diff options
Diffstat (limited to 'Eigen/Core')
| -rw-r--r-- | Eigen/Core | 290 |
1 files changed, 215 insertions, 75 deletions
diff --git a/Eigen/Core b/Eigen/Core index 509c529e1..0f7fa630d 100644 --- a/Eigen/Core +++ b/Eigen/Core @@ -14,6 +14,58 @@ // first thing Eigen does: stop the compiler from committing suicide #include "src/Core/util/DisableStupidWarnings.h" +// Handle NVCC/CUDA/SYCL +#if defined(__CUDACC__) || defined(__SYCL_DEVICE_ONLY__) + // Do not try asserts on CUDA and SYCL! + #ifndef EIGEN_NO_DEBUG + #define EIGEN_NO_DEBUG + #endif + + #ifdef EIGEN_INTERNAL_DEBUGGING + #undef EIGEN_INTERNAL_DEBUGGING + #endif + + #ifdef EIGEN_EXCEPTIONS + #undef EIGEN_EXCEPTIONS + #endif + + // All functions callable from CUDA code must be qualified with __device__ + #ifdef __CUDACC__ + // Do not try to vectorize on CUDA and SYCL! + #ifndef EIGEN_DONT_VECTORIZE + #define EIGEN_DONT_VECTORIZE + #endif + + #define EIGEN_DEVICE_FUNC __host__ __device__ + // We need math_functions.hpp to ensure that that EIGEN_USING_STD_MATH macro + // works properly on the device side + #include <math_functions.hpp> + #else + #define EIGEN_DEVICE_FUNC + #endif + +#else + #define EIGEN_DEVICE_FUNC + +#endif + +// When compiling CUDA device code with NVCC, pull in math functions from the +// global namespace. In host mode, and when device doee with clang, use the +// std versions. +#if defined(__CUDA_ARCH__) && defined(__NVCC__) + #define EIGEN_USING_STD_MATH(FUNC) using ::FUNC; +#else + #define EIGEN_USING_STD_MATH(FUNC) using std::FUNC; +#endif + +#if (defined(_CPPUNWIND) || defined(__EXCEPTIONS)) && !defined(__CUDA_ARCH__) && !defined(EIGEN_EXCEPTIONS) && !defined(EIGEN_USE_SYCL) + #define EIGEN_EXCEPTIONS +#endif + +#ifdef EIGEN_EXCEPTIONS + #include <new> +#endif + // then include this file where all our macros are defined. It's really important to do it first because // it's where we do all the alignment settings (platform detection and honoring the user's will if he // defined e.g. EIGEN_DONT_ALIGN) so it needs to be done before we do anything with vectorization. @@ -21,7 +73,7 @@ // Disable the ipa-cp-clone optimization flag with MinGW 6.x or newer (enabled by default with -O3) // See http://eigen.tuxfamily.org/bz/show_bug.cgi?id=556 for details. -#if defined(__MINGW32__) && EIGEN_GNUC_AT_LEAST(4,6) +#if EIGEN_COMP_MINGW && EIGEN_GNUC_AT_LEAST(4,6) #pragma GCC optimize ("-fno-ipa-cp-clone") #endif @@ -31,26 +83,26 @@ // and inclusion of their respective header files #include "src/Core/util/MKL_support.h" -// if alignment is disabled, then disable vectorization. Note: EIGEN_ALIGN is the proper check, it takes into -// account both the user's will (EIGEN_DONT_ALIGN) and our own platform checks -#if !EIGEN_ALIGN +// if alignment is disabled, then disable vectorization. Note: EIGEN_MAX_ALIGN_BYTES is the proper check, it takes into +// account both the user's will (EIGEN_MAX_ALIGN_BYTES,EIGEN_DONT_ALIGN) and our own platform checks +#if EIGEN_MAX_ALIGN_BYTES==0 #ifndef EIGEN_DONT_VECTORIZE #define EIGEN_DONT_VECTORIZE #endif #endif -#ifdef _MSC_VER +#if EIGEN_COMP_MSVC #include <malloc.h> // for _aligned_malloc -- need it regardless of whether vectorization is enabled - #if (_MSC_VER >= 1500) // 2008 or later + #if (EIGEN_COMP_MSVC >= 1500) // 2008 or later // Remember that usage of defined() in a #define is undefined by the standard. // a user reported that in 64-bit mode, MSVC doesn't care to define _M_IX86_FP. - #if (defined(_M_IX86_FP) && (_M_IX86_FP >= 2)) || defined(_M_X64) + #if (defined(_M_IX86_FP) && (_M_IX86_FP >= 2)) || EIGEN_ARCH_x86_64 #define EIGEN_SSE2_ON_MSVC_2008_OR_LATER #endif #endif #else // Remember that usage of defined() in a #define is undefined by the standard - #if (defined __SSE2__) && ( (!defined __GNUC__) || (defined __INTEL_COMPILER) || EIGEN_GNUC_AT_LEAST(4,2) ) + #if (defined __SSE2__) && ( (!EIGEN_COMP_GNUC) || EIGEN_COMP_ICC || EIGEN_GNUC_AT_LEAST(4,2) ) #define EIGEN_SSE2_ON_NON_MSVC_BUT_NOT_OLD_GCC #endif #endif @@ -82,6 +134,28 @@ #ifdef __SSE4_2__ #define EIGEN_VECTORIZE_SSE4_2 #endif + #ifdef __AVX__ + #define EIGEN_VECTORIZE_AVX + #define EIGEN_VECTORIZE_SSE3 + #define EIGEN_VECTORIZE_SSSE3 + #define EIGEN_VECTORIZE_SSE4_1 + #define EIGEN_VECTORIZE_SSE4_2 + #endif + #ifdef __AVX2__ + #define EIGEN_VECTORIZE_AVX2 + #endif + #ifdef __FMA__ + #define EIGEN_VECTORIZE_FMA + #endif + #if defined(__AVX512F__) && defined(EIGEN_ENABLE_AVX512) + #define EIGEN_VECTORIZE_AVX512 + #define EIGEN_VECTORIZE_AVX2 + #define EIGEN_VECTORIZE_AVX + #define EIGEN_VECTORIZE_FMA + #ifdef __AVX512DQ__ + #define EIGEN_VECTORIZE_AVX512DQ + #endif + #endif // include files @@ -95,9 +169,10 @@ extern "C" { // In theory we should only include immintrin.h and not the other *mmintrin.h header files directly. // Doing so triggers some issues with ICC. However old gcc versions seems to not have this file, thus: - #if defined(__INTEL_COMPILER) && __INTEL_COMPILER >= 1110 + #if EIGEN_COMP_ICC >= 1110 #include <immintrin.h> #else + #include <mmintrin.h> #include <emmintrin.h> #include <xmmintrin.h> #ifdef EIGEN_VECTORIZE_SSE3 @@ -112,8 +187,20 @@ #ifdef EIGEN_VECTORIZE_SSE4_2 #include <nmmintrin.h> #endif + #if defined(EIGEN_VECTORIZE_AVX) || defined(EIGEN_VECTORIZE_AVX512) + #include <immintrin.h> + #endif #endif } // end extern "C" + #elif defined __VSX__ + #define EIGEN_VECTORIZE + #define EIGEN_VECTORIZE_VSX + #include <altivec.h> + // We need to #undef all these ugly tokens defined in <altivec.h> + // => use __vector instead of vector + #undef bool + #undef vector + #undef pixel #elif defined __ALTIVEC__ #define EIGEN_VECTORIZE #define EIGEN_VECTORIZE_ALTIVEC @@ -123,13 +210,35 @@ #undef bool #undef vector #undef pixel - #elif defined __ARM_NEON + #elif (defined __ARM_NEON) || (defined __ARM_NEON__) #define EIGEN_VECTORIZE #define EIGEN_VECTORIZE_NEON #include <arm_neon.h> + #elif (defined __s390x__ && defined __VEC__) + #define EIGEN_VECTORIZE + #define EIGEN_VECTORIZE_ZVECTOR + #include <vecintrin.h> #endif #endif +#if defined(__F16C__) && !defined(EIGEN_COMP_CLANG) + // We can use the optimized fp16 to float and float to fp16 conversion routines + #define EIGEN_HAS_FP16_C +#endif + +#if defined __CUDACC__ + #define EIGEN_VECTORIZE_CUDA + #include <vector_types.h> + #if defined __CUDACC_VER__ && __CUDACC_VER__ >= 70500 + #define EIGEN_HAS_CUDA_FP16 + #endif +#endif + +#if defined EIGEN_HAS_CUDA_FP16 + #include <host_defines.h> + #include <cuda_fp16.h> +#endif + #if (defined _OPENMP) && (!defined EIGEN_DONT_PARALLELIZE) #define EIGEN_HAS_OPENMP #endif @@ -139,7 +248,7 @@ #endif // MSVC for windows mobile does not have the errno.h file -#if !(defined(_MSC_VER) && defined(_WIN32_WCE)) && !defined(__ARMCC_VERSION) +#if !(EIGEN_COMP_MSVC && EIGEN_OS_WINCE) && !EIGEN_COMP_ARM #define EIGEN_HAS_ERRNO #endif @@ -159,29 +268,30 @@ // for min/max: #include <algorithm> +// for std::is_nothrow_move_assignable +#ifdef EIGEN_INCLUDE_TYPE_TRAITS +#include <type_traits> +#endif + // for outputting debug info #ifdef EIGEN_DEBUG_ASSIGN #include <iostream> #endif // required for __cpuid, needs to be included after cmath -#if defined(_MSC_VER) && (defined(_M_IX86)||defined(_M_X64)) && (!defined(_WIN32_WCE)) +#if EIGEN_COMP_MSVC && EIGEN_ARCH_i386_OR_x86_64 && !EIGEN_OS_WINCE #include <intrin.h> #endif -#if defined(_CPPUNWIND) || defined(__EXCEPTIONS) - #define EIGEN_EXCEPTIONS -#endif - -#ifdef EIGEN_EXCEPTIONS - #include <new> -#endif - /** \brief Namespace containing all symbols from the %Eigen library. */ namespace Eigen { inline static const char *SimdInstructionSetsInUse(void) { -#if defined(EIGEN_VECTORIZE_SSE4_2) +#if defined(EIGEN_VECTORIZE_AVX512) + return "AVX512, FMA, AVX2, AVX, SSE, SSE2, SSE3, SSSE3, SSE4.1, SSE4.2"; +#elif defined(EIGEN_VECTORIZE_AVX) + return "AVX SSE, SSE2, SSE3, SSSE3, SSE4.1, SSE4.2"; +#elif defined(EIGEN_VECTORIZE_SSE4_2) return "SSE, SSE2, SSE3, SSSE3, SSE4.1, SSE4.2"; #elif defined(EIGEN_VECTORIZE_SSE4_1) return "SSE, SSE2, SSE3, SSSE3, SSE4.1"; @@ -193,8 +303,12 @@ inline static const char *SimdInstructionSetsInUse(void) { return "SSE, SSE2"; #elif defined(EIGEN_VECTORIZE_ALTIVEC) return "AltiVec"; +#elif defined(EIGEN_VECTORIZE_VSX) + return "VSX"; #elif defined(EIGEN_VECTORIZE_NEON) return "ARM NEON"; +#elif defined(EIGEN_VECTORIZE_ZVECTOR) + return "S390X ZVECTOR"; #else return "None"; #endif @@ -202,42 +316,21 @@ inline static const char *SimdInstructionSetsInUse(void) { } // end namespace Eigen -#define STAGE10_FULL_EIGEN2_API 10 -#define STAGE20_RESOLVE_API_CONFLICTS 20 -#define STAGE30_FULL_EIGEN3_API 30 -#define STAGE40_FULL_EIGEN3_STRICTNESS 40 -#define STAGE99_NO_EIGEN2_SUPPORT 99 - -#if defined EIGEN2_SUPPORT_STAGE40_FULL_EIGEN3_STRICTNESS - #define EIGEN2_SUPPORT - #define EIGEN2_SUPPORT_STAGE STAGE40_FULL_EIGEN3_STRICTNESS -#elif defined EIGEN2_SUPPORT_STAGE30_FULL_EIGEN3_API - #define EIGEN2_SUPPORT - #define EIGEN2_SUPPORT_STAGE STAGE30_FULL_EIGEN3_API -#elif defined EIGEN2_SUPPORT_STAGE20_RESOLVE_API_CONFLICTS - #define EIGEN2_SUPPORT - #define EIGEN2_SUPPORT_STAGE STAGE20_RESOLVE_API_CONFLICTS -#elif defined EIGEN2_SUPPORT_STAGE10_FULL_EIGEN2_API - #define EIGEN2_SUPPORT - #define EIGEN2_SUPPORT_STAGE STAGE10_FULL_EIGEN2_API -#elif defined EIGEN2_SUPPORT - // default to stage 3, that's what it's always meant - #define EIGEN2_SUPPORT_STAGE30_FULL_EIGEN3_API - #define EIGEN2_SUPPORT_STAGE STAGE30_FULL_EIGEN3_API -#else - #define EIGEN2_SUPPORT_STAGE STAGE99_NO_EIGEN2_SUPPORT +#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 +// This will generate an error message: +#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 #endif -#ifdef EIGEN2_SUPPORT -#undef minor -#endif +namespace Eigen { // we use size_t frequently and we'll never remember to prepend it with std:: everytime just to // ensure QNX/QCC support using std::size_t; -// gcc 4.6.0 wants std:: for ptrdiff_t +// gcc 4.6.0 wants std:: for ptrdiff_t using std::ptrdiff_t; +} + /** \defgroup Core_Module Core module * This is the main module of Eigen providing dense matrix and vector support * (both fixed and dynamic size) with all the features corresponding to a BLAS library @@ -249,8 +342,8 @@ using std::ptrdiff_t; */ #include "src/Core/util/Constants.h" -#include "src/Core/util/ForwardDeclarations.h" #include "src/Core/util/Meta.h" +#include "src/Core/util/ForwardDeclarations.h" #include "src/Core/util/StaticAssert.h" #include "src/Core/util/XprHelper.h" #include "src/Core/util/Memory.h" @@ -258,41 +351,92 @@ using std::ptrdiff_t; #include "src/Core/NumTraits.h" #include "src/Core/MathFunctions.h" #include "src/Core/GenericPacketMath.h" +#include "src/Core/MathFunctionsImpl.h" -#if defined EIGEN_VECTORIZE_SSE +#if defined EIGEN_VECTORIZE_AVX512 + #include "src/Core/arch/SSE/PacketMath.h" + #include "src/Core/arch/AVX/PacketMath.h" + #include "src/Core/arch/AVX512/PacketMath.h" + #include "src/Core/arch/AVX512/MathFunctions.h" +#elif defined EIGEN_VECTORIZE_AVX + // Use AVX for floats and doubles, SSE for integers + #include "src/Core/arch/SSE/PacketMath.h" + #include "src/Core/arch/SSE/Complex.h" + #include "src/Core/arch/SSE/MathFunctions.h" + #include "src/Core/arch/AVX/PacketMath.h" + #include "src/Core/arch/AVX/MathFunctions.h" + #include "src/Core/arch/AVX/Complex.h" + #include "src/Core/arch/AVX/TypeCasting.h" +#elif defined EIGEN_VECTORIZE_SSE #include "src/Core/arch/SSE/PacketMath.h" #include "src/Core/arch/SSE/MathFunctions.h" #include "src/Core/arch/SSE/Complex.h" -#elif defined EIGEN_VECTORIZE_ALTIVEC + #include "src/Core/arch/SSE/TypeCasting.h" +#elif defined(EIGEN_VECTORIZE_ALTIVEC) || defined(EIGEN_VECTORIZE_VSX) #include "src/Core/arch/AltiVec/PacketMath.h" + #include "src/Core/arch/AltiVec/MathFunctions.h" #include "src/Core/arch/AltiVec/Complex.h" #elif defined EIGEN_VECTORIZE_NEON #include "src/Core/arch/NEON/PacketMath.h" + #include "src/Core/arch/NEON/MathFunctions.h" #include "src/Core/arch/NEON/Complex.h" +#elif defined EIGEN_VECTORIZE_ZVECTOR + #include "src/Core/arch/ZVector/PacketMath.h" + #include "src/Core/arch/ZVector/MathFunctions.h" + #include "src/Core/arch/ZVector/Complex.h" +#endif + +// Half float support +#include "src/Core/arch/CUDA/Half.h" +#include "src/Core/arch/CUDA/PacketMathHalf.h" +#include "src/Core/arch/CUDA/TypeCasting.h" + +#if defined EIGEN_VECTORIZE_CUDA + #include "src/Core/arch/CUDA/PacketMath.h" + #include "src/Core/arch/CUDA/MathFunctions.h" #endif #include "src/Core/arch/Default/Settings.h" -#include "src/Core/Functors.h" +#include "src/Core/functors/TernaryFunctors.h" +#include "src/Core/functors/BinaryFunctors.h" +#include "src/Core/functors/UnaryFunctors.h" +#include "src/Core/functors/NullaryFunctors.h" +#include "src/Core/functors/StlFunctors.h" +#include "src/Core/functors/AssignmentFunctors.h" + +// Specialized functors to enable the processing of complex numbers +// on CUDA devices +#include "src/Core/arch/CUDA/Complex.h" + +#include "src/Core/IO.h" #include "src/Core/DenseCoeffsBase.h" #include "src/Core/DenseBase.h" #include "src/Core/MatrixBase.h" #include "src/Core/EigenBase.h" +#include "src/Core/Product.h" +#include "src/Core/CoreEvaluators.h" +#include "src/Core/AssignEvaluator.h" + #ifndef EIGEN_PARSED_BY_DOXYGEN // work around Doxygen bug triggered by Assign.h r814874 // at least confirmed with Doxygen 1.5.5 and 1.5.6 #include "src/Core/Assign.h" #endif +#include "src/Core/ArrayBase.h" #include "src/Core/util/BlasUtil.h" #include "src/Core/DenseStorage.h" #include "src/Core/NestByValue.h" -#include "src/Core/ForceAlignedAccess.h" + +// #include "src/Core/ForceAlignedAccess.h" + #include "src/Core/ReturnByValue.h" #include "src/Core/NoAlias.h" #include "src/Core/PlainObjectBase.h" #include "src/Core/Matrix.h" #include "src/Core/Array.h" +#include "src/Core/CwiseTernaryOp.h" #include "src/Core/CwiseBinaryOp.h" #include "src/Core/CwiseUnaryOp.h" #include "src/Core/CwiseNullaryOp.h" @@ -300,32 +444,32 @@ using std::ptrdiff_t; #include "src/Core/SelfCwiseBinaryOp.h" #include "src/Core/Dot.h" #include "src/Core/StableNorm.h" -#include "src/Core/MapBase.h" #include "src/Core/Stride.h" +#include "src/Core/MapBase.h" #include "src/Core/Map.h" +#include "src/Core/Ref.h" #include "src/Core/Block.h" #include "src/Core/VectorBlock.h" -#include "src/Core/Ref.h" #include "src/Core/Transpose.h" #include "src/Core/DiagonalMatrix.h" #include "src/Core/Diagonal.h" #include "src/Core/DiagonalProduct.h" -#include "src/Core/PermutationMatrix.h" -#include "src/Core/Transpositions.h" #include "src/Core/Redux.h" #include "src/Core/Visitor.h" #include "src/Core/Fuzzy.h" -#include "src/Core/IO.h" #include "src/Core/Swap.h" #include "src/Core/CommaInitializer.h" -#include "src/Core/Flagged.h" -#include "src/Core/ProductBase.h" #include "src/Core/GeneralProduct.h" +#include "src/Core/Solve.h" +#include "src/Core/Inverse.h" +#include "src/Core/SolverBase.h" +#include "src/Core/PermutationMatrix.h" +#include "src/Core/Transpositions.h" #include "src/Core/TriangularMatrix.h" #include "src/Core/SelfAdjointView.h" #include "src/Core/products/GeneralBlockPanelKernel.h" #include "src/Core/products/Parallelizer.h" -#include "src/Core/products/CoeffBasedProduct.h" +#include "src/Core/ProductEvaluators.h" #include "src/Core/products/GeneralMatrixVector.h" #include "src/Core/products/GeneralMatrixMatrix.h" #include "src/Core/SolveTriangular.h" @@ -340,6 +484,7 @@ using std::ptrdiff_t; #include "src/Core/products/TriangularSolverVector.h" #include "src/Core/BandMatrix.h" #include "src/Core/CoreIterators.h" +#include "src/Core/ConditionEstimator.h" #include "src/Core/BooleanRedux.h" #include "src/Core/Select.h" @@ -347,18 +492,17 @@ using std::ptrdiff_t; #include "src/Core/Random.h" #include "src/Core/Replicate.h" #include "src/Core/Reverse.h" -#include "src/Core/ArrayBase.h" #include "src/Core/ArrayWrapper.h" #ifdef EIGEN_USE_BLAS -#include "src/Core/products/GeneralMatrixMatrix_MKL.h" -#include "src/Core/products/GeneralMatrixVector_MKL.h" -#include "src/Core/products/GeneralMatrixMatrixTriangular_MKL.h" -#include "src/Core/products/SelfadjointMatrixMatrix_MKL.h" -#include "src/Core/products/SelfadjointMatrixVector_MKL.h" -#include "src/Core/products/TriangularMatrixMatrix_MKL.h" -#include "src/Core/products/TriangularMatrixVector_MKL.h" -#include "src/Core/products/TriangularSolverMatrix_MKL.h" +#include "src/Core/products/GeneralMatrixMatrix_BLAS.h" +#include "src/Core/products/GeneralMatrixVector_BLAS.h" +#include "src/Core/products/GeneralMatrixMatrixTriangular_BLAS.h" +#include "src/Core/products/SelfadjointMatrixMatrix_BLAS.h" +#include "src/Core/products/SelfadjointMatrixVector_BLAS.h" +#include "src/Core/products/TriangularMatrixMatrix_BLAS.h" +#include "src/Core/products/TriangularMatrixVector_BLAS.h" +#include "src/Core/products/TriangularSolverMatrix_BLAS.h" #endif // EIGEN_USE_BLAS #ifdef EIGEN_USE_MKL_VML @@ -369,8 +513,4 @@ using std::ptrdiff_t; #include "src/Core/util/ReenableStupidWarnings.h" -#ifdef EIGEN2_SUPPORT -#include "Eigen2Support" -#endif - #endif // EIGEN_CORE_H |