A vector math library (like package:vector_math) optimized for 2D use-cases.
3D is supported to, but there's no current goal to beat package:vector_math at
general 3D operations, merely to be on par. 2D use-cases should be many times
faster though.
This package makes the following hypotheses about vector math usage in typical 2D apps, such as the Flutter framework and engine, and apps using them:
- Identity and 2D translations are the most common kinds of matrices used. Therefore the object representation is optimized for these kinds of matrices.
- Precision higher than float32 is overkill for the 2D UI use-case.
With these assumptions, the package makes the following design choices to maximize performance:
- The matrix class is monomorphic for fastest method/field access.
- The matrix class is deeply immutable and const.
- Zero and identity matrices are canonicalized: there's exactly one constant instance for each. This makes checking for identity trivial.
- Matrix initialization lowers to the most specific matrix kind. If a 4x4 matrix is actually an identity matrix, the constructor will return the identity constant.
- Specialized algorithms based on matrix kind.
- Input types are always the most concrete numeric types (doubles, typed
list), unlike
package:vector_maththat takes varargs and dynamic types and performs runtime pattern matching on arguments. - Minimize boxing: a 2D translation (including identity) only needs x and y values. No need for extra arrays.