Vulkan Samples 阅读 -- Compute Shader（三）Cloth Simulation & Cull And LOD

Cloth Simulation

prepare

• queueFamilyIndices.graphics != queueFamilyIndices.compute
  • 区分图形管线与计算管线
• loadAssets：加球模型与布料纹理
• prepareStorageBuffers
  • std::vector: 布料分为两类. 一类覆盖在球上, 一类为下垂的布料
  • createBuffer: stagingBuffer
  • createBuffer: storageBuffers: 输入
  • createBuffer: createBuffer:输出
  • Copy from staging buffer
    • createCommandBuffer
    • vkCmdCopyBuffer: 输入
    • vkCmdCopyBuffer:输出
  • addGraphicsToComputeBarriers: 这里将Barrier单独封装一个方法
    • vkCmdPipelineBarrier
  • flushCommandBuffer
  • destroy
  • std::vector<uint32_t> indices: 准备网格的ibo
  • createBuffer: stagingBuffer
  • createBuffer: indexBuffer
  • Copy from staging buffer
    • createCommandBuffer
    • vkCmdCopyBuffer
    • flushCommandBuffer
  • stagingBuffer.destroy()
• prepareUniformBuffers
  • createBuffer: Compute shader uniform buffer block
  • updateComputeUBO: 准备布料仿真所需的参数
  • createBuffer: 图形管线的uniform
  • updateGraphicsUBO: 图形管线使用的参数
• setupDescriptorPool
  • 这里多设置了一个池: VK_DESCRIPTOR_TYPE_STORAGE_BUFFER
  • vkCreateDescriptorPool
• setupLayoutsAndDescriptors
  • vkCreateDescriptorSetLayout
  • vkCreatePipelineLayout
  • vkAllocateDescriptorSets
  • vkUpdateDescriptorSets
• preparePipelines
  • Rendering pipeline
    • vkCreateGraphicsPipelines: 布料显示
    • vkCreateGraphicsPipelines: 球模型显示
• prepareCompute
  • vkGetDeviceQueue
  • vkCreateDescriptorSetLayout
  • pushConstantRange
  • vkCreatePipelineLayout
  • vkAllocateDescriptorSets
  • vkUpdateDescriptorSets
  • Create pipeline
    • vkCreateComputePipelines
    • vkCreateCommandPool
    • vkAllocateCommandBuffers
    • vkCreateSemaphore: ready
    • vkCreateSemaphore: complete
    • buildComputeCommandBuffer
      • loop 2
        • vkBeginCommandBuffer
        • addGraphicsToComputeBarriers
        • vkCmdBindPipeline
        • vkCmdPushConstants
        • loop Dispatch the compute job: 执行64次
          • vkCmdBindDescriptorSets
          • vkCmdPushConstants
          • vkCmdDispatch: 这里分配工作组
          • addComputeToComputeBarriers
          • addComputeToGraphicsBarriers
          • vkEndCommandBuffer
• buildCommandBuffers
  • loop
    • vkBeginCommandBuffer
    • addComputeToGraphicsBarriers
    • vkCmdBeginRenderPass
    • vkCmdSetViewport
    • vkCmdSetScissor
    • Render sphere
      • vkCmdBindPipeline
      • vkCmdBindDescriptorSets
      • draw
    • Render cloth
      • vkCmdBindPipeline
      • vkCmdBindDescriptorSets
      • vkCmdBindIndexBuffer
      • vkCmdBindVertexBuffers
      • vkCmdDrawIndexed
      • drawUI
      • vkCmdEndRenderPass
      • addGraphicsToComputeBarriers
    • vkEndCommandBuffer

render

• draw
  • vkQueueSubmit
  • prepareFrame
  • vkQueueSubmit
  • submitFrame

shader

• CS
  • layout(std430, binding = 0) buffer ParticleIn
  • layout(std430, binding = 1) buffer ParticleOut
    • 注意std430布局的使用
  • layout (binding = 2) uniform UBO
  • layout (push_constant) uniform PushConsts
  • main
    • 获取索引
      • uvec3 id = gl_GlobalInvocationID;
      • uint index = id.y * params.particleCount.x + id.x;
    • 这里布料仿真分三个阶段
      • 计算受力方向和力
      • 力的作用下布料顶点的偏移
      • 和球模型的碰撞检测
      • 更新布料模型的法线
      • 输出布料数据
• VS: 正常渲染
• FS: 正常渲染

小结

   布料仿真主要还是注意Barrier的使用, 布料模拟有基础, 这里可以忽略.

Cull And LOD

prepare

• loadAssets：加载资源
• prepareBuffers
  • objectCount =64^3
  • std::vector instanceData(objectCount)
  • std::vector indirectCommands;
    • 存储index offsets
    • 存储instance count
    • 初始化instanceCount = 1
    • firstInstance = index
      • index为Object idx
    • createBuffer: stagingBuffer
    • createBuffer: storagebuffer
    • copyBuffer
    • stagingBuffer.destroy()
  • InstanceData
    • createBuffer: storagebuffer: 用于CS写入结果
      • indirectDrawCountBuffer
    • instanceData: 初始化数据的顶点和缩放属性
    • createBuffer: instanceData: stagingBuffer
    • createBuffer: storagebuffer
    • copyBuffer
    • stagingBuffer.destroy();
  • std::vector LODLevels;
    • createBuffer: stagingBuffer
    • createBuffer: storagebuffer
    • copyBuffer
    • destroy
  • createBuffer: Scene uniform buffer
  • updateUniformBuffer
• setupDescriptorSetLayout
  • vkCreateDescriptorSetLayout
  • vkCreatePipelineLayout
• preparePipelines
  - vkCreateGraphicsPipelines
• setupDescriptorPool
  • vkCreateDescriptorPool
• setupDescriptorSet
  • vkAllocateDescriptorSets
  • writeDescriptorSet
  • vkUpdateDescriptorSets
• prepareCompute
  • vkGetDeviceQueue
  • vkCreateDescriptorSetLayout
  • vkCreatePipelineLayout
  • vkAllocateDescriptorSets
  • writeDescriptorSet
    • 共4个
  • vkUpdateDescriptorSets
  • vkCreateComputePipelines
  • vkCreateCommandPool
  • vkAllocateCommandBuffers
  • vkCreateFence
  • vkCreateSemaphore
  • buildComputeCommandBuffer
    • vkBeginCommandBuffer
    • vkCmdPipelineBarrier
    • vkCmdBindPipeline
    • vkCmdBindDescriptorSets
    • vkCmdDispatch: objectCount/16
    • vkCmdPipelineBarrier
    • vkEndCommandBuffer
• buildCommandBuffers
  • loop
    • vkBeginCommandBuffer
    • addComputeToGraphicsBarriers
    • vkCmdBeginRenderPass
    • vkCmdSetViewport
    • vkCmdSetScissor
    • vkCmdBindPipeline
    • vkCmdBindVertexBuffers: 模型
    • vkCmdBindVertexBuffers: instanceBuffer
    • vkCmdBindIndexBuffer
    • vkCmdDrawIndexedIndirect
    • drawUI
    • vkCmdEndRenderPass
    • vkEndCommandBuffer

render

• draw
  • prepareFrame
  • vkWaitForFences
  • vkResetFences
  • vkQueueSubmit
  • prepareFrame
  • memcpy(&indirectStats, indirectDrawCountBuffer.mapped, sizeof(indirectStats))
    • Get draw count from compute
• updateUniformBuffer

shader

• CS
  • layout (binding = 0, std140) buffer Instances
    • 实例属性, pos和scale
  • layout (binding = 1, std430) writeonly buffer IndirectDraws
    • 注意std430布局的使用
  • layout (binding = 2) uniform UBO
  • layout (binding = 3) buffer UBOOut
  • layout (binding = 4) readonly buffer LODs
  • uint idx = gl_GlobalInvocationID.x + gl_GlobalInvocationID.y * gl_NumWorkGroups.x * gl_WorkGroupSize.x;
    • 获取全局idx
  • atomicExchange: atomicExchange(uboOut.drawCount, 0);
    • 读取并修改被封装的值
  • vec4 pos = vec4(instances[idx].pos.xyz, 1.0);
    • 获取当前实例的位置信息
  • atomicAdd(uboOut.drawCount, 1)
    • atomicAdd函数是先赋值后进行加法计算的
    • i++
  • distance
    • 距离函数
• VS:
  • vec4 pos = vec4((inPos.xyz * instanceScale) + instancePos, 1.0);
  • 根据实例属性设置顶点位置
• FS: 正常渲染即可

小结

   本节代码没有看太懂, 但是核心讲的是如何将CS计算的结果写回内存中. 可以回来复习.