Directx11教程(13) D3D11管线(1)

      从本篇教程开始,我们暂停代码的学习,先来了解一下D3D11的管线,这些管线不涉及具体的硬件,而是着重于理解能够支持D3D11的管线实现。

参考资料:

http://fgiesen.wordpress.com/2011/07/01/a-trip-through-the-graphics-pipeline-2011-part-1/

通过前面的教程,我们知道,要用D3D11画一个三角形,我们需要做以下步骤:

 

image

       这些步骤大致分为四个阶段,初始化阶段,数据装配阶段,shader执行阶段,以及合并输出阶段。这些步骤最终会在D3D11硬件设备上上执行。

     我们知道,显卡都有驱动程序,也就是driver,通过driver,windows系统才能和显卡硬件进行交互,完成渲染任务。

     通常显卡(GPU)的driver有D3D driver和OpenGL driver。 由于微软各代D3D 之间并不完全兼容,所以D3D driver有分为D3D9 driver,D3D10 driver,D3D11 driver,  D3D12 driver 等等。

image

 

     首先是我们的3D应用程序, 它通过调用D3D11 API函数,实现创建资源(比如顶点缓冲),设置状态(比如深度模版状态),调用drawIndex函数等等。

     D3D运行库会跟踪我们设置的状态,验证函数的参数是否正确,验证shader代码以及shader链接库, 编译shader(把shader从原始的HLSL编译成DX ASM, dx的汇编格式shader),另外它还有内存管理以及设备管理的功能。DX运行库的会把应用程序的调用最终传送到用户模式driver(UMD)中,从某种程度上来说,我们可以把DX运行库看成一个包装器,它是应用程序和UMD之间的接口

 

    UMD(user mode driver)是指用户模式driver,它其实就是一些动态链接库(dll),完全运行在cpu端。GPU厂商都愿意把更多的功能写入UMD,因为其仅是一个dll,容易调试、可以实现多线程操作(比如一个线程编译shader,一个线程处理纹理),即使UMD崩溃了,也不会引起系统蓝屏,因为它和我们普通的应用程序没有本质区别。

    UMD主要功能:编译shader(把DX ASM编译成特殊的IL,中间语言,再编译成对应硬件的机器码),转化应用程序的状态设置和drawcall到硬件识别的packet,并把packet放入到command buffer,另外UMD也有一些内存管理功能,比如虚拟地址管理。

    UMD最终会产生GPU中各个引擎的workload,比如图形引擎,视频编解码引擎,DMA引擎,computer引擎等等,这些workload都以command buffer的形式传到KMD中,再传给相应的硬件引擎,让它们去做这些工作。

    在UMD中,BLT操作管理是处理2D功能,主要就是buffer操作,比如stretch blt,color buffer clear,msaa resolve,内存copy等待。地址库决定memory的布局,比如tile mode,对齐方式,swizzle操作等待。

    UMD中还有特殊硬件处理层和特殊软件处理层,因为现在driver都是统一架构的,一套driver驱动不同代的显卡,所以UMD中特殊硬件处理层就是对特殊硬件进行一些操作,比如某代显卡有bug,可能就要在这个特殊处理层中进行一些补救操作。针对某些特殊软件的优化就放在特殊软件处理层中处理,比如某些跑分程序。

image

     在UMD和KMD之间还有DXGI,这是微软的DirectX图形基础架构,它的设计主要是进行一些底层的操作,它可以看作是KMD的运行库,现在微软把越来越多的底层管理放入到了DXGI中。比如显存管理,commandbuffer的调度,同步操作,present后缓冲到前缓冲,显示器的管理等等。3D应用程序也可以直接访问DXGI一部分功能,比如枚举系统中使用的显卡。

 

      KMD( kernel mode driver),是指Kernel模式driver,KMD负责直接和硬件打交道,可能在系统中有多个UMD实例,但KMD只能有一个。一旦KMD崩溃,操作系统可能会出现蓝屏错误,KMD主要功能包括:

1、在多个应用程序使用GPU的情况下,KMD通过slice time分时操作来管理应用程序。

    比如在一个时间片内一个app进行物理内存操作,另一个时间片内另一个app初始化GPU,设置显示模式等,在不同app间切换时,就需要context switch,响应中断等。

2、ASCI的初始化,一套driver会对应多代的显卡,ASIC这儿就是指不同代的显卡,它们可能架构不同,KMD要针对当前的硬件,选择合适的ASCI设置。

3. 电源管理,GPU内存资源的分配,回收。

4. command buffer提交到GPU硬件,以便GPU硬件开始buffer中的packet。

下图为d3d11 api和driver的交互框图,从中我们可以清晰的理解D3d api如何转化为硬件的packet,并被提交到硬件引擎的ring buffer。

        image
  • 0
    点赞
  • 1
    收藏
    觉得还不错? 一键收藏
  • 0
    评论
This book describes the Direct3D graphics pipeline, from presentation of scene data to pixels appearing on the screen. The book is organized sequentially following the data °ow through the pipeline from the application to the image displayed on the monitor. Each major section of the pipeline is treated by a part of the book, with chapters and subsections detailing each discrete stage of the pipeline. This section summarizes the contents of the book. Part I begins with a review of basic concepts used in 3D computer graphics and their representations in Direct3D. The IDirect3D9 interface is introduced and device selection is described. The IDirect3DDevice9 interface is introduced and an overview of device methods and internal state is given. Finally, a basic framework is given for a 2D application. Chapter 1 begins with an overview of the entire book. A review is given of display technology and the important concept of gamma correction. The representation of color in Direct3D and the macros for manipulating color values are described. The relevant mathematics of vectors, geometry and matrices are reviewed and summarized. A summary of COM and the IUnknown interface is COM: Component Object Model given. Finally, the coding style conventions followed in this book are presented along with some useful C++ coding techniques. Chapter 2 describes the Direct3D object. Every application instantiates this object to select a device from those available. Available devices advertise their location in the Win32 virtual desktop and their capabilities to applications 34 CHAPTER 1. INTRODUCTION through the Direct3D object. Selecting a device from those available and exam- ining a device's capabilities are described. Multiple monitor considerations are also discussed. Chapter 3 describes the Direct3D device object which provides access to the rendering pipeline. The device is the interface an application will use most often and it has a large amount of internal state that controls every stage of the rendering pipeline. This chapter provides a high-level overview of the device and its associated internal state. Detailed discussion of the device state appears throughout the rest of the book. Chapter 4 describes the basic architecture of a typical Direct3D application. Every 3D application can use 2D operations for manipulating frame bu®er con- tents directly. An application can run in full-screen or windowed modes and the di®erences are presented here. The handling of Windows messages and a ba- sic display processing loop are presented. At times it may be convenient to use GDI in a Direct3D application window and a method for mixing these two Win- dows subsystems is presented. Almost every full-screen application will want to use the cursor management provided by the device. Device color palettes and methods for gamma correction are presented. Part II describes the geometry processing portion of the graphics pipeline. The application delivers scene data to the pipeline in the form of geometric primitives. The pipeline processes the geometric primitives through a series of stages that results in pixels displayed on the monitor. This part describes the start of the pipeline where the processing of geometry takes place. Chapter 5 describes how to construct a scene representing the digital world that is imaged by the imaginary camera of the device. A scene consists of a collection of models drawn in sequence. Models are composed of a collection of graphic primitives. Graphic primitives are composed from streams of vertex and index data de¯ning the shape and appearance of objects in the scene. Vertices and indices are stored in resources created through the device. Chapter 6 covers vertex transformations, vertex blending and user-de¯ned clipping planes. With transformations, primitives can be positioned relative to each other in space. Vertex blending, also called \skinning", allows for smooth mesh interpolation. User-de¯ned clipping planes can be used to provide cut away views of primitives. Chapter 7 covers viewing with a virtual camera and projection onto the viewing plane which is displayed as pixels on the monitor. After modeling, objects are positioned relative to a camera. Objects are then projected from 3D camera space into the viewing plane for conversion into 2D screen pixels. Chapter 8 describes the lighting of geometric primitives. The lighting model is introduced and the supported shading algorithms and light types are de- scribed. Chapter 9 covers programmable vertex shading. Programmable vertex shaders can process the vertex data streams with custom code, producing a single ver- tex that is used for rasterization. The vertex shading machine architecture and instruction set are presented. Part III covers the rasterization portion of the pipeline where geometry is1.1. OVERVIEW 5 converted to a series of pixels for display on the monitor. Geometric primitives are lit based on the lighting of their environment and their material properties. After light has been applied to a primitive, it is scan converted into pixels for processing into the frame bu®er. Textures can be used to provide detailed surface appearance without extensive geometric modeling. Pixel shaders can be used to provide custom per-pixel appearance processing instead of the ¯xed- function pixel processing provided by the stock pipeline. Finally, the pixels generated from the scan conversion process are incorporated into the render target surface by the frame bu®er. Chapter 10 describes the scanline conversion of primitives into pixel frag- ments. Lighting and shading are used to process vertex positions and their associated data into a series of pixel fragments to be processed by the frame bu®er. Chapter 11 describes textures and volumes. Textures provide many e±cient per-pixel e®ects and can be used in a variety of manners. Volumes extend texture images to three dimensions and can be used for a volumetric per-pixel rendering e®ects. Chapter 13 describes programmable pixel shaders. Programmable pixel shaders combine texture map information and interpolated vertex information to produce a source pixel fragment. The pixel shading machine architecture and instruction set are presented. Chapter 14 describes how fragments are processed into the frame bu®er. After pixel shading, fragments are processed by the fog, alpha test, depth test, stencil test, alpha blending, dither, and color channel mask stages of the pipeline before being incorporated into the render target. A render target is presented for display on the monitor and video scan out. Part IV covers the D3DX utility library. D3DX provides an implementation of common operations used by Direct3D client programs. The code in D3DX consists entirely of client code and no system components. An application is free to reimplement the operations provided by D3DX, if necessary. Chapter 15 introduces D3DX and summarizes features not described else- where. Chapter 16 describes the abstract data types provided by D3DX. D3DX provides support for RGBA color, point, vector, plane, quaternion, and matrix data types. Chapter 17 describes the helper COM objects provided by D3DX. D3DX provides a matrix stack object to assist in rendering frame hierarchies, a font object to assist in the rendering of text, a sprite object to assist in the rendering of 2D images, an object to assist in rendering to a surface or an environment map and objects for the rendering of special e®ects. Chapter 19 describes the mesh objects provided by D3DX. The mesh objects provided by D3DX encompass rendering of indexed triangle lists as well as progressive meshes, mesh simpli¯cation and skinned meshes. Chapter 21 describes the X ¯le format with the ¯le extension .x. The X ¯le format provides for extensible hierarchical storage of data objects with object instancing.6 CHAPTER 1. INTRODUCTION Part V covers application level considerations. This part of the book de- scribes issues that are important to applications but aren't strictly part of the graphics pipeline. Debugging strategies for applications are presented. Almost all Direct3D applications will be concerned with performance; API related per- formance issues are discussed here. Finally, installation and deployment issues for Direct3D applications are discussed. Chapter 22 describes debugging strategies for Direct3D applications. This includes using the debug run-time for DirectX 9.0c, techniques for debugging full-screen applications and remote debugging. Chapter 23 covers application performance considerations. All real devices have limitations that a®ect performance. A general consideration of how the pipeline state a®ects performance is given. Chapter 24 covers application installation and setup. Appendix A provides a guided tour of the DirectX SDK materials.

“相关推荐”对你有帮助么?

  • 非常没帮助
  • 没帮助
  • 一般
  • 有帮助
  • 非常有帮助
提交
评论
添加红包

请填写红包祝福语或标题

红包个数最小为10个

红包金额最低5元

当前余额3.43前往充值 >
需支付:10.00
成就一亿技术人!
领取后你会自动成为博主和红包主的粉丝 规则
hope_wisdom
发出的红包
实付
使用余额支付
点击重新获取
扫码支付
钱包余额 0

抵扣说明:

1.余额是钱包充值的虚拟货币,按照1:1的比例进行支付金额的抵扣。
2.余额无法直接购买下载,可以购买VIP、付费专栏及课程。

余额充值