1/*===---- cmath - CUDA wrapper for <cmath> ---------------------------------=== 2 * 3 * Permission is hereby granted, free of charge, to any person obtaining a copy 4 * of this software and associated documentation files (the "Software"), to deal 5 * in the Software without restriction, including without limitation the rights 6 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell 7 * copies of the Software, and to permit persons to whom the Software is 8 * furnished to do so, subject to the following conditions: 9 * 10 * The above copyright notice and this permission notice shall be included in 11 * all copies or substantial portions of the Software. 12 * 13 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR 14 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, 15 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE 16 * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER 17 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, 18 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN 19 * THE SOFTWARE. 20 * 21 *===-----------------------------------------------------------------------=== 22 */ 23 24#ifndef __CLANG_CUDA_WRAPPERS_CMATH 25#define __CLANG_CUDA_WRAPPERS_CMATH 26 27#include_next <cmath> 28 29#if defined(_LIBCPP_STD_VER) 30 31// libc++ will need long double variants of these functions, but CUDA does not 32// provide them. We'll provide their declarations, which should allow the 33// headers to parse, but would not allow accidental use of them on a GPU. 34 35__attribute__((device)) long double logb(long double); 36__attribute__((device)) long double scalbn(long double, int); 37 38namespace std { 39 40// For __constexpr_fmin/fmax we only need device-side overloads before c++14 41// where they are not constexpr. 42#if _LIBCPP_STD_VER < 14 43 44__attribute__((device)) 45inline _LIBCPP_HIDE_FROM_ABI _LIBCPP_CONSTEXPR_SINCE_CXX14 float __constexpr_fmax(float __x, float __y) _NOEXCEPT { 46 return __builtin_fmaxf(__x, __y); 47} 48 49__attribute__((device)) 50inline _LIBCPP_HIDE_FROM_ABI _LIBCPP_CONSTEXPR_SINCE_CXX14 double __constexpr_fmax(double __x, double __y) _NOEXCEPT { 51 return __builtin_fmax(__x, __y); 52} 53 54__attribute__((device)) 55inline _LIBCPP_HIDE_FROM_ABI _LIBCPP_CONSTEXPR_SINCE_CXX14 long double 56__constexpr_fmax(long double __x, long double __y) _NOEXCEPT { 57 return __builtin_fmaxl(__x, __y); 58} 59 60template <class _Tp, class _Up, __enable_if_t<is_arithmetic<_Tp>::value && is_arithmetic<_Up>::value, int> = 0> 61__attribute__((device)) 62_LIBCPP_HIDE_FROM_ABI _LIBCPP_CONSTEXPR_SINCE_CXX14 typename __promote<_Tp, _Up>::type 63__constexpr_fmax(_Tp __x, _Up __y) _NOEXCEPT { 64 using __result_type = typename __promote<_Tp, _Up>::type; 65 return std::__constexpr_fmax(static_cast<__result_type>(__x), static_cast<__result_type>(__y)); 66} 67#endif // _LIBCPP_STD_VER < 14 68 69// For logb/scalbn templates we must always provide device overloads because 70// libc++ implementation uses __builtin_XXX which gets translated into a libcall 71// which we can't handle on GPU. We need to forward those to CUDA-provided 72// implementations. 73 74template <class _Tp> 75__attribute__((device)) 76_LIBCPP_HIDE_FROM_ABI _LIBCPP_CONSTEXPR_SINCE_CXX14 _Tp __constexpr_logb(_Tp __x) { 77 return ::logb(__x); 78} 79 80template <class _Tp> 81__attribute__((device)) 82_LIBCPP_HIDE_FROM_ABI _LIBCPP_CONSTEXPR_SINCE_CXX20 _Tp __constexpr_scalbn(_Tp __x, int __exp) { 83 return ::scalbn(__x, __exp); 84} 85 86} // namespace std// 87 88#endif // _LIBCPP_STD_VER 89 90#endif // include guard 91