Best Practices for Shader Model 3.0.pdfVIP

Shader Model 3.0, Best Practices Phil Scott Technical Developer Relations, EMEA Overview Short Pipeline Overview CPU Bound – new optimization opportunities Obscure bits of the pipeline that can trip you up Pixel Bound – new optimization opportunities 3.0 shader performance characteristics Pipelined Architecture (simplified view) Frame buffer Fragment Processor Texture Storage + Filtering RasterizerGeometry Processor Geometry StorageCPU Vertices Pixels Bottlenecks Frame buffer Fragment Processor Texture Storage + Filtering RasterizerGeometry Processor Geometry StorageCPU Limits the speed of the pipeline CPU / Fragment – focus of this talk CPU / Fragment Bound Still the two most likely cases these days in modern apps: CPU bound Becomes more and more likely the faster GPUs get Fragment bound Becomes more and more likely the longer shaders get Neither of these trends are likely to change soon Some new weapons for combating these Instancing HW Shadow Maps Shader model 3.0 DirectX 9 Instancing API What is it? Allows you to avoid DIP calls and minimise batching overhead Allows a single draw call to draw multiple instances of the same model What is required to use it? Microsoft DirectX 9.0c VS/PS 3.0 hardware Why use instancing? Speed. Still the single most common performance suck in most games today is draw calls Yeah. Yeah. We all know draw calls are bad But world matrices and other state often force us to separate draw calls The instancing API pushes the per instance draw logic down into the driver Saves DIP call overhead in both D3D and Driver Allows the driver to ensure minimal state changes between instances When to use instancing? Scene contains many instances of the same model Forest of Trees, Particles, Sprites If you can encode per instance data in 2nd streams. I.e instance transforms, model color, indices to textures/constants. Less useful if your batch size is large 1k polygons per draw There is some fixed overhead to using instancing How does