Deferred Rendering in Killzone 2
25 likes20,047 views
This document outlines a presentation by Michal Valient from Guerrilla at the Develop Conference in July 2007, focusing on deferred rendering techniques used in Killzone 2. It covers various aspects including g-buffer layout, rendering passes, light accumulation, and optimization strategies, emphasizing the advantages and challenges of deferred rendering. The presentation concludes that deferred rendering allows for artistic freedom and efficient handling of complex geometry and dynamic lighting.
![G-Buffer : Our approach
R8 G8 B8 A8
Depth 24bpp Stencil DS
Lighting Accumulation RGB Intensity RT0
Normal X (FP16) Normal Y (FP16) RT1
Motion Vectors XY Spec-Power Spec-Intensity RT2
Diffuse Albedo RGB Sun-Occlusion RT3
‣ Lighting accumulation – output buffer
‣ Intensity – luminance of Lighting accumulation
‣ Scaled to range [0…2]
‣ Normal.z = sqrt(1.0f - Normal.x2 - Normal.y2)
GUERRILLA | DEVELOP CONFERENCE | JULY ‘07 | BRIGHTON](https://image.slidesharecdn.com/develop07valientdeferredrenderinginkillzone2-111014045249-phpapp01/85/Deferred-Rendering-in-Killzone-2-19-320.jpg)
![Geometry Pass
‣ Fill the G-Buffer with all geometry (static, skinned, etc.)
‣ Write depth, motion, specular, etc. properties
‣ Initialize light accumulation buffer with pre-baked light
‣ Ambient, Incandescence, Constant specular
‣ Lightmaps on static geometry
‣ YUV color space, S3TC5 with Y in Alpha
‣ Sun occlusion in B channel
‣ Dynamic range [0..2]
‣ Image based lighting on dynamic geometry
GUERRILLA | DEVELOP CONFERENCE | JULY ‘07 | BRIGHTON](https://image.slidesharecdn.com/develop07valientdeferredrenderinginkillzone2-111014045249-phpapp01/85/Deferred-Rendering-in-Killzone-2-23-320.jpg)
![Image Based Lighting
‣ Artist placed light probes
‣ Arbitrary location and density
‣ Sampled and stored as 2nd order spherical harmonics
‣ Updated per frame for each object
‣ Blend four closest SHs based on distance
‣ Rotate into view space
‣ Encode into 8x8 envmap IBL texture
‣ Dynamic range [0..2]
‣ Generated on SPUs in parallel to other rendering tasks
GUERRILLA | DEVELOP CONFERENCE | JULY ‘07 | BRIGHTON](https://image.slidesharecdn.com/develop07valientdeferredrenderinginkillzone2-111014045249-phpapp01/85/Deferred-Rendering-in-Killzone-2-24-320.jpg)
Deferred Rendering in Killzone 2
- 1. GUERRILLA | DEVELOP CONFERENCE | JULY ‘07 | BRIGHTON
- 2. Deferred Rendering in Killzone 2 Michal Valient Senior Programmer, Guerrilla GUERRILLA | DEVELOP CONFERENCE | JULY ‘07 | BRIGHTON
- 3. Talk Outline ‣ Forward & Deferred Rendering Overview ‣ G-Buffer Layout ‣ Shader Creation ‣ Deferred Rendering in Detail ‣ Rendering Passes ‣ Light and Shadows ‣ Post-Processing ‣ SPU Usage / Architecture GUERRILLA | DEVELOP CONFERENCE | JULY ‘07 | BRIGHTON
- 4. Forward & Deferred Rendering Overview GUERRILLA | DEVELOP CONFERENCE | JULY ‘07 | BRIGHTON
- 5. Forward Rendering – Single Pass ‣ For each object ‣ Find all lights affecting object ‣ Render all lighting and material in a single shader ‣ Shader combinations explosion ‣ Shader for each material vs. light setup combination ‣ All shadow maps have to be in memory ‣ Wasted shader cycles ‣ Invisible surfaces / overdraw ‣ Triangles outside light influence GUERRILLA | DEVELOP CONFERENCE | JULY ‘07 | BRIGHTON
- 6. Forward Rendering – Multi-Pass ‣ For each light ‣ For each object ‣ Add lighting from single light to frame buffer ‣ Shader for each material and light type ‣ Wasted shader cycles ‣ Invisible surfaces / overdraw ‣ Triangles outside light influence ‣ Lots of repeated work ‣ Full vertex shaders, texture filtering GUERRILLA | DEVELOP CONFERENCE | JULY ‘07 | BRIGHTON
- 7. Deferred Rendering ‣ For each object ‣ Render surface properties into the G-Buffer ‣ For each light and lit pixel ‣ Use G-Buffer to compute lighting ‣ Add result to frame buffer ‣ Simpler shaders ‣ Scales well with number of lit pixels ‣ Does not handle transparent objects GUERRILLA | DEVELOP CONFERENCE | JULY ‘07 | BRIGHTON
- 8. G-Buffer Layout GUERRILLA | DEVELOP CONFERENCE | JULY ‘07 | BRIGHTON
- 9. Target Image
- 10. Depth
- 13. Specular roughness / Power
- 14. Screen-space 2D motion vector
- 17. Image with post-processing (depth of field, bloom, motion blur, colorize, ILR)
- 18. G-Buffer : Our approach R8 G8 B8 A8 Depth 24bpp Stencil DS Lighting Accumulation RGB Intensity RT0 Normal X (FP16) Normal Y (FP16) RT1 Motion Vectors XY Spec-Power Spec-Intensity RT2 Diffuse Albedo RGB Sun-Occlusion RT3 ‣ MRT - 4xRGBA8 + 24D8S (approx 36 MB) ‣ 720p with Quincunx MSAA ‣ Position computed from depth buffer and pixel coordinates GUERRILLA | DEVELOP CONFERENCE | JULY ‘07 | BRIGHTON
- 19. G-Buffer : Our approach R8 G8 B8 A8 Depth 24bpp Stencil DS Lighting Accumulation RGB Intensity RT0 Normal X (FP16) Normal Y (FP16) RT1 Motion Vectors XY Spec-Power Spec-Intensity RT2 Diffuse Albedo RGB Sun-Occlusion RT3 ‣ Lighting accumulation – output buffer ‣ Intensity – luminance of Lighting accumulation ‣ Scaled to range [0…2] ‣ Normal.z = sqrt(1.0f - Normal.x2 - Normal.y2) GUERRILLA | DEVELOP CONFERENCE | JULY ‘07 | BRIGHTON
- 20. G-Buffer : Our approach R8 G8 B8 A8 Depth 24bpp Stencil DS Lighting Accumulation RGB Intensity RT0 Normal X (FP16) Normal Y (FP16) RT1 Motion Vectors XY Spec-Power Spec-Intensity RT2 Diffuse Albedo RGB Sun-Occlusion RT3 ‣ Motion vectors – screen space ‣ Specular power - stored as log2(original)/10.5 ‣ High range and still high precision for low shininess ‣ Sun Occlusion - pre-rendered static sun shadows ‣ Mixed with real-time sun shadow for higher quality GUERRILLA | DEVELOP CONFERENCE | JULY ‘07 | BRIGHTON
- 21. G-Buffer Analysis ‣ Pros: ‣ Highly packed data structure ‣ Many extra attributes ‣ Allows MSAA with hardware support ‣ Cons: ‣ Limited output precision and dynamic range ‣ Lighting accumulation in gamma space ‣ Can use different color space (LogLuv) ‣ Attribute packing and unpacking overhead GUERRILLA | DEVELOP CONFERENCE | JULY ‘07 | BRIGHTON
- 22. Deferred Rendering Passes GUERRILLA | DEVELOP CONFERENCE | JULY ‘07 | BRIGHTON
- 23. Geometry Pass ‣ Fill the G-Buffer with all geometry (static, skinned, etc.) ‣ Write depth, motion, specular, etc. properties ‣ Initialize light accumulation buffer with pre-baked light ‣ Ambient, Incandescence, Constant specular ‣ Lightmaps on static geometry ‣ YUV color space, S3TC5 with Y in Alpha ‣ Sun occlusion in B channel ‣ Dynamic range [0..2] ‣ Image based lighting on dynamic geometry GUERRILLA | DEVELOP CONFERENCE | JULY ‘07 | BRIGHTON
- 24. Image Based Lighting ‣ Artist placed light probes ‣ Arbitrary location and density ‣ Sampled and stored as 2nd order spherical harmonics ‣ Updated per frame for each object ‣ Blend four closest SHs based on distance ‣ Rotate into view space ‣ Encode into 8x8 envmap IBL texture ‣ Dynamic range [0..2] ‣ Generated on SPUs in parallel to other rendering tasks GUERRILLA | DEVELOP CONFERENCE | JULY ‘07 | BRIGHTON
- 25. Scene lighting
- 26. Decals and Weapon Passes ‣ Primitives updating subset of the G-Buffer ‣ Bullet holes, posters, cracks, stains ‣ Reuse lighting of underlying surface ‣ Blend with albedo buffer ‣ Use G-Buffer Intensity channel to fix accumulation ‣ Same principle as particles with motion blur ‣ Separate weapon pass with different projection ‣ Different near plane ‣ Rendered into first 5% of depth buffer range ‣ Still reacts to lights and post-processing GUERRILLA | DEVELOP CONFERENCE | JULY ‘07 | BRIGHTON
- 29. Light Accumulation Pass ‣ Light is rendered as convex geometry ‣ Point light – sphere ‣ Spot light – cone ‣ Sun – full-screen quad ‣ For each light… ‣ Find and mark visible lit pixels ‣ If light contributes to screen ‣ Render shadow map ‣ Shade lit pixels and add to framebuffer GUERRILLA | DEVELOP CONFERENCE | JULY ‘07 | BRIGHTON
- 30. Determine Lit Pixels ‣ Marks pixels in front of the far light boundary ‣ Render back-faces of light volume ‣ Depth test GREATER-EQUAL ‣ Write to stencil on depth pass ‣ Skipped for very small distant lights GUERRILLA | DEVELOP CONFERENCE | JULY ‘07 | BRIGHTON
- 31. Determine Lit Pixels ‣ Find amount of lit pixels inside the volume ‣ Start pixel query ‣ Render front faces of light volume ‣ Depth test LESS-EQUAL ‣ Don’t write anything – only EQUAL stencil test GUERRILLA | DEVELOP CONFERENCE | JULY ‘07 | BRIGHTON
- 32. Render Shadow Map ‣ Enable conditional rendering ‣ Based on query results from previous stage ‣ GPU skips rendering for invisible lights ‣ Max 1024x1024xD16 shadow map ‣ Fast and with hardware filtering support ‣ Single map reused for all lights ‣ Skip small objects ‣ Small in shadow map and on screen ‣ Artist defined thresholds for lights and objects GUERRILLA | DEVELOP CONFERENCE | JULY ‘07 | BRIGHTON
- 33. Shade Lit Pixels ‣ Render front-faces of light volume ‣ Depth test - LESS-EQUAL ‣ Stencil test - EQUAL ‣ Runs only on marked pixels inside light ‣ Compute light equation ‣ Read and unpack G-Buffer attributes ‣ Calculate Light vector, Color, Distance Attenuation ‣ Perform shadow map filtering ‣ Add Phong lighting to frame buffer GUERRILLA | DEVELOP CONFERENCE | JULY ‘07 | BRIGHTON
- 34. Light Optimization ‣ Determine light size on the screen ‣ Approximation - angular size of light volume ‣ If light is “very small” ‣ Don’t do any stencil marking ‣ Switch to non-shadow casting type ‣ Shadows fade-out range ‣ Artist defined light sizes at which: ‣ Shadows start to fade out ‣ Switch to non-shadow casting light GUERRILLA | DEVELOP CONFERENCE | JULY ‘07 | BRIGHTON
- 35. Sun Rendering ‣ Full screen quad ‣ Stencil mark potentially lit pixels ‣ Use only sun occlusion from G-Buffer ‣ Run final shader on marked pixels ‣ Approx. 50% of pixels skipped thanks 1st pass ‣ Also skybox/background ‣ Simple directional light model ‣ Shadow = min(RealTimeShadow, Occlusion) GUERRILLA | DEVELOP CONFERENCE | JULY ‘07 | BRIGHTON
- 36. Sun – Real-Time Shadows ‣ Cascade shadow maps ‣ Provide more shadow detail where required ‣ Divide view frustum into several areas ‣ Split along view distance ‣ Split distances defined by artist ‣ Render shadow map for each area ‣ Max 4 cascades ‣ Max 512x512 pixels each in single texture ‣ Easy to address cascade in final render GUERRILLA | DEVELOP CONFERENCE | JULY ‘07 | BRIGHTON
- 37. Sun – Real-Time Shadows ‣ Issue: Shadow shimmering ‣ Light cascade frustums follow camera ‣ Sub pixel changes in shadow map ‣ Solution! ‣ Don’t rotate shadow map cascade ‣ Make bounding sphere of cascade frustum ‣ Use it to generate cascade light matrix ‣ Remove sub-pixel movements ‣ Project world origin onto shadow map ‣ Use it to round light matrix to nearest shadow pixel corner GUERRILLA | DEVELOP CONFERENCE | JULY ‘07 | BRIGHTON
- 38. Sun - Colored shadow Cascades - Unstable shadow artifacts
- 39. MSAA Lighting Details ‣ Run light shader at pixel resolution ‣ Read G-Buffer for both pixel samples ‣ Compute lighting for both samples ‣ Average results and add to frame buffer ‣ Optimization in shadow map filtering ‣ Max 12 shadow taps per pixel ‣ Alternate taps between both samples ‣ Half quality on edges, full quality elsewhere ‣ Performance equal to non-MSAA case GUERRILLA | DEVELOP CONFERENCE | JULY ‘07 | BRIGHTON
- 40. Forward Rendering Pass ‣ Used for transparent geometry ‣ Single pass solution ‣ Shader has four uberlights ‣ No shadows ‣ Per-vertex lighting version for particles ‣ Lower resolution rendering available ‣ Fill-rate intensive effects ‣ Half and quarter screen size rendering ‣ Half resolution rendering using MSAA HW GUERRILLA | DEVELOP CONFERENCE | JULY ‘07 | BRIGHTON
- 41. Post-Processing Pass ‣ Highly customizable color correction ‣ Separate curves for shadows, mid-tones, highlight colors, contrast and brightness ‣ Everything Depth dependent ‣ Per-frame LUT textures generated on SPU ‣ Image based motion blur and depth of field ‣ Internal lens reflection ‣ Film grain filter GUERRILLA | DEVELOP CONFERENCE | JULY ‘07 | BRIGHTON
- 42. SPU Usage and Architecture Putting it all together GUERRILLA | DEVELOP CONFERENCE | JULY ‘07 | BRIGHTON
- 43. SPU Usage ‣ We use SPU a lot during rendering ‣ Display list generation ‣ Main display list ‣ Lights and Shadow Maps ‣ Forward rendering ‣ Scene graph traversal / visibility culling ‣ Skinning ‣ Triangle trimming ‣ IBL generation ‣ Particles GUERRILLA | DEVELOP CONFERENCE | JULY ‘07 | BRIGHTON
- 44. SPU Usage (cont.) ‣ Everything is data driven ‣ No “virtual void Draw()” calls on objects ‣ Objects store a decision-tree with DrawParts ‣ DrawParts link shader, geometry and flags ‣ Decision tree used for LODs, etc. ‣ SPUs pull rendering data directly from objects ‣ Traverse scenegraph to find objects ‣ Process object's decision-tree to find DrawParts ‣ Create displaylist from DrawParts GUERRILLA | DEVELOP CONFERENCE | JULY ‘07 | BRIGHTON
- 45. SPU Architecture PPU SPU 0 SPU 1 SPU 2 SPU 3 Particles, Skinning Main scenegraph + displaylist IBL generation edgeGeom Shadow scenegraph + displaylist GUERRILLA | DEVELOP CONFERENCE | JULY ‘07 | BRIGHTON
- 46. SPU Architecture GAME, AI PHYSICS PPU SPU 0 SPU 1 SPU 2 SPU 3 Particles, Skinning Main scenegraph + displaylist IBL generation edgeGeom Shadow scenegraph + displaylist GUERRILLA | DEVELOP CONFERENCE | JULY ‘07 | BRIGHTON
- 47. SPU Architecture GAME, AI PREPARE PHYSICS DRAW PPU SPU 0 SPU 1 SPU 2 SPU 3 Particles, Skinning Main scenegraph + displaylist IBL generation edgeGeom Shadow scenegraph + displaylist GUERRILLA | DEVELOP CONFERENCE | JULY ‘07 | BRIGHTON
- 48. SPU Architecture GAME, AI PREPARE PHYSICS DRAW PPU SPU 0 SPU 1 SPU 2 SPU 3 Particles, Skinning Main scenegraph + displaylist IBL generation edgeGeom Shadow scenegraph + displaylist GUERRILLA | DEVELOP CONFERENCE | JULY ‘07 | BRIGHTON
- 49. SPU Architecture GAME, AI PREPARE PHYSICS DRAW PPU SPU 0 SPU 1 SPU 2 SPU 3 Particles, Skinning Main scenegraph + displaylist IBL generation edgeGeom Shadow scenegraph + displaylist GUERRILLA | DEVELOP CONFERENCE | JULY ‘07 | BRIGHTON
- 50. SPU Architecture GAME, AI PREPARE PHYSICS DRAW PPU SPU 0 SPU 1 SPU 2 SPU 3 Particles, Skinning Main scenegraph + displaylist IBL generation edgeGeom Shadow scenegraph + displaylist GUERRILLA | DEVELOP CONFERENCE | JULY ‘07 | BRIGHTON
- 51. SPU Architecture GAME, AI PREPARE PHYSICS DRAW PPU SPU 0 SPU 1 SPU 2 SPU 3 Particles, Skinning Main scenegraph + displaylist IBL generation edgeGeom Shadow scenegraph + displaylist GUERRILLA | DEVELOP CONFERENCE | JULY ‘07 | BRIGHTON
- 52. SPU Architecture DRAW GAME, AI PREPARE GAME, AI PREPARE PHYSICS DRAW DATA PHYSICS DRAW PPU LOCK SPU 0 SPU 1 SPU 2 SPU 3 Particles, Skinning Main scenegraph + displaylist IBL generation edgeGeom Shadow scenegraph + displaylist GUERRILLA | DEVELOP CONFERENCE | JULY ‘07 | BRIGHTON
- 53. Conclusion ‣ Deferred rendering works well and gives us artistic freedom to create distinctive Killzone look ‣ MSAA did not prove to be an issue ‣ Complex geometry with no resubmit ‣ Highly dynamic lighting in environments ‣ Extensive post-process ‣ Still a lot of features planned ‣ Ambient occlusion / contact shadows ‣ Shadows on transparent geometry ‣ More efficient anti-aliasing ‣ Dynamic radiosity GUERRILLA | DEVELOP CONFERENCE | JULY ‘07 | BRIGHTON