ShaderFoundry.hpp

Go to the documentation of this file.

#pragma once
 
#include "ShaderUtils.hpp"
 
#include "ShaderSpec.hpp"
 
#include "MayaFlux/Transitive/Memory/RingBuffer.hpp"
 
#include <span>
 
namespace MayaFlux::Core {
class VulkanBackend;
class VKShaderModule;
class VKComputePipeline;
class VKDescriptorManager;
}
 
namespace MayaFlux::Portal::Graphics {
 
class ComputePress;
class RenderFlow;
 
/**
 * @struct DescriptorBufferWrite
 * @brief One descriptor set slot to update, for ShaderFoundry::update_descriptor_buffers.
 *
 * Same fields as update_descriptor_buffer's parameter list, collected so
 * several slots, across one or more descriptor sets, can be written with a
 * single vkUpdateDescriptorSets call instead of one call per slot.
 */
struct DescriptorBufferWrite {
    DescriptorSetID descriptor_set_id;
    uint32_t binding;
    vk::DescriptorType type;
    vk::Buffer buffer;
    size_t offset;
    size_t size;
};
 
/**
 * @struct DescriptorImageWrite
 * @brief One descriptor set slot to update, for ShaderFoundry::update_descriptor_images.
 *
 * Covers both update_descriptor_image (combined image sampler) and
 * update_descriptor_storage_image (storage image) shapes; sampler is
 * ignored for a storage image write. Collected the same way as
 * DescriptorBufferWrite, so several image slots can be written with a
 * single vkUpdateDescriptorSets call instead of one call per slot.
 */
struct DescriptorImageWrite {
    DescriptorSetID descriptor_set_id;
    uint32_t binding;
    vk::DescriptorType type;
    vk::ImageView image_view;
    vk::Sampler sampler;
    vk::ImageLayout layout;
};
 
namespace detail {
    /**
     * @brief Emit complete SPIR-V assembly text for a generated compute kernel.
     * @param spec ShaderSpec produced by ShaderSpec::Assemble::build().
     * @return SPIR-V assembly string suitable for spvTextToBinary.
     */
    MAYAFLUX_API std::string emit_spirv_asm(const ShaderSpec& spec);
 
    /**
     * Emit a complete GLSL compute shader from spec metadata and a KernelSource body.
     *
     * Binding names are used directly as GLSL identifiers. The body text from
     * KernelSource::parse() is injected verbatim inside main() after uint i and
     * optional ivec2 coord are declared. Push constant fields are accessible as
     * pc.<name>. SSBO arrays are accessible as <name>[i].
     */
    MAYAFLUX_API std::string emit_glsl_kernel(const ShaderSpec& spec);
}
 
/**
 * @class ShaderFoundry
 * @brief Portal-level shader compilation and caching
 *
 * ShaderFoundry is a thin glue layer that:
 * - Wraps Core::VKShaderModule for convenient shader creation
 * - Provides caching to avoid redundant compilation
 * - Supports hot-reload workflows (watch files, recompile)
 * - Returns VKShaderModule directly for use in pipelines
 *
 * Design Philosophy:
 * - Manages compilation, NOT execution (that's Pipeline/Compute)
 * - Returns VKShaderModule directly (no wrapping)
 * - Simple, focused API aligned with VKShaderModule capabilities
 * - Integrates with existing Core shader infrastructure
 *
 * Consumers:
 * - VKBufferProcessor subclasses (compute shaders)
 * - Future Yantra::ComputePipeline (compute shaders)
 * - Future Yantra::RenderPipeline (graphics shaders)
 * - Future DataProcessors (image processing shaders)
 *
 * Usage:
 *   auto& compiler = Portal::Graphics::ShaderFoundry::instance();
 *
 *   // Compile from file
 *   auto shader = compiler.compile_from_file("shaders/my_kernel.comp");
 *
 *   // Compile from string
 *   auto shader = compiler.compile_from_source(glsl_code, ShaderStage::COMPUTE);
 *
 *   // Use in pipeline
 *   my_buffer_processor->set_shader(shader);
 *   my_compute_pipeline->set_shader(shader);
 */
class MAYAFLUX_API ShaderFoundry {
public:
    enum class CommandBufferType : uint8_t {
        GRAPHICS,
        COMPUTE,
        TRANSFER
    };
 
    enum class CommandBufferLevel : uint8_t {
        PRIMARY,
        SECONDARY
    };
 
private:
    struct DescriptorSetState {
        vk::DescriptorSet descriptor_set;
    };
 
    struct CommandBufferState {
        vk::CommandBuffer cmd;
        CommandBufferType type;
        CommandBufferLevel level { CommandBufferLevel::PRIMARY };
        bool is_active;
        vk::QueryPool timestamp_pool;
        std::unordered_map<std::string, uint32_t> timestamp_queries;
    };
 
    struct FenceState {
        vk::Fence fence;
        bool signaled;
 
        CommandBufferID cmd_id { INVALID_COMMAND_BUFFER };
    };
 
    struct SemaphoreState {
        vk::Semaphore semaphore;
    };
 
    struct ShaderState {
        std::shared_ptr<Core::VKShaderModule> module;
        std::string filepath;
        ShaderStage stage;
        std::string entry_point;
    };
 
public:
    static ShaderFoundry& instance()
    {
        static ShaderFoundry compiler;
        return compiler;
    }
 
    ShaderFoundry(const ShaderFoundry&) = delete;
    ShaderFoundry& operator=(const ShaderFoundry&) = delete;
    ShaderFoundry(ShaderFoundry&&) noexcept = delete;
    ShaderFoundry& operator=(ShaderFoundry&&) noexcept = delete;
 
    /**
     * @brief Initialize shader compiler
     * @param backend Shared pointer to VulkanBackend
     * @param config Compiler configuration
     * @return True if initialization succeeded
     *
     * Must be called before compiling any shaders.
     */
    bool initialize(
        const std::shared_ptr<Core::VulkanBackend>& backend,
        const ShaderCompilerConfig& config = {});
 
    /**
     * @brief Stop active command recording and free command buffers
     *
     * Frees all command buffers back to pool and destroys query pools.
     * Call this BEFORE destroying pipelines/resources that command buffers reference.
     * Does NOT destroy the command pool itself - that happens in shutdown().
     */
    void stop();
 
    /**
     * @brief Shutdown and cleanup all ShaderFoundry resources
     *
     * Destroys sync objects, descriptor resources, and shader modules.
     * Must be called AFTER stop() and AFTER pipeline consumers (RenderFlow/ComputePress) shutdown.
     */
    void shutdown();
 
    /**
     * @brief Check if compiler is initialized
     */
    [[nodiscard]] bool is_initialized() const { return m_backend != nullptr; }
 
    //==========================================================================
    // Shader Compilation - Primary API
    //==========================================================================
 
    /**
     * @brief Universal shader loader - auto-detects source type
     * @param content File path, GLSL source string, or SPIR-V path
     * @param stage Optional stage override (auto-detected if omitted)
     * @param entry_point Entry point function name (default: "main")
     * @return ShaderID, or INVALID_SHADER on failure
     *
     * Supports:
     * - GLSL files: .comp, .vert, .frag, .geom, .tesc, .tese
     * - SPIR-V files: .spv
     *
     * Stage auto-detection:
     *   .comp → COMPUTE
     *   .vert → VERTEX
     *   .frag → FRAGMENT
     *   .geom → GEOMETRY
     *   .tesc → TESS_CONTROL
     *   .tese → TESS_EVALUATION
     *   .mesh → MESH
     *   .task → TASK
     *
     * Examples:
     *   load_shader("shaders/kernel.comp");              // File
     *   load_shader("shaders/kernel.spv", COMPUTE);      // SPIR-V
     *   load_shader("#version 460\nvoid main(){}", COMPUTE); // Source
     */
    ShaderID load_shader(
        const std::string& content,
        std::optional<ShaderStage> stage = std::nullopt,
        const std::string& entry_point = "main");
 
    /**
     * @brief Load shader from explicit ShaderSource descriptor
     * @param source Complete shader source specification
     * @return ShaderID, or INVALID_SHADER on failure
     */
    ShaderID load_shader(const ShaderSource& source);
 
    /**
     * @brief Compile and cache a compute shader from a declarative ShaderSpec.
     *
     * Emits SPIR-V assembly text from the spec via detail::emit_spirv_asm(),
     * assembles it via VKShaderModule::create_from_spirv_asm() using
     * SPIRV-Tools, and caches the result by content hash. No shaderc or
     * GLSL toolchain is involved.
     *
     * Subsequent calls with a spec that produces identical assembly are cache
     * hits and return the existing ShaderID at no cost.
     *
     * The emitted shader is always a compute stage. Entry point is "main".
     * Binding 0 at set 0 is never emitted; it is engine-reserved.
     *
     * @param spec ShaderSpec produced by ShaderSpec::Assemble::build().
     * @return ShaderID on success, INVALID_SHADER on compilation failure.
     */
    ShaderID load_shader(const ShaderSpec& spec);
 
    /**
     * @brief Hot-reload shader (returns new ID)
     */
    ShaderID reload_shader(const std::string& filepath);
 
    /**
     * @brief Destroy shader (cleanup internal state)
     */
    void destroy_shader(ShaderID shader_id);
 
    /**
     * @brief Compile shader from ShaderSource descriptor
     * @param shader_source Shader descriptor (path or source + type + stage)
     * @return Compiled shader module, or nullptr on failure
     *
     * Unified interface that dispatches to appropriate compile method.
     * Useful for abstracted shader loading pipelines.
     */
    std::shared_ptr<Core::VKShaderModule> compile(const ShaderSource& shader_source);
 
    //==========================================================================
    // Shader Introspection
    //==========================================================================
 
    /**
     * @brief Get reflection info for compiled shader
     * @param shader_id ID of compiled shader
     * @return Reflection information
     *
     * Extracted during compilation if enabled in config.
     * Includes descriptor bindings, push constant ranges, workgroup size, etc.
     */
    ShaderReflectionInfo get_shader_reflection(ShaderID shader_id);
 
    /**
     * @brief Get the descriptor bindings a shader's entry point statically uses
     * @param shader_id ID of compiled shader
     * @return Used bindings only, excluding declared but unused resources
     */
    std::vector<DescriptorBindingInfo> get_active_descriptor_bindings(ShaderID shader_id);
 
    /**
     * @brief Get shader stage for compiled shader
     * @param shader_id ID of compiled shader
     * @return Shader stage (COMPUTE, VERTEX, FRAGMENT, etc.)
     */
    ShaderStage get_shader_stage(ShaderID shader_id);
 
    /**
     * @brief Get entry point name for compiled shader
     * @param shader_id ID of compiled shader
     * @return Entry point function name
     */
    std::string get_shader_entry_point(ShaderID shader_id);
 
    //==========================================================================
    // Hot-Reload Support
    //==========================================================================
 
    /**
     * @brief Invalidate cache for specific shader
     * @param cache_key File path or cache key
     *
     * Forces next compilation to recompile from source.
     * Useful for hot-reload workflows.
     */
    void invalidate_cache(const std::string& cache_key);
 
    /**
     * @brief Invalidate entire shader cache
     *
     * Forces all subsequent compilations to recompile.
     * Does NOT destroy existing shader modules (they remain valid).
     */
    void clear_cache();
 
    /**
     * @brief Hot-reload a shader from file
     * @param filepath Path to shader file
     * @return Recompiled shader module, or nullptr on failure
     *
     * Convenience method: invalidate_cache() + compile_from_file().
     * Returns new shader module; consumers must update references.
     */
    std::shared_ptr<Core::VKShaderModule> hot_reload(const std::string& filepath);
 
    //==========================================================================
    // Configuration
    //==========================================================================
 
    /**
     * @brief Update compiler configuration
     * @param config New configuration
     *
     * Affects future compilations.
     * Does NOT recompile existing shaders.
     */
    void set_config(const ShaderCompilerConfig& config);
 
    /**
     * @brief Get current compiler configuration
     */
    [[nodiscard]] const ShaderCompilerConfig& get_config() const { return m_config; }
 
    /**
     * @brief Add include directory for shader compilation
     * @param directory Path to directory containing shader includes
     *
     * Used for #include resolution in GLSL files.
     */
    void add_include_directory(const std::string& directory);
 
    /**
     * @brief Add preprocessor define for shader compilation
     * @param name Macro name
     * @param value Macro value (optional)
     *
     * Example: define("DEBUG", "1") → #define DEBUG 1
     */
    void add_define(const std::string& name, const std::string& value = "");
 
    //==========================================================================
    // Introspection
    //==========================================================================
 
    /**
     * @brief Check if shader is cached
     * @param cache_key File path or cache key
     */
    [[nodiscard]] bool is_cached(const std::string& cache_key) const;
 
    /**
     * @brief Get all cached shader keys
     */
    [[nodiscard]] std::vector<std::string> get_cached_keys() const;
 
    /**
     * @brief Get number of cached shaders
     */
    [[nodiscard]] size_t get_cache_size() const { return m_shader_cache.size(); }
 
    //==========================================================================
    // Descriptor Set Management - ShaderFoundry allocates and tracks
    //==========================================================================
 
    /**
     * @brief Allocate descriptor set for a pipeline
     * @param pipeline_id Which pipeline this is for
     * @param set_index Which descriptor set (0, 1, 2...)
     * @return Descriptor set ID
     */
    DescriptorSetID allocate_descriptor_set(vk::DescriptorSetLayout layout);
 
    /**
     * @brief Update descriptor set with buffer binding
     * @param descriptor_set_id ID of descriptor set to update
     * @param binding Binding index within the descriptor set
     * @param type Descriptor type (e.g., eStorageBuffer, eUniformBuffer)
     * @param buffer Vulkan buffer to bind
     * @param offset Offset within the buffer
     * @param size Size of the buffer region
     */
    void update_descriptor_buffer(
        DescriptorSetID descriptor_set_id,
        uint32_t binding,
        vk::DescriptorType type,
        vk::Buffer buffer,
        size_t offset,
        size_t size);
 
    /**
     * @brief Update several buffer descriptor slots in one Vulkan call.
     *
     * Equivalent to calling update_descriptor_buffer once per entry, except
     * every entry is submitted through a single vkUpdateDescriptorSets call
     * instead of one call per entry. Entries may target different descriptor
     * sets; order among them does not matter since each names a distinct
     * (descriptor set, binding). An entry naming an unknown descriptor_set_id
     * is skipped, matching update_descriptor_buffer's behaviour for the same
     * case. A call with an empty span does nothing.
     *
     * @param writes Slots to update.
     */
    void update_descriptor_buffers(std::span<const DescriptorBufferWrite> writes);
 
    /**
     * @brief Update descriptor set with image binding
     * @param descriptor_set_id ID of descriptor set to update
     * @param binding Binding index within the descriptor set
     * @param image_view Vulkan image view to bind
     * @param sampler Vulkan sampler to bind
     * @param layout Image layout (default: eShaderReadOnlyOptimal)
     * @param array_element Array index for array bindings (default: 0)
     */
    void update_descriptor_image(
        DescriptorSetID descriptor_set_id,
        uint32_t binding,
        vk::ImageView image_view,
        vk::Sampler sampler,
        vk::ImageLayout layout = vk::ImageLayout::eShaderReadOnlyOptimal,
        uint32_t array_element = 0);
 
    /**
     * @brief Update descriptor set with storage image binding
     * @param descriptor_set_id ID of descriptor set to update
     * @param binding Binding index within the descriptor set
     * @param image_view Vulkan image view to bind
     * @param layout Image layout (default: eGeneral)
     */
    void update_descriptor_storage_image(
        DescriptorSetID descriptor_set_id,
        uint32_t binding,
        vk::ImageView image_view,
        vk::ImageLayout layout = vk::ImageLayout::eGeneral);
 
    /**
     * @brief Update several image descriptor slots in one Vulkan call.
     *
     * Equivalent to calling update_descriptor_image or
     * update_descriptor_storage_image once per entry, except every entry is
     * submitted through a single vkUpdateDescriptorSets call. An entry
     * naming an unknown descriptor_set_id is skipped. A call with an empty
     * span does nothing.
     *
     * @param writes Slots to update.
     */
    void update_descriptor_images(std::span<const DescriptorImageWrite> writes);
 
    /**
     * @brief Get Vulkan descriptor set handle from DescriptorSetID
     * @param descriptor_set_id Descriptor set ID
     * @return Vulkan descriptor set handle
     */
    vk::DescriptorSet get_descriptor_set(DescriptorSetID descriptor_set_id);
 
    //==========================================================================
    // Command Recording - ShaderFoundry manages command buffers
    //==========================================================================
 
    /**
     * @brief Begin recording command buffer
     * @param type Command buffer type (GRAPHICS, COMPUTE, TRANSFER)
     * @return Command buffer ID
     */
    CommandBufferID begin_commands(CommandBufferType type);
 
    /**
     * @brief Begin recording a secondary command buffer for dynamic rendering
     * @param color_format Format of the color attachment (from swapchain)
     * @return Command buffer ID
     *
     * With dynamic rendering, secondary buffers don't need render pass objects.
     * They only need to know the attachment formats they'll render to.
     */
    CommandBufferID begin_secondary_commands(
        vk::Format color_format,
        vk::Format depth_format = vk::Format::eUndefined);
 
    /**
     * @brief Get Vulkan command buffer handle from CommandBufferID
     * @param cmd_id Command buffer ID
     */
    vk::CommandBuffer get_command_buffer(CommandBufferID cmd_id);
 
    /**
     * @brief End recording command buffer
     * @param cmd_id Command buffer ID to end
     * @return True if successful, false if invalid ID or not active
     */
    bool end_commands(CommandBufferID cmd_id);
 
    /**
     * @brief Free all allocated command buffers
     */
    void free_all_command_buffers();
 
    //==========================================================================
    // Memory Barriers and Synchronization
    //==========================================================================
 
    /**
     * @brief Submit command buffer and wait for completion
     * @param cmd_id Command buffer ID to submit
     */
    void submit_and_wait(CommandBufferID cmd_id);
 
    /**
     * @brief Submit command buffer asynchronously, returning a fence
     * @param cmd_id Command buffer ID to submit
     * @return Fence ID to wait on later
     */
    FenceID submit_async(CommandBufferID cmd_id);
 
    /**
     * @brief Submit command buffer asynchronously, returning a semaphore
     * @param cmd_id Command buffer ID to submit
     * @return Semaphore ID to wait on later
     */
    SemaphoreID submit_with_signal(CommandBufferID cmd_id);
 
    /**
     * @brief Wait for fence to be signaled
     * @param fence_id Fence ID to wait on
     */
    void wait_for_fence(FenceID fence_id);
 
    /**
     * @brief Wait for multiple fences to be signaled
     * @param fence_ids Vector of fence IDs to wait on
     */
    void wait_for_fences(const std::vector<FenceID>& fence_ids);
 
    /**
     * @brief Destroy the fence and free its associated command buffer.
     *
     * Must be called once per FenceID returned by submit_async, after
     * is_fence_signaled returns true. Failure to call this leaks a VkFence
     * and a command buffer slot every dispatch cycle.
     *
     * Safe to call with INVALID_FENCE.
     *
     * @param fence_id FenceID returned by submit_async.
     */
    void release_fence(FenceID fence_id);
 
    /**
     * @brief Check if fence is signaled
     * @param fence_id Fence ID to check
     * @return True if fence is signaled, false otherwise
     */
    bool is_fence_signaled(FenceID fence_id);
 
    /**
     * @brief Begin command buffer that waits on a semaphore
     * @param type Command buffer type (GRAPHICS, COMPUTE, TRANSFER)
     * @param wait_semaphore Semaphore ID to wait on
     * @param wait_stage Pipeline stage to wait at
     * @return Command buffer ID
     */
    CommandBufferID begin_commands_with_wait(
        CommandBufferType type,
        SemaphoreID wait_semaphore,
        vk::PipelineStageFlags wait_stage);
 
    /**
     * @brief Get Vulkan fence handle from FenceID
     * @param fence_id Fence ID
     */
    vk::Semaphore get_semaphore_handle(SemaphoreID semaphore_id);
 
    /**
     * @brief Insert buffer memory barrier
     */
    void buffer_barrier(
        CommandBufferID cmd_id,
        vk::Buffer buffer,
        vk::AccessFlags src_access,
        vk::AccessFlags dst_access,
        vk::PipelineStageFlags src_stage,
        vk::PipelineStageFlags dst_stage);
 
    /**
     * @brief Insert image memory barrier
     */
    void image_barrier(
        CommandBufferID cmd_id,
        vk::Image image,
        vk::ImageLayout old_layout,
        vk::ImageLayout new_layout,
        vk::AccessFlags src_access,
        vk::AccessFlags dst_access,
        vk::PipelineStageFlags src_stage,
        vk::PipelineStageFlags dst_stage);
 
    //==========================================================================
    // Queue Management
    //==========================================================================
 
    /**
     * @brief Set Vulkan queues for command submission
     */
    void set_graphics_queue(vk::Queue queue);
 
    /**
     * @brief Set Vulkan queues for command submission
     */
    void set_compute_queue(vk::Queue queue);
 
    /**
     * @brief Set Vulkan queues for command submission
     */
    void set_transfer_queue(vk::Queue queue);
 
    /**
     * @brief Get Vulkan graphics queue
     */
    [[nodiscard]] vk::Queue get_graphics_queue() const;
 
    /**
     * @brief Get Vulkan compute queue
     */
    [[nodiscard]] vk::Queue get_compute_queue() const;
 
    /**
     * @brief Get Vulkan transfer queue
     */
    [[nodiscard]] vk::Queue get_transfer_queue() const;
 
    //==========================================================================
    // Profiling
    //==========================================================================
 
    void begin_timestamp(CommandBufferID cmd_id, const std::string& label = "");
    void end_timestamp(CommandBufferID cmd_id, const std::string& label = "");
 
    struct TimestampResult {
        std::string label;
        uint64_t duration_ns;
        bool valid;
    };
 
    TimestampResult get_timestamp_result(CommandBufferID cmd_id, const std::string& label);
 
    //==========================================================================
    // Utilities
    //==========================================================================
 
    /**
     * @brief Convert Portal ShaderStage to Vulkan ShaderStageFlagBits
     */
    static vk::ShaderStageFlagBits to_vulkan_stage(ShaderStage stage);
 
    /**
     * @brief Auto-detect shader stage from file extension
     * @param filepath Path to shader file
     * @return Detected stage, or nullopt if unknown
     *
     * Delegates to VKShaderModule::detect_stage_from_extension().
     */
    static std::optional<ShaderStage> detect_stage_from_extension(const std::string& filepath);
 
    //==========================================================================
    // Device Access
    //==========================================================================
 
    /**
     * @brief Get logical device handle
     */
    [[nodiscard]] vk::Device get_device() const;
 
    /**
     * @brief Get physical device handle
     *
     * Required by consumers that allocate raw Vulkan buffers directly
     * (e.g. GpuComputeOperation) and need memory type selection via
     * vk::PhysicalDevice::getMemoryProperties().
     */
    [[nodiscard]] vk::PhysicalDevice get_physical_device() const;
 
private:
    /**
     * @enum DetectedSourceType
     * @brief Internal enum for source type detection
     */
    enum class DetectedSourceType : uint8_t {
        FILE_GLSL,
        FILE_SPIRV,
        SOURCE_STRING,
        UNKNOWN
    };
 
    ShaderFoundry() = default;
    ~ShaderFoundry() { shutdown(); }
 
    std::shared_ptr<Core::VulkanBackend> m_backend;
    ShaderCompilerConfig m_config;
 
    std::unordered_map<std::string, std::shared_ptr<Core::VKShaderModule>> m_shader_cache;
    std::unordered_map<ShaderID, ShaderState> m_shaders;
    std::unordered_map<std::string, ShaderID> m_shader_filepath_cache;
 
    std::shared_ptr<Core::VKDescriptorManager> m_global_descriptor_manager;
    std::unordered_map<DescriptorSetID, DescriptorSetState> m_descriptor_sets;
 
    std::unordered_map<CommandBufferID, CommandBufferState> m_command_buffers;
 
    struct FinishedTimestamp {
        CommandBufferID cmd_id { INVALID_COMMAND_BUFFER };
        TimestampResult result {};
    };
 
    static constexpr size_t k_finished_timestamp_capacity = 1024;
    Memory::HistoryBuffer<FinishedTimestamp> m_finished_timestamps { k_finished_timestamp_capacity };
 
    std::unordered_map<FenceID, FenceState> m_fences;
    std::unordered_map<SemaphoreID, SemaphoreState> m_semaphores;
 
    vk::Queue m_graphics_queue;
    vk::Queue m_compute_queue;
    vk::Queue m_transfer_queue;
 
    std::atomic<uint64_t> m_next_shader_id { 1 };
    std::atomic<uint64_t> m_next_descriptor_set_id { 1 };
    std::atomic<uint64_t> m_next_command_id { 1 };
    std::atomic<uint64_t> m_next_fence_id { 1 };
    std::atomic<uint64_t> m_next_semaphore_id { 1 };
 
    DetectedSourceType detect_source_type(const std::string& content) const;
    std::optional<std::filesystem::path> resolve_shader_path(const std::string& filepath) const;
    std::string generate_source_cache_key(const std::string& source, ShaderStage stage) const;
 
    std::shared_ptr<Core::VKShaderModule> create_shader_module();
 
    TimestampResult resolve_timestamp(const CommandBufferState& state, const std::string& label, bool wait);
 
    void cleanup_sync_objects();
    void cleanup_descriptor_resources();
    void cleanup_shader_modules();
 
    //==========================================================================
    // INTERNAL Shader Compilation Methods
    //==========================================================================
 
    std::shared_ptr<Core::VKShaderModule> compile_from_file(
        const std::string& filepath,
        std::optional<ShaderStage> stage = std::nullopt,
        const std::string& entry_point = "main");
 
    std::shared_ptr<Core::VKShaderModule> compile_from_source(
        const std::string& source,
        ShaderStage stage,
        const std::string& entry_point = "main");
 
    std::shared_ptr<Core::VKShaderModule> compile_from_source_cached(
        const std::string& source,
        ShaderStage stage,
        const std::string& cache_key,
        const std::string& entry_point = "main");
 
    std::shared_ptr<Core::VKShaderModule> compile_from_spirv(
        const std::string& spirv_path,
        ShaderStage stage,
        const std::string& entry_point = "main");
 
    std::shared_ptr<Core::VKShaderModule> compile_from_spirv_asm(
        const std::string& spirv_asm,
        ShaderStage stage,
        const std::string& entry_point = "main");
 
    std::shared_ptr<Core::VKShaderModule> get_vk_shader_module(ShaderID shader_id);
 
    friend class ComputePress;
    friend class RenderFlow;
 
    static bool s_initialized;
};
 
/**
 * @brief Get the global shader compiler instance
 * @return Reference to singleton shader compiler
 *
 * Must call initialize() before first use.
 * Thread-safe after initialization.
 */
inline MAYAFLUX_API ShaderFoundry& get_shader_foundry()
{
    return ShaderFoundry::instance();
}
 
} // namespace MayaFlux::Portal::Graphics