Vulkan Samples 阅读 -- Basics（三）: Push Constants & Specialization Constants & Texture Mapping

Push Constants

prepare

• loadAssets: 加载模型等资源

• setupSpheres: 为球模型设置随机颜色

• prepareUniformBuffers

  • updateUniformBuffers

• setupDescriptorSetLayout：正常操作，常量在此处进行设置

  	// Define the push constant range used by the pipeline layout
  	// Note that the spec only requires a minimum of 128 bytes, so for passing larger blocks of data you'd use UBOs or SSBOs
  	VkPushConstantRange pushConstantRange{};
  	pushConstantRange.stageFlags = VK_SHADER_STAGE_VERTEX_BIT;
  	pushConstantRange.offset = 0;
  	pushConstantRange.size = sizeof(SpherePushConstantData);
  
  	VkPipelineLayoutCreateInfo pipelineLayoutCreateInfo = vks::initializers::pipelineLayoutCreateInfo(&descriptorSetLayout, 1);
  	pipelineLayoutCreateInfo.pushConstantRangeCount  = 1;
  	pipelineLayoutCreateInfo.pPushConstantRanges = &pushConstantRange;
  	VK_CHECK_RESULT(vkCreatePipelineLayout(device, &pipelineLayoutCreateInfo, nullptr, &pipelineLayout));
  

• setupDescriptorSet

• preparePipelines

• setupDescriptorPool

• setupDescriptorSet

• buildCommandBuffers：常量绑定在这里

  	vkCmdPushConstants(
  			   drawCmdBuffers[i],
  			    pipelineLayout,
  			    VK_SHADER_STAGE_VERTEX_BIT,
  			    0,
  			    sizeof(SpherePushConstantData),
  			    &spheres[j]);
  			model.draw(drawCmdBuffers[i]);
  

• shader中常量的获取

  layout(push_constant) uniform PushConsts {
  	vec4 color;
  	vec4 position;
  } pushConsts;
  

render

• draw
• updateUniformBuffers

小结

   常量的设置比较简单，setupDescriptorSetLayout中设置， buildCommandBuffers中绑定，shader获取即可。

Specialization Constants

prepare

• loadAssets:

• prepareUniformBuffers

  • updateUniformBuffers

• setupDescriptorSetLayout：与常量不同 Specialization Constants 设置只在 preparePipelines 中

• preparePipelines

  	// Host data to take specialization constants from
  	// 常量结构体定义，命名随意，只要靠绑定id获取值
  	struct SpecializationData {
  		// Sets the lighting model used in the fragment "uber" shader
  		uint32_t lightingModel;
  		// Parameter for the toon shading part of the fragment shader
  		float toonDesaturationFactor = 0.5f;
  	} specializationData;
  
  	// Each shader constant of a shader stage corresponds to one map entry
  	//绑定常量实体
  	std::array<VkSpecializationMapEntry, 2> specializationMapEntries;
  	// Shader bindings based on specialization constants are marked by the new "constant_id" layout qualifier:
  	//	layout (constant_id = 0) const int LIGHTING_MODEL = 0;
  	//	layout (constant_id = 1) const float PARAM_TOON_DESATURATION = 0.0f;
  	
  	// Map entry for the lighting model to be used by the fragment shader
  	//id是shader中获取值的索引
  	specializationMapEntries[0].constantID = 0;
  	specializationMapEntries[0].size = sizeof(specializationData.lightingModel);
  	specializationMapEntries[0].offset = 0;
  
  	// Map entry for the toon shader parameter
  	specializationMapEntries[1].constantID = 1;
  	specializationMapEntries[1].size = sizeof(specializationData.toonDesaturationFactor);
  	specializationMapEntries[1].offset = offsetof(SpecializationData, toonDesaturationFactor);
  
  	// Prepare specialization info block for the shader stage
  	VkSpecializationInfo specializationInfo{};
  	specializationInfo.dataSize = sizeof(specializationData);
  	specializationInfo.mapEntryCount = static_cast<uint32_t>(specializationMapEntries.size());
  	specializationInfo.pMapEntries = specializationMapEntries.data();
  	specializationInfo.pData = &specializationData;
  
  	// Create pipelines
  	// All pipelines will use the same "uber" shader and specialization constants to change branching and parameters of that shader
  	shaderStages[0] = loadShader(getShadersPath() + "specializationconstants/uber.vert.spv", VK_SHADER_STAGE_VERTEX_BIT);
  	shaderStages[1] = loadShader(getShadersPath() + "specializationconstants/uber.frag.spv", VK_SHADER_STAGE_FRAGMENT_BIT);
  	// Specialization info is assigned is part of the shader stage (modul) and must be set after creating the module and before creating the pipeline
  	//选择常量绑定在哪个shader中
  	shaderStages[1].pSpecializationInfo = &specializationInfo;
  

  • setupDescriptorPool
  • setupDescriptorSet
  • buildCommandBuffers
  • shader中常量的获取

  	//变量名称可以自定义
  	layout (constant_id = 0) const int LIGHTING_MODEL = 0;
  	layout (constant_id = 1) const float PARAM_TOON_DESATURATION = 0.0f;
  

render

• draw
• updateUniformBuffers

小结

   Specialization 常量是shader域绑定的，只在preparePipelines中设置即可。与 Constants 的区别是，Specialization Constants 作用域是shader域，而 Constants 的作用域为 pipeline。

Texture Mapping

prepare

• loadTexture: 加载图片
  • ktx_uint8_t *textureData 从读图API中将纹理读入
  • 纹理结构体

    VkSampler sampler;
    VkImage image;
    VkImageLayout imageLayout;
    VkDeviceMemory deviceMemory;
    VkImageView view;
    uint32_t width, height; //从textureData 中获取
    uint32_t mipLevels; //从textureData 中获取
    

  • 官方给出两种将图片资源导入GPU的方式, 一种是线性式平铺, 一种是优化式平铺.
    • 线性平铺是可以直接将存储图片信息的内存空间数据直接复制到GPU上, 因此对于纹理的格式和功能支持非常有限
      • vkCreateImage
      • vkGetImageMemoryRequirements
      • vkAllocateMemory
      • vkBindImageMemory
      • vkMapMemory – memcpy – vkUnmapMemory
      • vkCmdPipelineBarrier： 直接将图片从 HOST Barrier 到 FRAGMENT_SHADER
      • flushCommandBuffer
    • 优化式平铺将图片内存空间数据与硬件支持的功能进行匹配, 然后生成specific layout, 这样能支持更多格式和相应功能. 而且Copy速度更快. 但是这种方式存储图片信息, 是存储在drive中, 主机无法访问.
      • vkCreateBuffer: 创建stagingBuffer
      • vkGetBufferMemoryRequirements: 申请buffer使用的内存
      • vkAllocateMemory: 分配内存
      • vkBindBufferMemory: 绑定内存
      • vkMapMemory – memcpy – vkUnmapMemory: 将图片信息复制到 host local staging buffer (specific layout)
      • std::vector bufferCopyRegions: 为每个 mip 设置缓冲区复制区域
      • vkCreateImage: Create optimal tiled target image on the device
        • 其中主要的是 imageCreateInfo.tiling = VK_IMAGE_TILING_OPTIMAL;
      • vkGetImageMemoryRequirements: 申请内存
      • vkAllocateMemory: 分配内存
      • vkBindImageMemory: 绑定资源
      • createCommandBuffer：创建复制命令
      • imageMemoryBarrier：: Image memory barriers for the texture image
        • 为每个pipeline 设置 image layout 过度，下面的操作就是从stagingBuffer中将图片数据复制到GPU
      • vkCmdPipelineBarrier：从 HOST Barrier 到 TRANSFER
      • vkCmdCopyBufferToImage：Copy mip levels from staging buffer
      • vkCmdPipelineBarrier：从TRANSFER Barrier 到 FRAGMENT_SHADER
      • flushCommandBuffer
      • vkFreeMemory：Clean up staging resources
      • vkDestroyBuffer：Clean up staging resources
  • ktxTexture_Destroy：释放从磁盘读取图片的内存
  • vkCreateSampler
  • vkCreateImageView
• generateQuad：创建一个面片
• setupVertexDescriptions
• prepareUniformBuffers
  • updateUniformBuffers
• setupDescriptorSetLayout
• preparePipelines
• setupDescriptorPool
• setupDescriptorSet
• buildCommandBuffers

render

• draw

小结

   本节主要讲如何更高效的读取贴图。与一般的直接从本地读取贴图数据，然后直接复制到GPU相比，使用 specific layout，能支持更多的格式和功能。并且有更高效的性能。