Anthony Barbier | 6ff3b19 | 2017-09-04 18:44:23 +0100 | [diff] [blame] | 1 | /* |
| 2 | * Copyright (c) 2016, 2017 ARM Limited. |
| 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 | */ |
| 24 | #include "arm_compute/core/NEON/kernels/NEDepthConvertKernel.h" |
| 25 | |
| 26 | #include "arm_compute/core/Error.h" |
| 27 | #include "arm_compute/core/Helpers.h" |
| 28 | #include "arm_compute/core/ITensor.h" |
| 29 | #include "arm_compute/core/NEON/NEFixedPoint.h" |
| 30 | #include "arm_compute/core/TensorInfo.h" |
| 31 | #include "arm_compute/core/Validate.h" |
| 32 | |
| 33 | #include <arm_neon.h> |
| 34 | |
| 35 | using namespace arm_compute; |
| 36 | |
| 37 | namespace arm_compute |
| 38 | { |
| 39 | class Coordinates; |
| 40 | } // namespace arm_compute |
| 41 | |
| 42 | NEDepthConvertKernel::NEDepthConvertKernel() |
| 43 | : _policy(), _shift(0) |
| 44 | { |
| 45 | } |
| 46 | |
| 47 | void NEDepthConvertKernel::configure(const ITensor *input, ITensor *output, ConvertPolicy policy, uint32_t shift) |
| 48 | { |
| 49 | ARM_COMPUTE_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input, 1, DataType::U8, DataType::QS8, DataType::S16, DataType::U16, DataType::F32); |
| 50 | ARM_COMPUTE_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(output, 1, DataType::U8, DataType::QS8, DataType::S16, DataType::U16, DataType::U32, DataType::S32, DataType::F32); |
| 51 | ARM_COMPUTE_ERROR_ON(shift >= 8); |
| 52 | ARM_COMPUTE_ERROR_ON(input == output); |
| 53 | ARM_COMPUTE_ERROR_ON_MSG(input->info()->data_type() == output->info()->data_type(), "Input and output data_types must be different"); |
| 54 | |
| 55 | ARM_COMPUTE_ERROR_ON_MSG(input->info()->data_type() == DataType::QS8 && (output->info()->data_type() != DataType::F32), |
| 56 | "Only data_types supported [in] QS8 -> [out] F32"); |
| 57 | |
| 58 | ARM_COMPUTE_ERROR_ON_MSG(input->info()->data_type() == DataType::U8 && (output->info()->data_type() != DataType::S16 && output->info()->data_type() != DataType::U16 |
| 59 | && output->info()->data_type() != DataType::S32), |
| 60 | "Only data_types supported [in] U8 -> [out] U16, S16, S32"); |
| 61 | |
| 62 | ARM_COMPUTE_ERROR_ON_MSG(input->info()->data_type() == DataType::U16 && (output->info()->data_type() != DataType::U8 && output->info()->data_type() != DataType::U32), |
| 63 | "Only data_types supported [in] U16 -> [out] U8, U32"); |
| 64 | |
| 65 | ARM_COMPUTE_ERROR_ON_MSG(input->info()->data_type() == DataType::S16 && (output->info()->data_type() != DataType::U8 && output->info()->data_type() != DataType::S32), |
| 66 | "Only data_types supported [in] S16 -> [out] U8, S32"); |
| 67 | |
| 68 | ARM_COMPUTE_ERROR_ON_MSG(input->info()->data_type() == DataType::F32 && (output->info()->data_type() != DataType::QS8), |
| 69 | "Only data_types supported [in] F32 -> [out] QS8"); |
| 70 | |
| 71 | _policy = policy; |
| 72 | _shift = shift; |
| 73 | |
| 74 | constexpr unsigned int num_elems_processed_per_iteration = 16; |
| 75 | INESimpleKernel::configure(input, output, num_elems_processed_per_iteration); |
| 76 | } |
| 77 | |
| 78 | void NEDepthConvertKernel::run(const Window &window) |
| 79 | { |
| 80 | ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL(this); |
| 81 | ARM_COMPUTE_ERROR_ON_INVALID_SUBWINDOW(INESimpleKernel::window(), window); |
| 82 | ARM_COMPUTE_ERROR_ON(nullptr == _input); |
| 83 | ARM_COMPUTE_ERROR_ON(nullptr == _output); |
| 84 | ARM_COMPUTE_ERROR_ON(_input == _output); |
| 85 | |
| 86 | Iterator input(_input, window); |
| 87 | Iterator output(_output, window); |
| 88 | |
| 89 | switch(_input->info()->data_type()) |
| 90 | { |
| 91 | case DataType::QS8: |
| 92 | { |
| 93 | const int fixed_point_position = _input->info()->fixed_point_position(); |
| 94 | |
| 95 | switch(_output->info()->data_type()) |
| 96 | { |
| 97 | case DataType::F32: |
| 98 | { |
| 99 | /* Up-conversion QS8 -> F32 */ |
| 100 | execute_window_loop(window, [&](const Coordinates & id) |
| 101 | { |
| 102 | const int8x16_t texels_s8 = vld1q_s8(reinterpret_cast<const int8_t *>(input.ptr())); |
| 103 | |
| 104 | float32x4x2_t texels_low = vcvt_f32_qs8(vget_low_s8(texels_s8), fixed_point_position); |
| 105 | float32x4x2_t texels_high = vcvt_f32_qs8(vget_high_s8(texels_s8), fixed_point_position); |
| 106 | |
| 107 | vst1q_f32(reinterpret_cast<float *>(output.ptr()), texels_low.val[0]); |
| 108 | vst1q_f32(reinterpret_cast<float *>(output.ptr()) + 4, texels_low.val[1]); |
| 109 | vst1q_f32(reinterpret_cast<float *>(output.ptr()) + 8, texels_high.val[0]); |
| 110 | vst1q_f32(reinterpret_cast<float *>(output.ptr()) + 12, texels_high.val[1]); |
| 111 | }, |
| 112 | input, output); |
| 113 | break; |
| 114 | } |
| 115 | default: |
| 116 | ARM_COMPUTE_ERROR("Output data type not supported"); |
| 117 | } |
| 118 | break; |
| 119 | } |
| 120 | case DataType::U8: |
| 121 | { |
| 122 | const int16x8_t b = vdupq_n_s16(_shift); |
| 123 | |
| 124 | switch(_output->info()->data_type()) |
| 125 | { |
| 126 | case DataType::S16: |
| 127 | { |
| 128 | /* Up-conversion U8 -> S16 */ |
| 129 | execute_window_loop(window, [&](const Coordinates & id) |
| 130 | { |
| 131 | const uint8x16_t texels_u8 = vld1q_u8(input.ptr()); |
| 132 | |
| 133 | const int16x8x2_t texels = |
| 134 | { |
| 135 | { |
| 136 | vshlq_s16(vreinterpretq_s16_u16(vmovl_u8(vget_low_u8(texels_u8))), b), |
| 137 | vshlq_s16(vreinterpretq_s16_u16(vmovl_u8(vget_high_u8(texels_u8))), b) |
| 138 | } |
| 139 | }; |
| 140 | |
| 141 | vst1q_s16(reinterpret_cast<int16_t *>(output.ptr()), texels.val[0]); |
| 142 | vst1q_s16(reinterpret_cast<int16_t *>(output.ptr()) + 8, texels.val[1]); |
| 143 | }, |
| 144 | input, output); |
| 145 | break; |
| 146 | } |
| 147 | case DataType::S32: |
| 148 | { |
| 149 | /* Up-conversion U8 -> S32 */ |
| 150 | execute_window_loop(window, [&](const Coordinates & id) |
| 151 | { |
| 152 | const uint8x16_t texels_u8 = vld1q_u8(input.ptr()); |
| 153 | |
| 154 | const int16x8x2_t texels = |
| 155 | { |
| 156 | { |
| 157 | vshlq_s16(vreinterpretq_s16_u16(vmovl_u8(vget_low_u8(texels_u8))), b), |
| 158 | vshlq_s16(vreinterpretq_s16_u16(vmovl_u8(vget_high_u8(texels_u8))), b) |
| 159 | } |
| 160 | }; |
| 161 | |
| 162 | vst1q_s32(reinterpret_cast<int32_t *>(output.ptr()), vmovl_s16(vget_low_s16(texels.val[0]))); |
| 163 | vst1q_s32(reinterpret_cast<int32_t *>(output.ptr()) + 4, vmovl_s16(vget_high_s16(texels.val[0]))); |
| 164 | vst1q_s32(reinterpret_cast<int32_t *>(output.ptr()) + 8, vmovl_s16(vget_low_s16(texels.val[1]))); |
| 165 | vst1q_s32(reinterpret_cast<int32_t *>(output.ptr()) + 12, vmovl_s16(vget_high_s16(texels.val[1]))); |
| 166 | }, |
| 167 | input, output); |
| 168 | break; |
| 169 | } |
| 170 | case DataType::U16: |
| 171 | { |
| 172 | /* Up-conversion U8 -> U16 */ |
| 173 | execute_window_loop(window, [&](const Coordinates & id) |
| 174 | { |
| 175 | const uint8x16_t texels_u8 = vld1q_u8(input.ptr()); |
| 176 | |
| 177 | const uint16x8x2_t texels = |
| 178 | { |
| 179 | { |
| 180 | vshlq_u16(vmovl_u8(vget_low_u8(texels_u8)), b), |
| 181 | vshlq_u16(vmovl_u8(vget_high_u8(texels_u8)), b) |
| 182 | } |
| 183 | }; |
| 184 | |
| 185 | vst1q_u16(reinterpret_cast<uint16_t *>(output.ptr()), texels.val[0]); |
| 186 | vst1q_u16(reinterpret_cast<uint16_t *>(output.ptr()) + 8, texels.val[1]); |
| 187 | }, |
| 188 | input, output); |
| 189 | break; |
| 190 | } |
| 191 | default: |
| 192 | ARM_COMPUTE_ERROR("Output data type not supported"); |
| 193 | } |
| 194 | break; |
| 195 | } |
| 196 | case DataType::S16: |
| 197 | { |
| 198 | switch(_output->info()->data_type()) |
| 199 | { |
| 200 | case DataType::U8: |
| 201 | { |
| 202 | const int16x8_t b = vdupq_n_s16(-static_cast<int16_t>(_shift)); |
| 203 | |
| 204 | /* Down-conversion S16 -> U8 */ |
| 205 | if(ConvertPolicy::SATURATE == _policy) |
| 206 | { |
| 207 | execute_window_loop(window, [&](const Coordinates & id) |
| 208 | { |
| 209 | const int16x8x2_t texels = |
| 210 | { |
| 211 | { |
| 212 | vqshlq_s16(vld1q_s16(reinterpret_cast<int16_t *>(input.ptr())), b), |
| 213 | vqshlq_s16(vld1q_s16(reinterpret_cast<int16_t *>(input.ptr()) + 8), b) |
| 214 | } |
| 215 | }; |
| 216 | |
| 217 | vst1q_u8(output.ptr(), vcombine_u8(vqmovun_s16(texels.val[0]), vqmovun_s16(texels.val[1]))); |
| 218 | }, |
| 219 | input, output); |
| 220 | } |
| 221 | else |
| 222 | { |
| 223 | execute_window_loop(window, [&](const Coordinates & id) |
| 224 | { |
| 225 | const int16x8x2_t texels = |
| 226 | { |
| 227 | { |
| 228 | vshlq_s16(vld1q_s16(reinterpret_cast<int16_t *>(input.ptr())), b), |
| 229 | vshlq_s16(vld1q_s16(reinterpret_cast<int16_t *>(input.ptr()) + 8), b) |
| 230 | } |
| 231 | }; |
| 232 | |
| 233 | vst1q_u8(output.ptr(), vcombine_u8(vmovn_u16(vreinterpretq_u16_s16(texels.val[0])), |
| 234 | vmovn_u16(vreinterpretq_u16_s16(texels.val[1])))); |
| 235 | }, |
| 236 | input, output); |
| 237 | } |
| 238 | break; |
| 239 | } |
| 240 | case DataType::S32: |
| 241 | { |
| 242 | const int32x4_t b = vdupq_n_s32(_shift); |
| 243 | |
| 244 | /* Up-conversion S16 -> S32 */ |
| 245 | execute_window_loop(window, [&](const Coordinates & id) |
| 246 | { |
| 247 | const int16x8x2_t texels = |
| 248 | { |
| 249 | { |
| 250 | vld1q_s16(reinterpret_cast<int16_t *>(input.ptr())), |
| 251 | vld1q_s16(reinterpret_cast<int16_t *>(input.ptr()) + 8) |
| 252 | } |
| 253 | }; |
| 254 | |
| 255 | const int32x4x4_t texels_s32 = |
| 256 | { |
| 257 | { |
| 258 | vshlq_s32(vmovl_s16(vget_low_s16(texels.val[0])), b), |
| 259 | vshlq_s32(vmovl_s16(vget_high_s16(texels.val[0])), b), |
| 260 | vshlq_s32(vmovl_s16(vget_low_s16(texels.val[1])), b), |
| 261 | vshlq_s32(vmovl_s16(vget_high_s16(texels.val[1])), b) |
| 262 | } |
| 263 | }; |
| 264 | |
| 265 | vst1q_s32(reinterpret_cast<int32_t *>(output.ptr()), texels_s32.val[0]); |
| 266 | vst1q_s32(reinterpret_cast<int32_t *>(output.ptr()) + 4, texels_s32.val[1]); |
| 267 | vst1q_s32(reinterpret_cast<int32_t *>(output.ptr()) + 8, texels_s32.val[2]); |
| 268 | vst1q_s32(reinterpret_cast<int32_t *>(output.ptr()) + 12, texels_s32.val[3]); |
| 269 | }, |
| 270 | input, output); |
| 271 | break; |
| 272 | } |
| 273 | default: |
| 274 | ARM_COMPUTE_ERROR("Output data type not supported"); |
| 275 | } |
| 276 | break; |
| 277 | } |
| 278 | case DataType::U16: |
| 279 | { |
| 280 | switch(_output->info()->data_type()) |
| 281 | { |
| 282 | case DataType::U8: |
| 283 | { |
| 284 | const int16x8_t b = vdupq_n_s16(-static_cast<int16_t>(_shift)); |
| 285 | |
| 286 | /* Down-conversion U16 -> U8 */ |
| 287 | if(ConvertPolicy::SATURATE == _policy) |
| 288 | { |
| 289 | execute_window_loop(window, [&](const Coordinates & id) |
| 290 | { |
| 291 | const uint16x8x2_t texels = |
| 292 | { |
| 293 | { |
| 294 | vqshlq_u16(vld1q_u16(reinterpret_cast<uint16_t *>(input.ptr())), b), |
| 295 | vqshlq_u16(vld1q_u16(reinterpret_cast<uint16_t *>(input.ptr()) + 8), b) |
| 296 | } |
| 297 | }; |
| 298 | |
| 299 | vst1q_u8(output.ptr(), vcombine_u8(vqmovn_u16(texels.val[0]), vqmovn_u16(texels.val[1]))); |
| 300 | }, |
| 301 | input, output); |
| 302 | } |
| 303 | else |
| 304 | { |
| 305 | execute_window_loop(window, [&](const Coordinates & id) |
| 306 | { |
| 307 | const uint16x8x2_t texels = |
| 308 | { |
| 309 | { |
| 310 | vshlq_u16(vld1q_u16(reinterpret_cast<uint16_t *>(input.ptr())), b), |
| 311 | vshlq_u16(vld1q_u16(reinterpret_cast<uint16_t *>(input.ptr()) + 8), b) |
| 312 | } |
| 313 | }; |
| 314 | |
| 315 | vst1q_u8(output.ptr(), vcombine_u8(vmovn_u16(texels.val[0]), vmovn_u16(texels.val[1]))); |
| 316 | }, |
| 317 | input, output); |
| 318 | } |
| 319 | break; |
| 320 | } |
| 321 | case DataType::U32: |
| 322 | { |
| 323 | const int32x4_t b = vdupq_n_s32(_shift); |
| 324 | |
| 325 | /* Up-conversion U16 -> U32 */ |
| 326 | execute_window_loop(window, [&](const Coordinates & id) |
| 327 | { |
| 328 | const uint16x8x2_t texels = |
| 329 | { |
| 330 | { |
| 331 | vld1q_u16(reinterpret_cast<uint16_t *>(input.ptr())), |
| 332 | vld1q_u16(reinterpret_cast<uint16_t *>(input.ptr()) + 8) |
| 333 | } |
| 334 | }; |
| 335 | |
| 336 | vst1q_u32(reinterpret_cast<uint32_t *>(output.ptr()), vshlq_u32(vmovl_u16(vget_low_u16(texels.val[0])), b)); |
| 337 | vst1q_u32(reinterpret_cast<uint32_t *>(output.ptr()) + 4, vshlq_u32(vmovl_u16(vget_high_u16(texels.val[0])), b)); |
| 338 | vst1q_u32(reinterpret_cast<uint32_t *>(output.ptr()) + 8, vshlq_u32(vmovl_u16(vget_low_u16(texels.val[1])), b)); |
| 339 | vst1q_u32(reinterpret_cast<uint32_t *>(output.ptr()) + 12, vshlq_u32(vmovl_u16(vget_high_u16(texels.val[1])), b)); |
| 340 | }, |
| 341 | input, output); |
| 342 | break; |
| 343 | } |
| 344 | default: |
| 345 | ARM_COMPUTE_ERROR("Output data type not supported"); |
| 346 | } |
| 347 | break; |
| 348 | } |
| 349 | case DataType::F32: |
| 350 | { |
| 351 | switch(_output->info()->data_type()) |
| 352 | { |
| 353 | case DataType::QS8: |
| 354 | { |
| 355 | const int fixed_point_position = _output->info()->fixed_point_position(); |
| 356 | /* Down-conversion F32 -> QS8 */ |
| 357 | execute_window_loop(window, [&](const Coordinates & id) |
| 358 | { |
| 359 | const float32x4x4_t texels_f32 = |
| 360 | { |
| 361 | { |
| 362 | vld1q_f32(reinterpret_cast<const float *>(input.ptr())), |
| 363 | vld1q_f32(reinterpret_cast<const float *>(input.ptr()) + 4), |
| 364 | vld1q_f32(reinterpret_cast<const float *>(input.ptr()) + 8), |
| 365 | vld1q_f32(reinterpret_cast<const float *>(input.ptr()) + 12) |
| 366 | } |
| 367 | }; |
| 368 | |
| 369 | const qint8x16_t texels_s8 = vcvtq_qs8_f32(texels_f32, fixed_point_position); |
| 370 | |
| 371 | vst1q_s8(reinterpret_cast<int8_t *>(output.ptr()), texels_s8); |
| 372 | }, |
| 373 | input, output); |
| 374 | break; |
| 375 | } |
| 376 | default: |
| 377 | ARM_COMPUTE_ERROR("Output data type not supported"); |
| 378 | } |
| 379 | break; |
| 380 | } |
| 381 | default: |
| 382 | ARM_COMPUTE_ERROR("Not supported"); |
| 383 | } |
| 384 | } |