Anthony Barbier | 6ff3b19 | 2017-09-04 18:44:23 +0100 | [diff] [blame] | 1 | /* |
Giorgio Arena | a66eaa2 | 2017-12-21 19:50:06 +0000 | [diff] [blame] | 2 | * Copyright (c) 2016-2018 ARM Limited. |
Anthony Barbier | 6ff3b19 | 2017-09-04 18:44:23 +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 | */ |
| 24 | #ifndef __UTILS_UTILS_H__ |
| 25 | #define __UTILS_UTILS_H__ |
| 26 | |
| 27 | #include "arm_compute/core/Helpers.h" |
| 28 | #include "arm_compute/core/ITensor.h" |
| 29 | #include "arm_compute/core/Types.h" |
| 30 | #include "arm_compute/core/Validate.h" |
steniu01 | bee466b | 2017-06-21 16:45:41 +0100 | [diff] [blame] | 31 | #include "arm_compute/core/Window.h" |
Anthony Barbier | 6ff3b19 | 2017-09-04 18:44:23 +0100 | [diff] [blame] | 32 | #include "arm_compute/runtime/Tensor.h" |
Giorgio Arena | cf3935f | 2017-10-26 17:14:13 +0100 | [diff] [blame] | 33 | #include "libnpy/npy.hpp" |
Anthony Barbier | 2a07e18 | 2017-08-04 18:20:27 +0100 | [diff] [blame] | 34 | #include "support/ToolchainSupport.h" |
Anthony Barbier | 6ff3b19 | 2017-09-04 18:44:23 +0100 | [diff] [blame] | 35 | |
| 36 | #ifdef ARM_COMPUTE_CL |
| 37 | #include "arm_compute/core/CL/OpenCL.h" |
Isabella Gottardi | 02aabcc | 2017-10-12 17:28:51 +0100 | [diff] [blame] | 38 | #include "arm_compute/runtime/CL/CLDistribution1D.h" |
Anthony Barbier | 6ff3b19 | 2017-09-04 18:44:23 +0100 | [diff] [blame] | 39 | #include "arm_compute/runtime/CL/CLTensor.h" |
| 40 | #endif /* ARM_COMPUTE_CL */ |
Anthony Barbier | 7068f99 | 2017-10-26 15:23:08 +0100 | [diff] [blame] | 41 | #ifdef ARM_COMPUTE_GC |
| 42 | #include "arm_compute/runtime/GLES_COMPUTE/GCTensor.h" |
| 43 | #endif /* ARM_COMPUTE_GC */ |
Anthony Barbier | 6ff3b19 | 2017-09-04 18:44:23 +0100 | [diff] [blame] | 44 | |
| 45 | #include <cstdlib> |
| 46 | #include <cstring> |
| 47 | #include <fstream> |
| 48 | #include <iostream> |
Giorgio Arena | cf3935f | 2017-10-26 17:14:13 +0100 | [diff] [blame] | 49 | #include <random> |
| 50 | #include <string> |
| 51 | #include <tuple> |
| 52 | #include <vector> |
Anthony Barbier | 6ff3b19 | 2017-09-04 18:44:23 +0100 | [diff] [blame] | 53 | |
| 54 | namespace arm_compute |
| 55 | { |
| 56 | namespace utils |
| 57 | { |
Anthony Barbier | 6db0ff5 | 2018-01-05 10:59:12 +0000 | [diff] [blame] | 58 | /** Abstract Example class. |
| 59 | * |
| 60 | * All examples have to inherit from this class. |
| 61 | */ |
| 62 | class Example |
| 63 | { |
| 64 | public: |
| 65 | virtual void do_setup(int argc, char **argv) {}; |
| 66 | virtual void do_run() {}; |
| 67 | virtual void do_teardown() {}; |
| 68 | |
| 69 | /** Default destructor. */ |
| 70 | virtual ~Example() = default; |
| 71 | }; |
| 72 | |
| 73 | /** Run an example and handle the potential exceptions it throws |
| 74 | * |
| 75 | * @param[in] argc Number of command line arguments |
| 76 | * @param[in] argv Command line arguments |
| 77 | * @param[in] example Example to run |
| 78 | */ |
| 79 | int run_example(int argc, char **argv, Example &example); |
| 80 | |
| 81 | template <typename T> |
| 82 | int run_example(int argc, char **argv) |
| 83 | { |
| 84 | T example; |
| 85 | return run_example(argc, argv, example); |
| 86 | } |
Anthony Barbier | 6ff3b19 | 2017-09-04 18:44:23 +0100 | [diff] [blame] | 87 | |
| 88 | /** Draw a RGB rectangular window for the detected object |
| 89 | * |
| 90 | * @param[in, out] tensor Input tensor where the rectangle will be drawn on. Format supported: RGB888 |
| 91 | * @param[in] rect Geometry of the rectangular window |
| 92 | * @param[in] r Red colour to use |
| 93 | * @param[in] g Green colour to use |
| 94 | * @param[in] b Blue colour to use |
| 95 | */ |
| 96 | void draw_detection_rectangle(arm_compute::ITensor *tensor, const arm_compute::DetectionWindow &rect, uint8_t r, uint8_t g, uint8_t b); |
| 97 | |
| 98 | /** Parse the ppm header from an input file stream. At the end of the execution, |
| 99 | * the file position pointer will be located at the first pixel stored in the ppm file |
| 100 | * |
| 101 | * @param[in] fs Input file stream to parse |
| 102 | * |
| 103 | * @return The width, height and max value stored in the header of the PPM file |
| 104 | */ |
| 105 | std::tuple<unsigned int, unsigned int, int> parse_ppm_header(std::ifstream &fs); |
| 106 | |
Giorgio Arena | cf3935f | 2017-10-26 17:14:13 +0100 | [diff] [blame] | 107 | /** Parse the npy header from an input file stream. At the end of the execution, |
| 108 | * the file position pointer will be located at the first pixel stored in the npy file //TODO |
| 109 | * |
| 110 | * @param[in] fs Input file stream to parse |
| 111 | * |
| 112 | * @return The width and height stored in the header of the NPY file |
| 113 | */ |
| 114 | std::tuple<std::vector<unsigned long>, bool, std::string> parse_npy_header(std::ifstream &fs); |
| 115 | |
| 116 | /** Obtain numpy type string from DataType. |
| 117 | * |
| 118 | * @param[in] data_type Data type. |
| 119 | * |
| 120 | * @return numpy type string. |
| 121 | */ |
| 122 | inline std::string get_typestring(DataType data_type) |
| 123 | { |
| 124 | // Check endianness |
| 125 | const unsigned int i = 1; |
| 126 | const char *c = reinterpret_cast<const char *>(&i); |
| 127 | std::string endianness; |
| 128 | if(*c == 1) |
| 129 | { |
| 130 | endianness = std::string("<"); |
| 131 | } |
| 132 | else |
| 133 | { |
| 134 | endianness = std::string(">"); |
| 135 | } |
| 136 | const std::string no_endianness("|"); |
| 137 | |
| 138 | switch(data_type) |
| 139 | { |
| 140 | case DataType::U8: |
Giorgio Arena | a66eaa2 | 2017-12-21 19:50:06 +0000 | [diff] [blame] | 141 | case DataType::QASYMM8: |
Giorgio Arena | cf3935f | 2017-10-26 17:14:13 +0100 | [diff] [blame] | 142 | return no_endianness + "u" + support::cpp11::to_string(sizeof(uint8_t)); |
| 143 | case DataType::S8: |
| 144 | return no_endianness + "i" + support::cpp11::to_string(sizeof(int8_t)); |
| 145 | case DataType::U16: |
| 146 | return endianness + "u" + support::cpp11::to_string(sizeof(uint16_t)); |
| 147 | case DataType::S16: |
| 148 | return endianness + "i" + support::cpp11::to_string(sizeof(int16_t)); |
| 149 | case DataType::U32: |
| 150 | return endianness + "u" + support::cpp11::to_string(sizeof(uint32_t)); |
| 151 | case DataType::S32: |
| 152 | return endianness + "i" + support::cpp11::to_string(sizeof(int32_t)); |
| 153 | case DataType::U64: |
| 154 | return endianness + "u" + support::cpp11::to_string(sizeof(uint64_t)); |
| 155 | case DataType::S64: |
| 156 | return endianness + "i" + support::cpp11::to_string(sizeof(int64_t)); |
| 157 | case DataType::F32: |
| 158 | return endianness + "f" + support::cpp11::to_string(sizeof(float)); |
| 159 | case DataType::F64: |
| 160 | return endianness + "f" + support::cpp11::to_string(sizeof(double)); |
| 161 | case DataType::SIZET: |
| 162 | return endianness + "u" + support::cpp11::to_string(sizeof(size_t)); |
| 163 | default: |
| 164 | ARM_COMPUTE_ERROR("NOT SUPPORTED!"); |
| 165 | } |
| 166 | } |
| 167 | |
Georgios Pinitas | dc836b6 | 2017-09-20 14:02:37 +0100 | [diff] [blame] | 168 | /** Maps a tensor if needed |
| 169 | * |
| 170 | * @param[in] tensor Tensor to be mapped |
| 171 | * @param[in] blocking Specified if map is blocking or not |
| 172 | */ |
| 173 | template <typename T> |
Gian Marco Iodice | ae27e94 | 2017-09-28 18:31:26 +0100 | [diff] [blame] | 174 | inline void map(T &tensor, bool blocking) |
Georgios Pinitas | dc836b6 | 2017-09-20 14:02:37 +0100 | [diff] [blame] | 175 | { |
| 176 | ARM_COMPUTE_UNUSED(tensor); |
| 177 | ARM_COMPUTE_UNUSED(blocking); |
| 178 | } |
| 179 | |
| 180 | /** Unmaps a tensor if needed |
| 181 | * |
| 182 | * @param tensor Tensor to be unmapped |
| 183 | */ |
| 184 | template <typename T> |
Gian Marco Iodice | ae27e94 | 2017-09-28 18:31:26 +0100 | [diff] [blame] | 185 | inline void unmap(T &tensor) |
Georgios Pinitas | dc836b6 | 2017-09-20 14:02:37 +0100 | [diff] [blame] | 186 | { |
| 187 | ARM_COMPUTE_UNUSED(tensor); |
| 188 | } |
| 189 | |
| 190 | #ifdef ARM_COMPUTE_CL |
| 191 | /** Maps a tensor if needed |
| 192 | * |
| 193 | * @param[in] tensor Tensor to be mapped |
| 194 | * @param[in] blocking Specified if map is blocking or not |
| 195 | */ |
Gian Marco Iodice | ae27e94 | 2017-09-28 18:31:26 +0100 | [diff] [blame] | 196 | inline void map(CLTensor &tensor, bool blocking) |
Georgios Pinitas | dc836b6 | 2017-09-20 14:02:37 +0100 | [diff] [blame] | 197 | { |
| 198 | tensor.map(blocking); |
| 199 | } |
| 200 | |
| 201 | /** Unmaps a tensor if needed |
| 202 | * |
| 203 | * @param tensor Tensor to be unmapped |
| 204 | */ |
Gian Marco Iodice | ae27e94 | 2017-09-28 18:31:26 +0100 | [diff] [blame] | 205 | inline void unmap(CLTensor &tensor) |
Georgios Pinitas | dc836b6 | 2017-09-20 14:02:37 +0100 | [diff] [blame] | 206 | { |
| 207 | tensor.unmap(); |
| 208 | } |
Isabella Gottardi | 02aabcc | 2017-10-12 17:28:51 +0100 | [diff] [blame] | 209 | |
| 210 | /** Maps a distribution if needed |
| 211 | * |
| 212 | * @param[in] distribution Distribution to be mapped |
| 213 | * @param[in] blocking Specified if map is blocking or not |
| 214 | */ |
| 215 | inline void map(CLDistribution1D &distribution, bool blocking) |
| 216 | { |
| 217 | distribution.map(blocking); |
| 218 | } |
| 219 | |
| 220 | /** Unmaps a distribution if needed |
| 221 | * |
| 222 | * @param distribution Distribution to be unmapped |
| 223 | */ |
| 224 | inline void unmap(CLDistribution1D &distribution) |
| 225 | { |
| 226 | distribution.unmap(); |
| 227 | } |
Georgios Pinitas | dc836b6 | 2017-09-20 14:02:37 +0100 | [diff] [blame] | 228 | #endif /* ARM_COMPUTE_CL */ |
| 229 | |
Anthony Barbier | 7068f99 | 2017-10-26 15:23:08 +0100 | [diff] [blame] | 230 | #ifdef ARM_COMPUTE_GC |
| 231 | /** Maps a tensor if needed |
| 232 | * |
| 233 | * @param[in] tensor Tensor to be mapped |
| 234 | * @param[in] blocking Specified if map is blocking or not |
| 235 | */ |
| 236 | inline void map(GCTensor &tensor, bool blocking) |
| 237 | { |
| 238 | tensor.map(blocking); |
| 239 | } |
| 240 | |
| 241 | /** Unmaps a tensor if needed |
| 242 | * |
| 243 | * @param tensor Tensor to be unmapped |
| 244 | */ |
| 245 | inline void unmap(GCTensor &tensor) |
| 246 | { |
| 247 | tensor.unmap(); |
| 248 | } |
| 249 | #endif /* ARM_COMPUTE_GC */ |
| 250 | |
Anthony Barbier | 6ff3b19 | 2017-09-04 18:44:23 +0100 | [diff] [blame] | 251 | /** Class to load the content of a PPM file into an Image |
| 252 | */ |
| 253 | class PPMLoader |
| 254 | { |
| 255 | public: |
| 256 | PPMLoader() |
| 257 | : _fs(), _width(0), _height(0) |
| 258 | { |
| 259 | } |
| 260 | /** Open a PPM file and reads its metadata (Width, height) |
| 261 | * |
| 262 | * @param[in] ppm_filename File to open |
| 263 | */ |
| 264 | void open(const std::string &ppm_filename) |
| 265 | { |
| 266 | ARM_COMPUTE_ERROR_ON(is_open()); |
| 267 | try |
| 268 | { |
| 269 | _fs.exceptions(std::ifstream::failbit | std::ifstream::badbit); |
| 270 | _fs.open(ppm_filename, std::ios::in | std::ios::binary); |
| 271 | |
| 272 | unsigned int max_val = 0; |
| 273 | std::tie(_width, _height, max_val) = parse_ppm_header(_fs); |
| 274 | |
| 275 | ARM_COMPUTE_ERROR_ON_MSG(max_val >= 256, "2 bytes per colour channel not supported in file %s", ppm_filename.c_str()); |
| 276 | } |
Anthony Barbier | 6db0ff5 | 2018-01-05 10:59:12 +0000 | [diff] [blame] | 277 | catch(std::runtime_error &e) |
Anthony Barbier | 6ff3b19 | 2017-09-04 18:44:23 +0100 | [diff] [blame] | 278 | { |
| 279 | ARM_COMPUTE_ERROR("Accessing %s: %s", ppm_filename.c_str(), e.what()); |
| 280 | } |
| 281 | } |
| 282 | /** Return true if a PPM file is currently open |
| 283 | */ |
| 284 | bool is_open() |
| 285 | { |
| 286 | return _fs.is_open(); |
| 287 | } |
| 288 | |
| 289 | /** Initialise an image's metadata with the dimensions of the PPM file currently open |
| 290 | * |
| 291 | * @param[out] image Image to initialise |
| 292 | * @param[in] format Format to use for the image (Must be RGB888 or U8) |
| 293 | */ |
| 294 | template <typename T> |
| 295 | void init_image(T &image, arm_compute::Format format) |
| 296 | { |
| 297 | ARM_COMPUTE_ERROR_ON(!is_open()); |
| 298 | ARM_COMPUTE_ERROR_ON(format != arm_compute::Format::RGB888 && format != arm_compute::Format::U8); |
| 299 | |
| 300 | // Use the size of the input PPM image |
| 301 | arm_compute::TensorInfo image_info(_width, _height, format); |
| 302 | image.allocator()->init(image_info); |
| 303 | } |
| 304 | |
| 305 | /** Fill an image with the content of the currently open PPM file. |
| 306 | * |
| 307 | * @note If the image is a CLImage, the function maps and unmaps the image |
| 308 | * |
| 309 | * @param[in,out] image Image to fill (Must be allocated, and of matching dimensions with the opened PPM). |
| 310 | */ |
| 311 | template <typename T> |
| 312 | void fill_image(T &image) |
| 313 | { |
| 314 | ARM_COMPUTE_ERROR_ON(!is_open()); |
| 315 | ARM_COMPUTE_ERROR_ON(image.info()->dimension(0) != _width || image.info()->dimension(1) != _height); |
| 316 | ARM_COMPUTE_ERROR_ON_FORMAT_NOT_IN(&image, arm_compute::Format::U8, arm_compute::Format::RGB888); |
| 317 | try |
| 318 | { |
Anthony Barbier | 7068f99 | 2017-10-26 15:23:08 +0100 | [diff] [blame] | 319 | // Map buffer if creating a CLTensor/GCTensor |
Georgios Pinitas | dc836b6 | 2017-09-20 14:02:37 +0100 | [diff] [blame] | 320 | map(image, true); |
| 321 | |
Anthony Barbier | 6ff3b19 | 2017-09-04 18:44:23 +0100 | [diff] [blame] | 322 | // Check if the file is large enough to fill the image |
| 323 | const size_t current_position = _fs.tellg(); |
| 324 | _fs.seekg(0, std::ios_base::end); |
| 325 | const size_t end_position = _fs.tellg(); |
| 326 | _fs.seekg(current_position, std::ios_base::beg); |
| 327 | |
| 328 | ARM_COMPUTE_ERROR_ON_MSG((end_position - current_position) < image.info()->tensor_shape().total_size() * image.info()->element_size(), |
| 329 | "Not enough data in file"); |
| 330 | ARM_COMPUTE_UNUSED(end_position); |
| 331 | |
| 332 | switch(image.info()->format()) |
| 333 | { |
| 334 | case arm_compute::Format::U8: |
| 335 | { |
| 336 | // We need to convert the data from RGB to grayscale: |
| 337 | // Iterate through every pixel of the image |
| 338 | arm_compute::Window window; |
| 339 | window.set(arm_compute::Window::DimX, arm_compute::Window::Dimension(0, _width, 1)); |
| 340 | window.set(arm_compute::Window::DimY, arm_compute::Window::Dimension(0, _height, 1)); |
| 341 | |
| 342 | arm_compute::Iterator out(&image, window); |
| 343 | |
| 344 | unsigned char red = 0; |
| 345 | unsigned char green = 0; |
| 346 | unsigned char blue = 0; |
| 347 | |
| 348 | arm_compute::execute_window_loop(window, [&](const arm_compute::Coordinates & id) |
| 349 | { |
| 350 | red = _fs.get(); |
| 351 | green = _fs.get(); |
| 352 | blue = _fs.get(); |
| 353 | |
| 354 | *out.ptr() = 0.2126f * red + 0.7152f * green + 0.0722f * blue; |
| 355 | }, |
| 356 | out); |
| 357 | |
| 358 | break; |
| 359 | } |
| 360 | case arm_compute::Format::RGB888: |
| 361 | { |
| 362 | // There is no format conversion needed: we can simply copy the content of the input file to the image one row at the time. |
| 363 | // Create a vertical window to iterate through the image's rows: |
| 364 | arm_compute::Window window; |
| 365 | window.set(arm_compute::Window::DimY, arm_compute::Window::Dimension(0, _height, 1)); |
| 366 | |
| 367 | arm_compute::Iterator out(&image, window); |
| 368 | |
| 369 | arm_compute::execute_window_loop(window, [&](const arm_compute::Coordinates & id) |
| 370 | { |
| 371 | // Copy one row from the input file to the current row of the image: |
| 372 | _fs.read(reinterpret_cast<std::fstream::char_type *>(out.ptr()), _width * image.info()->element_size()); |
| 373 | }, |
| 374 | out); |
| 375 | |
| 376 | break; |
| 377 | } |
| 378 | default: |
| 379 | ARM_COMPUTE_ERROR("Unsupported format"); |
| 380 | } |
| 381 | |
Anthony Barbier | 7068f99 | 2017-10-26 15:23:08 +0100 | [diff] [blame] | 382 | // Unmap buffer if creating a CLTensor/GCTensor |
Georgios Pinitas | dc836b6 | 2017-09-20 14:02:37 +0100 | [diff] [blame] | 383 | unmap(image); |
Anthony Barbier | 6ff3b19 | 2017-09-04 18:44:23 +0100 | [diff] [blame] | 384 | } |
| 385 | catch(const std::ifstream::failure &e) |
| 386 | { |
| 387 | ARM_COMPUTE_ERROR("Loading PPM file: %s", e.what()); |
| 388 | } |
| 389 | } |
| 390 | |
Gian Marco | 44ec2e7 | 2017-10-19 14:13:38 +0100 | [diff] [blame] | 391 | /** Fill a tensor with 3 planes (one for each channel) with the content of the currently open PPM file. |
| 392 | * |
| 393 | * @note If the image is a CLImage, the function maps and unmaps the image |
| 394 | * |
| 395 | * @param[in,out] tensor Tensor with 3 planes to fill (Must be allocated, and of matching dimensions with the opened PPM). Data types supported: U8/F32 |
| 396 | * @param[in] bgr (Optional) Fill the first plane with blue channel (default = false) |
| 397 | */ |
| 398 | template <typename T> |
| 399 | void fill_planar_tensor(T &tensor, bool bgr = false) |
| 400 | { |
| 401 | ARM_COMPUTE_ERROR_ON(!is_open()); |
| 402 | ARM_COMPUTE_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(&tensor, 1, DataType::U8, DataType::F32); |
| 403 | ARM_COMPUTE_ERROR_ON(tensor.info()->dimension(0) != _width || tensor.info()->dimension(1) != _height || tensor.info()->dimension(2) != 3); |
| 404 | |
| 405 | try |
| 406 | { |
| 407 | // Map buffer if creating a CLTensor |
| 408 | map(tensor, true); |
| 409 | |
| 410 | // Check if the file is large enough to fill the image |
| 411 | const size_t current_position = _fs.tellg(); |
| 412 | _fs.seekg(0, std::ios_base::end); |
| 413 | const size_t end_position = _fs.tellg(); |
| 414 | _fs.seekg(current_position, std::ios_base::beg); |
| 415 | |
| 416 | ARM_COMPUTE_ERROR_ON_MSG((end_position - current_position) < tensor.info()->tensor_shape().total_size(), |
| 417 | "Not enough data in file"); |
| 418 | ARM_COMPUTE_UNUSED(end_position); |
| 419 | |
| 420 | // Iterate through every pixel of the image |
| 421 | arm_compute::Window window; |
| 422 | window.set(arm_compute::Window::DimX, arm_compute::Window::Dimension(0, _width, 1)); |
| 423 | window.set(arm_compute::Window::DimY, arm_compute::Window::Dimension(0, _height, 1)); |
| 424 | window.set(arm_compute::Window::DimZ, arm_compute::Window::Dimension(0, 1, 1)); |
| 425 | |
| 426 | arm_compute::Iterator out(&tensor, window); |
| 427 | |
| 428 | unsigned char red = 0; |
| 429 | unsigned char green = 0; |
| 430 | unsigned char blue = 0; |
| 431 | |
| 432 | size_t stride_z = tensor.info()->strides_in_bytes()[2]; |
| 433 | |
| 434 | arm_compute::execute_window_loop(window, [&](const arm_compute::Coordinates & id) |
| 435 | { |
| 436 | red = _fs.get(); |
| 437 | green = _fs.get(); |
| 438 | blue = _fs.get(); |
| 439 | |
| 440 | switch(tensor.info()->data_type()) |
| 441 | { |
| 442 | case arm_compute::DataType::U8: |
| 443 | { |
| 444 | *(out.ptr() + 0 * stride_z) = bgr ? blue : red; |
| 445 | *(out.ptr() + 1 * stride_z) = green; |
| 446 | *(out.ptr() + 2 * stride_z) = bgr ? red : blue; |
| 447 | break; |
| 448 | } |
| 449 | case arm_compute::DataType::F32: |
| 450 | { |
| 451 | *reinterpret_cast<float *>(out.ptr() + 0 * stride_z) = static_cast<float>(bgr ? blue : red); |
| 452 | *reinterpret_cast<float *>(out.ptr() + 1 * stride_z) = static_cast<float>(green); |
| 453 | *reinterpret_cast<float *>(out.ptr() + 2 * stride_z) = static_cast<float>(bgr ? red : blue); |
| 454 | break; |
| 455 | } |
| 456 | default: |
| 457 | { |
| 458 | ARM_COMPUTE_ERROR("Unsupported data type"); |
| 459 | } |
| 460 | } |
| 461 | }, |
| 462 | out); |
| 463 | |
| 464 | // Unmap buffer if creating a CLTensor |
| 465 | unmap(tensor); |
| 466 | } |
| 467 | catch(const std::ifstream::failure &e) |
| 468 | { |
| 469 | ARM_COMPUTE_ERROR("Loading PPM file: %s", e.what()); |
| 470 | } |
| 471 | } |
| 472 | |
Isabella Gottardi | a4c6188 | 2017-11-03 12:11:55 +0000 | [diff] [blame] | 473 | /** Return the width of the currently open PPM file. |
| 474 | */ |
| 475 | unsigned int width() const |
| 476 | { |
| 477 | return _width; |
| 478 | } |
| 479 | |
| 480 | /** Return the height of the currently open PPM file. |
| 481 | */ |
| 482 | unsigned int height() const |
| 483 | { |
| 484 | return _height; |
| 485 | } |
| 486 | |
Anthony Barbier | 6ff3b19 | 2017-09-04 18:44:23 +0100 | [diff] [blame] | 487 | private: |
| 488 | std::ifstream _fs; |
| 489 | unsigned int _width, _height; |
| 490 | }; |
| 491 | |
Giorgio Arena | cf3935f | 2017-10-26 17:14:13 +0100 | [diff] [blame] | 492 | class NPYLoader |
| 493 | { |
| 494 | public: |
| 495 | NPYLoader() |
| 496 | : _fs(), _shape(), _fortran_order(false), _typestring() |
| 497 | { |
| 498 | } |
| 499 | |
| 500 | /** Open a NPY file and reads its metadata |
| 501 | * |
| 502 | * @param[in] npy_filename File to open |
| 503 | */ |
| 504 | void open(const std::string &npy_filename) |
| 505 | { |
| 506 | ARM_COMPUTE_ERROR_ON(is_open()); |
| 507 | try |
| 508 | { |
| 509 | _fs.exceptions(std::ifstream::failbit | std::ifstream::badbit); |
| 510 | _fs.open(npy_filename, std::ios::in | std::ios::binary); |
| 511 | |
| 512 | std::tie(_shape, _fortran_order, _typestring) = parse_npy_header(_fs); |
| 513 | } |
| 514 | catch(const std::ifstream::failure &e) |
| 515 | { |
| 516 | ARM_COMPUTE_ERROR("Accessing %s: %s", npy_filename.c_str(), e.what()); |
| 517 | } |
| 518 | } |
| 519 | /** Return true if a NPY file is currently open */ |
| 520 | bool is_open() |
| 521 | { |
| 522 | return _fs.is_open(); |
| 523 | } |
| 524 | |
| 525 | /** Return true if a NPY file is in fortran order */ |
| 526 | bool is_fortran() |
| 527 | { |
| 528 | return _fortran_order; |
| 529 | } |
| 530 | |
Gian Marco | 0bc5a25 | 2017-12-04 13:55:08 +0000 | [diff] [blame] | 531 | /** Initialise the tensor's metadata with the dimensions of the NPY file currently open |
Giorgio Arena | cf3935f | 2017-10-26 17:14:13 +0100 | [diff] [blame] | 532 | * |
| 533 | * @param[out] tensor Tensor to initialise |
Gian Marco | 0bc5a25 | 2017-12-04 13:55:08 +0000 | [diff] [blame] | 534 | * @param[in] dt Data type to use for the tensor |
Giorgio Arena | cf3935f | 2017-10-26 17:14:13 +0100 | [diff] [blame] | 535 | */ |
| 536 | template <typename T> |
Gian Marco | 0bc5a25 | 2017-12-04 13:55:08 +0000 | [diff] [blame] | 537 | void init_tensor(T &tensor, arm_compute::DataType dt) |
Giorgio Arena | cf3935f | 2017-10-26 17:14:13 +0100 | [diff] [blame] | 538 | { |
| 539 | ARM_COMPUTE_ERROR_ON(!is_open()); |
Gian Marco | 0bc5a25 | 2017-12-04 13:55:08 +0000 | [diff] [blame] | 540 | ARM_COMPUTE_ERROR_ON(dt != arm_compute::DataType::F32); |
Giorgio Arena | cf3935f | 2017-10-26 17:14:13 +0100 | [diff] [blame] | 541 | |
| 542 | // Use the size of the input NPY tensor |
| 543 | TensorShape shape; |
| 544 | shape.set_num_dimensions(_shape.size()); |
| 545 | for(size_t i = 0; i < _shape.size(); ++i) |
| 546 | { |
| 547 | shape.set(i, _shape.at(i)); |
| 548 | } |
| 549 | |
Gian Marco | 0bc5a25 | 2017-12-04 13:55:08 +0000 | [diff] [blame] | 550 | arm_compute::TensorInfo tensor_info(shape, 1, dt); |
Giorgio Arena | cf3935f | 2017-10-26 17:14:13 +0100 | [diff] [blame] | 551 | tensor.allocator()->init(tensor_info); |
| 552 | } |
| 553 | |
| 554 | /** Fill a tensor with the content of the currently open NPY file. |
| 555 | * |
| 556 | * @note If the tensor is a CLTensor, the function maps and unmaps the tensor |
| 557 | * |
| 558 | * @param[in,out] tensor Tensor to fill (Must be allocated, and of matching dimensions with the opened NPY). |
| 559 | */ |
| 560 | template <typename T> |
| 561 | void fill_tensor(T &tensor) |
| 562 | { |
| 563 | ARM_COMPUTE_ERROR_ON(!is_open()); |
Anthony Barbier | 6db0ff5 | 2018-01-05 10:59:12 +0000 | [diff] [blame] | 564 | ARM_COMPUTE_ERROR_ON_DATA_TYPE_NOT_IN(&tensor, arm_compute::DataType::F32); |
Giorgio Arena | cf3935f | 2017-10-26 17:14:13 +0100 | [diff] [blame] | 565 | try |
| 566 | { |
| 567 | // Map buffer if creating a CLTensor |
| 568 | map(tensor, true); |
| 569 | |
| 570 | // Check if the file is large enough to fill the tensor |
| 571 | const size_t current_position = _fs.tellg(); |
| 572 | _fs.seekg(0, std::ios_base::end); |
| 573 | const size_t end_position = _fs.tellg(); |
| 574 | _fs.seekg(current_position, std::ios_base::beg); |
| 575 | |
| 576 | ARM_COMPUTE_ERROR_ON_MSG((end_position - current_position) < tensor.info()->tensor_shape().total_size() * tensor.info()->element_size(), |
| 577 | "Not enough data in file"); |
| 578 | ARM_COMPUTE_UNUSED(end_position); |
| 579 | |
| 580 | // Check if the typestring matches the given one |
| 581 | std::string expect_typestr = get_typestring(tensor.info()->data_type()); |
| 582 | ARM_COMPUTE_ERROR_ON_MSG(_typestring != expect_typestr, "Typestrings mismatch"); |
| 583 | |
| 584 | // Validate tensor shape |
Anthony Barbier | 6db0ff5 | 2018-01-05 10:59:12 +0000 | [diff] [blame] | 585 | ARM_COMPUTE_ERROR_ON_MSG(_shape.size() != tensor.info()->tensor_shape().num_dimensions(), "Tensor ranks mismatch"); |
Giorgio Arena | cf3935f | 2017-10-26 17:14:13 +0100 | [diff] [blame] | 586 | if(_fortran_order) |
| 587 | { |
| 588 | for(size_t i = 0; i < _shape.size(); ++i) |
| 589 | { |
Anthony Barbier | 6db0ff5 | 2018-01-05 10:59:12 +0000 | [diff] [blame] | 590 | ARM_COMPUTE_ERROR_ON_MSG(tensor.info()->tensor_shape()[i] != _shape[i], "Tensor dimensions mismatch"); |
Giorgio Arena | cf3935f | 2017-10-26 17:14:13 +0100 | [diff] [blame] | 591 | } |
| 592 | } |
| 593 | else |
| 594 | { |
| 595 | for(size_t i = 0; i < _shape.size(); ++i) |
| 596 | { |
Anthony Barbier | 6db0ff5 | 2018-01-05 10:59:12 +0000 | [diff] [blame] | 597 | ARM_COMPUTE_ERROR_ON_MSG(tensor.info()->tensor_shape()[i] != _shape[_shape.size() - i - 1], "Tensor dimensions mismatch"); |
Giorgio Arena | cf3935f | 2017-10-26 17:14:13 +0100 | [diff] [blame] | 598 | } |
| 599 | } |
| 600 | |
Gian Marco | 0bc5a25 | 2017-12-04 13:55:08 +0000 | [diff] [blame] | 601 | switch(tensor.info()->data_type()) |
Giorgio Arena | cf3935f | 2017-10-26 17:14:13 +0100 | [diff] [blame] | 602 | { |
Gian Marco | 0bc5a25 | 2017-12-04 13:55:08 +0000 | [diff] [blame] | 603 | case arm_compute::DataType::F32: |
Giorgio Arena | cf3935f | 2017-10-26 17:14:13 +0100 | [diff] [blame] | 604 | { |
| 605 | // Read data |
| 606 | if(tensor.info()->padding().empty()) |
| 607 | { |
| 608 | // If tensor has no padding read directly from stream. |
| 609 | _fs.read(reinterpret_cast<char *>(tensor.buffer()), tensor.info()->total_size()); |
| 610 | } |
| 611 | else |
| 612 | { |
| 613 | // If tensor has padding accessing tensor elements through execution window. |
| 614 | Window window; |
| 615 | window.use_tensor_dimensions(tensor.info()->tensor_shape()); |
| 616 | |
| 617 | execute_window_loop(window, [&](const Coordinates & id) |
| 618 | { |
| 619 | _fs.read(reinterpret_cast<char *>(tensor.ptr_to_element(id)), tensor.info()->element_size()); |
| 620 | }); |
| 621 | } |
| 622 | |
| 623 | break; |
| 624 | } |
| 625 | default: |
Gian Marco | 0bc5a25 | 2017-12-04 13:55:08 +0000 | [diff] [blame] | 626 | ARM_COMPUTE_ERROR("Unsupported data type"); |
Giorgio Arena | cf3935f | 2017-10-26 17:14:13 +0100 | [diff] [blame] | 627 | } |
| 628 | |
| 629 | // Unmap buffer if creating a CLTensor |
| 630 | unmap(tensor); |
| 631 | } |
| 632 | catch(const std::ifstream::failure &e) |
| 633 | { |
| 634 | ARM_COMPUTE_ERROR("Loading NPY file: %s", e.what()); |
| 635 | } |
| 636 | } |
| 637 | |
| 638 | private: |
| 639 | std::ifstream _fs; |
| 640 | std::vector<unsigned long> _shape; |
| 641 | bool _fortran_order; |
| 642 | std::string _typestring; |
| 643 | }; |
| 644 | |
Anthony Barbier | 6ff3b19 | 2017-09-04 18:44:23 +0100 | [diff] [blame] | 645 | /** Template helper function to save a tensor image to a PPM file. |
| 646 | * |
| 647 | * @note Only U8 and RGB888 formats supported. |
| 648 | * @note Only works with 2D tensors. |
| 649 | * @note If the input tensor is a CLTensor, the function maps and unmaps the image |
| 650 | * |
| 651 | * @param[in] tensor The tensor to save as PPM file |
| 652 | * @param[in] ppm_filename Filename of the file to create. |
| 653 | */ |
| 654 | template <typename T> |
| 655 | void save_to_ppm(T &tensor, const std::string &ppm_filename) |
| 656 | { |
| 657 | ARM_COMPUTE_ERROR_ON_FORMAT_NOT_IN(&tensor, arm_compute::Format::RGB888, arm_compute::Format::U8); |
| 658 | ARM_COMPUTE_ERROR_ON(tensor.info()->num_dimensions() > 2); |
| 659 | |
| 660 | std::ofstream fs; |
| 661 | |
| 662 | try |
| 663 | { |
| 664 | fs.exceptions(std::ofstream::failbit | std::ofstream::badbit | std::ofstream::eofbit); |
| 665 | fs.open(ppm_filename, std::ios::out | std::ios::binary); |
| 666 | |
| 667 | const unsigned int width = tensor.info()->tensor_shape()[0]; |
| 668 | const unsigned int height = tensor.info()->tensor_shape()[1]; |
| 669 | |
| 670 | fs << "P6\n" |
| 671 | << width << " " << height << " 255\n"; |
| 672 | |
Anthony Barbier | 7068f99 | 2017-10-26 15:23:08 +0100 | [diff] [blame] | 673 | // Map buffer if creating a CLTensor/GCTensor |
Georgios Pinitas | dc836b6 | 2017-09-20 14:02:37 +0100 | [diff] [blame] | 674 | map(tensor, true); |
Anthony Barbier | 6ff3b19 | 2017-09-04 18:44:23 +0100 | [diff] [blame] | 675 | |
| 676 | switch(tensor.info()->format()) |
| 677 | { |
| 678 | case arm_compute::Format::U8: |
| 679 | { |
| 680 | arm_compute::Window window; |
| 681 | window.set(arm_compute::Window::DimX, arm_compute::Window::Dimension(0, width, 1)); |
| 682 | window.set(arm_compute::Window::DimY, arm_compute::Window::Dimension(0, height, 1)); |
| 683 | |
| 684 | arm_compute::Iterator in(&tensor, window); |
| 685 | |
| 686 | arm_compute::execute_window_loop(window, [&](const arm_compute::Coordinates & id) |
| 687 | { |
| 688 | const unsigned char value = *in.ptr(); |
| 689 | |
| 690 | fs << value << value << value; |
| 691 | }, |
| 692 | in); |
| 693 | |
| 694 | break; |
| 695 | } |
| 696 | case arm_compute::Format::RGB888: |
| 697 | { |
| 698 | arm_compute::Window window; |
| 699 | window.set(arm_compute::Window::DimX, arm_compute::Window::Dimension(0, width, width)); |
| 700 | window.set(arm_compute::Window::DimY, arm_compute::Window::Dimension(0, height, 1)); |
| 701 | |
| 702 | arm_compute::Iterator in(&tensor, window); |
| 703 | |
| 704 | arm_compute::execute_window_loop(window, [&](const arm_compute::Coordinates & id) |
| 705 | { |
| 706 | fs.write(reinterpret_cast<std::fstream::char_type *>(in.ptr()), width * tensor.info()->element_size()); |
| 707 | }, |
| 708 | in); |
| 709 | |
| 710 | break; |
| 711 | } |
| 712 | default: |
| 713 | ARM_COMPUTE_ERROR("Unsupported format"); |
| 714 | } |
Georgios Pinitas | dc836b6 | 2017-09-20 14:02:37 +0100 | [diff] [blame] | 715 | |
Anthony Barbier | 7068f99 | 2017-10-26 15:23:08 +0100 | [diff] [blame] | 716 | // Unmap buffer if creating a CLTensor/GCTensor |
Georgios Pinitas | dc836b6 | 2017-09-20 14:02:37 +0100 | [diff] [blame] | 717 | unmap(tensor); |
Anthony Barbier | 6ff3b19 | 2017-09-04 18:44:23 +0100 | [diff] [blame] | 718 | } |
| 719 | catch(const std::ofstream::failure &e) |
| 720 | { |
| 721 | ARM_COMPUTE_ERROR("Writing %s: (%s)", ppm_filename.c_str(), e.what()); |
| 722 | } |
| 723 | } |
steniu01 | bee466b | 2017-06-21 16:45:41 +0100 | [diff] [blame] | 724 | |
Giorgio Arena | cf3935f | 2017-10-26 17:14:13 +0100 | [diff] [blame] | 725 | /** Template helper function to save a tensor image to a NPY file. |
| 726 | * |
Gian Marco | bfa3b52 | 2017-12-12 10:08:38 +0000 | [diff] [blame] | 727 | * @note Only F32 data type supported. |
Giorgio Arena | cf3935f | 2017-10-26 17:14:13 +0100 | [diff] [blame] | 728 | * @note Only works with 2D tensors. |
| 729 | * @note If the input tensor is a CLTensor, the function maps and unmaps the image |
| 730 | * |
| 731 | * @param[in] tensor The tensor to save as NPY file |
| 732 | * @param[in] npy_filename Filename of the file to create. |
| 733 | * @param[in] fortran_order If true, save matrix in fortran order. |
| 734 | */ |
| 735 | template <typename T> |
| 736 | void save_to_npy(T &tensor, const std::string &npy_filename, bool fortran_order) |
| 737 | { |
Gian Marco | bfa3b52 | 2017-12-12 10:08:38 +0000 | [diff] [blame] | 738 | ARM_COMPUTE_ERROR_ON_DATA_TYPE_NOT_IN(&tensor, arm_compute::DataType::F32); |
Giorgio Arena | cf3935f | 2017-10-26 17:14:13 +0100 | [diff] [blame] | 739 | ARM_COMPUTE_ERROR_ON(tensor.info()->num_dimensions() > 2); |
| 740 | |
| 741 | std::ofstream fs; |
| 742 | |
| 743 | try |
| 744 | { |
| 745 | fs.exceptions(std::ofstream::failbit | std::ofstream::badbit | std::ofstream::eofbit); |
| 746 | fs.open(npy_filename, std::ios::out | std::ios::binary); |
| 747 | |
Anthony Barbier | 87f21cd | 2017-11-10 16:27:32 +0000 | [diff] [blame] | 748 | const unsigned int width = tensor.info()->tensor_shape()[0]; |
| 749 | const unsigned int height = tensor.info()->tensor_shape()[1]; |
| 750 | std::vector<npy::ndarray_len_t> shape(2); |
Giorgio Arena | cf3935f | 2017-10-26 17:14:13 +0100 | [diff] [blame] | 751 | |
| 752 | if(!fortran_order) |
| 753 | { |
| 754 | shape[0] = height, shape[1] = width; |
| 755 | } |
| 756 | else |
| 757 | { |
| 758 | shape[0] = width, shape[1] = height; |
| 759 | } |
| 760 | |
| 761 | // Map buffer if creating a CLTensor |
| 762 | map(tensor, true); |
| 763 | |
Gian Marco | bfa3b52 | 2017-12-12 10:08:38 +0000 | [diff] [blame] | 764 | switch(tensor.info()->data_type()) |
Giorgio Arena | cf3935f | 2017-10-26 17:14:13 +0100 | [diff] [blame] | 765 | { |
Gian Marco | bfa3b52 | 2017-12-12 10:08:38 +0000 | [diff] [blame] | 766 | case arm_compute::DataType::F32: |
Giorgio Arena | cf3935f | 2017-10-26 17:14:13 +0100 | [diff] [blame] | 767 | { |
| 768 | std::vector<float> tmp; /* Used only to get the typestring */ |
| 769 | npy::Typestring typestring_o{ tmp }; |
| 770 | std::string typestring = typestring_o.str(); |
| 771 | |
| 772 | std::ofstream stream(npy_filename, std::ofstream::binary); |
Anthony Barbier | 87f21cd | 2017-11-10 16:27:32 +0000 | [diff] [blame] | 773 | npy::write_header(stream, typestring, fortran_order, shape); |
Giorgio Arena | cf3935f | 2017-10-26 17:14:13 +0100 | [diff] [blame] | 774 | |
| 775 | arm_compute::Window window; |
| 776 | window.set(arm_compute::Window::DimX, arm_compute::Window::Dimension(0, width, 1)); |
| 777 | window.set(arm_compute::Window::DimY, arm_compute::Window::Dimension(0, height, 1)); |
| 778 | |
| 779 | arm_compute::Iterator in(&tensor, window); |
| 780 | |
| 781 | arm_compute::execute_window_loop(window, [&](const arm_compute::Coordinates & id) |
| 782 | { |
| 783 | stream.write(reinterpret_cast<const char *>(in.ptr()), sizeof(float)); |
| 784 | }, |
| 785 | in); |
| 786 | |
| 787 | break; |
| 788 | } |
| 789 | default: |
| 790 | ARM_COMPUTE_ERROR("Unsupported format"); |
| 791 | } |
| 792 | |
| 793 | // Unmap buffer if creating a CLTensor |
| 794 | unmap(tensor); |
| 795 | } |
| 796 | catch(const std::ofstream::failure &e) |
| 797 | { |
| 798 | ARM_COMPUTE_ERROR("Writing %s: (%s)", npy_filename.c_str(), e.what()); |
| 799 | } |
| 800 | } |
| 801 | |
steniu01 | bee466b | 2017-06-21 16:45:41 +0100 | [diff] [blame] | 802 | /** Load the tensor with pre-trained data from a binary file |
| 803 | * |
| 804 | * @param[in] tensor The tensor to be filled. Data type supported: F32. |
| 805 | * @param[in] filename Filename of the binary file to load from. |
| 806 | */ |
| 807 | template <typename T> |
| 808 | void load_trained_data(T &tensor, const std::string &filename) |
| 809 | { |
| 810 | ARM_COMPUTE_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(&tensor, 1, DataType::F32); |
| 811 | |
| 812 | std::ifstream fs; |
| 813 | |
| 814 | try |
| 815 | { |
| 816 | fs.exceptions(std::ofstream::failbit | std::ofstream::badbit | std::ofstream::eofbit); |
| 817 | // Open file |
| 818 | fs.open(filename, std::ios::in | std::ios::binary); |
| 819 | |
| 820 | if(!fs.good()) |
| 821 | { |
| 822 | throw std::runtime_error("Could not load binary data: " + filename); |
| 823 | } |
| 824 | |
Anthony Barbier | 7068f99 | 2017-10-26 15:23:08 +0100 | [diff] [blame] | 825 | // Map buffer if creating a CLTensor/GCTensor |
Georgios Pinitas | dc836b6 | 2017-09-20 14:02:37 +0100 | [diff] [blame] | 826 | map(tensor, true); |
| 827 | |
steniu01 | bee466b | 2017-06-21 16:45:41 +0100 | [diff] [blame] | 828 | Window window; |
| 829 | |
| 830 | window.set(arm_compute::Window::DimX, arm_compute::Window::Dimension(0, 1, 1)); |
| 831 | |
| 832 | for(unsigned int d = 1; d < tensor.info()->num_dimensions(); ++d) |
| 833 | { |
| 834 | window.set(d, Window::Dimension(0, tensor.info()->tensor_shape()[d], 1)); |
| 835 | } |
| 836 | |
| 837 | arm_compute::Iterator in(&tensor, window); |
| 838 | |
| 839 | execute_window_loop(window, [&](const Coordinates & id) |
| 840 | { |
| 841 | fs.read(reinterpret_cast<std::fstream::char_type *>(in.ptr()), tensor.info()->tensor_shape()[0] * tensor.info()->element_size()); |
| 842 | }, |
| 843 | in); |
| 844 | |
Anthony Barbier | 7068f99 | 2017-10-26 15:23:08 +0100 | [diff] [blame] | 845 | // Unmap buffer if creating a CLTensor/GCTensor |
Georgios Pinitas | dc836b6 | 2017-09-20 14:02:37 +0100 | [diff] [blame] | 846 | unmap(tensor); |
steniu01 | bee466b | 2017-06-21 16:45:41 +0100 | [diff] [blame] | 847 | } |
| 848 | catch(const std::ofstream::failure &e) |
| 849 | { |
| 850 | ARM_COMPUTE_ERROR("Writing %s: (%s)", filename.c_str(), e.what()); |
| 851 | } |
| 852 | } |
| 853 | |
Giorgio Arena | cf3935f | 2017-10-26 17:14:13 +0100 | [diff] [blame] | 854 | template <typename T> |
| 855 | void fill_random_tensor(T &tensor, float lower_bound, float upper_bound) |
Anthony Barbier | 2a07e18 | 2017-08-04 18:20:27 +0100 | [diff] [blame] | 856 | { |
Giorgio Arena | cf3935f | 2017-10-26 17:14:13 +0100 | [diff] [blame] | 857 | std::random_device rd; |
| 858 | std::mt19937 gen(rd()); |
Anthony Barbier | 2a07e18 | 2017-08-04 18:20:27 +0100 | [diff] [blame] | 859 | |
Giorgio Arena | cf3935f | 2017-10-26 17:14:13 +0100 | [diff] [blame] | 860 | TensorShape shape(tensor.info()->dimension(0), tensor.info()->dimension(1)); |
| 861 | |
| 862 | Window window; |
| 863 | window.set(Window::DimX, Window::Dimension(0, shape.x(), 1)); |
| 864 | window.set(Window::DimY, Window::Dimension(0, shape.y(), 1)); |
| 865 | |
| 866 | map(tensor, true); |
| 867 | |
| 868 | Iterator it(&tensor, window); |
| 869 | |
Gian Marco | bfa3b52 | 2017-12-12 10:08:38 +0000 | [diff] [blame] | 870 | switch(tensor.info()->data_type()) |
Anthony Barbier | 2a07e18 | 2017-08-04 18:20:27 +0100 | [diff] [blame] | 871 | { |
Gian Marco | bfa3b52 | 2017-12-12 10:08:38 +0000 | [diff] [blame] | 872 | case arm_compute::DataType::F32: |
Giorgio Arena | cf3935f | 2017-10-26 17:14:13 +0100 | [diff] [blame] | 873 | { |
| 874 | std::uniform_real_distribution<float> dist(lower_bound, upper_bound); |
| 875 | |
| 876 | execute_window_loop(window, [&](const Coordinates & id) |
| 877 | { |
| 878 | *reinterpret_cast<float *>(it.ptr()) = dist(gen); |
| 879 | }, |
| 880 | it); |
| 881 | |
| 882 | break; |
| 883 | } |
Anthony Barbier | 2a07e18 | 2017-08-04 18:20:27 +0100 | [diff] [blame] | 884 | default: |
Giorgio Arena | cf3935f | 2017-10-26 17:14:13 +0100 | [diff] [blame] | 885 | { |
| 886 | ARM_COMPUTE_ERROR("Unsupported format"); |
| 887 | } |
Anthony Barbier | 2a07e18 | 2017-08-04 18:20:27 +0100 | [diff] [blame] | 888 | } |
Giorgio Arena | cf3935f | 2017-10-26 17:14:13 +0100 | [diff] [blame] | 889 | |
| 890 | unmap(tensor); |
Anthony Barbier | 2a07e18 | 2017-08-04 18:20:27 +0100 | [diff] [blame] | 891 | } |
Giorgio Arena | cf3935f | 2017-10-26 17:14:13 +0100 | [diff] [blame] | 892 | |
| 893 | template <typename T> |
Gian Marco | 0bc5a25 | 2017-12-04 13:55:08 +0000 | [diff] [blame] | 894 | void init_sgemm_output(T &dst, T &src0, T &src1, arm_compute::DataType dt) |
Giorgio Arena | cf3935f | 2017-10-26 17:14:13 +0100 | [diff] [blame] | 895 | { |
Gian Marco | 0bc5a25 | 2017-12-04 13:55:08 +0000 | [diff] [blame] | 896 | dst.allocator()->init(TensorInfo(TensorShape(src1.info()->dimension(0), src0.info()->dimension(1)), 1, dt)); |
Giorgio Arena | cf3935f | 2017-10-26 17:14:13 +0100 | [diff] [blame] | 897 | } |
Gian Marco | 5ca7409 | 2018-02-08 16:21:54 +0000 | [diff] [blame] | 898 | /** This function returns the amount of memory free reading from /proc/meminfo |
| 899 | * |
| 900 | * @return The free memory in kB |
| 901 | */ |
| 902 | uint64_t get_mem_free_from_meminfo(); |
Anthony Barbier | 6ff3b19 | 2017-09-04 18:44:23 +0100 | [diff] [blame] | 903 | } // namespace utils |
| 904 | } // namespace arm_compute |
| 905 | #endif /* __UTILS_UTILS_H__*/ |