George Wort | 05398a9 | 2019-01-25 15:38:33 +0000 | [diff] [blame] | 1 | /* |
| 2 | * Copyright (c) 2019 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/NECropKernel.h" |
| 25 | |
| 26 | #include "arm_compute/core/CPP/Validate.h" |
| 27 | #include "arm_compute/core/IAccessWindow.h" |
| 28 | #include "arm_compute/core/ITensor.h" |
| 29 | #include "arm_compute/core/TensorInfo.h" |
| 30 | #include "arm_compute/core/Window.h" |
| 31 | |
| 32 | #include "arm_compute/core/NEON/wrapper/wrapper.h" |
| 33 | #include "arm_compute/core/Types.h" |
| 34 | #include "arm_compute/core/utils/helpers/bit_ops.h" |
| 35 | #include "arm_compute/core/utils/helpers/tensor_transform.h" |
| 36 | #include "arm_compute/core/utils/misc/ShapeCalculator.h" |
| 37 | |
| 38 | #include <map> |
| 39 | |
| 40 | namespace arm_compute |
| 41 | { |
| 42 | namespace |
| 43 | { |
| 44 | template <typename T> |
| 45 | inline float32x4_t load_as_f32(T *ptr) |
| 46 | { |
| 47 | ARM_COMPUTE_UNUSED(ptr); |
| 48 | ARM_COMPUTE_ERROR("Type not supported."); |
| 49 | } |
| 50 | |
| 51 | template <> |
| 52 | inline float32x4_t load_as_f32(float *ptr) |
| 53 | { |
| 54 | return wrapper::vloadq(ptr); |
| 55 | } |
| 56 | |
| 57 | template <> |
| 58 | inline float32x4_t load_as_f32(int32_t *ptr) |
| 59 | { |
| 60 | return vcvtq_f32_s32(wrapper::vloadq(ptr)); |
| 61 | } |
| 62 | |
| 63 | template <> |
| 64 | inline float32x4_t load_as_f32(uint32_t *ptr) |
| 65 | { |
| 66 | return vcvtq_f32_u32(wrapper::vloadq(ptr)); |
| 67 | } |
| 68 | |
| 69 | template <> |
| 70 | inline float32x4_t load_as_f32(int16_t *ptr) |
| 71 | { |
| 72 | return vcvtq_f32_s32(vmovl_s16(wrapper::vload(ptr))); |
| 73 | } |
| 74 | |
| 75 | template <> |
| 76 | inline float32x4_t load_as_f32(uint16_t *ptr) |
| 77 | { |
| 78 | return vcvtq_f32_u32(vmovl_u16(wrapper::vload(ptr))); |
| 79 | } |
| 80 | |
| 81 | #ifdef __ARM_FEATURE_FP16_VECTOR_ARITHMETIC |
| 82 | template <> |
| 83 | inline float32x4_t load_as_f32(float16_t *ptr) |
| 84 | { |
| 85 | return vcvt_f32_f16(wrapper::vload(ptr)); |
| 86 | } |
| 87 | #endif /* __ARM_FEATURE_FP16_VECTOR_ARITHMETIC */ |
| 88 | |
| 89 | template <typename T, bool input_has_single_channel, bool is_width_flipped> |
| 90 | inline void in_bounds_crop_window(const ITensor *input, const ITensor *output, float *output_ptr, Coordinates input_offset, |
| 91 | int32_t window_step_x, int32_t output_width_start, int32_t output_width_limit) |
| 92 | { |
| 93 | // Reverse elements if width flipped. |
| 94 | if(is_width_flipped) |
| 95 | { |
| 96 | // Collapse first dimension if possible. |
| 97 | if(input_has_single_channel) |
| 98 | { |
| 99 | int32_t x = output_width_start; |
| 100 | Coordinates negative_offset(input_offset); |
| 101 | negative_offset.set(1, negative_offset[1] - window_step_x + 1); |
| 102 | for(; x <= output_width_limit - window_step_x; x += window_step_x, negative_offset[1] -= window_step_x) |
| 103 | { |
| 104 | auto in = load_as_f32(reinterpret_cast<T *>(input->ptr_to_element(negative_offset))); |
| 105 | |
| 106 | in = wrapper::vrev64(in); |
| 107 | in = wrapper::vcombine(wrapper::vgethigh(in), wrapper::vgetlow(in)); |
| 108 | |
| 109 | wrapper::vstore(output_ptr + x, in); |
| 110 | } |
| 111 | input_offset[1] = negative_offset[1] + window_step_x - 1; |
| 112 | for(; x < output_width_limit; ++x, --input_offset[1]) |
| 113 | { |
| 114 | *(output_ptr + x) = static_cast<float>(*reinterpret_cast<T *>(input->ptr_to_element(input_offset))); |
| 115 | } |
| 116 | } |
| 117 | else |
| 118 | { |
| 119 | for(int32_t x = output_width_start; x < output_width_limit; ++x, --input_offset[1]) |
| 120 | { |
| 121 | input_offset.set(0, 0); |
| 122 | int32_t c = 0; |
| 123 | for(; c <= static_cast<int32_t>(input->info()->dimension(0)) - window_step_x; c += window_step_x, input_offset[0] += window_step_x) |
| 124 | { |
| 125 | auto in = load_as_f32(reinterpret_cast<T *>(input->ptr_to_element(input_offset))); |
| 126 | wrapper::vstore(output_ptr + x * output->info()->dimension(0) + c, in); |
| 127 | } |
| 128 | for(; c < static_cast<int32_t>(input->info()->dimension(0)); ++c, ++input_offset[0]) |
| 129 | { |
| 130 | *(output_ptr + x * output->info()->dimension(0) + c) = static_cast<float>(*reinterpret_cast<T *>(input->ptr_to_element(input_offset))); |
| 131 | } |
| 132 | } |
| 133 | } |
| 134 | } |
| 135 | else |
| 136 | { |
| 137 | // Use memcpy if the elements don't need converting to float. |
| 138 | if(std::is_same<T, float>::value) |
| 139 | { |
| 140 | memcpy(static_cast<void *>(output_ptr + output_width_start * output->info()->dimension(0)), |
| 141 | reinterpret_cast<const void *>(input->ptr_to_element(input_offset)), |
| 142 | (output_width_limit - output_width_start) * output->info()->dimension(0) * output->info()->element_size()); |
| 143 | } |
| 144 | else |
| 145 | { |
| 146 | int32_t x = 0; |
| 147 | int32_t limit = (output_width_limit - output_width_start) * static_cast<int32_t>(output->info()->dimension(0)); |
| 148 | float *output_start_ptr = output_ptr + output_width_start * output->info()->dimension(0); |
| 149 | for(; x <= limit - window_step_x; x += window_step_x, input_offset[0] += window_step_x) |
| 150 | { |
| 151 | auto in = load_as_f32(reinterpret_cast<T *>(input->ptr_to_element(input_offset))); |
| 152 | wrapper::vstore(output_start_ptr + x, in); |
| 153 | } |
| 154 | for(; x < limit; ++x, ++input_offset[0]) |
| 155 | { |
| 156 | *(output_start_ptr + x) = static_cast<float>(*reinterpret_cast<T *>(input->ptr_to_element(input_offset))); |
| 157 | } |
| 158 | } |
| 159 | } |
| 160 | } |
| 161 | |
| 162 | inline void out_of_bounds_crop_window(const ITensor *output, float *output_ptr, float extrapolation_value, |
| 163 | int32_t window_step_x, int32_t output_width_start, int32_t output_width_limit) |
| 164 | { |
| 165 | auto in = wrapper::vdup_n(extrapolation_value, wrapper::traits::vector_128_tag()); |
| 166 | int32_t x = 0; |
| 167 | int32_t limit = (output_width_limit - output_width_start) * static_cast<int32_t>(output->info()->dimension(0)); |
| 168 | float *output_start_ptr = output_ptr + output_width_start * output->info()->dimension(0); |
| 169 | for(; x <= limit - window_step_x; x += window_step_x) |
| 170 | { |
| 171 | wrapper::vstore(output_start_ptr + x, in); |
| 172 | } |
| 173 | for(; x < limit; ++x) |
| 174 | { |
| 175 | *(output_start_ptr + x) = extrapolation_value; |
| 176 | } |
| 177 | } |
| 178 | |
| 179 | template <bool is_height_flipped, bool has_cols_in_bounds, bool has_cols_out_of_bounds_before, bool has_cols_out_of_bounds_after> |
| 180 | inline void execute_window(const ITensor *input, const ITensor *output, Coordinates input_offset, float extrapolation_value, |
Michalis Spyrou | bcfd09a | 2019-05-01 13:03:59 +0100 | [diff] [blame] | 181 | const std::array<uint32_t, 2> &rows_out_of_bounds, const std::array<uint32_t, 2> &cols_out_of_bounds, NECropKernel::InBoundsCropFunction *in_bounds_crop_function) |
George Wort | 05398a9 | 2019-01-25 15:38:33 +0000 | [diff] [blame] | 182 | { |
| 183 | // Output is always float. |
| 184 | const int window_step_x = 16 / sizeof(float); |
| 185 | auto *output_ptr = reinterpret_cast<float *>(output->buffer()); |
| 186 | // Output window: |
| 187 | // -------------------------------- |
| 188 | // | Out of bounds | |
| 189 | // | rows before | |
| 190 | // |------------------------------| |
| 191 | // | Out of | In | Out of | |
| 192 | // | bounds | bounds | bounds | |
| 193 | // | cols | elements | cols | |
| 194 | // | before | copied | after | |
| 195 | // | | from input | | |
| 196 | // -------------------------------- |
| 197 | // | Out of bounds | |
| 198 | // | rows after | |
| 199 | // |------------------------------| |
| 200 | // Fill all output rows that have no elements that are within the input bounds with the extrapolation value. |
| 201 | // First for the rows before the in bounds rows. |
| 202 | out_of_bounds_crop_window(output, output_ptr, extrapolation_value, window_step_x, 0, rows_out_of_bounds[0] * output->info()->dimension(1)); |
| 203 | output_ptr += rows_out_of_bounds[0] * output->info()->dimension(1) * output->info()->dimension(0); |
| 204 | // Iterate through each row that has any elements within the input bounds. |
| 205 | for(uint32_t row = rows_out_of_bounds[0]; static_cast<int32_t>(row) < static_cast<int32_t>(output->info()->dimension(2) - rows_out_of_bounds[1]); |
| 206 | ++row, is_height_flipped ? --input_offset[2] : ++input_offset[2]) |
| 207 | { |
| 208 | // Fill all elements in the row that are out of bounds with the extrapolation value. |
| 209 | // First for the elements before the in bounds elements. |
| 210 | if(has_cols_out_of_bounds_before) |
| 211 | { |
| 212 | out_of_bounds_crop_window(output, output_ptr, extrapolation_value, window_step_x, 0, cols_out_of_bounds[0]); |
| 213 | } |
| 214 | // Copy all elements within the input bounds from the input tensor. |
| 215 | if(has_cols_in_bounds) |
| 216 | { |
| 217 | (*in_bounds_crop_function)(input, output, output_ptr, input_offset, window_step_x, cols_out_of_bounds[0], output->info()->dimension(1) - cols_out_of_bounds[1]); |
| 218 | } |
| 219 | // Fill all elements after the in bounds elements with the extrapolation value. |
| 220 | if(has_cols_out_of_bounds_after) |
| 221 | { |
| 222 | out_of_bounds_crop_window(output, output_ptr, extrapolation_value, window_step_x, output->info()->dimension(1) - cols_out_of_bounds[1], output->info()->dimension(1)); |
| 223 | } |
| 224 | output_ptr += output->info()->dimension(1) * output->info()->dimension(0); |
| 225 | } |
| 226 | // Fill all rows after the in bounds elements with the extrapolation value. |
| 227 | out_of_bounds_crop_window(output, output_ptr, extrapolation_value, window_step_x, 0, rows_out_of_bounds[1] * output->info()->dimension(1)); |
| 228 | } |
| 229 | } // namespace |
| 230 | |
| 231 | NECropKernel::NECropKernel() |
| 232 | : _input(nullptr), _crop_boxes(nullptr), _box_ind(nullptr), _output(nullptr), _start(), _end(), _crop_box_ind(0), _extrapolation_value(0), _rows_out_of_bounds(), _cols_out_of_bounds(), |
| 233 | _in_bounds_crop_functions(), _in_bounds_crop_function(nullptr), _crop_function(nullptr) |
| 234 | { |
| 235 | } |
| 236 | |
| 237 | void NECropKernel::configure(const ITensor *input, const ITensor *crop_boxes, const ITensor *box_ind, ITensor *output, uint32_t crop_box_ind, float extrapolation_value) |
| 238 | { |
| 239 | ARM_COMPUTE_ERROR_ON_NULLPTR(input, output); |
| 240 | ARM_COMPUTE_ERROR_THROW_ON(validate(input->info(), crop_boxes->info(), box_ind->info(), output->info(), crop_box_ind, extrapolation_value)); |
| 241 | |
| 242 | _input = input; |
| 243 | _crop_boxes = crop_boxes; |
| 244 | _box_ind = box_ind; |
| 245 | _output = output; |
| 246 | _crop_box_ind = crop_box_ind; |
| 247 | _extrapolation_value = extrapolation_value; |
| 248 | |
| 249 | const static std::map<std::pair<DataType, bool>, std::pair<NECropKernel::InBoundsCropFunction *, NECropKernel::InBoundsCropFunction *>> in_map_function = |
| 250 | { |
| 251 | { { DataType::F32, false }, { &in_bounds_crop_window<float, false, false>, &in_bounds_crop_window<float, false, true> } }, |
| 252 | { { DataType::F32, true }, { &in_bounds_crop_window<float, true, false>, &in_bounds_crop_window<float, true, true> } }, |
| 253 | { { DataType::U16, false }, { &in_bounds_crop_window<uint16_t, false, false>, &in_bounds_crop_window<uint16_t, false, true> } }, |
| 254 | { { DataType::U16, true }, { &in_bounds_crop_window<uint16_t, true, false>, &in_bounds_crop_window<uint16_t, true, true> } }, |
| 255 | { { DataType::S16, false }, { &in_bounds_crop_window<int16_t, false, false>, &in_bounds_crop_window<int16_t, false, true> } }, |
| 256 | { { DataType::S16, true }, { &in_bounds_crop_window<int16_t, true, false>, &in_bounds_crop_window<int16_t, true, true> } }, |
| 257 | { { DataType::U32, false }, { &in_bounds_crop_window<uint32_t, false, false>, &in_bounds_crop_window<uint32_t, false, true> } }, |
| 258 | { { DataType::U32, true }, { &in_bounds_crop_window<uint32_t, true, false>, &in_bounds_crop_window<uint32_t, true, true> } }, |
| 259 | { { DataType::S32, false }, { &in_bounds_crop_window<int32_t, false, false>, &in_bounds_crop_window<int32_t, false, true> } }, |
| 260 | { { DataType::S32, true }, { &in_bounds_crop_window<int32_t, true, false>, &in_bounds_crop_window<int32_t, true, true> } }, |
| 261 | #ifdef __ARM_FEATURE_FP16_VECTOR_ARITHMETIC |
| 262 | { { DataType::F16, false }, { &in_bounds_crop_window<float16_t, false, false>, &in_bounds_crop_window<float16_t, false, true> } }, |
| 263 | { { DataType::F16, false }, { &in_bounds_crop_window<float16_t, true, false>, &in_bounds_crop_window<float16_t, true, true> } } |
| 264 | #endif /* __ARM_FEATURE_FP16_VECTOR_ARITHMETIC */ |
| 265 | }; |
| 266 | |
| 267 | auto in_it = in_map_function.find({ input->info()->data_type(), input->info()->dimension(0) == 1 }); |
| 268 | |
| 269 | if(in_it != in_map_function.end()) |
| 270 | { |
| 271 | _in_bounds_crop_functions = in_it->second; |
| 272 | } |
| 273 | } |
| 274 | |
| 275 | Status NECropKernel::validate(const ITensorInfo *input, const ITensorInfo *crop_boxes, const ITensorInfo *box_ind, const ITensorInfo *output, uint32_t crop_box_ind, float extrapolation_value) |
| 276 | { |
| 277 | ARM_COMPUTE_UNUSED(extrapolation_value); |
| 278 | ARM_COMPUTE_RETURN_ERROR_ON_CPU_F16_UNSUPPORTED(input); |
| 279 | ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input, 1, DataType::U16, DataType::S16, DataType::F16, DataType::U32, DataType::S32, DataType::F32); |
| 280 | ARM_COMPUTE_RETURN_ERROR_ON_DATA_LAYOUT_NOT_IN(input, DataLayout::NHWC); |
| 281 | ARM_COMPUTE_RETURN_ERROR_ON(input->tensor_shape().num_dimensions() > 4); |
| 282 | ARM_COMPUTE_RETURN_ERROR_ON(crop_boxes->tensor_shape()[0] != 4); |
| 283 | ARM_COMPUTE_RETURN_ERROR_ON(crop_boxes->tensor_shape()[1] != box_ind->tensor_shape()[0]); |
| 284 | ARM_COMPUTE_RETURN_ERROR_ON(crop_boxes->tensor_shape()[1] <= crop_box_ind); |
| 285 | ARM_COMPUTE_RETURN_ERROR_ON(box_ind->tensor_shape()[0] <= crop_box_ind); |
| 286 | if(output->total_size() > 0) |
| 287 | { |
| 288 | ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_NOT_IN(output, DataType::F32); |
| 289 | ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_LAYOUT(input, output); |
| 290 | ARM_COMPUTE_RETURN_ERROR_ON(output->num_dimensions() != 3); |
| 291 | ARM_COMPUTE_RETURN_ERROR_ON(output->has_padding()); |
| 292 | } |
| 293 | return Status{}; |
| 294 | } |
| 295 | |
| 296 | void NECropKernel::configure_output_shape() |
| 297 | { |
| 298 | // _crop_box_ind is used to index _crop_boxes and retrieve the appropriate crop box. |
| 299 | // The crop box is specified by normalized coordinates [y0, x0, y1, x1]. |
| 300 | const float x0 = *reinterpret_cast<const float *>(_crop_boxes->ptr_to_element(Coordinates(1, _crop_box_ind))); |
| 301 | const float y0 = *reinterpret_cast<const float *>(_crop_boxes->ptr_to_element(Coordinates(0, _crop_box_ind))); |
| 302 | const float x1 = *reinterpret_cast<const float *>(_crop_boxes->ptr_to_element(Coordinates(3, _crop_box_ind))); |
| 303 | const float y1 = *reinterpret_cast<const float *>(_crop_boxes->ptr_to_element(Coordinates(2, _crop_box_ind))); |
| 304 | // The normalized coordiantes are scaled to retrieve the floating point image coordinates which are rounded to integers. |
| 305 | _start = Coordinates(std::floor(x0 * (_input->info()->tensor_shape()[1] - 1) + 0.5f), |
| 306 | std::floor(y0 * (_input->info()->tensor_shape()[2] - 1) + 0.5f)); |
| 307 | _end = Coordinates(std::floor(x1 * (_input->info()->tensor_shape()[1] - 1) + 0.5f), |
| 308 | std::floor(y1 * (_input->info()->tensor_shape()[2] - 1) + 0.5f)); |
| 309 | const TensorShape out_shape(_input->info()->tensor_shape()[0], abs(_end[0] - _start[0]) + 1, abs(_end[1] - _start[1]) + 1); |
| 310 | _output->info()->set_tensor_shape(out_shape); |
| 311 | |
| 312 | _in_bounds_crop_function = _start[0] <= _end[0] ? _in_bounds_crop_functions.first : _in_bounds_crop_functions.second; |
| 313 | |
| 314 | bool is_width_flipped = _end[0] < _start[0]; |
| 315 | bool is_height_flipped = _end[1] < _start[1]; |
| 316 | if(is_height_flipped) |
| 317 | { |
| 318 | _rows_out_of_bounds[0] = _start[1] >= static_cast<int32_t>(_input->info()->dimension(2)) ? std::min(static_cast<uint32_t>(_start[1] - _input->info()->dimension(2) + 1), |
| 319 | static_cast<uint32_t>(_output->info()->dimension(2))) : |
| 320 | 0; |
| 321 | _rows_out_of_bounds[1] = _end[1] < 0 ? std::min(static_cast<uint32_t>(-_end[1]), |
| 322 | static_cast<uint32_t>(_output->info()->dimension(2))) : |
| 323 | 0; |
| 324 | } |
| 325 | else |
| 326 | { |
| 327 | _rows_out_of_bounds[0] = _start[1] < 0 ? std::min(static_cast<uint32_t>(-_start[1]), |
| 328 | static_cast<uint32_t>(_output->info()->dimension(2))) : |
| 329 | 0; |
| 330 | _rows_out_of_bounds[1] = _end[1] >= static_cast<int32_t>(_input->info()->dimension(2)) ? std::min(static_cast<uint32_t>(_end[1] - _input->info()->dimension(2) + 1), |
| 331 | static_cast<uint32_t>(_output->info()->dimension(2))) : |
| 332 | 0; |
| 333 | } |
| 334 | if(is_width_flipped) |
| 335 | { |
| 336 | _cols_out_of_bounds[0] = _start[0] >= static_cast<int32_t>(_input->info()->dimension(1)) ? std::min(static_cast<uint32_t>(_start[0] - _input->info()->dimension(1) + 1), |
| 337 | static_cast<uint32_t>(_output->info()->dimension(1))) : |
| 338 | 0; |
| 339 | _cols_out_of_bounds[1] = _end[0] < 0 ? std::min(static_cast<uint32_t>(-_end[0]), |
| 340 | static_cast<uint32_t>(_output->info()->dimension(1))) : |
| 341 | 0; |
| 342 | } |
| 343 | else |
| 344 | { |
| 345 | _cols_out_of_bounds[0] = _start[0] < 0 ? std::min(static_cast<uint32_t>(-_start[0]), |
| 346 | static_cast<uint32_t>(_output->info()->dimension(1))) : |
| 347 | 0; |
| 348 | _cols_out_of_bounds[1] = _end[0] >= static_cast<int32_t>(_input->info()->dimension(1)) ? std::min(static_cast<uint32_t>(_end[0] - _input->info()->dimension(1) + 1), |
| 349 | static_cast<uint32_t>(_output->info()->dimension(1))) : |
| 350 | 0; |
| 351 | } |
| 352 | |
| 353 | const static std::map<std::tuple<bool, bool, bool, bool>, NECropKernel::CropFunction *> map_function = |
| 354 | { |
| 355 | { std::make_tuple(false, false, false, false), &execute_window<false, false, false, false> }, |
| 356 | { std::make_tuple(false, false, false, true), &execute_window<false, false, false, true> }, |
| 357 | { std::make_tuple(false, false, true, false), &execute_window<false, false, true, false> }, |
| 358 | { std::make_tuple(false, false, true, true), &execute_window<false, false, true, true> }, |
| 359 | { std::make_tuple(false, true, false, false), &execute_window<false, true, false, false> }, |
| 360 | { std::make_tuple(false, true, false, true), &execute_window<false, true, false, true> }, |
| 361 | { std::make_tuple(false, true, true, false), &execute_window<false, true, true, false> }, |
| 362 | { std::make_tuple(false, true, true, true), &execute_window<false, true, true, true> }, |
| 363 | { std::make_tuple(true, false, false, false), &execute_window<true, false, false, false> }, |
| 364 | { std::make_tuple(true, false, false, true), &execute_window<true, false, false, true> }, |
| 365 | { std::make_tuple(true, false, true, false), &execute_window<true, false, true, false> }, |
| 366 | { std::make_tuple(true, false, true, true), &execute_window<true, false, true, true> }, |
| 367 | { std::make_tuple(true, true, false, false), &execute_window<true, true, false, false> }, |
| 368 | { std::make_tuple(true, true, false, true), &execute_window<true, true, false, true> }, |
| 369 | { std::make_tuple(true, true, true, false), &execute_window<true, true, true, false> }, |
| 370 | { std::make_tuple(true, true, true, true), &execute_window<true, true, true, true> }, |
| 371 | }; |
| 372 | |
| 373 | auto it = map_function.find(std::make_tuple(is_height_flipped, |
| 374 | _cols_out_of_bounds[0] + _cols_out_of_bounds[1] < _output->info()->dimension(1), |
| 375 | _cols_out_of_bounds[0] > 0, |
| 376 | _cols_out_of_bounds[1] > 0)); |
| 377 | |
| 378 | if(it != map_function.end()) |
| 379 | { |
| 380 | _crop_function = it->second; |
| 381 | } |
| 382 | |
| 383 | INEKernel::configure(calculate_max_window(*_output->info())); |
| 384 | } |
| 385 | |
| 386 | void NECropKernel::run(const Window &window, const ThreadInfo &info) |
| 387 | { |
| 388 | ARM_COMPUTE_UNUSED(window, info); |
| 389 | ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL(this); |
| 390 | ARM_COMPUTE_ERROR_ON_INVALID_SUBWINDOW(INEKernel::window(), window); |
| 391 | |
| 392 | ARM_COMPUTE_ERROR_ON(_input->info()->has_padding()); |
| 393 | ARM_COMPUTE_ERROR_ON(_output->info()->has_padding()); |
| 394 | |
| 395 | uint32_t batch_index = *(reinterpret_cast<int32_t *>(_box_ind->ptr_to_element(Coordinates(_crop_box_ind)))); |
| 396 | Coordinates input_offset(0, _end[0] < _start[0] ? _start[0] - _cols_out_of_bounds[0] : _start[0] + _cols_out_of_bounds[0], |
| 397 | _end[1] < _start[1] ? _start[1] - _rows_out_of_bounds[0] : _start[1] + _rows_out_of_bounds[0], batch_index); |
| 398 | (*_crop_function)(_input, _output, input_offset, _extrapolation_value, _rows_out_of_bounds, _cols_out_of_bounds, _in_bounds_crop_function); |
| 399 | } |
| 400 | } // namespace arm_compute |