|
| 1 | +// Copyright (c) 2021 PaddlePaddle Authors. All Rights Reserved. |
| 2 | +// |
| 3 | +// Licensed under the Apache License, Version 2.0 (the "License"); |
| 4 | +// you may not use this file except in compliance with the License. |
| 5 | +// You may obtain a copy of the License at |
| 6 | +// |
| 7 | +// http://www.apache.org/licenses/LICENSE-2.0 |
| 8 | +// |
| 9 | +// Unless required by applicable law or agreed to in writing, software |
| 10 | +// distributed under the License is distributed on an "AS IS" BASIS, |
| 11 | +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. |
| 12 | +// See the License for the specific language governing permissions and |
| 13 | +// limitations under the License. |
| 14 | + |
| 15 | +#pragma once |
| 16 | + |
| 17 | +namespace paddle { |
| 18 | +namespace operators { |
| 19 | +namespace kernel_primitives { |
| 20 | +namespace details { |
| 21 | + |
| 22 | +static __device__ __forceinline__ platform::float16 Exp(platform::float16 x) { |
| 23 | + return ::Eigen::numext::exp(x); |
| 24 | +} |
| 25 | + |
| 26 | +static __device__ __forceinline__ float Exp(float x) { return expf(x); } |
| 27 | + |
| 28 | +static __device__ __forceinline__ double Exp(double x) { return exp(x); } |
| 29 | + |
| 30 | +static __device__ __forceinline__ platform::float16 Log(platform::float16 x) { |
| 31 | + return ::Eigen::numext::log(x); |
| 32 | +} |
| 33 | + |
| 34 | +static __device__ __forceinline__ float Log(float x) { return logf(x); } |
| 35 | + |
| 36 | +static __device__ __forceinline__ double Log(double x) { return log(x); } |
| 37 | + |
| 38 | +} // namespace details |
| 39 | + |
| 40 | +/******************************** Unary Functor *******************************/ |
| 41 | + |
| 42 | +/** |
| 43 | + * @brief Default unary exp functor |
| 44 | + */ |
| 45 | +template <typename Tx, typename Ty = Tx> |
| 46 | +struct ExpFunctor { |
| 47 | + HOSTDEVICE inline ExpFunctor() {} |
| 48 | + |
| 49 | + HOSTDEVICE explicit inline ExpFunctor(int n) {} |
| 50 | + |
| 51 | + HOSTDEVICE inline Ty operator()(const Tx& x) const { |
| 52 | + return static_cast<Ty>(details::Exp(x)); |
| 53 | + } |
| 54 | +}; |
| 55 | + |
| 56 | +/** |
| 57 | + * @brief Default unary identity functor |
| 58 | + */ |
| 59 | +template <typename Tx, typename Ty = Tx> |
| 60 | +struct IdentityFunctor { |
| 61 | + HOSTDEVICE inline IdentityFunctor() {} |
| 62 | + |
| 63 | + HOSTDEVICE explicit inline IdentityFunctor(int n) {} |
| 64 | + |
| 65 | + HOSTDEVICE inline Ty operator()(const Tx& x) const { |
| 66 | + return static_cast<Ty>(x); |
| 67 | + } |
| 68 | +}; |
| 69 | + |
| 70 | +/** |
| 71 | + * @brief Default unary div functor. Divide by a constant |
| 72 | + */ |
| 73 | +template <typename Tx, typename Ty = Tx> |
| 74 | +struct DivideFunctor { |
| 75 | + HOSTDEVICE inline DivideFunctor() { n_inv = static_cast<Tx>(1.0f); } |
| 76 | + |
| 77 | + HOSTDEVICE explicit inline DivideFunctor(int n) : n_inv((Tx)(1.0 / n)) {} |
| 78 | + |
| 79 | + HOSTDEVICE inline Ty operator()(const Tx& x) const { |
| 80 | + return static_cast<Ty>(x * n_inv); |
| 81 | + } |
| 82 | + |
| 83 | + private: |
| 84 | + Tx n_inv; |
| 85 | +}; |
| 86 | + |
| 87 | +/** |
| 88 | + * @brief Default unary square functor |
| 89 | + */ |
| 90 | +template <typename Tx, typename Ty = Tx> |
| 91 | +struct SquareFunctor { |
| 92 | + HOSTDEVICE inline SquareFunctor() {} |
| 93 | + |
| 94 | + HOSTDEVICE explicit inline SquareFunctor(int n) {} |
| 95 | + |
| 96 | + HOSTDEVICE inline Ty operator()(const Tx& x) const { |
| 97 | + return static_cast<Ty>(x) * static_cast<Ty>(x); |
| 98 | + } |
| 99 | +}; |
| 100 | + |
| 101 | +/****************************** Binary Functor ********************************/ |
| 102 | + |
| 103 | +/** |
| 104 | + * @brief Default binary min functor |
| 105 | + */ |
| 106 | +template <typename T> |
| 107 | +struct MinFunctor { |
| 108 | + inline T initial() { return static_cast<T>(std::numeric_limits<T>::max()); } |
| 109 | + |
| 110 | + __device__ __forceinline__ T operator()(const T& a, const T& b) const { |
| 111 | + return (b < a) ? b : a; |
| 112 | + } |
| 113 | +}; |
| 114 | + |
| 115 | +/** |
| 116 | + * @brief Default binary max functor |
| 117 | + */ |
| 118 | +template <typename T> |
| 119 | +struct MaxFunctor { |
| 120 | + inline T initial() { |
| 121 | + return static_cast<T>(std::numeric_limits<T>::lowest()); |
| 122 | + } |
| 123 | + |
| 124 | + __device__ __forceinline__ T operator()(const T& a, const T& b) const { |
| 125 | + return (b > a) ? b : a; |
| 126 | + } |
| 127 | +}; |
| 128 | + |
| 129 | +/** |
| 130 | + * @brief Default binary add functor |
| 131 | + */ |
| 132 | +template <typename T> |
| 133 | +struct AddFunctor { |
| 134 | + inline T initial() { return static_cast<T>(0.0f); } |
| 135 | + |
| 136 | + __device__ __forceinline__ T operator()(const T& a, const T& b) const { |
| 137 | + return b + a; |
| 138 | + } |
| 139 | +}; |
| 140 | + |
| 141 | +/** |
| 142 | + * @brief Default binary add functor |
| 143 | + */ |
| 144 | +template <typename T> |
| 145 | +struct MulFunctor { |
| 146 | + inline T initial() { return static_cast<T>(1.0f); } |
| 147 | + |
| 148 | + __device__ __forceinline__ T operator()(const T& a, const T& b) const { |
| 149 | + return b * a; |
| 150 | + } |
| 151 | +}; |
| 152 | + |
| 153 | +/** |
| 154 | + * @brief Default binary logic or functor |
| 155 | + */ |
| 156 | +template <typename T> |
| 157 | +struct LogicalOrFunctor { |
| 158 | + inline T initial() { return static_cast<T>(false); } |
| 159 | + |
| 160 | + __device__ __forceinline__ T operator()(const T& a, const T& b) const { |
| 161 | + return b || a; |
| 162 | + } |
| 163 | +}; |
| 164 | + |
| 165 | +/** |
| 166 | + * @brief Default binary logic and functor |
| 167 | + */ |
| 168 | +template <typename T> |
| 169 | +struct LogicalAndFunctor { |
| 170 | + inline T initial() { return static_cast<T>(true); } |
| 171 | + |
| 172 | + __device__ __forceinline__ T operator()(const T& a, const T& b) const { |
| 173 | + return b && a; |
| 174 | + } |
| 175 | +}; |
| 176 | + |
| 177 | +/** |
| 178 | + * @brief Default binary sub functor |
| 179 | + */ |
| 180 | +template <typename T> |
| 181 | +struct SubFunctor { |
| 182 | + inline T initial() { return static_cast<T>(0.0f); } |
| 183 | + |
| 184 | + inline HOSTDEVICE T operator()(const T& a, const T& b) const { return a - b; } |
| 185 | +}; |
| 186 | + |
| 187 | +/** |
| 188 | + * @brief Default binary div functor |
| 189 | + */ |
| 190 | +template <typename T, typename Enable = void> |
| 191 | +struct DivFunctor { |
| 192 | + inline T initial() { return static_cast<T>(1.0f); } |
| 193 | + |
| 194 | + inline HOSTDEVICE T operator()(const T& a, const T& b) const { return a / b; } |
| 195 | +}; |
| 196 | + |
| 197 | +template <typename T> |
| 198 | +struct DivFunctor<T, |
| 199 | + typename std::enable_if<std::is_integral<T>::value>::type> { |
| 200 | + inline T initial() { return static_cast<T>(1.0f); } |
| 201 | + |
| 202 | + inline HOSTDEVICE T operator()(const T& a, const T& b) const { |
| 203 | + // For int32/int64, need to check whether the divison is zero. |
| 204 | + PADDLE_ENFORCE_NE(b, 0, |
| 205 | + platform::errors::InvalidArgument( |
| 206 | + "Integer division by zero encountered " |
| 207 | + "in (floor) divide. Please check the input value.")); |
| 208 | + return a / b; |
| 209 | + } |
| 210 | +}; |
| 211 | + |
| 212 | +/** |
| 213 | + * @brief Default binary floor divide functor |
| 214 | + */ |
| 215 | +template <typename T> |
| 216 | +struct FloorDivFunctor { |
| 217 | + inline T initial() { return static_cast<T>(1.0f); } |
| 218 | + |
| 219 | + inline HOSTDEVICE T operator()(const T& a, const T& b) const { |
| 220 | + PADDLE_ENFORCE_NE(b, 0, |
| 221 | + platform::errors::InvalidArgument( |
| 222 | + "Integer division by zero encountered " |
| 223 | + "in (floor) divide. Please check the input value.")); |
| 224 | + return static_cast<T>(std::trunc(a / b)); |
| 225 | + } |
| 226 | +}; |
| 227 | + |
| 228 | +} // namespace kernel_primitives |
| 229 | +} // namespace operators |
| 230 | +} // namespace paddle |
0 commit comments