Vulkan Samples 阅读 -- Advanced（一）: Multi Sampling & High Dynamic Range & Shadow Mapping

Multi Sampling

prepare

• getMaxUsableSampleCount：获取当前使用显卡支持最大的多重采样个数
• setupRenderPass：重写父类方法，设置多重采样 attachment
  • std::array<VkAttachmentDescription, 3>
    • attachments[0]： Multisampled attachment
    • attachments[1]：从Multisampled attachment读取渲染结果
    • attachments[2]：Multisampled depth attachment
  • std::array<VkSubpassDependency, 2>
    • dependencies[0]: 向 Multisampled attachment写数据
    • dependencies[1]:从 Multisampled attachment 写入 color attachment
• setupFrameBuffer: 重写父类方法, 设置多重采样 Target
  • setupMultisampleTarget
    • imageCreateInfo(color)
    • vkCreateImage
    • vkGetImageMemoryRequirements-vkAllocateMemory-vkBindImageMemory
    • vkCreateImageView
    • imageCreateInfo(depth)
    • vkCreateImage
    • vkGetImageMemoryRequirements-vkAllocateMemory-vkBindImageMemory
    • vkCreateImageView
  • vkCreateFramebuffer
• loadAssets: 加载模型
• prepareUniformBuffers
  • updateUniformBuffers
• setupDescriptorSetLayout
• preparePipelines:
  • VkGraphicsPipelineCreateInfo: 加入多重采样信息
    • VkPipelineMultisampleStateCreateInfo
      • multisampleState.sType = VK_STRUCTURE_TYPE_PIPELINE_MULTISAMPLE_STATE_CREATE_INFO;
      • multisampleState.rasterizationSamples = sampleCount;
      • multisampleState.sampleShadingEnable = VK_TRUE;
      • multisampleState.minSampleShading = 0.25f;
  • vkCreateGraphicsPipelines
• setupDescriptorPool
• setupDescriptorSet
• buildCommandBuffers

render

• draw
• updateUniformBuffers

shader

- 使用的光照模型为Blinn-Phong

小结

   本节主要讲MSAA的设置方法, 重点是增加了 Multisampled attachment.

High Dynamic Range

prepare

• loadAssets：加载模型, 和HDR贴图(cube map)
• prepareUniformBuffers
  • updateUniformBuffers
• prepareoffscreenfer: 离屏渲染资源
  • createAttachment: 两个color, 一个深度
  • vkCreateRenderPass
  • vkCreateFramebuffer
  • vkCreateSampler
  • Bloom separable filter pass
    • createAttachment: color
    • vkCreateRenderPass
    • vkCreateFramebuffer
    • vkCreateSampler
• setupDescriptorSetLayout
  • vkCreateDescriptorSetLayout-vkCreatePipelineLayout(正常初始化流程)
  • Bloom filter
    • vkCreateDescriptorSetLayout
    • vkCreatePipelineLayout
  • G-Buffer composition
    • vkCreateDescriptorSetLayout
    • vkCreatePipelineLayout
• preparePipelines: 有四个Pipelines
  • 最终渲染管线
  • Bloom pass
  • Second blur pass
  • Skybox pipeline
    • 有常量设置
      • specializationMapEntries
        • 0 为渲染天空盒
        • 1 为渲染物体
  • Object rendering pipeline
• setupDescriptorPool
• setupDescriptorSet : 为四个渲染管线设置 Descriptor
• buildCommandBuffers: 三个阶段
  • Render scene to offscreen framebuffer
    • Skybox
    • 3D object
  • First bloom pass
  • Scene rendering with applied second bloom pass

render

• draw
• updateUniformBuffers

shader

- gbuffer阶段
	- 如果是渲染天空盒, 读取cubmap颜色
	- 如果是渲染物体: 则读取光反射天空盒颜色
- bloom 阶段: 也有常量
	- 模糊天空盒光源部分做泛光
- 渲染场景 composition
	- 读取gbuffer中颜色, 
	- 如果开启泛光, 渲染物体时使用使用泛光

小结

   本节讲的功能是, 如何渲染天空盒. 其实就是cubmap的复习, 常量的使用. 也重点展示了, bloom 的用法. 学习OpenGL中有对应章节教程, 所以可以无视泛光算法.

Shadow Mapping

prepare

• loadAssets：加载场景资源
• prepareOffscreenFramebuffer
  • 准备用于阴影的深度贴图
    • vkCreateImage
    • vkAllocateMemory-vkBindImageMemory-
    • vkCreateImageView
    • vkCreateSampler
  • prepareOffscreenRenderpass
    • attachmentDescription
    • std::array<VkSubpassDependency, 2> dependencies;
    • VkRenderPassCreateInfo
    • vkCreateRenderPass
  • vkCreateFramebuffer
• prepareUniformBuffers
  • updateLight
  • updateUniformBufferOffscreen
  • updateUniformBuffers
• setupDescriptorSetLayout
• preparePipelines: 有三组Pipelines
  • 阴影深度贴图显示
  • 渲染带阴影的场景
    • No filtering
    • PCF filtering
  • 渲染阴影深度贴图
• setupDescriptorPool
• setupDescriptorSet : 为三个渲染管线设置 Descriptor
• buildCommandBuffers: 三个阶段
  • Generate shadow map
    • vkCmdSetDepthBias: vulkan专门设置DepthBias的命令
  • Visualize shadow map: 如果选择显示阴影贴图走此Command
  • 渲染有阴影的场景

render

• draw
• updateLight
• updateUniformBufferOffscreen
• updateUniformBuffers

shader

- offscreen : 只需要相机视角的深度信息
- quad: 渲染深度贴图信息
- scene: 渲染场景

小结

   本节主要讲定向光阴影算法的实现, 基础中的基础. 重点在于, vulkan的vkCmdSetDepthBias命令. 后两节的阴影算法暂时不看, 后期在算法研习阶段再回来阅读.