1 | // This file is part of Eigen, a lightweight C++ template library
|
---|
2 | // for linear algebra.
|
---|
3 | //
|
---|
4 | // Copyright (C) 2008-2009 Gael Guennebaud <gael.guennebaud@inria.fr>
|
---|
5 | // Copyright (C) 2009 Mathieu Gautier <mathieu.gautier@cea.fr>
|
---|
6 | //
|
---|
7 | // This Source Code Form is subject to the terms of the Mozilla
|
---|
8 | // Public License v. 2.0. If a copy of the MPL was not distributed
|
---|
9 | // with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
|
---|
10 |
|
---|
11 | #include "main.h"
|
---|
12 | #include <Eigen/Geometry>
|
---|
13 | #include <Eigen/LU>
|
---|
14 | #include <Eigen/SVD>
|
---|
15 |
|
---|
16 | template<typename T> T bounded_acos(T v)
|
---|
17 | {
|
---|
18 | using std::acos;
|
---|
19 | using std::min;
|
---|
20 | using std::max;
|
---|
21 | return acos((max)(T(-1),(min)(v,T(1))));
|
---|
22 | }
|
---|
23 |
|
---|
24 | template<typename QuatType> void check_slerp(const QuatType& q0, const QuatType& q1)
|
---|
25 | {
|
---|
26 | using std::abs;
|
---|
27 | typedef typename QuatType::Scalar Scalar;
|
---|
28 | typedef AngleAxis<Scalar> AA;
|
---|
29 |
|
---|
30 | Scalar largeEps = test_precision<Scalar>();
|
---|
31 |
|
---|
32 | Scalar theta_tot = AA(q1*q0.inverse()).angle();
|
---|
33 | if(theta_tot>M_PI)
|
---|
34 | theta_tot = Scalar(2.*M_PI)-theta_tot;
|
---|
35 | for(Scalar t=0; t<=Scalar(1.001); t+=Scalar(0.1))
|
---|
36 | {
|
---|
37 | QuatType q = q0.slerp(t,q1);
|
---|
38 | Scalar theta = AA(q*q0.inverse()).angle();
|
---|
39 | VERIFY(abs(q.norm() - 1) < largeEps);
|
---|
40 | if(theta_tot==0) VERIFY(theta_tot==0);
|
---|
41 | else VERIFY(abs(theta - t * theta_tot) < largeEps);
|
---|
42 | }
|
---|
43 | }
|
---|
44 |
|
---|
45 | template<typename Scalar, int Options> void quaternion(void)
|
---|
46 | {
|
---|
47 | /* this test covers the following files:
|
---|
48 | Quaternion.h
|
---|
49 | */
|
---|
50 | using std::abs;
|
---|
51 | typedef Matrix<Scalar,3,1> Vector3;
|
---|
52 | typedef Matrix<Scalar,4,1> Vector4;
|
---|
53 | typedef Quaternion<Scalar,Options> Quaternionx;
|
---|
54 | typedef AngleAxis<Scalar> AngleAxisx;
|
---|
55 |
|
---|
56 | Scalar largeEps = test_precision<Scalar>();
|
---|
57 | if (internal::is_same<Scalar,float>::value)
|
---|
58 | largeEps = 1e-3f;
|
---|
59 |
|
---|
60 | Scalar eps = internal::random<Scalar>() * Scalar(1e-2);
|
---|
61 |
|
---|
62 | Vector3 v0 = Vector3::Random(),
|
---|
63 | v1 = Vector3::Random(),
|
---|
64 | v2 = Vector3::Random(),
|
---|
65 | v3 = Vector3::Random();
|
---|
66 |
|
---|
67 | Scalar a = internal::random<Scalar>(-Scalar(M_PI), Scalar(M_PI)),
|
---|
68 | b = internal::random<Scalar>(-Scalar(M_PI), Scalar(M_PI));
|
---|
69 |
|
---|
70 | // Quaternion: Identity(), setIdentity();
|
---|
71 | Quaternionx q1, q2;
|
---|
72 | q2.setIdentity();
|
---|
73 | VERIFY_IS_APPROX(Quaternionx(Quaternionx::Identity()).coeffs(), q2.coeffs());
|
---|
74 | q1.coeffs().setRandom();
|
---|
75 | VERIFY_IS_APPROX(q1.coeffs(), (q1*q2).coeffs());
|
---|
76 |
|
---|
77 | // concatenation
|
---|
78 | q1 *= q2;
|
---|
79 |
|
---|
80 | q1 = AngleAxisx(a, v0.normalized());
|
---|
81 | q2 = AngleAxisx(a, v1.normalized());
|
---|
82 |
|
---|
83 | // angular distance
|
---|
84 | Scalar refangle = abs(AngleAxisx(q1.inverse()*q2).angle());
|
---|
85 | if (refangle>Scalar(M_PI))
|
---|
86 | refangle = Scalar(2)*Scalar(M_PI) - refangle;
|
---|
87 |
|
---|
88 | if((q1.coeffs()-q2.coeffs()).norm() > 10*largeEps)
|
---|
89 | {
|
---|
90 | VERIFY_IS_MUCH_SMALLER_THAN(abs(q1.angularDistance(q2) - refangle), Scalar(1));
|
---|
91 | }
|
---|
92 |
|
---|
93 | // rotation matrix conversion
|
---|
94 | VERIFY_IS_APPROX(q1 * v2, q1.toRotationMatrix() * v2);
|
---|
95 | VERIFY_IS_APPROX(q1 * q2 * v2,
|
---|
96 | q1.toRotationMatrix() * q2.toRotationMatrix() * v2);
|
---|
97 |
|
---|
98 | VERIFY( (q2*q1).isApprox(q1*q2, largeEps)
|
---|
99 | || !(q2 * q1 * v2).isApprox(q1.toRotationMatrix() * q2.toRotationMatrix() * v2));
|
---|
100 |
|
---|
101 | q2 = q1.toRotationMatrix();
|
---|
102 | VERIFY_IS_APPROX(q1*v1,q2*v1);
|
---|
103 |
|
---|
104 |
|
---|
105 | // angle-axis conversion
|
---|
106 | AngleAxisx aa = AngleAxisx(q1);
|
---|
107 | VERIFY_IS_APPROX(q1 * v1, Quaternionx(aa) * v1);
|
---|
108 |
|
---|
109 | // Do not execute the test if the rotation angle is almost zero, or
|
---|
110 | // the rotation axis and v1 are almost parallel.
|
---|
111 | if (abs(aa.angle()) > 5*test_precision<Scalar>()
|
---|
112 | && (aa.axis() - v1.normalized()).norm() < 1.99
|
---|
113 | && (aa.axis() + v1.normalized()).norm() < 1.99)
|
---|
114 | {
|
---|
115 | VERIFY_IS_NOT_APPROX(q1 * v1, Quaternionx(AngleAxisx(aa.angle()*2,aa.axis())) * v1);
|
---|
116 | }
|
---|
117 |
|
---|
118 | // from two vector creation
|
---|
119 | VERIFY_IS_APPROX( v2.normalized(),(q2.setFromTwoVectors(v1, v2)*v1).normalized());
|
---|
120 | VERIFY_IS_APPROX( v1.normalized(),(q2.setFromTwoVectors(v1, v1)*v1).normalized());
|
---|
121 | VERIFY_IS_APPROX(-v1.normalized(),(q2.setFromTwoVectors(v1,-v1)*v1).normalized());
|
---|
122 | if (internal::is_same<Scalar,double>::value)
|
---|
123 | {
|
---|
124 | v3 = (v1.array()+eps).matrix();
|
---|
125 | VERIFY_IS_APPROX( v3.normalized(),(q2.setFromTwoVectors(v1, v3)*v1).normalized());
|
---|
126 | VERIFY_IS_APPROX(-v3.normalized(),(q2.setFromTwoVectors(v1,-v3)*v1).normalized());
|
---|
127 | }
|
---|
128 |
|
---|
129 | // from two vector creation static function
|
---|
130 | VERIFY_IS_APPROX( v2.normalized(),(Quaternionx::FromTwoVectors(v1, v2)*v1).normalized());
|
---|
131 | VERIFY_IS_APPROX( v1.normalized(),(Quaternionx::FromTwoVectors(v1, v1)*v1).normalized());
|
---|
132 | VERIFY_IS_APPROX(-v1.normalized(),(Quaternionx::FromTwoVectors(v1,-v1)*v1).normalized());
|
---|
133 | if (internal::is_same<Scalar,double>::value)
|
---|
134 | {
|
---|
135 | v3 = (v1.array()+eps).matrix();
|
---|
136 | VERIFY_IS_APPROX( v3.normalized(),(Quaternionx::FromTwoVectors(v1, v3)*v1).normalized());
|
---|
137 | VERIFY_IS_APPROX(-v3.normalized(),(Quaternionx::FromTwoVectors(v1,-v3)*v1).normalized());
|
---|
138 | }
|
---|
139 |
|
---|
140 | // inverse and conjugate
|
---|
141 | VERIFY_IS_APPROX(q1 * (q1.inverse() * v1), v1);
|
---|
142 | VERIFY_IS_APPROX(q1 * (q1.conjugate() * v1), v1);
|
---|
143 |
|
---|
144 | // test casting
|
---|
145 | Quaternion<float> q1f = q1.template cast<float>();
|
---|
146 | VERIFY_IS_APPROX(q1f.template cast<Scalar>(),q1);
|
---|
147 | Quaternion<double> q1d = q1.template cast<double>();
|
---|
148 | VERIFY_IS_APPROX(q1d.template cast<Scalar>(),q1);
|
---|
149 |
|
---|
150 | // test bug 369 - improper alignment.
|
---|
151 | Quaternionx *q = new Quaternionx;
|
---|
152 | delete q;
|
---|
153 |
|
---|
154 | q1 = AngleAxisx(a, v0.normalized());
|
---|
155 | q2 = AngleAxisx(b, v1.normalized());
|
---|
156 | check_slerp(q1,q2);
|
---|
157 |
|
---|
158 | q1 = AngleAxisx(b, v1.normalized());
|
---|
159 | q2 = AngleAxisx(b+Scalar(M_PI), v1.normalized());
|
---|
160 | check_slerp(q1,q2);
|
---|
161 |
|
---|
162 | q1 = AngleAxisx(b, v1.normalized());
|
---|
163 | q2 = AngleAxisx(-b, -v1.normalized());
|
---|
164 | check_slerp(q1,q2);
|
---|
165 |
|
---|
166 | q1.coeffs() = Vector4::Random().normalized();
|
---|
167 | q2.coeffs() = -q1.coeffs();
|
---|
168 | check_slerp(q1,q2);
|
---|
169 | }
|
---|
170 |
|
---|
171 | template<typename Scalar> void mapQuaternion(void){
|
---|
172 | typedef Map<Quaternion<Scalar>, Aligned> MQuaternionA;
|
---|
173 | typedef Map<const Quaternion<Scalar>, Aligned> MCQuaternionA;
|
---|
174 | typedef Map<Quaternion<Scalar> > MQuaternionUA;
|
---|
175 | typedef Map<const Quaternion<Scalar> > MCQuaternionUA;
|
---|
176 | typedef Quaternion<Scalar> Quaternionx;
|
---|
177 | typedef Matrix<Scalar,3,1> Vector3;
|
---|
178 | typedef AngleAxis<Scalar> AngleAxisx;
|
---|
179 |
|
---|
180 | Vector3 v0 = Vector3::Random(),
|
---|
181 | v1 = Vector3::Random();
|
---|
182 | Scalar a = internal::random<Scalar>(-Scalar(M_PI), Scalar(M_PI));
|
---|
183 |
|
---|
184 | EIGEN_ALIGN16 Scalar array1[4];
|
---|
185 | EIGEN_ALIGN16 Scalar array2[4];
|
---|
186 | EIGEN_ALIGN16 Scalar array3[4+1];
|
---|
187 | Scalar* array3unaligned = array3+1;
|
---|
188 |
|
---|
189 | MQuaternionA mq1(array1);
|
---|
190 | MCQuaternionA mcq1(array1);
|
---|
191 | MQuaternionA mq2(array2);
|
---|
192 | MQuaternionUA mq3(array3unaligned);
|
---|
193 | MCQuaternionUA mcq3(array3unaligned);
|
---|
194 |
|
---|
195 | // std::cerr << array1 << " " << array2 << " " << array3 << "\n";
|
---|
196 | mq1 = AngleAxisx(a, v0.normalized());
|
---|
197 | mq2 = mq1;
|
---|
198 | mq3 = mq1;
|
---|
199 |
|
---|
200 | Quaternionx q1 = mq1;
|
---|
201 | Quaternionx q2 = mq2;
|
---|
202 | Quaternionx q3 = mq3;
|
---|
203 | Quaternionx q4 = MCQuaternionUA(array3unaligned);
|
---|
204 |
|
---|
205 | VERIFY_IS_APPROX(q1.coeffs(), q2.coeffs());
|
---|
206 | VERIFY_IS_APPROX(q1.coeffs(), q3.coeffs());
|
---|
207 | VERIFY_IS_APPROX(q4.coeffs(), q3.coeffs());
|
---|
208 | #ifdef EIGEN_VECTORIZE
|
---|
209 | if(internal::packet_traits<Scalar>::Vectorizable)
|
---|
210 | VERIFY_RAISES_ASSERT((MQuaternionA(array3unaligned)));
|
---|
211 | #endif
|
---|
212 |
|
---|
213 | VERIFY_IS_APPROX(mq1 * (mq1.inverse() * v1), v1);
|
---|
214 | VERIFY_IS_APPROX(mq1 * (mq1.conjugate() * v1), v1);
|
---|
215 |
|
---|
216 | VERIFY_IS_APPROX(mcq1 * (mcq1.inverse() * v1), v1);
|
---|
217 | VERIFY_IS_APPROX(mcq1 * (mcq1.conjugate() * v1), v1);
|
---|
218 |
|
---|
219 | VERIFY_IS_APPROX(mq3 * (mq3.inverse() * v1), v1);
|
---|
220 | VERIFY_IS_APPROX(mq3 * (mq3.conjugate() * v1), v1);
|
---|
221 |
|
---|
222 | VERIFY_IS_APPROX(mcq3 * (mcq3.inverse() * v1), v1);
|
---|
223 | VERIFY_IS_APPROX(mcq3 * (mcq3.conjugate() * v1), v1);
|
---|
224 |
|
---|
225 | VERIFY_IS_APPROX(mq1*mq2, q1*q2);
|
---|
226 | VERIFY_IS_APPROX(mq3*mq2, q3*q2);
|
---|
227 | VERIFY_IS_APPROX(mcq1*mq2, q1*q2);
|
---|
228 | VERIFY_IS_APPROX(mcq3*mq2, q3*q2);
|
---|
229 | }
|
---|
230 |
|
---|
231 | template<typename Scalar> void quaternionAlignment(void){
|
---|
232 | typedef Quaternion<Scalar,AutoAlign> QuaternionA;
|
---|
233 | typedef Quaternion<Scalar,DontAlign> QuaternionUA;
|
---|
234 |
|
---|
235 | EIGEN_ALIGN16 Scalar array1[4];
|
---|
236 | EIGEN_ALIGN16 Scalar array2[4];
|
---|
237 | EIGEN_ALIGN16 Scalar array3[4+1];
|
---|
238 | Scalar* arrayunaligned = array3+1;
|
---|
239 |
|
---|
240 | QuaternionA *q1 = ::new(reinterpret_cast<void*>(array1)) QuaternionA;
|
---|
241 | QuaternionUA *q2 = ::new(reinterpret_cast<void*>(array2)) QuaternionUA;
|
---|
242 | QuaternionUA *q3 = ::new(reinterpret_cast<void*>(arrayunaligned)) QuaternionUA;
|
---|
243 |
|
---|
244 | q1->coeffs().setRandom();
|
---|
245 | *q2 = *q1;
|
---|
246 | *q3 = *q1;
|
---|
247 |
|
---|
248 | VERIFY_IS_APPROX(q1->coeffs(), q2->coeffs());
|
---|
249 | VERIFY_IS_APPROX(q1->coeffs(), q3->coeffs());
|
---|
250 | #if defined(EIGEN_VECTORIZE) && EIGEN_ALIGN_STATICALLY
|
---|
251 | if(internal::packet_traits<Scalar>::Vectorizable)
|
---|
252 | VERIFY_RAISES_ASSERT((::new(reinterpret_cast<void*>(arrayunaligned)) QuaternionA));
|
---|
253 | #endif
|
---|
254 | }
|
---|
255 |
|
---|
256 | template<typename PlainObjectType> void check_const_correctness(const PlainObjectType&)
|
---|
257 | {
|
---|
258 | // there's a lot that we can't test here while still having this test compile!
|
---|
259 | // the only possible approach would be to run a script trying to compile stuff and checking that it fails.
|
---|
260 | // CMake can help with that.
|
---|
261 |
|
---|
262 | // verify that map-to-const don't have LvalueBit
|
---|
263 | typedef typename internal::add_const<PlainObjectType>::type ConstPlainObjectType;
|
---|
264 | VERIFY( !(internal::traits<Map<ConstPlainObjectType> >::Flags & LvalueBit) );
|
---|
265 | VERIFY( !(internal::traits<Map<ConstPlainObjectType, Aligned> >::Flags & LvalueBit) );
|
---|
266 | VERIFY( !(Map<ConstPlainObjectType>::Flags & LvalueBit) );
|
---|
267 | VERIFY( !(Map<ConstPlainObjectType, Aligned>::Flags & LvalueBit) );
|
---|
268 | }
|
---|
269 |
|
---|
270 | void test_geo_quaternion()
|
---|
271 | {
|
---|
272 | for(int i = 0; i < g_repeat; i++) {
|
---|
273 | CALL_SUBTEST_1(( quaternion<float,AutoAlign>() ));
|
---|
274 | CALL_SUBTEST_1( check_const_correctness(Quaternionf()) );
|
---|
275 | CALL_SUBTEST_2(( quaternion<double,AutoAlign>() ));
|
---|
276 | CALL_SUBTEST_2( check_const_correctness(Quaterniond()) );
|
---|
277 | CALL_SUBTEST_3(( quaternion<float,DontAlign>() ));
|
---|
278 | CALL_SUBTEST_4(( quaternion<double,DontAlign>() ));
|
---|
279 | CALL_SUBTEST_5(( quaternionAlignment<float>() ));
|
---|
280 | CALL_SUBTEST_6(( quaternionAlignment<double>() ));
|
---|
281 | CALL_SUBTEST_1( mapQuaternion<float>() );
|
---|
282 | CALL_SUBTEST_2( mapQuaternion<double>() );
|
---|
283 | }
|
---|
284 | }
|
---|