Designing a Game s Shader Library for Current and Next Generation Hardware

Transcription

1 ATI TECHNOLOGIES INC. Designing a Game s Shader Library for Current and Next Generation Hardware Alex Vlachos AVlachos@ati.com 1

2 Outline Intro & Motivation Demos Why a Shader Library is Needed ATI Demo Shader File Format Multi-pass Rendering Fallback Shaders Default State Run-Time Implementation 2

3 The Design Dilemma Do you design your engine for yesterday s graphics chips or tomorrow s? Too often, the answer is yesterday s. Why not both? With a little engineering time spent up front, you ll save yourself twice as much time during the rest of your development process. 3

4 What is a Shader Library? Library in the sense of a linkable.lib file or collection of source code (not a collection of shaders). API that deals completely with setting render state, shaders, shader constants, etc. Provides a text file format for describing how to render a piece of geometry from preprocessing to runtime. Completely abstracts the graphics API. Necessary for cross-platform development. Often coupled with a windowing API that abstracts window creation and associated functionality. 4

5 RADEON 8500 Nature Demo 5

6 RADEON 8500 Rachel Demo 6

7 RADEON 8500 Ocean Demo 7

8 RADEON 8500 Tubes Demo 8

9 Future of the Industry Advances in the graphics industry are becoming extremely shader-centric and will continue on that path. Designing a forward-looking engine equates to having a powerful and flexible shader library at the center. Every part of your engine needs to be designed with the shader library in mind. 9

10 Current Games Games are too conservative with shaders these days even on the latest consoles! Stop hard-coding shaders in C/C++ code! You re only slowing yourself down. 10

11 Advantages of Having a Shader Library Add completely new shaders without having to write a line of C code or recompiling your engine! Fix rendering bugs by editing a text file while the app is still running and see instant feedback. Makes debugging vertex and pixel shaders a trivial task in comparison to the hardcoded method. Falling back to older hardware becomes a much simpler task and follows naturally. 11

12 Anatomy of an ATI Demo Shader File at a Glance Art Notes What artists see within Maya or 3DS MAX Stream Mappings Defines how to preprocess your vertex and index buffers. Texture Declaration Defines how to preprocess and store textures for runtime. Variables Define variables for use within the shader file. Artists can set variables from within the art tool. Sub-Shaders There can be many of these to describe fallbacks or rendering wire frame, NPR, etc. Passes Supports multi-pass rendering Set Texture State Set Render State Vertex and Pixel Shader Constant Setup Vertex and Pixel Shader Code 12

13 Artist s Point-of-View Artists associate a shader file with a given material within the art tool. The art tool pulls instructions out of the shader file for the artists. To the RIGHT are our Maya and 3DS MAX material plug-ins. 13

14 Art Notes Art Notes section is for our Maya and 3D Studio MAX material plug-ins. This is what the artists see when they select a shader within the material editor. StartArtNotes Base = 24-bit base map T1 = Need 8-bit fur density map T2 = Anisotropic lighting lookup map EndArtNotes 14

15 Streams & Stream Mappings This describes how to fully preprocess our vertex and index buffers, and the data associated with each stream element. There can be a unique index buffer associated with each Stream Map. StartStream sposnorm Normal float3 POSITION Position float3 NORMAL VertexNormal EndStream StartStream stexcoords Normal float2 TEX0 UV0 //BaseTexU, BaseTexV float3 TEX1 Tangent0 EndStream StartStream sfurfins FurFins float3 POSITION Position float3 NORMAL VertexNormal float4 TEX0 FinFaceData0 //FinTexU, FinTexV, BaseTexUVDist, RandOffset float4 TEX1 FinFaceData1 //BaseTexU, BaseTexV, BaseTexDu, BaseTexDv float3 TEX2 FaceNormal EndStream StreamMap smbasepass(sposnorm, stexcoords) StreamMap smshellpass(sposnorm, stexcoords) StreamMap smfinspass(sfurfins) 15

16 Texture Declaration Describes how to fully preprocess our textures including input formats, mip map generation, and output format. Our language allows for each output texture to be created from up to two input textures that the artists specify. Texture tbase 2D DXT1("Base", RGB, KaiserGamma) Texture tfurshelltexture 2D RGBA( T1", RGB, Kaiser, T2", GRAY, KaiserGamma) Texture tbump 2D DXT5("Bump", HEIGHT, Box, "Opacity", GRAY, Box) Texture tanisolookup 2D RGBA( T5", RGB, Box, T6", GRAY, Box) Texture tnormcube CM RGBX("Base3", RGB, Box) Texture tenvmap CMAuto //This is an auto-generated cube map Texture twater Renderable("WaterReflection") 16

17 Variables Can be bound to render state or shader constants. If a variable is editable, the artist can override the variable from within the art tool. Variables can also be tied to internal state such as camera position or transform matrices. Float furheight(4.0) EDITABLE Vector furfadescalebias(0.5, 0.5, 0.0, 0.0) Vector objectspacecampos(camera_position, OBJECT_SPACE) Vector objectspacelightpos(light_position, OBJECT_SPACE, 0) Matrix wvp(matrix_wvp) 17

18 Miscellaneous This section is reserved for application specific string pairs. All of these are application specific, and the shader library knows nothing about them. StartMisc Transparent TRUE Animation Object EndMisc Skinned StencilShadows 18

19 Sub-Shaders A shader file can contain several sub-shaders which describe how to completely render the geometry in an arbitrary number of passes. A shader file can contain one of these for each supported graphics card and acts as the fallback mechanism. Each sub-shader has a type in the form of a string. At load time, the first sub-shader that validates (found from top to bottom) of a given type is kept in memory and the rest are skipped. 19

20 Sub-Shaders (Cont d) // StartShader //For RADEON 8500 or ps.1.4 minimum Type Normal //Can be anything, like Wire frame, Invincible StartPass EndPass EndShader // StartShader //For RADEON or fixed function transform Type Normal StartPass EndPass StartPass EndPass EndShader 20

21 Pass Each pass has a unique Stream Mapping, unique set of render state and textures, and vertex and pixel shader code. StartPass //Set Stream Map //Set Textures //Set Render State //Set Vertex Shader Constants //Vertex Shader Code //Set Pixel Shader Constants //Pixel Shader Code EndPass Within a sub-shader, state is sticky between passes. 21

22 Pass - Set Stream Map A single stream map is bound per pass. This is useful if your multi-pass algorithm requires different vertex buffers per pass (fur shells vs. fur fins). SetStreamMap smfinspass 22

23 Pass - Set Textures Binds current texture, associated filter state, LOD bias, etc. in a cleaner language than current API s. SetTexture 0 tfurfintexture Bilinear LodBias(-0.5) Aniso(4) SetTexture 1 tfurfinoffsettexture Bilinear SetTexture 2 tbase Trilinear SetTexture 3 tanisolookup Linear 23

24 Pass - Set Render State Describes all render state in a easier to read language ZTest FALSE ZWrite FALSE Cull CCW Stencil TRUE ZFail(DecSat) Func(Always) Blend TRUE ONE SRCALPHA AlphaTest TRUE 0.9 NPatch Order(Cubic, Quadratic) Make your shader files more readable than bare OpenGL or D3D code! 24

25 Pass - Vertex & Pixel Shader Constants Constants can be set to either variables which were previously defined or constant values. VsConst 0 const(0.0, 0.25, 0.5, 1.0) VsConst 1 commonconst VsConst 2 objectspacecampos VsConst 3 objectspacelightpos VsConst 4 wvpmatrix VsConst 8 furfadescalebias VsConst 9 furheight VsConst 10 furfintextiling PsConst 1 invfintextiling //1.0 / fintextilingfactor PsConst 2 lightcolor //Color of light 25

26 Pass - Vertex & Pixel Shaders Vertex and pixel shader code can either be embedded in the shader file or reference an external file. StartPixelShader ps.1.4 texld r0, t0 texld r1, t0 texcrd r2, t2 texcrd r3, t3 texcrd r4, t4 texcrd r5, t1 dp3_d2 r3.r, r2, r0_bx2 dp3_d2 r3.gba, r3, r0_bx2 mul r1, r1_bias, r4 mad r5.rg, r1, c1, r5 mov r5.b, r0.a //direction of aniso and opacity //offset texture //tan space L //tan space V //du,dv that fin spans in base map coords //base map coords //T.L //T.V //hair offset value * du,dv //base map lookup for hair color perturbed by offset map //pass base map opacity through phase texld r2, r5 //base map lookup texld r3, r3 //aniso lighting texcrd r5, t5 // mul_x2 r3, r3, c2 //scale diffuse and specular components of light mad r0.rgb, r2, r3, r3.a //diffuse * base + specular +mul r0.a, r5.b, r2.a //hair opacity * base map opacity mul_sat r0.rgb, r0, v0 //scale hair color by hair height +mul_sat r0.a, r0.a, v0.a //opacity * edge on fading EndPixelShader 26

27 Default Shader Defines all default state for your engine within a shader file. You don t need to rely on D3D s or OpenGL s default state since you override it with what is most useful to your app. This also helps reduce overall state change. Your shaders ultimately have less redundancy since you can rely on the defaults you set. 27

28 Cleanup Shader Since you set a default shader at startup, cleaning up state becomes trivial and fully automated! When the engine is done rendering with a given shader, the shader library automatically sets any changed render state back to the default. Cleanup can be very smart by delaying cleanup until the next shader is activated and change only the absolute minimal state required. Reduces human error of forgetting to set a given render state back to what it was. 28

29 More on Fallbacks Some older hardware lacks multi-stream vertex buffers. You often have to store redundant data on disk to do the most optimal thing at runtime. If you ve generalized your preprocessing based on earlier slides, this should mean no additional work on your part. At runtime, load all shaders first, then load only the necessary vertex buffers, index buffers, and textures. More interesting shaders also require additional textures that are often not needed for the fallback sub-shaders. 29

30 Vertex Buffer Creation When organizing vertex buffers during preprocessing, keep buffers common to the same shader together so it s easier to discard them. You may want to classify each stream and group them to make the loadtime rejection easier: StartStream sposnorm Normal For

31 Example Run-Time Code AtiShader *sh = GetShaderPtr(); for (int i=0; i<sh->numpasses; i++) { //Figure out which streams are needed for this pass } //For each draw call // Set current textures through shader lib // Set current vertex and index buffers (streams) // Draw 31

32 State Change Priority In the past, we prioritized texture changes as the most important state change. Now prioritize based on the following order: Vertex and Pixel Shader Code Vertex and Pixel Shader Constants Textures Vertex and Index Buffers 32

33 Summary Shader library should be central to every part of your rendering pipeline from artist creation, to preprocessing, to run-time. Provides an approachable method for true fallback shaders. Facilitates cross-platform development. Separates shader writing from engine development! Better allocation of internal resources. 33

34 More Information Slides will be available on the ATI Developer Relations Website: ati.com/developer 34

35 Questions? This outtake was captured from the Rachel demo when we first rendered this model in realtime. Her face polygons remained in object space accidentally We corrected the problem by simply editing the text shader file without requiring a recompile or restart of the app. 35