DeltaMaths 0.0.7

DeltaMaths

Portable engine-independent mathematics for .NET. Runtime targets netstandard2.0, netstandard2.1, net8.0 and net10.0, and has no Unity, renderer or System.Numerics dependency.

The API includes scalar utilities and deterministic 16.16 fix, bool/int/uint/float/double/half/fix vectors, geometry/interpolation/trigonometry, swizzles, generated GLSL-compatible float/double matrices and quaternion, plus normalized GLSL pack/unpack operations for float2 and float4 and bit-preserving double-word packing. It also provides conventional DeltaMaths.* and shader-like lowercase maths.* entry points.

fix stores a signed 16.16 value. Its public raw field is intentionally retained as a narrow ABI and serialization escape hatch for existing integrations; ordinary code should use the typed conversions and operators.

using Delta.Maths;
using static Delta.Maths.maths;

var direction = normalize(new float3(1f, 2f, 3f));
var transform = float4x4.CreateTRS(position, rotation, scale);

The generated matrix types cover every floatCxR and doubleCxR combination where C and R are 2, 3 or 4, matching GLSL's matCxR and dmatCxR names. C is the number of columns and R is the number of rows. Each matrix stores sequential column vectors (c0, c1, ...), so M * v uses column vectors and matches GLSL 4.60. In std430, a three-component float column has a 16-byte stride and a three-component double column has a 32-byte stride; the generated values contain explicit padding where required. The manifest records the column count, row count, alignment, stride and total size. Square matrices also provide determinant, inverse and a safe TryInverse; every matrix provides arithmetic, matrix/vector multiplication, transpose, matrixCompMult and outerProduct.

Each matrix has constructors from its column vectors, a diagonal scalar, all row-major Mij scalar components, and every other generated float or double matrix. The matrix-to-matrix constructor follows GLSL conversion rules: overlapping columns and rows are copied, missing off-diagonal elements are zero, and a missing diagonal element is one. The Mij scalar constructor retains the CLR row-major parameter order for compatibility; use the column constructor when spelling a GLSL column-major scalar sequence explicitly.

float4x4 is column-major as four sequential float4 columns. Translation is in c3.xyz; T * R * S applies scale, rotation, then translation. CPU code, GLSL and std430 share this convention.

src/DeltaMaths/Vectors/shader-contract.json is generated from the same declarations and is the committed generated DeltaMaths ABI consumed and validated by DeltaShader. DeltaMathsGen owns its declarations/generation; neither DeltaMaths nor DeltaShader recreates it. fix remains CPU-only. Float, double and half shader entries carry the std430, float64 or float16 capability required by their GLSL type; actual device/compiler support is decided by the shader consumer. GPU-only intrinsics stay in DeltaShader.

See CPU_GPU_CONFORMANCE.md for the cross-project CPU/Shader/Vulkan correctness protocol. Its bundles are test tooling, not part of the runtime DeltaMaths API.

See WORKFLOW.md for generation/build/test commands, TODO.md for selected work and AGENTS.md for routing.