GP1 Software Rasterization Pipeline PDF

GP1 – Software Rasterization Pipeline (High-Level)...

GP1 – Software Rasterization Pipeline (High-Level) INPUT (Interpolated Vertex Attributes for Pixel) INPUT (VERTEX_IN) INPUT Position : Vector4 Vertices (in World Space) Vertices (in ‘NDC’ space) (X & Y > ScreenSpace || Z > Interpolated Normalized Depth || W > Interpolated Camera-Space Depth) Each Vertex has at least a POSITION attribute, but can be decorated with extra attributes such as COLOR, UV,... Each Vertex has at least a POSITION attribute, but can be decorated with extra attributes such as COLOR, UV,... Interpolated Color : ColorRGB Interpolated UV : Vector2 Interpolated Normal : Vector3 Interpolated Tangent : Vector3 PROJECTION STAGE RASTERIZATION STAGE Interpolated ViewDirection : Vector3 [All vertices] [Each Triangle > Iterate Pixels] PIXEL SHADING STAGE [Positions] Frustum Culling (+ Clipping) [Calculate Color for given Pixel] Model(space) > NDC(space) (NDC) Vertex inside Camera Frustum? X & Y between [-1,1] Z between [0.1] CONSTANTS (Directional Light) Every Pixel that passed the Depth Test Multiply vertex positions with WVP matrix Ignore Triangle if not in Frustum World > Model Transformation LightDirection > {.577f, -.577f,.577f } View > Camera Transformation LightIntensity > 7.f Every Vertex Projection > Projection based on AspectRatio & FOV NDC (space) > Raster/Screen (space) Optimization > Calculate Triangle Every Triangle Specular Shininess > 25.f + Apply Perspective Divide BoundingBox (BB) in Rasterspace > See slides for formula. After this transformation your Position X & Y should be in ScreenSpace (2D raster coordinates). Only iterate the pixels covered by this BB. NORMAL MAPPING (Normal Texture) [Normals & Tangents] Calculate TangentSpaceAxis, multiply with sampled normal from Model(space) > World(space) NormalMap > Use this resulting Normal as new Normal PIXEL inside TRIANGLE? Multiply Normals & Tangents with (3x3) World matrix Cross product between each Triangle edge and Vertex-2-Pixel vectors (2D Barycentric Coordinates (We only want to rotate them, so cast (4x4) World matrix to a (3x3) vectors) > IF signs of all cross-product results are the same THEN pixel is DIFFUSE (Diffuse Texture) matrix [FMatrix3] > Discards the translational component) inside the Triangle. Cross Product Results can be Basic Lambert Diffuse used to calculate Every Pixel interpolation weights for [ViewDirection] Depth Interpolation + Depth Test SPECULAR (Specular + Gloss Texture) Depth & Attribute Phong Specular (similar to RayTracer) Correct ‘Depth Interpolation’ > LAB 3 Interpolation. Normalize(WorldPosition – CameraPosition) WorldPosition > ModelSpace position multiplied with (4x4) World matrix Vertex Attribute Interpolation + See LAB 3 for ‘correct’ Depth & Vertex Interpolation AMBIENT (optional) Solid Color > RGB{.025f,.025f,.025f } Correct ‘Attribute Interpolation’ > LAB 3 OUTPUT (VERTEX_OUT) Vertices (in ‘NDC’ space) OUTPUT (Only if DepthTest succeeds) OUTPUT > Final Pixel Color (Diffuse + Specular + Ambient) Positions are Projected, Normals & Tangents are transformed, ViewDirection is calculated, other attributes are (Normalize Color before writing to backbuffer) passed through Position X & Y > NDC (Range [-1,1] ) Single VERTEX_OUT structure containing interpolated attributes Position Z > NDC (Range [0,1] ] (Position, Color, UV, Normal, Tangent, ViewDirection) Position W > ViewSpace (Distance from camera) CULL MODE Depends on the cross-product sign * This is only a High-Level overview/guide! > Use the slides for deeper explanation & All Negative vs All Positive == FrontFace proper implementation guidelines. vs BackFace culling