Siggraph 2011: Occlusion culling in Alan Wake
Download as PPTX, PDF7 likes5,484 views
The document discusses occlusion culling techniques used in the game Alan Wake, focusing on rendering challenges in large outdoor worlds and the management of dynamic objects. It highlights the use of the Umbra visibility middleware to optimize rendering through efficient data handling, hierarchical hardware occlusion culling, and shadow mapping strategies. Key strategies include adaptive query bounds, shadow caster culling, and maintaining a balanced hierarchy for dynamic objects to improve performance and reduce memory usage.
Siggraph 2011: Occlusion culling in Alan Wake
- 2. Occlusion Culling in Alan WakeAri Silvennoinen
- 3. Rendering Challenges in Alan Wake
- 4. IntroductionRendering challenges in Alan Wake
- 10. Techniques and ideas used to tackle the rendering challenges
- 11. Umbra is the visibility middleware used in Alan Wake
- 12. Supports both hardware and software occlusion Large Outdoor WorldsOcclusion Culling and LOD
- 14. Only a subset of the world fits in the memoryRequirements for VisibilityScalable culling (view frustum, distance and occlusion)
- 15. Out-of-core objects should be cheap
- 17. Incremental updates for dynamic objectsUmbra Architecture ClientUmbraInsert/UpdateSpatial databaseQuery beginQuery traversalCallbacksQuery end
- 18. Spatial DatabaseAxis-aligned BSP-tree with varying SAH split planes
- 19. Premultiplied OBB representation for objects
- 20. 4x3 matrix for mapping the unit cube to world space OBB
- 21. Lazy update mechanism based on camera frustum
- 22. Mark nodes as dirty, update when actually traversed
- 24. Split plane optimization to improve spatial subdivision
- 25. Dead node cleanup to save memory on out-of-core locationsOcclusion CullingHierarchical hardware occlusion culling
- 26. Avoiding latency by interleaving work
- 27. Umbra does hierarchy traversal and database update tasks to prevent CPU stalling
- 28. Works well in practice, however 100% utilization impossible due to GPU starvation
- 29. Avoiding latency by delayed reads
- 30. Artifacts!
- 31. Can not use full hierarchy -> not scalableOcclusion CullingHybrid solution brings the best of both worlds
- 32. Screen space error bounds for artifacts
- 34. Some queries can be guaranteed to be ready with high probability
- 35. Keep the query count low
- 36. Queries are cheap but not free
- 38. Special rules for expensive objectsOcclusion Culling
- 39. Dynamic ObjectsDynamic objects live in the same database
- 41. Linear prediction with conservative dynamics model
- 42. Stable and fast
- 43. Temporal bounds are AABBs
- 44. Has even been used with bullets driven by physics
- 45. Take advantage of lazy and amortized updates
- 46. Keeps hierarchy balanced during movement
- 47. Ensures optimal node placement when object becomes staticDynamic Objects
- 48. Local ShadowsAccurate bounding geometry
- 51. Distance culling
- 52. Test every frame
- 53. No batching
- 54. Shadow casters use frustum cullingGlobal ShadowsShadow map costs
- 56. Shadow map sampling + filtering
- 57. Lot of objects + cascaded shadows = high shadow map generation cost
- 58. What do we really need to draw to a shadow map?Observation ICull objects in shadow
- 59. Observation IICull objects in which do not cast a visible shadow
- 60. Shadow Caster CullingUse camera visibility to determine relevant shadow casters
- 61. Test each cascade for visibility
- 62. Check if the cascade contains any visible objects
- 63. Can potentially skip shadow map generation
- 64. Compute light space scissor rectangle for each cascade
- 67. Bittner et. Al, Shadow Caster Culling for Efficient Shadow Mapping, I3D 2011Cascaded Shadow Maps
- 70. Cascaded Shadow MapsSmall scissorEmpty cascade
- 71. Conclusions and Future WorkEfficient spatial database in a key role
- 74. Streaming?
- 76. Rethink visibility and spatial connectivity
- 77. See our slides from “Advances in Real-Time Rendering in Games” course ThanksRemedy Entertainment
- 78. Petri Häkkinen
- 79. Markus Mäki
- 80. Olli Tervo
- 81. TuukkaTaipalvesi
- 82. Colleagues at Umbra Software
- 83. TimoAila