Compute Library
 20.08
CLDepthwiseConvolutionLayer3x3NHWCKernel.cpp
Go to the documentation of this file.
1 /*
2  * Copyright (c) 2018-2020 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 
32 #include "arm_compute/core/Error.h"
35 #include "arm_compute/core/Types.h"
36 #include "arm_compute/core/Utils.h"
39 #include "support/StringSupport.h"
40 
41 namespace arm_compute
42 {
43 namespace
44 {
45 Status validate_arguments(const ITensorInfo *input, const ITensorInfo *weights, const ITensorInfo *biases, const ITensorInfo *output,
46  const PadStrideInfo &conv_info, unsigned int depth_multiplier, const ActivationLayerInfo &act_info, const Size2D &dilation,
47  const ITensorInfo *output_multipliers, const ITensorInfo *output_shifts)
48 {
51  ARM_COMPUTE_RETURN_ERROR_ON_MSG((act_info.enabled()) && (input->data_type() == DataType::QASYMM8 || input->data_type() == DataType::QASYMM8_SIGNED)
53  && (act_info.activation() != ActivationLayerInfo::ActivationFunction::BOUNDED_RELU)
54  && (act_info.activation() != ActivationLayerInfo::ActivationFunction::RELU)
55  && (act_info.activation() != ActivationLayerInfo::ActivationFunction::LOGISTIC),
56  "For QASYMM8 only logistic, relu, lower bounded relu and lower-upper bounded relu are supported");
57  ARM_COMPUTE_RETURN_ERROR_ON(depth_multiplier > 1); // COMPMID-1071 Add depth multiplier support for NHWC
58 
59  ARM_COMPUTE_RETURN_ERROR_ON(conv_info.stride().first < 1);
60  ARM_COMPUTE_RETURN_ERROR_ON(std::max(conv_info.pad_top(), conv_info.pad_bottom()) > 4);
61 
62  ARM_COMPUTE_RETURN_ERROR_ON((dilation.x() < 1) || (dilation.y() < 1));
63 
64  const bool is_qasymm = is_data_type_quantized_asymmetric(input->data_type());
65  const size_t weights_width = 3;
66  const size_t weights_height = 3;
67 
69  *input, TensorInfo(TensorShape(weights_width, weights_height), 1, weights->data_type()).set_data_layout(DataLayout::NCHW), conv_info, depth_multiplier, dilation);
70  if(is_qasymm)
71  {
72  DepthwiseConvolutionReshapeInfo info;
73  info.c0 = 4;
74  ARM_COMPUTE_RETURN_ERROR_ON((weights->dimension(0) / info.c0) != weights_width * weights_height);
75 
76  ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(output_multipliers, output_shifts);
79  ARM_COMPUTE_RETURN_ERROR_ON(output_multipliers->num_dimensions() > 1);
80  ARM_COMPUTE_RETURN_ERROR_ON(output_shifts->num_dimensions() > 1);
81 
83  {
84  ARM_COMPUTE_RETURN_ERROR_ON(output_shape[0] != output_multipliers->dimension(0));
85  ARM_COMPUTE_RETURN_ERROR_ON(output_shape[0] != output_shifts->dimension(0));
86  }
87  else
88  {
90  ARM_COMPUTE_RETURN_ERROR_ON(1 != output_multipliers->dimension(0));
91  ARM_COMPUTE_RETURN_ERROR_ON(1 != output_shifts->dimension(0));
92  }
93  }
94  else
95  {
97  ARM_COMPUTE_RETURN_ERROR_ON((weights->dimension(1) != weights_width) || (weights->dimension(2) != weights_height));
98  }
99 
100  if(biases != nullptr)
101  {
102  ARM_COMPUTE_RETURN_ERROR_ON(biases->dimension(0) != output_shape[0]);
103  if(is_qasymm)
104  {
106  }
107  else
108  {
110  }
111 
112  ARM_COMPUTE_RETURN_ERROR_ON(biases->num_dimensions() > 1);
113  }
114 
115  if(output->total_size() != 0)
116  {
118  }
119 
120  return Status{};
121 }
122 
123 std::pair<Status, Window> validate_and_configure_window(ITensorInfo *input, ITensorInfo *weights, ITensorInfo *bias, ITensorInfo *output,
124  const PadStrideInfo &conv_info, unsigned int depth_multiplier, const Size2D &dilation,
125  ITensorInfo *output_multipliers, ITensorInfo *output_shifts)
126 {
127  const size_t weights_width = 3;
128  const size_t weights_height = 3;
129 
130  // Get convolved dimensions
132  *input, TensorInfo(TensorShape(weights_width, weights_height), 1, weights->data_type()).set_data_layout(DataLayout::NCHW), conv_info, depth_multiplier, dilation);
133 
134  auto_init_if_empty(*output, input->clone()->set_tensor_shape(output_shape).set_quantization_info(output->quantization_info()));
135 
136  const bool is_qasymm = is_data_type_quantized_asymmetric(input->data_type());
137  const bool is_stride_1_dilation_1 = ((conv_info.stride().first == conv_info.stride().second) && (conv_info.stride().first == 1) && dilation.x() == 1 && dilation.y() == 1);
138 
139  const unsigned int num_rows_processed_per_iteration = is_stride_1_dilation_1 ? 2 : 1;
140  const unsigned int num_elems_accessed_per_iteration = is_qasymm ? 4 : (8 / input->element_size());
141  const unsigned int num_rows_read_per_iteration = num_rows_processed_per_iteration + 2;
142  const unsigned int num_rows_written_per_iteration = std::ceil(num_rows_processed_per_iteration / static_cast<float>(conv_info.stride().first));
143 
144  BorderSize border_size;
145  border_size = BorderSize(conv_info.pad_left(), 0, std::max(std::max(conv_info.pad_right(), conv_info.pad_bottom()), conv_info.pad_top()), 0);
146 
147  // Configure kernel window
148  Window win = calculate_max_window(*output, Steps(num_elems_accessed_per_iteration, num_rows_written_per_iteration));
149 
150  AccessWindowStatic input_access(input, 0, -border_size.top, ceil_to_multiple(input->dimension(0), num_elems_accessed_per_iteration),
151  ceil_to_multiple(input->dimension(1) + border_size.bottom, num_rows_read_per_iteration));
152  AccessWindowRectangle output_access(output, 0, 0, num_elems_accessed_per_iteration, num_rows_written_per_iteration);
153 
154  bool window_changed = false;
155 
156  if(is_qasymm)
157  {
158  if((output_multipliers != nullptr) && (output_shifts != nullptr))
159  {
160  AccessWindowHorizontal output_multipliers_access(output_multipliers, 0, num_elems_accessed_per_iteration);
161  AccessWindowHorizontal output_shifts_access(output_shifts, 0, num_elems_accessed_per_iteration);
162  window_changed = window_changed || update_window_and_padding(win, input_access, output_access, output_multipliers_access, output_shifts_access);
163  }
164  else
165  {
166  Status err = ARM_COMPUTE_CREATE_ERROR(ErrorCode::RUNTIME_ERROR, "output_multipliers and output_shifts must be non-nullptr for quantized input");
167  return std::make_pair(err, win);
168  }
169  }
170  else
171  {
172  AccessWindowStatic weights_access(weights, 0, 0, ceil_to_multiple(weights->dimension(0), num_elems_accessed_per_iteration), weights->dimension(1));
173  window_changed = update_window_and_padding(win, input_access, weights_access, output_access);
174  }
175 
176  if(bias != nullptr)
177  {
178  AccessWindowHorizontal bias_access(bias, 0, num_elems_accessed_per_iteration);
179  window_changed = window_changed || update_window_and_padding(win, bias_access);
180  }
181  output_access.set_valid_region(win, ValidRegion(Coordinates(), output->tensor_shape()));
182 
183  Status err = (window_changed) ? ARM_COMPUTE_CREATE_ERROR(ErrorCode::RUNTIME_ERROR, "Insufficient Padding!") : Status{};
184  return std::make_pair(err, win);
185 }
186 } // namespace
187 
189  : _num_rows_processed_per_iteration(1), _num_planes_processed_per_iteration(1)
190 {
191 }
192 
194 {
195  return _border_size;
196 }
197 
199  const PadStrideInfo &conv_info, unsigned int depth_multiplier, ActivationLayerInfo act_info, const Size2D &dilation,
200  const ICLTensor *output_multipliers, const ICLTensor *output_shifts)
201 {
202  configure(CLKernelLibrary::get().get_compile_context(), input, weights, biases, output, conv_info, depth_multiplier, act_info, dilation, output_multipliers, output_shifts);
203 }
204 
206  const PadStrideInfo &conv_info, unsigned int depth_multiplier, ActivationLayerInfo act_info, const Size2D &dilation,
207  const ICLTensor *output_multipliers, const ICLTensor *output_shifts)
208 {
210  ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(input->info(), weights->info(), (biases != nullptr) ? biases->info() : nullptr, output->info(),
211  conv_info, depth_multiplier, act_info, dilation,
212  (output_multipliers != nullptr) ? output_multipliers->info() : nullptr,
213  (output_shifts != nullptr) ? output_shifts->info() : nullptr));
214  auto win_config = validate_and_configure_window(input->info(), weights->info(), biases != nullptr ? biases->info() : nullptr, output->info(),
215  conv_info, depth_multiplier, dilation,
216  (output_multipliers != nullptr) ? output_multipliers->info() : nullptr,
217  (output_shifts != nullptr) ? output_shifts->info() : nullptr);
218  ARM_COMPUTE_ERROR_THROW_ON(win_config.first);
219 
220  const bool is_stride_1 = ((conv_info.stride().first == conv_info.stride().second) && (conv_info.stride().first == 1));
221  const bool is_stride_1_dilation_1 = (is_stride_1 && dilation.x() == 1 && dilation.y() == 1);
222 
223  const bool is_quantized_per_channel = is_data_type_quantized_per_channel(weights->info()->data_type());
224  const bool is_dot8_supported = dot8_supported(CLKernelLibrary::get().get_device()) && !is_quantized_per_channel;
225 
226  _input = input;
227  _output = output;
228  _weights = weights;
229  _biases = biases;
230  _conv_stride_y = conv_info.stride().second;
231  _num_rows_processed_per_iteration = is_stride_1_dilation_1 ? 2 : 1;
232  _num_planes_processed_per_iteration = is_stride_1_dilation_1 ? 2 : 1;
233  _output_multipliers = output_multipliers;
234  _output_shifts = output_shifts;
235  _is_quantized = is_data_type_quantized_asymmetric(input->info()->data_type());
236 
237  // If QASYMM8 and the 8 bit dot product is available, force _num_planes_processed_per_iteration to 1
238  if(is_dot8_supported && _is_quantized)
239  {
240  _num_planes_processed_per_iteration = 1;
241  }
242 
243  _border_size = BorderSize(_is_quantized && is_stride_1 ? 0 : conv_info.pad_left(), 0, std::max(std::max(conv_info.pad_right(), conv_info.pad_bottom()), conv_info.pad_top()), 0);
244 
245  const unsigned int num_elems_accessed_per_iteration = _is_quantized ? 4 : (8 / input->info()->element_size());
246 
247  CLBuildOptions build_opts;
248  build_opts.add_option("-DACTIVATION_TYPE=" + lower_string(string_from_activation_func(act_info.activation())));
249  build_opts.add_option_if(_biases != nullptr, "-DHAS_BIAS");
250  build_opts.add_option("-DVEC_SIZE=" + support::cpp11::to_string(num_elems_accessed_per_iteration));
251  build_opts.add_option("-DSRC_DIM_2=" + support::cpp11::to_string(_input->info()->dimension(2)));
252  build_opts.add_option("-DCONV_PAD_TOP=" + support::cpp11::to_string(conv_info.pad_top()));
253  build_opts.add_option("-DCONV_PAD_LEFT=" + support::cpp11::to_string(conv_info.pad_left()));
254  build_opts.add_option("-DDILATION_X=" + support::cpp11::to_string(dilation.x()));
255  build_opts.add_option("-DDILATION_Y=" + support::cpp11::to_string(dilation.y()));
256 
257  if(_is_quantized)
258  {
259  const UniformQuantizationInfo iq_info = _input->info()->quantization_info().uniform();
260  const UniformQuantizationInfo wq_info = _weights->info()->quantization_info().uniform();
261  const UniformQuantizationInfo oq_info = _output->info()->quantization_info().uniform();
262 
263  build_opts.add_option("-DSRC_DIM_1=" + support::cpp11::to_string(_input->info()->dimension(1)));
264  build_opts.add_option("-DINPUT_OFFSET=" + support::cpp11::to_string(-iq_info.offset));
265  build_opts.add_option("-DWEIGHTS_OFFSET=" + support::cpp11::to_string(-wq_info.offset));
266  build_opts.add_option("-DOUTPUT_OFFSET=" + support::cpp11::to_string(oq_info.offset));
267  build_opts.add_option("-DK_OFFSET=" + support::cpp11::to_string(9 * iq_info.offset * wq_info.offset));
268  build_opts.add_option_if(is_quantized_per_channel, "-DPER_CHANNEL_QUANTIZATION");
269  build_opts.add_option_if(is_dot8_supported, "-DIS_DOT8");
270 
271  // Compute non-per-channel multiplier and shift anyway to make OpenCL kernel simpler
272  float multiplier = iq_info.scale * wq_info.scale / oq_info.scale;
273  int output_multiplier = 0;
274  int output_shift = 0;
275  quantization::calculate_quantized_multiplier(multiplier, &output_multiplier, &output_shift);
276  build_opts.add_option("-DOUTPUT_MULTIPLIER=" + support::cpp11::to_string(output_multiplier));
277  build_opts.add_option("-DOUTPUT_SHIFT=" + support::cpp11::to_string(output_shift));
278 
279  if(act_info.enabled())
280  {
281  const int a_val = quantize_qasymm8(act_info.a(), oq_info);
282  const int b_val = quantize_qasymm8(act_info.b(), oq_info);
283  const int o1 = oq_info.offset;
284 
285  build_opts.add_option("-DA_VAL=" + support::cpp11::to_string(a_val));
286  build_opts.add_option("-DB_VAL=" + support::cpp11::to_string(b_val));
287  build_opts.add_option("-DCONST_0=" + support::cpp11::to_string(o1));
288 
289  const float s1 = iq_info.scale;
290  build_opts.add_option("-DS1_VAL=" + float_to_string_with_full_precision(s1));
291  build_opts.add_option("-DO1_VAL=" + support::cpp11::to_string(o1));
292  }
293 
294  build_opts.add_option("-DDATA_TYPE=" + get_cl_type_from_data_type(input->info()->data_type()));
295  build_opts.add_option("-DWEIGHTS_TYPE=" + get_cl_type_from_data_type(weights->info()->data_type()));
296  build_opts.add_option("-DWEIGHTS_PROMOTED_TYPE=" + get_cl_promoted_type_from_data_type(weights->info()->data_type()));
297  }
298  else
299  {
300  build_opts.add_option_if(act_info.enabled(), "-DA_VAL=" + float_to_string_with_full_precision(act_info.a()));
301  build_opts.add_option_if(act_info.enabled(), "-DB_VAL=" + float_to_string_with_full_precision(act_info.b()));
302  build_opts.add_option("-DDATA_TYPE=" + get_cl_type_from_data_type(_input->info()->data_type()));
303  }
304 
305  if(is_stride_1_dilation_1)
306  {
307  build_opts.add_option("-DNUM_ROWS_PROCESSED=" + support::cpp11::to_string(_num_rows_processed_per_iteration));
308  build_opts.add_option("-DNUM_PLANES_PROCESSED=" + support::cpp11::to_string(_num_planes_processed_per_iteration));
309  build_opts.add_option("-DDST_DIM_2=" + support::cpp11::to_string(_output->info()->dimension(2)));
310  }
311  else
312  {
313  build_opts.add_option("-DCONV_STRIDE_X=" + support::cpp11::to_string(conv_info.stride().first));
314  build_opts.add_option("-DCONV_STRIDE_Y=" + support::cpp11::to_string(_conv_stride_y));
315  }
316  build_opts.add_option_if(_input->info()->tensor_shape().total_size_upper(3) > 1,
317  "-DDST_DEPTH=" + support::cpp11::to_string(static_cast<int>(std::ceil(_output->info()->dimension(2) / static_cast<float>(_num_planes_processed_per_iteration)))));
318 
319  std::string kernel_name;
320  // Create kernel
321  if(_is_quantized)
322  {
323  kernel_name = std::string("dwc_3x3_reshaped_quantized8");
324  kernel_name += (is_dot8_supported && is_stride_1_dilation_1 ? "_dot8" : "");
325  kernel_name += (is_stride_1_dilation_1 ? "_stride1" : "");
326  kernel_name += "_nhwc";
327  }
328  else
329  {
330  kernel_name = std::string("depthwise_convolution_3x3_nhwc");
331  kernel_name += (is_stride_1_dilation_1 ? "_stride1" : "");
332  }
333 
334  build_opts.add_option_if(input->info()->data_type() == DataType::F16, "-DIS_F16");
335  build_opts.add_option_if(input->info()->data_type() == DataType::F32, "-DIS_F32");
336 
337  ICLKernel::configure_internal(win_config.second);
338  _kernel = create_kernel(compile_context, kernel_name, build_opts.options());
339 
340  // Set config_id for enabling LWS tuning
341  _config_id = kernel_name;
342  _config_id += "_";
343  _config_id += support::cpp11::to_string(input->info()->dimension(0));
344  _config_id += "_";
345  _config_id += support::cpp11::to_string(input->info()->dimension(1));
346  _config_id += "_";
347  _config_id += support::cpp11::to_string(input->info()->dimension(2));
348  _config_id += "_";
349  _config_id += support::cpp11::to_string(output->info()->dimension(0));
350  _config_id += "_";
351  _config_id += support::cpp11::to_string(output->info()->dimension(1));
352  _config_id += "_";
353  _config_id += string_from_data_type(input->info()->data_type());
354 }
355 
357  const PadStrideInfo &conv_info, unsigned int depth_multiplier, ActivationLayerInfo act_info, const Size2D &dilation,
358  const ITensorInfo *output_multipliers, const ITensorInfo *output_shifts)
359 {
360  ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments(input, weights, biases, output, conv_info, depth_multiplier, act_info, dilation, output_multipliers, output_shifts));
361  ARM_COMPUTE_RETURN_ON_ERROR(validate_and_configure_window(input->clone().get(), weights->clone().get(),
362  biases != nullptr ? biases->clone().get() : nullptr,
363  output->clone().get(), conv_info, depth_multiplier, dilation,
364  (output_multipliers != nullptr) ? output_multipliers->clone().get() : nullptr,
365  (output_shifts != nullptr) ? output_shifts->clone().get() : nullptr)
366  .first);
367 
368  return Status{};
369 }
370 
371 void CLDepthwiseConvolutionLayer3x3NHWCKernel::run(const Window &window, cl::CommandQueue &queue)
372 {
375 
376  // Collapse window
378  const size_t total_batches = _input->info()->tensor_shape().total_size_upper(3);
379 
380  Window win = window_collapsed;
381  win.set(Window::DimZ, Window::Dimension(0, std::ceil(_output->info()->dimension(2) / static_cast<float>(_num_planes_processed_per_iteration)) * total_batches, 1));
382 
383  // Create input window and adjust
384  Window win_in = win;
385  win_in.set_dimension_step(Window::DimY, _num_rows_processed_per_iteration);
386  win_in.set_dimension_step(Window::DimZ, _conv_stride_y);
387 
388  ARM_COMPUTE_ERROR_ON((win_in.y().step() < window.y().step()) || (win_in.z().step() < window.z().step()));
389 
390  Window slice_in = win_in.first_slice_window_4D();
391  Window slice_out = win.first_slice_window_4D();
392 
393  unsigned int idx = 2 * num_arguments_per_4D_tensor() + (_is_quantized ? num_arguments_per_2D_tensor() : num_arguments_per_3D_tensor());
394 
395  if(_is_quantized)
396  {
397  Window slice;
398  slice.use_tensor_dimensions(_output_multipliers->info()->tensor_shape());
399  slice.set_dimension_step(Window::DimX, window.x().step());
400  add_1D_tensor_argument(idx, _output_multipliers, slice);
401  add_1D_tensor_argument(idx, _output_shifts, slice);
402  }
403 
404  if(_biases != nullptr)
405  {
406  Window win_biases;
407  win_biases.use_tensor_dimensions(_biases->info()->tensor_shape());
408  win_biases.set_dimension_step(Window::DimX, window.x().step());
409  add_1D_tensor_argument(idx, _biases, win_biases);
410  }
411 
412  // Calculate the max_offset.
413  // max_offset is the offset for the last NOT valid value in the Z dimension (spatial dimension Y for NHWC)
414  // |******************|
415  // | pad_top |
416  // |******************|
417  // | |
418  // | plane0 |
419  // | batch0 |
420  // |__________________|
421  // |******************| Batch 0
422  // | pad_bottom |
423  // | pad_top |
424  // |******************|
425  // | |
426  // | plane1 |
427  // | batch0 |
428  // |__________________|-----> max_offset
429  // |******************|
430  // | pad_bottom |
431  // | pad_top |
432  // |******************|
433  // | |
434  // | plane0 |
435  // | batch1 |
436  // |__________________|
437  // |******************| Batch 1
438  // | pad_bottom |
439  // | pad_top |
440  // |******************|
441  // | |
442  // | plane1 |
443  // | batch1 |
444  // |__________________|
445  // | pad_bottom |
446  // |******************|
447  const int max_offset = _input->info()->strides_in_bytes().z() * _input->info()->dimension(2) - (_input->info()->padding().bottom + _input->info()->padding().top) *
448  _input->info()->strides_in_bytes().y();
449  _kernel.setArg(idx, max_offset);
450 
451  do
452  {
453  unsigned int idx = 0;
454  add_4D_tensor_argument(idx, _input, slice_in);
455  add_4D_tensor_argument(idx, _output, slice_out);
456  if(_is_quantized)
457  {
458  add_2D_tensor_argument(idx, _weights, slice_out);
459  }
460  else
461  {
462  add_3D_tensor_argument(idx, _weights, slice_out);
463  }
464  enqueue(queue, *this, slice_out, lws_hint());
465  }
466  while(win.slide_window_slice_4D(slice_out) && win_in.slide_window_slice_4D(slice_in));
467 }
468 } // namespace arm_compute
void run(const Window &window, cl::CommandQueue &queue) override
Enqueue the OpenCL kernel to process the given window on the passed OpenCL command queue.
#define ARM_COMPUTE_RETURN_ERROR_ON_F16_UNSUPPORTED(tensor)
Definition: CLValidate.h:34
const Window & window() const
The maximum window the kernel can be executed on.
Definition: IKernel.cpp:28
bool dot8_supported(const cl::Device &device)
Helper function to check whether the cl_arm_integer_dot_product_int8 extension is supported.
Definition: CLHelpers.cpp:239
TensorShape compute_depthwise_convolution_shape(const ITensorInfo &input, const ITensorInfo &weights, PadStrideInfo conv_info, unsigned int depth_multiplier, const Size2D &dilation=Size2D(1U, 1U))
Calculate the depthwise convolution output shape of a tensor.
virtual size_t dimension(size_t index) const =0
Return the size of the requested dimension.
Container for 2D border size.
Definition: Types.h:272
void enqueue(cl::CommandQueue &queue, ICLKernel &kernel, const Window &window, const cl::NDRange &lws_hint=CLKernelLibrary::get().default_ndrange(), bool use_dummy_work_items=false)
Add the kernel to the command queue with the given window.
Definition: ICLKernel.cpp:39
const StringSet & options() const
Gets the current options list set.
#define ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(...)
Definition: Validate.h:545
uint8_t quantize_qasymm8(float value, const INFO_TYPE &qinfo, RoundingPolicy rounding_policy=RoundingPolicy::TO_NEAREST_UP)
Quantize a value given an unsigned 8-bit asymmetric quantization scheme.
constexpr int step() const
Return the step of the dimension.
Definition: Window.h:102
cl::NDRange lws_hint() const
Return the Local-Workgroup-Size hint.
Definition: ICLKernel.h:263
#define ARM_COMPUTE_RETURN_ON_ERROR(status)
Checks if a status contains an error and returns it.
Definition: Error.h:204
std::string to_string(T &&value)
Convert integer and float values to string.
virtual DataType data_type() const =0
Data type used for each element of the tensor.
#define ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(t, c,...)
Definition: Validate.h:792
1 channel, 1 F32 per channel
#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
const std::string & string_from_activation_func(ActivationLayerInfo::ActivationFunction act)
Translates a given activation function to a string.
Definition: Utils.cpp:163
static CLKernelLibrary & get()
Access the KernelLibrary singleton.
Store the tensor's metadata.
Definition: ITensorInfo.h:40
#define ARM_COMPUTE_ERROR_THROW_ON(status)
Definition: Error.h:455
Quantization info when assuming per layer quantization.
Describe one of the image's dimensions with a start, end and step.
Definition: Window.h:75
Status calculate_quantized_multiplier(float multiplier, int32_t *quant_multiplier, int32_t *shift, bool ignore_epsilon=false)
Calculate quantized representation of multiplier.
constexpr const Dimension & z() const
Alias to access the third dimension of the window.
Definition: Window.h:161
Status class.
Definition: Error.h:52
std::string lower_string(const std::string &val)
Lower a given string.
Definition: Utils.cpp:326
#define ARM_COMPUTE_RETURN_ERROR_ON(cond)
If the condition is true, an error is returned.
Definition: Error.h:296
Activation Layer Information class.
Definition: Types.h:1517
Window calculate_max_window(const ValidRegion &valid_region, const Steps &steps=Steps(), bool skip_border=false, BorderSize border_size=BorderSize())
Calculate the maximum window for a given tensor shape and border setting.
Definition: Helpers.cpp:28
void add_3D_tensor_argument(unsigned int &idx, const ICLTensor *tensor, const Window &window)
Add the passed 3D tensor's parameters to the object's kernel's arguments starting from the index idx.
Definition: ICLKernel.h:159
void use_tensor_dimensions(const TensorShape &shape, size_t first_dimension=Window::DimX)
Use the tensor's dimensions to fill the window dimensions.
Definition: Window.inl:276
Copyright (c) 2017-2020 Arm Limited.
bool auto_init_if_empty(ITensorInfo &info, const TensorShape &shape, int num_channels, DataType data_type, QuantizationInfo quantization_info=QuantizationInfo())
Auto initialize the tensor info (shape, number of channels and data type) if the current assignment i...
Definition: Helpers.inl:207
1 channel, 1 F16 per channel
ITensorInfo * info() const override
Interface to be implemented by the child class to return the tensor's metadata.
Definition: Tensor.cpp:33
1 channel, 1 S32 per channel
void add_option(std::string option)
Adds option to the existing build option list.
static Status validate(const ITensorInfo *input, const ITensorInfo *weights, const ITensorInfo *biases, const ITensorInfo *output, const PadStrideInfo &conv_info, unsigned int depth_multiplier=1, ActivationLayerInfo act_info=ActivationLayerInfo(), const Size2D &dilation=Size2D(1U, 1U), const ITensorInfo *output_multipliers=nullptr, const ITensorInfo *output_shifts=nullptr)
Static function to check if given info will lead to a valid configuration of CLDepthwiseConvolutionLa...
cl::Kernel create_kernel(const CLCompileContext &ctx, const std::string &kernel_name, const std::set< std::string > &build_opts=std::set< std::string >())
Creates an opencl kernel using a compile context.
Definition: CLHelpers.cpp:403
const std::string & string_from_data_type(DataType dt)
Convert a data type identity into a string.
Definition: Utils.cpp:135
static constexpr size_t DimX
Alias for dimension 0 also known as X dimension.
Definition: Window.h:43
bool update_window_and_padding(Window &win, Ts &&... patterns)
Update window and padding size for each of the access patterns.
Definition: Helpers.h:437
static constexpr unsigned int num_arguments_per_3D_tensor()
Returns the number of arguments enqueued per 3D tensor object.
Definition: ICLKernel.h:201
BorderSize border_size() const override
The size of the border for that kernel.
Window collapse_if_possible(const Window &full_window, size_t first, size_t last, bool *has_collapsed=nullptr) const
Collapse the dimensions between first and last if possible.
Definition: Window.inl:68
bool is_data_type_quantized_per_channel(DataType dt)
Check if a given data type is of per channel type.
Definition: Utils.h:1198
std::string float_to_string_with_full_precision(float val)
Create a string with the float in full precision.
Definition: Utils.h:1215
auto ceil_to_multiple(S value, T divisor) -> decltype(((value+divisor - 1)/divisor) *divisor)
Computes the smallest number larger or equal to value that is a multiple of divisor.
Definition: Utils.h:67
#define ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DIMENSIONS(...)
Definition: Validate.h:288
quantized, asymmetric fixed-point 8-bit number unsigned
void configure(const ICLTensor *input, const ICLTensor *weights, const ICLTensor *biases, ICLTensor *output, const PadStrideInfo &conv_info, unsigned int depth_multiplier=1, ActivationLayerInfo act_info=ActivationLayerInfo(), const Size2D &dilation=Size2D(1U, 1U), const ICLTensor *output_multipliers=nullptr, const ICLTensor *output_shifts=nullptr) override
Default move assignment operator.
std::string kernel_name
std::string get_cl_type_from_data_type(const DataType &dt)
Translates a tensor data type to the appropriate OpenCL type.
Definition: CLHelpers.cpp:37
virtual std::unique_ptr< T > clone() const =0
Provide a clone of the current object of class T.
virtual ITensorInfo * info() const =0
Interface to be implemented by the child class to return the tensor's metadata.
void add_option_if(bool cond, std::string option)
Adds option if a given condition is true;.
Padding and stride information class.
Definition: Types.h:689
void set(size_t dimension, const Dimension &dim)
Set the values of a given dimension.
Definition: Window.inl:49
static constexpr unsigned int num_arguments_per_2D_tensor()
Returns the number of arguments enqueued per 2D tensor object.
Definition: ICLKernel.h:193
static constexpr unsigned int num_arguments_per_4D_tensor()
Returns the number of arguments enqueued per 4D tensor object.
Definition: ICLKernel.h:209
Num samples, channels, height, width.
CLCompileContext class.
src_info set_data_layout(data_layout)
bool is_data_type_quantized_asymmetric(DataType dt)
Check if a given data type is of asymmetric quantized type.
Definition: Utils.h:1143
std::string get_cl_promoted_type_from_data_type(const DataType &dt)
Translates a tensor data type to the appropriate OpenCL promoted type.
Definition: CLHelpers.cpp:73
#define ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(...)
Definition: Validate.h:163
static constexpr size_t DimY
Alias for dimension 1 also known as Y dimension.
Definition: Window.h:45
void set_dimension_step(size_t dimension, int step)
Set the step of a given dimension.
Definition: Window.inl:167
void add_2D_tensor_argument(unsigned int &idx, const ICLTensor *tensor, const Window &window)
Add the passed 2D tensor's parameters to the object's kernel's arguments starting from the index idx.
Definition: ICLKernel.h:135
#define ARM_COMPUTE_ERROR_ON_NULLPTR(...)
Definition: Validate.h:161
Interface for OpenCL tensor.
Definition: ICLTensor.h:42
#define ARM_COMPUTE_CREATE_ERROR(error_code, msg)
Creates an error with a given message.
Definition: Error.h:159
static constexpr size_t DimZ
Alias for dimension 2 also known as Z dimension.
Definition: Window.h:47
Class for specifying the size of an image or rectangle.
Definition: Size2D.h:34
constexpr const Dimension & y() const
Alias to access the second dimension of the window.
Definition: Window.h:152
Status validate_arguments(const ITensorInfo *input, const ITensorInfo *bias, const ITensorInfo *output, const GEMMLowpOutputStageInfo *output_stage)
Window first_slice_window_4D() const
First 4D slice of the window.
Definition: Window.h:297
bool slide_window_slice_4D(Window &slice) const
Slide the passed 4D window slice.
Definition: Window.h:345
#define ARM_COMPUTE_RETURN_ERROR_ON_MSG(cond, msg)
If the condition is true, an error is returned.
Definition: Error.h:244
quantized, asymmetric fixed-point 8-bit number signed
void add_1D_tensor_argument(unsigned int &idx, const ICLTensor *tensor, const Window &window)
Add the passed 1D tensor's parameters to the object's kernel's arguments starting from the index idx.
Definition: ICLKernel.h:111
#define ARM_COMPUTE_ERROR_ON_INVALID_SUBWINDOW(f, s)
Definition: Validate.h:205
void add_4D_tensor_argument(unsigned int &idx, const ICLTensor *tensor, const Window &window)
Add the passed 4D tensor's parameters to the object's kernel's arguments starting from the index idx.
Definition: ICLKernel.h:169
Describe a multidimensional execution window.
Definition: Window.h:39
#define ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL(k)
Definition: Validate.h:941
SimpleTensor< T > slice(const SimpleTensor< T > &src, Coordinates starts, Coordinates ends)
constexpr const Dimension & x() const
Alias to access the first dimension of the window.
Definition: Window.h:143