1// This file is part of Eigen, a lightweight C++ template library
2// for linear algebra.
3//
4// Copyright (C) 2008 Gael Guennebaud <gael.guennebaud@inria.fr>
5// Copyright (C) 2007-2011 Benoit Jacob <jacob.benoit.1@gmail.com>
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#ifndef EIGEN_CORE_H
12#define EIGEN_CORE_H
13
14#define Eigen itkeigen
15
16// first thing Eigen does: stop the compiler from committing suicide
17#include "src/Core/util/DisableStupidWarnings.h"
18
19#if defined(__CUDACC__) && !defined(EIGEN_NO_CUDA)
20  #define EIGEN_CUDACC __CUDACC__
21#endif
22
23#if defined(__CUDA_ARCH__) && !defined(EIGEN_NO_CUDA)
24  #define EIGEN_CUDA_ARCH __CUDA_ARCH__
25#endif
26
27#if defined(__CUDACC_VER_MAJOR__) && (__CUDACC_VER_MAJOR__ >= 9)
28#define EIGEN_CUDACC_VER  ((__CUDACC_VER_MAJOR__ * 10000) + (__CUDACC_VER_MINOR__ * 100))
29#elif defined(__CUDACC_VER__)
30#define EIGEN_CUDACC_VER __CUDACC_VER__
31#else
32#define EIGEN_CUDACC_VER 0
33#endif
34
35// Handle NVCC/CUDA/SYCL
36#if defined(__CUDACC__) || defined(__SYCL_DEVICE_ONLY__)
37  // Do not try asserts on CUDA and SYCL!
38  #ifndef EIGEN_NO_DEBUG
39  #define EIGEN_NO_DEBUG
40  #endif
41
42  #ifdef EIGEN_INTERNAL_DEBUGGING
43  #undef EIGEN_INTERNAL_DEBUGGING
44  #endif
45
46  #ifdef EIGEN_EXCEPTIONS
47  #undef EIGEN_EXCEPTIONS
48  #endif
49
50  // All functions callable from CUDA code must be qualified with __device__
51  #ifdef __CUDACC__
52    // Do not try to vectorize on CUDA and SYCL!
53    #ifndef EIGEN_DONT_VECTORIZE
54    #define EIGEN_DONT_VECTORIZE
55    #endif
56
57    #define EIGEN_DEVICE_FUNC __host__ __device__
58    // We need cuda_runtime.h to ensure that that EIGEN_USING_STD_MATH macro
59    // works properly on the device side
60    #include <cuda_runtime.h>
61  #else
62    #define EIGEN_DEVICE_FUNC
63  #endif
64
65#else
66  #define EIGEN_DEVICE_FUNC
67
68#endif
69
70// When compiling CUDA device code with NVCC, pull in math functions from the
71// global namespace.  In host mode, and when device doee with clang, use the
72// std versions.
73#if defined(__CUDA_ARCH__) && defined(__NVCC__)
74  #define EIGEN_USING_STD_MATH(FUNC) using ::FUNC;
75#else
76  #define EIGEN_USING_STD_MATH(FUNC) using std::FUNC;
77#endif
78
79#if (defined(_CPPUNWIND) || defined(__EXCEPTIONS)) && !defined(__CUDA_ARCH__) && !defined(EIGEN_EXCEPTIONS) && !defined(EIGEN_USE_SYCL)
80  #define EIGEN_EXCEPTIONS
81#endif
82
83#ifdef EIGEN_EXCEPTIONS
84  #include <new>
85#endif
86
87// then include this file where all our macros are defined. It's really important to do it first because
88// it's where we do all the alignment settings (platform detection and honoring the user's will if he
89// defined e.g. EIGEN_DONT_ALIGN) so it needs to be done before we do anything with vectorization.
90#include "src/Core/util/Macros.h"
91
92// Disable the ipa-cp-clone optimization flag with MinGW 6.x or newer (enabled by default with -O3)
93// See http://eigen.tuxfamily.org/bz/show_bug.cgi?id=556 for details.
94#if EIGEN_COMP_MINGW && EIGEN_GNUC_AT_LEAST(4,6)
95  #pragma GCC optimize ("-fno-ipa-cp-clone")
96#endif
97
98#include <complex>
99
100// this include file manages BLAS and MKL related macros
101// and inclusion of their respective header files
102#include "src/Core/util/MKL_support.h"
103
104// if alignment is disabled, then disable vectorization. Note: EIGEN_MAX_ALIGN_BYTES is the proper check, it takes into
105// account both the user's will (EIGEN_MAX_ALIGN_BYTES,EIGEN_DONT_ALIGN) and our own platform checks
106#if EIGEN_MAX_ALIGN_BYTES==0
107  #ifndef EIGEN_DONT_VECTORIZE
108    #define EIGEN_DONT_VECTORIZE
109  #endif
110#endif
111
112#if EIGEN_COMP_MSVC
113  #include <malloc.h> // for _aligned_malloc -- need it regardless of whether vectorization is enabled
114  #if (EIGEN_COMP_MSVC >= 1500) // 2008 or later
115    // Remember that usage of defined() in a #define is undefined by the standard.
116    // a user reported that in 64-bit mode, MSVC doesn't care to define _M_IX86_FP.
117    #if (defined(_M_IX86_FP) && (_M_IX86_FP >= 2)) || EIGEN_ARCH_x86_64
118      #define EIGEN_SSE2_ON_MSVC_2008_OR_LATER
119    #endif
120  #endif
121#else
122  // Remember that usage of defined() in a #define is undefined by the standard
123  #if (defined __SSE2__) && ( (!EIGEN_COMP_GNUC) || EIGEN_COMP_ICC || EIGEN_GNUC_AT_LEAST(4,2) )
124    #define EIGEN_SSE2_ON_NON_MSVC_BUT_NOT_OLD_GCC
125  #endif
126#endif
127
128#ifndef EIGEN_DONT_VECTORIZE
129
130  #if defined (EIGEN_SSE2_ON_NON_MSVC_BUT_NOT_OLD_GCC) || defined(EIGEN_SSE2_ON_MSVC_2008_OR_LATER)
131
132    // Defines symbols for compile-time detection of which instructions are
133    // used.
134    // EIGEN_VECTORIZE_YY is defined if and only if the instruction set YY is used
135    #define EIGEN_VECTORIZE
136    #define EIGEN_VECTORIZE_SSE
137    #define EIGEN_VECTORIZE_SSE2
138
139    // Detect sse3/ssse3/sse4:
140    // gcc and icc defines __SSE3__, ...
141    // there is no way to know about this on msvc. You can define EIGEN_VECTORIZE_SSE* if you
142    // want to force the use of those instructions with msvc.
143    #ifdef __SSE3__
144      #define EIGEN_VECTORIZE_SSE3
145    #endif
146    #ifdef __SSSE3__
147      #define EIGEN_VECTORIZE_SSSE3
148    #endif
149    #ifdef __SSE4_1__
150      #define EIGEN_VECTORIZE_SSE4_1
151    #endif
152    #ifdef __SSE4_2__
153      #define EIGEN_VECTORIZE_SSE4_2
154    #endif
155    #ifdef __AVX__
156      #define EIGEN_VECTORIZE_AVX
157      #define EIGEN_VECTORIZE_SSE3
158      #define EIGEN_VECTORIZE_SSSE3
159      #define EIGEN_VECTORIZE_SSE4_1
160      #define EIGEN_VECTORIZE_SSE4_2
161    #endif
162    #ifdef __AVX2__
163      #define EIGEN_VECTORIZE_AVX2
164    #endif
165    #ifdef __FMA__
166      #define EIGEN_VECTORIZE_FMA
167    #endif
168    #if defined(__AVX512F__) && defined(EIGEN_ENABLE_AVX512)
169      #define EIGEN_VECTORIZE_AVX512
170      #define EIGEN_VECTORIZE_AVX2
171      #define EIGEN_VECTORIZE_AVX
172      #define EIGEN_VECTORIZE_FMA
173      #ifdef __AVX512DQ__
174        #define EIGEN_VECTORIZE_AVX512DQ
175      #endif
176      #ifdef __AVX512ER__
177        #define EIGEN_VECTORIZE_AVX512ER
178      #endif
179    #endif
180
181    // include files
182
183    // This extern "C" works around a MINGW-w64 compilation issue
184    // https://sourceforge.net/tracker/index.php?func=detail&aid=3018394&group_id=202880&atid=983354
185    // In essence, intrin.h is included by windows.h and also declares intrinsics (just as emmintrin.h etc. below do).
186    // However, intrin.h uses an extern "C" declaration, and g++ thus complains of duplicate declarations
187    // with conflicting linkage.  The linkage for intrinsics doesn't matter, but at that stage the compiler doesn't know;
188    // so, to avoid compile errors when windows.h is included after Eigen/Core, ensure intrinsics are extern "C" here too.
189    // notice that since these are C headers, the extern "C" is theoretically needed anyways.
190    extern "C" {
191      // In theory we should only include immintrin.h and not the other *mmintrin.h header files directly.
192      // Doing so triggers some issues with ICC. However old gcc versions seems to not have this file, thus:
193      #if EIGEN_COMP_ICC >= 1110
194        #include <immintrin.h>
195      #else
196        #include <mmintrin.h>
197        #include <emmintrin.h>
198        #include <xmmintrin.h>
199        #ifdef  EIGEN_VECTORIZE_SSE3
200        #include <pmmintrin.h>
201        #endif
202        #ifdef EIGEN_VECTORIZE_SSSE3
203        #include <tmmintrin.h>
204        #endif
205        #ifdef EIGEN_VECTORIZE_SSE4_1
206        #include <smmintrin.h>
207        #endif
208        #ifdef EIGEN_VECTORIZE_SSE4_2
209        #include <nmmintrin.h>
210        #endif
211        #if defined(EIGEN_VECTORIZE_AVX) || defined(EIGEN_VECTORIZE_AVX512)
212        #include <immintrin.h>
213        #endif
214      #endif
215    } // end extern "C"
216  #elif defined __VSX__
217    #define EIGEN_VECTORIZE
218    #define EIGEN_VECTORIZE_VSX
219    #include <altivec.h>
220    // We need to #undef all these ugly tokens defined in <altivec.h>
221    // => use __vector instead of vector
222    #undef bool
223    #undef vector
224    #undef pixel
225  #elif defined __ALTIVEC__
226    #define EIGEN_VECTORIZE
227    #define EIGEN_VECTORIZE_ALTIVEC
228    #include <altivec.h>
229    // We need to #undef all these ugly tokens defined in <altivec.h>
230    // => use __vector instead of vector
231    #undef bool
232    #undef vector
233    #undef pixel
234  #elif (defined  __ARM_NEON) || (defined __ARM_NEON__)
235    #define EIGEN_VECTORIZE
236    #define EIGEN_VECTORIZE_NEON
237    #include <arm_neon.h>
238  #elif (defined __s390x__ && defined __VEC__)
239    #define EIGEN_VECTORIZE
240    #define EIGEN_VECTORIZE_ZVECTOR
241    #include <vecintrin.h>
242  #endif
243#endif
244
245#if defined(__F16C__) && !defined(EIGEN_COMP_CLANG)
246  // We can use the optimized fp16 to float and float to fp16 conversion routines
247  #define EIGEN_HAS_FP16_C
248#endif
249
250#if defined __CUDACC__
251  #define EIGEN_VECTORIZE_CUDA
252  #include <vector_types.h>
253  #if EIGEN_CUDACC_VER >= 70500
254    #define EIGEN_HAS_CUDA_FP16
255  #endif
256#endif
257
258#if defined EIGEN_HAS_CUDA_FP16
259  #include <host_defines.h>
260  #include <cuda_fp16.h>
261#endif
262
263#if (defined _OPENMP) && (!defined EIGEN_DONT_PARALLELIZE)
264  #define EIGEN_HAS_OPENMP
265#endif
266
267#ifdef EIGEN_HAS_OPENMP
268#include <omp.h>
269#endif
270
271// MSVC for windows mobile does not have the errno.h file
272#if !(EIGEN_COMP_MSVC && EIGEN_OS_WINCE) && !EIGEN_COMP_ARM
273#define EIGEN_HAS_ERRNO
274#endif
275
276#ifdef EIGEN_HAS_ERRNO
277#include <cerrno>
278#endif
279#include <cstddef>
280#include <cstdlib>
281#include <cmath>
282#include <cassert>
283#include <functional>
284#include <iosfwd>
285#include <cstring>
286#include <string>
287#include <limits>
288#include <climits> // for CHAR_BIT
289// for min/max:
290#include <algorithm>
291
292// for std::is_nothrow_move_assignable
293#ifdef EIGEN_INCLUDE_TYPE_TRAITS
294#include <type_traits>
295#endif
296
297// for outputting debug info
298#ifdef EIGEN_DEBUG_ASSIGN
299#include <iostream>
300#endif
301
302// required for __cpuid, needs to be included after cmath
303#if EIGEN_COMP_MSVC && EIGEN_ARCH_i386_OR_x86_64 && !EIGEN_OS_WINCE
304  #include <intrin.h>
305#endif
306
307/** \brief Namespace containing all symbols from the %Eigen library. */
308namespace Eigen {
309
310inline static const char *SimdInstructionSetsInUse(void) {
311#if defined(EIGEN_VECTORIZE_AVX512)
312  return "AVX512, FMA, AVX2, AVX, SSE, SSE2, SSE3, SSSE3, SSE4.1, SSE4.2";
313#elif defined(EIGEN_VECTORIZE_AVX)
314  return "AVX SSE, SSE2, SSE3, SSSE3, SSE4.1, SSE4.2";
315#elif defined(EIGEN_VECTORIZE_SSE4_2)
316  return "SSE, SSE2, SSE3, SSSE3, SSE4.1, SSE4.2";
317#elif defined(EIGEN_VECTORIZE_SSE4_1)
318  return "SSE, SSE2, SSE3, SSSE3, SSE4.1";
319#elif defined(EIGEN_VECTORIZE_SSSE3)
320  return "SSE, SSE2, SSE3, SSSE3";
321#elif defined(EIGEN_VECTORIZE_SSE3)
322  return "SSE, SSE2, SSE3";
323#elif defined(EIGEN_VECTORIZE_SSE2)
324  return "SSE, SSE2";
325#elif defined(EIGEN_VECTORIZE_ALTIVEC)
326  return "AltiVec";
327#elif defined(EIGEN_VECTORIZE_VSX)
328  return "VSX";
329#elif defined(EIGEN_VECTORIZE_NEON)
330  return "ARM NEON";
331#elif defined(EIGEN_VECTORIZE_ZVECTOR)
332  return "S390X ZVECTOR";
333#else
334  return "None";
335#endif
336}
337
338} // end namespace Eigen
339
340#if defined EIGEN2_SUPPORT_STAGE40_FULL_EIGEN3_STRICTNESS || defined EIGEN2_SUPPORT_STAGE30_FULL_EIGEN3_API || defined EIGEN2_SUPPORT_STAGE20_RESOLVE_API_CONFLICTS || defined EIGEN2_SUPPORT_STAGE10_FULL_EIGEN2_API || defined EIGEN2_SUPPORT
341// This will generate an error message:
342#error Eigen2-support is only available up to version 3.2. Please go to "http://eigen.tuxfamily.org/index.php?title=Eigen2" for further information
343#endif
344
345namespace Eigen {
346
347// we use size_t frequently and we'll never remember to prepend it with std:: everytime just to
348// ensure QNX/QCC support
349using std::size_t;
350// gcc 4.6.0 wants std:: for ptrdiff_t
351using std::ptrdiff_t;
352
353}
354
355/** \defgroup Core_Module Core module
356  * This is the main module of Eigen providing dense matrix and vector support
357  * (both fixed and dynamic size) with all the features corresponding to a BLAS library
358  * and much more...
359  *
360  * \code
361  * #include <Eigen/Core>
362  * \endcode
363  */
364
365#include "src/Core/util/Constants.h"
366#include "src/Core/util/Meta.h"
367#include "src/Core/util/ForwardDeclarations.h"
368#include "src/Core/util/StaticAssert.h"
369#include "src/Core/util/XprHelper.h"
370#include "src/Core/util/Memory.h"
371
372#include "src/Core/NumTraits.h"
373#include "src/Core/MathFunctions.h"
374#include "src/Core/GenericPacketMath.h"
375#include "src/Core/MathFunctionsImpl.h"
376#include "src/Core/arch/Default/ConjHelper.h"
377
378#if defined EIGEN_VECTORIZE_AVX512
379  #include "src/Core/arch/SSE/PacketMath.h"
380  #include "src/Core/arch/AVX/PacketMath.h"
381  #include "src/Core/arch/AVX512/PacketMath.h"
382  #include "src/Core/arch/AVX512/MathFunctions.h"
383#elif defined EIGEN_VECTORIZE_AVX
384  // Use AVX for floats and doubles, SSE for integers
385  #include "src/Core/arch/SSE/PacketMath.h"
386  #include "src/Core/arch/SSE/Complex.h"
387  #include "src/Core/arch/SSE/MathFunctions.h"
388  #include "src/Core/arch/AVX/PacketMath.h"
389  #include "src/Core/arch/AVX/MathFunctions.h"
390  #include "src/Core/arch/AVX/Complex.h"
391  #include "src/Core/arch/AVX/TypeCasting.h"
392  #include "src/Core/arch/SSE/TypeCasting.h"
393#elif defined EIGEN_VECTORIZE_SSE
394  #include "src/Core/arch/SSE/PacketMath.h"
395  #include "src/Core/arch/SSE/MathFunctions.h"
396  #include "src/Core/arch/SSE/Complex.h"
397  #include "src/Core/arch/SSE/TypeCasting.h"
398#elif defined(EIGEN_VECTORIZE_ALTIVEC) || defined(EIGEN_VECTORIZE_VSX)
399  #include "src/Core/arch/AltiVec/PacketMath.h"
400  #include "src/Core/arch/AltiVec/MathFunctions.h"
401  #include "src/Core/arch/AltiVec/Complex.h"
402#elif defined EIGEN_VECTORIZE_NEON
403  #include "src/Core/arch/NEON/PacketMath.h"
404  #include "src/Core/arch/NEON/MathFunctions.h"
405  #include "src/Core/arch/NEON/Complex.h"
406#elif defined EIGEN_VECTORIZE_ZVECTOR
407  #include "src/Core/arch/ZVector/PacketMath.h"
408  #include "src/Core/arch/ZVector/MathFunctions.h"
409  #include "src/Core/arch/ZVector/Complex.h"
410#endif
411
412// Half float support
413#include "src/Core/arch/CUDA/Half.h"
414#include "src/Core/arch/CUDA/PacketMathHalf.h"
415#include "src/Core/arch/CUDA/TypeCasting.h"
416
417#if defined EIGEN_VECTORIZE_CUDA
418  #include "src/Core/arch/CUDA/PacketMath.h"
419  #include "src/Core/arch/CUDA/MathFunctions.h"
420#endif
421
422#include "src/Core/arch/Default/Settings.h"
423
424#include "src/Core/functors/TernaryFunctors.h"
425#include "src/Core/functors/BinaryFunctors.h"
426#include "src/Core/functors/UnaryFunctors.h"
427#include "src/Core/functors/NullaryFunctors.h"
428#include "src/Core/functors/StlFunctors.h"
429#include "src/Core/functors/AssignmentFunctors.h"
430
431// Specialized functors to enable the processing of complex numbers
432// on CUDA devices
433#include "src/Core/arch/CUDA/Complex.h"
434
435#include "src/Core/IO.h"
436#include "src/Core/DenseCoeffsBase.h"
437#include "src/Core/DenseBase.h"
438#include "src/Core/MatrixBase.h"
439#include "src/Core/EigenBase.h"
440
441#include "src/Core/Product.h"
442#include "src/Core/CoreEvaluators.h"
443#include "src/Core/AssignEvaluator.h"
444
445#ifndef EIGEN_PARSED_BY_DOXYGEN // work around Doxygen bug triggered by Assign.h r814874
446                                // at least confirmed with Doxygen 1.5.5 and 1.5.6
447  #include "src/Core/Assign.h"
448#endif
449
450#include "src/Core/ArrayBase.h"
451#include "src/Core/util/BlasUtil.h"
452#include "src/Core/DenseStorage.h"
453#include "src/Core/NestByValue.h"
454
455// #include "src/Core/ForceAlignedAccess.h"
456
457#include "src/Core/ReturnByValue.h"
458#include "src/Core/NoAlias.h"
459#include "src/Core/PlainObjectBase.h"
460#include "src/Core/Matrix.h"
461#include "src/Core/Array.h"
462#include "src/Core/CwiseTernaryOp.h"
463#include "src/Core/CwiseBinaryOp.h"
464#include "src/Core/CwiseUnaryOp.h"
465#include "src/Core/CwiseNullaryOp.h"
466#include "src/Core/CwiseUnaryView.h"
467#include "src/Core/SelfCwiseBinaryOp.h"
468#include "src/Core/Dot.h"
469#include "src/Core/StableNorm.h"
470#include "src/Core/Stride.h"
471#include "src/Core/MapBase.h"
472#include "src/Core/Map.h"
473#include "src/Core/Ref.h"
474#include "src/Core/Block.h"
475#include "src/Core/VectorBlock.h"
476#include "src/Core/Transpose.h"
477#include "src/Core/DiagonalMatrix.h"
478#include "src/Core/Diagonal.h"
479#include "src/Core/DiagonalProduct.h"
480#include "src/Core/Redux.h"
481#include "src/Core/Visitor.h"
482#include "src/Core/Fuzzy.h"
483#include "src/Core/Swap.h"
484#include "src/Core/CommaInitializer.h"
485#include "src/Core/GeneralProduct.h"
486#include "src/Core/Solve.h"
487#include "src/Core/Inverse.h"
488#include "src/Core/SolverBase.h"
489#include "src/Core/PermutationMatrix.h"
490#include "src/Core/Transpositions.h"
491#include "src/Core/TriangularMatrix.h"
492#include "src/Core/SelfAdjointView.h"
493#include "src/Core/products/GeneralBlockPanelKernel.h"
494#include "src/Core/products/Parallelizer.h"
495#include "src/Core/ProductEvaluators.h"
496#include "src/Core/products/GeneralMatrixVector.h"
497#include "src/Core/products/GeneralMatrixMatrix.h"
498#include "src/Core/SolveTriangular.h"
499#include "src/Core/products/GeneralMatrixMatrixTriangular.h"
500#include "src/Core/products/SelfadjointMatrixVector.h"
501#include "src/Core/products/SelfadjointMatrixMatrix.h"
502#include "src/Core/products/SelfadjointProduct.h"
503#include "src/Core/products/SelfadjointRank2Update.h"
504#include "src/Core/products/TriangularMatrixVector.h"
505#include "src/Core/products/TriangularMatrixMatrix.h"
506#include "src/Core/products/TriangularSolverMatrix.h"
507#include "src/Core/products/TriangularSolverVector.h"
508#include "src/Core/BandMatrix.h"
509#include "src/Core/CoreIterators.h"
510#include "src/Core/ConditionEstimator.h"
511
512#include "src/Core/BooleanRedux.h"
513#include "src/Core/Select.h"
514#include "src/Core/VectorwiseOp.h"
515#include "src/Core/Random.h"
516#include "src/Core/Replicate.h"
517#include "src/Core/Reverse.h"
518#include "src/Core/ArrayWrapper.h"
519
520#ifdef EIGEN_USE_BLAS
521#include "src/Core/products/GeneralMatrixMatrix_BLAS.h"
522#include "src/Core/products/GeneralMatrixVector_BLAS.h"
523#include "src/Core/products/GeneralMatrixMatrixTriangular_BLAS.h"
524#include "src/Core/products/SelfadjointMatrixMatrix_BLAS.h"
525#include "src/Core/products/SelfadjointMatrixVector_BLAS.h"
526#include "src/Core/products/TriangularMatrixMatrix_BLAS.h"
527#include "src/Core/products/TriangularMatrixVector_BLAS.h"
528#include "src/Core/products/TriangularSolverMatrix_BLAS.h"
529#endif // EIGEN_USE_BLAS
530
531#ifdef EIGEN_USE_MKL_VML
532#include "src/Core/Assign_MKL.h"
533#endif
534
535#include "src/Core/GlobalFunctions.h"
536
537#include "src/Core/util/ReenableStupidWarnings.h"
538
539#endif // EIGEN_CORE_H
540