devtools/eigen
1
0
Fork 0
mirror of https://gitlab.com/libeigen/eigen.git synced 2026-04-10 11:34:33 +08:00
Code Issues Packages Projects Releases Wiki Activity

Merged eigen/eigen into default

Browse Source
This commit is contained in:
Deven Desai
2019-03-19 16:52:38 -04:00
parent e7e6809e6b 5c93b38c5f
commit 2dbea5510f
258 changed files with 9032 additions and 3992 deletions
Download Patch File Download Diff File Expand all files Collapse all files

12
Eigen/src/Core/util/BlasUtil.h
View File

@@ -274,7 +274,8 @@ template<typename XprType> struct blas_traits
HasUsableDirectAccess = ( (int(XprType::Flags)&DirectAccessBit)
&& ( bool(XprType::IsVectorAtCompileTime)
|| int(inner_stride_at_compile_time<XprType>::ret) == 1)
) ? 1 : 0
) ? 1 : 0,
HasScalarFactor = false
};
typedef typename conditional<bool(HasUsableDirectAccess),
ExtractType,
@@ -306,6 +307,9 @@ template<typename Scalar, typename NestedXpr, typename Plain>
struct blas_traits<CwiseBinaryOp<scalar_product_op<Scalar>, const CwiseNullaryOp<scalar_constant_op<Scalar>,Plain>, NestedXpr> >
: blas_traits<NestedXpr>
{
enum {
HasScalarFactor = true
};
typedef blas_traits<NestedXpr> Base;
typedef CwiseBinaryOp<scalar_product_op<Scalar>, const CwiseNullaryOp<scalar_constant_op<Scalar>,Plain>, NestedXpr> XprType;
typedef typename Base::ExtractType ExtractType;
@@ -317,6 +321,9 @@ template<typename Scalar, typename NestedXpr, typename Plain>
struct blas_traits<CwiseBinaryOp<scalar_product_op<Scalar>, NestedXpr, const CwiseNullaryOp<scalar_constant_op<Scalar>,Plain> > >
: blas_traits<NestedXpr>
{
enum {
HasScalarFactor = true
};
typedef blas_traits<NestedXpr> Base;
typedef CwiseBinaryOp<scalar_product_op<Scalar>, NestedXpr, const CwiseNullaryOp<scalar_constant_op<Scalar>,Plain> > XprType;
typedef typename Base::ExtractType ExtractType;
@@ -335,6 +342,9 @@ template<typename Scalar, typename NestedXpr>
struct blas_traits<CwiseUnaryOp<scalar_opposite_op<Scalar>, NestedXpr> >
: blas_traits<NestedXpr>
{
enum {
HasScalarFactor = true
};
typedef blas_traits<NestedXpr> Base;
typedef CwiseUnaryOp<scalar_opposite_op<Scalar>, NestedXpr> XprType;
typedef typename Base::ExtractType ExtractType;

39
Eigen/src/Core/util/ConfigureVectorization.h
View File

@@ -10,13 +10,6 @@
#ifndef EIGEN_CONFIGURE_VECTORIZATION_H
#define EIGEN_CONFIGURE_VECTORIZATION_H
// FIXME: not sure why this is needed, perhaps it is not needed anymore.
#ifdef __NVCC__
#ifndef EIGEN_DONT_VECTORIZE
#define EIGEN_DONT_VECTORIZE
#endif
#endif
//------------------------------------------------------------------------------------------
// Static and dynamic alignment control
//
@@ -36,10 +29,15 @@
*
* If we made alignment depend on whether or not EIGEN_VECTORIZE is defined, it would be impossible to link
* vectorized and non-vectorized code.
*
* FIXME: this code can be cleaned up once we switch to proper C++11 only.
*/
#if (defined EIGEN_CUDACC)
#define EIGEN_ALIGN_TO_BOUNDARY(n) __align__(n)
#define EIGEN_ALIGNOF(x) __alignof(x)
#elif EIGEN_HAS_ALIGNAS
#define EIGEN_ALIGN_TO_BOUNDARY(n) alignas(n)
#define EIGEN_ALIGNOF(x) alignof(x)
#elif EIGEN_COMP_GNUC || EIGEN_COMP_PGI || EIGEN_COMP_IBM || EIGEN_COMP_ARM
#define EIGEN_ALIGN_TO_BOUNDARY(n) __attribute__((aligned(n)))
#define EIGEN_ALIGNOF(x) __alignof(x)
@@ -51,12 +49,18 @@
#define EIGEN_ALIGN_TO_BOUNDARY(n) __attribute__((aligned(n)))
#define EIGEN_ALIGNOF(x) __alignof(x)
#else
#error Please tell me what is the equivalent of __attribute__((aligned(n))) and __alignof(x) for your compiler
#error Please tell me what is the equivalent of alignas(n) and alignof(x) for your compiler
#endif
// If the user explicitly disable vectorization, then we also disable alignment
#if defined(EIGEN_DONT_VECTORIZE)
#define EIGEN_IDEAL_MAX_ALIGN_BYTES 0
#if defined(EIGEN_GPUCC)
// GPU code is always vectorized and requires memory alignment for
// statically allocated buffers.
#define EIGEN_IDEAL_MAX_ALIGN_BYTES 16
#else
#define EIGEN_IDEAL_MAX_ALIGN_BYTES 0
#endif
#elif defined(__AVX512F__)
// 64 bytes static alignment is preferred only if really required
#define EIGEN_IDEAL_MAX_ALIGN_BYTES 64
@@ -183,8 +187,6 @@
//----------------------------------------------------------------------
// 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
@@ -211,7 +213,7 @@
#endif
#ifndef EIGEN_DONT_VECTORIZE
#if !(defined(EIGEN_DONT_VECTORIZE) || defined(EIGEN_GPUCC))
#if defined (EIGEN_SSE2_ON_NON_MSVC_BUT_NOT_OLD_GCC) || defined(EIGEN_SSE2_ON_MSVC_2008_OR_LATER)
@@ -253,10 +255,19 @@
#define EIGEN_VECTORIZE_SSE4_1
#define EIGEN_VECTORIZE_SSE4_2
#endif
#ifdef __FMA__
#if defined(__FMA__) || (EIGEN_COMP_MSVC && defined(__AVX2__))
// MSVC does not expose a switch dedicated for FMA
// For MSVC, AVX2 => FMA
#define EIGEN_VECTORIZE_FMA
#endif
#if defined(__AVX512F__)
#ifndef EIGEN_VECTORIZE_FMA
#if EIGEN_COMP_GNUC
#error Please add -mfma to your compiler flags: compiling with -mavx512f alone without SSE/AVX FMA is not supported (bug 1638).
#else
#error Please enable FMA in your compiler flags (e.g. -mfma): compiling with AVX512 alone without SSE/AVX FMA is not supported (bug 1638).
#endif
#endif
#define EIGEN_VECTORIZE_AVX512
#define EIGEN_VECTORIZE_AVX2
#define EIGEN_VECTORIZE_AVX
@@ -375,7 +386,7 @@
#endif
#if defined(EIGEN_HAS_CUDA_FP16)
#include <host_defines.h>
#include <cuda_runtime_api.h>
#include <cuda_fp16.h>
#endif

7
Eigen/src/Core/util/ForwardDeclarations.h
View File

@@ -47,11 +47,7 @@ template<typename T> struct NumTraits;
template<typename Derived> struct EigenBase;
template<typename Derived> class DenseBase;
template<typename Derived> class PlainObjectBase;
template<typename Derived,
int Level = internal::accessors_level<Derived>::value >
class DenseCoeffsBase;
template<typename Derived, int Level> class DenseCoeffsBase;
template<typename _Scalar, int _Rows, int _Cols,
int _Options = AutoAlign |
@@ -260,6 +256,7 @@ template<typename MatrixType> class HouseholderQR;
template<typename MatrixType> class ColPivHouseholderQR;
template<typename MatrixType> class FullPivHouseholderQR;
template<typename MatrixType> class CompleteOrthogonalDecomposition;
template<typename MatrixType> class SVDBase;
template<typename MatrixType, int QRPreconditioner = ColPivHouseholderQRPreconditioner> class JacobiSVD;
template<typename MatrixType> class BDCSVD;
template<typename MatrixType, int UpLo = Lower> class LLT;

4
Eigen/src/Core/util/IndexedViewHelper.h
View File

@@ -23,7 +23,7 @@ struct symbolic_last_tag {};
* Can be used as a parameter to Eigen::seq and Eigen::seqN functions to symbolically reference the last element/row/columns
* of the underlying vector or matrix once passed to DenseBase::operator()(const RowIndices&, const ColIndices&).
*
* This symbolic placeholder support standard arithmetic operation.
* This symbolic placeholder supports standard arithmetic operations.
*
* A typical usage example would be:
* \code
@@ -44,7 +44,7 @@ static const symbolic::SymbolExpr<internal::symbolic_last_tag> last; // PLEASE u
* reference the last+1 element/row/columns of the underlying vector or matrix once
* passed to DenseBase::operator()(const RowIndices&, const ColIndices&).
*
* This symbolic placeholder support standard arithmetic operation.
* This symbolic placeholder supports standard arithmetic operations.
* It is essentially an alias to last+fix<1>.
*
* \sa last

98
Eigen/src/Core/util/Macros.h
View File

@@ -129,16 +129,21 @@
#define EIGEN_COMP_MSVC_STRICT 0
#endif
/// \internal EIGEN_COMP_IBM set to 1 if the compiler is IBM XL C++
#if defined(__IBMCPP__) || defined(__xlc__)
#define EIGEN_COMP_IBM 1
/// \internal EIGEN_COMP_IBM set to xlc version if the compiler is IBM XL C++
// XLC version
// 3.1 0x0301
// 4.5 0x0405
// 5.0 0x0500
// 12.1 0x0C01
#if defined(__IBMCPP__) || defined(__xlc__) || defined(__ibmxl__)
#define EIGEN_COMP_IBM __xlC__
#else
#define EIGEN_COMP_IBM 0
#endif
/// \internal EIGEN_COMP_PGI set to 1 if the compiler is Portland Group Compiler
/// \internal EIGEN_COMP_PGI set to PGI version if the compiler is Portland Group Compiler
#if defined(__PGI)
#define EIGEN_COMP_PGI 1
#define EIGEN_COMP_PGI (__PGIC__*100+__PGIC_MINOR__)
#else
#define EIGEN_COMP_PGI 0
#endif
@@ -347,9 +352,17 @@
#define EIGEN_OS_WIN_STRICT 0
#endif
/// \internal EIGEN_OS_SUN set to 1 if the OS is SUN
/// \internal EIGEN_OS_SUN set to __SUNPRO_C if the OS is SUN
// compiler solaris __SUNPRO_C
// version studio
// 5.7 10 0x570
// 5.8 11 0x580
// 5.9 12 0x590
// 5.10 12.1 0x5100
// 5.11 12.2 0x5110
// 5.12 12.3 0x5120
#if (defined(sun) || defined(__sun)) && !(defined(__SVR4) || defined(__svr4__))
#define EIGEN_OS_SUN 1
#define EIGEN_OS_SUN __SUNPRO_C
#else
#define EIGEN_OS_SUN 0
#endif
@@ -495,13 +508,33 @@
#define EIGEN_HAS_STATIC_ARRAY_TEMPLATE 0
#endif
#if EIGEN_MAX_CPP_VER>=11 && (defined(__cplusplus) && (__cplusplus >= 201103L) || EIGEN_COMP_MSVC >= 1900)
// The macro EIGEN_COMP_CXXVER defines the c++ verson expected by the compiler.
// For instance, if compiling with gcc and -std=c++17, then EIGEN_COMP_CXXVER
// is defined to 17.
#if (defined(__cplusplus) && (__cplusplus > 201402L) || EIGEN_COMP_MSVC >= 1914)
#define EIGEN_COMP_CXXVER 17
#elif (defined(__cplusplus) && (__cplusplus > 201103L) || EIGEN_COMP_MSVC >= 1910)
#define EIGEN_COMP_CXXVER 14
#elif (defined(__cplusplus) && (__cplusplus >= 201103L) || EIGEN_COMP_MSVC >= 1900)
#define EIGEN_COMP_CXXVER 11
#else
#define EIGEN_COMP_CXXVER 03
#endif
// The macros EIGEN_HAS_CXX?? defines a rough estimate of available c++ features
// but in practice we should not rely on them but rather on the availabilty of
// individual features as defined later.
// This is why there is no EIGEN_HAS_CXX17.
// FIXME: get rid of EIGEN_HAS_CXX14 and maybe even EIGEN_HAS_CXX11.
#if EIGEN_MAX_CPP_VER>=11 && EIGEN_COMP_CXXVER>=11
#define EIGEN_HAS_CXX11 1
#else
#define EIGEN_HAS_CXX11 0
#endif
#if EIGEN_MAX_CPP_VER>=14 && (defined(__cplusplus) && (__cplusplus > 201103L) || EIGEN_COMP_MSVC >= 1910)
#if EIGEN_MAX_CPP_VER>=14 && EIGEN_COMP_CXXVER>=14
#define EIGEN_HAS_CXX14 1
#else
#define EIGEN_HAS_CXX14 0
@@ -546,6 +579,22 @@
#endif
#endif
#ifndef EIGEN_HAS_ALIGNAS
#if EIGEN_MAX_CPP_VER>=11 && EIGEN_HAS_CXX11 && \
( __has_feature(cxx_alignas) \
|| EIGEN_HAS_CXX14 \
|| (EIGEN_COMP_MSVC >= 1800) \
|| (EIGEN_GNUC_AT_LEAST(4,8)) \
|| (EIGEN_COMP_CLANG>=305) \
|| (EIGEN_COMP_ICC>=1500) \
|| (EIGEN_COMP_PGI>=1500) \
|| (EIGEN_COMP_SUNCC>=0x5130))
#define EIGEN_HAS_ALIGNAS 1
#else
#define EIGEN_HAS_ALIGNAS 0
#endif
#endif
// Does the compiler support type_traits?
// - full support of type traits was added only to GCC 5.1.0.
// - 20150626 corresponds to the last release of 4.x libstdc++
@@ -649,6 +698,23 @@
#endif
#endif
// NOTE: the required Apple's clang version is very conservative
// and it could be that XCode 9 works just fine.
// NOTE: the MSVC version is based on https://en.cppreference.com/w/cpp/compiler_support
// and not tested.
#ifndef EIGEN_HAS_CXX17_OVERALIGN
#if EIGEN_MAX_CPP_VER>=17 && EIGEN_COMP_CXXVER>=17 && ( \
(EIGEN_COMP_MSVC >= 1912) \
|| (EIGEN_GNUC_AT_LEAST(7,0)) \
|| ((!defined(__apple_build_version__)) && (EIGEN_COMP_CLANG>=500)) \
|| (( defined(__apple_build_version__)) && (__apple_build_version__>=10000000)) \
)
#define EIGEN_HAS_CXX17_OVERALIGN 1
#else
#define EIGEN_HAS_CXX17_OVERALIGN 0
#endif
#endif
#if defined(EIGEN_CUDACC) && EIGEN_HAS_CONSTEXPR
// While available already with c++11, this is useful mostly starting with c++14 and relaxed constexpr rules
#if defined(__NVCC__)
@@ -742,10 +808,6 @@
// All functions callable from CUDA/HIP code must be qualified with __device__
#ifdef EIGEN_GPUCC
#ifndef EIGEN_DONT_VECTORIZE
#define EIGEN_DONT_VECTORIZE
#endif
#define EIGEN_DEVICE_FUNC __host__ __device__
#else
#define EIGEN_DEVICE_FUNC
@@ -841,7 +903,7 @@
// Suppresses 'unused variable' warnings.
namespace Eigen {
namespace internal {
template<typename T> EIGEN_DEVICE_FUNC void ignore_unused_variable(const T&) {}
template<typename T> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void ignore_unused_variable(const T&) {}
}
}
#define EIGEN_UNUSED_VARIABLE(var) Eigen::internal::ignore_unused_variable(var);
@@ -1035,7 +1097,7 @@ namespace Eigen {
#endif
#define EIGEN_MAKE_SCALAR_BINARY_OP_ONTHERIGHT(METHOD,OPNAME) \
template <typename T> EIGEN_DEVICE_FUNC inline \
template <typename T> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE \
EIGEN_MSVC10_WORKAROUND_BINARYOP_RETURN_TYPE(const EIGEN_EXPR_BINARYOP_SCALAR_RETURN_TYPE(Derived,typename internal::promote_scalar_arg<Scalar EIGEN_COMMA T EIGEN_COMMA EIGEN_SCALAR_BINARY_SUPPORTED(OPNAME,Scalar,T)>::type,OPNAME))\
(METHOD)(const T& scalar) const { \
typedef typename internal::promote_scalar_arg<Scalar,T,EIGEN_SCALAR_BINARY_SUPPORTED(OPNAME,Scalar,T)>::type PromotedT; \
@@ -1044,7 +1106,7 @@ namespace Eigen {
}
#define EIGEN_MAKE_SCALAR_BINARY_OP_ONTHELEFT(METHOD,OPNAME) \
template <typename T> EIGEN_DEVICE_FUNC inline friend \
template <typename T> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE friend \
EIGEN_MSVC10_WORKAROUND_BINARYOP_RETURN_TYPE(const EIGEN_SCALAR_BINARYOP_EXPR_RETURN_TYPE(typename internal::promote_scalar_arg<Scalar EIGEN_COMMA T EIGEN_COMMA EIGEN_SCALAR_BINARY_SUPPORTED(OPNAME,T,Scalar)>::type,Derived,OPNAME)) \
(METHOD)(const T& scalar, const StorageBaseType& matrix) { \
typedef typename internal::promote_scalar_arg<Scalar,T,EIGEN_SCALAR_BINARY_SUPPORTED(OPNAME,T,Scalar)>::type PromotedT; \
@@ -1093,9 +1155,9 @@ namespace Eigen {
# define EIGEN_NOEXCEPT
# define EIGEN_NOEXCEPT_IF(x)
# define EIGEN_NO_THROW throw()
# if EIGEN_COMP_MSVC
# if EIGEN_COMP_MSVC || EIGEN_COMP_CXXVER>=17
// MSVC does not support exception specifications (warning C4290),
// and they are deprecated in c++11 anyway.
// and they are deprecated in c++11 anyway. This is even an error in c++17.
# define EIGEN_EXCEPTION_SPEC(X) throw()
# else
# define EIGEN_EXCEPTION_SPEC(X) throw(X)

15
Eigen/src/Core/util/Memory.h
View File

@@ -360,7 +360,7 @@ template<typename T, bool Align> EIGEN_DEVICE_FUNC inline T* conditional_aligned
template<typename T> EIGEN_DEVICE_FUNC inline void aligned_delete(T *ptr, std::size_t size)
{
destruct_elements_of_array<T>(ptr, size);
aligned_free(ptr);
Eigen::internal::aligned_free(ptr);
}
/** \internal Deletes objects constructed with conditional_aligned_new
@@ -768,6 +768,17 @@ template<typename T> void swap(scoped_array<T> &a,scoped_array<T> &b)
*** Implementation of EIGEN_MAKE_ALIGNED_OPERATOR_NEW [_IF] ***
*****************************************************************************/
#if EIGEN_HAS_CXX17_OVERALIGN
// C++17 -> no need to bother about alignment anymore :)
#define EIGEN_MAKE_ALIGNED_OPERATOR_NEW_NOTHROW(NeedsToAlign)
#define EIGEN_MAKE_ALIGNED_OPERATOR_NEW_IF(NeedsToAlign)
#define EIGEN_MAKE_ALIGNED_OPERATOR_NEW
#define EIGEN_MAKE_ALIGNED_OPERATOR_NEW_IF_VECTORIZABLE_FIXED_SIZE(Scalar,Size)
#else
#if EIGEN_MAX_ALIGN_BYTES!=0
#define EIGEN_MAKE_ALIGNED_OPERATOR_NEW_NOTHROW(NeedsToAlign) \
void* operator new(std::size_t size, const std::nothrow_t&) EIGEN_NO_THROW { \
@@ -810,6 +821,8 @@ template<typename T> void swap(scoped_array<T> &a,scoped_array<T> &b)
((EIGEN_MAX_ALIGN_BYTES>=32) && ((sizeof(Scalar)*(Size))%(EIGEN_MAX_ALIGN_BYTES/2)==0)) || \
((EIGEN_MAX_ALIGN_BYTES>=64) && ((sizeof(Scalar)*(Size))%(EIGEN_MAX_ALIGN_BYTES/4)==0)) )))
#endif
/****************************************************************************/
/** \class aligned_allocator

33
Eigen/src/Core/util/Meta.h
View File

@@ -636,8 +636,41 @@ template<> EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC
bool not_equal_strict(const double& x,const double& y) { return std::not_equal_to<double>()(x,y); }
#endif
/** \internal extract the bits of the float \a x */
inline unsigned int as_uint(float x)
{
unsigned int ret;
std::memcpy(&ret, &x, sizeof(float));
return ret;
}
} // end namespace numext
} // end namespace Eigen
// Define portable (u)int{32,64} types
#if EIGEN_HAS_CXX11
#include <cstdint>
namespace Eigen {
namespace numext {
typedef std::uint32_t uint32_t;
typedef std::int32_t int32_t;
typedef std::uint64_t uint64_t;
typedef std::int64_t int64_t;
}
}
#else
// Without c++11, all compilers able to compile Eigen also
// provides the C99 stdint.h header file.
#include <stdint.h>
namespace Eigen {
namespace numext {
typedef ::uint32_t uint32_t;
typedef ::int32_t int32_t;
typedef ::uint64_t uint64_t;
typedef ::int64_t int64_t;
}
}
#endif
#endif // EIGEN_META_H

3
Eigen/src/Core/util/StaticAssert.h
View File

@@ -104,7 +104,8 @@
STORAGE_INDEX_MUST_MATCH=1,
CHOLMOD_SUPPORTS_DOUBLE_PRECISION_ONLY=1,
SELFADJOINTVIEW_ACCEPTS_UPPER_AND_LOWER_MODE_ONLY=1,
INVALID_TEMPLATE_PARAMETER=1
INVALID_TEMPLATE_PARAMETER=1,
GPU_TENSOR_CONTRACTION_DOES_NOT_SUPPORT_OUTPUT_KERNELS=1
};
};

3
Eigen/src/Core/util/XprHelper.h
View File

@@ -184,7 +184,8 @@ template<typename T> struct unpacket_traits
enum
{
size = 1,
alignment = 1
alignment = 1,
vectorizable = false
};
};