| /*===---- cmath - CUDA wrapper for <cmath> ---------------------------------=== |
| * |
| * Permission is hereby granted, free of charge, to any person obtaining a copy |
| * of this software and associated documentation files (the "Software"), to deal |
| * in the Software without restriction, including without limitation the rights |
| * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell |
| * copies of the Software, and to permit persons to whom the Software is |
| * furnished to do so, subject to the following conditions: |
| * |
| * The above copyright notice and this permission notice shall be included in |
| * all copies or substantial portions of the Software. |
| * |
| * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR |
| * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, |
| * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE |
| * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER |
| * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, |
| * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN |
| * THE SOFTWARE. |
| * |
| *===-----------------------------------------------------------------------=== |
| */ |
| |
| #ifndef __CLANG_CUDA_WRAPPERS_CMATH |
| #define __CLANG_CUDA_WRAPPERS_CMATH |
| |
| #include_next <cmath> |
| |
| #if defined(_LIBCPP_STD_VER) |
| |
| // libc++ will need long double variants of these functions, but CUDA does not |
| // provide them. We'll provide their declarations, which should allow the |
| // headers to parse, but would not allow accidental use of them on a GPU. |
| |
| __attribute__((device)) long double logb(long double); |
| __attribute__((device)) long double scalbn(long double, int); |
| |
| namespace std { |
| |
| // For __constexpr_fmin/fmax we only need device-side overloads before c++14 |
| // where they are not constexpr. |
| #if _LIBCPP_STD_VER < 14 |
| |
| __attribute__((device)) |
| inline _LIBCPP_HIDE_FROM_ABI _LIBCPP_CONSTEXPR_SINCE_CXX14 float __constexpr_fmax(float __x, float __y) _NOEXCEPT { |
| return __builtin_fmaxf(__x, __y); |
| } |
| |
| __attribute__((device)) |
| inline _LIBCPP_HIDE_FROM_ABI _LIBCPP_CONSTEXPR_SINCE_CXX14 double __constexpr_fmax(double __x, double __y) _NOEXCEPT { |
| return __builtin_fmax(__x, __y); |
| } |
| |
| __attribute__((device)) |
| inline _LIBCPP_HIDE_FROM_ABI _LIBCPP_CONSTEXPR_SINCE_CXX14 long double |
| __constexpr_fmax(long double __x, long double __y) _NOEXCEPT { |
| return __builtin_fmaxl(__x, __y); |
| } |
| |
| template <class _Tp, class _Up, __enable_if_t<is_arithmetic<_Tp>::value && is_arithmetic<_Up>::value, int> = 0> |
| __attribute__((device)) |
| _LIBCPP_HIDE_FROM_ABI _LIBCPP_CONSTEXPR_SINCE_CXX14 typename __promote<_Tp, _Up>::type |
| __constexpr_fmax(_Tp __x, _Up __y) _NOEXCEPT { |
| using __result_type = typename __promote<_Tp, _Up>::type; |
| return std::__constexpr_fmax(static_cast<__result_type>(__x), static_cast<__result_type>(__y)); |
| } |
| #endif // _LIBCPP_STD_VER < 14 |
| |
| // For logb/scalbn templates we must always provide device overloads because |
| // libc++ implementation uses __builtin_XXX which gets translated into a libcall |
| // which we can't handle on GPU. We need to forward those to CUDA-provided |
| // implementations. |
| |
| template <class _Tp> |
| __attribute__((device)) |
| _LIBCPP_HIDE_FROM_ABI _LIBCPP_CONSTEXPR_SINCE_CXX14 _Tp __constexpr_logb(_Tp __x) { |
| return ::logb(__x); |
| } |
| |
| template <class _Tp> |
| __attribute__((device)) |
| _LIBCPP_HIDE_FROM_ABI _LIBCPP_CONSTEXPR_SINCE_CXX20 _Tp __constexpr_scalbn(_Tp __x, int __exp) { |
| return ::scalbn(__x, __exp); |
| } |
| |
| } // namespace std// |
| |
| #endif // _LIBCPP_STD_VER |
| |
| #endif // include guard |