Vulkan Samples 阅读 -- Basics（二）: Pipelines & Descriptor Sets & Dynamic Uniform Buffers

Pipelines

prepare

• loadAssets: 加载模型等资源
• prepareUniformBuffers： VulkanDevice中有创建buffer的函数
  • updateUniformBuffers：将MVP矩阵写入UniformBuffers
• setupDescriptorSetLayout
  • descriptorSetLayoutBinding：绑定UniformBuffers
  • vkCreateDescriptorSetLayout：创建DescriptorSetLayout
  • vkCreatePipelineLayout：创建PipelineLayout
• preparePipelines
  • 准备的 info 有
    • VkPipelineInputAssemblyStateCreateInfo
    • VkPipelineRasterizationStateCreateInfo
    • VkPipelineColorBlendAttachmentState
    • VkPipelineColorBlendStateCreateInfo
    • VkPipelineDepthStencilStateCreateInfo
    • VkPipelineViewportStateCreateInfo
    • VkPipelineMultisampleStateCreateInfo
    • VkPipelineDynamicStateCreateInfo
    • std::array<VkPipelineShaderStageCreateInfo, 2>
  • 将上面的 info 绑定到 VkGraphicsPipelineCreateInfo 中
  • vkCreateGraphicsPipelines：因为需要渲染三个不同窗口。因此需要创建三个Pipeline。

		// We are using this pipeline as the base for the other pipelines (derivatives)
		// Pipeline derivatives can be used for pipelines that share most of their state
		// Depending on the implementation this may result in better performance for pipeline
		// switching and faster creation time
		pipelineCI.flags = VK_PIPELINE_CREATE_ALLOW_DERIVATIVES_BIT;

		// Textured pipeline
		// Phong shading pipeline
		shaderStages[0] = loadShader(getShadersPath() + "pipelines/phong.vert.spv", VK_SHADER_STAGE_VERTEX_BIT);
		shaderStages[1] = loadShader(getShadersPath() + "pipelines/phong.frag.spv", VK_SHADER_STAGE_FRAGMENT_BIT);
		VK_CHECK_RESULT(vkCreateGraphicsPipelines(device, pipelineCache, 1, &pipelineCI, nullptr, &pipelines.phong));

		// All pipelines created after the base pipeline will be derivatives
		pipelineCI.flags = VK_PIPELINE_CREATE_DERIVATIVE_BIT;
		// Base pipeline will be our first created pipeline
		pipelineCI.basePipelineHandle = pipelines.phong;
		// It's only allowed to either use a handle or index for the base pipeline
		// As we use the handle, we must set the index to -1 (see section 9.5 of the specification)
		pipelineCI.basePipelineIndex = -1;

		// Toon shading pipeline
		shaderStages[0] = loadShader(getShadersPath() + "pipelines/toon.vert.spv", VK_SHADER_STAGE_VERTEX_BIT);
		shaderStages[1] = loadShader(getShadersPath() + "pipelines/toon.frag.spv", VK_SHADER_STAGE_FRAGMENT_BIT);
		VK_CHECK_RESULT(vkCreateGraphicsPipelines(device, pipelineCache, 1, &pipelineCI, nullptr, &pipelines.toon));
		// Pipeline for wire frame rendering
		// Non solid rendering is not a mandatory Vulkan feature
		if (deviceFeatures.fillModeNonSolid)
		{
			rasterizationState.polygonMode = VK_POLYGON_MODE_LINE;
			shaderStages[0] = loadShader(getShadersPath() + "pipelines/wireframe.vert.spv", VK_SHADER_STAGE_VERTEX_BIT);
			shaderStages[1] = loadShader(getShadersPath() + "pipelines/wireframe.frag.spv", VK_SHADER_STAGE_FRAGMENT_BIT);
			VK_CHECK_RESULT(vkCreateGraphicsPipelines(device, pipelineCache, 1, &pipelineCI, nullptr, &pipelines.wireframe));
		}

• setupDescriptorPool
  • vkCreateDescriptorPool: 暂时池的 Size 为 VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER=6
• setupDescriptorSet
  • VkDescriptorSetAllocateInfo：需要的已准备的资源有
    - descriptorPool
    - descriptorSetLayout
  • vkAllocateDescriptorSets：分配池
  • vkUpdateDescriptorSets：将资源写入（UniformBuffers）
• buildCommandBuffers
  • VkRenderPassBeginInfo
    • renderPass
    • renderArea
    • pClearValues = VkClearValue
  • loop drawCmdBuffers：drawCmd次数与frameBuffers个数相关，都等于 swapChain.imageCount
    • vkBeginCommandBuffer
    • vkCmdBeginRenderPass
    • vkCmdSetViewport
    • vkCmdSetScissor
    • vkCmdBindDescriptorSets
    • scene.bindBuffers(drawCmdBuffers[i]);
    • 三个视窗渲染命令 loop
      • vkCmdSetViewport：更新ViewPort
      • vkCmdBindPipeline
      • scene.draw(drawCmdBuffers[i]);
        • vkCmdSetLineWidth：如果是画线可以设置线宽
    • drawUI(drawCmdBuffers[i]);
    • vkCmdEndRenderPass
    • vkEndCommandBuffer

render

• draw
  • VulkanExampleBase::prepareFrame()
  • vkQueueSubmit：渲染需要提交的是 drawCmdBuffers
    • commandBufferCount
    • pCommandBuffers = drawCmdBuffers[currentBuffer]
  • VulkanExampleBase::submitFrame()
• updateUniformBuffers

小结

   从运行程序的结果来看，Pipelines 主要展示如何使用多个VkPipeline来渲染场景。应用场景是：可以将一个场景以不同风格的方式，同时渲染到屏幕上。

Dynamic Uniform Buffers

prepare

• loadAssets: 加载模型等资源
• prepareUniformBuffers： VulkanDevice中有创建buffer的函数
  • updateUniformBuffers：将MVP矩阵写入UniformBuffers
• setupDescriptor
  • descriptorSetLayoutBinding：绑定UniformBuffers和 sampler
  • vkCreateDescriptorSetLayout：创建DescriptorSetLayout
  • vkCreateDescriptorPool: 这里比上一节多了一个Pool
  • vkAllocateDescriptorSets
  • vkUpdateDescriptorSets
• preparePipelines
  • vkCreatePipelineLayout：上一节vkCreatePipelineLayout 在 setupDescriptorSetLayout 中
  • 准备的 info 有
    • VkPipelineInputAssemblyStateCreateInfo
    • VkPipelineRasterizationStateCreateInfo
    • VkPipelineColorBlendAttachmentState
    • VkPipelineColorBlendStateCreateInfo
    • VkPipelineDepthStencilStateCreateInfo
    • VkPipelineViewportStateCreateInfo
    • VkPipelineMultisampleStateCreateInfo
    • VkPipelineDynamicStateCreateInfo
    • std::array<VkPipelineShaderStageCreateInfo, 2>
  • 将上面的 info 绑定到 VkGraphicsPipelineCreateInfo 中
  • vkCreateGraphicsPipelines：这里只创建一个 Pipeline
• setupDescriptorPool
  • vkCreateDescriptorPool: 暂时池的 Size 为 VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER=6
• setupDescriptorSet
  • VkDescriptorSetAllocateInfo：需要的已准备的资源有
    - descriptorPool
    - descriptorSetLayout
  • vkAllocateDescriptorSets：分配池
  • vkUpdateDescriptorSets：将资源写入（UniformBuffers）
• buildCommandBuffers
  • VkRenderPassBeginInfo
    • renderPass
    • renderArea
    • pClearValues = VkClearValue
  • loop drawCmdBuffers：drawCmd次数与frameBuffers个数相关，都等于 swapChain.imageCount
    • vkBeginCommandBuffer
    • vkCmdBeginRenderPass
    • vkCmdBindPipeline：多了一个绑定的pipline
    • vkCmdSetViewport
    • vkCmdSetScissor
    • //vkCmdBindDescriptorSets 这里移到下面
    • scene.bindBuffers(drawCmdBuffers[i]);
    • loop 物体数量：重点在这里。使用DescriptorSets可以使用不同的uniform或sample渲染同一个物体
      • vkCmdBindDescriptorSets
      • model.draw(drawCmdBuffers[i]);
    • drawUI(drawCmdBuffers[i]);
    • vkCmdEndRenderPass
    • vkEndCommandBuffer

render

• draw
  • VulkanExampleBase::prepareFrame()
  • vkQueueSubmit：渲染需要提交的是 drawCmdBuffers
    • commandBufferCount
    • pCommandBuffers = drawCmdBuffers[currentBuffer]
  • VulkanExampleBase::submitFrame()
• updateUniformBuffers

小结

   DescriptorSets可以使用不同的uniform或sample渲染同一个物体，这样处理的好处是，如果场景物体相同，可以只加载一个模型，使用 DescriptorSets 小结方式渲染出所有物体。

Descriptor Sets

prepare

• generateCube: 生成模型等资源
• setupVertexDescriptions：绑定模型资源数据 info
  • bindingDescriptions
  • attributeDescriptions
  • pipelineVertexInputStateCreateInfo
• prepareUniformBuffers
  • size_t bufferSize = OBJECT_INSTANCES * dynamicAlignment: 获得所需的 buffer size

  	//没明白这里的作用，暂时理解为判断sizeof(glm::mat4)是否为够minUboAlignment的字节数 相当于alignas(minUboAlignment)
  	size_t minUboAlignment = vulkanDevice->properties.limits.minUniformBufferOffsetAlignment;
  	dynamicAlignment = sizeof(glm::mat4);
  	if (minUboAlignment > 0) {
  		dynamicAlignment = (dynamicAlignment + minUboAlignment - 1) & ~(minUboAlignment - 1);
  	}
  

  • alignedAlloc：对齐分配
  • createBuffer：创建 view 矩阵 uniform 和 dynamic 的 uniform
  • updateUniformBuffers: 更新 uniform 的数据
  • updateDynamicUniformBuffer: 更新 dynamic 的 uniform 数据
  • vkFlushMappedMemoryRanges: 刷新映射范围, 即提交更新
• setupDescriptorSetLayout
  • 这里注意setLayoutBindings绑定的时候多出一个 VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER_DYNAMIC
• setupDescriptorSet
  • writeDescriptorSet: 多出一个 VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER_DYNAMIC
• preparePipelines
  • 多出 VkPipelineDynamicStateCreateInfo 信息
• setupDescriptorPool
  • vkCreateDescriptorPool: 多出 VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER_DYNAMIC
• setupDescriptorSet
  • writeDescriptorSets中 多出 VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER_DYNAMIC
• buildCommandBuffers: drawCmdBuffers具体分配 UNIFORM_BUFFER_DYNAMIC 内容

	// Render multiple objects using different model matrices by dynamically offsetting into one uniform buffer
	for (uint32_t j = 0; j < OBJECT_INSTANCES; j++)
		{
			// One dynamic offset per dynamic descriptor to offset into the ubo containing all model matrices
			uint32_t dynamicOffset = j * static_cast<uint32_t>(dynamicAlignment);
			// Bind the descriptor set for rendering a mesh using the dynamic offset
			vkCmdBindDescriptorSets(drawCmdBuffers[i], VK_PIPELINE_BIND_POINT_GRAPHICS, pipelineLayout, 0, 1, &descriptorSet, 1, &dynamicOffset);

			vkCmdDrawIndexed(drawCmdBuffers[i], indexCount, 1, 0, 0, 0);
		}

render

• draw: 渲染只需更新uniform信息即可
• VulkanExampleBase::prepareFrame()
• vkQueueSubmit：渲染需要提交的是 drawCmdBuffers
  • commandBufferCount
  • pCommandBuffers = drawCmdBuffers[currentBuffer]
• VulkanExampleBase::submitFrame()
• updateDynamicUniformBuffer()

小结

   Dynamic Uniform Buffers：可以使我们自由绑定任意多个 Uniform；应用场景很多。