Michalis Spyrou | a6825a4 | 2018-09-13 12:24:03 +0100 | [diff] [blame] | 1 | /* |
Georgios Pinitas | ddb93bb | 2020-10-02 16:38:59 +0100 | [diff] [blame] | 2 | * Copyright (c) 2018-2020 Arm Limited. |
Michalis Spyrou | a6825a4 | 2018-09-13 12:24:03 +0100 | [diff] [blame] | 3 | * |
| 4 | * SPDX-License-Identifier: MIT |
| 5 | * |
| 6 | * Permission is hereby granted, free of charge, to any person obtaining a copy |
| 7 | * of this software and associated documentation files (the "Software"), to |
| 8 | * deal in the Software without restriction, including without limitation the |
| 9 | * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or |
| 10 | * sell copies of the Software, and to permit persons to whom the Software is |
| 11 | * furnished to do so, subject to the following conditions: |
| 12 | * |
| 13 | * The above copyright notice and this permission notice shall be included in all |
| 14 | * copies or substantial portions of the Software. |
| 15 | * |
| 16 | * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR |
| 17 | * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, |
| 18 | * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE |
| 19 | * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER |
| 20 | * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, |
| 21 | * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE |
| 22 | * SOFTWARE. |
| 23 | */ |
Michalis Spyrou | f464337 | 2019-11-29 16:17:13 +0000 | [diff] [blame] | 24 | #ifndef ARM_COMPUTE_WRAPPER_ADD_H |
| 25 | #define ARM_COMPUTE_WRAPPER_ADD_H |
Michalis Spyrou | a6825a4 | 2018-09-13 12:24:03 +0100 | [diff] [blame] | 26 | |
| 27 | #include <arm_neon.h> |
| 28 | |
| 29 | namespace arm_compute |
| 30 | { |
| 31 | namespace wrapper |
| 32 | { |
| 33 | #define VADD_IMPL(stype, vtype, prefix, postfix) \ |
| 34 | inline vtype vadd(const vtype &a, const vtype &b) \ |
| 35 | { \ |
| 36 | return prefix##_##postfix(a, b); \ |
| 37 | } |
| 38 | |
| 39 | VADD_IMPL(uint8x8_t, uint8x8_t, vadd, u8) |
| 40 | VADD_IMPL(int8x8_t, int8x8_t, vadd, s8) |
| 41 | VADD_IMPL(uint16x4_t, uint16x4_t, vadd, u16) |
| 42 | VADD_IMPL(int16x4_t, int16x4_t, vadd, s16) |
| 43 | VADD_IMPL(uint32x2_t, uint32x2_t, vadd, u32) |
| 44 | VADD_IMPL(int32x2_t, int32x2_t, vadd, s32) |
| 45 | VADD_IMPL(uint64x1_t, uint64x1_t, vadd, u64) |
| 46 | VADD_IMPL(int64x1_t, int64x1_t, vadd, s64) |
| 47 | VADD_IMPL(float32x2_t, float32x2_t, vadd, f32) |
| 48 | #ifdef __ARM_FEATURE_FP16_VECTOR_ARITHMETIC |
| 49 | VADD_IMPL(float16x4_t, float16x4_t, vadd, f16) |
| 50 | #endif // __ARM_FEATURE_FP16_VECTOR_ARITHMETIC |
| 51 | |
| 52 | VADD_IMPL(uint8x16_t, uint8x16_t, vaddq, u8) |
| 53 | VADD_IMPL(int8x16_t, int8x16_t, vaddq, s8) |
| 54 | VADD_IMPL(uint16x8_t, uint16x8_t, vaddq, u16) |
| 55 | VADD_IMPL(int16x8_t, int16x8_t, vaddq, s16) |
| 56 | VADD_IMPL(uint32x4_t, uint32x4_t, vaddq, u32) |
| 57 | VADD_IMPL(int32x4_t, int32x4_t, vaddq, s32) |
| 58 | VADD_IMPL(uint64x2_t, uint64x2_t, vaddq, u64) |
| 59 | VADD_IMPL(int64x2_t, int64x2_t, vaddq, s64) |
| 60 | VADD_IMPL(float32x4_t, float32x4_t, vaddq, f32) |
| 61 | #ifdef __ARM_FEATURE_FP16_VECTOR_ARITHMETIC |
| 62 | VADD_IMPL(float16x8_t, float16x8_t, vaddq, f16) |
| 63 | #endif // __ARM_FEATURE_FP16_VECTOR_ARITHMETIC |
Michalis Spyrou | a6825a4 | 2018-09-13 12:24:03 +0100 | [diff] [blame] | 64 | #undef VADD_IMPL |
Georgios Pinitas | 5a59453 | 2018-12-03 14:30:05 +0000 | [diff] [blame] | 65 | |
Georgios Pinitas | dbdea0d | 2019-10-16 19:21:40 +0100 | [diff] [blame] | 66 | // VQADD: Vector saturating add (No notion of saturation for floating point) |
Georgios Pinitas | 5a59453 | 2018-12-03 14:30:05 +0000 | [diff] [blame] | 67 | #define VQADD_IMPL(stype, vtype, prefix, postfix) \ |
| 68 | inline vtype vqadd(const vtype &a, const vtype &b) \ |
| 69 | { \ |
| 70 | return prefix##_##postfix(a, b); \ |
| 71 | } |
| 72 | |
Georgios Pinitas | 5a59453 | 2018-12-03 14:30:05 +0000 | [diff] [blame] | 73 | VQADD_IMPL(uint8x8_t, uint8x8_t, vqadd, u8) |
| 74 | VQADD_IMPL(int8x8_t, int8x8_t, vqadd, s8) |
| 75 | VQADD_IMPL(uint16x4_t, uint16x4_t, vqadd, u16) |
| 76 | VQADD_IMPL(int16x4_t, int16x4_t, vqadd, s16) |
| 77 | VQADD_IMPL(uint32x2_t, uint32x2_t, vqadd, u32) |
| 78 | VQADD_IMPL(int32x2_t, int32x2_t, vqadd, s32) |
| 79 | VQADD_IMPL(uint64x1_t, uint64x1_t, vqadd, u64) |
| 80 | VQADD_IMPL(int64x1_t, int64x1_t, vqadd, s64) |
| 81 | VQADD_IMPL(float32x2_t, float32x2_t, vadd, f32) |
| 82 | #ifdef __ARM_FEATURE_FP16_VECTOR_ARITHMETIC |
| 83 | VQADD_IMPL(float16x4_t, float16x4_t, vadd, f16) |
| 84 | #endif // __ARM_FEATURE_FP16_VECTOR_ARITHMETIC |
| 85 | |
| 86 | VQADD_IMPL(uint8x16_t, uint8x16_t, vqaddq, u8) |
| 87 | VQADD_IMPL(int8x16_t, int8x16_t, vqaddq, s8) |
| 88 | VQADD_IMPL(uint16x8_t, uint16x8_t, vqaddq, u16) |
| 89 | VQADD_IMPL(int16x8_t, int16x8_t, vqaddq, s16) |
| 90 | VQADD_IMPL(uint32x4_t, uint32x4_t, vqaddq, u32) |
| 91 | VQADD_IMPL(int32x4_t, int32x4_t, vqaddq, s32) |
| 92 | VQADD_IMPL(uint64x2_t, uint64x2_t, vqaddq, u64) |
| 93 | VQADD_IMPL(int64x2_t, int64x2_t, vqaddq, s64) |
| 94 | VQADD_IMPL(float32x4_t, float32x4_t, vaddq, f32) |
| 95 | #ifdef __ARM_FEATURE_FP16_VECTOR_ARITHMETIC |
| 96 | VQADD_IMPL(float16x8_t, float16x8_t, vaddq, f16) |
| 97 | #endif // __ARM_FEATURE_FP16_VECTOR_ARITHMETIC |
| 98 | #undef VQADD_IMPL |
Georgios Pinitas | dbdea0d | 2019-10-16 19:21:40 +0100 | [diff] [blame] | 99 | |
| 100 | // VADDW: Vector widening add |
| 101 | #define VADDW_IMPL(wtype, vtype, prefix, postfix) \ |
| 102 | inline wtype vaddw(const wtype &a, const vtype &b) \ |
| 103 | { \ |
| 104 | return prefix##_##postfix(a, b); \ |
| 105 | } |
| 106 | |
| 107 | VADDW_IMPL(uint16x8_t, uint8x8_t, vaddw, u8) |
| 108 | VADDW_IMPL(int16x8_t, int8x8_t, vaddw, s8) |
| 109 | VADDW_IMPL(uint32x4_t, uint16x4_t, vaddw, u16) |
| 110 | VADDW_IMPL(int32x4_t, int16x4_t, vaddw, s16) |
| 111 | VADDW_IMPL(uint64x2_t, uint32x2_t, vaddw, u32) |
| 112 | VADDW_IMPL(int64x2_t, int32x2_t, vaddw, s32) |
| 113 | #undef VADDW_IMPL |
| 114 | |
| 115 | // VADDL: Vector long add |
| 116 | #define VADDL_IMPL(wtype, vtype, prefix, postfix) \ |
| 117 | inline wtype vaddl(const vtype &a, const vtype &b) \ |
| 118 | { \ |
| 119 | return prefix##_##postfix(a, b); \ |
| 120 | } |
| 121 | |
| 122 | VADDL_IMPL(uint16x8_t, uint8x8_t, vaddl, u8) |
| 123 | VADDL_IMPL(int16x8_t, int8x8_t, vaddl, s8) |
| 124 | VADDL_IMPL(uint32x4_t, uint16x4_t, vaddl, u16) |
| 125 | VADDL_IMPL(int32x4_t, int16x4_t, vaddl, s16) |
| 126 | VADDL_IMPL(uint64x2_t, uint32x2_t, vaddl, u32) |
| 127 | VADDL_IMPL(int64x2_t, int32x2_t, vaddl, s32) |
| 128 | #undef VADDL_IMPL |
| 129 | |
| 130 | #if defined(__aarch64__) |
| 131 | // VADDV: Across vector add |
| 132 | #define VADDV_IMPL(stype, vtype, prefix, postfix) \ |
| 133 | inline stype vaddv(const vtype &a) \ |
| 134 | { \ |
| 135 | return prefix##_##postfix(a); \ |
| 136 | } |
| 137 | |
| 138 | VADDV_IMPL(uint8_t, uint8x8_t, vaddv, u8) |
| 139 | VADDV_IMPL(int8_t, int8x8_t, vaddv, s8) |
| 140 | VADDV_IMPL(uint16_t, uint16x4_t, vaddv, u16) |
| 141 | VADDV_IMPL(int16_t, int16x4_t, vaddv, s16) |
| 142 | VADDV_IMPL(uint32_t, uint32x2_t, vaddv, u32) |
| 143 | VADDV_IMPL(int32_t, int32x2_t, vaddv, s32) |
| 144 | VADDV_IMPL(float, float32x2_t, vaddv, f32) |
| 145 | |
| 146 | VADDV_IMPL(uint8_t, uint8x16_t, vaddvq, u8) |
| 147 | VADDV_IMPL(int8_t, int8x16_t, vaddvq, s8) |
| 148 | VADDV_IMPL(uint16_t, uint16x8_t, vaddvq, u16) |
| 149 | VADDV_IMPL(int16_t, int16x8_t, vaddvq, s16) |
| 150 | VADDV_IMPL(uint32_t, uint32x4_t, vaddvq, u32) |
| 151 | VADDV_IMPL(int32_t, int32x4_t, vaddvq, s32) |
| 152 | VADDV_IMPL(uint64_t, uint64x2_t, vaddvq, u64) |
| 153 | VADDV_IMPL(int64_t, int64x2_t, vaddvq, s64) |
| 154 | VADDV_IMPL(float, float32x4_t, vaddvq, f32) |
| 155 | #undef VADDV_IMPL |
| 156 | #endif // defined(__aarch64__) |
| 157 | |
| 158 | // VPADDL: Signed add long pairwise |
| 159 | #define VPADDL_IMPL(ltype, vtype, prefix, postfix) \ |
| 160 | inline ltype vpaddl(const vtype &a) \ |
| 161 | { \ |
| 162 | return prefix##_##postfix(a); \ |
| 163 | } |
| 164 | |
| 165 | VPADDL_IMPL(uint16x4_t, uint8x8_t, vpaddl, u8) |
| 166 | VPADDL_IMPL(int16x4_t, int8x8_t, vpaddl, s8) |
| 167 | VPADDL_IMPL(uint32x2_t, uint16x4_t, vpaddl, u16) |
| 168 | VPADDL_IMPL(int32x2_t, int16x4_t, vpaddl, s16) |
| 169 | VPADDL_IMPL(uint64x1_t, uint32x2_t, vpaddl, u32) |
| 170 | VPADDL_IMPL(int64x1_t, int32x2_t, vpaddl, s32) |
| 171 | |
| 172 | VPADDL_IMPL(uint16x8_t, uint8x16_t, vpaddlq, u8) |
| 173 | VPADDL_IMPL(int16x8_t, int8x16_t, vpaddlq, s8) |
| 174 | VPADDL_IMPL(uint32x4_t, uint16x8_t, vpaddlq, u16) |
| 175 | VPADDL_IMPL(int32x4_t, int16x8_t, vpaddlq, s16) |
| 176 | VPADDL_IMPL(uint64x2_t, uint32x4_t, vpaddlq, u32) |
| 177 | VPADDL_IMPL(int64x2_t, int32x4_t, vpaddlq, s32) |
| 178 | #undef VPADDL_IMPL |
Manuel Bottini | 581f178 | 2019-11-13 17:24:43 +0000 | [diff] [blame] | 179 | |
| 180 | // VPADD: Add pairwise |
| 181 | #define VPADD_IMPL(stype, vtype, prefix, postfix) \ |
| 182 | inline vtype vpadd(const vtype &a, const vtype &b) \ |
| 183 | { \ |
| 184 | return prefix##_##postfix(a, b); \ |
| 185 | } |
| 186 | |
| 187 | VPADD_IMPL(uint8x8_t, uint8x8_t, vpadd, u8) |
| 188 | VPADD_IMPL(int8x8_t, int8x8_t, vpadd, s8) |
| 189 | VPADD_IMPL(uint16x4_t, uint16x4_t, vpadd, u16) |
| 190 | VPADD_IMPL(int16x4_t, int16x4_t, vpadd, s16) |
| 191 | VPADD_IMPL(uint32x2_t, uint32x2_t, vpadd, u32) |
| 192 | VPADD_IMPL(int32x2_t, int32x2_t, vpadd, s32) |
| 193 | VPADD_IMPL(float32x2_t, float32x2_t, vpadd, f32) |
| 194 | #ifdef __ARM_FEATURE_FP16_VECTOR_ARITHMETIC |
| 195 | VPADD_IMPL(float16x4_t, float16x4_t, vpadd, f16) |
| 196 | #endif // __ARM_FEATURE_FP16_VECTOR_ARITHMETIC |
| 197 | |
| 198 | #undef VPADD_IMPL |
Michalis Spyrou | a6825a4 | 2018-09-13 12:24:03 +0100 | [diff] [blame] | 199 | } // namespace wrapper |
| 200 | } // namespace arm_compute |
Michalis Spyrou | f464337 | 2019-11-29 16:17:13 +0000 | [diff] [blame] | 201 | #endif /* ARM_COMPUTE_WRAPPER_ADD_H */ |