[EPIC]: Improve rasterization tests coverage

[EmilioLaiso]

Legend

🏗️ - Requires backend/yaml work

👍 - Testable

Vertex Shader

Semantics

• SV_VertexID
  • Add GraphicsSystemValues.test for graphics-stage system-value semantics #1142
• SV_Position
  • Already covered by all graphics tests
• SV_ClipDistance[n] - 👍
• SV_CullDistance[n] - 👍
• SV_InstanceID - 🏗️
• SV_StartVertexLocation - 🏗️
• SV_StartInstanceLocation - 🏗️
• SV_RenderTargetArrayIndex - 🏗️
• SV_ViewportArrayIndex - 🏗️
• SV_ViewID - 🏗️

Functional

• Non-Float32 vertex attribute formats (Int32, UInt32, Float16, Int16, UInt16) - 👍
• Normalized integer attribute formats (UNorm8, SNorm8, UNorm16, SNorm16) - 👍
• Multiple varyings to PS — many simultaneous user semantics, mixed integer/float types - 👍
• Struct-typed varyings between VS and PS - 👍
• CBuffer read in VS (real mul(MVP, position) instead of pass-through) - 👍
• Push constants in VS - 👍
• StructuredBuffer<T> read from VS - 👍
• ByteAddressBuffer read from VS - 👍
• RWStructuredBuffer write from VS (different UAV path than PS-side write) - 👍
• Texture2D.SampleLevel(...) from VS (no derivatives in VS) - 👍
• Texture2D.Load(...) from VS - 👍
• SV_Position with W ≠ 1 (verify perspective divide downstream) - 👍
• Multiple vertex buffer slots (stream 0 + stream 1) - 🏗️
• Per-instance attributes (input rate INSTANCE) - 🏗️
• Indexed draws + BaseVertex - 🏗️
• Strip topologies + primitive restart - 🏗️

Niche / vendor behavior

• Provoking-vertex consistency (FIRST is the default on D3D12 / VK / MTL) with a nointerpolation integer varying - 👍
• Vertex at clip-plane boundary (x = ±w, y = ±w, z = 0 / w) - 👍
• Vertex with W close to 0 (numerical stability) - 👍
• NaN / Inf in SV_Position (must be clipped, no rasterized output) - 👍
• Subnormal positions (FTZ vs preserve at the rasterizer) - 👍
• Negative-area / degenerate triangles - 👍

Pixel Shader

Semantics

• SV_Position (input)
  • Implicitly used by every graphics test; explicitly asserted at pixel center in GraphicsSystemValues.test
• SV_PrimitiveID
  • Add GraphicsSystemValues.test for graphics-stage system-value semantics #1142
• SV_IsFrontFace
  • Add GraphicsSystemValues.test for graphics-stage system-value semantics #1142
• SV_ClipDistance[n] (input) - 👍
• SV_CullDistance[n] (input) - 👍
• SV_Depth (output) - 👍
• SV_DepthGreaterEqual (output) - 👍
• SV_DepthLessEqual (output) - 👍
• SV_Target1..SV_Target7 (MRT) - 🏗️
• SV_StencilRef (output) - 🏗️
• SV_SampleIndex (input) - 🏗️
• SV_Coverage (input) - 🏗️
• SV_Coverage (output) - 🏗️
• SV_InnerCoverage (input) - 🏗️
• SV_Barycentrics (input) - 🏗️
• SV_RenderTargetArrayIndex (input from VS/GS) - 🏗️
• SV_ViewportArrayIndex (input from VS/GS) - 🏗️

Functional

• discard
  • test/Graphics/discard.test (Add pixel shader discard test #1119)
• ddx_fine, ddy_fine, fwidth
  • test/Graphics/..
• clip() intrinsic (companion to discard) - 👍
• ddx / ddy (default precision) - 👍
• ddx_coarse / ddy_coarse - 👍
• fwidth_coarse / fwidth_fine - 👍
• Interpolation modifier: nointerpolation (flat) - Add InterpolationModifiers.test for HLSL pixel-shader interpolation modifiers #1151
• Interpolation modifier: noperspective - Add InterpolationModifiers.test for HLSL pixel-shader interpolation modifiers #1151
• Interpolation modifier: centroid - Add InterpolationModifiers.test for HLSL pixel-shader interpolation modifiers #1151
• Interpolation modifier: linear - Add InterpolationModifiers.test for HLSL pixel-shader interpolation modifiers #1151
• EvaluateAttributeAtCentroid - 👍
• EvaluateAttributeSnapped - 👍
• EvaluateAttributeAtSample - 🏗️
• Interpolation modifier: sample - 🏗️
• PS UAV write to RWBuffer<T> - 👍
• PS UAV write to RWByteAddressBuffer - 👍
• PS UAV write to RWTexture2D - 👍
• PS reads Texture2D with all sample overloads (Sample, SampleBias, SampleGrad, SampleLevel, SampleCmp, Gather) - 👍

Niche / vendor behavior

• Helper-invocation suppression — discarded lane's writes blocked, but its derivatives still drive the quad's other lanes - 👍
• Quad message-passing (QuadReadAcrossX/Y/Diagonal) under partial coverage - 👍
• SV_Position.xy exactly at pixel center (x + 0.5, y + 0.5) - 👍
• Depth clamping outside [MinDepth, MaxDepth] (clipped vs clamped vs undefined) - 👍
• SV_IsFrontFace for zero-area / degenerate triangles - 👍
• discard + SV_Depth write interaction (depth must not be written) - 👍
• clip() with NaN / Inf operand - 👍
• Pixel coverage at exact triangle edge (top-left rule) - 👍
• Sample-frequency shading (PS runs per-sample, not per-pixel) - 🏗️

Hull Shader

🏗️ Stage-level prerequisites (must land before any test below):

• Extend Stages enum with Hull.
• Patch-list topologies (PatchList<3>, PatchList<4>, ..., PatchList<32>) per backend.
• HS attribute reflection: [domain], [partitioning], [outputtopology], [outputcontrolpoints], [patchconstantfunc], [maxtessfactor].
• Tessellation pipeline state on each backend (D3D12 already supports it; VK needs tessellationShader feature; MTL via metal-irconverter).

Semantics

• SV_TessFactor (3 / 4 / 2 floats by domain) - 🏗️
• SV_InsideTessFactor (1 / 2 / 0 floats by domain) - 🏗️
• SV_OutputControlPointID - 🏗️
• SV_PrimitiveID (HS input) - 🏗️
• HullSystemValues.test — bundled test - 🏗️

Functional

• Pass-through HS (factor 1, identity control points) - 🏗️
• InputPatch<T, N> reading from VS output - 🏗️
• OutputPatch<T, M> to DS - 🏗️
• Patch-constant function reading CBuffer - 🏗️
• Per-edge varying tess factors - 🏗️
• Patch culling via SV_TessFactor = 0 (entire patch discarded) - 🏗️
• Negative tess factor (must be treated as 0) - 🏗️
• SV_TessFactor clamping at [maxtessfactor] - 🏗️
• HS reads UAV / SRV - 🏗️

Niche / vendor behavior

• Tess factor Inf / NaN handling (spec says clamped to max) - 🏗️
• Patch with all coincident control points (degenerate) - 🏗️
• Tess factor > 64 (spec says clamp; observe driver behavior) - 🏗️
• HS invocations seeing partial control-point arrays - 🏗️

Domain Shader

🏗️ Stage-level prerequisites:

• Extend Stages enum with Domain.
• DS pipeline state on each backend (paired with HS).

Semantics

• SV_DomainLocation (float2 for tri/quad, float for isoline) - 🏗️
• SV_PrimitiveID (DS input) - 🏗️
• SV_Position (DS output to rasterizer) - 🏗️
• SV_ClipDistance / SV_CullDistance (DS output) - 🏗️
• DomainSystemValues.test — bundled test - 🏗️

Functional

• Domain: tri (barycentric SV_DomainLocation) - 🏗️
• Domain: quad (UV SV_DomainLocation) - 🏗️
• Domain: isoline - 🏗️
• Partitioning: integer - 🏗️
• Partitioning: fractional_even - 🏗️
• Partitioning: fractional_odd - 🏗️
• Partitioning: pow2 - 🏗️
• Output topology: triangle_cw - 🏗️
• Output topology: triangle_ccw - 🏗️
• Output topology: line - 🏗️
• Output topology: point - 🏗️
• Adaptive tessellation: varying per-edge factors, no T-junctions between adjacent patches - 🏗️
• DS displacement (sample heightmap, write modified SV_Position) - 🏗️
• DS reads CBuffer / SRV - 🏗️

Niche / vendor behavior

• fractional_* partitioning vertex-count is implementation-defined within bounds — assert with tolerance, not exact - 🏗️
• SV_DomainLocation precision near 0 / 1 boundaries - 🏗️
• Isoline outside.x (density) vs outside.y (detail) — easy to invert - 🏗️
• CW / CCW output topology must match the rasterizer's front-face convention (vendors sometimes invert one path) - 🏗️
• Sub-pixel tessellation precision (zoom in on a quad and verify smoothness) - 🏗️

Geometry Shader

🏗️ Stage-level prerequisites:

• Extend Stages enum with Geometry.
• GS pipeline state on each backend.
• [maxvertexcount(N)] and [instance(N)] reflection.
• Variable input topology (point, line, triangle, lineadj, triangleadj).
• Output stream types (PointStream<T>, LineStream<T>, TriangleStream<T>).

Semantics

• SV_PrimitiveID (GS input — auto-generated per input primitive) - 🏗️
• SV_GSInstanceID (when [instance(N)] is used) - 🏗️
• SV_RenderTargetArrayIndex (GS output) - 🏗️
• SV_ViewportArrayIndex (GS output) - 🏗️
• GeometrySystemValues.test — bundled test - 🏗️

Functional

• Pass-through GS (1 in, 1 out, same topology — verifies pipeline plumbing) - 🏗️
• Point-to-quad expansion (point sprite — input point, 4 vertices out as triangle strip) - 🏗️
• Line-to-triangle ribbons (line in, two triangles out) - 🏗️
• Triangle subdivision (1 input → 4 output triangles) - 🏗️
• Topology change: triangle in → line out (wireframe) - 🏗️
• Selective output: if (cond) stream.Append(v); — some primitives produce 0 outputs - 🏗️
• [instance(N)] GS instancing — N invocations per input primitive - 🏗️
• Adjacency input triangleadj (6 vertices: 3 triangle + 3 neighbors), use neighbor data for silhouette detection - 🏗️
• [maxvertexcount(1024)] — spec minimum upper bound; verify large outputs - 🏗️
• GS UAV write (each invocation writes a unique slot) - 🏗️
• GS reads CBuffer / SRV - 🏗️
• RestartStrip() mid-output - 🏗️

Niche / vendor behavior

• Empty stream — [maxvertexcount(N)] declared, 0 emitted, primitive vanishes - 🏗️
• Provoking vertex of GS-emitted primitives (implementation-defined, assert consistency) - 🏗️
• Exceeding [maxvertexcount] (undefined per spec, observe behavior) - 🏗️
• GS performance falls — soak test with high output rate (correctness under throughput pressure) - 🏗️
• GS-emitted primitive interaction with SV_RenderTargetArrayIndex — different primitives to different RT slices - 🏗️

Cross-cutting / Pipeline

Topologies

• Triangle list
  • All current graphics tests
• Triangle strip - 🏗️
• Line list - 🏗️
• Line strip - 🏗️
• Point list - 🏗️
• Triangle list with adjacency - 🏗️
• Line list with adjacency - 🏗️
• Patch lists (3, 4, 16, 32 control points) - 🏗️
• Primitive restart with strip topologies - 🏗️

Pipeline state

• Multiple render targets (NumRenderTargets > 1) - 🏗️
• Depth-only rendering (no color attachment) - 🏗️
• Stencil read / write - 🏗️
• Configurable blend state - 🏗️
• Conservative rasterization - 🏗️
• Variable-rate shading - 🏗️
• MSAA pipeline state (SampleCount > 1) - 🏗️

Multi-stage pipelines

• VS → HS → DS → PS (no GS) - 🏗️
• VS → GS → PS (no tessellation) - 🏗️
• VS → HS → DS → GS → PS (full pipeline) - 🏗️
• Resource visibility flags per stage (D3D12_SHADER_VISIBILITY_*) - 👍
• CBuffer accessed from multiple stages (deduplication, single binding) - 👍
• Push constants accessed from multiple stages - 👍
• Stage-interface signature compatibility (mismatched semantics between adjacent stages must error) - 👍

Cross-backend determinism

• Same shader produces byte-exact PNG across DX / VK / MTL on simple geometry - 👍
• Document DiffRMS / Intervals tolerance per goldenimage test - 👍
• WARP vs hardware GPU drift profile per test (existing tests already absorb it; document explicitly) - 👍