diff options
author | Stanislaw Halik <sthalik@misaki.pl> | 2017-03-25 14:17:07 +0100 |
---|---|---|
committer | Stanislaw Halik <sthalik@misaki.pl> | 2017-03-25 14:17:07 +0100 |
commit | 35f7829af10c61e33dd2e2a7a015058e11a11ea0 (patch) | |
tree | 7135010dcf8fd0a49f3020d52112709bcb883bd6 /eigen/Eigen/src/Core/util/Macros.h | |
parent | 6e8724193e40a932faf9064b664b529e7301c578 (diff) |
update
Diffstat (limited to 'eigen/Eigen/src/Core/util/Macros.h')
-rw-r--r-- | eigen/Eigen/src/Core/util/Macros.h | 578 |
1 files changed, 441 insertions, 137 deletions
diff --git a/eigen/Eigen/src/Core/util/Macros.h b/eigen/Eigen/src/Core/util/Macros.h index 16c248c..14ec87d 100644 --- a/eigen/Eigen/src/Core/util/Macros.h +++ b/eigen/Eigen/src/Core/util/Macros.h @@ -1,7 +1,7 @@ // This file is part of Eigen, a lightweight C++ template library // for linear algebra. // -// Copyright (C) 2008-2010 Gael Guennebaud <gael.guennebaud@inria.fr> +// 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 @@ -12,26 +12,25 @@ #define EIGEN_MACROS_H #define EIGEN_WORLD_VERSION 3 -#define EIGEN_MAJOR_VERSION 2 -#define EIGEN_MINOR_VERSION 9 +#define EIGEN_MAJOR_VERSION 3 +#define EIGEN_MINOR_VERSION 90 #define EIGEN_VERSION_AT_LEAST(x,y,z) (EIGEN_WORLD_VERSION>x || (EIGEN_WORLD_VERSION>=x && \ (EIGEN_MAJOR_VERSION>y || (EIGEN_MAJOR_VERSION>=y && \ EIGEN_MINOR_VERSION>=z)))) - // Compiler identification, EIGEN_COMP_* /// \internal EIGEN_COMP_GNUC set to 1 for all compilers compatible with GCC #ifdef __GNUC__ - #define EIGEN_COMP_GNUC 1 + #define EIGEN_COMP_GNUC (__GNUC__*10+__GNUC_MINOR__) #else #define EIGEN_COMP_GNUC 0 #endif -/// \internal EIGEN_COMP_CLANG set to 1 if the compiler is clang (alias for __clang__) +/// \internal EIGEN_COMP_CLANG set to major+minor version (e.g., 307 for clang 3.7) if the compiler is clang #if defined(__clang__) - #define EIGEN_COMP_CLANG 1 + #define EIGEN_COMP_CLANG (__clang_major__*100+__clang_minor__) #else #define EIGEN_COMP_CLANG 0 #endif @@ -72,8 +71,17 @@ #define EIGEN_COMP_MSVC 0 #endif -/// \internal EIGEN_COMP_MSVC_STRICT set to 1 if the compiler is really Microsoft Visual C++ and not ,e.g., ICC -#if EIGEN_COMP_MSVC && !(EIGEN_COMP_ICC) +// For the record, here is a table summarizing the possible values for EIGEN_COMP_MSVC: +// name ver MSC_VER +// 2008 9 1500 +// 2010 10 1600 +// 2012 11 1700 +// 2013 12 1800 +// 2015 14 1900 +// "15" 15 1900 + +/// \internal EIGEN_COMP_MSVC_STRICT set to 1 if the compiler is really Microsoft Visual C++ and not ,e.g., ICC or clang-cl +#if EIGEN_COMP_MSVC && !(EIGEN_COMP_ICC || EIGEN_COMP_LLVM || EIGEN_COMP_CLANG) #define EIGEN_COMP_MSVC_STRICT _MSC_VER #else #define EIGEN_COMP_MSVC_STRICT 0 @@ -100,9 +108,16 @@ #define EIGEN_COMP_ARM 0 #endif +/// \internal EIGEN_COMP_ARM set to 1 if the compiler is ARM Compiler +#if defined(__EMSCRIPTEN__) + #define EIGEN_COMP_EMSCRIPTEN 1 +#else + #define EIGEN_COMP_EMSCRIPTEN 0 +#endif + /// \internal EIGEN_GNUC_STRICT set to 1 if the compiler is really GCC and not a compatible compiler (e.g., ICC, clang, mingw, etc.) -#if EIGEN_COMP_GNUC && !(EIGEN_COMP_CLANG || EIGEN_COMP_ICC || EIGEN_COMP_MINGW || EIGEN_COMP_PGI || EIGEN_COMP_IBM || EIGEN_COMP_ARM ) +#if EIGEN_COMP_GNUC && !(EIGEN_COMP_CLANG || EIGEN_COMP_ICC || EIGEN_COMP_MINGW || EIGEN_COMP_PGI || EIGEN_COMP_IBM || EIGEN_COMP_ARM || EIGEN_COMP_EMSCRIPTEN) #define EIGEN_COMP_GNUC_STRICT 1 #else #define EIGEN_COMP_GNUC_STRICT 0 @@ -299,91 +314,176 @@ #endif -#if EIGEN_GNUC_AT_MOST(4,3) && !defined(__clang__) + +#if EIGEN_GNUC_AT_MOST(4,3) && !EIGEN_COMP_CLANG // see bug 89 #define EIGEN_SAFE_TO_USE_STANDARD_ASSERT_MACRO 0 #else #define EIGEN_SAFE_TO_USE_STANDARD_ASSERT_MACRO 1 #endif -// 16 byte alignment is only useful for vectorization. Since it affects the ABI, we need to enable -// 16 byte alignment on all platforms where vectorization might be enabled. In theory we could always -// enable alignment, but it can be a cause of problems on some platforms, so we just disable it in -// certain common platform (compiler+architecture combinations) to avoid these problems. -// Only static alignment is really problematic (relies on nonstandard compiler extensions that don't -// work everywhere, for example don't work on GCC/ARM), try to keep heap alignment even -// when we have to disable static alignment. -#if defined(__GNUC__) && !(defined(__i386__) || defined(__x86_64__) || defined(__powerpc__) || defined(__ppc__) || defined(__ia64__)) -#define EIGEN_GCC_AND_ARCH_DOESNT_WANT_STACK_ALIGNMENT 1 +// This macro can be used to prevent from macro expansion, e.g.: +// std::max EIGEN_NOT_A_MACRO(a,b) +#define EIGEN_NOT_A_MACRO + +#ifdef EIGEN_DEFAULT_TO_ROW_MAJOR +#define EIGEN_DEFAULT_MATRIX_STORAGE_ORDER_OPTION Eigen::RowMajor #else -#define EIGEN_GCC_AND_ARCH_DOESNT_WANT_STACK_ALIGNMENT 0 +#define EIGEN_DEFAULT_MATRIX_STORAGE_ORDER_OPTION Eigen::ColMajor #endif -// static alignment is completely disabled with GCC 3, Sun Studio, and QCC/QNX -#if !EIGEN_GCC_AND_ARCH_DOESNT_WANT_STACK_ALIGNMENT \ - && !EIGEN_GCC3_OR_OLDER \ - && !defined(__SUNPRO_CC) \ - && !defined(__QNXNTO__) - #define EIGEN_ARCH_WANTS_STACK_ALIGNMENT 1 -#else - #define EIGEN_ARCH_WANTS_STACK_ALIGNMENT 0 +#ifndef EIGEN_DEFAULT_DENSE_INDEX_TYPE +#define EIGEN_DEFAULT_DENSE_INDEX_TYPE std::ptrdiff_t #endif -#ifdef EIGEN_DONT_ALIGN - #ifndef EIGEN_DONT_ALIGN_STATICALLY - #define EIGEN_DONT_ALIGN_STATICALLY - #endif - #define EIGEN_ALIGN 0 +// Cross compiler wrapper around LLVM's __has_builtin +#ifdef __has_builtin +# define EIGEN_HAS_BUILTIN(x) __has_builtin(x) #else - #define EIGEN_ALIGN 1 +# define EIGEN_HAS_BUILTIN(x) 0 #endif -// EIGEN_ALIGN_STATICALLY is the true test whether we want to align arrays on the stack or not. It takes into account both the user choice to explicitly disable -// alignment (EIGEN_DONT_ALIGN_STATICALLY) and the architecture config (EIGEN_ARCH_WANTS_STACK_ALIGNMENT). Henceforth, only EIGEN_ALIGN_STATICALLY should be used. -#if EIGEN_ARCH_WANTS_STACK_ALIGNMENT && !defined(EIGEN_DONT_ALIGN_STATICALLY) - #define EIGEN_ALIGN_STATICALLY 1 -#else - #define EIGEN_ALIGN_STATICALLY 0 - #ifndef EIGEN_DISABLE_UNALIGNED_ARRAY_ASSERT - #define EIGEN_DISABLE_UNALIGNED_ARRAY_ASSERT - #endif +// A Clang feature extension to determine compiler features. +// We use it to determine 'cxx_rvalue_references' +#ifndef __has_feature +# define __has_feature(x) 0 #endif -#ifdef EIGEN_DEFAULT_TO_ROW_MAJOR -#define EIGEN_DEFAULT_MATRIX_STORAGE_ORDER_OPTION RowMajor +// Some old compilers do not support template specializations like: +// template<typename T,int N> void foo(const T x[N]); +#if !( EIGEN_COMP_CLANG && ((EIGEN_COMP_CLANG<309) || defined(__apple_build_version__)) || EIGEN_COMP_GNUC_STRICT && EIGEN_COMP_GNUC<49) +#define EIGEN_HAS_STATIC_ARRAY_TEMPLATE 1 #else -#define EIGEN_DEFAULT_MATRIX_STORAGE_ORDER_OPTION ColMajor +#define EIGEN_HAS_STATIC_ARRAY_TEMPLATE 0 #endif -#ifndef EIGEN_DEFAULT_DENSE_INDEX_TYPE -#define EIGEN_DEFAULT_DENSE_INDEX_TYPE std::ptrdiff_t +// Upperbound on the C++ version to use. +// Expected values are 03, 11, 14, 17, etc. +// By default, let's use an arbitrarily large C++ version. +#ifndef EIGEN_MAX_CPP_VER +#define EIGEN_MAX_CPP_VER 99 #endif -// A Clang feature extension to determine compiler features. -// We use it to determine 'cxx_rvalue_references' -#ifndef __has_feature -# define __has_feature(x) 0 +#if EIGEN_MAX_CPP_VER>=11 && (defined(__cplusplus) && (__cplusplus >= 201103L) || EIGEN_COMP_MSVC >= 1900) +#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) +#define EIGEN_HAS_CXX14 1 +#else +#define EIGEN_HAS_CXX14 0 #endif // Do we support r-value references? -#if (__has_feature(cxx_rvalue_references) || \ - (defined(__cplusplus) && __cplusplus >= 201103L) || \ - (defined(_MSC_VER) && _MSC_VER >= 1600)) - #define EIGEN_HAVE_RVALUE_REFERENCES +#ifndef EIGEN_HAS_RVALUE_REFERENCES +#if EIGEN_MAX_CPP_VER>=11 && \ + (__has_feature(cxx_rvalue_references) || \ + (defined(__cplusplus) && __cplusplus >= 201103L) || \ + (EIGEN_COMP_MSVC >= 1600)) + #define EIGEN_HAS_RVALUE_REFERENCES 1 +#else + #define EIGEN_HAS_RVALUE_REFERENCES 0 +#endif #endif +// Does the compiler support C99? +#ifndef EIGEN_HAS_C99_MATH +#if EIGEN_MAX_CPP_VER>=11 && \ + ((defined(__STDC_VERSION__) && (__STDC_VERSION__ >= 199901)) \ + || (defined(__GNUC__) && defined(_GLIBCXX_USE_C99)) \ + || (defined(_LIBCPP_VERSION) && !defined(_MSC_VER)) \ + || (EIGEN_COMP_MSVC >= 1900) || defined(__SYCL_DEVICE_ONLY__)) + #define EIGEN_HAS_C99_MATH 1 +#else + #define EIGEN_HAS_C99_MATH 0 +#endif +#endif -// Cross compiler wrapper around LLVM's __has_builtin -#ifdef __has_builtin -# define EIGEN_HAS_BUILTIN(x) __has_builtin(x) +// Does the compiler support result_of? +#ifndef EIGEN_HAS_STD_RESULT_OF +#if EIGEN_MAX_CPP_VER>=11 && ((__has_feature(cxx_lambdas) || (defined(__cplusplus) && __cplusplus >= 201103L))) +#define EIGEN_HAS_STD_RESULT_OF 1 #else -# define EIGEN_HAS_BUILTIN(x) 0 +#define EIGEN_HAS_STD_RESULT_OF 0 +#endif +#endif + +// Does the compiler support variadic templates? +#ifndef EIGEN_HAS_VARIADIC_TEMPLATES +#if EIGEN_MAX_CPP_VER>=11 && (__cplusplus > 199711L || EIGEN_COMP_MSVC >= 1900) \ + && (!defined(__NVCC__) || !EIGEN_ARCH_ARM_OR_ARM64 || (defined __CUDACC_VER__ && __CUDACC_VER__ >= 80000) ) + // ^^ Disable the use of variadic templates when compiling with versions of nvcc older than 8.0 on ARM devices: + // this prevents nvcc from crashing when compiling Eigen on Tegra X1 +#define EIGEN_HAS_VARIADIC_TEMPLATES 1 +#elif EIGEN_MAX_CPP_VER>=11 && (__cplusplus > 199711L || EIGEN_COMP_MSVC >= 1900) && defined(__SYCL_DEVICE_ONLY__) +#define EIGEN_HAS_VARIADIC_TEMPLATES 1 +#else +#define EIGEN_HAS_VARIADIC_TEMPLATES 0 +#endif +#endif + +// Does the compiler fully support const expressions? (as in c++14) +#ifndef EIGEN_HAS_CONSTEXPR + +#if defined(__CUDACC__) +// Const expressions are supported provided that c++11 is enabled and we're using either clang or nvcc 7.5 or above +#if EIGEN_MAX_CPP_VER>=14 && (__cplusplus > 199711L && defined(__CUDACC_VER__) && (EIGEN_COMP_CLANG || __CUDACC_VER__ >= 70500)) + #define EIGEN_HAS_CONSTEXPR 1 +#endif +#elif EIGEN_MAX_CPP_VER>=14 && (__has_feature(cxx_relaxed_constexpr) || (defined(__cplusplus) && __cplusplus >= 201402L) || \ + (EIGEN_GNUC_AT_LEAST(4,8) && (__cplusplus > 199711L)) || \ + (EIGEN_COMP_CLANG >= 306 && (__cplusplus > 199711L))) +#define EIGEN_HAS_CONSTEXPR 1 +#endif + +#ifndef EIGEN_HAS_CONSTEXPR +#define EIGEN_HAS_CONSTEXPR 0 +#endif + +#endif + +// Does the compiler support C++11 math? +// Let's be conservative and enable the default C++11 implementation only if we are sure it exists +#ifndef EIGEN_HAS_CXX11_MATH + #if EIGEN_MAX_CPP_VER>=11 && ((__cplusplus > 201103L) || (__cplusplus >= 201103L) && (EIGEN_COMP_GNUC_STRICT || EIGEN_COMP_CLANG || EIGEN_COMP_MSVC || EIGEN_COMP_ICC) \ + && (EIGEN_ARCH_i386_OR_x86_64) && (EIGEN_OS_GNULINUX || EIGEN_OS_WIN_STRICT || EIGEN_OS_MAC)) + #define EIGEN_HAS_CXX11_MATH 1 + #else + #define EIGEN_HAS_CXX11_MATH 0 + #endif +#endif + +// Does the compiler support proper C++11 containers? +#ifndef EIGEN_HAS_CXX11_CONTAINERS + #if EIGEN_MAX_CPP_VER>=11 && \ + ((__cplusplus > 201103L) \ + || ((__cplusplus >= 201103L) && (EIGEN_COMP_GNUC_STRICT || EIGEN_COMP_CLANG || EIGEN_COMP_ICC>=1400)) \ + || EIGEN_COMP_MSVC >= 1900) + #define EIGEN_HAS_CXX11_CONTAINERS 1 + #else + #define EIGEN_HAS_CXX11_CONTAINERS 0 + #endif +#endif + +// Does the compiler support C++11 noexcept? +#ifndef EIGEN_HAS_CXX11_NOEXCEPT + #if EIGEN_MAX_CPP_VER>=11 && \ + (__has_feature(cxx_noexcept) \ + || (__cplusplus > 201103L) \ + || ((__cplusplus >= 201103L) && (EIGEN_COMP_GNUC_STRICT || EIGEN_COMP_CLANG || EIGEN_COMP_ICC>=1400)) \ + || EIGEN_COMP_MSVC >= 1900) + #define EIGEN_HAS_CXX11_NOEXCEPT 1 + #else + #define EIGEN_HAS_CXX11_NOEXCEPT 0 + #endif #endif /** Allows to disable some optimizations which might affect the accuracy of the result. * Such optimization are enabled by default, and set EIGEN_FAST_MATH to 0 to disable them. * They currently include: - * - single precision Cwise::sin() and Cwise::cos() when SSE vectorization is enabled. + * - single precision ArrayBase::sin() and ArrayBase::cos() for SSE and AVX vectorization. */ #ifndef EIGEN_FAST_MATH #define EIGEN_FAST_MATH 1 @@ -395,6 +495,8 @@ #define EIGEN_CAT2(a,b) a ## b #define EIGEN_CAT(a,b) EIGEN_CAT2(a,b) +#define EIGEN_COMMA , + // convert a token to a string #define EIGEN_MAKESTRING2(a) #a #define EIGEN_MAKESTRING(a) EIGEN_MAKESTRING2(a) @@ -402,7 +504,7 @@ // EIGEN_STRONG_INLINE is a stronger version of the inline, using __forceinline on MSVC, // but it still doesn't use GCC's always_inline. This is useful in (common) situations where MSVC needs forceinline // but GCC is still doing fine with just inline. -#if (defined _MSC_VER) || (defined __INTEL_COMPILER) +#if EIGEN_COMP_MSVC || EIGEN_COMP_ICC #define EIGEN_STRONG_INLINE __forceinline #else #define EIGEN_STRONG_INLINE inline @@ -412,24 +514,25 @@ // attribute to maximize inlining. This should only be used when really necessary: in particular, // it uses __attribute__((always_inline)) on GCC, which most of the time is useless and can severely harm compile times. // FIXME with the always_inline attribute, -// gcc 3.4.x reports the following compilation error: +// gcc 3.4.x and 4.1 reports the following compilation error: // Eval.h:91: sorry, unimplemented: inlining failed in call to 'const Eigen::Eval<Derived> Eigen::MatrixBase<Scalar, Derived>::eval() const' // : function body not available -#if EIGEN_GNUC_AT_LEAST(4,0) +// See also bug 1367 +#if EIGEN_GNUC_AT_LEAST(4,2) #define EIGEN_ALWAYS_INLINE __attribute__((always_inline)) inline #else #define EIGEN_ALWAYS_INLINE EIGEN_STRONG_INLINE #endif -#if (defined __GNUC__) +#if EIGEN_COMP_GNUC #define EIGEN_DONT_INLINE __attribute__((noinline)) -#elif (defined _MSC_VER) +#elif EIGEN_COMP_MSVC #define EIGEN_DONT_INLINE __declspec(noinline) #else #define EIGEN_DONT_INLINE #endif -#if (defined __GNUC__) +#if EIGEN_COMP_GNUC #define EIGEN_PERMISSIVE_EXPR __extension__ #else #define EIGEN_PERMISSIVE_EXPR @@ -439,8 +542,8 @@ // - static is not very good because it prevents definitions from different object files to be merged. // So static causes the resulting linked executable to be bloated with multiple copies of the same function. // - inline is not perfect either as it unwantedly hints the compiler toward inlining the function. -#define EIGEN_DECLARE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS -#define EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS inline +#define EIGEN_DECLARE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_DEVICE_FUNC +#define EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_DEVICE_FUNC inline #ifdef NDEBUG # ifndef EIGEN_NO_DEBUG @@ -498,15 +601,15 @@ #endif #ifdef EIGEN_NO_DEBUG -#define EIGEN_ONLY_USED_FOR_DEBUG(x) (void)x +#define EIGEN_ONLY_USED_FOR_DEBUG(x) EIGEN_UNUSED_VARIABLE(x) #else #define EIGEN_ONLY_USED_FOR_DEBUG(x) #endif #ifndef EIGEN_NO_DEPRECATED_WARNING - #if (defined __GNUC__) + #if EIGEN_COMP_GNUC #define EIGEN_DEPRECATED __attribute__((deprecated)) - #elif (defined _MSC_VER) + #elif EIGEN_COMP_MSVC #define EIGEN_DEPRECATED __declspec(deprecated) #else #define EIGEN_DEPRECATED @@ -515,7 +618,7 @@ #define EIGEN_DEPRECATED #endif -#if (defined __GNUC__) +#if EIGEN_COMP_GNUC #define EIGEN_UNUSED __attribute__((unused)) #else #define EIGEN_UNUSED @@ -524,19 +627,41 @@ // Suppresses 'unused variable' warnings. namespace Eigen { namespace internal { - template<typename T> void ignore_unused_variable(const T&) {} + template<typename T> EIGEN_DEVICE_FUNC void ignore_unused_variable(const T&) {} } } #define EIGEN_UNUSED_VARIABLE(var) Eigen::internal::ignore_unused_variable(var); #if !defined(EIGEN_ASM_COMMENT) - #if (defined __GNUC__) && ( defined(__i386__) || defined(__x86_64__) ) + #if EIGEN_COMP_GNUC && (EIGEN_ARCH_i386_OR_x86_64 || EIGEN_ARCH_ARM_OR_ARM64) #define EIGEN_ASM_COMMENT(X) __asm__("#" X) #else #define EIGEN_ASM_COMMENT(X) #endif #endif + +#if EIGEN_COMP_MSVC + // NOTE MSVC often gives C4127 warnings with compiletime if statements. See bug 1362. + // This workaround is ugly, but it does the job. +# define EIGEN_CONST_CONDITIONAL(cond) (void)0, cond +#else +# define EIGEN_CONST_CONDITIONAL(cond) cond +#endif + +//------------------------------------------------------------------------------------------ +// Static and dynamic alignment control +// +// The main purpose of this section is to define EIGEN_MAX_ALIGN_BYTES and EIGEN_MAX_STATIC_ALIGN_BYTES +// as the maximal boundary in bytes on which dynamically and statically allocated data may be alignment respectively. +// The values of EIGEN_MAX_ALIGN_BYTES and EIGEN_MAX_STATIC_ALIGN_BYTES can be specified by the user. If not, +// a default value is automatically computed based on architecture, compiler, and OS. +// +// This section also defines macros EIGEN_ALIGN_TO_BOUNDARY(N) and the shortcuts EIGEN_ALIGN{8,16,32,_MAX} +// to be used to declare statically aligned buffers. +//------------------------------------------------------------------------------------------ + + /* EIGEN_ALIGN_TO_BOUNDARY(n) forces data to be n-byte aligned. This is used to satisfy SIMD requirements. * However, we do that EVEN if vectorization (EIGEN_VECTORIZE) is disabled, * so that vectorization doesn't affect binary compatibility. @@ -544,28 +669,149 @@ namespace Eigen { * If we made alignment depend on whether or not EIGEN_VECTORIZE is defined, it would be impossible to link * vectorized and non-vectorized code. */ -#if (defined __GNUC__) || (defined __PGI) || (defined __IBMCPP__) || (defined __ARMCC_VERSION) +#if (defined __CUDACC__) + #define EIGEN_ALIGN_TO_BOUNDARY(n) __align__(n) +#elif EIGEN_COMP_GNUC || EIGEN_COMP_PGI || EIGEN_COMP_IBM || EIGEN_COMP_ARM #define EIGEN_ALIGN_TO_BOUNDARY(n) __attribute__((aligned(n))) -#elif (defined _MSC_VER) +#elif EIGEN_COMP_MSVC #define EIGEN_ALIGN_TO_BOUNDARY(n) __declspec(align(n)) -#elif (defined __SUNPRO_CC) +#elif EIGEN_COMP_SUNCC // FIXME not sure about this one: #define EIGEN_ALIGN_TO_BOUNDARY(n) __attribute__((aligned(n))) #else #error Please tell me what is the equivalent of __attribute__((aligned(n))) 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 +#elif defined(EIGEN_VECTORIZE_AVX512) + // 64 bytes static alignmeent is preferred only if really required + #define EIGEN_IDEAL_MAX_ALIGN_BYTES 64 +#elif defined(__AVX__) + // 32 bytes static alignmeent is preferred only if really required + #define EIGEN_IDEAL_MAX_ALIGN_BYTES 32 +#else + #define EIGEN_IDEAL_MAX_ALIGN_BYTES 16 +#endif + + +// EIGEN_MIN_ALIGN_BYTES defines the minimal value for which the notion of explicit alignment makes sense +#define EIGEN_MIN_ALIGN_BYTES 16 + +// Defined the boundary (in bytes) on which the data needs to be aligned. Note +// that unless EIGEN_ALIGN is defined and not equal to 0, the data may not be +// aligned at all regardless of the value of this #define. + +#if (defined(EIGEN_DONT_ALIGN_STATICALLY) || defined(EIGEN_DONT_ALIGN)) && defined(EIGEN_MAX_STATIC_ALIGN_BYTES) && EIGEN_MAX_STATIC_ALIGN_BYTES>0 +#error EIGEN_MAX_STATIC_ALIGN_BYTES and EIGEN_DONT_ALIGN[_STATICALLY] are both defined with EIGEN_MAX_STATIC_ALIGN_BYTES!=0. Use EIGEN_MAX_STATIC_ALIGN_BYTES=0 as a synonym of EIGEN_DONT_ALIGN_STATICALLY. +#endif + +// EIGEN_DONT_ALIGN_STATICALLY and EIGEN_DONT_ALIGN are deprectated +// They imply EIGEN_MAX_STATIC_ALIGN_BYTES=0 +#if defined(EIGEN_DONT_ALIGN_STATICALLY) || defined(EIGEN_DONT_ALIGN) + #ifdef EIGEN_MAX_STATIC_ALIGN_BYTES + #undef EIGEN_MAX_STATIC_ALIGN_BYTES + #endif + #define EIGEN_MAX_STATIC_ALIGN_BYTES 0 +#endif + +#ifndef EIGEN_MAX_STATIC_ALIGN_BYTES + + // Try to automatically guess what is the best default value for EIGEN_MAX_STATIC_ALIGN_BYTES + + // 16 byte alignment is only useful for vectorization. Since it affects the ABI, we need to enable + // 16 byte alignment on all platforms where vectorization might be enabled. In theory we could always + // enable alignment, but it can be a cause of problems on some platforms, so we just disable it in + // certain common platform (compiler+architecture combinations) to avoid these problems. + // Only static alignment is really problematic (relies on nonstandard compiler extensions), + // try to keep heap alignment even when we have to disable static alignment. + #if EIGEN_COMP_GNUC && !(EIGEN_ARCH_i386_OR_x86_64 || EIGEN_ARCH_ARM_OR_ARM64 || EIGEN_ARCH_PPC || EIGEN_ARCH_IA64) + #define EIGEN_GCC_AND_ARCH_DOESNT_WANT_STACK_ALIGNMENT 1 + #elif EIGEN_ARCH_ARM_OR_ARM64 && EIGEN_COMP_GNUC_STRICT && EIGEN_GNUC_AT_MOST(4, 6) + // Old versions of GCC on ARM, at least 4.4, were once seen to have buggy static alignment support. + // Not sure which version fixed it, hopefully it doesn't affect 4.7, which is still somewhat in use. + // 4.8 and newer seem definitely unaffected. + #define EIGEN_GCC_AND_ARCH_DOESNT_WANT_STACK_ALIGNMENT 1 + #else + #define EIGEN_GCC_AND_ARCH_DOESNT_WANT_STACK_ALIGNMENT 0 + #endif + + // static alignment is completely disabled with GCC 3, Sun Studio, and QCC/QNX + #if !EIGEN_GCC_AND_ARCH_DOESNT_WANT_STACK_ALIGNMENT \ + && !EIGEN_GCC3_OR_OLDER \ + && !EIGEN_COMP_SUNCC \ + && !EIGEN_OS_QNX + #define EIGEN_ARCH_WANTS_STACK_ALIGNMENT 1 + #else + #define EIGEN_ARCH_WANTS_STACK_ALIGNMENT 0 + #endif + + #if EIGEN_ARCH_WANTS_STACK_ALIGNMENT + #define EIGEN_MAX_STATIC_ALIGN_BYTES EIGEN_IDEAL_MAX_ALIGN_BYTES + #else + #define EIGEN_MAX_STATIC_ALIGN_BYTES 0 + #endif + +#endif + +// If EIGEN_MAX_ALIGN_BYTES is defined, then it is considered as an upper bound for EIGEN_MAX_ALIGN_BYTES +#if defined(EIGEN_MAX_ALIGN_BYTES) && EIGEN_MAX_ALIGN_BYTES<EIGEN_MAX_STATIC_ALIGN_BYTES +#undef EIGEN_MAX_STATIC_ALIGN_BYTES +#define EIGEN_MAX_STATIC_ALIGN_BYTES EIGEN_MAX_ALIGN_BYTES +#endif + +#if EIGEN_MAX_STATIC_ALIGN_BYTES==0 && !defined(EIGEN_DISABLE_UNALIGNED_ARRAY_ASSERT) + #define EIGEN_DISABLE_UNALIGNED_ARRAY_ASSERT +#endif + +// At this stage, EIGEN_MAX_STATIC_ALIGN_BYTES>0 is the true test whether we want to align arrays on the stack or not. +// It takes into account both the user choice to explicitly enable/disable alignment (by settting EIGEN_MAX_STATIC_ALIGN_BYTES) +// and the architecture config (EIGEN_ARCH_WANTS_STACK_ALIGNMENT). +// Henceforth, only EIGEN_MAX_STATIC_ALIGN_BYTES should be used. + + +// Shortcuts to EIGEN_ALIGN_TO_BOUNDARY #define EIGEN_ALIGN8 EIGEN_ALIGN_TO_BOUNDARY(8) #define EIGEN_ALIGN16 EIGEN_ALIGN_TO_BOUNDARY(16) +#define EIGEN_ALIGN32 EIGEN_ALIGN_TO_BOUNDARY(32) +#define EIGEN_ALIGN64 EIGEN_ALIGN_TO_BOUNDARY(64) +#if EIGEN_MAX_STATIC_ALIGN_BYTES>0 +#define EIGEN_ALIGN_MAX EIGEN_ALIGN_TO_BOUNDARY(EIGEN_MAX_STATIC_ALIGN_BYTES) +#else +#define EIGEN_ALIGN_MAX +#endif + + +// Dynamic alignment control + +#if defined(EIGEN_DONT_ALIGN) && defined(EIGEN_MAX_ALIGN_BYTES) && EIGEN_MAX_ALIGN_BYTES>0 +#error EIGEN_MAX_ALIGN_BYTES and EIGEN_DONT_ALIGN are both defined with EIGEN_MAX_ALIGN_BYTES!=0. Use EIGEN_MAX_ALIGN_BYTES=0 as a synonym of EIGEN_DONT_ALIGN. +#endif -#if EIGEN_ALIGN_STATICALLY -#define EIGEN_USER_ALIGN_TO_BOUNDARY(n) EIGEN_ALIGN_TO_BOUNDARY(n) -#define EIGEN_USER_ALIGN16 EIGEN_ALIGN16 +#ifdef EIGEN_DONT_ALIGN + #ifdef EIGEN_MAX_ALIGN_BYTES + #undef EIGEN_MAX_ALIGN_BYTES + #endif + #define EIGEN_MAX_ALIGN_BYTES 0 +#elif !defined(EIGEN_MAX_ALIGN_BYTES) + #define EIGEN_MAX_ALIGN_BYTES EIGEN_IDEAL_MAX_ALIGN_BYTES +#endif + +#if EIGEN_IDEAL_MAX_ALIGN_BYTES > EIGEN_MAX_ALIGN_BYTES +#define EIGEN_DEFAULT_ALIGN_BYTES EIGEN_IDEAL_MAX_ALIGN_BYTES #else -#define EIGEN_USER_ALIGN_TO_BOUNDARY(n) -#define EIGEN_USER_ALIGN16 +#define EIGEN_DEFAULT_ALIGN_BYTES EIGEN_MAX_ALIGN_BYTES #endif + +#ifndef EIGEN_UNALIGNED_VECTORIZE +#define EIGEN_UNALIGNED_VECTORIZE 1 +#endif + +//---------------------------------------------------------------------- + + #ifdef EIGEN_DONT_USE_RESTRICT_KEYWORD #define EIGEN_RESTRICT #endif @@ -591,25 +837,26 @@ namespace Eigen { // just an empty macro ! #define EIGEN_EMPTY -#if defined(_MSC_VER) && (_MSC_VER < 1900) && (!defined(__INTEL_COMPILER)) -#define EIGEN_INHERIT_ASSIGNMENT_EQUAL_OPERATOR(Derived) \ - using Base::operator =; -#elif defined(__clang__) // workaround clang bug (see http://forum.kde.org/viewtopic.php?f=74&t=102653) -#define EIGEN_INHERIT_ASSIGNMENT_EQUAL_OPERATOR(Derived) \ - using Base::operator =; \ - EIGEN_STRONG_INLINE Derived& operator=(const Derived& other) { Base::operator=(other); return *this; } \ - template <typename OtherDerived> \ - EIGEN_STRONG_INLINE Derived& operator=(const DenseBase<OtherDerived>& other) { Base::operator=(other.derived()); return *this; } -#else -#define EIGEN_INHERIT_ASSIGNMENT_EQUAL_OPERATOR(Derived) \ - using Base::operator =; \ - EIGEN_STRONG_INLINE Derived& operator=(const Derived& other) \ - { \ - Base::operator=(other); \ - return *this; \ - } +#if EIGEN_COMP_MSVC_STRICT && (EIGEN_COMP_MSVC < 1900 || defined(__CUDACC_VER__)) // for older MSVC versions, as well as 1900 && CUDA 8, using the base operator is sufficient (cf Bugs 1000, 1324) + #define EIGEN_INHERIT_ASSIGNMENT_EQUAL_OPERATOR(Derived) \ + using Base::operator =; +#elif EIGEN_COMP_CLANG // workaround clang bug (see http://forum.kde.org/viewtopic.php?f=74&t=102653) + #define EIGEN_INHERIT_ASSIGNMENT_EQUAL_OPERATOR(Derived) \ + using Base::operator =; \ + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived& operator=(const Derived& other) { Base::operator=(other); return *this; } \ + template <typename OtherDerived> \ + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived& operator=(const DenseBase<OtherDerived>& other) { Base::operator=(other.derived()); return *this; } +#else + #define EIGEN_INHERIT_ASSIGNMENT_EQUAL_OPERATOR(Derived) \ + using Base::operator =; \ + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived& operator=(const Derived& other) \ + { \ + Base::operator=(other); \ + return *this; \ + } #endif + /** \internal * \brief Macro to manually inherit assignment operators. * This is necessary, because the implicitly defined assignment operator gets deleted when a custom operator= is defined. @@ -628,32 +875,13 @@ namespace Eigen { typedef typename Eigen::internal::traits<Derived>::Scalar Scalar; /*!< \brief Numeric type, e.g. float, double, int or std::complex<float>. */ \ typedef typename Eigen::NumTraits<Scalar>::Real RealScalar; /*!< \brief The underlying numeric type for composed scalar types. \details In cases where Scalar is e.g. std::complex<T>, T were corresponding to RealScalar. */ \ typedef typename Base::CoeffReturnType CoeffReturnType; /*!< \brief The return type for coefficient access. \details Depending on whether the object allows direct coefficient access (e.g. for a MatrixXd), this type is either 'const Scalar&' or simply 'Scalar' for objects that do not allow direct coefficient access. */ \ - typedef typename Eigen::internal::nested<Derived>::type Nested; \ + typedef typename Eigen::internal::ref_selector<Derived>::type Nested; \ typedef typename Eigen::internal::traits<Derived>::StorageKind StorageKind; \ - typedef typename Eigen::internal::traits<Derived>::Index Index; \ - enum { RowsAtCompileTime = Eigen::internal::traits<Derived>::RowsAtCompileTime, \ + typedef typename Eigen::internal::traits<Derived>::StorageIndex StorageIndex; \ + enum CompileTimeTraits \ + { RowsAtCompileTime = Eigen::internal::traits<Derived>::RowsAtCompileTime, \ ColsAtCompileTime = Eigen::internal::traits<Derived>::ColsAtCompileTime, \ Flags = Eigen::internal::traits<Derived>::Flags, \ - CoeffReadCost = Eigen::internal::traits<Derived>::CoeffReadCost, \ - SizeAtCompileTime = Base::SizeAtCompileTime, \ - MaxSizeAtCompileTime = Base::MaxSizeAtCompileTime, \ - IsVectorAtCompileTime = Base::IsVectorAtCompileTime }; - - -#define EIGEN_DENSE_PUBLIC_INTERFACE(Derived) \ - typedef typename Eigen::internal::traits<Derived>::Scalar Scalar; /*!< \brief Numeric type, e.g. float, double, int or std::complex<float>. */ \ - typedef typename Eigen::NumTraits<Scalar>::Real RealScalar; /*!< \brief The underlying numeric type for composed scalar types. \details In cases where Scalar is e.g. std::complex<T>, T were corresponding to RealScalar. */ \ - typedef typename Base::PacketScalar PacketScalar; \ - typedef typename Base::CoeffReturnType CoeffReturnType; /*!< \brief The return type for coefficient access. \details Depending on whether the object allows direct coefficient access (e.g. for a MatrixXd), this type is either 'const Scalar&' or simply 'Scalar' for objects that do not allow direct coefficient access. */ \ - typedef typename Eigen::internal::nested<Derived>::type Nested; \ - typedef typename Eigen::internal::traits<Derived>::StorageKind StorageKind; \ - typedef typename Eigen::internal::traits<Derived>::Index Index; \ - enum { RowsAtCompileTime = Eigen::internal::traits<Derived>::RowsAtCompileTime, \ - ColsAtCompileTime = Eigen::internal::traits<Derived>::ColsAtCompileTime, \ - MaxRowsAtCompileTime = Eigen::internal::traits<Derived>::MaxRowsAtCompileTime, \ - MaxColsAtCompileTime = Eigen::internal::traits<Derived>::MaxColsAtCompileTime, \ - Flags = Eigen::internal::traits<Derived>::Flags, \ - CoeffReadCost = Eigen::internal::traits<Derived>::CoeffReadCost, \ SizeAtCompileTime = Base::SizeAtCompileTime, \ MaxSizeAtCompileTime = Base::MaxSizeAtCompileTime, \ IsVectorAtCompileTime = Base::IsVectorAtCompileTime }; \ @@ -661,6 +889,12 @@ namespace Eigen { using Base::const_cast_derived; +// FIXME Maybe the EIGEN_DENSE_PUBLIC_INTERFACE could be removed as importing PacketScalar is rarely needed +#define EIGEN_DENSE_PUBLIC_INTERFACE(Derived) \ + EIGEN_GENERIC_PUBLIC_INTERFACE(Derived) \ + typedef typename Base::PacketScalar PacketScalar; + + #define EIGEN_PLAIN_ENUM_MIN(a,b) (((int)a <= (int)b) ? (int)a : (int)b) #define EIGEN_PLAIN_ENUM_MAX(a,b) (((int)a >= (int)b) ? (int)a : (int)b) @@ -686,24 +920,14 @@ namespace Eigen { #define EIGEN_SIZE_MAX(a,b) (((int)a == Dynamic || (int)b == Dynamic) ? Dynamic \ : ((int)a >= (int)b) ? (int)a : (int)b) -#define EIGEN_ADD_COST(a,b) int(a)==Dynamic || int(b)==Dynamic ? Dynamic : int(a)+int(b) - #define EIGEN_LOGICAL_XOR(a,b) (((a) || (b)) && !((a) && (b))) #define EIGEN_IMPLIES(a,b) (!(a) || (b)) -#define EIGEN_MAKE_CWISE_BINARY_OP(METHOD,FUNCTOR) \ - template<typename OtherDerived> \ - EIGEN_STRONG_INLINE const CwiseBinaryOp<FUNCTOR<Scalar>, const Derived, const OtherDerived> \ - (METHOD)(const EIGEN_CURRENT_STORAGE_BASE_CLASS<OtherDerived> &other) const \ - { \ - return CwiseBinaryOp<FUNCTOR<Scalar>, const Derived, const OtherDerived>(derived(), other.derived()); \ - } - -// the expression type of a cwise product -#define EIGEN_CWISE_PRODUCT_RETURN_TYPE(LHS,RHS) \ +// the expression type of a standard coefficient wise binary operation +#define EIGEN_CWISE_BINARY_RETURN_TYPE(LHS,RHS,OPNAME) \ CwiseBinaryOp< \ - internal::scalar_product_op< \ + EIGEN_CAT(EIGEN_CAT(internal::scalar_,OPNAME),_op)< \ typename internal::traits<LHS>::Scalar, \ typename internal::traits<RHS>::Scalar \ >, \ @@ -711,4 +935,84 @@ namespace Eigen { const RHS \ > +#define EIGEN_MAKE_CWISE_BINARY_OP(METHOD,OPNAME) \ + template<typename OtherDerived> \ + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const EIGEN_CWISE_BINARY_RETURN_TYPE(Derived,OtherDerived,OPNAME) \ + (METHOD)(const EIGEN_CURRENT_STORAGE_BASE_CLASS<OtherDerived> &other) const \ + { \ + return EIGEN_CWISE_BINARY_RETURN_TYPE(Derived,OtherDerived,OPNAME)(derived(), other.derived()); \ + } + +#define EIGEN_SCALAR_BINARY_SUPPORTED(OPNAME,TYPEA,TYPEB) \ + (Eigen::internal::has_ReturnType<Eigen::ScalarBinaryOpTraits<TYPEA,TYPEB,EIGEN_CAT(EIGEN_CAT(Eigen::internal::scalar_,OPNAME),_op)<TYPEA,TYPEB> > >::value) + +#define EIGEN_EXPR_BINARYOP_SCALAR_RETURN_TYPE(EXPR,SCALAR,OPNAME) \ + CwiseBinaryOp<EIGEN_CAT(EIGEN_CAT(internal::scalar_,OPNAME),_op)<typename internal::traits<EXPR>::Scalar,SCALAR>, const EXPR, \ + const typename internal::plain_constant_type<EXPR,SCALAR>::type> + +#define EIGEN_SCALAR_BINARYOP_EXPR_RETURN_TYPE(SCALAR,EXPR,OPNAME) \ + CwiseBinaryOp<EIGEN_CAT(EIGEN_CAT(internal::scalar_,OPNAME),_op)<SCALAR,typename internal::traits<EXPR>::Scalar>, \ + const typename internal::plain_constant_type<EXPR,SCALAR>::type, const EXPR> + +// Workaround for MSVC 2010 (see ML thread "patch with compile for for MSVC 2010") +#if EIGEN_COMP_MSVC_STRICT<=1600 +#define EIGEN_MSVC10_WORKAROUND_BINARYOP_RETURN_TYPE(X) typename internal::enable_if<true,X>::type +#else +#define EIGEN_MSVC10_WORKAROUND_BINARYOP_RETURN_TYPE(X) X +#endif + +#define EIGEN_MAKE_SCALAR_BINARY_OP_ONTHERIGHT(METHOD,OPNAME) \ + template <typename T> EIGEN_DEVICE_FUNC 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; \ + return EIGEN_EXPR_BINARYOP_SCALAR_RETURN_TYPE(Derived,PromotedT,OPNAME)(derived(), \ + typename internal::plain_constant_type<Derived,PromotedT>::type(derived().rows(), derived().cols(), internal::scalar_constant_op<PromotedT>(scalar))); \ + } + +#define EIGEN_MAKE_SCALAR_BINARY_OP_ONTHELEFT(METHOD,OPNAME) \ + template <typename T> EIGEN_DEVICE_FUNC 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; \ + return EIGEN_SCALAR_BINARYOP_EXPR_RETURN_TYPE(PromotedT,Derived,OPNAME)( \ + typename internal::plain_constant_type<Derived,PromotedT>::type(matrix.derived().rows(), matrix.derived().cols(), internal::scalar_constant_op<PromotedT>(scalar)), matrix.derived()); \ + } + +#define EIGEN_MAKE_SCALAR_BINARY_OP(METHOD,OPNAME) \ + EIGEN_MAKE_SCALAR_BINARY_OP_ONTHELEFT(METHOD,OPNAME) \ + EIGEN_MAKE_SCALAR_BINARY_OP_ONTHERIGHT(METHOD,OPNAME) + + +#ifdef EIGEN_EXCEPTIONS +# define EIGEN_THROW_X(X) throw X +# define EIGEN_THROW throw +# define EIGEN_TRY try +# define EIGEN_CATCH(X) catch (X) +#else +# ifdef __CUDA_ARCH__ +# define EIGEN_THROW_X(X) asm("trap;") +# define EIGEN_THROW asm("trap;") +# else +# define EIGEN_THROW_X(X) std::abort() +# define EIGEN_THROW std::abort() +# endif +# define EIGEN_TRY if (true) +# define EIGEN_CATCH(X) else +#endif + + +#if EIGEN_HAS_CXX11_NOEXCEPT +# define EIGEN_INCLUDE_TYPE_TRAITS +# define EIGEN_NOEXCEPT noexcept +# define EIGEN_NOEXCEPT_IF(x) noexcept(x) +# define EIGEN_NO_THROW noexcept(true) +# define EIGEN_EXCEPTION_SPEC(X) noexcept(false) +#else +# define EIGEN_NOEXCEPT +# define EIGEN_NOEXCEPT_IF(x) +# define EIGEN_NO_THROW throw() +# define EIGEN_EXCEPTION_SPEC(X) throw(X) +#endif + #endif // EIGEN_MACROS_H |