xref: /aosp_15_r20/external/eigen/test/stable_norm.cpp (revision bf2c37156dfe67e5dfebd6d394bad8b2ab5804d4)
1 // This file is part of Eigen, a lightweight C++ template library
2 // for linear algebra.
3 //
4 // Copyright (C) 2009-2014 Gael Guennebaud <[email protected]>
5 //
6 // This Source Code Form is subject to the terms of the Mozilla
7 // Public License v. 2.0. If a copy of the MPL was not distributed
8 // with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
9 
10 #include "main.h"
11 
copy(const T & x)12 template<typename T> EIGEN_DONT_INLINE T copy(const T& x)
13 {
14   return x;
15 }
16 
stable_norm(const MatrixType & m)17 template<typename MatrixType> void stable_norm(const MatrixType& m)
18 {
19   /* this test covers the following files:
20      StableNorm.h
21   */
22   using std::sqrt;
23   using std::abs;
24   typedef typename MatrixType::Scalar Scalar;
25   typedef typename NumTraits<Scalar>::Real RealScalar;
26 
27   bool complex_real_product_ok = true;
28 
29   // Check the basic machine-dependent constants.
30   {
31     int ibeta, it, iemin, iemax;
32 
33     ibeta = std::numeric_limits<RealScalar>::radix;         // base for floating-point numbers
34     it    = std::numeric_limits<RealScalar>::digits;        // number of base-beta digits in mantissa
35     iemin = std::numeric_limits<RealScalar>::min_exponent;  // minimum exponent
36     iemax = std::numeric_limits<RealScalar>::max_exponent;  // maximum exponent
37 
38     VERIFY( (!(iemin > 1 - 2*it || 1+it>iemax || (it==2 && ibeta<5) || (it<=4 && ibeta <= 3 ) || it<2))
39            && "the stable norm algorithm cannot be guaranteed on this computer");
40 
41     Scalar inf = std::numeric_limits<RealScalar>::infinity();
42     if(NumTraits<Scalar>::IsComplex && (numext::isnan)(inf*RealScalar(1)) )
43     {
44       complex_real_product_ok = false;
45       static bool first = true;
46       if(first)
47         std::cerr << "WARNING: compiler mess up complex*real product, " << inf << " * " << 1.0 << " = " << inf*RealScalar(1) << std::endl;
48       first = false;
49     }
50   }
51 
52 
53   Index rows = m.rows();
54   Index cols = m.cols();
55 
56   // get a non-zero random factor
57   Scalar factor = internal::random<Scalar>();
58   while(numext::abs2(factor)<RealScalar(1e-4))
59     factor = internal::random<Scalar>();
60   Scalar big = factor * ((std::numeric_limits<RealScalar>::max)() * RealScalar(1e-4));
61 
62   factor = internal::random<Scalar>();
63   while(numext::abs2(factor)<RealScalar(1e-4))
64     factor = internal::random<Scalar>();
65   Scalar small = factor * ((std::numeric_limits<RealScalar>::min)() * RealScalar(1e4));
66 
67   Scalar one(1);
68 
69   MatrixType  vzero = MatrixType::Zero(rows, cols),
70               vrand = MatrixType::Random(rows, cols),
71               vbig(rows, cols),
72               vsmall(rows,cols);
73 
74   vbig.fill(big);
75   vsmall.fill(small);
76 
77   VERIFY_IS_MUCH_SMALLER_THAN(vzero.norm(), static_cast<RealScalar>(1));
78   VERIFY_IS_APPROX(vrand.stableNorm(),      vrand.norm());
79   VERIFY_IS_APPROX(vrand.blueNorm(),        vrand.norm());
80   VERIFY_IS_APPROX(vrand.hypotNorm(),       vrand.norm());
81 
82   // test with expressions as input
83   VERIFY_IS_APPROX((one*vrand).stableNorm(),      vrand.norm());
84   VERIFY_IS_APPROX((one*vrand).blueNorm(),        vrand.norm());
85   VERIFY_IS_APPROX((one*vrand).hypotNorm(),       vrand.norm());
86   VERIFY_IS_APPROX((one*vrand+one*vrand-one*vrand).stableNorm(),      vrand.norm());
87   VERIFY_IS_APPROX((one*vrand+one*vrand-one*vrand).blueNorm(),        vrand.norm());
88   VERIFY_IS_APPROX((one*vrand+one*vrand-one*vrand).hypotNorm(),       vrand.norm());
89 
90   RealScalar size = static_cast<RealScalar>(m.size());
91 
92   // test numext::isfinite
93   VERIFY(!(numext::isfinite)( std::numeric_limits<RealScalar>::infinity()));
94   VERIFY(!(numext::isfinite)(sqrt(-abs(big))));
95 
96   // test overflow
97   VERIFY((numext::isfinite)(sqrt(size)*abs(big)));
98   VERIFY_IS_NOT_APPROX(sqrt(copy(vbig.squaredNorm())), abs(sqrt(size)*big)); // here the default norm must fail
99   VERIFY_IS_APPROX(vbig.stableNorm(), sqrt(size)*abs(big));
100   VERIFY_IS_APPROX(vbig.blueNorm(),   sqrt(size)*abs(big));
101   VERIFY_IS_APPROX(vbig.hypotNorm(),  sqrt(size)*abs(big));
102 
103   // test underflow
104   VERIFY((numext::isfinite)(sqrt(size)*abs(small)));
105   VERIFY_IS_NOT_APPROX(sqrt(copy(vsmall.squaredNorm())),   abs(sqrt(size)*small)); // here the default norm must fail
106   VERIFY_IS_APPROX(vsmall.stableNorm(), sqrt(size)*abs(small));
107   VERIFY_IS_APPROX(vsmall.blueNorm(),   sqrt(size)*abs(small));
108   VERIFY_IS_APPROX(vsmall.hypotNorm(),  sqrt(size)*abs(small));
109 
110   // Test compilation of cwise() version
111   VERIFY_IS_APPROX(vrand.colwise().stableNorm(),      vrand.colwise().norm());
112   VERIFY_IS_APPROX(vrand.colwise().blueNorm(),        vrand.colwise().norm());
113   VERIFY_IS_APPROX(vrand.colwise().hypotNorm(),       vrand.colwise().norm());
114   VERIFY_IS_APPROX(vrand.rowwise().stableNorm(),      vrand.rowwise().norm());
115   VERIFY_IS_APPROX(vrand.rowwise().blueNorm(),        vrand.rowwise().norm());
116   VERIFY_IS_APPROX(vrand.rowwise().hypotNorm(),       vrand.rowwise().norm());
117 
118   // test NaN, +inf, -inf
119   MatrixType v;
120   Index i = internal::random<Index>(0,rows-1);
121   Index j = internal::random<Index>(0,cols-1);
122 
123   // NaN
124   {
125     v = vrand;
126     v(i,j) = std::numeric_limits<RealScalar>::quiet_NaN();
127     VERIFY(!(numext::isfinite)(v.squaredNorm()));   VERIFY((numext::isnan)(v.squaredNorm()));
128     VERIFY(!(numext::isfinite)(v.norm()));          VERIFY((numext::isnan)(v.norm()));
129     VERIFY(!(numext::isfinite)(v.stableNorm()));    VERIFY((numext::isnan)(v.stableNorm()));
130     VERIFY(!(numext::isfinite)(v.blueNorm()));      VERIFY((numext::isnan)(v.blueNorm()));
131     VERIFY(!(numext::isfinite)(v.hypotNorm()));     VERIFY((numext::isnan)(v.hypotNorm()));
132   }
133 
134   // +inf
135   {
136     v = vrand;
137     v(i,j) = std::numeric_limits<RealScalar>::infinity();
138     VERIFY(!(numext::isfinite)(v.squaredNorm()));   VERIFY(isPlusInf(v.squaredNorm()));
139     VERIFY(!(numext::isfinite)(v.norm()));          VERIFY(isPlusInf(v.norm()));
140     VERIFY(!(numext::isfinite)(v.stableNorm()));
141     if(complex_real_product_ok){
142       VERIFY(isPlusInf(v.stableNorm()));
143     }
144     VERIFY(!(numext::isfinite)(v.blueNorm()));      VERIFY(isPlusInf(v.blueNorm()));
145     VERIFY(!(numext::isfinite)(v.hypotNorm()));     VERIFY(isPlusInf(v.hypotNorm()));
146   }
147 
148   // -inf
149   {
150     v = vrand;
151     v(i,j) = -std::numeric_limits<RealScalar>::infinity();
152     VERIFY(!(numext::isfinite)(v.squaredNorm()));   VERIFY(isPlusInf(v.squaredNorm()));
153     VERIFY(!(numext::isfinite)(v.norm()));          VERIFY(isPlusInf(v.norm()));
154     VERIFY(!(numext::isfinite)(v.stableNorm()));
155     if(complex_real_product_ok) {
156       VERIFY(isPlusInf(v.stableNorm()));
157     }
158     VERIFY(!(numext::isfinite)(v.blueNorm()));      VERIFY(isPlusInf(v.blueNorm()));
159     VERIFY(!(numext::isfinite)(v.hypotNorm()));     VERIFY(isPlusInf(v.hypotNorm()));
160   }
161 
162   // mix
163   {
164     Index i2 = internal::random<Index>(0,rows-1);
165     Index j2 = internal::random<Index>(0,cols-1);
166     v = vrand;
167     v(i,j) = -std::numeric_limits<RealScalar>::infinity();
168     v(i2,j2) = std::numeric_limits<RealScalar>::quiet_NaN();
169     VERIFY(!(numext::isfinite)(v.squaredNorm()));   VERIFY((numext::isnan)(v.squaredNorm()));
170     VERIFY(!(numext::isfinite)(v.norm()));          VERIFY((numext::isnan)(v.norm()));
171     VERIFY(!(numext::isfinite)(v.stableNorm()));    VERIFY((numext::isnan)(v.stableNorm()));
172     VERIFY(!(numext::isfinite)(v.blueNorm()));      VERIFY((numext::isnan)(v.blueNorm()));
173     if (i2 != i || j2 != j) {
174       // hypot propagates inf over NaN.
175       VERIFY(!(numext::isfinite)(v.hypotNorm()));     VERIFY((numext::isinf)(v.hypotNorm()));
176     } else {
177       // inf is overwritten by NaN, expect norm to be NaN.
178       VERIFY(!(numext::isfinite)(v.hypotNorm()));     VERIFY((numext::isnan)(v.hypotNorm()));
179     }
180   }
181 
182   // stableNormalize[d]
183   {
184     VERIFY_IS_APPROX(vrand.stableNormalized(), vrand.normalized());
185     MatrixType vcopy(vrand);
186     vcopy.stableNormalize();
187     VERIFY_IS_APPROX(vcopy, vrand.normalized());
188     VERIFY_IS_APPROX((vrand.stableNormalized()).norm(), RealScalar(1));
189     VERIFY_IS_APPROX(vcopy.norm(), RealScalar(1));
190     VERIFY_IS_APPROX((vbig.stableNormalized()).norm(), RealScalar(1));
191     VERIFY_IS_APPROX((vsmall.stableNormalized()).norm(), RealScalar(1));
192     RealScalar big_scaling = ((std::numeric_limits<RealScalar>::max)() * RealScalar(1e-4));
193     VERIFY_IS_APPROX(vbig/big_scaling, (vbig.stableNorm() * vbig.stableNormalized()).eval()/big_scaling);
194     VERIFY_IS_APPROX(vsmall, vsmall.stableNorm() * vsmall.stableNormalized());
195   }
196 }
197 
198 template<typename Scalar>
test_hypot()199 void test_hypot()
200 {
201   typedef typename NumTraits<Scalar>::Real RealScalar;
202   Scalar factor = internal::random<Scalar>();
203   while(numext::abs2(factor)<RealScalar(1e-4))
204     factor = internal::random<Scalar>();
205   Scalar big = factor * ((std::numeric_limits<RealScalar>::max)() * RealScalar(1e-4));
206 
207   factor = internal::random<Scalar>();
208   while(numext::abs2(factor)<RealScalar(1e-4))
209     factor = internal::random<Scalar>();
210   Scalar small = factor * ((std::numeric_limits<RealScalar>::min)() * RealScalar(1e4));
211 
212   Scalar  one   (1),
213           zero  (0),
214           sqrt2 (std::sqrt(2)),
215           nan   (std::numeric_limits<RealScalar>::quiet_NaN());
216 
217   Scalar a = internal::random<Scalar>(-1,1);
218   Scalar b = internal::random<Scalar>(-1,1);
219   VERIFY_IS_APPROX(numext::hypot(a,b),std::sqrt(numext::abs2(a)+numext::abs2(b)));
220   VERIFY_IS_EQUAL(numext::hypot(zero,zero), zero);
221   VERIFY_IS_APPROX(numext::hypot(one, one), sqrt2);
222   VERIFY_IS_APPROX(numext::hypot(big,big), sqrt2*numext::abs(big));
223   VERIFY_IS_APPROX(numext::hypot(small,small), sqrt2*numext::abs(small));
224   VERIFY_IS_APPROX(numext::hypot(small,big), numext::abs(big));
225   VERIFY((numext::isnan)(numext::hypot(nan,a)));
226   VERIFY((numext::isnan)(numext::hypot(a,nan)));
227 }
228 
EIGEN_DECLARE_TEST(stable_norm)229 EIGEN_DECLARE_TEST(stable_norm)
230 {
231   for(int i = 0; i < g_repeat; i++) {
232     CALL_SUBTEST_3( test_hypot<double>() );
233     CALL_SUBTEST_4( test_hypot<float>() );
234     CALL_SUBTEST_5( test_hypot<std::complex<double> >() );
235     CALL_SUBTEST_6( test_hypot<std::complex<float> >() );
236 
237     CALL_SUBTEST_1( stable_norm(Matrix<float, 1, 1>()) );
238     CALL_SUBTEST_2( stable_norm(Vector4d()) );
239     CALL_SUBTEST_3( stable_norm(VectorXd(internal::random<int>(10,2000))) );
240     CALL_SUBTEST_3( stable_norm(MatrixXd(internal::random<int>(10,200), internal::random<int>(10,200))) );
241     CALL_SUBTEST_4( stable_norm(VectorXf(internal::random<int>(10,2000))) );
242     CALL_SUBTEST_5( stable_norm(VectorXcd(internal::random<int>(10,2000))) );
243     CALL_SUBTEST_6( stable_norm(VectorXcf(internal::random<int>(10,2000))) );
244   }
245 }
246