未验证 提交 846ce406 编写于 作者: Q Qi Li 提交者: GitHub

[ROCM] update eigen cmake and patch, test=develop (#30602)

上级 173660be
...@@ -26,13 +26,6 @@ if(WIN32) ...@@ -26,13 +26,6 @@ if(WIN32)
set(EIGEN_TAG 917060c364181f33a735dc023818d5a54f60e54c) set(EIGEN_TAG 917060c364181f33a735dc023818d5a54f60e54c)
endif() endif()
# eigen on cuda9.1 missing header of math_funtions.hpp
# https://stackoverflow.com/questions/43113508/math-functions-hpp-not-found-when-using-cuda-with-eigen
if(WITH_ROCM_PLATFORM)
set(EIGEN_REPOSITORY ${GIT_URL}/sabreshao/hipeigen.git)
set(EIGEN_TAG 7cb2b6e5a4b4a1efe658abb215cd866c6fb2275e)
endif()
cache_third_party(extern_eigen3 cache_third_party(extern_eigen3
REPOSITORY ${EIGEN_REPOSITORY} REPOSITORY ${EIGEN_REPOSITORY}
TAG ${EIGEN_TAG} TAG ${EIGEN_TAG}
...@@ -56,14 +49,25 @@ elseif(LINUX) ...@@ -56,14 +49,25 @@ elseif(LINUX)
# add patch to avoid compilation error in c++11 # add patch to avoid compilation error in c++11
file(TO_NATIVE_PATH ${PADDLE_SOURCE_DIR}/patches/eigen/MathFunctions.h native_src2) file(TO_NATIVE_PATH ${PADDLE_SOURCE_DIR}/patches/eigen/MathFunctions.h native_src2)
file(TO_NATIVE_PATH ${EIGEN_SOURCE_DIR}/Eigen/src/Core/MathFunctions.h native_dst2) file(TO_NATIVE_PATH ${EIGEN_SOURCE_DIR}/Eigen/src/Core/MathFunctions.h native_dst2)
if(WITH_ROCM)
# For HIPCC Eigen::internal::device::numeric_limits is not EIGEN_DEVICE_FUNC
# which will cause compiler error of using __host__ funciont in __host__ __device__
file(TO_NATIVE_PATH ${PADDLE_SOURCE_DIR}/patches/eigen/Meta.h native_src3)
file(TO_NATIVE_PATH ${EIGEN_SOURCE_DIR}/Eigen/src/Core/util/Meta.h native_dst3)
# For HIPCC Eigen::internal::scalar_sum_op<bool,bool> is not EIGEN_DEVICE_FUNC
# which will cause compiler error of using __host__ funciont in __host__ __device__
file(TO_NATIVE_PATH ${PADDLE_SOURCE_DIR}/patches/eigen/BinaryFunctors.h native_src4)
file(TO_NATIVE_PATH ${EIGEN_SOURCE_DIR}/Eigen/src/Core/functors/BinaryFunctors.h native_dst4)
set(EIGEN_PATCH_COMMAND cp ${native_src1} ${native_dst1} && cp ${native_src2} ${native_dst2} && cp ${native_src3} ${native_dst3} && cp ${native_src4} ${native_dst4})
else()
set(EIGEN_PATCH_COMMAND cp ${native_src1} ${native_dst1} && cp ${native_src2} ${native_dst2}) set(EIGEN_PATCH_COMMAND cp ${native_src1} ${native_dst1} && cp ${native_src2} ${native_dst2})
endif()
endif() endif()
set(EIGEN_INCLUDE_DIR ${EIGEN_SOURCE_DIR}) set(EIGEN_INCLUDE_DIR ${EIGEN_SOURCE_DIR})
INCLUDE_DIRECTORIES(${EIGEN_INCLUDE_DIR}) INCLUDE_DIRECTORIES(${EIGEN_INCLUDE_DIR})
if(WITH_AMD_GPU) ExternalProject_Add(
ExternalProject_Add(
extern_eigen3 extern_eigen3
${EXTERNAL_PROJECT_LOG_ARGS} ${EXTERNAL_PROJECT_LOG_ARGS}
${SHALLOW_CLONE} ${SHALLOW_CLONE}
...@@ -76,23 +80,7 @@ if(WITH_AMD_GPU) ...@@ -76,23 +80,7 @@ if(WITH_AMD_GPU)
BUILD_COMMAND "" BUILD_COMMAND ""
INSTALL_COMMAND "" INSTALL_COMMAND ""
TEST_COMMAND "" TEST_COMMAND ""
) )
else()
ExternalProject_Add(
extern_eigen3
${EXTERNAL_PROJECT_LOG_ARGS}
${SHALLOW_CLONE}
"${EIGEN_DOWNLOAD_CMD}"
PREFIX ${EIGEN_PREFIX_DIR}
SOURCE_DIR ${EIGEN_SOURCE_DIR}
UPDATE_COMMAND ""
PATCH_COMMAND ${EIGEN_PATCH_COMMAND}
CONFIGURE_COMMAND ""
BUILD_COMMAND ""
INSTALL_COMMAND ""
TEST_COMMAND ""
)
endif()
add_library(eigen3 INTERFACE) add_library(eigen3 INTERFACE)
......
// This file is part of Eigen, a lightweight C++ template library
// for linear algebra.
//
// Copyright (C) 2008-2010 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/.
// clang-format off
#ifndef EIGEN_BINARY_FUNCTORS_H
#define EIGEN_BINARY_FUNCTORS_H
namespace Eigen {
namespace internal {
//---------- associative binary functors ----------
template<typename Arg1, typename Arg2>
struct binary_op_base
{
typedef Arg1 first_argument_type;
typedef Arg2 second_argument_type;
};
/** \internal
* \brief Template functor to compute the sum of two scalars
*
* \sa class CwiseBinaryOp, MatrixBase::operator+, class VectorwiseOp, DenseBase::sum()
*/
template<typename LhsScalar,typename RhsScalar>
struct scalar_sum_op : binary_op_base<LhsScalar,RhsScalar>
{
typedef typename ScalarBinaryOpTraits<LhsScalar,RhsScalar,scalar_sum_op>::ReturnType result_type;
#ifndef EIGEN_SCALAR_BINARY_OP_PLUGIN
EIGEN_EMPTY_STRUCT_CTOR(scalar_sum_op)
#else
scalar_sum_op() {
EIGEN_SCALAR_BINARY_OP_PLUGIN
}
#endif
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const result_type operator() (const LhsScalar& a, const RhsScalar& b) const { return a + b; }
template<typename Packet>
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const Packet packetOp(const Packet& a, const Packet& b) const
{ return internal::padd(a,b); }
template<typename Packet>
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const result_type predux(const Packet& a) const
{ return internal::predux(a); }
};
template<typename LhsScalar,typename RhsScalar>
struct functor_traits<scalar_sum_op<LhsScalar,RhsScalar> > {
enum {
Cost = (NumTraits<LhsScalar>::AddCost+NumTraits<RhsScalar>::AddCost)/2, // rough estimate!
PacketAccess = is_same<LhsScalar,RhsScalar>::value && packet_traits<LhsScalar>::HasAdd && packet_traits<RhsScalar>::HasAdd
// TODO vectorize mixed sum
};
};
/** \internal
* \brief Template specialization to deprecate the summation of boolean expressions.
* This is required to solve Bug 426.
* \sa DenseBase::count(), DenseBase::any(), ArrayBase::cast(), MatrixBase::cast()
*/
template<> struct scalar_sum_op<bool,bool> : scalar_sum_op<int,int> {
EIGEN_DEPRECATED EIGEN_DEVICE_FUNC
scalar_sum_op() {}
};
/** \internal
* \brief Template functor to compute the product of two scalars
*
* \sa class CwiseBinaryOp, Cwise::operator*(), class VectorwiseOp, MatrixBase::redux()
*/
template<typename LhsScalar,typename RhsScalar>
struct scalar_product_op : binary_op_base<LhsScalar,RhsScalar>
{
typedef typename ScalarBinaryOpTraits<LhsScalar,RhsScalar,scalar_product_op>::ReturnType result_type;
#ifndef EIGEN_SCALAR_BINARY_OP_PLUGIN
EIGEN_EMPTY_STRUCT_CTOR(scalar_product_op)
#else
scalar_product_op() {
EIGEN_SCALAR_BINARY_OP_PLUGIN
}
#endif
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const result_type operator() (const LhsScalar& a, const RhsScalar& b) const { return a * b; }
template<typename Packet>
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const Packet packetOp(const Packet& a, const Packet& b) const
{ return internal::pmul(a,b); }
template<typename Packet>
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const result_type predux(const Packet& a) const
{ return internal::predux_mul(a); }
};
template<typename LhsScalar,typename RhsScalar>
struct functor_traits<scalar_product_op<LhsScalar,RhsScalar> > {
enum {
Cost = (NumTraits<LhsScalar>::MulCost + NumTraits<RhsScalar>::MulCost)/2, // rough estimate!
PacketAccess = is_same<LhsScalar,RhsScalar>::value && packet_traits<LhsScalar>::HasMul && packet_traits<RhsScalar>::HasMul
// TODO vectorize mixed product
};
};
/** \internal
* \brief Template functor to compute the conjugate product of two scalars
*
* This is a short cut for conj(x) * y which is needed for optimization purpose; in Eigen2 support mode, this becomes x * conj(y)
*/
template<typename LhsScalar,typename RhsScalar>
struct scalar_conj_product_op : binary_op_base<LhsScalar,RhsScalar>
{
enum {
Conj = NumTraits<LhsScalar>::IsComplex
};
typedef typename ScalarBinaryOpTraits<LhsScalar,RhsScalar,scalar_conj_product_op>::ReturnType result_type;
EIGEN_EMPTY_STRUCT_CTOR(scalar_conj_product_op)
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const result_type operator() (const LhsScalar& a, const RhsScalar& b) const
{ return conj_helper<LhsScalar,RhsScalar,Conj,false>().pmul(a,b); }
template<typename Packet>
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const Packet packetOp(const Packet& a, const Packet& b) const
{ return conj_helper<Packet,Packet,Conj,false>().pmul(a,b); }
};
template<typename LhsScalar,typename RhsScalar>
struct functor_traits<scalar_conj_product_op<LhsScalar,RhsScalar> > {
enum {
Cost = NumTraits<LhsScalar>::MulCost,
PacketAccess = internal::is_same<LhsScalar, RhsScalar>::value && packet_traits<LhsScalar>::HasMul
};
};
/** \internal
* \brief Template functor to compute the min of two scalars
*
* \sa class CwiseBinaryOp, MatrixBase::cwiseMin, class VectorwiseOp, MatrixBase::minCoeff()
*/
template<typename LhsScalar,typename RhsScalar>
struct scalar_min_op : binary_op_base<LhsScalar,RhsScalar>
{
typedef typename ScalarBinaryOpTraits<LhsScalar,RhsScalar,scalar_min_op>::ReturnType result_type;
EIGEN_EMPTY_STRUCT_CTOR(scalar_min_op)
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const result_type operator() (const LhsScalar& a, const RhsScalar& b) const { return numext::mini(a, b); }
template<typename Packet>
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const Packet packetOp(const Packet& a, const Packet& b) const
{ return internal::pmin(a,b); }
template<typename Packet>
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const result_type predux(const Packet& a) const
{ return internal::predux_min(a); }
};
template<typename LhsScalar,typename RhsScalar>
struct functor_traits<scalar_min_op<LhsScalar,RhsScalar> > {
enum {
Cost = (NumTraits<LhsScalar>::AddCost+NumTraits<RhsScalar>::AddCost)/2,
PacketAccess = internal::is_same<LhsScalar, RhsScalar>::value && packet_traits<LhsScalar>::HasMin
};
};
/** \internal
* \brief Template functor to compute the max of two scalars
*
* \sa class CwiseBinaryOp, MatrixBase::cwiseMax, class VectorwiseOp, MatrixBase::maxCoeff()
*/
template<typename LhsScalar,typename RhsScalar>
struct scalar_max_op : binary_op_base<LhsScalar,RhsScalar>
{
typedef typename ScalarBinaryOpTraits<LhsScalar,RhsScalar,scalar_max_op>::ReturnType result_type;
EIGEN_EMPTY_STRUCT_CTOR(scalar_max_op)
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const result_type operator() (const LhsScalar& a, const RhsScalar& b) const { return numext::maxi(a, b); }
template<typename Packet>
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const Packet packetOp(const Packet& a, const Packet& b) const
{ return internal::pmax(a,b); }
template<typename Packet>
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const result_type predux(const Packet& a) const
{ return internal::predux_max(a); }
};
template<typename LhsScalar,typename RhsScalar>
struct functor_traits<scalar_max_op<LhsScalar,RhsScalar> > {
enum {
Cost = (NumTraits<LhsScalar>::AddCost+NumTraits<RhsScalar>::AddCost)/2,
PacketAccess = internal::is_same<LhsScalar, RhsScalar>::value && packet_traits<LhsScalar>::HasMax
};
};
/** \internal
* \brief Template functors for comparison of two scalars
* \todo Implement packet-comparisons
*/
template<typename LhsScalar, typename RhsScalar, ComparisonName cmp> struct scalar_cmp_op;
template<typename LhsScalar, typename RhsScalar, ComparisonName cmp>
struct functor_traits<scalar_cmp_op<LhsScalar,RhsScalar, cmp> > {
enum {
Cost = (NumTraits<LhsScalar>::AddCost+NumTraits<RhsScalar>::AddCost)/2,
PacketAccess = false
};
};
template<ComparisonName Cmp, typename LhsScalar, typename RhsScalar>
struct result_of<scalar_cmp_op<LhsScalar, RhsScalar, Cmp>(LhsScalar,RhsScalar)> {
typedef bool type;
};
template<typename LhsScalar, typename RhsScalar>
struct scalar_cmp_op<LhsScalar,RhsScalar, cmp_EQ> : binary_op_base<LhsScalar,RhsScalar>
{
typedef bool result_type;
EIGEN_EMPTY_STRUCT_CTOR(scalar_cmp_op)
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE bool operator()(const LhsScalar& a, const RhsScalar& b) const {return a==b;}
};
template<typename LhsScalar, typename RhsScalar>
struct scalar_cmp_op<LhsScalar,RhsScalar, cmp_LT> : binary_op_base<LhsScalar,RhsScalar>
{
typedef bool result_type;
EIGEN_EMPTY_STRUCT_CTOR(scalar_cmp_op)
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE bool operator()(const LhsScalar& a, const RhsScalar& b) const {return a<b;}
};
template<typename LhsScalar, typename RhsScalar>
struct scalar_cmp_op<LhsScalar,RhsScalar, cmp_LE> : binary_op_base<LhsScalar,RhsScalar>
{
typedef bool result_type;
EIGEN_EMPTY_STRUCT_CTOR(scalar_cmp_op)
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE bool operator()(const LhsScalar& a, const RhsScalar& b) const {return a<=b;}
};
template<typename LhsScalar, typename RhsScalar>
struct scalar_cmp_op<LhsScalar,RhsScalar, cmp_GT> : binary_op_base<LhsScalar,RhsScalar>
{
typedef bool result_type;
EIGEN_EMPTY_STRUCT_CTOR(scalar_cmp_op)
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE bool operator()(const LhsScalar& a, const RhsScalar& b) const {return a>b;}
};
template<typename LhsScalar, typename RhsScalar>
struct scalar_cmp_op<LhsScalar,RhsScalar, cmp_GE> : binary_op_base<LhsScalar,RhsScalar>
{
typedef bool result_type;
EIGEN_EMPTY_STRUCT_CTOR(scalar_cmp_op)
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE bool operator()(const LhsScalar& a, const RhsScalar& b) const {return a>=b;}
};
template<typename LhsScalar, typename RhsScalar>
struct scalar_cmp_op<LhsScalar,RhsScalar, cmp_UNORD> : binary_op_base<LhsScalar,RhsScalar>
{
typedef bool result_type;
EIGEN_EMPTY_STRUCT_CTOR(scalar_cmp_op)
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE bool operator()(const LhsScalar& a, const RhsScalar& b) const {return !(a<=b || b<=a);}
};
template<typename LhsScalar, typename RhsScalar>
struct scalar_cmp_op<LhsScalar,RhsScalar, cmp_NEQ> : binary_op_base<LhsScalar,RhsScalar>
{
typedef bool result_type;
EIGEN_EMPTY_STRUCT_CTOR(scalar_cmp_op)
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE bool operator()(const LhsScalar& a, const RhsScalar& b) const {return a!=b;}
};
/** \internal
* \brief Template functor to compute the hypot of two \b positive \b and \b real scalars
*
* \sa MatrixBase::stableNorm(), class Redux
*/
template<typename Scalar>
struct scalar_hypot_op<Scalar,Scalar> : binary_op_base<Scalar,Scalar>
{
EIGEN_EMPTY_STRUCT_CTOR(scalar_hypot_op)
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const Scalar operator() (const Scalar &x, const Scalar &y) const
{
// This functor is used by hypotNorm only for which it is faster to first apply abs
// on all coefficients prior to reduction through hypot.
// This way we avoid calling abs on positive and real entries, and this also permits
// to seamlessly handle complexes. Otherwise we would have to handle both real and complexes
// through the same functor...
return internal::positive_real_hypot(x,y);
}
};
template<typename Scalar>
struct functor_traits<scalar_hypot_op<Scalar,Scalar> > {
enum
{
Cost = 3 * NumTraits<Scalar>::AddCost +
2 * NumTraits<Scalar>::MulCost +
2 * scalar_div_cost<Scalar,false>::value,
PacketAccess = false
};
};
/** \internal
* \brief Template functor to compute the pow of two scalars
*/
template<typename Scalar, typename Exponent>
struct scalar_pow_op : binary_op_base<Scalar,Exponent>
{
typedef typename ScalarBinaryOpTraits<Scalar,Exponent,scalar_pow_op>::ReturnType result_type;
#ifndef EIGEN_SCALAR_BINARY_OP_PLUGIN
EIGEN_EMPTY_STRUCT_CTOR(scalar_pow_op)
#else
scalar_pow_op() {
typedef Scalar LhsScalar;
typedef Exponent RhsScalar;
EIGEN_SCALAR_BINARY_OP_PLUGIN
}
#endif
EIGEN_DEVICE_FUNC
inline result_type operator() (const Scalar& a, const Exponent& b) const { return numext::pow(a, b); }
};
template<typename Scalar, typename Exponent>
struct functor_traits<scalar_pow_op<Scalar,Exponent> > {
enum { Cost = 5 * NumTraits<Scalar>::MulCost, PacketAccess = false };
};
//---------- non associative binary functors ----------
/** \internal
* \brief Template functor to compute the difference of two scalars
*
* \sa class CwiseBinaryOp, MatrixBase::operator-
*/
template<typename LhsScalar,typename RhsScalar>
struct scalar_difference_op : binary_op_base<LhsScalar,RhsScalar>
{
typedef typename ScalarBinaryOpTraits<LhsScalar,RhsScalar,scalar_difference_op>::ReturnType result_type;
#ifndef EIGEN_SCALAR_BINARY_OP_PLUGIN
EIGEN_EMPTY_STRUCT_CTOR(scalar_difference_op)
#else
scalar_difference_op() {
EIGEN_SCALAR_BINARY_OP_PLUGIN
}
#endif
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const result_type operator() (const LhsScalar& a, const RhsScalar& b) const { return a - b; }
template<typename Packet>
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const Packet packetOp(const Packet& a, const Packet& b) const
{ return internal::psub(a,b); }
};
template<typename LhsScalar,typename RhsScalar>
struct functor_traits<scalar_difference_op<LhsScalar,RhsScalar> > {
enum {
Cost = (NumTraits<LhsScalar>::AddCost+NumTraits<RhsScalar>::AddCost)/2,
PacketAccess = is_same<LhsScalar,RhsScalar>::value && packet_traits<LhsScalar>::HasSub && packet_traits<RhsScalar>::HasSub
};
};
/** \internal
* \brief Template functor to compute the quotient of two scalars
*
* \sa class CwiseBinaryOp, Cwise::operator/()
*/
template<typename LhsScalar,typename RhsScalar>
struct scalar_quotient_op : binary_op_base<LhsScalar,RhsScalar>
{
typedef typename ScalarBinaryOpTraits<LhsScalar,RhsScalar,scalar_quotient_op>::ReturnType result_type;
#ifndef EIGEN_SCALAR_BINARY_OP_PLUGIN
EIGEN_EMPTY_STRUCT_CTOR(scalar_quotient_op)
#else
scalar_quotient_op() {
EIGEN_SCALAR_BINARY_OP_PLUGIN
}
#endif
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const result_type operator() (const LhsScalar& a, const RhsScalar& b) const { return a / b; }
template<typename Packet>
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const Packet packetOp(const Packet& a, const Packet& b) const
{ return internal::pdiv(a,b); }
};
template<typename LhsScalar,typename RhsScalar>
struct functor_traits<scalar_quotient_op<LhsScalar,RhsScalar> > {
typedef typename scalar_quotient_op<LhsScalar,RhsScalar>::result_type result_type;
enum {
PacketAccess = is_same<LhsScalar,RhsScalar>::value && packet_traits<LhsScalar>::HasDiv && packet_traits<RhsScalar>::HasDiv,
Cost = scalar_div_cost<result_type,PacketAccess>::value
};
};
/** \internal
* \brief Template functor to compute the and of two booleans
*
* \sa class CwiseBinaryOp, ArrayBase::operator&&
*/
struct scalar_boolean_and_op {
EIGEN_EMPTY_STRUCT_CTOR(scalar_boolean_and_op)
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE bool operator() (const bool& a, const bool& b) const { return a && b; }
};
template<> struct functor_traits<scalar_boolean_and_op> {
enum {
Cost = NumTraits<bool>::AddCost,
PacketAccess = false
};
};
/** \internal
* \brief Template functor to compute the or of two booleans
*
* \sa class CwiseBinaryOp, ArrayBase::operator||
*/
struct scalar_boolean_or_op {
EIGEN_EMPTY_STRUCT_CTOR(scalar_boolean_or_op)
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE bool operator() (const bool& a, const bool& b) const { return a || b; }
};
template<> struct functor_traits<scalar_boolean_or_op> {
enum {
Cost = NumTraits<bool>::AddCost,
PacketAccess = false
};
};
/** \internal
* \brief Template functor to compute the xor of two booleans
*
* \sa class CwiseBinaryOp, ArrayBase::operator^
*/
struct scalar_boolean_xor_op {
EIGEN_EMPTY_STRUCT_CTOR(scalar_boolean_xor_op)
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE bool operator() (const bool& a, const bool& b) const { return a ^ b; }
};
template<> struct functor_traits<scalar_boolean_xor_op> {
enum {
Cost = NumTraits<bool>::AddCost,
PacketAccess = false
};
};
/** \internal
* \brief Template functor to compute the absolute difference of two scalars
*
* \sa class CwiseBinaryOp, MatrixBase::absolute_difference
*/
template<typename LhsScalar,typename RhsScalar>
struct scalar_absolute_difference_op : binary_op_base<LhsScalar,RhsScalar>
{
typedef typename ScalarBinaryOpTraits<LhsScalar,RhsScalar,scalar_absolute_difference_op>::ReturnType result_type;
#ifndef EIGEN_SCALAR_BINARY_OP_PLUGIN
EIGEN_EMPTY_STRUCT_CTOR(scalar_absolute_difference_op)
#else
scalar_absolute_difference_op() {
EIGEN_SCALAR_BINARY_OP_PLUGIN
}
#endif
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const result_type operator() (const LhsScalar& a, const RhsScalar& b) const
{ return numext::absdiff(a,b); }
template<typename Packet>
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const Packet packetOp(const Packet& a, const Packet& b) const
{ return internal::pabsdiff(a,b); }
};
template<typename LhsScalar,typename RhsScalar>
struct functor_traits<scalar_absolute_difference_op<LhsScalar,RhsScalar> > {
enum {
Cost = (NumTraits<LhsScalar>::AddCost+NumTraits<RhsScalar>::AddCost)/2,
PacketAccess = is_same<LhsScalar,RhsScalar>::value && packet_traits<LhsScalar>::HasAbsDiff
};
};
//---------- binary functors bound to a constant, thus appearing as a unary functor ----------
// The following two classes permits to turn any binary functor into a unary one with one argument bound to a constant value.
// They are analogues to std::binder1st/binder2nd but with the following differences:
// - they are compatible with packetOp
// - they are portable across C++ versions (the std::binder* are deprecated in C++11)
template<typename BinaryOp> struct bind1st_op : BinaryOp {
typedef typename BinaryOp::first_argument_type first_argument_type;
typedef typename BinaryOp::second_argument_type second_argument_type;
typedef typename BinaryOp::result_type result_type;
EIGEN_DEVICE_FUNC explicit bind1st_op(const first_argument_type &val) : m_value(val) {}
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const result_type operator() (const second_argument_type& b) const { return BinaryOp::operator()(m_value,b); }
template<typename Packet>
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const Packet packetOp(const Packet& b) const
{ return BinaryOp::packetOp(internal::pset1<Packet>(m_value), b); }
first_argument_type m_value;
};
template<typename BinaryOp> struct functor_traits<bind1st_op<BinaryOp> > : functor_traits<BinaryOp> {};
template<typename BinaryOp> struct bind2nd_op : BinaryOp {
typedef typename BinaryOp::first_argument_type first_argument_type;
typedef typename BinaryOp::second_argument_type second_argument_type;
typedef typename BinaryOp::result_type result_type;
EIGEN_DEVICE_FUNC explicit bind2nd_op(const second_argument_type &val) : m_value(val) {}
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const result_type operator() (const first_argument_type& a) const { return BinaryOp::operator()(a,m_value); }
template<typename Packet>
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const Packet packetOp(const Packet& a) const
{ return BinaryOp::packetOp(a,internal::pset1<Packet>(m_value)); }
second_argument_type m_value;
};
template<typename BinaryOp> struct functor_traits<bind2nd_op<BinaryOp> > : functor_traits<BinaryOp> {};
} // end namespace internal
} // end namespace Eigen
#endif // EIGEN_BINARY_FUNCTORS_H
// clang-format on
// This file is part of Eigen, a lightweight C++ template library
// for linear algebra.
//
// Copyright (C) 2008-2015 Gael Guennebaud <gael.guennebaud@inria.fr>
// Copyright (C) 2006-2008 Benoit Jacob <jacob.benoit.1@gmail.com>
//
// 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/.
// clang-format off
#ifndef EIGEN_META_H
#define EIGEN_META_H
#if defined(EIGEN_GPU_COMPILE_PHASE)
#include <cfloat>
#if defined(EIGEN_CUDA_ARCH)
#include <math_constants.h>
#endif
#if defined(EIGEN_HIP_DEVICE_COMPILE)
#include "Eigen/src/Core/arch/HIP/hcc/math_constants.h"
#endif
#endif
#if EIGEN_COMP_ICC>=1600 && __cplusplus >= 201103L
#include <cstdint>
#endif
namespace Eigen {
typedef EIGEN_DEFAULT_DENSE_INDEX_TYPE DenseIndex;
/**
* \brief The Index type as used for the API.
* \details To change this, \c \#define the preprocessor symbol \c EIGEN_DEFAULT_DENSE_INDEX_TYPE.
* \sa \blank \ref TopicPreprocessorDirectives, StorageIndex.
*/
typedef EIGEN_DEFAULT_DENSE_INDEX_TYPE Index;
namespace internal {
/** \internal
* \file Meta.h
* This file contains generic metaprogramming classes which are not specifically related to Eigen.
* \note In case you wonder, yes we're aware that Boost already provides all these features,
* we however don't want to add a dependency to Boost.
*/
// Only recent versions of ICC complain about using ptrdiff_t to hold pointers,
// and older versions do not provide *intptr_t types.
#if EIGEN_COMP_ICC>=1600 && __cplusplus >= 201103L
typedef std::intptr_t IntPtr;
typedef std::uintptr_t UIntPtr;
#else
typedef std::ptrdiff_t IntPtr;
typedef std::size_t UIntPtr;
#endif
struct true_type { enum { value = 1 }; };
struct false_type { enum { value = 0 }; };
template<bool Condition>
struct bool_constant;
template<>
struct bool_constant<true> : true_type {};
template<>
struct bool_constant<false> : false_type {};
template<bool Condition, typename Then, typename Else>
struct conditional { typedef Then type; };
template<typename Then, typename Else>
struct conditional <false, Then, Else> { typedef Else type; };
template<typename T> struct remove_reference { typedef T type; };
template<typename T> struct remove_reference<T&> { typedef T type; };
template<typename T> struct remove_pointer { typedef T type; };
template<typename T> struct remove_pointer<T*> { typedef T type; };
template<typename T> struct remove_pointer<T*const> { typedef T type; };
template <class T> struct remove_const { typedef T type; };
template <class T> struct remove_const<const T> { typedef T type; };
template <class T> struct remove_const<const T[]> { typedef T type[]; };
template <class T, unsigned int Size> struct remove_const<const T[Size]> { typedef T type[Size]; };
template<typename T> struct remove_all { typedef T type; };
template<typename T> struct remove_all<const T> { typedef typename remove_all<T>::type type; };
template<typename T> struct remove_all<T const&> { typedef typename remove_all<T>::type type; };
template<typename T> struct remove_all<T&> { typedef typename remove_all<T>::type type; };
template<typename T> struct remove_all<T const*> { typedef typename remove_all<T>::type type; };
template<typename T> struct remove_all<T*> { typedef typename remove_all<T>::type type; };
template<typename T> struct is_arithmetic { enum { value = false }; };
template<> struct is_arithmetic<float> { enum { value = true }; };
template<> struct is_arithmetic<double> { enum { value = true }; };
template<> struct is_arithmetic<long double> { enum { value = true }; };
template<> struct is_arithmetic<bool> { enum { value = true }; };
template<> struct is_arithmetic<char> { enum { value = true }; };
template<> struct is_arithmetic<signed char> { enum { value = true }; };
template<> struct is_arithmetic<unsigned char> { enum { value = true }; };
template<> struct is_arithmetic<signed short> { enum { value = true }; };
template<> struct is_arithmetic<unsigned short>{ enum { value = true }; };
template<> struct is_arithmetic<signed int> { enum { value = true }; };
template<> struct is_arithmetic<unsigned int> { enum { value = true }; };
template<> struct is_arithmetic<signed long> { enum { value = true }; };
template<> struct is_arithmetic<unsigned long> { enum { value = true }; };
template<typename T, typename U> struct is_same { enum { value = 0 }; };
template<typename T> struct is_same<T,T> { enum { value = 1 }; };
template< class T >
struct is_void : is_same<void, typename remove_const<T>::type> {};
#if EIGEN_HAS_CXX11
template<> struct is_arithmetic<signed long long> { enum { value = true }; };
template<> struct is_arithmetic<unsigned long long> { enum { value = true }; };
using std::is_integral;
#else
template<typename T> struct is_integral { enum { value = false }; };
template<> struct is_integral<bool> { enum { value = true }; };
template<> struct is_integral<char> { enum { value = true }; };
template<> struct is_integral<signed char> { enum { value = true }; };
template<> struct is_integral<unsigned char> { enum { value = true }; };
template<> struct is_integral<signed short> { enum { value = true }; };
template<> struct is_integral<unsigned short> { enum { value = true }; };
template<> struct is_integral<signed int> { enum { value = true }; };
template<> struct is_integral<unsigned int> { enum { value = true }; };
template<> struct is_integral<signed long> { enum { value = true }; };
template<> struct is_integral<unsigned long> { enum { value = true }; };
#if EIGEN_COMP_MSVC
template<> struct is_integral<signed __int64> { enum { value = true }; };
template<> struct is_integral<unsigned __int64> { enum { value = true }; };
#endif
#endif
#if EIGEN_HAS_CXX11
using std::make_unsigned;
#else
// TODO: Possibly improve this implementation of make_unsigned.
// It is currently used only by
// template<typename Scalar> struct random_default_impl<Scalar, false, true>.
template<typename> struct make_unsigned;
template<> struct make_unsigned<char> { typedef unsigned char type; };
template<> struct make_unsigned<signed char> { typedef unsigned char type; };
template<> struct make_unsigned<unsigned char> { typedef unsigned char type; };
template<> struct make_unsigned<signed short> { typedef unsigned short type; };
template<> struct make_unsigned<unsigned short> { typedef unsigned short type; };
template<> struct make_unsigned<signed int> { typedef unsigned int type; };
template<> struct make_unsigned<unsigned int> { typedef unsigned int type; };
template<> struct make_unsigned<signed long> { typedef unsigned long type; };
template<> struct make_unsigned<unsigned long> { typedef unsigned long type; };
#if EIGEN_COMP_MSVC
template<> struct make_unsigned<signed __int64> { typedef unsigned __int64 type; };
template<> struct make_unsigned<unsigned __int64> { typedef unsigned __int64 type; };
#endif
#endif
template <typename T> struct add_const { typedef const T type; };
template <typename T> struct add_const<T&> { typedef T& type; };
template <typename T> struct is_const { enum { value = 0 }; };
template <typename T> struct is_const<T const> { enum { value = 1 }; };
template<typename T> struct add_const_on_value_type { typedef const T type; };
template<typename T> struct add_const_on_value_type<T&> { typedef T const& type; };
template<typename T> struct add_const_on_value_type<T*> { typedef T const* type; };
template<typename T> struct add_const_on_value_type<T* const> { typedef T const* const type; };
template<typename T> struct add_const_on_value_type<T const* const> { typedef T const* const type; };
#if EIGEN_HAS_CXX11
using std::is_convertible;
#else
template<typename From, typename To>
struct is_convertible_impl
{
private:
struct any_conversion
{
template <typename T> any_conversion(const volatile T&);
template <typename T> any_conversion(T&);
};
struct yes {int a[1];};
struct no {int a[2];};
template<typename T>
static yes test(T, int);
template<typename T>
static no test(any_conversion, ...);
public:
static typename internal::remove_reference<From>::type* ms_from;
#ifdef __INTEL_COMPILER
#pragma warning push
#pragma warning ( disable : 2259 )
#endif
enum { value = sizeof(test<To>(*ms_from, 0))==sizeof(yes) };
#ifdef __INTEL_COMPILER
#pragma warning pop
#endif
};
template<typename From, typename To>
struct is_convertible
{
enum { value = is_convertible_impl<From,To>::value };
};
template<typename T>
struct is_convertible<T,T&> { enum { value = false }; };
template<typename T>
struct is_convertible<const T,const T&> { enum { value = true }; };
#endif
/** \internal Allows to enable/disable an overload
* according to a compile time condition.
*/
template<bool Condition, typename T=void> struct enable_if;
template<typename T> struct enable_if<true,T>
{ typedef T type; };
#if defined(EIGEN_GPU_COMPILE_PHASE)
#if !defined(__FLT_EPSILON__)
#define __FLT_EPSILON__ FLT_EPSILON
#define __DBL_EPSILON__ DBL_EPSILON
#endif
namespace device {
template<typename T> struct numeric_limits
{
EIGEN_DEVICE_FUNC static T epsilon() { return 0; }
EIGEN_DEVICE_FUNC static T (max)() { assert(false && "Highest not supported for this type"); }
EIGEN_DEVICE_FUNC static T (min)() { assert(false && "Lowest not supported for this type"); }
EIGEN_DEVICE_FUNC static T infinity() { assert(false && "Infinity not supported for this type"); }
EIGEN_DEVICE_FUNC static T quiet_NaN() { assert(false && "quiet_NaN not supported for this type"); }
};
template<> struct numeric_limits<float>
{
EIGEN_DEVICE_FUNC
static float epsilon() { return __FLT_EPSILON__; }
EIGEN_DEVICE_FUNC
static float (max)() {
#if defined(EIGEN_CUDA_ARCH)
return CUDART_MAX_NORMAL_F;
#else
return HIPRT_MAX_NORMAL_F;
#endif
}
EIGEN_DEVICE_FUNC
static float (min)() { return FLT_MIN; }
EIGEN_DEVICE_FUNC
static float infinity() {
#if defined(EIGEN_CUDA_ARCH)
return CUDART_INF_F;
#else
return HIPRT_INF_F;
#endif
}
EIGEN_DEVICE_FUNC
static float quiet_NaN() {
#if defined(EIGEN_CUDA_ARCH)
return CUDART_NAN_F;
#else
return HIPRT_NAN_F;
#endif
}
};
template<> struct numeric_limits<double>
{
EIGEN_DEVICE_FUNC
static double epsilon() { return __DBL_EPSILON__; }
EIGEN_DEVICE_FUNC
static double (max)() { return DBL_MAX; }
EIGEN_DEVICE_FUNC
static double (min)() { return DBL_MIN; }
EIGEN_DEVICE_FUNC
static double infinity() {
#if defined(EIGEN_CUDA_ARCH)
return CUDART_INF;
#else
return HIPRT_INF;
#endif
}
EIGEN_DEVICE_FUNC
static double quiet_NaN() {
#if defined(EIGEN_CUDA_ARCH)
return CUDART_NAN;
#else
return HIPRT_NAN;
#endif
}
};
template<> struct numeric_limits<int>
{
EIGEN_DEVICE_FUNC
static int epsilon() { return 0; }
EIGEN_DEVICE_FUNC
static int (max)() { return INT_MAX; }
EIGEN_DEVICE_FUNC
static int (min)() { return INT_MIN; }
};
template<> struct numeric_limits<unsigned int>
{
EIGEN_DEVICE_FUNC
static unsigned int epsilon() { return 0; }
EIGEN_DEVICE_FUNC
static unsigned int (max)() { return UINT_MAX; }
EIGEN_DEVICE_FUNC
static unsigned int (min)() { return 0; }
};
template<> struct numeric_limits<long>
{
EIGEN_DEVICE_FUNC
static long epsilon() { return 0; }
EIGEN_DEVICE_FUNC
static long (max)() { return LONG_MAX; }
EIGEN_DEVICE_FUNC
static long (min)() { return LONG_MIN; }
};
template<> struct numeric_limits<unsigned long>
{
EIGEN_DEVICE_FUNC
static unsigned long epsilon() { return 0; }
EIGEN_DEVICE_FUNC
static unsigned long (max)() { return ULONG_MAX; }
EIGEN_DEVICE_FUNC
static unsigned long (min)() { return 0; }
};
template<> struct numeric_limits<long long>
{
EIGEN_DEVICE_FUNC
static long long epsilon() { return 0; }
EIGEN_DEVICE_FUNC
static long long (max)() { return LLONG_MAX; }
EIGEN_DEVICE_FUNC
static long long (min)() { return LLONG_MIN; }
};
template<> struct numeric_limits<unsigned long long>
{
EIGEN_DEVICE_FUNC
static unsigned long long epsilon() { return 0; }
EIGEN_DEVICE_FUNC
static unsigned long long (max)() { return ULLONG_MAX; }
EIGEN_DEVICE_FUNC
static unsigned long long (min)() { return 0; }
};
}
#endif
/** \internal
* A base class do disable default copy ctor and copy assignment operator.
*/
class noncopyable
{
EIGEN_DEVICE_FUNC noncopyable(const noncopyable&);
EIGEN_DEVICE_FUNC const noncopyable& operator=(const noncopyable&);
protected:
EIGEN_DEVICE_FUNC noncopyable() {}
EIGEN_DEVICE_FUNC ~noncopyable() {}
};
/** \internal
* Provides access to the number of elements in the object of as a compile-time constant expression.
* It "returns" Eigen::Dynamic if the size cannot be resolved at compile-time (default).
*
* Similar to std::tuple_size, but more general.
*
* It currently supports:
* - any types T defining T::SizeAtCompileTime
* - plain C arrays as T[N]
* - std::array (c++11)
* - some internal types such as SingleRange and AllRange
*
* The second template parameter eases SFINAE-based specializations.
*/
template<typename T, typename EnableIf = void> struct array_size {
enum { value = Dynamic };
};
template<typename T> struct array_size<T,typename internal::enable_if<((T::SizeAtCompileTime&0)==0)>::type> {
enum { value = T::SizeAtCompileTime };
};
template<typename T, int N> struct array_size<const T (&)[N]> {
enum { value = N };
};
template<typename T, int N> struct array_size<T (&)[N]> {
enum { value = N };
};
#if EIGEN_HAS_CXX11
template<typename T, std::size_t N> struct array_size<const std::array<T,N> > {
enum { value = N };
};
template<typename T, std::size_t N> struct array_size<std::array<T,N> > {
enum { value = N };
};
#endif
/** \internal
* Analogue of the std::size free function.
* It returns the size of the container or view \a x of type \c T
*
* It currently supports:
* - any types T defining a member T::size() const
* - plain C arrays as T[N]
*
*/
template<typename T>
Index size(const T& x) { return x.size(); }
template<typename T,std::size_t N>
Index size(const T (&) [N]) { return N; }
/** \internal
* Convenient struct to get the result type of a unary or binary functor.
*
* It supports both the current STL mechanism (using the result_type member) as well as
* upcoming next STL generation (using a templated result member).
* If none of these members is provided, then the type of the first argument is returned. FIXME, that behavior is a pretty bad hack.
*/
#if EIGEN_HAS_STD_RESULT_OF
template<typename T> struct result_of {
typedef typename std::result_of<T>::type type1;
typedef typename remove_all<type1>::type type;
};
#else
template<typename T> struct result_of { };
struct has_none {int a[1];};
struct has_std_result_type {int a[2];};
struct has_tr1_result {int a[3];};
template<typename Func, typename ArgType, int SizeOf=sizeof(has_none)>
struct unary_result_of_select {typedef typename internal::remove_all<ArgType>::type type;};
template<typename Func, typename ArgType>
struct unary_result_of_select<Func, ArgType, sizeof(has_std_result_type)> {typedef typename Func::result_type type;};
template<typename Func, typename ArgType>
struct unary_result_of_select<Func, ArgType, sizeof(has_tr1_result)> {typedef typename Func::template result<Func(ArgType)>::type type;};
template<typename Func, typename ArgType>
struct result_of<Func(ArgType)> {
template<typename T>
static has_std_result_type testFunctor(T const *, typename T::result_type const * = 0);
template<typename T>
static has_tr1_result testFunctor(T const *, typename T::template result<T(ArgType)>::type const * = 0);
static has_none testFunctor(...);
// note that the following indirection is needed for gcc-3.3
enum {FunctorType = sizeof(testFunctor(static_cast<Func*>(0)))};
typedef typename unary_result_of_select<Func, ArgType, FunctorType>::type type;
};
template<typename Func, typename ArgType0, typename ArgType1, int SizeOf=sizeof(has_none)>
struct binary_result_of_select {typedef typename internal::remove_all<ArgType0>::type type;};
template<typename Func, typename ArgType0, typename ArgType1>
struct binary_result_of_select<Func, ArgType0, ArgType1, sizeof(has_std_result_type)>
{typedef typename Func::result_type type;};
template<typename Func, typename ArgType0, typename ArgType1>
struct binary_result_of_select<Func, ArgType0, ArgType1, sizeof(has_tr1_result)>
{typedef typename Func::template result<Func(ArgType0,ArgType1)>::type type;};
template<typename Func, typename ArgType0, typename ArgType1>
struct result_of<Func(ArgType0,ArgType1)> {
template<typename T>
static has_std_result_type testFunctor(T const *, typename T::result_type const * = 0);
template<typename T>
static has_tr1_result testFunctor(T const *, typename T::template result<T(ArgType0,ArgType1)>::type const * = 0);
static has_none testFunctor(...);
// note that the following indirection is needed for gcc-3.3
enum {FunctorType = sizeof(testFunctor(static_cast<Func*>(0)))};
typedef typename binary_result_of_select<Func, ArgType0, ArgType1, FunctorType>::type type;
};
template<typename Func, typename ArgType0, typename ArgType1, typename ArgType2, int SizeOf=sizeof(has_none)>
struct ternary_result_of_select {typedef typename internal::remove_all<ArgType0>::type type;};
template<typename Func, typename ArgType0, typename ArgType1, typename ArgType2>
struct ternary_result_of_select<Func, ArgType0, ArgType1, ArgType2, sizeof(has_std_result_type)>
{typedef typename Func::result_type type;};
template<typename Func, typename ArgType0, typename ArgType1, typename ArgType2>
struct ternary_result_of_select<Func, ArgType0, ArgType1, ArgType2, sizeof(has_tr1_result)>
{typedef typename Func::template result<Func(ArgType0,ArgType1,ArgType2)>::type type;};
template<typename Func, typename ArgType0, typename ArgType1, typename ArgType2>
struct result_of<Func(ArgType0,ArgType1,ArgType2)> {
template<typename T>
static has_std_result_type testFunctor(T const *, typename T::result_type const * = 0);
template<typename T>
static has_tr1_result testFunctor(T const *, typename T::template result<T(ArgType0,ArgType1,ArgType2)>::type const * = 0);
static has_none testFunctor(...);
// note that the following indirection is needed for gcc-3.3
enum {FunctorType = sizeof(testFunctor(static_cast<Func*>(0)))};
typedef typename ternary_result_of_select<Func, ArgType0, ArgType1, ArgType2, FunctorType>::type type;
};
#endif
struct meta_yes { char a[1]; };
struct meta_no { char a[2]; };
// Check whether T::ReturnType does exist
template <typename T>
struct has_ReturnType
{
template <typename C> static meta_yes testFunctor(C const *, typename C::ReturnType const * = 0);
template <typename C> static meta_no testFunctor(...);
enum { value = sizeof(testFunctor<T>(static_cast<T*>(0))) == sizeof(meta_yes) };
};
template<typename T> const T* return_ptr();
template <typename T, typename IndexType=Index>
struct has_nullary_operator
{
template <typename C> static meta_yes testFunctor(C const *,typename enable_if<(sizeof(return_ptr<C>()->operator()())>0)>::type * = 0);
static meta_no testFunctor(...);
enum { value = sizeof(testFunctor(static_cast<T*>(0))) == sizeof(meta_yes) };
};
template <typename T, typename IndexType=Index>
struct has_unary_operator
{
template <typename C> static meta_yes testFunctor(C const *,typename enable_if<(sizeof(return_ptr<C>()->operator()(IndexType(0)))>0)>::type * = 0);
static meta_no testFunctor(...);
enum { value = sizeof(testFunctor(static_cast<T*>(0))) == sizeof(meta_yes) };
};
template <typename T, typename IndexType=Index>
struct has_binary_operator
{
template <typename C> static meta_yes testFunctor(C const *,typename enable_if<(sizeof(return_ptr<C>()->operator()(IndexType(0),IndexType(0)))>0)>::type * = 0);
static meta_no testFunctor(...);
enum { value = sizeof(testFunctor(static_cast<T*>(0))) == sizeof(meta_yes) };
};
/** \internal In short, it computes int(sqrt(\a Y)) with \a Y an integer.
* Usage example: \code meta_sqrt<1023>::ret \endcode
*/
template<int Y,
int InfX = 0,
int SupX = ((Y==1) ? 1 : Y/2),
bool Done = ((SupX-InfX)<=1 ? true : ((SupX*SupX <= Y) && ((SupX+1)*(SupX+1) > Y))) >
// use ?: instead of || just to shut up a stupid gcc 4.3 warning
class meta_sqrt
{
enum {
MidX = (InfX+SupX)/2,
TakeInf = MidX*MidX > Y ? 1 : 0,
NewInf = int(TakeInf) ? InfX : int(MidX),
NewSup = int(TakeInf) ? int(MidX) : SupX
};
public:
enum { ret = meta_sqrt<Y,NewInf,NewSup>::ret };
};
template<int Y, int InfX, int SupX>
class meta_sqrt<Y, InfX, SupX, true> { public: enum { ret = (SupX*SupX <= Y) ? SupX : InfX }; };
/** \internal Computes the least common multiple of two positive integer A and B
* at compile-time. It implements a naive algorithm testing all multiples of A.
* It thus works better if A>=B.
*/
template<int A, int B, int K=1, bool Done = ((A*K)%B)==0>
struct meta_least_common_multiple
{
enum { ret = meta_least_common_multiple<A,B,K+1>::ret };
};
template<int A, int B, int K>
struct meta_least_common_multiple<A,B,K,true>
{
enum { ret = A*K };
};
/** \internal determines whether the product of two numeric types is allowed and what the return type is */
template<typename T, typename U> struct scalar_product_traits
{
enum { Defined = 0 };
};
// FIXME quick workaround around current limitation of result_of
// template<typename Scalar, typename ArgType0, typename ArgType1>
// struct result_of<scalar_product_op<Scalar>(ArgType0,ArgType1)> {
// typedef typename scalar_product_traits<typename remove_all<ArgType0>::type, typename remove_all<ArgType1>::type>::ReturnType type;
// };
/** \internal Obtains a POD type suitable to use as storage for an object of a size
* of at most Len bytes, aligned as specified by \c Align.
*/
template<unsigned Len, unsigned Align>
struct aligned_storage {
struct type {
EIGEN_ALIGN_TO_BOUNDARY(Align) unsigned char data[Len];
};
};
} // end namespace internal
namespace numext {
#if defined(EIGEN_GPU_COMPILE_PHASE)
template<typename T> EIGEN_DEVICE_FUNC void swap(T &a, T &b) { T tmp = b; b = a; a = tmp; }
#else
template<typename T> EIGEN_STRONG_INLINE void swap(T &a, T &b) { std::swap(a,b); }
#endif
#if defined(EIGEN_GPU_COMPILE_PHASE)
using internal::device::numeric_limits;
#else
using std::numeric_limits;
#endif
// Integer division with rounding up.
// T is assumed to be an integer type with a>=0, and b>0
template<typename T>
EIGEN_DEVICE_FUNC
T div_ceil(const T &a, const T &b)
{
return (a+b-1) / b;
}
// The aim of the following functions is to bypass -Wfloat-equal warnings
// when we really want a strict equality comparison on floating points.
template<typename X, typename Y> EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC
bool equal_strict(const X& x,const Y& y) { return x == y; }
#if !defined(EIGEN_GPU_COMPILE_PHASE) || (!defined(EIGEN_CUDA_ARCH) && defined(EIGEN_CONSTEXPR_ARE_DEVICE_FUNC))
template<> EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC
bool equal_strict(const float& x,const float& y) { return std::equal_to<float>()(x,y); }
template<> EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC
bool equal_strict(const double& x,const double& y) { return std::equal_to<double>()(x,y); }
#endif
template<typename X, typename Y> EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC
bool not_equal_strict(const X& x,const Y& y) { return x != y; }
#if !defined(EIGEN_GPU_COMPILE_PHASE) || (!defined(EIGEN_CUDA_ARCH) && defined(EIGEN_CONSTEXPR_ARE_DEVICE_FUNC))
template<> EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC
bool not_equal_strict(const float& x,const float& y) { return std::not_equal_to<float>()(x,y); }
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::uint8_t uint8_t;
typedef std::int8_t int8_t;
typedef std::uint16_t uint16_t;
typedef std::int16_t int16_t;
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 ::uint8_t uint8_t;
typedef ::int8_t int8_t;
typedef ::uint16_t uint16_t;
typedef ::int16_t int16_t;
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
// clang-format on
Markdown is supported
0% .
You are about to add 0 people to the discussion. Proceed with caution.
先完成此消息的编辑!
想要评论请 注册