diff options
Diffstat (limited to 'Eigen/src/Core/MathFunctions.h')
| -rw-r--r-- | Eigen/src/Core/MathFunctions.h | 171 |
1 files changed, 145 insertions, 26 deletions
diff --git a/Eigen/src/Core/MathFunctions.h b/Eigen/src/Core/MathFunctions.h index 48cf565fb..8e7dd2b73 100644 --- a/Eigen/src/Core/MathFunctions.h +++ b/Eigen/src/Core/MathFunctions.h @@ -23,10 +23,10 @@ double abs(double x) { return (fabs(x)); } float abs(float x) { return (fabsf(x)); } long double abs(long double x) { return (fabsl(x)); } #endif - + namespace internal { -/** \internal \struct global_math_functions_filtering_base +/** \internal \class global_math_functions_filtering_base * * What it does: * Defines a typedef 'type' as follows: @@ -496,7 +496,7 @@ template<typename Scalar, bool IsInteger> struct pow_default_impl { typedef Scalar retval; - static inline Scalar run(const Scalar& x, const Scalar& y) + static EIGEN_DEVICE_FUNC inline Scalar run(const Scalar& x, const Scalar& y) { EIGEN_USING_STD_MATH(pow); return pow(x, y); @@ -506,7 +506,7 @@ struct pow_default_impl template<typename Scalar> struct pow_default_impl<Scalar, true> { - static inline Scalar run(Scalar x, Scalar y) + static EIGEN_DEVICE_FUNC inline Scalar run(Scalar x, Scalar y) { Scalar res(1); eigen_assert(!NumTraits<Scalar>::IsSigned || y >= 0); @@ -704,11 +704,13 @@ EIGEN_DEVICE_FUNC typename internal::enable_if<(!internal::is_integral<T>::value)&&(!NumTraits<T>::IsComplex),bool>::type isfinite_impl(const T& x) { - #if EIGEN_USE_STD_FPCLASSIFY + #ifdef __CUDA_ARCH__ + return (::isfinite)(x); + #elif EIGEN_USE_STD_FPCLASSIFY using std::isfinite; return isfinite EIGEN_NOT_A_MACRO (x); #else - return x<NumTraits<T>::highest() && x>NumTraits<T>::lowest(); + return x<=NumTraits<T>::highest() && x>=NumTraits<T>::lowest(); #endif } @@ -717,7 +719,9 @@ EIGEN_DEVICE_FUNC typename internal::enable_if<(!internal::is_integral<T>::value)&&(!NumTraits<T>::IsComplex),bool>::type isinf_impl(const T& x) { - #if EIGEN_USE_STD_FPCLASSIFY + #ifdef __CUDA_ARCH__ + return (::isinf)(x); + #elif EIGEN_USE_STD_FPCLASSIFY using std::isinf; return isinf EIGEN_NOT_A_MACRO (x); #else @@ -730,7 +734,9 @@ EIGEN_DEVICE_FUNC typename internal::enable_if<(!internal::is_integral<T>::value)&&(!NumTraits<T>::IsComplex),bool>::type isnan_impl(const T& x) { - #if EIGEN_USE_STD_FPCLASSIFY + #ifdef __CUDA_ARCH__ + return (::isnan)(x); + #elif EIGEN_USE_STD_FPCLASSIFY using std::isnan; return isnan EIGEN_NOT_A_MACRO (x); #else @@ -748,9 +754,9 @@ template<typename T> EIGEN_DEVICE_FUNC bool isinf_msvc_helper(T x) } //MSVC defines a _isnan builtin function, but for double only -EIGEN_DEVICE_FUNC inline bool isnan_impl(const long double& x) { return _isnan(x); } -EIGEN_DEVICE_FUNC inline bool isnan_impl(const double& x) { return _isnan(x); } -EIGEN_DEVICE_FUNC inline bool isnan_impl(const float& x) { return _isnan(x); } +EIGEN_DEVICE_FUNC inline bool isnan_impl(const long double& x) { return _isnan(x)!=0; } +EIGEN_DEVICE_FUNC inline bool isnan_impl(const double& x) { return _isnan(x)!=0; } +EIGEN_DEVICE_FUNC inline bool isnan_impl(const float& x) { return _isnan(x)!=0; } EIGEN_DEVICE_FUNC inline bool isinf_impl(const long double& x) { return isinf_msvc_helper(x); } EIGEN_DEVICE_FUNC inline bool isinf_impl(const double& x) { return isinf_msvc_helper(x); } @@ -780,9 +786,9 @@ template<> EIGEN_TMP_NOOPT_ATTRIB bool isinf_impl(const long double& x) { return #endif // The following overload are defined at the end of this file -template<typename T> bool isfinite_impl(const std::complex<T>& x); -template<typename T> bool isnan_impl(const std::complex<T>& x); -template<typename T> bool isinf_impl(const std::complex<T>& x); +template<typename T> EIGEN_DEVICE_FUNC bool isfinite_impl(const std::complex<T>& x); +template<typename T> EIGEN_DEVICE_FUNC bool isnan_impl(const std::complex<T>& x); +template<typename T> EIGEN_DEVICE_FUNC bool isinf_impl(const std::complex<T>& x); } // end namespace internal @@ -946,6 +952,14 @@ T (floor)(const T& x) return floor(x); } +#ifdef __CUDACC__ +template<> EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE +float floor(const float &x) { return ::floorf(x); } + +template<> EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE +double floor(const double &x) { return ::floor(x); } +#endif + template<typename T> EIGEN_DEVICE_FUNC T (ceil)(const T& x) @@ -954,8 +968,17 @@ T (ceil)(const T& x) return ceil(x); } -// Log base 2 for 32 bits positive integers. -// Conveniently returns 0 for x==0. +#ifdef __CUDACC__ +template<> EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE +float ceil(const float &x) { return ::ceilf(x); } + +template<> EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE +double ceil(const double &x) { return ::ceil(x); } +#endif + + +/** Log base 2 for 32 bits positive integers. + * Conveniently returns 0 for x==0. */ inline int log2(int x) { eigen_assert(x>=0); @@ -969,24 +992,122 @@ inline int log2(int x) return table[(v * 0x07C4ACDDU) >> 27]; } +/** \returns the square root of \a x. + * + * It is essentially equivalent to \code using std::sqrt; return sqrt(x); \endcode, + * but slightly faster for float/double and some compilers (e.g., gcc), thanks to + * specializations when SSE is enabled. + * + * It's usage is justified in performance critical functions, like norm/normalize. + */ +template<typename T> +EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE +T sqrt(const T &x) +{ + EIGEN_USING_STD_MATH(sqrt); + return sqrt(x); +} + +template<typename T> +EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE +T log(const T &x) { + EIGEN_USING_STD_MATH(log); + return log(x); +} + +#ifdef __CUDACC__ +template<> EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE +float log(const float &x) { return ::logf(x); } + +template<> EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE +double log(const double &x) { return ::log(x); } +#endif + +template<typename T> +EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE +T tan(const T &x) { + EIGEN_USING_STD_MATH(tan); + return tan(x); +} + +#ifdef __CUDACC__ +template<> EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE +float tan(const float &x) { return ::tanf(x); } + +template<> EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE +double tan(const double &x) { return ::tan(x); } +#endif + +template<typename T> +EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE +typename NumTraits<T>::Real abs(const T &x) { + EIGEN_USING_STD_MATH(abs); + return abs(x); +} + +#ifdef __CUDACC__ +template<> EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE +float abs(const float &x) { return ::fabsf(x); } + +template<> EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE +double abs(const double &x) { return ::fabs(x); } +#endif + +template<typename T> +EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE +T exp(const T &x) { + EIGEN_USING_STD_MATH(exp); + return exp(x); +} + +#ifdef __CUDACC__ +template<> EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE +float exp(const float &x) { return ::expf(x); } + +template<> EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE +double exp(const double &x) { return ::exp(x); } +#endif + + +template <typename T> +EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE +T fmod(const T& a, const T& b) { + EIGEN_USING_STD_MATH(floor); + return fmod(a, b); +} + +#ifdef __CUDACC__ +template <> +EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE +float fmod(const float& a, const float& b) { + return ::fmodf(a, b); +} + +template <> +EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE +double fmod(const double& a, const double& b) { + return ::fmod(a, b); +} +#endif + } // end namespace numext namespace internal { template<typename T> -bool isfinite_impl(const std::complex<T>& x) +EIGEN_DEVICE_FUNC bool isfinite_impl(const std::complex<T>& x) { return (numext::isfinite)(numext::real(x)) && (numext::isfinite)(numext::imag(x)); } template<typename T> -bool isnan_impl(const std::complex<T>& x) +EIGEN_DEVICE_FUNC bool isnan_impl(const std::complex<T>& x) { return (numext::isnan)(numext::real(x)) || (numext::isnan)(numext::imag(x)); } template<typename T> -bool isinf_impl(const std::complex<T>& x) +EIGEN_DEVICE_FUNC bool isinf_impl(const std::complex<T>& x) { return ((numext::isinf)(numext::real(x)) || (numext::isinf)(numext::imag(x))) && (!(numext::isnan)(x)); } @@ -1007,14 +1128,12 @@ struct scalar_fuzzy_default_impl<Scalar, false, false> template<typename OtherScalar> EIGEN_DEVICE_FUNC static inline bool isMuchSmallerThan(const Scalar& x, const OtherScalar& y, const RealScalar& prec) { - EIGEN_USING_STD_MATH(abs); - return abs(x) <= abs(y) * prec; + return numext::abs(x) <= numext::abs(y) * prec; } EIGEN_DEVICE_FUNC static inline bool isApprox(const Scalar& x, const Scalar& y, const RealScalar& prec) { - EIGEN_USING_STD_MATH(abs); - return abs(x - y) <= numext::mini(abs(x), abs(y)) * prec; + return numext::abs(x - y) <= numext::mini(numext::abs(x), numext::abs(y)) * prec; } EIGEN_DEVICE_FUNC static inline bool isApproxOrLessThan(const Scalar& x, const Scalar& y, const RealScalar& prec) @@ -1064,21 +1183,21 @@ struct scalar_fuzzy_impl : scalar_fuzzy_default_impl<Scalar, NumTraits<Scalar>:: template<typename Scalar, typename OtherScalar> EIGEN_DEVICE_FUNC inline bool isMuchSmallerThan(const Scalar& x, const OtherScalar& y, - typename NumTraits<Scalar>::Real precision = NumTraits<Scalar>::dummy_precision()) + const typename NumTraits<Scalar>::Real &precision = NumTraits<Scalar>::dummy_precision()) { return scalar_fuzzy_impl<Scalar>::template isMuchSmallerThan<OtherScalar>(x, y, precision); } template<typename Scalar> EIGEN_DEVICE_FUNC inline bool isApprox(const Scalar& x, const Scalar& y, - typename NumTraits<Scalar>::Real precision = NumTraits<Scalar>::dummy_precision()) + const typename NumTraits<Scalar>::Real &precision = NumTraits<Scalar>::dummy_precision()) { return scalar_fuzzy_impl<Scalar>::isApprox(x, y, precision); } template<typename Scalar> EIGEN_DEVICE_FUNC inline bool isApproxOrLessThan(const Scalar& x, const Scalar& y, - typename NumTraits<Scalar>::Real precision = NumTraits<Scalar>::dummy_precision()) + const typename NumTraits<Scalar>::Real &precision = NumTraits<Scalar>::dummy_precision()) { return scalar_fuzzy_impl<Scalar>::isApproxOrLessThan(x, y, precision); } |