窗口调整时Vulkan的pRenderPassBegin->framebuffer无效问题求助

问题现象 #

我的Vulkan应用在旧笔记本、搭载Intel UHD核显的笔记本上运行正常，但在搭载RTX4060m显卡的新笔记本上，调整窗口大小或切换全屏/窗口模式时，验证层报错renderPassCreateInfo->framebuffer无效，随后程序约1秒后崩溃。

控制台日志 #

Validationl layer: Searching for ICD drivers named .\nvoglv64.dll
Validationl layer: Loading layer library C:\msys2\publicPkg\VulkanSDK\1.3.296.0\Bin\.\VkLayer_khronos_validation.dll
Validationl layer: Loading layer library C:\WINDOWS\System32\DriverStore\FileRepository\nvcvegpu.inf_amd64_2b4a4257593cec26\.\nvoglv64.dll
Validationl layer: Validation Error: [ VUID-VkRenderPassBeginInfo-framebuffer-parameter ] Object 0: handle = 0x1b359d88860, type = VK_OBJECT_TYPE_INSTANCE; | MessageID = 0xe02fa864 | vkCmdBeginRenderPass(): pRenderPassBegin->framebuffer Invalid VkFramebuffer Object 0x1b35c20bd50.
The Vulkan spec states: framebuffer must be a valid VkFramebuffer handle

动态状态配置 #

VkDynamicState dynamicStates[] = {
    VK_DYNAMIC_STATE_VIEWPORT,
    VK_DYNAMIC_STATE_SCISSOR
};

相关代码片段 #

帧缓冲创建函数 #

void __fastcall createFramebuffers(void) {
    swapchainFramebuffers = g_array_sized_new(false, true, sizeof(VkFramebuffer), swapchainImageViews->len);
    g_array_set_size(swapchainFramebuffers, swapchainImageViews->len);
    
    for (uint_fast32_t i = 0; i < swapchainImageViews->len; ++i) {
        VkImageView imageAttachments[] = {
            g_array_index(swapchainImageViews, VkImageView, i)
        };

        VkFramebufferCreateInfo framebufferInfo = (const VkFramebufferCreateInfo){ };
        framebufferInfo.sType = VK_STRUCTURE_TYPE_FRAMEBUFFER_CREATE_INFO;
        framebufferInfo.renderPass = graphicsRenderPass;
        framebufferInfo.attachmentCount = 1;
        framebufferInfo.pAttachments = imageAttachments;
        framebufferInfo.layers = 1;
        framebufferInfo.width = swapchain2DExtent.width;
        framebufferInfo.height = swapchain2DExtent.height;

        if (vkCreateFramebuffer(logicalGpu, &framebufferInfo, NULL, &g_array_index(swapchainFramebuffers, VkFramebuffer, i)) != VK_SUCCESS)
            printf_s("Failed to create framebuffer!");
    }
}

命令缓冲录制函数（问题行标注） #

void __fastcall recordCommandBuffer(VkCommandBuffer graphicsCommandBuffer, uint_fast32_t imageIndex) {
    VkCommandBufferBeginInfo commandBufferUsageInfo = (const VkCommandBufferBeginInfo){ };
    commandBufferUsageInfo.sType = VK_STRUCTURE_TYPE_COMMAND_BUFFER_BEGIN_INFO;
    commandBufferUsageInfo.flags = 0;
    commandBufferUsageInfo.pInheritanceInfo = NULL;

    if (vkBeginCommandBuffer(graphicsCommandBuffer, &commandBufferUsageInfo) != VK_SUCCESS)
        printf_s("Failed to begin recording command buffer!");

    VkRenderPassBeginInfo graphicsRenderPassInfo = (const VkRenderPassBeginInfo){ };
    graphicsRenderPassInfo.sType = VK_STRUCTURE_TYPE_RENDER_PASS_BEGIN_INFO;
    graphicsRenderPassInfo.renderPass = graphicsRenderPass;
    graphicsRenderPassInfo.framebuffer = g_array_index(swapchainFramebuffers, VkFramebuffer, imageIndex); //! PROBLEM
    graphicsRenderPassInfo.renderArea.offset = (VkOffset2D){0, 0};
    graphicsRenderPassInfo.renderArea.extent = swapchain2DExtent;
    graphicsRenderPassInfo.clearValueCount = 1;
    graphicsRenderPassInfo.pClearValues = (VkClearValue[4]){ 0.0f, 0.0f, 0.0f, 1.0f };

    vkCmdBeginRenderPass(graphicsCommandBuffer, &graphicsRenderPassInfo, VK_SUBPASS_CONTENTS_INLINE);
    vkCmdBindPipeline(graphicsCommandBuffer, VK_PIPELINE_BIND_POINT_GRAPHICS, graphicsPipeline);

    VkViewport viewport = (const VkViewport){ };
    viewport.x = 0.0f;
    viewport.y = 0.0f;
    viewport.width = (float)swapchain2DExtent.width;
    viewport.height = (float)swapchain2DExtent.height;
    viewport.minDepth = 0.0f;
    viewport.maxDepth = 1.0f;

    VkRect2D scissor = (const VkRect2D){ };
    scissor.offset = (VkOffset2D){0, 0};
    scissor.extent = swapchain2DExtent;
    
    vkCmdSetViewport(graphicsCommandBuffer, 0, 1, &viewport);
    vkCmdSetScissor(graphicsCommandBuffer, 0, 1, &scissor);

    vkCmdDraw(graphicsCommandBuffer, 3, 1, 0, 0);
    vkCmdEndRenderPass(graphicsCommandBuffer);

    if (vkEndCommandBuffer(graphicsCommandBuffer) != VK_SUCCESS)
        printf_s("Failed to end recording command buffer!");
}

问题原因分析 #

这个问题的核心是窗口尺寸变更后，未正确更新帧缓冲及命令缓冲，导致引用了已失效的VkFramebuffer对象，而NVIDIA驱动对Vulkan对象有效性的校验比Intel核显驱动更严格：

1. 帧缓冲与swapchain尺寸不匹配：窗口resize或全屏切换时，swapchain的extent会发生变化，原来基于旧尺寸创建的帧缓冲会因为尺寸不匹配而失效，必须销毁并重建适配新尺寸的帧缓冲。你的代码仅实现了初始帧缓冲创建，但没有在窗口尺寸变更时执行销毁旧帧缓冲、更新swapchain2DExtent、重建新帧缓冲的流程。

2. 命令缓冲复用未重新录制：如果你的命令缓冲是一次性录制后复用的（当前recordCommandBuffer的flags为0，没有设置允许复用的标记），旧命令缓冲中会保留对旧帧缓冲的引用，即使重建了新帧缓冲，旧命令缓冲仍会引用无效对象，触发验证层报错。

3. 驱动校验差异：Intel核显驱动可能对短暂的无效对象引用容忍度更高，但这属于违反Vulkan规范的行为；NVIDIA驱动的验证层会严格执行规范，及时检测到无效帧缓冲并报错，最终导致程序崩溃。

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