Vulkan重建交换链时出现内存泄漏问题求助

正在学习Vulkan教程时，发现反复最小化窗口会引发内存异常：每次最小化再最大化窗口，应用内存都会激增约100MB。窗口最小化时调用vkAcquireNextImageKHR会触发VK_ERROR_OUT_OF_DATE_KHR错误，因此会执行交换链重建逻辑。

渲染循环相关代码 #

vkWaitForFences(_deviceManager.getDevice(), 1, &_inFlightFences[_currentFrame], VK_TRUE, UINT64_MAX);

uint32_t imageIndex;

VkResult result = vkAcquireNextImageKHR(_deviceManager.getDevice(), _swapChainManager.getSwapChain(), UINT64_MAX,
    _imageAvailableSemaphores[_currentFrame], VK_NULL_HANDLE, &imageIndex);

if (result == VK_ERROR_OUT_OF_DATE_KHR) {
    _swapChainManager.recreateSwapChain(_windowManager.getWindow(), _deviceManager.getDevice(), 
        _physicalDeviceManager.getPhysicalDevice(), _surfaceManager.getSurface(), 
        _physicalDeviceManager.getMsaaSamples(), _renderPass);
    return;
}
else if (result != VK_SUCCESS && result != VK_SUBOPTIMAL_KHR) {
    throw std::runtime_error("failed to acquire swap chain image!");
}

// Only reset the fence if we are submitting work
vkResetFences(_deviceManager.getDevice(), 1, &_inFlightFences[_currentFrame]);

代码执行顺序：

• vkWaitForFences(...)
• vkAcquireNextImageKHR(...)
• recreateSwapChain(...)
• vkResetFences(...)

交换链重建代码 #

void SwapChainManager::recreateSwapChain(GLFWwindow* window, VkDevice device, VkPhysicalDevice physicalDevice, 
    VkSurfaceKHR surface, VkSampleCountFlagBits msaaSamples, VkRenderPass renderPass)
{
    int width = 0, height = 0;
    glfwGetFramebufferSize(window, &width, &height);
    while (width == 0 || height == 0) {
        glfwGetFramebufferSize(window, &width, &height);
        glfwWaitEvents();
    }

    vkDeviceWaitIdle(device);

    cleanupSwapChain(device);

    createSwapChain(window, physicalDevice, surface, device, msaaSamples);
    createImageViews(device);

    colorResources.createResources(_swapChainExtent.width, _swapChainExtent.height,
        _swapChainImageFormat, physicalDevice, msaaSamples, device);

    depthResources.createResources(_swapChainExtent.width, _swapChainExtent.height,
        physicalDevice, msaaSamples, device);

    createFramebuffers(device, renderPass);

}

已知信息：cleanupSwapChain()负责销毁交换链占用的内存，已尝试将设备属性查询改为仅执行一次但无效；退出应用时控制台提示未在销毁VkDevice前销毁VkImages、VkImageView，还有部分VkDeviceMemory和VkBuffer（疑似独立问题）。

排查方向与修复建议 #

1. 检查cleanupSwapChain()的资源销毁完整性
   确认cleanupSwapChain()是否完整销毁所有交换链关联资源：

   • 交换链本身（vkDestroySwapchainKHR）
   • 交换链图像对应的VkImageView
   • 颜色缓冲、深度缓冲的VkImage、VkImageView及绑定的VkDeviceMemory
   • 帧缓冲（vkDestroyFramebuffer）
     若colorResources和depthResources的资源未在cleanupSwapChain()中销毁，每次重建都会新增资源，直接导致内存泄漏。

2. 修正渲染循环中的栅栏重置逻辑
   当前触发VK_ERROR_OUT_OF_DATE_KHR后直接return，vkResetFences()无法执行，会导致当前帧栅栏未重置，后续循环可能出现异常甚至资源释放问题。调整逻辑：

   if (result == VK_ERROR_OUT_OF_DATE_KHR) {
       // 重置栅栏后再重建交换链
       vkResetFences(_deviceManager.getDevice(), 1, &_inFlightFences[_currentFrame]);
       _swapChainManager.recreateSwapChain(_windowManager.getWindow(), _deviceManager.getDevice(), 
           _physicalDeviceManager.getPhysicalDevice(), _surfaceManager.getSurface(), 
           _physicalDeviceManager.getMsaaSamples(), _renderPass);
       return;
   }
   

3. 严格遵循资源销毁顺序
   Vulkan对资源销毁顺序有严格要求：

   • 先销毁依赖资源（帧缓冲、ImageView），再销毁底层Image和DeviceMemory
   • 确保cleanupSwapChain()内的销毁顺序与recreateSwapChain()中的创建顺序完全相反：销毁帧缓冲 → 销毁颜色/深度资源 → 销毁ImageView → 销毁交换链

4. 启用Vulkan验证层
   开启VK_LAYER_KHRONOS_validation验证层，它会精准定位未正确销毁的资源，给出具体泄漏点信息，是新手排查内存问题的关键工具。

内容的提问来源于stack exchange，提问作者Caleb Derrickson