1*bdd1243dSDimitry Andric/*===---- cmath - CUDA wrapper for <cmath> ---------------------------------=== 2*bdd1243dSDimitry Andric * 3*bdd1243dSDimitry Andric * Permission is hereby granted, free of charge, to any person obtaining a copy 4*bdd1243dSDimitry Andric * of this software and associated documentation files (the "Software"), to deal 5*bdd1243dSDimitry Andric * in the Software without restriction, including without limitation the rights 6*bdd1243dSDimitry Andric * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell 7*bdd1243dSDimitry Andric * copies of the Software, and to permit persons to whom the Software is 8*bdd1243dSDimitry Andric * furnished to do so, subject to the following conditions: 9*bdd1243dSDimitry Andric * 10*bdd1243dSDimitry Andric * The above copyright notice and this permission notice shall be included in 11*bdd1243dSDimitry Andric * all copies or substantial portions of the Software. 12*bdd1243dSDimitry Andric * 13*bdd1243dSDimitry Andric * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR 14*bdd1243dSDimitry Andric * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, 15*bdd1243dSDimitry Andric * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE 16*bdd1243dSDimitry Andric * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER 17*bdd1243dSDimitry Andric * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, 18*bdd1243dSDimitry Andric * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN 19*bdd1243dSDimitry Andric * THE SOFTWARE. 20*bdd1243dSDimitry Andric * 21*bdd1243dSDimitry Andric *===-----------------------------------------------------------------------=== 22*bdd1243dSDimitry Andric */ 23*bdd1243dSDimitry Andric 24*bdd1243dSDimitry Andric#ifndef __CLANG_CUDA_WRAPPERS_CMATH 25*bdd1243dSDimitry Andric#define __CLANG_CUDA_WRAPPERS_CMATH 26*bdd1243dSDimitry Andric 27*bdd1243dSDimitry Andric#include_next <cmath> 28*bdd1243dSDimitry Andric 29*bdd1243dSDimitry Andric#if defined(_LIBCPP_STD_VER) 30*bdd1243dSDimitry Andric 31*bdd1243dSDimitry Andric// libc++ will need long double variants of these functions, but CUDA does not 32*bdd1243dSDimitry Andric// provide them. We'll provide their declarations, which should allow the 33*bdd1243dSDimitry Andric// headers to parse, but would not allow accidental use of them on a GPU. 34*bdd1243dSDimitry Andric 35*bdd1243dSDimitry Andric__attribute__((device)) long double logb(long double); 36*bdd1243dSDimitry Andric__attribute__((device)) long double scalbn(long double, int); 37*bdd1243dSDimitry Andric 38*bdd1243dSDimitry Andricnamespace std { 39*bdd1243dSDimitry Andric 40*bdd1243dSDimitry Andric// For __constexpr_fmin/fmax we only need device-side overloads before c++14 41*bdd1243dSDimitry Andric// where they are not constexpr. 42*bdd1243dSDimitry Andric#if _LIBCPP_STD_VER < 14 43*bdd1243dSDimitry Andric 44*bdd1243dSDimitry Andric__attribute__((device)) 45*bdd1243dSDimitry Andricinline _LIBCPP_HIDE_FROM_ABI _LIBCPP_CONSTEXPR_SINCE_CXX14 float __constexpr_fmax(float __x, float __y) _NOEXCEPT { 46*bdd1243dSDimitry Andric return __builtin_fmaxf(__x, __y); 47*bdd1243dSDimitry Andric} 48*bdd1243dSDimitry Andric 49*bdd1243dSDimitry Andric__attribute__((device)) 50*bdd1243dSDimitry Andricinline _LIBCPP_HIDE_FROM_ABI _LIBCPP_CONSTEXPR_SINCE_CXX14 double __constexpr_fmax(double __x, double __y) _NOEXCEPT { 51*bdd1243dSDimitry Andric return __builtin_fmax(__x, __y); 52*bdd1243dSDimitry Andric} 53*bdd1243dSDimitry Andric 54*bdd1243dSDimitry Andric__attribute__((device)) 55*bdd1243dSDimitry Andricinline _LIBCPP_HIDE_FROM_ABI _LIBCPP_CONSTEXPR_SINCE_CXX14 long double 56*bdd1243dSDimitry Andric__constexpr_fmax(long double __x, long double __y) _NOEXCEPT { 57*bdd1243dSDimitry Andric return __builtin_fmaxl(__x, __y); 58*bdd1243dSDimitry Andric} 59*bdd1243dSDimitry Andric 60*bdd1243dSDimitry Andrictemplate <class _Tp, class _Up, __enable_if_t<is_arithmetic<_Tp>::value && is_arithmetic<_Up>::value, int> = 0> 61*bdd1243dSDimitry Andric__attribute__((device)) 62*bdd1243dSDimitry Andric_LIBCPP_HIDE_FROM_ABI _LIBCPP_CONSTEXPR_SINCE_CXX14 typename __promote<_Tp, _Up>::type 63*bdd1243dSDimitry Andric__constexpr_fmax(_Tp __x, _Up __y) _NOEXCEPT { 64*bdd1243dSDimitry Andric using __result_type = typename __promote<_Tp, _Up>::type; 65*bdd1243dSDimitry Andric return std::__constexpr_fmax(static_cast<__result_type>(__x), static_cast<__result_type>(__y)); 66*bdd1243dSDimitry Andric} 67*bdd1243dSDimitry Andric#endif // _LIBCPP_STD_VER < 14 68*bdd1243dSDimitry Andric 69*bdd1243dSDimitry Andric// For logb/scalbn templates we must always provide device overloads because 70*bdd1243dSDimitry Andric// libc++ implementation uses __builtin_XXX which gets translated into a libcall 71*bdd1243dSDimitry Andric// which we can't handle on GPU. We need to forward those to CUDA-provided 72*bdd1243dSDimitry Andric// implementations. 73*bdd1243dSDimitry Andric 74*bdd1243dSDimitry Andrictemplate <class _Tp> 75*bdd1243dSDimitry Andric__attribute__((device)) 76*bdd1243dSDimitry Andric_LIBCPP_HIDE_FROM_ABI _LIBCPP_CONSTEXPR_SINCE_CXX14 _Tp __constexpr_logb(_Tp __x) { 77*bdd1243dSDimitry Andric return ::logb(__x); 78*bdd1243dSDimitry Andric} 79*bdd1243dSDimitry Andric 80*bdd1243dSDimitry Andrictemplate <class _Tp> 81*bdd1243dSDimitry Andric__attribute__((device)) 82*bdd1243dSDimitry Andric_LIBCPP_HIDE_FROM_ABI _LIBCPP_CONSTEXPR_SINCE_CXX20 _Tp __constexpr_scalbn(_Tp __x, int __exp) { 83*bdd1243dSDimitry Andric return ::scalbn(__x, __exp); 84*bdd1243dSDimitry Andric} 85*bdd1243dSDimitry Andric 86*bdd1243dSDimitry Andric} // namespace std// 87*bdd1243dSDimitry Andric 88*bdd1243dSDimitry Andric#endif // _LIBCPP_STD_VER 89*bdd1243dSDimitry Andric 90*bdd1243dSDimitry Andric#endif // include guard 91