I'm just curious what is the real motivation to disable the old way, i.e. how exactly it affect security ?

the function if somebody needs it see https://en.wikipedia.org/wiki/Invertible_matrix#Inversion_of_3_%C3%97_3_matrices static float3x3 inverse(float3x3 const m) { float const A = m[1][1] * m[2][2] - m[2][1] * m[1][2]; float const B = -(m[0][1] * m[2][2] - m[2][1] * m[0][2]); float const C = m[0][1] * m[1][2] - m[1][1] * m[0][2]; float const D = -(m[1][0] * m[2][2] - m[2][0] * m[1][2]); float const E = m[0][0] * m[2][2] - m[2][0] * m[0][2]; float const F = -(m[0][0] * m[1][2] - m[1][0] * m[0][2]); float const G = m[1][0] * m[2][1] - m[2][0] * m[1][1]; float const H = -(m[0][0] * m[2][1] - m[2][0] * m[0][1]); float const I = m[0][0] * m[1][1] - m[1][0] * m[0][1]; float const det = m[0][0] * A + m[1][0] * B + m[2][0] * C; float const inv_det = 1.f / det; return inv_det * float3x3{ float3{A, B, C}, float3{D, E, F}, float3{G, H, I} }; }

Nowadays the metal shading language specification states that it's possible, but still it doesn't specify the order, section 2.3.2 Matrix Constructors says: Since Metal 2, a matrix of type T with n columns and m rows > can also be constructed from n * m scalars of type T. The following examples are legal constructors: float2x2(float, float, float, float); float3x2(float, float, float, float, float, float); I'm too lazy to bounce data to the shader and back just to check, but I'd appreciate if anybody who knows would answer to the initial question, and even better would make a clarification in that document BTW the document: https://developer.apple.com/metal/Metal-Shading-Language-Specification.pdf