Kelsidavis-WoWee/src/rendering/vk_context.cpp

#define VMA_IMPLEMENTATION
#include "rendering/vk_context.hpp"
#include "core/logger.hpp"
#include <VkBootstrap.h>
#include <SDL2/SDL_vulkan.h>
#include <imgui_impl_vulkan.h>
#include <algorithm>
#include <cstring>

namespace wowee {
namespace rendering {

static VKAPI_ATTR VkBool32 VKAPI_CALL debugCallback(
    VkDebugUtilsMessageSeverityFlagBitsEXT severity,
    [[maybe_unused]] VkDebugUtilsMessageTypeFlagsEXT type,
    const VkDebugUtilsMessengerCallbackDataEXT* callbackData,
    [[maybe_unused]] void* userData)
{
    if (severity >= VK_DEBUG_UTILS_MESSAGE_SEVERITY_ERROR_BIT_EXT) {
        LOG_ERROR("Vulkan: ", callbackData->pMessage);
    } else if (severity >= VK_DEBUG_UTILS_MESSAGE_SEVERITY_WARNING_BIT_EXT) {
        LOG_WARNING("Vulkan: ", callbackData->pMessage);
    }
    return VK_FALSE;
}

VkContext::~VkContext() {
    shutdown();
}

bool VkContext::initialize(SDL_Window* window) {
    LOG_INFO("Initializing Vulkan context");

    if (!createInstance(window)) return false;
    if (!createSurface(window)) return false;
    if (!selectPhysicalDevice()) return false;
    if (!createLogicalDevice()) return false;
    if (!createAllocator()) return false;

    int w, h;
    SDL_Vulkan_GetDrawableSize(window, &w, &h);
    if (!createSwapchain(w, h)) return false;

    if (!createCommandPools()) return false;
    if (!createSyncObjects()) return false;
    if (!createImGuiResources()) return false;

    LOG_INFO("Vulkan context initialized successfully");
    return true;
}

void VkContext::shutdown() {
    LOG_WARNING("VkContext::shutdown - vkDeviceWaitIdle...");
    if (device) {
        vkDeviceWaitIdle(device);
    }

    LOG_WARNING("VkContext::shutdown - destroyImGuiResources...");
    destroyImGuiResources();

    // Destroy sync objects
    for (auto& frame : frames) {
        if (frame.inFlightFence) vkDestroyFence(device, frame.inFlightFence, nullptr);
        if (frame.renderFinishedSemaphore) vkDestroySemaphore(device, frame.renderFinishedSemaphore, nullptr);
        if (frame.imageAvailableSemaphore) vkDestroySemaphore(device, frame.imageAvailableSemaphore, nullptr);
        if (frame.commandPool) vkDestroyCommandPool(device, frame.commandPool, nullptr);
        frame = {};
    }

    // Clean up any in-flight async upload batches (device already idle)
    for (auto& batch : inFlightBatches_) {
        // Staging buffers: skip destroy — allocator is about to be torn down
        vkDestroyFence(device, batch.fence, nullptr);
        // Command buffer freed when pool is destroyed below
    }
    inFlightBatches_.clear();

    if (immFence) { vkDestroyFence(device, immFence, nullptr); immFence = VK_NULL_HANDLE; }
    if (immCommandPool) { vkDestroyCommandPool(device, immCommandPool, nullptr); immCommandPool = VK_NULL_HANDLE; }

    LOG_WARNING("VkContext::shutdown - destroySwapchain...");
    destroySwapchain();

    // Skip vmaDestroyAllocator — it walks every allocation to free it, which
    // takes many seconds with thousands of loaded textures/models.  The driver
    // reclaims all device memory when we destroy the device, and the OS reclaims
    // everything on process exit.  Skipping this makes shutdown instant.
    allocator = VK_NULL_HANDLE;

    LOG_WARNING("VkContext::shutdown - vkDestroyDevice...");
    if (device) { vkDestroyDevice(device, nullptr); device = VK_NULL_HANDLE; }
    if (surface) { vkDestroySurfaceKHR(instance, surface, nullptr); surface = VK_NULL_HANDLE; }

    if (debugMessenger) {
        auto func = reinterpret_cast<PFN_vkDestroyDebugUtilsMessengerEXT>(
            vkGetInstanceProcAddr(instance, "vkDestroyDebugUtilsMessengerEXT"));
        if (func) func(instance, debugMessenger, nullptr);
        debugMessenger = VK_NULL_HANDLE;
    }

    if (instance) { vkDestroyInstance(instance, nullptr); instance = VK_NULL_HANDLE; }

    LOG_WARNING("Vulkan context shutdown complete");
}

bool VkContext::createInstance(SDL_Window* window) {
    // Get required SDL extensions
    unsigned int sdlExtCount = 0;
    SDL_Vulkan_GetInstanceExtensions(window, &sdlExtCount, nullptr);
    std::vector<const char*> sdlExts(sdlExtCount);
    SDL_Vulkan_GetInstanceExtensions(window, &sdlExtCount, sdlExts.data());

    vkb::InstanceBuilder builder;
    builder.set_app_name("Wowee")
           .set_app_version(VK_MAKE_VERSION(1, 0, 0))
           .require_api_version(1, 1, 0);

    for (auto ext : sdlExts) {
        builder.enable_extension(ext);
    }

    if (enableValidation) {
        builder.request_validation_layers(true)
               .set_debug_callback(debugCallback);
    }

    auto instRet = builder.build();
    if (!instRet) {
        LOG_ERROR("Failed to create Vulkan instance: ", instRet.error().message());
        return false;
    }

    vkbInstance_ = instRet.value();
    instance = vkbInstance_.instance;
    debugMessenger = vkbInstance_.debug_messenger;

    LOG_INFO("Vulkan instance created");
    return true;
}

bool VkContext::createSurface(SDL_Window* window) {
    if (!SDL_Vulkan_CreateSurface(window, instance, &surface)) {
        LOG_ERROR("Failed to create Vulkan surface: ", SDL_GetError());
        return false;
    }
    return true;
}

bool VkContext::selectPhysicalDevice() {
    vkb::PhysicalDeviceSelector selector{vkbInstance_};
    selector.set_surface(surface)
            .set_minimum_version(1, 1)
            .prefer_gpu_device_type(vkb::PreferredDeviceType::discrete);

    auto physRet = selector.select();
    if (!physRet) {
        LOG_ERROR("Failed to select Vulkan physical device: ", physRet.error().message());
        return false;
    }

    vkbPhysicalDevice_ = physRet.value();
    physicalDevice = vkbPhysicalDevice_.physical_device;

    VkPhysicalDeviceProperties props;
    vkGetPhysicalDeviceProperties(physicalDevice, &props);
    uint32_t apiVersion = props.apiVersion;

    VkPhysicalDeviceDepthStencilResolveProperties dsResolveProps{};
    dsResolveProps.sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_DEPTH_STENCIL_RESOLVE_PROPERTIES;
    VkPhysicalDeviceProperties2 props2{};
    props2.sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_PROPERTIES_2;
    props2.pNext = &dsResolveProps;
    vkGetPhysicalDeviceProperties2(physicalDevice, &props2);

    if (apiVersion >= VK_API_VERSION_1_2) {
        VkResolveModeFlags modes = dsResolveProps.supportedDepthResolveModes;
        if (modes & VK_RESOLVE_MODE_SAMPLE_ZERO_BIT) {
            depthResolveMode_ = VK_RESOLVE_MODE_SAMPLE_ZERO_BIT;
            depthResolveSupported_ = true;
        } else if (modes & VK_RESOLVE_MODE_MIN_BIT) {
            depthResolveMode_ = VK_RESOLVE_MODE_MIN_BIT;
            depthResolveSupported_ = true;
        } else if (modes & VK_RESOLVE_MODE_MAX_BIT) {
            depthResolveMode_ = VK_RESOLVE_MODE_MAX_BIT;
            depthResolveSupported_ = true;
        } else if (modes & VK_RESOLVE_MODE_AVERAGE_BIT) {
            depthResolveMode_ = VK_RESOLVE_MODE_AVERAGE_BIT;
            depthResolveSupported_ = true;
        }
    } else {
        depthResolveSupported_ = false;
        depthResolveMode_ = VK_RESOLVE_MODE_NONE;
    }

    LOG_INFO("Vulkan device: ", props.deviceName);
    LOG_INFO("Vulkan API version: ", VK_VERSION_MAJOR(props.apiVersion), ".",
             VK_VERSION_MINOR(props.apiVersion), ".", VK_VERSION_PATCH(props.apiVersion));
    LOG_INFO("Depth resolve support: ", depthResolveSupported_ ? "YES" : "NO");

    return true;
}

bool VkContext::createLogicalDevice() {
    vkb::DeviceBuilder deviceBuilder{vkbPhysicalDevice_};
    auto devRet = deviceBuilder.build();
    if (!devRet) {
        LOG_ERROR("Failed to create Vulkan logical device: ", devRet.error().message());
        return false;
    }

    auto vkbDevice = devRet.value();
    device = vkbDevice.device;

    auto gqRet = vkbDevice.get_queue(vkb::QueueType::graphics);
    if (!gqRet) {
        LOG_ERROR("Failed to get graphics queue");
        return false;
    }
    graphicsQueue = gqRet.value();
    graphicsQueueFamily = vkbDevice.get_queue_index(vkb::QueueType::graphics).value();

    auto pqRet = vkbDevice.get_queue(vkb::QueueType::present);
    if (!pqRet) {
        // Fall back to graphics queue for presentation
        presentQueue = graphicsQueue;
        presentQueueFamily = graphicsQueueFamily;
    } else {
        presentQueue = pqRet.value();
        presentQueueFamily = vkbDevice.get_queue_index(vkb::QueueType::present).value();
    }

    LOG_INFO("Vulkan logical device created");
    return true;
}

bool VkContext::createAllocator() {
    VmaAllocatorCreateInfo allocInfo{};
    allocInfo.instance = instance;
    allocInfo.physicalDevice = physicalDevice;
    allocInfo.device = device;
    allocInfo.vulkanApiVersion = VK_API_VERSION_1_1;

    if (vmaCreateAllocator(&allocInfo, &allocator) != VK_SUCCESS) {
        LOG_ERROR("Failed to create VMA allocator");
        return false;
    }

    LOG_INFO("VMA allocator created");
    return true;
}

bool VkContext::createSwapchain(int width, int height) {
    vkb::SwapchainBuilder swapchainBuilder{physicalDevice, device, surface};

    auto& builder = swapchainBuilder
        .set_desired_format({VK_FORMAT_B8G8R8A8_UNORM, VK_COLOR_SPACE_SRGB_NONLINEAR_KHR})
        .set_desired_extent(static_cast<uint32_t>(width), static_cast<uint32_t>(height))
        .set_image_usage_flags(VK_IMAGE_USAGE_COLOR_ATTACHMENT_BIT | VK_IMAGE_USAGE_TRANSFER_SRC_BIT)
        .set_desired_min_image_count(2)
        .set_old_swapchain(swapchain);

    if (vsync_) {
        builder.set_desired_present_mode(VK_PRESENT_MODE_FIFO_KHR);
    } else {
        builder.set_desired_present_mode(VK_PRESENT_MODE_IMMEDIATE_KHR);
        builder.add_fallback_present_mode(VK_PRESENT_MODE_MAILBOX_KHR);
        builder.add_fallback_present_mode(VK_PRESENT_MODE_FIFO_RELAXED_KHR);
    }

    auto swapRet = builder.build();

    if (!swapRet) {
        LOG_ERROR("Failed to create Vulkan swapchain: ", swapRet.error().message());
        return false;
    }

    // Destroy old swapchain if recreating
    if (swapchain != VK_NULL_HANDLE) {
        destroySwapchain();
    }

    auto vkbSwap = swapRet.value();
    swapchain = vkbSwap.swapchain;
    swapchainFormat = vkbSwap.image_format;
    swapchainExtent = vkbSwap.extent;
    swapchainImages = vkbSwap.get_images().value();
    swapchainImageViews = vkbSwap.get_image_views().value();

    // Create framebuffers for ImGui render pass (created after ImGui resources)
    // Will be created in createImGuiResources or recreateSwapchain

    LOG_INFO("Vulkan swapchain created: ", swapchainExtent.width, "x", swapchainExtent.height,
             " (", swapchainImages.size(), " images)");
    swapchainDirty = false;
    return true;
}

void VkContext::destroySwapchain() {
    for (auto fb : swapchainFramebuffers) {
        if (fb) vkDestroyFramebuffer(device, fb, nullptr);
    }
    swapchainFramebuffers.clear();

    for (auto iv : swapchainImageViews) {
        if (iv) vkDestroyImageView(device, iv, nullptr);
    }
    swapchainImageViews.clear();
    swapchainImages.clear();

    if (swapchain) {
        vkDestroySwapchainKHR(device, swapchain, nullptr);
        swapchain = VK_NULL_HANDLE;
    }
}

bool VkContext::createCommandPools() {
    // Per-frame command pools (resettable)
    for (uint32_t i = 0; i < MAX_FRAMES_IN_FLIGHT; i++) {
        VkCommandPoolCreateInfo poolInfo{};
        poolInfo.sType = VK_STRUCTURE_TYPE_COMMAND_POOL_CREATE_INFO;
        poolInfo.flags = VK_COMMAND_POOL_CREATE_RESET_COMMAND_BUFFER_BIT;
        poolInfo.queueFamilyIndex = graphicsQueueFamily;

        if (vkCreateCommandPool(device, &poolInfo, nullptr, &frames[i].commandPool) != VK_SUCCESS) {
            LOG_ERROR("Failed to create command pool for frame ", i);
            return false;
        }

        VkCommandBufferAllocateInfo allocInfo{};
        allocInfo.sType = VK_STRUCTURE_TYPE_COMMAND_BUFFER_ALLOCATE_INFO;
        allocInfo.commandPool = frames[i].commandPool;
        allocInfo.level = VK_COMMAND_BUFFER_LEVEL_PRIMARY;
        allocInfo.commandBufferCount = 1;

        if (vkAllocateCommandBuffers(device, &allocInfo, &frames[i].commandBuffer) != VK_SUCCESS) {
            LOG_ERROR("Failed to allocate command buffer for frame ", i);
            return false;
        }
    }

    // Immediate submit pool
    VkCommandPoolCreateInfo immPoolInfo{};
    immPoolInfo.sType = VK_STRUCTURE_TYPE_COMMAND_POOL_CREATE_INFO;
    immPoolInfo.flags = VK_COMMAND_POOL_CREATE_RESET_COMMAND_BUFFER_BIT;
    immPoolInfo.queueFamilyIndex = graphicsQueueFamily;

    if (vkCreateCommandPool(device, &immPoolInfo, nullptr, &immCommandPool) != VK_SUCCESS) {
        LOG_ERROR("Failed to create immediate command pool");
        return false;
    }

    return true;
}

bool VkContext::createSyncObjects() {
    VkSemaphoreCreateInfo semInfo{};
    semInfo.sType = VK_STRUCTURE_TYPE_SEMAPHORE_CREATE_INFO;

    VkFenceCreateInfo fenceInfo{};
    fenceInfo.sType = VK_STRUCTURE_TYPE_FENCE_CREATE_INFO;
    fenceInfo.flags = VK_FENCE_CREATE_SIGNALED_BIT; // Start signaled so first frame doesn't block

    for (uint32_t i = 0; i < MAX_FRAMES_IN_FLIGHT; i++) {
        if (vkCreateSemaphore(device, &semInfo, nullptr, &frames[i].imageAvailableSemaphore) != VK_SUCCESS ||
            vkCreateSemaphore(device, &semInfo, nullptr, &frames[i].renderFinishedSemaphore) != VK_SUCCESS ||
            vkCreateFence(device, &fenceInfo, nullptr, &frames[i].inFlightFence) != VK_SUCCESS) {
            LOG_ERROR("Failed to create sync objects for frame ", i);
            return false;
        }
    }

    // Immediate submit fence (not signaled initially)
    VkFenceCreateInfo immFenceInfo{};
    immFenceInfo.sType = VK_STRUCTURE_TYPE_FENCE_CREATE_INFO;
    if (vkCreateFence(device, &immFenceInfo, nullptr, &immFence) != VK_SUCCESS) {
        LOG_ERROR("Failed to create immediate submit fence");
        return false;
    }

    return true;
}

bool VkContext::createDepthBuffer() {
    VkImageCreateInfo imgInfo{};
    imgInfo.sType = VK_STRUCTURE_TYPE_IMAGE_CREATE_INFO;
    imgInfo.imageType = VK_IMAGE_TYPE_2D;
    imgInfo.format = depthFormat;
    imgInfo.extent = {swapchainExtent.width, swapchainExtent.height, 1};
    imgInfo.mipLevels = 1;
    imgInfo.arrayLayers = 1;
    imgInfo.samples = msaaSamples_;
    imgInfo.tiling = VK_IMAGE_TILING_OPTIMAL;
    imgInfo.usage = VK_IMAGE_USAGE_DEPTH_STENCIL_ATTACHMENT_BIT | VK_IMAGE_USAGE_TRANSFER_SRC_BIT;

    VmaAllocationCreateInfo allocInfo{};
    allocInfo.usage = VMA_MEMORY_USAGE_GPU_ONLY;

    if (vmaCreateImage(allocator, &imgInfo, &allocInfo, &depthImage, &depthAllocation, nullptr) != VK_SUCCESS) {
        LOG_ERROR("Failed to create depth image");
        return false;
    }

    VkImageViewCreateInfo viewInfo{};
    viewInfo.sType = VK_STRUCTURE_TYPE_IMAGE_VIEW_CREATE_INFO;
    viewInfo.image = depthImage;
    viewInfo.viewType = VK_IMAGE_VIEW_TYPE_2D;
    viewInfo.format = depthFormat;
    viewInfo.subresourceRange.aspectMask = VK_IMAGE_ASPECT_DEPTH_BIT;
    viewInfo.subresourceRange.levelCount = 1;
    viewInfo.subresourceRange.layerCount = 1;

    if (vkCreateImageView(device, &viewInfo, nullptr, &depthImageView) != VK_SUCCESS) {
        LOG_ERROR("Failed to create depth image view");
        return false;
    }

    return true;
}

void VkContext::destroyDepthBuffer() {
    if (depthImageView) { vkDestroyImageView(device, depthImageView, nullptr); depthImageView = VK_NULL_HANDLE; }
    if (depthImage) { vmaDestroyImage(allocator, depthImage, depthAllocation); depthImage = VK_NULL_HANDLE; depthAllocation = VK_NULL_HANDLE; }
}

bool VkContext::createMsaaColorImage() {
    if (msaaSamples_ == VK_SAMPLE_COUNT_1_BIT) return true; // No MSAA image needed

    VkImageCreateInfo imgInfo{};
    imgInfo.sType = VK_STRUCTURE_TYPE_IMAGE_CREATE_INFO;
    imgInfo.imageType = VK_IMAGE_TYPE_2D;
    imgInfo.format = swapchainFormat;
    imgInfo.extent = {swapchainExtent.width, swapchainExtent.height, 1};
    imgInfo.mipLevels = 1;
    imgInfo.arrayLayers = 1;
    imgInfo.samples = msaaSamples_;
    imgInfo.tiling = VK_IMAGE_TILING_OPTIMAL;
    imgInfo.usage = VK_IMAGE_USAGE_COLOR_ATTACHMENT_BIT | VK_IMAGE_USAGE_TRANSIENT_ATTACHMENT_BIT;

    VmaAllocationCreateInfo allocInfo{};
    allocInfo.usage = VMA_MEMORY_USAGE_GPU_ONLY;
    allocInfo.preferredFlags = VK_MEMORY_PROPERTY_LAZILY_ALLOCATED_BIT;

    if (vmaCreateImage(allocator, &imgInfo, &allocInfo, &msaaColorImage_, &msaaColorAllocation_, nullptr) != VK_SUCCESS) {
        // Retry without TRANSIENT (some drivers reject it at high sample counts)
        imgInfo.usage = VK_IMAGE_USAGE_COLOR_ATTACHMENT_BIT;
        allocInfo.preferredFlags = 0;
        if (vmaCreateImage(allocator, &imgInfo, &allocInfo, &msaaColorImage_, &msaaColorAllocation_, nullptr) != VK_SUCCESS) {
            LOG_ERROR("Failed to create MSAA color image");
            return false;
        }
    }

    VkImageViewCreateInfo viewInfo{};
    viewInfo.sType = VK_STRUCTURE_TYPE_IMAGE_VIEW_CREATE_INFO;
    viewInfo.image = msaaColorImage_;
    viewInfo.viewType = VK_IMAGE_VIEW_TYPE_2D;
    viewInfo.format = swapchainFormat;
    viewInfo.subresourceRange.aspectMask = VK_IMAGE_ASPECT_COLOR_BIT;
    viewInfo.subresourceRange.levelCount = 1;
    viewInfo.subresourceRange.layerCount = 1;

    if (vkCreateImageView(device, &viewInfo, nullptr, &msaaColorView_) != VK_SUCCESS) {
        LOG_ERROR("Failed to create MSAA color image view");
        return false;
    }

    return true;
}

void VkContext::destroyMsaaColorImage() {
    if (msaaColorView_) { vkDestroyImageView(device, msaaColorView_, nullptr); msaaColorView_ = VK_NULL_HANDLE; }
    if (msaaColorImage_) { vmaDestroyImage(allocator, msaaColorImage_, msaaColorAllocation_); msaaColorImage_ = VK_NULL_HANDLE; msaaColorAllocation_ = VK_NULL_HANDLE; }
}

bool VkContext::createDepthResolveImage() {
    if (msaaSamples_ == VK_SAMPLE_COUNT_1_BIT || !depthResolveSupported_) return true;

    VkImageCreateInfo imgInfo{};
    imgInfo.sType = VK_STRUCTURE_TYPE_IMAGE_CREATE_INFO;
    imgInfo.imageType = VK_IMAGE_TYPE_2D;
    imgInfo.format = depthFormat;
    imgInfo.extent = {swapchainExtent.width, swapchainExtent.height, 1};
    imgInfo.mipLevels = 1;
    imgInfo.arrayLayers = 1;
    imgInfo.samples = VK_SAMPLE_COUNT_1_BIT;
    imgInfo.tiling = VK_IMAGE_TILING_OPTIMAL;
    imgInfo.usage = VK_IMAGE_USAGE_DEPTH_STENCIL_ATTACHMENT_BIT | VK_IMAGE_USAGE_TRANSFER_SRC_BIT;

    VmaAllocationCreateInfo allocInfo{};
    allocInfo.usage = VMA_MEMORY_USAGE_GPU_ONLY;

    if (vmaCreateImage(allocator, &imgInfo, &allocInfo, &depthResolveImage, &depthResolveAllocation, nullptr) != VK_SUCCESS) {
        LOG_ERROR("Failed to create depth resolve image");
        return false;
    }

    VkImageViewCreateInfo viewInfo{};
    viewInfo.sType = VK_STRUCTURE_TYPE_IMAGE_VIEW_CREATE_INFO;
    viewInfo.image = depthResolveImage;
    viewInfo.viewType = VK_IMAGE_VIEW_TYPE_2D;
    viewInfo.format = depthFormat;
    viewInfo.subresourceRange.aspectMask = VK_IMAGE_ASPECT_DEPTH_BIT;
    viewInfo.subresourceRange.levelCount = 1;
    viewInfo.subresourceRange.layerCount = 1;
    if (vkCreateImageView(device, &viewInfo, nullptr, &depthResolveImageView) != VK_SUCCESS) {
        LOG_ERROR("Failed to create depth resolve image view");
        return false;
    }

    return true;
}

void VkContext::destroyDepthResolveImage() {
    if (depthResolveImageView) {
        vkDestroyImageView(device, depthResolveImageView, nullptr);
        depthResolveImageView = VK_NULL_HANDLE;
    }
    if (depthResolveImage) {
        vmaDestroyImage(allocator, depthResolveImage, depthResolveAllocation);
        depthResolveImage = VK_NULL_HANDLE;
        depthResolveAllocation = VK_NULL_HANDLE;
    }
}

VkSampleCountFlagBits VkContext::getMaxUsableSampleCount() const {
    VkPhysicalDeviceProperties props;
    vkGetPhysicalDeviceProperties(physicalDevice, &props);
    VkSampleCountFlags counts = props.limits.framebufferColorSampleCounts
                               & props.limits.framebufferDepthSampleCounts;
    if (counts & VK_SAMPLE_COUNT_8_BIT) return VK_SAMPLE_COUNT_8_BIT;
    if (counts & VK_SAMPLE_COUNT_4_BIT) return VK_SAMPLE_COUNT_4_BIT;
    if (counts & VK_SAMPLE_COUNT_2_BIT) return VK_SAMPLE_COUNT_2_BIT;
    return VK_SAMPLE_COUNT_1_BIT;
}

void VkContext::setMsaaSamples(VkSampleCountFlagBits samples) {
    // Clamp to max supported
    VkSampleCountFlagBits maxSamples = getMaxUsableSampleCount();
    if (samples > maxSamples) samples = maxSamples;
    msaaSamples_ = samples;
    swapchainDirty = true;
}

bool VkContext::createImGuiResources() {
    // Create depth buffer first
    if (!createDepthBuffer()) return false;

    // Create MSAA color image if needed
    if (!createMsaaColorImage()) return false;
    // Create single-sample depth resolve image for MSAA path (if supported)
    if (!createDepthResolveImage()) return false;

    bool useMsaa = (msaaSamples_ > VK_SAMPLE_COUNT_1_BIT);

    if (useMsaa) {
        const bool useDepthResolve = (depthResolveImageView != VK_NULL_HANDLE);
        // MSAA render pass: 3 or 4 attachments
        VkAttachmentDescription attachments[4] = {};

        // Attachment 0: MSAA color target
        attachments[0].format = swapchainFormat;
        attachments[0].samples = msaaSamples_;
        attachments[0].loadOp = VK_ATTACHMENT_LOAD_OP_CLEAR;
        attachments[0].storeOp = VK_ATTACHMENT_STORE_OP_DONT_CARE;
        attachments[0].stencilLoadOp = VK_ATTACHMENT_LOAD_OP_DONT_CARE;
        attachments[0].stencilStoreOp = VK_ATTACHMENT_STORE_OP_DONT_CARE;
        attachments[0].initialLayout = VK_IMAGE_LAYOUT_UNDEFINED;
        attachments[0].finalLayout = VK_IMAGE_LAYOUT_COLOR_ATTACHMENT_OPTIMAL;

        // Attachment 1: Depth (multisampled)
        attachments[1].format = depthFormat;
        attachments[1].samples = msaaSamples_;
        attachments[1].loadOp = VK_ATTACHMENT_LOAD_OP_CLEAR;
        attachments[1].storeOp = VK_ATTACHMENT_STORE_OP_STORE;
        attachments[1].stencilLoadOp = VK_ATTACHMENT_LOAD_OP_DONT_CARE;
        attachments[1].stencilStoreOp = VK_ATTACHMENT_STORE_OP_DONT_CARE;
        attachments[1].initialLayout = VK_IMAGE_LAYOUT_UNDEFINED;
        attachments[1].finalLayout = VK_IMAGE_LAYOUT_DEPTH_STENCIL_ATTACHMENT_OPTIMAL;

        // Attachment 2: Resolve target (swapchain image)
        attachments[2].format = swapchainFormat;
        attachments[2].samples = VK_SAMPLE_COUNT_1_BIT;
        attachments[2].loadOp = VK_ATTACHMENT_LOAD_OP_DONT_CARE;
        attachments[2].storeOp = VK_ATTACHMENT_STORE_OP_STORE;
        attachments[2].stencilLoadOp = VK_ATTACHMENT_LOAD_OP_DONT_CARE;
        attachments[2].stencilStoreOp = VK_ATTACHMENT_STORE_OP_DONT_CARE;
        attachments[2].initialLayout = VK_IMAGE_LAYOUT_UNDEFINED;
        attachments[2].finalLayout = VK_IMAGE_LAYOUT_PRESENT_SRC_KHR;

        if (useDepthResolve) {
            attachments[3].format = depthFormat;
            attachments[3].samples = VK_SAMPLE_COUNT_1_BIT;
            attachments[3].loadOp = VK_ATTACHMENT_LOAD_OP_DONT_CARE;
            attachments[3].storeOp = VK_ATTACHMENT_STORE_OP_STORE;
            attachments[3].stencilLoadOp = VK_ATTACHMENT_LOAD_OP_DONT_CARE;
            attachments[3].stencilStoreOp = VK_ATTACHMENT_STORE_OP_DONT_CARE;
            attachments[3].initialLayout = VK_IMAGE_LAYOUT_UNDEFINED;
            attachments[3].finalLayout = VK_IMAGE_LAYOUT_DEPTH_STENCIL_ATTACHMENT_OPTIMAL;
        }

        if (useDepthResolve) {
            VkAttachmentDescription2 attachments2[4]{};
            for (int i = 0; i < 4; ++i) {
                attachments2[i].sType = VK_STRUCTURE_TYPE_ATTACHMENT_DESCRIPTION_2;
                attachments2[i].format = attachments[i].format;
                attachments2[i].samples = attachments[i].samples;
                attachments2[i].loadOp = attachments[i].loadOp;
                attachments2[i].storeOp = attachments[i].storeOp;
                attachments2[i].stencilLoadOp = attachments[i].stencilLoadOp;
                attachments2[i].stencilStoreOp = attachments[i].stencilStoreOp;
                attachments2[i].initialLayout = attachments[i].initialLayout;
                attachments2[i].finalLayout = attachments[i].finalLayout;
            }

            VkAttachmentReference2 colorRef2{};
            colorRef2.sType = VK_STRUCTURE_TYPE_ATTACHMENT_REFERENCE_2;
            colorRef2.attachment = 0;
            colorRef2.layout = VK_IMAGE_LAYOUT_COLOR_ATTACHMENT_OPTIMAL;
            VkAttachmentReference2 depthRef2{};
            depthRef2.sType = VK_STRUCTURE_TYPE_ATTACHMENT_REFERENCE_2;
            depthRef2.attachment = 1;
            depthRef2.layout = VK_IMAGE_LAYOUT_DEPTH_STENCIL_ATTACHMENT_OPTIMAL;
            VkAttachmentReference2 resolveRef2{};
            resolveRef2.sType = VK_STRUCTURE_TYPE_ATTACHMENT_REFERENCE_2;
            resolveRef2.attachment = 2;
            resolveRef2.layout = VK_IMAGE_LAYOUT_COLOR_ATTACHMENT_OPTIMAL;
            VkAttachmentReference2 depthResolveRef2{};
            depthResolveRef2.sType = VK_STRUCTURE_TYPE_ATTACHMENT_REFERENCE_2;
            depthResolveRef2.attachment = 3;
            depthResolveRef2.layout = VK_IMAGE_LAYOUT_DEPTH_STENCIL_ATTACHMENT_OPTIMAL;

            VkSubpassDescriptionDepthStencilResolve dsResolve{};
            dsResolve.sType = VK_STRUCTURE_TYPE_SUBPASS_DESCRIPTION_DEPTH_STENCIL_RESOLVE;
            dsResolve.depthResolveMode = depthResolveMode_;
            dsResolve.stencilResolveMode = VK_RESOLVE_MODE_NONE;
            dsResolve.pDepthStencilResolveAttachment = &depthResolveRef2;

            VkSubpassDescription2 subpass2{};
            subpass2.sType = VK_STRUCTURE_TYPE_SUBPASS_DESCRIPTION_2;
            subpass2.pNext = &dsResolve;
            subpass2.pipelineBindPoint = VK_PIPELINE_BIND_POINT_GRAPHICS;
            subpass2.colorAttachmentCount = 1;
            subpass2.pColorAttachments = &colorRef2;
            subpass2.pDepthStencilAttachment = &depthRef2;
            subpass2.pResolveAttachments = &resolveRef2;

            VkSubpassDependency2 dep2{};
            dep2.sType = VK_STRUCTURE_TYPE_SUBPASS_DEPENDENCY_2;
            dep2.srcSubpass = VK_SUBPASS_EXTERNAL;
            dep2.dstSubpass = 0;
            dep2.srcStageMask = VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT | VK_PIPELINE_STAGE_EARLY_FRAGMENT_TESTS_BIT;
            dep2.dstStageMask = VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT | VK_PIPELINE_STAGE_EARLY_FRAGMENT_TESTS_BIT;
            dep2.dstAccessMask = VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT | VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_WRITE_BIT;

            VkRenderPassCreateInfo2 rpInfo2{};
            rpInfo2.sType = VK_STRUCTURE_TYPE_RENDER_PASS_CREATE_INFO_2;
            rpInfo2.attachmentCount = 4;
            rpInfo2.pAttachments = attachments2;
            rpInfo2.subpassCount = 1;
            rpInfo2.pSubpasses = &subpass2;
            rpInfo2.dependencyCount = 1;
            rpInfo2.pDependencies = &dep2;

            if (vkCreateRenderPass2(device, &rpInfo2, nullptr, &imguiRenderPass) != VK_SUCCESS) {
                LOG_ERROR("Failed to create MSAA render pass (depth resolve)");
                return false;
            }
        } else {
            VkAttachmentReference colorRef{};
            colorRef.attachment = 0;
            colorRef.layout = VK_IMAGE_LAYOUT_COLOR_ATTACHMENT_OPTIMAL;

            VkAttachmentReference depthRef{};
            depthRef.attachment = 1;
            depthRef.layout = VK_IMAGE_LAYOUT_DEPTH_STENCIL_ATTACHMENT_OPTIMAL;

            VkAttachmentReference resolveRef{};
            resolveRef.attachment = 2;
            resolveRef.layout = VK_IMAGE_LAYOUT_COLOR_ATTACHMENT_OPTIMAL;

            VkSubpassDescription subpass{};
            subpass.pipelineBindPoint = VK_PIPELINE_BIND_POINT_GRAPHICS;
            subpass.colorAttachmentCount = 1;
            subpass.pColorAttachments = &colorRef;
            subpass.pDepthStencilAttachment = &depthRef;
            subpass.pResolveAttachments = &resolveRef;

            VkSubpassDependency dependency{};
            dependency.srcSubpass = VK_SUBPASS_EXTERNAL;
            dependency.dstSubpass = 0;
            dependency.srcStageMask = VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT | VK_PIPELINE_STAGE_EARLY_FRAGMENT_TESTS_BIT;
            dependency.dstStageMask = VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT | VK_PIPELINE_STAGE_EARLY_FRAGMENT_TESTS_BIT;
            dependency.srcAccessMask = 0;
            dependency.dstAccessMask = VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT | VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_WRITE_BIT;

            VkRenderPassCreateInfo rpInfo{};
            rpInfo.sType = VK_STRUCTURE_TYPE_RENDER_PASS_CREATE_INFO;
            rpInfo.attachmentCount = 3;
            rpInfo.pAttachments = attachments;
            rpInfo.subpassCount = 1;
            rpInfo.pSubpasses = &subpass;
            rpInfo.dependencyCount = 1;
            rpInfo.pDependencies = &dependency;

            if (vkCreateRenderPass(device, &rpInfo, nullptr, &imguiRenderPass) != VK_SUCCESS) {
                LOG_ERROR("Failed to create MSAA render pass");
                return false;
            }
        }

        // Framebuffers: [msaaColorView, depthView, swapchainView, depthResolveView?]
        swapchainFramebuffers.resize(swapchainImageViews.size());
        for (size_t i = 0; i < swapchainImageViews.size(); i++) {
            VkImageView fbAttachments[4] = {msaaColorView_, depthImageView, swapchainImageViews[i], depthResolveImageView};

            VkFramebufferCreateInfo fbInfo{};
            fbInfo.sType = VK_STRUCTURE_TYPE_FRAMEBUFFER_CREATE_INFO;
            fbInfo.renderPass = imguiRenderPass;
            fbInfo.attachmentCount = useDepthResolve ? 4 : 3;
            fbInfo.pAttachments = fbAttachments;
            fbInfo.width = swapchainExtent.width;
            fbInfo.height = swapchainExtent.height;
            fbInfo.layers = 1;

            if (vkCreateFramebuffer(device, &fbInfo, nullptr, &swapchainFramebuffers[i]) != VK_SUCCESS) {
                LOG_ERROR("Failed to create MSAA swapchain framebuffer ", i);
                return false;
            }
        }
    } else {
        // Non-MSAA render pass: 2 attachments (color + depth) — original path
        VkAttachmentDescription attachments[2] = {};

        // Color attachment (swapchain image)
        attachments[0].format = swapchainFormat;
        attachments[0].samples = VK_SAMPLE_COUNT_1_BIT;
        attachments[0].loadOp = VK_ATTACHMENT_LOAD_OP_CLEAR;
        attachments[0].storeOp = VK_ATTACHMENT_STORE_OP_STORE;
        attachments[0].stencilLoadOp = VK_ATTACHMENT_LOAD_OP_DONT_CARE;
        attachments[0].stencilStoreOp = VK_ATTACHMENT_STORE_OP_DONT_CARE;
        attachments[0].initialLayout = VK_IMAGE_LAYOUT_UNDEFINED;
        attachments[0].finalLayout = VK_IMAGE_LAYOUT_PRESENT_SRC_KHR;

        // Depth attachment
        attachments[1].format = depthFormat;
        attachments[1].samples = VK_SAMPLE_COUNT_1_BIT;
        attachments[1].loadOp = VK_ATTACHMENT_LOAD_OP_CLEAR;
        attachments[1].storeOp = VK_ATTACHMENT_STORE_OP_STORE;
        attachments[1].stencilLoadOp = VK_ATTACHMENT_LOAD_OP_DONT_CARE;
        attachments[1].stencilStoreOp = VK_ATTACHMENT_STORE_OP_DONT_CARE;
        attachments[1].initialLayout = VK_IMAGE_LAYOUT_UNDEFINED;
        attachments[1].finalLayout = VK_IMAGE_LAYOUT_DEPTH_STENCIL_ATTACHMENT_OPTIMAL;

        VkAttachmentReference colorRef{};
        colorRef.attachment = 0;
        colorRef.layout = VK_IMAGE_LAYOUT_COLOR_ATTACHMENT_OPTIMAL;

        VkAttachmentReference depthRef{};
        depthRef.attachment = 1;
        depthRef.layout = VK_IMAGE_LAYOUT_DEPTH_STENCIL_ATTACHMENT_OPTIMAL;

        VkSubpassDescription subpass{};
        subpass.pipelineBindPoint = VK_PIPELINE_BIND_POINT_GRAPHICS;
        subpass.colorAttachmentCount = 1;
        subpass.pColorAttachments = &colorRef;
        subpass.pDepthStencilAttachment = &depthRef;

        VkSubpassDependency dependency{};
        dependency.srcSubpass = VK_SUBPASS_EXTERNAL;
        dependency.dstSubpass = 0;
        dependency.srcStageMask = VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT | VK_PIPELINE_STAGE_EARLY_FRAGMENT_TESTS_BIT;
        dependency.dstStageMask = VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT | VK_PIPELINE_STAGE_EARLY_FRAGMENT_TESTS_BIT;
        dependency.srcAccessMask = 0;
        dependency.dstAccessMask = VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT | VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_WRITE_BIT;

        VkRenderPassCreateInfo rpInfo{};
        rpInfo.sType = VK_STRUCTURE_TYPE_RENDER_PASS_CREATE_INFO;
        rpInfo.attachmentCount = 2;
        rpInfo.pAttachments = attachments;
        rpInfo.subpassCount = 1;
        rpInfo.pSubpasses = &subpass;
        rpInfo.dependencyCount = 1;
        rpInfo.pDependencies = &dependency;

        if (vkCreateRenderPass(device, &rpInfo, nullptr, &imguiRenderPass) != VK_SUCCESS) {
            LOG_ERROR("Failed to create render pass");
            return false;
        }

        // Framebuffers: [swapchainView, depthView]
        swapchainFramebuffers.resize(swapchainImageViews.size());
        for (size_t i = 0; i < swapchainImageViews.size(); i++) {
            VkImageView fbAttachments[2] = {swapchainImageViews[i], depthImageView};

            VkFramebufferCreateInfo fbInfo{};
            fbInfo.sType = VK_STRUCTURE_TYPE_FRAMEBUFFER_CREATE_INFO;
            fbInfo.renderPass = imguiRenderPass;
            fbInfo.attachmentCount = 2;
            fbInfo.pAttachments = fbAttachments;
            fbInfo.width = swapchainExtent.width;
            fbInfo.height = swapchainExtent.height;
            fbInfo.layers = 1;

            if (vkCreateFramebuffer(device, &fbInfo, nullptr, &swapchainFramebuffers[i]) != VK_SUCCESS) {
                LOG_ERROR("Failed to create swapchain framebuffer ", i);
                return false;
            }
        }
    }

    // Create descriptor pool for ImGui
    VkDescriptorPoolSize poolSizes[] = {
        {VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER, 100},
    };

    VkDescriptorPoolCreateInfo dpInfo{};
    dpInfo.sType = VK_STRUCTURE_TYPE_DESCRIPTOR_POOL_CREATE_INFO;
    dpInfo.flags = VK_DESCRIPTOR_POOL_CREATE_FREE_DESCRIPTOR_SET_BIT;
    dpInfo.maxSets = 100;
    dpInfo.poolSizeCount = 1;
    dpInfo.pPoolSizes = poolSizes;

    if (vkCreateDescriptorPool(device, &dpInfo, nullptr, &imguiDescriptorPool) != VK_SUCCESS) {
        LOG_ERROR("Failed to create ImGui descriptor pool");
        return false;
    }

    return true;
}

void VkContext::destroyImGuiResources() {
    // Destroy uploaded UI textures
    for (auto& tex : uiTextures_) {
        if (tex.view) vkDestroyImageView(device, tex.view, nullptr);
        if (tex.image) vkDestroyImage(device, tex.image, nullptr);
        if (tex.memory) vkFreeMemory(device, tex.memory, nullptr);
    }
    uiTextures_.clear();
    if (uiTextureSampler_) {
        vkDestroySampler(device, uiTextureSampler_, nullptr);
        uiTextureSampler_ = VK_NULL_HANDLE;
    }

    if (imguiDescriptorPool) {
        vkDestroyDescriptorPool(device, imguiDescriptorPool, nullptr);
        imguiDescriptorPool = VK_NULL_HANDLE;
    }
    destroyMsaaColorImage();
    destroyDepthResolveImage();
    destroyDepthBuffer();
    // Framebuffers are destroyed in destroySwapchain()
    if (imguiRenderPass) {
        vkDestroyRenderPass(device, imguiRenderPass, nullptr);
        imguiRenderPass = VK_NULL_HANDLE;
    }
}

static uint32_t findMemType(VkPhysicalDevice physDev, uint32_t typeFilter, VkMemoryPropertyFlags props) {
    VkPhysicalDeviceMemoryProperties memProps;
    vkGetPhysicalDeviceMemoryProperties(physDev, &memProps);
    for (uint32_t i = 0; i < memProps.memoryTypeCount; i++) {
        if ((typeFilter & (1 << i)) && (memProps.memoryTypes[i].propertyFlags & props) == props)
            return i;
    }
    return 0;
}

VkDescriptorSet VkContext::uploadImGuiTexture(const uint8_t* rgba, int width, int height) {
    if (!device || !physicalDevice || width <= 0 || height <= 0 || !rgba)
        return VK_NULL_HANDLE;

    VkDeviceSize imageSize = static_cast<VkDeviceSize>(width) * height * 4;

    // Create shared sampler on first call
    if (!uiTextureSampler_) {
        VkSamplerCreateInfo si{};
        si.sType = VK_STRUCTURE_TYPE_SAMPLER_CREATE_INFO;
        si.magFilter = VK_FILTER_LINEAR;
        si.minFilter = VK_FILTER_LINEAR;
        si.addressModeU = VK_SAMPLER_ADDRESS_MODE_CLAMP_TO_EDGE;
        si.addressModeV = VK_SAMPLER_ADDRESS_MODE_CLAMP_TO_EDGE;
        si.addressModeW = VK_SAMPLER_ADDRESS_MODE_CLAMP_TO_EDGE;
        if (vkCreateSampler(device, &si, nullptr, &uiTextureSampler_) != VK_SUCCESS) {
            LOG_ERROR("Failed to create UI texture sampler");
            return VK_NULL_HANDLE;
        }
    }

    // Staging buffer
    VkBuffer stagingBuffer;
    VkDeviceMemory stagingMemory;
    {
        VkBufferCreateInfo bufInfo{};
        bufInfo.sType = VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO;
        bufInfo.size = imageSize;
        bufInfo.usage = VK_BUFFER_USAGE_TRANSFER_SRC_BIT;
        bufInfo.sharingMode = VK_SHARING_MODE_EXCLUSIVE;
        if (vkCreateBuffer(device, &bufInfo, nullptr, &stagingBuffer) != VK_SUCCESS)
            return VK_NULL_HANDLE;

        VkMemoryRequirements memReqs;
        vkGetBufferMemoryRequirements(device, stagingBuffer, &memReqs);

        VkMemoryAllocateInfo allocInfo{};
        allocInfo.sType = VK_STRUCTURE_TYPE_MEMORY_ALLOCATE_INFO;
        allocInfo.allocationSize = memReqs.size;
        allocInfo.memoryTypeIndex = findMemType(physicalDevice, memReqs.memoryTypeBits,
            VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT | VK_MEMORY_PROPERTY_HOST_COHERENT_BIT);
        if (vkAllocateMemory(device, &allocInfo, nullptr, &stagingMemory) != VK_SUCCESS) {
            vkDestroyBuffer(device, stagingBuffer, nullptr);
            return VK_NULL_HANDLE;
        }
        vkBindBufferMemory(device, stagingBuffer, stagingMemory, 0);

        void* mapped;
        vkMapMemory(device, stagingMemory, 0, imageSize, 0, &mapped);
        memcpy(mapped, rgba, imageSize);
        vkUnmapMemory(device, stagingMemory);
    }

    // Create image
    VkImage image;
    VkDeviceMemory imageMemory;
    {
        VkImageCreateInfo imgInfo{};
        imgInfo.sType = VK_STRUCTURE_TYPE_IMAGE_CREATE_INFO;
        imgInfo.imageType = VK_IMAGE_TYPE_2D;
        imgInfo.format = VK_FORMAT_R8G8B8A8_UNORM;
        imgInfo.extent = {static_cast<uint32_t>(width), static_cast<uint32_t>(height), 1};
        imgInfo.mipLevels = 1;
        imgInfo.arrayLayers = 1;
        imgInfo.samples = VK_SAMPLE_COUNT_1_BIT;
        imgInfo.tiling = VK_IMAGE_TILING_OPTIMAL;
        imgInfo.usage = VK_IMAGE_USAGE_TRANSFER_DST_BIT | VK_IMAGE_USAGE_SAMPLED_BIT;
        imgInfo.sharingMode = VK_SHARING_MODE_EXCLUSIVE;
        imgInfo.initialLayout = VK_IMAGE_LAYOUT_UNDEFINED;
        if (vkCreateImage(device, &imgInfo, nullptr, &image) != VK_SUCCESS) {
            vkDestroyBuffer(device, stagingBuffer, nullptr);
            vkFreeMemory(device, stagingMemory, nullptr);
            return VK_NULL_HANDLE;
        }

        VkMemoryRequirements memReqs;
        vkGetImageMemoryRequirements(device, image, &memReqs);

        VkMemoryAllocateInfo allocInfo{};
        allocInfo.sType = VK_STRUCTURE_TYPE_MEMORY_ALLOCATE_INFO;
        allocInfo.allocationSize = memReqs.size;
        allocInfo.memoryTypeIndex = findMemType(physicalDevice, memReqs.memoryTypeBits,
            VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT);
        if (vkAllocateMemory(device, &allocInfo, nullptr, &imageMemory) != VK_SUCCESS) {
            vkDestroyImage(device, image, nullptr);
            vkDestroyBuffer(device, stagingBuffer, nullptr);
            vkFreeMemory(device, stagingMemory, nullptr);
            return VK_NULL_HANDLE;
        }
        vkBindImageMemory(device, image, imageMemory, 0);
    }

    // Upload via immediate submit
    immediateSubmit([&](VkCommandBuffer cmd) {
        VkImageMemoryBarrier barrier{};
        barrier.sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER;
        barrier.oldLayout = VK_IMAGE_LAYOUT_UNDEFINED;
        barrier.newLayout = VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL;
        barrier.srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED;
        barrier.dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED;
        barrier.image = image;
        barrier.subresourceRange = {VK_IMAGE_ASPECT_COLOR_BIT, 0, 1, 0, 1};
        barrier.srcAccessMask = 0;
        barrier.dstAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT;
        vkCmdPipelineBarrier(cmd, VK_PIPELINE_STAGE_TOP_OF_PIPE_BIT,
            VK_PIPELINE_STAGE_TRANSFER_BIT, 0, 0, nullptr, 0, nullptr, 1, &barrier);

        VkBufferImageCopy region{};
        region.imageSubresource = {VK_IMAGE_ASPECT_COLOR_BIT, 0, 0, 1};
        region.imageExtent = {static_cast<uint32_t>(width), static_cast<uint32_t>(height), 1};
        vkCmdCopyBufferToImage(cmd, stagingBuffer, image,
            VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL, 1, &region);

        barrier.oldLayout = VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL;
        barrier.newLayout = VK_IMAGE_LAYOUT_SHADER_READ_ONLY_OPTIMAL;
        barrier.srcAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT;
        barrier.dstAccessMask = VK_ACCESS_SHADER_READ_BIT;
        vkCmdPipelineBarrier(cmd, VK_PIPELINE_STAGE_TRANSFER_BIT,
            VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT, 0, 0, nullptr, 0, nullptr, 1, &barrier);
    });

    // Cleanup staging
    vkDestroyBuffer(device, stagingBuffer, nullptr);
    vkFreeMemory(device, stagingMemory, nullptr);

    // Create image view
    VkImageView imageView;
    {
        VkImageViewCreateInfo viewInfo{};
        viewInfo.sType = VK_STRUCTURE_TYPE_IMAGE_VIEW_CREATE_INFO;
        viewInfo.image = image;
        viewInfo.viewType = VK_IMAGE_VIEW_TYPE_2D;
        viewInfo.format = VK_FORMAT_R8G8B8A8_UNORM;
        viewInfo.subresourceRange = {VK_IMAGE_ASPECT_COLOR_BIT, 0, 1, 0, 1};
        if (vkCreateImageView(device, &viewInfo, nullptr, &imageView) != VK_SUCCESS) {
            vkDestroyImage(device, image, nullptr);
            vkFreeMemory(device, imageMemory, nullptr);
            return VK_NULL_HANDLE;
        }
    }

    // Register with ImGui
    VkDescriptorSet ds = ImGui_ImplVulkan_AddTexture(uiTextureSampler_, imageView,
        VK_IMAGE_LAYOUT_SHADER_READ_ONLY_OPTIMAL);

    // Track for cleanup
    uiTextures_.push_back({image, imageMemory, imageView});

    return ds;
}

bool VkContext::recreateSwapchain(int width, int height) {
    vkDeviceWaitIdle(device);

    // Destroy old framebuffers
    for (auto fb : swapchainFramebuffers) {
        if (fb) vkDestroyFramebuffer(device, fb, nullptr);
    }
    swapchainFramebuffers.clear();

    // Destroy old image views
    for (auto iv : swapchainImageViews) {
        if (iv) vkDestroyImageView(device, iv, nullptr);
    }
    swapchainImageViews.clear();

    VkSwapchainKHR oldSwapchain = swapchain;

    vkb::SwapchainBuilder swapchainBuilder{physicalDevice, device, surface};
    auto& builder = swapchainBuilder
        .set_desired_format({VK_FORMAT_B8G8R8A8_UNORM, VK_COLOR_SPACE_SRGB_NONLINEAR_KHR})
        .set_desired_extent(static_cast<uint32_t>(width), static_cast<uint32_t>(height))
        .set_image_usage_flags(VK_IMAGE_USAGE_COLOR_ATTACHMENT_BIT | VK_IMAGE_USAGE_TRANSFER_SRC_BIT)
        .set_desired_min_image_count(2)
        .set_old_swapchain(oldSwapchain);

    if (vsync_) {
        builder.set_desired_present_mode(VK_PRESENT_MODE_FIFO_KHR);
    } else {
        builder.set_desired_present_mode(VK_PRESENT_MODE_IMMEDIATE_KHR);
        builder.add_fallback_present_mode(VK_PRESENT_MODE_MAILBOX_KHR);
        builder.add_fallback_present_mode(VK_PRESENT_MODE_FIFO_RELAXED_KHR);
    }

    auto swapRet = builder.build();

    if (!swapRet) {
        // Destroy old swapchain now that we failed (it can't be used either)
        if (oldSwapchain) {
            vkDestroySwapchainKHR(device, oldSwapchain, nullptr);
            swapchain = VK_NULL_HANDLE;
        }
        LOG_ERROR("Failed to recreate swapchain: ", swapRet.error().message());
        // Keep swapchainDirty=true so the next frame retries
        swapchainDirty = true;
        return false;
    }

    // Success — safe to retire the old swapchain
    if (oldSwapchain) {
        vkDestroySwapchainKHR(device, oldSwapchain, nullptr);
    }

    auto vkbSwap = swapRet.value();
    swapchain = vkbSwap.swapchain;
    swapchainFormat = vkbSwap.image_format;
    swapchainExtent = vkbSwap.extent;
    swapchainImages = vkbSwap.get_images().value();
    swapchainImageViews = vkbSwap.get_image_views().value();

    // Recreate depth buffer + MSAA color image + depth resolve image
    destroyMsaaColorImage();
    destroyDepthResolveImage();
    destroyDepthBuffer();

    // Destroy old render pass (needs recreation if MSAA changed)
    if (imguiRenderPass) {
        vkDestroyRenderPass(device, imguiRenderPass, nullptr);
        imguiRenderPass = VK_NULL_HANDLE;
    }

    if (!createDepthBuffer()) return false;
    if (!createMsaaColorImage()) return false;
    if (!createDepthResolveImage()) return false;

    bool useMsaa = (msaaSamples_ > VK_SAMPLE_COUNT_1_BIT);

    if (useMsaa) {
        const bool useDepthResolve = (depthResolveImageView != VK_NULL_HANDLE);
        // MSAA render pass: 3 or 4 attachments
        VkAttachmentDescription attachments[4] = {};
        attachments[0].format = swapchainFormat;
        attachments[0].samples = msaaSamples_;
        attachments[0].loadOp = VK_ATTACHMENT_LOAD_OP_CLEAR;
        attachments[0].storeOp = VK_ATTACHMENT_STORE_OP_DONT_CARE;
        attachments[0].stencilLoadOp = VK_ATTACHMENT_LOAD_OP_DONT_CARE;
        attachments[0].stencilStoreOp = VK_ATTACHMENT_STORE_OP_DONT_CARE;
        attachments[0].initialLayout = VK_IMAGE_LAYOUT_UNDEFINED;
        attachments[0].finalLayout = VK_IMAGE_LAYOUT_COLOR_ATTACHMENT_OPTIMAL;

        attachments[1].format = depthFormat;
        attachments[1].samples = msaaSamples_;
        attachments[1].loadOp = VK_ATTACHMENT_LOAD_OP_CLEAR;
        attachments[1].storeOp = VK_ATTACHMENT_STORE_OP_STORE;
        attachments[1].stencilLoadOp = VK_ATTACHMENT_LOAD_OP_DONT_CARE;
        attachments[1].stencilStoreOp = VK_ATTACHMENT_STORE_OP_DONT_CARE;
        attachments[1].initialLayout = VK_IMAGE_LAYOUT_UNDEFINED;
        attachments[1].finalLayout = VK_IMAGE_LAYOUT_DEPTH_STENCIL_ATTACHMENT_OPTIMAL;

        attachments[2].format = swapchainFormat;
        attachments[2].samples = VK_SAMPLE_COUNT_1_BIT;
        attachments[2].loadOp = VK_ATTACHMENT_LOAD_OP_DONT_CARE;
        attachments[2].storeOp = VK_ATTACHMENT_STORE_OP_STORE;
        attachments[2].stencilLoadOp = VK_ATTACHMENT_LOAD_OP_DONT_CARE;
        attachments[2].stencilStoreOp = VK_ATTACHMENT_STORE_OP_DONT_CARE;
        attachments[2].initialLayout = VK_IMAGE_LAYOUT_UNDEFINED;
        attachments[2].finalLayout = VK_IMAGE_LAYOUT_PRESENT_SRC_KHR;

        if (useDepthResolve) {
            attachments[3].format = depthFormat;
            attachments[3].samples = VK_SAMPLE_COUNT_1_BIT;
            attachments[3].loadOp = VK_ATTACHMENT_LOAD_OP_DONT_CARE;
            attachments[3].storeOp = VK_ATTACHMENT_STORE_OP_STORE;
            attachments[3].stencilLoadOp = VK_ATTACHMENT_LOAD_OP_DONT_CARE;
            attachments[3].stencilStoreOp = VK_ATTACHMENT_STORE_OP_DONT_CARE;
            attachments[3].initialLayout = VK_IMAGE_LAYOUT_UNDEFINED;
            attachments[3].finalLayout = VK_IMAGE_LAYOUT_DEPTH_STENCIL_ATTACHMENT_OPTIMAL;
        }

        if (useDepthResolve) {
            VkAttachmentDescription2 attachments2[4]{};
            for (int i = 0; i < 4; ++i) {
                attachments2[i].sType = VK_STRUCTURE_TYPE_ATTACHMENT_DESCRIPTION_2;
                attachments2[i].format = attachments[i].format;
                attachments2[i].samples = attachments[i].samples;
                attachments2[i].loadOp = attachments[i].loadOp;
                attachments2[i].storeOp = attachments[i].storeOp;
                attachments2[i].stencilLoadOp = attachments[i].stencilLoadOp;
                attachments2[i].stencilStoreOp = attachments[i].stencilStoreOp;
                attachments2[i].initialLayout = attachments[i].initialLayout;
                attachments2[i].finalLayout = attachments[i].finalLayout;
            }

            VkAttachmentReference2 colorRef2{};
            colorRef2.sType = VK_STRUCTURE_TYPE_ATTACHMENT_REFERENCE_2;
            colorRef2.attachment = 0;
            colorRef2.layout = VK_IMAGE_LAYOUT_COLOR_ATTACHMENT_OPTIMAL;
            VkAttachmentReference2 depthRef2{};
            depthRef2.sType = VK_STRUCTURE_TYPE_ATTACHMENT_REFERENCE_2;
            depthRef2.attachment = 1;
            depthRef2.layout = VK_IMAGE_LAYOUT_DEPTH_STENCIL_ATTACHMENT_OPTIMAL;
            VkAttachmentReference2 resolveRef2{};
            resolveRef2.sType = VK_STRUCTURE_TYPE_ATTACHMENT_REFERENCE_2;
            resolveRef2.attachment = 2;
            resolveRef2.layout = VK_IMAGE_LAYOUT_COLOR_ATTACHMENT_OPTIMAL;
            VkAttachmentReference2 depthResolveRef2{};
            depthResolveRef2.sType = VK_STRUCTURE_TYPE_ATTACHMENT_REFERENCE_2;
            depthResolveRef2.attachment = 3;
            depthResolveRef2.layout = VK_IMAGE_LAYOUT_DEPTH_STENCIL_ATTACHMENT_OPTIMAL;

            VkSubpassDescriptionDepthStencilResolve dsResolve{};
            dsResolve.sType = VK_STRUCTURE_TYPE_SUBPASS_DESCRIPTION_DEPTH_STENCIL_RESOLVE;
            dsResolve.depthResolveMode = depthResolveMode_;
            dsResolve.stencilResolveMode = VK_RESOLVE_MODE_NONE;
            dsResolve.pDepthStencilResolveAttachment = &depthResolveRef2;

            VkSubpassDescription2 subpass2{};
            subpass2.sType = VK_STRUCTURE_TYPE_SUBPASS_DESCRIPTION_2;
            subpass2.pNext = &dsResolve;
            subpass2.pipelineBindPoint = VK_PIPELINE_BIND_POINT_GRAPHICS;
            subpass2.colorAttachmentCount = 1;
            subpass2.pColorAttachments = &colorRef2;
            subpass2.pDepthStencilAttachment = &depthRef2;
            subpass2.pResolveAttachments = &resolveRef2;

            VkSubpassDependency2 dep2{};
            dep2.sType = VK_STRUCTURE_TYPE_SUBPASS_DEPENDENCY_2;
            dep2.srcSubpass = VK_SUBPASS_EXTERNAL;
            dep2.dstSubpass = 0;
            dep2.srcStageMask = VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT | VK_PIPELINE_STAGE_EARLY_FRAGMENT_TESTS_BIT;
            dep2.dstStageMask = VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT | VK_PIPELINE_STAGE_EARLY_FRAGMENT_TESTS_BIT;
            dep2.dstAccessMask = VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT | VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_WRITE_BIT;

            VkRenderPassCreateInfo2 rpInfo2{};
            rpInfo2.sType = VK_STRUCTURE_TYPE_RENDER_PASS_CREATE_INFO_2;
            rpInfo2.attachmentCount = 4;
            rpInfo2.pAttachments = attachments2;
            rpInfo2.subpassCount = 1;
            rpInfo2.pSubpasses = &subpass2;
            rpInfo2.dependencyCount = 1;
            rpInfo2.pDependencies = &dep2;

            if (vkCreateRenderPass2(device, &rpInfo2, nullptr, &imguiRenderPass) != VK_SUCCESS) {
                LOG_ERROR("Failed to recreate MSAA render pass (depth resolve)");
                return false;
            }
        } else {
            VkAttachmentReference colorRef{0, VK_IMAGE_LAYOUT_COLOR_ATTACHMENT_OPTIMAL};
            VkAttachmentReference depthRef{1, VK_IMAGE_LAYOUT_DEPTH_STENCIL_ATTACHMENT_OPTIMAL};
            VkAttachmentReference resolveRef{2, VK_IMAGE_LAYOUT_COLOR_ATTACHMENT_OPTIMAL};

            VkSubpassDescription subpass{};
            subpass.pipelineBindPoint = VK_PIPELINE_BIND_POINT_GRAPHICS;
            subpass.colorAttachmentCount = 1;
            subpass.pColorAttachments = &colorRef;
            subpass.pDepthStencilAttachment = &depthRef;
            subpass.pResolveAttachments = &resolveRef;

            VkSubpassDependency dependency{};
            dependency.srcSubpass = VK_SUBPASS_EXTERNAL;
            dependency.dstSubpass = 0;
            dependency.srcStageMask = VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT | VK_PIPELINE_STAGE_EARLY_FRAGMENT_TESTS_BIT;
            dependency.dstStageMask = VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT | VK_PIPELINE_STAGE_EARLY_FRAGMENT_TESTS_BIT;
            dependency.srcAccessMask = 0;
            dependency.dstAccessMask = VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT | VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_WRITE_BIT;

            VkRenderPassCreateInfo rpInfo{};
            rpInfo.sType = VK_STRUCTURE_TYPE_RENDER_PASS_CREATE_INFO;
            rpInfo.attachmentCount = 3;
            rpInfo.pAttachments = attachments;
            rpInfo.subpassCount = 1;
            rpInfo.pSubpasses = &subpass;
            rpInfo.dependencyCount = 1;
            rpInfo.pDependencies = &dependency;

            if (vkCreateRenderPass(device, &rpInfo, nullptr, &imguiRenderPass) != VK_SUCCESS) {
                LOG_ERROR("Failed to recreate MSAA render pass");
                return false;
            }
        }

        swapchainFramebuffers.resize(swapchainImageViews.size());
        for (size_t i = 0; i < swapchainImageViews.size(); i++) {
            VkImageView fbAttachments[4] = {msaaColorView_, depthImageView, swapchainImageViews[i], depthResolveImageView};
            VkFramebufferCreateInfo fbInfo{};
            fbInfo.sType = VK_STRUCTURE_TYPE_FRAMEBUFFER_CREATE_INFO;
            fbInfo.renderPass = imguiRenderPass;
            fbInfo.attachmentCount = useDepthResolve ? 4 : 3;
            fbInfo.pAttachments = fbAttachments;
            fbInfo.width = swapchainExtent.width;
            fbInfo.height = swapchainExtent.height;
            fbInfo.layers = 1;
            if (vkCreateFramebuffer(device, &fbInfo, nullptr, &swapchainFramebuffers[i]) != VK_SUCCESS) {
                LOG_ERROR("Failed to recreate MSAA swapchain framebuffer ", i);
                return false;
            }
        }
    } else {
        // Non-MSAA render pass: 2 attachments
        VkAttachmentDescription attachments[2] = {};
        attachments[0].format = swapchainFormat;
        attachments[0].samples = VK_SAMPLE_COUNT_1_BIT;
        attachments[0].loadOp = VK_ATTACHMENT_LOAD_OP_CLEAR;
        attachments[0].storeOp = VK_ATTACHMENT_STORE_OP_STORE;
        attachments[0].stencilLoadOp = VK_ATTACHMENT_LOAD_OP_DONT_CARE;
        attachments[0].stencilStoreOp = VK_ATTACHMENT_STORE_OP_DONT_CARE;
        attachments[0].initialLayout = VK_IMAGE_LAYOUT_UNDEFINED;
        attachments[0].finalLayout = VK_IMAGE_LAYOUT_PRESENT_SRC_KHR;

        attachments[1].format = depthFormat;
        attachments[1].samples = VK_SAMPLE_COUNT_1_BIT;
        attachments[1].loadOp = VK_ATTACHMENT_LOAD_OP_CLEAR;
        attachments[1].storeOp = VK_ATTACHMENT_STORE_OP_STORE;
        attachments[1].stencilLoadOp = VK_ATTACHMENT_LOAD_OP_DONT_CARE;
        attachments[1].stencilStoreOp = VK_ATTACHMENT_STORE_OP_DONT_CARE;
        attachments[1].initialLayout = VK_IMAGE_LAYOUT_UNDEFINED;
        attachments[1].finalLayout = VK_IMAGE_LAYOUT_DEPTH_STENCIL_ATTACHMENT_OPTIMAL;

        VkAttachmentReference colorRef{0, VK_IMAGE_LAYOUT_COLOR_ATTACHMENT_OPTIMAL};
        VkAttachmentReference depthRef{1, VK_IMAGE_LAYOUT_DEPTH_STENCIL_ATTACHMENT_OPTIMAL};

        VkSubpassDescription subpass{};
        subpass.pipelineBindPoint = VK_PIPELINE_BIND_POINT_GRAPHICS;
        subpass.colorAttachmentCount = 1;
        subpass.pColorAttachments = &colorRef;
        subpass.pDepthStencilAttachment = &depthRef;

        VkSubpassDependency dependency{};
        dependency.srcSubpass = VK_SUBPASS_EXTERNAL;
        dependency.dstSubpass = 0;
        dependency.srcStageMask = VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT | VK_PIPELINE_STAGE_EARLY_FRAGMENT_TESTS_BIT;
        dependency.dstStageMask = VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT | VK_PIPELINE_STAGE_EARLY_FRAGMENT_TESTS_BIT;
        dependency.srcAccessMask = 0;
        dependency.dstAccessMask = VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT | VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_WRITE_BIT;

        VkRenderPassCreateInfo rpInfo{};
        rpInfo.sType = VK_STRUCTURE_TYPE_RENDER_PASS_CREATE_INFO;
        rpInfo.attachmentCount = 2;
        rpInfo.pAttachments = attachments;
        rpInfo.subpassCount = 1;
        rpInfo.pSubpasses = &subpass;
        rpInfo.dependencyCount = 1;
        rpInfo.pDependencies = &dependency;

        if (vkCreateRenderPass(device, &rpInfo, nullptr, &imguiRenderPass) != VK_SUCCESS) {
            LOG_ERROR("Failed to recreate render pass");
            return false;
        }

        swapchainFramebuffers.resize(swapchainImageViews.size());
        for (size_t i = 0; i < swapchainImageViews.size(); i++) {
            VkImageView fbAttachments[2] = {swapchainImageViews[i], depthImageView};
            VkFramebufferCreateInfo fbInfo{};
            fbInfo.sType = VK_STRUCTURE_TYPE_FRAMEBUFFER_CREATE_INFO;
            fbInfo.renderPass = imguiRenderPass;
            fbInfo.attachmentCount = 2;
            fbInfo.pAttachments = fbAttachments;
            fbInfo.width = swapchainExtent.width;
            fbInfo.height = swapchainExtent.height;
            fbInfo.layers = 1;
            if (vkCreateFramebuffer(device, &fbInfo, nullptr, &swapchainFramebuffers[i]) != VK_SUCCESS) {
                LOG_ERROR("Failed to recreate swapchain framebuffer ", i);
                return false;
            }
        }
    }

    swapchainDirty = false;
    LOG_INFO("Swapchain recreated: ", swapchainExtent.width, "x", swapchainExtent.height);
    return true;
}

VkCommandBuffer VkContext::beginFrame(uint32_t& imageIndex) {
    if (deviceLost_) return VK_NULL_HANDLE;
    if (swapchain == VK_NULL_HANDLE) return VK_NULL_HANDLE;  // Swapchain lost; recreate pending

    auto& frame = frames[currentFrame];

    // Wait for this frame's fence (with timeout to detect GPU hangs)
    static int beginFrameCounter = 0;
    beginFrameCounter++;
    VkResult fenceResult = vkWaitForFences(device, 1, &frame.inFlightFence, VK_TRUE, 5000000000ULL); // 5 second timeout
    if (fenceResult == VK_TIMEOUT) {
        LOG_ERROR("beginFrame[", beginFrameCounter, "] FENCE TIMEOUT (5s) on frame slot ", currentFrame, " — GPU hang detected!");
        return VK_NULL_HANDLE;
    }
    if (fenceResult != VK_SUCCESS) {
        LOG_ERROR("beginFrame[", beginFrameCounter, "] fence wait failed: ", (int)fenceResult);
        if (fenceResult == VK_ERROR_DEVICE_LOST) {
            deviceLost_ = true;
        }
        return VK_NULL_HANDLE;
    }

    // Acquire next swapchain image
    VkResult result = vkAcquireNextImageKHR(device, swapchain, UINT64_MAX,
        frame.imageAvailableSemaphore, VK_NULL_HANDLE, &imageIndex);

    if (result == VK_ERROR_OUT_OF_DATE_KHR) {
        swapchainDirty = true;
        return VK_NULL_HANDLE;
    }
    if (result != VK_SUCCESS && result != VK_SUBOPTIMAL_KHR) {
        LOG_ERROR("Failed to acquire swapchain image");
        return VK_NULL_HANDLE;
    }

    vkResetFences(device, 1, &frame.inFlightFence);
    vkResetCommandBuffer(frame.commandBuffer, 0);

    VkCommandBufferBeginInfo beginInfo{};
    beginInfo.sType = VK_STRUCTURE_TYPE_COMMAND_BUFFER_BEGIN_INFO;
    beginInfo.flags = VK_COMMAND_BUFFER_USAGE_ONE_TIME_SUBMIT_BIT;

    vkBeginCommandBuffer(frame.commandBuffer, &beginInfo);

    return frame.commandBuffer;
}

void VkContext::endFrame(VkCommandBuffer cmd, uint32_t imageIndex) {
    static int endFrameCounter = 0;
    endFrameCounter++;

    VkResult endResult = vkEndCommandBuffer(cmd);
    if (endResult != VK_SUCCESS) {
        LOG_ERROR("endFrame[", endFrameCounter, "] vkEndCommandBuffer FAILED: ", (int)endResult);
    }

    auto& frame = frames[currentFrame];

    VkPipelineStageFlags waitStage = VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT;

    VkSubmitInfo submitInfo{};
    submitInfo.sType = VK_STRUCTURE_TYPE_SUBMIT_INFO;
    submitInfo.waitSemaphoreCount = 1;
    submitInfo.pWaitSemaphores = &frame.imageAvailableSemaphore;
    submitInfo.pWaitDstStageMask = &waitStage;
    submitInfo.commandBufferCount = 1;
    submitInfo.pCommandBuffers = &cmd;
    submitInfo.signalSemaphoreCount = 1;
    submitInfo.pSignalSemaphores = &frame.renderFinishedSemaphore;

    VkResult submitResult = vkQueueSubmit(graphicsQueue, 1, &submitInfo, frame.inFlightFence);
    if (submitResult != VK_SUCCESS) {
        LOG_ERROR("endFrame[", endFrameCounter, "] vkQueueSubmit FAILED: ", (int)submitResult);
        if (submitResult == VK_ERROR_DEVICE_LOST) {
            deviceLost_ = true;
        }
    }

    VkPresentInfoKHR presentInfo{};
    presentInfo.sType = VK_STRUCTURE_TYPE_PRESENT_INFO_KHR;
    presentInfo.waitSemaphoreCount = 1;
    presentInfo.pWaitSemaphores = &frame.renderFinishedSemaphore;
    presentInfo.swapchainCount = 1;
    presentInfo.pSwapchains = &swapchain;
    presentInfo.pImageIndices = &imageIndex;

    VkResult result = vkQueuePresentKHR(presentQueue, &presentInfo);
    if (result == VK_ERROR_OUT_OF_DATE_KHR || result == VK_SUBOPTIMAL_KHR) {
        swapchainDirty = true;
    }

    currentFrame = (currentFrame + 1) % MAX_FRAMES_IN_FLIGHT;
}

VkCommandBuffer VkContext::beginSingleTimeCommands() {
    VkCommandBufferAllocateInfo allocInfo{};
    allocInfo.sType = VK_STRUCTURE_TYPE_COMMAND_BUFFER_ALLOCATE_INFO;
    allocInfo.commandPool = immCommandPool;
    allocInfo.level = VK_COMMAND_BUFFER_LEVEL_PRIMARY;
    allocInfo.commandBufferCount = 1;

    VkCommandBuffer cmd;
    vkAllocateCommandBuffers(device, &allocInfo, &cmd);

    VkCommandBufferBeginInfo beginInfo{};
    beginInfo.sType = VK_STRUCTURE_TYPE_COMMAND_BUFFER_BEGIN_INFO;
    beginInfo.flags = VK_COMMAND_BUFFER_USAGE_ONE_TIME_SUBMIT_BIT;
    vkBeginCommandBuffer(cmd, &beginInfo);

    return cmd;
}

void VkContext::endSingleTimeCommands(VkCommandBuffer cmd) {
    vkEndCommandBuffer(cmd);

    VkSubmitInfo submitInfo{};
    submitInfo.sType = VK_STRUCTURE_TYPE_SUBMIT_INFO;
    submitInfo.commandBufferCount = 1;
    submitInfo.pCommandBuffers = &cmd;

    vkQueueSubmit(graphicsQueue, 1, &submitInfo, immFence);
    vkWaitForFences(device, 1, &immFence, VK_TRUE, UINT64_MAX);
    vkResetFences(device, 1, &immFence);

    vkFreeCommandBuffers(device, immCommandPool, 1, &cmd);
}

void VkContext::immediateSubmit(std::function<void(VkCommandBuffer cmd)>&& function) {
    if (inUploadBatch_) {
        // Record into the batch command buffer — no submit, no fence wait
        function(batchCmd_);
        return;
    }
    VkCommandBuffer cmd = beginSingleTimeCommands();
    function(cmd);
    endSingleTimeCommands(cmd);
}

void VkContext::beginUploadBatch() {
    uploadBatchDepth_++;
    if (inUploadBatch_) return; // already in a batch (nested call)
    inUploadBatch_ = true;
    batchCmd_ = beginSingleTimeCommands();
}

void VkContext::endUploadBatch() {
    if (uploadBatchDepth_ <= 0) return;
    uploadBatchDepth_--;
    if (uploadBatchDepth_ > 0) return; // still inside an outer batch

    inUploadBatch_ = false;

    if (batchStagingBuffers_.empty()) {
        // No GPU copies were recorded — skip the submit entirely.
        vkEndCommandBuffer(batchCmd_);
        vkFreeCommandBuffers(device, immCommandPool, 1, &batchCmd_);
        batchCmd_ = VK_NULL_HANDLE;
        return;
    }

    // Submit commands with a NEW fence — don't wait, let GPU work in parallel.
    vkEndCommandBuffer(batchCmd_);

    VkFenceCreateInfo fenceInfo{};
    fenceInfo.sType = VK_STRUCTURE_TYPE_FENCE_CREATE_INFO;
    VkFence fence = VK_NULL_HANDLE;
    vkCreateFence(device, &fenceInfo, nullptr, &fence);

    VkSubmitInfo submitInfo{};
    submitInfo.sType = VK_STRUCTURE_TYPE_SUBMIT_INFO;
    submitInfo.commandBufferCount = 1;
    submitInfo.pCommandBuffers = &batchCmd_;
    vkQueueSubmit(graphicsQueue, 1, &submitInfo, fence);

    // Stash everything for later cleanup when fence signals
    InFlightBatch batch;
    batch.fence = fence;
    batch.cmd = batchCmd_;
    batch.stagingBuffers = std::move(batchStagingBuffers_);
    inFlightBatches_.push_back(std::move(batch));

    batchCmd_ = VK_NULL_HANDLE;
    batchStagingBuffers_.clear();
}

void VkContext::endUploadBatchSync() {
    if (uploadBatchDepth_ <= 0) return;
    uploadBatchDepth_--;
    if (uploadBatchDepth_ > 0) return;

    inUploadBatch_ = false;

    if (batchStagingBuffers_.empty()) {
        vkEndCommandBuffer(batchCmd_);
        vkFreeCommandBuffers(device, immCommandPool, 1, &batchCmd_);
        batchCmd_ = VK_NULL_HANDLE;
        return;
    }

    // Synchronous path for load screens — submit and wait
    endSingleTimeCommands(batchCmd_);
    batchCmd_ = VK_NULL_HANDLE;

    for (auto& staging : batchStagingBuffers_) {
        destroyBuffer(allocator, staging);
    }
    batchStagingBuffers_.clear();
}

void VkContext::pollUploadBatches() {
    if (inFlightBatches_.empty()) return;

    for (auto it = inFlightBatches_.begin(); it != inFlightBatches_.end(); ) {
        VkResult result = vkGetFenceStatus(device, it->fence);
        if (result == VK_SUCCESS) {
            // GPU finished — free resources
            for (auto& staging : it->stagingBuffers) {
                destroyBuffer(allocator, staging);
            }
            vkFreeCommandBuffers(device, immCommandPool, 1, &it->cmd);
            vkDestroyFence(device, it->fence, nullptr);
            it = inFlightBatches_.erase(it);
        } else {
            ++it;
        }
    }
}

void VkContext::waitAllUploads() {
    for (auto& batch : inFlightBatches_) {
        vkWaitForFences(device, 1, &batch.fence, VK_TRUE, UINT64_MAX);
        for (auto& staging : batch.stagingBuffers) {
            destroyBuffer(allocator, staging);
        }
        vkFreeCommandBuffers(device, immCommandPool, 1, &batch.cmd);
        vkDestroyFence(device, batch.fence, nullptr);
    }
    inFlightBatches_.clear();
}

void VkContext::deferStagingCleanup(AllocatedBuffer staging) {
    batchStagingBuffers_.push_back(staging);
}

} // namespace rendering
} // namespace wowee