Compute Library
 21.02
NENonMaximaSuppression3x3Kernel.cpp
Go to the documentation of this file.
1 /*
2  * Copyright (c) 2016-2020 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  */
25 
26 #include "arm_compute/core/Error.h"
30 #include "arm_compute/core/Types.h"
31 #include "arm_compute/core/Utils.h"
35 
36 #include <arm_neon.h>
37 #include <cstddef>
38 
39 using namespace arm_compute;
40 
41 namespace arm_compute
42 {
43 class Coordinates;
44 } // namespace arm_compute
45 
46 #ifdef __ARM_FEATURE_FP16_VECTOR_ARITHMETIC
47 namespace fp16
48 {
49 inline void mask_top(const float16x8_t &vc, const float16x8_t &in0, const float16x8_t &in1, uint16x8_t &mask)
50 {
51  // vc > nc.val[0], vc > nc.val[1], vc > nc.val[2]
52  mask = vandq_u16(mask, vcgeq_f16(vc, in0));
53  mask = vandq_u16(mask, vcgeq_f16(vc, vextq_f16(in0, in1, 1)));
54  mask = vandq_u16(mask, vcgeq_f16(vc, vextq_f16(in0, in1, 2)));
55 }
56 
57 inline void mask_middle(const float16x8_t &vc, const float16x8_t &in0, const float16x8_t &in1, uint16x8_t &mask)
58 {
59  // vc >= nc.val[0], vc > nc.val[2]
60  mask = vandq_u16(mask, vcgeq_f16(vc, in0));
61  mask = vandq_u16(mask, vcgtq_f16(vc, vextq_f16(in0, in1, 2)));
62 }
63 
64 inline void mask_bottom(const float16x8_t &vc, const float16x8_t &in0, const float16x8_t &in1, uint16x8_t &mask)
65 {
66  // vc > nc.val[0], vc > nc.val[1], vc > nc.val[2]
67  mask = vandq_u16(mask, vcgtq_f16(vc, in0));
68  mask = vandq_u16(mask, vcgtq_f16(vc, vextq_f16(in0, in1, 1)));
69  mask = vandq_u16(mask, vcgtq_f16(vc, vextq_f16(in0, in1, 2)));
70 }
71 
72 inline void non_maxima_suppression3x3_F32_F32(const void *__restrict in_ptr, void *__restrict out_ptr, const uint32_t in_stride)
73 {
74  auto in = static_cast<const float *__restrict>(in_ptr) - 1;
75  const auto out = static_cast<float *__restrict>(out_ptr);
76 
77  // Get centre scores
78  const float16x8x2_t vc =
79  {
80  vcombine_f16(vcvt_f16_f32(vld1q_f32(in + 1)), vcvt_f16_f32(vld1q_f32(in + 5))),
81  vcombine_f16(vcvt_f16_f32(vld1q_f32(in + 9)), vcvt_f16_f32(vld1q_f32(in + 13)))
82  };
83 
84  // Neighboring pixels
85  in -= in_stride;
86 
87  static const float16x4_t zero_f16x4 = vdup_n_f16(0);
88  static const uint16x8_t zero_u16 = vdupq_n_u16(0);
89  static const uint16x8_t true_mask = vceqq_u16(zero_u16, zero_u16);
90  static const uint16x8x2_t true_mask_x2 =
91  {
92  true_mask,
93  true_mask
94  };
95 
96  uint16x8x2_t mask = true_mask_x2;
97 
98  // Top row
99  const float16x8_t tmp_top0 = vcombine_f16(vcvt_f16_f32(vld1q_f32(in)), vcvt_f16_f32(vld1q_f32(in + 4)));
100  const float16x8_t tmp_top1 = vcombine_f16(vcvt_f16_f32(vld1q_f32(in + 8)), vcvt_f16_f32(vld1q_f32(in + 12)));
101  const float16x8_t tmp_top2 = vcombine_f16(vcvt_f16_f32(vld1q_f32(in + 16)), zero_f16x4);
102 
103  // vc >= nc.val[0], vc >= nc.val[1], vc >= nc.val[2]
104  mask_top(vc.val[0], tmp_top0, tmp_top1, mask.val[0]);
105  mask_top(vc.val[1], tmp_top1, tmp_top2, mask.val[1]);
106 
107  in += in_stride;
108 
109  // Middle row
110  const float16x8_t tmp_mid0 = vcombine_f16(vcvt_f16_f32(vld1q_f32(in)), vcvt_f16_f32(vld1q_f32(in + 4)));
111  const float16x8_t tmp_mid1 = vcombine_f16(vcvt_f16_f32(vld1q_f32(in + 8)), vcvt_f16_f32(vld1q_f32(in + 12)));
112  const float16x8_t tmp_mid2 = vcombine_f16(vcvt_f16_f32(vld1q_f32(in + 16)), zero_f16x4);
113 
114  // vc >= nc.val[0], vc > nc.val[2]
115  mask_middle(vc.val[0], tmp_mid0, tmp_mid1, mask.val[0]);
116  mask_middle(vc.val[1], tmp_mid1, tmp_mid2, mask.val[1]);
117 
118  in += in_stride;
119 
120  // Bottom row
121  const float16x8_t tmp_bot0 = vcombine_f16(vcvt_f16_f32(vld1q_f32(in)), vcvt_f16_f32(vld1q_f32(in + 4)));
122  const float16x8_t tmp_bot1 = vcombine_f16(vcvt_f16_f32(vld1q_f32(in + 8)), vcvt_f16_f32(vld1q_f32(in + 12)));
123  const float16x8_t tmp_bot2 = vcombine_f16(vcvt_f16_f32(vld1q_f32(in + 16)), zero_f16x4);
124 
125  // vc > nc.val[0], vc > nc.val[1], vc > nc.val[2]
126  mask_bottom(vc.val[0], tmp_bot0, tmp_bot1, mask.val[0]);
127  mask_bottom(vc.val[1], tmp_bot1, tmp_bot2, mask.val[1]);
128 
129  // Store
130  static const float16x8_t zero_f16x8 = vdupq_n_f16(0);
131 
132  const float16x8_t suppressed0 = vbslq_f16(mask.val[0], vc.val[0], zero_f16x8);
133  vst1q_f32(out + 0, vcvt_f32_f16(vget_low_f16(suppressed0)));
134  vst1q_f32(out + 4, vcvt_f32_f16(vget_high_f16(suppressed0)));
135 
136  const float16x8_t suppressed1 = vbslq_f16(mask.val[1], vc.val[1], zero_f16x8);
137  vst1q_f32(out + 8, vcvt_f32_f16(vget_low_f16(suppressed1)));
138  vst1q_f32(out + 12, vcvt_f32_f16(vget_high_f16(suppressed1)));
139 }
140 
141 inline void non_maxima_suppression3x3_U8_U8(const void *__restrict in_ptr, void *__restrict out_ptr, const uint32_t in_stride)
142 {
143  auto in = static_cast<const uint8_t *__restrict>(in_ptr) - 1;
144  const auto out = static_cast<uint8_t *__restrict>(out_ptr);
145 
146  // Get centre scores
147  const uint8x16_t vc = vld1q_u8(in + 1);
148 
149  // Neighboring pixels
150  in -= in_stride;
151 
152  // Top row
153  const uint8x16_t l_nc_0 = vld1q_u8(in);
154  const uint8x16_t m_nc_0 = vld1q_u8(in + 1);
155  const uint8x16_t r_nc_0 = vld1q_u8(in + 2);
156 
157  // Keep center scores if ...
158  // vc >= l_nc_0, vc >= m_nc_0, vc >= r_nc_0
159  uint8x16_t mask = vcgeq_u8(vc, l_nc_0);
160  mask = vandq_u8(mask, vcgeq_u8(vc, m_nc_0));
161  mask = vandq_u8(mask, vcgeq_u8(vc, r_nc_0));
162 
163  in += in_stride;
164 
165  // Middle row
166  const uint8x16_t l_nc_1 = vld1q_u8(in);
167  const uint8x16_t r_nc_1 = vld1q_u8(in + 2);
168 
169  // ... and ...
170  // vc >= l_nc_1, vc > r_nc_1
171  mask = vandq_u8(mask, vcgeq_u8(vc, l_nc_1));
172  mask = vandq_u8(mask, vcgtq_u8(vc, r_nc_1));
173 
174  in += in_stride;
175 
176  // Bottom row
177  const uint8x16_t l_nc_2 = vld1q_u8(in);
178  const uint8x16_t m_nc_2 = vld1q_u8(in + 1);
179  const uint8x16_t r_nc_2 = vld1q_u8(in + 2);
180 
181  // ... and ...
182  // vc > l_nc_2, vc > m_nc_2, vc > r_nc_2
183  mask = vandq_u8(mask, vcgtq_u8(vc, l_nc_2));
184  mask = vandq_u8(mask, vcgtq_u8(vc, m_nc_2));
185  mask = vandq_u8(mask, vcgtq_u8(vc, r_nc_2));
186 
187  // Store
188  static const uint8x16_t zero = vdupq_n_u8(0);
189  vst1q_u8(out, vbslq_u8(mask, vc, zero));
190 }
191 } // namespace fp16
192 
193 void NENonMaximaSuppression3x3FP16Kernel::configure(const ITensor *input, ITensor *output, bool border_undefined)
194 {
198 
199  _input = input;
200  _output = output;
201 
202  switch(input->info()->data_type())
203  {
204  case DataType::U8:
205  _func = &fp16::non_maxima_suppression3x3_U8_U8;
206  break;
207  default:
208  _func = &fp16::non_maxima_suppression3x3_F32_F32;
209  break;
210  }
211 
212  constexpr unsigned int num_elems_processed_per_iteration = 16;
213  const unsigned int num_elems_read_per_iteration = 16 + 2 * border_size().left + (input->info()->data_type() == DataType::U8 ? 0 : 3);
214  constexpr unsigned int num_elems_written_per_iteration = 16;
215  constexpr unsigned int num_rows_read_per_iteration = 3;
216 
217  // Configure kernel window
218  Window win = calculate_max_window(*input->info(), Steps(num_elems_processed_per_iteration), border_undefined, border_size());
219  AccessWindowHorizontal output_access(output->info(), 0, num_elems_written_per_iteration);
220 
222  AccessWindowRectangle(input->info(), -border_size().left, -border_size().top, num_elems_read_per_iteration, num_rows_read_per_iteration),
223  output_access);
224 
225  output_access.set_valid_region(win, input->info()->valid_region(), border_undefined, border_size());
226 
227  INEKernel::configure(win);
228 }
229 #endif /* __ARM_FEATURE_FP16_VECTOR_ARITHMETIC */
230 
231 namespace
232 {
233 inline void non_maxima_suppression3x3_FLOAT_FLOAT(const void *__restrict input_ptr, void *__restrict output_ptr, const uint32_t input_stride)
234 {
235  auto input = static_cast<const float *__restrict>(input_ptr) - 1;
236  const auto output = static_cast<float *__restrict>(output_ptr);
237 
238  // Get centre scores
239  const float32x4x4_t vc =
240  {
241  {
242  vld1q_f32(input + 1),
243  vld1q_f32(input + 5),
244  vld1q_f32(input + 9),
245  vld1q_f32(input + 13)
246  }
247  };
248 
249  // Neighboring pixels
250  float32x4x4_t l_nc{ {} };
251  float32x4x4_t m_nc{ {} };
252  float32x4x4_t r_nc{ {} };
253 
254  input -= input_stride;
255 
256  // Row0 - Low part
257  float32x4_t tmp_low = vld1q_f32(input);
258  float32x4_t tmp_high = vld1q_f32(input + 4);
259  float32x4_t tmp_high1 = vld1q_f32(input + 8);
260 
261  l_nc.val[0] = tmp_low;
262  m_nc.val[0] = vextq_f32(tmp_low, tmp_high, 1);
263  r_nc.val[0] = vextq_f32(tmp_low, tmp_high, 2);
264 
265  tmp_low = tmp_high;
266  tmp_high = tmp_high1;
267 
268  l_nc.val[1] = tmp_low;
269  m_nc.val[1] = vextq_f32(tmp_low, tmp_high, 1);
270  r_nc.val[1] = vextq_f32(tmp_low, tmp_high, 2);
271 
272  // Row0 - High part
273  tmp_low = tmp_high1;
274  tmp_high = vld1q_f32(input + 12);
275  tmp_high1 = vld1q_f32(input + 16);
276 
277  l_nc.val[2] = tmp_low;
278  m_nc.val[2] = vextq_f32(tmp_low, tmp_high, 1);
279  r_nc.val[2] = vextq_f32(tmp_low, tmp_high, 2);
280 
281  tmp_low = tmp_high;
282  tmp_high = tmp_high1;
283 
284  l_nc.val[3] = tmp_low;
285  m_nc.val[3] = vextq_f32(tmp_low, tmp_high, 1);
286  r_nc.val[3] = vextq_f32(tmp_low, tmp_high, 2);
287 
288  // mc >= nc.val[0], mc >= nc.val[1], mc >= nc.val[2]
289  uint32x4x4_t mask{ {} };
290  mask.val[0] = vcgeq_f32(vc.val[0], l_nc.val[0]);
291  mask.val[0] = vandq_u32(mask.val[0], vcgeq_f32(vc.val[0], m_nc.val[0]));
292  mask.val[0] = vandq_u32(mask.val[0], vcgeq_f32(vc.val[0], r_nc.val[0]));
293  mask.val[1] = vcgeq_f32(vc.val[1], l_nc.val[1]);
294  mask.val[1] = vandq_u32(mask.val[1], vcgeq_f32(vc.val[1], m_nc.val[1]));
295  mask.val[1] = vandq_u32(mask.val[1], vcgeq_f32(vc.val[1], r_nc.val[1]));
296  mask.val[2] = vcgeq_f32(vc.val[2], l_nc.val[2]);
297  mask.val[2] = vandq_u32(mask.val[2], vcgeq_f32(vc.val[2], m_nc.val[2]));
298  mask.val[2] = vandq_u32(mask.val[2], vcgeq_f32(vc.val[2], r_nc.val[2]));
299  mask.val[3] = vcgeq_f32(vc.val[3], l_nc.val[3]);
300  mask.val[3] = vandq_u32(mask.val[3], vcgeq_f32(vc.val[3], m_nc.val[3]));
301  mask.val[3] = vandq_u32(mask.val[3], vcgeq_f32(vc.val[3], r_nc.val[3]));
302 
303  input += input_stride;
304 
305  // Row1 - Low part
306  tmp_low = vld1q_f32(input);
307  tmp_high = vld1q_f32(input + 4);
308  tmp_high1 = vld1q_f32(input + 8);
309 
310  l_nc.val[0] = tmp_low;
311  r_nc.val[0] = vextq_f32(tmp_low, tmp_high, 2);
312 
313  tmp_low = tmp_high;
314  tmp_high = tmp_high1;
315 
316  l_nc.val[1] = tmp_low;
317  r_nc.val[1] = vextq_f32(tmp_low, tmp_high, 2);
318 
319  // Row1 - High part
320  tmp_low = tmp_high1;
321  tmp_high = vld1q_f32(input + 12);
322  tmp_high1 = vld1q_f32(input + 16);
323 
324  l_nc.val[2] = tmp_low;
325  r_nc.val[2] = vextq_f32(tmp_low, tmp_high, 2);
326 
327  tmp_low = tmp_high;
328  tmp_high = tmp_high1;
329 
330  l_nc.val[3] = tmp_low;
331  r_nc.val[3] = vextq_f32(tmp_low, tmp_high, 2);
332 
333  // mc >= nc.val[0], mc > nc.val[2]
334  mask.val[0] = vandq_u32(mask.val[0], vcgeq_f32(vc.val[0], l_nc.val[0]));
335  mask.val[0] = vandq_u32(mask.val[0], vcgtq_f32(vc.val[0], r_nc.val[0]));
336  mask.val[1] = vandq_u32(mask.val[1], vcgeq_f32(vc.val[1], l_nc.val[1]));
337  mask.val[1] = vandq_u32(mask.val[1], vcgtq_f32(vc.val[1], r_nc.val[1]));
338  mask.val[2] = vandq_u32(mask.val[2], vcgeq_f32(vc.val[2], l_nc.val[2]));
339  mask.val[2] = vandq_u32(mask.val[2], vcgtq_f32(vc.val[2], r_nc.val[2]));
340  mask.val[3] = vandq_u32(mask.val[3], vcgeq_f32(vc.val[3], l_nc.val[3]));
341  mask.val[3] = vandq_u32(mask.val[3], vcgtq_f32(vc.val[3], r_nc.val[3]));
342 
343  input += input_stride;
344 
345  // Row2 - Low part
346  tmp_low = vld1q_f32(input);
347  tmp_high = vld1q_f32(input + 4);
348  tmp_high1 = vld1q_f32(input + 8);
349 
350  l_nc.val[0] = tmp_low;
351  m_nc.val[0] = vextq_f32(tmp_low, tmp_high, 1);
352  r_nc.val[0] = vextq_f32(tmp_low, tmp_high, 2);
353 
354  tmp_low = tmp_high;
355  tmp_high = tmp_high1;
356 
357  l_nc.val[1] = tmp_low;
358  m_nc.val[1] = vextq_f32(tmp_low, tmp_high, 1);
359  r_nc.val[1] = vextq_f32(tmp_low, tmp_high, 2);
360 
361  // Row2 - High part
362  tmp_low = tmp_high1;
363  tmp_high = vld1q_f32(input + 12);
364  tmp_high1 = vld1q_f32(input + 16);
365 
366  l_nc.val[2] = tmp_low;
367  m_nc.val[2] = vextq_f32(tmp_low, tmp_high, 1);
368  r_nc.val[2] = vextq_f32(tmp_low, tmp_high, 2);
369 
370  tmp_low = tmp_high;
371  tmp_high = tmp_high1;
372 
373  l_nc.val[3] = tmp_low;
374  m_nc.val[3] = vextq_f32(tmp_low, tmp_high, 1);
375  r_nc.val[3] = vextq_f32(tmp_low, tmp_high, 2);
376 
377  // mc > nc.val[0], mc > nc.val[1], mc > nc.val[2]
378  mask.val[0] = vandq_u32(mask.val[0], vcgtq_f32(vc.val[0], l_nc.val[0]));
379  mask.val[0] = vandq_u32(mask.val[0], vcgtq_f32(vc.val[0], m_nc.val[0]));
380  mask.val[0] = vandq_u32(mask.val[0], vcgtq_f32(vc.val[0], r_nc.val[0]));
381  mask.val[1] = vandq_u32(mask.val[1], vcgtq_f32(vc.val[1], l_nc.val[1]));
382  mask.val[1] = vandq_u32(mask.val[1], vcgtq_f32(vc.val[1], m_nc.val[1]));
383  mask.val[1] = vandq_u32(mask.val[1], vcgtq_f32(vc.val[1], r_nc.val[1]));
384  mask.val[2] = vandq_u32(mask.val[2], vcgtq_f32(vc.val[2], l_nc.val[2]));
385  mask.val[2] = vandq_u32(mask.val[2], vcgtq_f32(vc.val[2], m_nc.val[2]));
386  mask.val[2] = vandq_u32(mask.val[2], vcgtq_f32(vc.val[2], r_nc.val[2]));
387  mask.val[3] = vandq_u32(mask.val[3], vcgtq_f32(vc.val[3], l_nc.val[3]));
388  mask.val[3] = vandq_u32(mask.val[3], vcgtq_f32(vc.val[3], m_nc.val[3]));
389  mask.val[3] = vandq_u32(mask.val[3], vcgtq_f32(vc.val[3], r_nc.val[3]));
390 
391  static const float32x4_t zero = vdupq_n_f32(0.f);
392 
393  // Store
394  vst1q_f32(output + 0, vbslq_f32(mask.val[0], vc.val[0], zero));
395  vst1q_f32(output + 4, vbslq_f32(mask.val[1], vc.val[1], zero));
396  vst1q_f32(output + 8, vbslq_f32(mask.val[2], vc.val[2], zero));
397  vst1q_f32(output + 12, vbslq_f32(mask.val[3], vc.val[3], zero));
398 }
399 
400 inline void non_maxima_suppression3x3_U8_U8(const void *__restrict input_ptr, void *__restrict output_ptr, const uint32_t input_stride)
401 {
402  auto input = static_cast<const uint8_t *__restrict>(input_ptr) - 1;
403  const auto output = static_cast<uint8_t *__restrict>(output_ptr);
404 
405  // Get centre scores
406  const uint8x16_t vc = vld1q_u8(input + 1);
407 
408  // Neighboring pixels
409  uint8x16_t l_nc{};
410  uint8x16_t m_nc{};
411  uint8x16_t r_nc{};
412 
413  input -= input_stride;
414 
415  // Row0
416  l_nc = vld1q_u8(input);
417  m_nc = vld1q_u8(input + 1);
418  r_nc = vld1q_u8(input + 2);
419 
420  // mc >= l_nc, mc >= m_nc, mc >= r_nc
421  uint8x16_t mask = vcgeq_u8(vc, l_nc);
422  mask = vandq_u8(mask, vcgeq_u8(vc, m_nc));
423  mask = vandq_u8(mask, vcgeq_u8(vc, r_nc));
424 
425  input += input_stride;
426 
427  // Row1
428  l_nc = vld1q_u8(input);
429  r_nc = vld1q_u8(input + 2);
430 
431  // mc >= l_nc, mc > r_nc
432  mask = vandq_u8(mask, vcgeq_u8(vc, l_nc));
433  mask = vandq_u8(mask, vcgtq_u8(vc, r_nc));
434 
435  input += input_stride;
436 
437  // Row2
438  l_nc = vld1q_u8(input);
439  m_nc = vld1q_u8(input + 1);
440  r_nc = vld1q_u8(input + 2);
441 
442  // mc > l_nc, mc > m_nc, mc > r_nc
443  mask = vandq_u8(mask, vcgtq_u8(vc, l_nc));
444  mask = vandq_u8(mask, vcgtq_u8(vc, m_nc));
445  mask = vandq_u8(mask, vcgtq_u8(vc, r_nc));
446 
447  static const uint8x16_t zero = vdupq_n_u8(0);
448 
449  // Store
450  vst1q_u8(output, vbslq_u8(mask, vc, zero));
451 }
452 } // namespace
453 
455  : _func(nullptr), _input(nullptr), _output(nullptr)
456 {
457 }
458 
460 {
461  return BorderSize(1);
462 }
463 
464 void NENonMaximaSuppression3x3Kernel::configure(const ITensor *input, ITensor *output, bool border_undefined)
465 {
469 
470  _input = input;
471  _output = output;
472 
473  if(input->info()->data_type() == DataType::U8)
474  {
475  _func = &non_maxima_suppression3x3_U8_U8;
476  }
477  else
478  {
479  _func = &non_maxima_suppression3x3_FLOAT_FLOAT;
480  }
481 
482  constexpr unsigned int num_elems_processed_per_iteration = 16;
483  const unsigned int num_elems_read_per_iteration = 16 + 2 * border_size().left + (input->info()->data_type() == DataType::U8 ? 0 : 3);
484  constexpr unsigned int num_elems_written_per_iteration = 16;
485  constexpr unsigned int num_rows_read_per_iteration = 3;
486 
487  // Configure kernel window
488  Window win = calculate_max_window(*input->info(), Steps(num_elems_processed_per_iteration), border_undefined, border_size());
489  AccessWindowHorizontal output_access(output->info(), 0, num_elems_written_per_iteration);
490 
492  AccessWindowRectangle(input->info(), -border_size().left, -border_size().top, num_elems_read_per_iteration, num_rows_read_per_iteration),
493  output_access);
494 
495  output_access.set_valid_region(win, input->info()->valid_region(), border_undefined, border_size());
496 
497  INEKernel::configure(win);
498 }
499 
501 {
502  ARM_COMPUTE_UNUSED(info);
505  ARM_COMPUTE_ERROR_ON(_func == nullptr);
506  Iterator input(_input, window);
507  Iterator output(_output, window);
508 
509  const size_t input_stride = _input->info()->strides_in_bytes()[1] / element_size_from_data_type(_input->info()->data_type());
510 
511  execute_window_loop(window, [&](const Coordinates &)
512  {
513  _func(input.ptr(), output.ptr(), input_stride);
514  },
515  input, output);
516 }
unsigned int top
top of the border
Definition: Types.h:375
Window calculate_max_window(const ValidRegion &valid_region, const Steps &steps, bool skip_border, BorderSize border_size)
const Window & window() const
The maximum window the kernel can be executed on.
Definition: IKernel.cpp:28
Container for 2D border size.
Definition: Types.h:273
1 channel, 1 U8 per channel
size_t element_size_from_data_type(DataType dt)
The size in bytes of the data type.
Definition: Utils.h:185
virtual DataType data_type() const =0
Data type used for each element of the tensor.
1 channel, 1 F32 per channel
float16x8_t vextq_f16(float16x8_t, float16x8_t, int)
Definition: clang-tidy.h:108
#define ARM_COMPUTE_ERROR_ON(cond)
If the condition is true then an error message is printed and an exception thrown.
Definition: Error.h:466
BorderSize border_size() const override
The size of the border for that kernel.
Interface for Neon tensor.
Definition: ITensor.h:36
Copyright (c) 2017-2021 Arm Limited.
virtual ValidRegion valid_region() const =0
Valid region of the tensor.
void configure(const ITensor *input, ITensor *output, bool border_undefined)
Initialise the kernel&#39;s sources, destinations and border mode.
Implementation of a rectangular access pattern.
uint16x8_t vcgeq_f16(float16x8_t, float16x8_t)
Definition: clang-tidy.h:93
bool update_window_and_padding(Window &win, Ts &&... patterns)
Update window and padding size for each of the access patterns.
Definition: WindowHelpers.h:46
#define ARM_COMPUTE_UNUSED(...)
To avoid unused variables warnings.
Definition: Error.h:152
#define ARM_COMPUTE_ERROR_ON_MISMATCHING_DATA_TYPES(...)
Definition: Validate.h:543
Class to describe a number of elements in each dimension.
Definition: Steps.h:40
Coordinates of an item.
Definition: Coordinates.h:37
Implementation of a row access pattern.
virtual ITensorInfo * info() const =0
Interface to be implemented by the child class to return the tensor&#39;s metadata.
constexpr uint8_t * ptr() const
Return a pointer to the current pixel.
Definition: Helpers.inl:139
void run(const Window &window, const ThreadInfo &info) override
Execute the kernel on the passed window.
unsigned int left
left of the border
Definition: Types.h:378
#define ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL(k)
Definition: Validate.h:941
#define ARM_COMPUTE_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(t, c,...)
Definition: Validate.h:790
float16x8_t vbslq_f16(uint16x8_t, float16x8_t, float16x8_t)
Definition: clang-tidy.h:103
ScaleKernelInfo info(interpolation_policy, default_border_mode, PixelValue(), sampling_policy, false)
Information about executing thread and CPU.
Definition: CPPTypes.h:235
unsigned int num_elems_processed_per_iteration
void execute_window_loop(const Window &w, L &&lambda_function, Ts &&... iterators)
Iterate through the passed window, automatically adjusting the iterators and calling the lambda_funct...
Definition: Helpers.inl:77
virtual const Strides & strides_in_bytes() const =0
The strides in bytes for accessing each dimension of the tensor.
Iterator updated by execute_window_loop for each window element.
Definition: Helpers.h:46
uint16x8_t vcgtq_f16(float16x8_t, float16x8_t)
Definition: clang-tidy.h:98
Describe a multidimensional execution window.
Definition: Window.h:39
#define ARM_COMPUTE_ERROR_ON_INVALID_SUBWINDOW(f, s)
Definition: Validate.h:205