1 #include "btPolarDecomposition.h"
2 #include "btMinMax.h"
3 
4 namespace
5 {
abs_column_sum(const btMatrix3x3 & a,int i)6   btScalar abs_column_sum(const btMatrix3x3& a, int i)
7   {
8     return btFabs(a[0][i]) + btFabs(a[1][i]) + btFabs(a[2][i]);
9   }
10 
abs_row_sum(const btMatrix3x3 & a,int i)11   btScalar abs_row_sum(const btMatrix3x3& a, int i)
12   {
13     return btFabs(a[i][0]) + btFabs(a[i][1]) + btFabs(a[i][2]);
14   }
15 
p1_norm(const btMatrix3x3 & a)16   btScalar p1_norm(const btMatrix3x3& a)
17   {
18     const btScalar sum0 = abs_column_sum(a,0);
19     const btScalar sum1 = abs_column_sum(a,1);
20     const btScalar sum2 = abs_column_sum(a,2);
21     return btMax(btMax(sum0, sum1), sum2);
22   }
23 
pinf_norm(const btMatrix3x3 & a)24   btScalar pinf_norm(const btMatrix3x3& a)
25   {
26     const btScalar sum0 = abs_row_sum(a,0);
27     const btScalar sum1 = abs_row_sum(a,1);
28     const btScalar sum2 = abs_row_sum(a,2);
29     return btMax(btMax(sum0, sum1), sum2);
30   }
31 }
32 
33 
34 
btPolarDecomposition(btScalar tolerance,unsigned int maxIterations)35 btPolarDecomposition::btPolarDecomposition(btScalar tolerance, unsigned int maxIterations)
36 : m_tolerance(tolerance)
37 , m_maxIterations(maxIterations)
38 {
39 }
40 
decompose(const btMatrix3x3 & a,btMatrix3x3 & u,btMatrix3x3 & h) const41 unsigned int btPolarDecomposition::decompose(const btMatrix3x3& a, btMatrix3x3& u, btMatrix3x3& h) const
42 {
43   // Use the 'u' and 'h' matrices for intermediate calculations
44   u = a;
45   h = a.inverse();
46 
47   for (unsigned int i = 0; i < m_maxIterations; ++i)
48   {
49     const btScalar h_1 = p1_norm(h);
50     const btScalar h_inf = pinf_norm(h);
51     const btScalar u_1 = p1_norm(u);
52     const btScalar u_inf = pinf_norm(u);
53 
54     const btScalar h_norm = h_1 * h_inf;
55     const btScalar u_norm = u_1 * u_inf;
56 
57     // The matrix is effectively singular so we cannot invert it
58     if (btFuzzyZero(h_norm) || btFuzzyZero(u_norm))
59       break;
60 
61     const btScalar gamma = btPow(h_norm / u_norm, 0.25f);
62     const btScalar inv_gamma = btScalar(1.0) / gamma;
63 
64     // Determine the delta to 'u'
65     const btMatrix3x3 delta = (u * (gamma - btScalar(2.0)) + h.transpose() * inv_gamma) * btScalar(0.5);
66 
67     // Update the matrices
68     u += delta;
69     h = u.inverse();
70 
71     // Check for convergence
72     if (p1_norm(delta) <= m_tolerance * u_1)
73     {
74       h = u.transpose() * a;
75       h = (h + h.transpose()) * 0.5;
76       return i;
77     }
78   }
79 
80   // The algorithm has failed to converge to the specified tolerance, but we
81   // want to make sure that the matrices returned are in the right form.
82   h = u.transpose() * a;
83   h = (h + h.transpose()) * 0.5;
84 
85   return m_maxIterations;
86 }
87 
maxIterations() const88 unsigned int btPolarDecomposition::maxIterations() const
89 {
90   return m_maxIterations;
91 }
92 
polarDecompose(const btMatrix3x3 & a,btMatrix3x3 & u,btMatrix3x3 & h)93 unsigned int polarDecompose(const btMatrix3x3& a, btMatrix3x3& u, btMatrix3x3& h)
94 {
95   static btPolarDecomposition polar;
96   return polar.decompose(a, u, h);
97 }
98 
99