk::Sampler> m_sampler_cache;
 
    vk::Semaphore m_deferred_semaphore {};
 
    size_t compute_sampler_hash(vk::Filter filter, vk::SamplerAddressMode address_mode, float max_anisotropy) const;
};
 
} // namespace MayaFlux::Core