1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
|
// This file is part of Eigen, a lightweight C++ template library
// for linear algebra.
//
// Copyright (C) 2008 Benoit Jacob <jacob.benoit.1@gmail.com>
// Copyright (C) 2015 Gael Guennebaud <gael.guennebaud@inria.fr>
//
// This Source Code Form is subject to the terms of the Mozilla
// Public License v. 2.0. If a copy of the MPL was not distributed
// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
#if defined(EIGEN_TEST_PART_1)
// default
#elif defined(EIGEN_TEST_PART_2)
#define EIGEN_MAX_STATIC_ALIGN_BYTES 16
#define EIGEN_MAX_ALIGN_BYTES 16
#elif defined(EIGEN_TEST_PART_3)
#define EIGEN_MAX_STATIC_ALIGN_BYTES 32
#define EIGEN_MAX_ALIGN_BYTES 32
#elif defined(EIGEN_TEST_PART_4)
#define EIGEN_MAX_STATIC_ALIGN_BYTES 64
#define EIGEN_MAX_ALIGN_BYTES 64
#endif
#include "main.h"
typedef Matrix<float, 6,1> Vector6f;
typedef Matrix<float, 8,1> Vector8f;
typedef Matrix<float, 12,1> Vector12f;
typedef Matrix<double, 5,1> Vector5d;
typedef Matrix<double, 6,1> Vector6d;
typedef Matrix<double, 7,1> Vector7d;
typedef Matrix<double, 8,1> Vector8d;
typedef Matrix<double, 9,1> Vector9d;
typedef Matrix<double,10,1> Vector10d;
typedef Matrix<double,12,1> Vector12d;
struct TestNew1
{
MatrixXd m; // good: m will allocate its own array, taking care of alignment.
TestNew1() : m(20,20) {}
};
struct TestNew2
{
Matrix3d m; // good: m's size isn't a multiple of 16 bytes, so m doesn't have to be 16-byte aligned,
// 8-byte alignment is good enough here, which we'll get automatically
};
struct TestNew3
{
Vector2f m; // good: m's size isn't a multiple of 16 bytes, so m doesn't have to be 16-byte aligned
};
struct TestNew4
{
EIGEN_MAKE_ALIGNED_OPERATOR_NEW
Vector2d m;
float f; // make the struct have sizeof%16!=0 to make it a little more tricky when we allow an array of 2 such objects
};
struct TestNew5
{
EIGEN_MAKE_ALIGNED_OPERATOR_NEW
float f; // try the f at first -- the EIGEN_ALIGN_MAX attribute of m should make that still work
Matrix4f m;
};
struct TestNew6
{
Matrix<float,2,2,DontAlign> m; // good: no alignment requested
float f;
};
template<bool Align> struct Depends
{
EIGEN_MAKE_ALIGNED_OPERATOR_NEW_IF(Align)
Vector2d m;
float f;
};
template<typename T>
void check_unalignedassert_good()
{
T *x, *y;
x = new T;
delete x;
y = new T[2];
delete[] y;
}
#if EIGEN_MAX_STATIC_ALIGN_BYTES>0
template<typename T>
void construct_at_boundary(int boundary)
{
char buf[sizeof(T)+256];
size_t _buf = reinterpret_cast<internal::UIntPtr>(buf);
_buf += (EIGEN_MAX_ALIGN_BYTES - (_buf % EIGEN_MAX_ALIGN_BYTES)); // make 16/32/...-byte aligned
_buf += boundary; // make exact boundary-aligned
T *x = ::new(reinterpret_cast<void*>(_buf)) T;
x[0].setZero(); // just in order to silence warnings
x->~T();
}
#endif
void unalignedassert()
{
#if EIGEN_MAX_STATIC_ALIGN_BYTES>0
construct_at_boundary<Vector2f>(4);
construct_at_boundary<Vector3f>(4);
construct_at_boundary<Vector4f>(16);
construct_at_boundary<Vector6f>(4);
construct_at_boundary<Vector8f>(EIGEN_MAX_ALIGN_BYTES);
construct_at_boundary<Vector12f>(16);
construct_at_boundary<Matrix2f>(16);
construct_at_boundary<Matrix3f>(4);
construct_at_boundary<Matrix4f>(EIGEN_MAX_ALIGN_BYTES);
construct_at_boundary<Vector2d>(16);
construct_at_boundary<Vector3d>(4);
construct_at_boundary<Vector4d>(EIGEN_MAX_ALIGN_BYTES);
construct_at_boundary<Vector5d>(4);
construct_at_boundary<Vector6d>(16);
construct_at_boundary<Vector7d>(4);
construct_at_boundary<Vector8d>(EIGEN_MAX_ALIGN_BYTES);
construct_at_boundary<Vector9d>(4);
construct_at_boundary<Vector10d>(16);
construct_at_boundary<Vector12d>(EIGEN_MAX_ALIGN_BYTES);
construct_at_boundary<Matrix2d>(EIGEN_MAX_ALIGN_BYTES);
construct_at_boundary<Matrix3d>(4);
construct_at_boundary<Matrix4d>(EIGEN_MAX_ALIGN_BYTES);
construct_at_boundary<Vector2cf>(16);
construct_at_boundary<Vector3cf>(4);
construct_at_boundary<Vector2cd>(EIGEN_MAX_ALIGN_BYTES);
construct_at_boundary<Vector3cd>(16);
#endif
check_unalignedassert_good<TestNew1>();
check_unalignedassert_good<TestNew2>();
check_unalignedassert_good<TestNew3>();
check_unalignedassert_good<TestNew4>();
check_unalignedassert_good<TestNew5>();
check_unalignedassert_good<TestNew6>();
check_unalignedassert_good<Depends<true> >();
#if EIGEN_MAX_STATIC_ALIGN_BYTES>0
if(EIGEN_MAX_ALIGN_BYTES>=16)
{
VERIFY_RAISES_ASSERT(construct_at_boundary<Vector4f>(8));
VERIFY_RAISES_ASSERT(construct_at_boundary<Vector8f>(8));
VERIFY_RAISES_ASSERT(construct_at_boundary<Vector12f>(8));
VERIFY_RAISES_ASSERT(construct_at_boundary<Vector2d>(8));
VERIFY_RAISES_ASSERT(construct_at_boundary<Vector4d>(8));
VERIFY_RAISES_ASSERT(construct_at_boundary<Vector6d>(8));
VERIFY_RAISES_ASSERT(construct_at_boundary<Vector8d>(8));
VERIFY_RAISES_ASSERT(construct_at_boundary<Vector10d>(8));
VERIFY_RAISES_ASSERT(construct_at_boundary<Vector12d>(8));
// Complexes are disabled because the compiler might aggressively vectorize
// the initialization of complex coeffs to 0 before we can check for alignedness
//VERIFY_RAISES_ASSERT(construct_at_boundary<Vector2cf>(8));
VERIFY_RAISES_ASSERT(construct_at_boundary<Vector4i>(8));
}
for(int b=8; b<EIGEN_MAX_ALIGN_BYTES; b+=8)
{
if(b<32) VERIFY_RAISES_ASSERT(construct_at_boundary<Vector8f>(b));
if(b<64) VERIFY_RAISES_ASSERT(construct_at_boundary<Matrix4f>(b));
if(b<32) VERIFY_RAISES_ASSERT(construct_at_boundary<Vector4d>(b));
if(b<32) VERIFY_RAISES_ASSERT(construct_at_boundary<Matrix2d>(b));
if(b<128) VERIFY_RAISES_ASSERT(construct_at_boundary<Matrix4d>(b));
//if(b<32) VERIFY_RAISES_ASSERT(construct_at_boundary<Vector2cd>(b));
}
#endif
}
EIGEN_DECLARE_TEST(unalignedassert)
{
CALL_SUBTEST(unalignedassert());
}
|
