| 1 | /////////////////////////////////////////////////////////////// |
| 2 | // Copyright 2020 Madhur Chauhan. |
| 3 | // Copyright 2020 John Maddock. |
| 4 | // Distributed under the Boost |
| 5 | // Software License, Version 1.0. (See accompanying file |
| 6 | // LICENSE_1_0.txt or copy at https://www.boost.org/LICENSE_1_0.txt |
| 7 | |
| 8 | #ifndef BOOST_MP_INTEL_INTRINSICS_HPP |
| 9 | #define BOOST_MP_INTEL_INTRINSICS_HPP |
| 10 | // |
| 11 | // Select which actual implementation header to use: |
| 12 | // |
| 13 | #ifdef __has_include |
| 14 | #if __has_include(<immintrin.h>) |
| 15 | #define BOOST_MP_HAS_IMMINTRIN_H |
| 16 | #endif |
| 17 | #endif |
| 18 | // |
| 19 | // If this is GCC/clang, then check that the actual intrinsic exists: |
| 20 | // |
| 21 | #if defined(__has_builtin) && defined(__GNUC__) |
| 22 | #if !__has_builtin(__builtin_ia32_addcarryx_u64) && defined(BOOST_MP_HAS_IMMINTRIN_H) \ |
| 23 | && !(defined(BOOST_GCC) && (__GNUC__ >= 9) \ |
| 24 | && (defined(__amd64__) || defined(__amd64) || defined(__x86_64__) || defined(__x86_64)\ |
| 25 | || defined(i386) || defined(__i386) || defined(__i386__) || defined(_M_AMD64) \ |
| 26 | || defined(_M_X64) || defined(__amd64__) || defined(_M_X64))) |
| 27 | #undef BOOST_MP_HAS_IMMINTRIN_H |
| 28 | #endif |
| 29 | #elif defined(BOOST_MP_HAS_IMMINTRIN_H) && defined(__GNUC__) && !(defined(BOOST_GCC) && (__GNUC__ >= 9)) |
| 30 | #undef BOOST_MP_HAS_IMMINTRIN_H |
| 31 | #endif |
| 32 | |
| 33 | #if defined(__clang_major__) && (__clang_major__ < 9) |
| 34 | // We appear to crash the compiler if we try to use these intrinsics? |
| 35 | #undef BOOST_MP_HAS_IMMINTRIN_H |
| 36 | #endif |
| 37 | |
| 38 | #if defined(_WIN32) && (defined(_M_ARM64) || defined(_M_ARM)) |
| 39 | // |
| 40 | // When targeting platforms such as ARM, msvc (and also clang when emulating msvc) still has the |
| 41 | // Intel headers in its include path even though they're not usable. |
| 42 | // See https://github.com/boostorg/multiprecision/issues/321 |
| 43 | // Also https://github.com/boostorg/multiprecision/issues/475 |
| 44 | // |
| 45 | #undef BOOST_MP_HAS_IMMINTRIN_H |
| 46 | #endif |
| 47 | |
| 48 | #if defined(__APPLE_CC__) && defined(__clang_major__) && (__clang_major__ < 11) && defined(BOOST_MP_HAS_IMMINTRIN_H) |
| 49 | // Apple clang has it's own version numbers. |
| 50 | #undef BOOST_MP_HAS_IMMINTRIN_H |
| 51 | #endif |
| 52 | |
| 53 | |
| 54 | // |
| 55 | // If the compiler supports the intrinsics used by GCC internally |
| 56 | // inside <immintrin.h> then we'll use them directly. |
| 57 | // This is a bit of defensive programming, mostly for a modern clang |
| 58 | // sitting on top of an older GCC header install. |
| 59 | // |
| 60 | #if defined(__has_builtin) && !defined(BOOST_INTEL) |
| 61 | |
| 62 | # if __has_builtin(__builtin_ia32_addcarryx_u64) |
| 63 | # define BOOST_MP_ADDC __builtin_ia32_addcarryx_u |
| 64 | # endif |
| 65 | |
| 66 | # if __has_builtin(__builtin_ia32_subborrow_u64) |
| 67 | # define BOOST_MP_SUBB __builtin_ia32_subborrow_u |
| 68 | # elif __has_builtin(__builtin_ia32_sbb_u64) |
| 69 | # define BOOST_MP_SUBB __builtin_ia32_sbb_u |
| 70 | # endif |
| 71 | |
| 72 | #endif |
| 73 | |
| 74 | #ifndef BOOST_MP_ADDC |
| 75 | #define BOOST_MP_ADDC _addcarry_u |
| 76 | #endif |
| 77 | #ifndef BOOST_MP_SUBB |
| 78 | #define BOOST_MP_SUBB _subborrow_u |
| 79 | #endif |
| 80 | |
| 81 | #ifdef BOOST_MP_HAS_IMMINTRIN_H |
| 82 | |
| 83 | #ifdef BOOST_MSVC |
| 84 | // |
| 85 | // This is a subset of the full <immintrin.h> : |
| 86 | // |
| 87 | #include <intrin.h> |
| 88 | #else |
| 89 | #include <immintrin.h> |
| 90 | #endif |
| 91 | |
| 92 | #if defined(BOOST_HAS_INT128) |
| 93 | |
| 94 | namespace boost { namespace multiprecision { namespace detail { |
| 95 | |
| 96 | BOOST_MP_FORCEINLINE unsigned char addcarry_limb(unsigned char carry, limb_type a, limb_type b, limb_type* p_result) |
| 97 | { |
| 98 | #ifdef BOOST_INTEL |
| 99 | using cast_type = unsigned __int64; |
| 100 | #else |
| 101 | using cast_type = unsigned long long; |
| 102 | #endif |
| 103 | return BOOST_JOIN(BOOST_MP_ADDC, 64)(carry, a, b, reinterpret_cast<cast_type*>(p_result)); |
| 104 | } |
| 105 | |
| 106 | BOOST_MP_FORCEINLINE unsigned char subborrow_limb(unsigned char carry, limb_type a, limb_type b, limb_type* p_result) |
| 107 | { |
| 108 | #ifdef BOOST_INTEL |
| 109 | using cast_type = unsigned __int64; |
| 110 | #else |
| 111 | using cast_type = unsigned long long; |
| 112 | #endif |
| 113 | return BOOST_JOIN(BOOST_MP_SUBB, 64)(carry, a, b, reinterpret_cast<cast_type*>(p_result)); |
| 114 | } |
| 115 | |
| 116 | }}} // namespace boost::multiprecision::detail |
| 117 | |
| 118 | #else |
| 119 | |
| 120 | namespace boost { namespace multiprecision { namespace detail { |
| 121 | |
| 122 | BOOST_MP_FORCEINLINE unsigned char addcarry_limb(unsigned char carry, limb_type a, limb_type b, limb_type* p_result) |
| 123 | { |
| 124 | return BOOST_JOIN(BOOST_MP_ADDC, 32)(carry, a, b, reinterpret_cast<unsigned int*>(p_result)); |
| 125 | } |
| 126 | |
| 127 | BOOST_MP_FORCEINLINE unsigned char subborrow_limb(unsigned char carry, limb_type a, limb_type b, limb_type* p_result) |
| 128 | { |
| 129 | return BOOST_JOIN(BOOST_MP_SUBB, 32)(carry, a, b, reinterpret_cast<unsigned int*>(p_result)); |
| 130 | } |
| 131 | |
| 132 | }}} // namespace boost::multiprecision::detail |
| 133 | |
| 134 | #endif |
| 135 | |
| 136 | #endif |
| 137 | |
| 138 | #endif |
| 139 | |