Vulkan Samples 阅读 -- Performance（二）: Occlusion Queries & Pipeline Statistics

Occlusion Queries

prepare

• loadAssets: 加载模型(茶壶, 遮挡平面, 球)
• setupQueryPool
  • vkCreateQueryPool: 创建一个查询池
• prepareUniformBuffers
  • createBuffer: Vertex shader, Teapot, Sphere
  • updateUniformBuffers
    • uboVS.visible = (passedSamples[0] > 0) ? 1.0f : 0.0f;
    • 需要将查询结果存入UniformBuffers
• setupDescriptorSetLayout
• preparePipelines:
  • vkCreateGraphicsPipelines: Solid rendering pipeline
  • vkCreateGraphicsPipelines: 画遮挡pass
  • vkCreateGraphicsPipelines: 绘制, 开启blend
• setupDescriptorPool
• setupDescriptorSets
• buildCommandBuffers
  • loop
    • vkBeginCommandBuffer
    • vkCmdResetQueryPool: 必须在renderpass之前初始化
    • vkCmdBeginRenderPass
    • vkCmdSetViewport
    • scissor
    • vkCmdSetScissor
    • 遮挡计算pass
      • vkCmdBindPipeline:
      • vkCmdBindDescriptorSets
      • draw: Occluder 遮挡物先绘制
      • draw: Teapot
      • draw: Sphere
    • 显示可见性pass
      • vkCmdClearAttachments
      • 茶壶
        • vkCmdBindPipeline
        • vkCmdBindDescriptorSets
        • draw
      • Sphere
        • vkCmdBindDescriptorSets
        • draw
      • 遮挡物
        • vkCmdBindPipeline
        • vkCmdBindDescriptorSets
        • draw
    • vkCmdEndRenderPass
    • vkEndCommandBuffer

render

• draw
  • updateUniformBuffers
  • getQueryResults
    • vkGetQueryPoolResults： 绑定为uniform写入可见性的Samples，在updateUniformBuffers写入

shader

• 绘制遮挡shader没有特殊操作，应该是vulkan根据深度信息自动计算遮挡
• 绘制物体阶段根据uniform 的visible参数判断是否被遮挡

小结

   遮挡处理在游戏中应用比较多。工业使用场景感觉还是很少。本节主要讲了QueryPool的使用方法。以后可以开发一下这个池的应用场景。

Pipeline Statistics

prepare

• loadAssets: 加载模型
• setupQueryPool: 创建用于统计的QueryPool
  • VkQueryPoolCreateInfo
    • VK_QUERY_PIPELINE_STATISTIC_INPUT_ASSEMBLY_VERTICES_BIT
    • VK_QUERY_PIPELINE_STATISTIC_INPUT_ASSEMBLY_PRIMITIVES_BIT
    • VK_QUERY_PIPELINE_STATISTIC_VERTEX_SHADER_INVOCATIONS_BIT
    • VK_QUERY_PIPELINE_STATISTIC_CLIPPING_INVOCATIONS_BIT
    • VK_QUERY_PIPELINE_STATISTIC_CLIPPING_PRIMITIVES_BIT
    • VK_QUERY_PIPELINE_STATISTIC_FRAGMENT_SHADER_INVOCATIONS_BIT
    • VK_QUERY_PIPELINE_STATISTIC_TESSELLATION_CONTROL_SHADER_PATCHES_BIT
    • VK_QUERY_PIPELINE_STATISTIC_TESSELLATION_EVALUATION_SHADER_INVOCATIONS_BIT
  • vkCreateQueryPool: 创建查询池
• prepareUniformBuffers
  • updateUniformBuffers
• setupDescriptorSetLayout
• preparePipelines:
  • VkPipelineRasterizationStateCreateInfo: 设置面剔除方法
  • 开启blend: blendEnable = VK_TRUE;
  • 开启discard: rasterizerDiscardEnable = VK_TRUE;
  • 开启wireframe模式: polygonMode = VK_POLYGON_MODE_LINE;
  • vkCreateGraphicsPipelines: 如果细分开启, shaderStages新增两维用于存放细分阶段shader
• setupDescriptorPool
• setupDescriptorSets
• buildCommandBuffers
  • loop
    • vkBeginCommandBuffer
    • vkCmdResetQueryPool: 必须在renderpass之前初始化
    • vkCmdBeginRenderPass
    • vkCmdSetViewport
    • vkCmdSetScissor
    • vkCmdBeginQuery
    • vkCmdBindPipeline
    • vkCmdBindDescriptorSets
    • vkCmdBindVertexBuffers
    • vkCmdBindIndexBuffer
    • loop
      -vkCmdPushConstants
    • vkCmdEndQuery
    • drawUI
    • vkCmdEndRenderPass
    • vkEndCommandBuffer

render

• draw
  • getQueryResults
    • vkGetQueryPoolResult

shader

• VS: 正常渲染流程
• TESE: 细分写入阶段
  • layout (triangles) in;
  • 设置细分顶点位置
  • 输出更新的细分顶点的其他属性
• TESC: 细分输出阶段

  #version 450
  
  layout (vertices = 3) out;
  
  layout (location = 0) in vec3 inNormal[];
  layout (location = 1) in vec3 inColor[];
  layout (location = 2) in vec3 inViewVec[];
  layout (location = 3) in vec3 inLightVec[];
  
  layout (location = 0) out vec3 outNormal[3];
  layout (location = 1) out vec3 outColor[3];
  layout (location = 2) out vec3 outViewVec[3];
  layout (location = 3) out vec3 outLightVec[3];
  
  void main(void)
  {
      if (gl_InvocationID == 0)
      {
          gl_TessLevelInner[0] = 2.0;
          gl_TessLevelOuter[0] = 1.0;
          gl_TessLevelOuter[1] = 1.0;
          gl_TessLevelOuter[2] = 1.0;
      }
  
      gl_out[gl_InvocationID].gl_Position = gl_in[gl_InvocationID].gl_Position;
  	outNormal[gl_InvocationID] = inNormal[gl_InvocationID];
  	outColor[gl_InvocationID] = inColor[gl_InvocationID];	
  	outViewVec[gl_InvocationID] = inViewVec[gl_InvocationID];	
  	outLightVec[gl_InvocationID] = inLightVec[gl_InvocationID];	
  }
  

• FS: 正常渲染阶段
• 注意
  • 细分最大支持GL_MAX_PATCH_VERTICES
  • 面片的顶点数量默认为3，小于3不产生任何几何图元
  • gl_out保存细分后输出的面片
  • 细分层次控制生成图元操作，gl_TessLevelInner 和 gl_TessLevelOuter
    • gl_TessLevelInner : 表示细分面片的内部区域
    • gl_TessLevelOuter: 表示细分面片的外部区域
    • 三角形细分使用gl_TessLevelOuter 前三个值
    • 三角形细分使用gl_TessLevelInner 第一个值
      • 最大值应该只有四个值
    • gl_TessLevelInner 和gl_TessLevelOuter数值的大小应该是绘制线的分段数量,越多细分面片越多
  • gl_InvocationID应该表示当前是输出图元(PATCH)的id，是gl_out的索引

小结

   本节依然介绍QueryPool的其他使用方法, 但是里面增加了细分的功能, 和一些常用渲染模式的设置. QueryPool的确实用性很强。