Vulkan动态Uniform Buffer渲染崩溃：循环超21次即冻结异常

问题现象 #

实现Vulkan动态Uniform Buffer（Dynamic UBO）时，循环绘制21次程序运行正常，但循环次数设为22次及以上时，程序直接冻结崩溃。设备为GTX 3060，尝试渲染数百个球体时，访问uniformBufferObject[4032]（对应偏移192*21字节）时触发崩溃。bufferSize为UniformBufferObject结构体大小，该结构体包含3个glm::mat4，计算得192字节。

相关代码片段 #

绘制循环代码 #

uint32_t dynamicOffset = 0;

glm::mat4 proj = glm::perspective(fov, (float)screenWidth / (float)screenHeight, zNear, zFar);
proj[1][1] *= -1;
glm::mat4 view = glm::lookAt(glm::vec3(0.0f, 0.0f, 2.0f), glm::vec3(0.0f, 0.0f, 0.0f), glm::vec3(0.0f, 1.0f, 0.0f));
    
for (int i = 0; i < 21; i++)
{
    dynamicOffset = bufferSize * i;

    // Update and draw the cube
    uniformBufferObject[i].proj = proj;
    uniformBufferObject[i].view = view;
    uniformBufferObject[i].model = glm::mat4(1.0f);
    uniformBufferObject[i].model = glm::translate(uniformBufferObject[i].model, glm::vec3(0.0f, 0.0f, -3.0f));
    uniformBufferObject[i].model = glm::rotate(uniformBufferObject[i].model, 0.0f, glm::vec3(1.0f, 1.0f, 1.0f));
    uniformBufferObject[i].model = glm::scale(uniformBufferObject[i].model, glm::vec3(0.45f));
    
    vkCmdBindDescriptorSets(commandBuffer, VK_PIPELINE_BIND_POINT_GRAPHICS, graphicsPipelineLayout, 0, 1, &descriptorSet, 1, &dynamicOffset);
    vkCmdDraw(commandBuffer, static_cast<uint32_t>(vertices.size()), 1, 0, 0);
}

Uniform Buffer创建代码 #

void createUniformBuffer()
{
    createBuffer(bufferSize, VK_BUFFER_USAGE_UNIFORM_BUFFER_BIT, VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT | VK_MEMORY_PROPERTY_HOST_COHERENT_BIT, uniformBuffer, uniformBufferMemory);
    
    vkMapMemory(device, uniformBufferMemory, 0, bufferSize, 0, &uniformBufferMapped);
    uniformBufferObject = static_cast<UniformBufferObject*>(uniformBufferMapped);
}

Create Buffer函数 #

void createBuffer(VkDeviceSize size, VkBufferUsageFlags usage, VkMemoryPropertyFlags properties, VkBuffer& buffer, VkDeviceMemory& bufferMemory) 
{
    VkBufferCreateInfo bufferInfo{};
    bufferInfo.sType = VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO;
    bufferInfo.size = size;
    bufferInfo.usage = usage;
    bufferInfo.sharingMode = VK_SHARING_MODE_EXCLUSIVE;

    vkCreateBuffer(device, &bufferInfo, nullptr, &buffer);

    VkMemoryRequirements memRequirements;
    vkGetBufferMemoryRequirements(device, buffer, &memRequirements);

    VkMemoryAllocateInfo allocInfo{};
    allocInfo.sType = VK_STRUCTURE_TYPE_MEMORY_ALLOCATE_INFO;
    allocInfo.allocationSize = memRequirements.size;
    allocInfo.memoryTypeIndex = findMemoryType(memRequirements.memoryTypeBits, properties);

    vkAllocateMemory(device, &allocInfo, nullptr, &bufferMemory);

    vkBindBufferMemory(device, buffer, bufferMemory, 0);
}

findMemoryType函数 #

uint32_t findMemoryType(uint32_t typeFilter, VkMemoryPropertyFlags properties) {
    VkPhysicalDeviceMemoryProperties memProperties;
    vkGetPhysicalDeviceMemoryProperties(physicalDevice, &memProperties);

    for (uint32_t i = 0; i < memProperties.memoryTypeCount; i++) {
        if ((typeFilter & (1 << i)) && (memProperties.memoryTypes[i].propertyFlags & properties) == properties) {
            return i;
        }
    }
}

UBO结构体与描述符相关代码 #

struct UniformBufferObject 
{
    glm::mat4 model;
    glm::mat4 view;
    glm::mat4 proj;
};

VkDescriptorSetLayoutBinding uboLayoutBinding{};
uboLayoutBinding.binding = 0;
uboLayoutBinding.descriptorCount = 1;
uboLayoutBinding.descriptorType = VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER_DYNAMIC;
uboLayoutBinding.stageFlags = VK_SHADER_STAGE_VERTEX_BIT;

std::array<VkDescriptorPoolSize, 2> poolSizes{};
poolSizes[0].type = VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER_DYNAMIC;
poolSizes[0].descriptorCount = static_cast<uint32_t>(1);
poolSizes[1].type = VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER;
poolSizes[1].descriptorCount = static_cast<uint32_t>(1);

descriptorWrites[0].sType = VK_STRUCTURE_TYPE_WRITE_DESCRIPTOR_SET;
descriptorWrites[0].dstSet = descriptorSet;
descriptorWrites[0].dstBinding = 0;
descriptorWrites[0].dstArrayElement = 0;
descriptorWrites[0].descriptorType = VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER_DYNAMIC;
descriptorWrites[0].descriptorCount = 1;
descriptorWrites[0].pBufferInfo = &bufferInfo;

问题原因与修复方案 #

核心问题 #

1. Uniform Buffer内存容量不足：创建Buffer时仅分配了单个UBO的内存（192字节），但循环中尝试访问第22个实例时，偏移量达到192*21=4032字节，远超实际分配的内存，导致内存越界访问触发崩溃。
2. 动态偏移未满足对齐要求：Vulkan要求动态UBO的偏移量必须是设备minUniformBufferOffsetAlignment的整数倍，GTX3060的该值为256字节，当前192字节的UBO大小作为偏移步长违反规范，可能引发设备异常。

修复步骤 #

1. 扩容Uniform Buffer内存

根据需要支持的最大实例数分配足够内存，同时对齐到设备要求的最小偏移量：

// 假设需要支持1000个实例
const uint32_t maxInstanceCount = 1000;
VkPhysicalDeviceProperties deviceProps;
vkGetPhysicalDeviceProperties(physicalDevice, &deviceProps);
uint32_t minUboAlign = deviceProps.limits.minUniformBufferOffsetAlignment;

// 计算单个UBO的对齐后大小
VkDeviceSize alignedUboSize = (sizeof(UniformBufferObject) + minUboAlign - 1) 
                            & ~(minUboAlign - 1);
// 总缓冲区大小
VkDeviceSize totalBufferSize = alignedUboSize * maxInstanceCount;

// 创建扩容后的Buffer
createBuffer(totalBufferSize, VK_BUFFER_USAGE_UNIFORM_BUFFER_BIT, 
             VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT | VK_MEMORY_PROPERTY_HOST_COHERENT_BIT, 
             uniformBuffer, uniformBufferMemory);

// 映射整个缓冲区内存
vkMapMemory(device, uniformBufferMemory, 0, VK_WHOLE_SIZE, 0, &uniformBufferMapped);
uniformBufferObject = static_cast<UniformBufferObject*>(uniformBufferMapped);

2. 使用对齐后的偏移量

循环中计算动态偏移时，改用对齐后的UBO大小作为步长：

// 替换原bufferSize为alignedUboSize
dynamicOffset = alignedUboSize * i;

3. 更新描述符缓冲区信息

确保描述符写入时指向整个大缓冲区：

VkDescriptorBufferInfo bufferInfo{};
bufferInfo.buffer = uniformBuffer;
bufferInfo.offset = 0;
bufferInfo.range = totalBufferSize; // 或使用VK_WHOLE_SIZE

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