Inverting a 4x4 matrix

Question

I am looking for a sample code implementation on how to invert a 4x4 matrix. I know there is Gaussian eleminiation, LU decomposition, etc., but instead of looking at them in detail I am really just looking for the code to do this.

Language ideally C++, data is available in array of 16 floats in column-major order.

Solution 1

here:

bool gluInvertMatrix(const double m[16], double invOut[16])
{
    double inv[16], det;
    int i;

    inv[0] = m[5]  * m[10] * m[15] - 
             m[5]  * m[11] * m[14] - 
             m[9]  * m[6]  * m[15] + 
             m[9]  * m[7]  * m[14] +
             m[13] * m[6]  * m[11] - 
             m[13] * m[7]  * m[10];

    inv[4] = -m[4]  * m[10] * m[15] + 
              m[4]  * m[11] * m[14] + 
              m[8]  * m[6]  * m[15] - 
              m[8]  * m[7]  * m[14] - 
              m[12] * m[6]  * m[11] + 
              m[12] * m[7]  * m[10];

    inv[8] = m[4]  * m[9] * m[15] - 
             m[4]  * m[11] * m[13] - 
             m[8]  * m[5] * m[15] + 
             m[8]  * m[7] * m[13] + 
             m[12] * m[5] * m[11] - 
             m[12] * m[7] * m[9];

    inv[12] = -m[4]  * m[9] * m[14] + 
               m[4]  * m[10] * m[13] +
               m[8]  * m[5] * m[14] - 
               m[8]  * m[6] * m[13] - 
               m[12] * m[5] * m[10] + 
               m[12] * m[6] * m[9];

    inv[1] = -m[1]  * m[10] * m[15] + 
              m[1]  * m[11] * m[14] + 
              m[9]  * m[2] * m[15] - 
              m[9]  * m[3] * m[14] - 
              m[13] * m[2] * m[11] + 
              m[13] * m[3] * m[10];

    inv[5] = m[0]  * m[10] * m[15] - 
             m[0]  * m[11] * m[14] - 
             m[8]  * m[2] * m[15] + 
             m[8]  * m[3] * m[14] + 
             m[12] * m[2] * m[11] - 
             m[12] * m[3] * m[10];

    inv[9] = -m[0]  * m[9] * m[15] + 
              m[0]  * m[11] * m[13] + 
              m[8]  * m[1] * m[15] - 
              m[8]  * m[3] * m[13] - 
              m[12] * m[1] * m[11] + 
              m[12] * m[3] * m[9];

    inv[13] = m[0]  * m[9] * m[14] - 
              m[0]  * m[10] * m[13] - 
              m[8]  * m[1] * m[14] + 
              m[8]  * m[2] * m[13] + 
              m[12] * m[1] * m[10] - 
              m[12] * m[2] * m[9];

    inv[2] = m[1]  * m[6] * m[15] - 
             m[1]  * m[7] * m[14] - 
             m[5]  * m[2] * m[15] + 
             m[5]  * m[3] * m[14] + 
             m[13] * m[2] * m[7] - 
             m[13] * m[3] * m[6];

    inv[6] = -m[0]  * m[6] * m[15] + 
              m[0]  * m[7] * m[14] + 
              m[4]  * m[2] * m[15] - 
              m[4]  * m[3] * m[14] - 
              m[12] * m[2] * m[7] + 
              m[12] * m[3] * m[6];

    inv[10] = m[0]  * m[5] * m[15] - 
              m[0]  * m[7] * m[13] - 
              m[4]  * m[1] * m[15] + 
              m[4]  * m[3] * m[13] + 
              m[12] * m[1] * m[7] - 
              m[12] * m[3] * m[5];

    inv[14] = -m[0]  * m[5] * m[14] + 
               m[0]  * m[6] * m[13] + 
               m[4]  * m[1] * m[14] - 
               m[4]  * m[2] * m[13] - 
               m[12] * m[1] * m[6] + 
               m[12] * m[2] * m[5];

    inv[3] = -m[1] * m[6] * m[11] + 
              m[1] * m[7] * m[10] + 
              m[5] * m[2] * m[11] - 
              m[5] * m[3] * m[10] - 
              m[9] * m[2] * m[7] + 
              m[9] * m[3] * m[6];

    inv[7] = m[0] * m[6] * m[11] - 
             m[0] * m[7] * m[10] - 
             m[4] * m[2] * m[11] + 
             m[4] * m[3] * m[10] + 
             m[8] * m[2] * m[7] - 
             m[8] * m[3] * m[6];

    inv[11] = -m[0] * m[5] * m[11] + 
               m[0] * m[7] * m[9] + 
               m[4] * m[1] * m[11] - 
               m[4] * m[3] * m[9] - 
               m[8] * m[1] * m[7] + 
               m[8] * m[3] * m[5];

    inv[15] = m[0] * m[5] * m[10] - 
              m[0] * m[6] * m[9] - 
              m[4] * m[1] * m[10] + 
              m[4] * m[2] * m[9] + 
              m[8] * m[1] * m[6] - 
              m[8] * m[2] * m[5];

    det = m[0] * inv[0] + m[1] * inv[4] + m[2] * inv[8] + m[3] * inv[12];

    if (det == 0)
        return false;

    det = 1.0 / det;

    for (i = 0; i < 16; i++)
        invOut[i] = inv[i] * det;

    return true;
}

This was lifted from MESA implementation of the GLU library.

Solution 2

If anyone looking for more costumized code and "easier to read", then I got this

var A2323 = m.m22 * m.m33 - m.m23 * m.m32 ;
var A1323 = m.m21 * m.m33 - m.m23 * m.m31 ;
var A1223 = m.m21 * m.m32 - m.m22 * m.m31 ;
var A0323 = m.m20 * m.m33 - m.m23 * m.m30 ;
var A0223 = m.m20 * m.m32 - m.m22 * m.m30 ;
var A0123 = m.m20 * m.m31 - m.m21 * m.m30 ;
var A2313 = m.m12 * m.m33 - m.m13 * m.m32 ;
var A1313 = m.m11 * m.m33 - m.m13 * m.m31 ;
var A1213 = m.m11 * m.m32 - m.m12 * m.m31 ;
var A2312 = m.m12 * m.m23 - m.m13 * m.m22 ;
var A1312 = m.m11 * m.m23 - m.m13 * m.m21 ;
var A1212 = m.m11 * m.m22 - m.m12 * m.m21 ;
var A0313 = m.m10 * m.m33 - m.m13 * m.m30 ;
var A0213 = m.m10 * m.m32 - m.m12 * m.m30 ;
var A0312 = m.m10 * m.m23 - m.m13 * m.m20 ;
var A0212 = m.m10 * m.m22 - m.m12 * m.m20 ;
var A0113 = m.m10 * m.m31 - m.m11 * m.m30 ;
var A0112 = m.m10 * m.m21 - m.m11 * m.m20 ;

var det = m.m00 * ( m.m11 * A2323 - m.m12 * A1323 + m.m13 * A1223 ) 
    - m.m01 * ( m.m10 * A2323 - m.m12 * A0323 + m.m13 * A0223 ) 
    + m.m02 * ( m.m10 * A1323 - m.m11 * A0323 + m.m13 * A0123 ) 
    - m.m03 * ( m.m10 * A1223 - m.m11 * A0223 + m.m12 * A0123 ) ;
det = 1 / det;

return new Matrix4x4() {
   m00 = det *   ( m.m11 * A2323 - m.m12 * A1323 + m.m13 * A1223 ),
   m01 = det * - ( m.m01 * A2323 - m.m02 * A1323 + m.m03 * A1223 ),
   m02 = det *   ( m.m01 * A2313 - m.m02 * A1313 + m.m03 * A1213 ),
   m03 = det * - ( m.m01 * A2312 - m.m02 * A1312 + m.m03 * A1212 ),
   m10 = det * - ( m.m10 * A2323 - m.m12 * A0323 + m.m13 * A0223 ),
   m11 = det *   ( m.m00 * A2323 - m.m02 * A0323 + m.m03 * A0223 ),
   m12 = det * - ( m.m00 * A2313 - m.m02 * A0313 + m.m03 * A0213 ),
   m13 = det *   ( m.m00 * A2312 - m.m02 * A0312 + m.m03 * A0212 ),
   m20 = det *   ( m.m10 * A1323 - m.m11 * A0323 + m.m13 * A0123 ),
   m21 = det * - ( m.m00 * A1323 - m.m01 * A0323 + m.m03 * A0123 ),
   m22 = det *   ( m.m00 * A1313 - m.m01 * A0313 + m.m03 * A0113 ),
   m23 = det * - ( m.m00 * A1312 - m.m01 * A0312 + m.m03 * A0112 ),
   m30 = det * - ( m.m10 * A1223 - m.m11 * A0223 + m.m12 * A0123 ),
   m31 = det *   ( m.m00 * A1223 - m.m01 * A0223 + m.m02 * A0123 ),
   m32 = det * - ( m.m00 * A1213 - m.m01 * A0213 + m.m02 * A0113 ),
   m33 = det *   ( m.m00 * A1212 - m.m01 * A0212 + m.m02 * A0112 ),
};

I don't write the code, but my program did. I made a small program to make a program that calculate the determinant and inverse of any N-matrix.

I do it because once in the past I need a code that inverses 5x5 matrix, but nobody in the earth have done this so I made one.

Take a look about the program here.

EDIT: The matrix layout is row-by-row (meaning m01 is in the first row and second column). Also the language is C#, but should be easy to convert into C.

Solution 3

This is the C++ version for @willnode's answer

template<typename Matrix>
static inline void InvertMatrix4(const Matrix& m, Matrix& im, double& det)
{
    double A2323 = m(2, 2) * m(3, 3) - m(2, 3) * m(3, 2);
    double A1323 = m(2, 1) * m(3, 3) - m(2, 3) * m(3, 1);
    double A1223 = m(2, 1) * m(3, 2) - m(2, 2) * m(3, 1);
    double A0323 = m(2, 0) * m(3, 3) - m(2, 3) * m(3, 0);
    double A0223 = m(2, 0) * m(3, 2) - m(2, 2) * m(3, 0);
    double A0123 = m(2, 0) * m(3, 1) - m(2, 1) * m(3, 0);
    double A2313 = m(1, 2) * m(3, 3) - m(1, 3) * m(3, 2);
    double A1313 = m(1, 1) * m(3, 3) - m(1, 3) * m(3, 1);
    double A1213 = m(1, 1) * m(3, 2) - m(1, 2) * m(3, 1);
    double A2312 = m(1, 2) * m(2, 3) - m(1, 3) * m(2, 2);
    double A1312 = m(1, 1) * m(2, 3) - m(1, 3) * m(2, 1);
    double A1212 = m(1, 1) * m(2, 2) - m(1, 2) * m(2, 1);
    double A0313 = m(1, 0) * m(3, 3) - m(1, 3) * m(3, 0);
    double A0213 = m(1, 0) * m(3, 2) - m(1, 2) * m(3, 0);
    double A0312 = m(1, 0) * m(2, 3) - m(1, 3) * m(2, 0);
    double A0212 = m(1, 0) * m(2, 2) - m(1, 2) * m(2, 0);
    double A0113 = m(1, 0) * m(3, 1) - m(1, 1) * m(3, 0);
    double A0112 = m(1, 0) * m(2, 1) - m(1, 1) * m(2, 0);

    det = m(0, 0) * ( m(1, 1) * A2323 - m(1, 2) * A1323 + m(1, 3) * A1223 )
        - m(0, 1) * ( m(1, 0) * A2323 - m(1, 2) * A0323 + m(1, 3) * A0223 )
        + m(0, 2) * ( m(1, 0) * A1323 - m(1, 1) * A0323 + m(1, 3) * A0123 )
        - m(0, 3) * ( m(1, 0) * A1223 - m(1, 1) * A0223 + m(1, 2) * A0123 );
    det = 1 / det;

    im(0, 0) = det *   ( m(1, 1) * A2323 - m(1, 2) * A1323 + m(1, 3) * A1223 );
    im(0, 1) = det * - ( m(0, 1) * A2323 - m(0, 2) * A1323 + m(0, 3) * A1223 );
    im(0, 2) = det *   ( m(0, 1) * A2313 - m(0, 2) * A1313 + m(0, 3) * A1213 );
    im(0, 3) = det * - ( m(0, 1) * A2312 - m(0, 2) * A1312 + m(0, 3) * A1212 );
    im(1, 0) = det * - ( m(1, 0) * A2323 - m(1, 2) * A0323 + m(1, 3) * A0223 );
    im(1, 1) = det *   ( m(0, 0) * A2323 - m(0, 2) * A0323 + m(0, 3) * A0223 );
    im(1, 2) = det * - ( m(0, 0) * A2313 - m(0, 2) * A0313 + m(0, 3) * A0213 );
    im(1, 3) = det *   ( m(0, 0) * A2312 - m(0, 2) * A0312 + m(0, 3) * A0212 );
    im(2, 0) = det *   ( m(1, 0) * A1323 - m(1, 1) * A0323 + m(1, 3) * A0123 );
    im(2, 1) = det * - ( m(0, 0) * A1323 - m(0, 1) * A0323 + m(0, 3) * A0123 );
    im(2, 2) = det *   ( m(0, 0) * A1313 - m(0, 1) * A0313 + m(0, 3) * A0113 );
    im(2, 3) = det * - ( m(0, 0) * A1312 - m(0, 1) * A0312 + m(0, 3) * A0112 );
    im(3, 0) = det * - ( m(1, 0) * A1223 - m(1, 1) * A0223 + m(1, 2) * A0123 );
    im(3, 1) = det *   ( m(0, 0) * A1223 - m(0, 1) * A0223 + m(0, 2) * A0123 );
    im(3, 2) = det * - ( m(0, 0) * A1213 - m(0, 1) * A0213 + m(0, 2) * A0113 );
    im(3, 3) = det *   ( m(0, 0) * A1212 - m(0, 1) * A0212 + m(0, 2) * A0112 );
}

Solution 4

You can use the GNU Scientific Library or look the code up in it.

Edit: You seem to want the Linear Algebra section.

Solution 5

Here is a small (just one header) C++ vector math library (geared towards 3D programming). If you use it, keep in mind that layout of its matrices in memory is inverted comparing to what OpenGL expects, I had fun time figuring it out...

View more solutions

Comments

[shoosh]

You probably wouldn't want it any other way.

[Imagist]

Yes I would. Compilers are perfectly capable of unrolling loops, especially when you tell them to.

[Maciej Piechotka]

Even if they wouldn't that might be time to invest into metaprogramming (here's place for loop). I workarounded one compiler which did not unroll trivial loops by using m4.

[fluffy]

Sadly, that code isn't really that straightforward to make in a loopable way to begin with, much less a way that a compiler can adequately unroll. Also, that code comes from a rather old C library which has a LOT of very fiddly optimizations, and it's code that works already (and has been thoroughly tested and proven by thousands of Linux OpenGL programs at this point) so why rewrite it?

[Zoomulator]

Is this for column major or row major ordered matrices?

[Timmmm]

Imagist: It wasn't derived from an unrolled loop. That really is the simplest way of writing it.

[Timmmm]

Zoomulator: Awesomely it is for both! This is because inverse(transpose(A)) = transpose(inverse(A)).

[dfeuer]

Why not calculate the determinant first? Is this attempting realtime consistency or something?

[Admin]

@Imagist you can check my answer if you're looking for a 'looping variant'. Aggressive compiler optimisations won't fix the performance hit...

[crazyjoe]

@dfeuer It would take 72 multiplications to get the determinant first. While doing it first would actually be faster in case of non-invertible matrix, for an invertible matrix it saves 68 multiplications by doing it after. Also there's the bonus of having less code this way.

[Alexandre C.]

Comatrix-based inversion like this is going to be very prone to roundoff error (look at all the potential for catastrophic cancellation). Stabler versions include branches (ie. pivoting) however, so this is kind of a trade-off. I suspect that Mesa programmers prefer performance here.

[Bim]

Strainghtforward implementation. @crazyjoe: You could move the calculation of inv[0], inv[4], inv[8], inv[12] to the top and calculate the determinant earlier, so you optimize the non-invertible path too...

[Alundaio]

I don't understand, this code is moving m12,m13,m14 to m3,m7,m11 where other algorithms will just leave the last column in place.

[paddyg]

116 float multiplications compared with 200 for the accepted answer. (and determinant checking before doing majority of calculation)

[PhilT]

@BrynnMahsman From the looks of it, it's based on Minors, cofactors and adjugate - mathsisfun.com/algebra/…. The other method would be column and row operations but I think this would be difficult to code as you'd have to use conditional logic

[Stuntddude]

This answer feels like a gift from God. You even use the same naming convention for matrix elements as I do.

[Krupip]

@fluffy This is just using determinants isn't it? This should be fairly "loopable".

[fluffy]

@whn Where are the access patterns that can be turned into unrollable loops in a way that the compiler would be likely to be able to unroll this concisely? Also why ping me specifically on a comment I posted on an answer 9 years ago?

[wychmaster]

Really good answer, but regarding this " I need a code that inverses 5x5 matrix, but nobody in the earth have done this" --- The reason for that is probably, that it is cheaper to use a direct solver (Gauss, LU) than using a formula based on determinants (Cramers rule?).

[anymous.asker]

I should add, this code works very well in general, but it will produce quite incorrect results when the determinant of the matrix is low (say, <= 1e-4).