Compute Library
 20.02.1
CLDepthwiseConvolutionLayerNativeKernel.cpp
Go to the documentation of this file.
1 /*
2  * Copyright (c) 2019-2020 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 
31 #include "arm_compute/core/Error.h"
35 #include "arm_compute/core/Types.h"
36 #include "arm_compute/core/Utils.h"
39 
40 namespace arm_compute
41 {
42 namespace
43 {
44 Status validate_arguments(const ITensorInfo *input, const ITensorInfo *weights, const ITensorInfo *biases, const ITensorInfo *output, const DWCWeightsKernelInfo &dwc_weights_info,
45  const DWCKernelInfo &dwc_info, const PadStrideInfo &conv_info, unsigned int depth_multiplier, const Size2D &dilation,
46  const ITensorInfo *output_multipliers, const ITensorInfo *output_shifts)
47 {
48  ARM_COMPUTE_UNUSED(dwc_info);
53  ARM_COMPUTE_RETURN_ERROR_ON(depth_multiplier > 1 && dwc_weights_info.n0 != 1);
54  ARM_COMPUTE_RETURN_ERROR_ON(conv_info.stride().first < 1);
55  ARM_COMPUTE_RETURN_ERROR_ON(conv_info.stride().second < 1);
56  ARM_COMPUTE_RETURN_ERROR_ON((dilation.x() < 1) || (dilation.y() < 1));
57  const size_t idx_c = get_data_layout_dimension_index(input->data_layout(), DataLayoutDimension::CHANNEL);
58  ARM_COMPUTE_UNUSED(idx_c);
59  ARM_COMPUTE_RETURN_ERROR_ON(weights->dimension(idx_c) != (input->dimension(idx_c) * depth_multiplier));
60 
62 
63  const bool is_quantized = is_data_type_quantized(input->data_type());
64 
65  if(biases != nullptr)
66  {
67  ARM_COMPUTE_RETURN_ERROR_ON(biases->dimension(0) != output_shape[idx_c]);
68  ARM_COMPUTE_RETURN_ERROR_ON(biases->num_dimensions() > 1);
69 
70  if(is_quantized)
71  {
73  }
74  else
75  {
77  }
78  }
79 
80  if(is_quantized)
81  {
82  ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(output_multipliers, output_shifts);
85  ARM_COMPUTE_RETURN_ERROR_ON(output_multipliers->num_dimensions() > 1);
86  ARM_COMPUTE_RETURN_ERROR_ON(output_shifts->num_dimensions() > 1);
87 
89  {
91  ARM_COMPUTE_RETURN_ERROR_ON(output_shape[idx_c] != output_multipliers->dimension(0));
92  ARM_COMPUTE_RETURN_ERROR_ON(output_shape[idx_c] != output_shifts->dimension(0));
93  }
94  else
95  {
97  ARM_COMPUTE_RETURN_ERROR_ON(1 != output_multipliers->dimension(0));
98  ARM_COMPUTE_RETURN_ERROR_ON(1 != output_shifts->dimension(0));
99  }
100  }
101  else
102  {
104  }
105 
106  if(output->total_size() != 0)
107  {
110  }
111 
112  if(is_data_type_quantized(input->data_type()))
113  {
114  const UniformQuantizationInfo iq_info = input->quantization_info().uniform();
115  const UniformQuantizationInfo wq_info = weights->quantization_info().uniform();
116  const UniformQuantizationInfo oq_info = (output->total_size() != 0) ? output->quantization_info().uniform() : iq_info;
117 
118  float multiplier = iq_info.scale * wq_info.scale / oq_info.scale;
119  int output_multiplier = 0;
120  int output_shift = 0;
121  ARM_COMPUTE_RETURN_ON_ERROR(quantization::calculate_quantized_multiplier(multiplier, &output_multiplier, &output_shift));
122  }
123 
124  return Status{};
125 }
126 
127 std::pair<Status, Window> validate_and_configure_window(ITensorInfo *input, ITensorInfo *weights, ITensorInfo *bias, ITensorInfo *output, const DWCWeightsKernelInfo &dwc_weights_info,
128  const DWCKernelInfo &dwc_info, const PadStrideInfo &conv_info, unsigned int depth_multiplier, const Size2D &dilation,
129  ITensorInfo *output_multipliers, ITensorInfo *output_shifts)
130 {
131  ARM_COMPUTE_UNUSED(dwc_info);
132 
133  // Get convolved dimensions
135 
136  auto_init_if_empty(*output, input->clone()->set_tensor_shape(output_shape).set_quantization_info(output->quantization_info()));
137 
138  const unsigned int n0 = dwc_weights_info.n0;
139 
140  // Configure kernel window
141  Window win = calculate_max_window(*output, Steps(n0));
142 
143  // The following access windows are only valid in case of NHWC and because n0 must unit in case depth_multiplier > 1
144  AccessWindowHorizontal input_access(input, 0, n0);
145  AccessWindowHorizontal weights_access(weights, 0, n0);
146  AccessWindowHorizontal output_access(output, 0, n0);
147 
148  bool window_changed = false;
149 
150  if(bias != nullptr)
151  {
152  AccessWindowHorizontal bias_access(bias, 0, n0);
153  window_changed = update_window_and_padding(win, input_access, weights_access, bias_access, output_access);
154  }
155  else
156  {
157  window_changed = update_window_and_padding(win, input_access, weights_access, output_access);
158  }
159 
160  if(is_data_type_quantized(input->data_type()))
161  {
162  if((output_multipliers != nullptr) && (output_shifts != nullptr))
163  {
164  AccessWindowHorizontal output_multipliers_access(output_multipliers, 0, n0);
165  AccessWindowHorizontal output_shifts_access(output_shifts, 0, n0);
166  window_changed = window_changed || update_window_and_padding(win, output_multipliers_access, output_shifts_access);
167  }
168  else
169  {
170  Status err = ARM_COMPUTE_CREATE_ERROR(ErrorCode::RUNTIME_ERROR, "output_multipliers and output_shifts must be non-nullptr for quantized input");
171  return std::make_pair(err, win);
172  }
173  }
174 
175  output_access.set_valid_region(win, ValidRegion(Coordinates(), output->tensor_shape()));
176 
177  Status err = (window_changed) ? ARM_COMPUTE_CREATE_ERROR(ErrorCode::RUNTIME_ERROR, "Insufficient Padding!") : Status{};
178  return std::make_pair(err, win);
179 }
180 } // namespace
181 
183  : _input(nullptr),
184  _weights(nullptr),
185  _biases(nullptr),
186  _output(nullptr),
187  _depth_multiplier(1),
188  _output_multipliers(nullptr),
189  _output_shifts(nullptr),
190  _is_quantized(false)
191 {
192 }
193 
195  const DWCKernelInfo &dwc_info, const PadStrideInfo &conv_info, unsigned int depth_multiplier, const Size2D &dilation,
196  const ICLTensor *output_multipliers, const ICLTensor *output_shifts)
197 {
199  ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(input->info(), weights->info(), (biases != nullptr) ? biases->info() : nullptr, output->info(),
200  dwc_weights_info, dwc_info, conv_info, depth_multiplier, dilation,
201  (output_multipliers != nullptr) ? output_multipliers->info() : nullptr, (output_shifts != nullptr) ? output_shifts->info() : nullptr));
202 
203  auto win_config = validate_and_configure_window(input->info(), weights->info(), biases != nullptr ? biases->info() : nullptr, output->info(),
204  dwc_weights_info, dwc_info, conv_info, depth_multiplier, dilation,
205  (output_multipliers != nullptr) ? output_multipliers->info() : nullptr, (output_shifts != nullptr) ? output_shifts->info() : nullptr);
206  ARM_COMPUTE_ERROR_THROW_ON(win_config.first);
207 
208  _input = input;
209  _output = output;
210  _weights = weights;
211  _biases = biases;
212  _depth_multiplier = depth_multiplier;
213  _output_multipliers = output_multipliers;
214  _output_shifts = output_shifts;
215  _is_quantized = is_data_type_quantized(input->info()->data_type());
216 
217  const size_t idx_w = get_data_layout_dimension_index(input->info()->data_layout(), DataLayoutDimension::WIDTH);
218  const size_t idx_h = get_data_layout_dimension_index(input->info()->data_layout(), DataLayoutDimension::HEIGHT);
219  const size_t weights_width = weights->info()->dimension(idx_w);
220  const size_t weights_height = weights->info()->dimension(idx_h);
221 
222  CLBuildOptions build_opts;
223  build_opts.add_option_if(_biases != nullptr, "-DHAS_BIAS");
224  build_opts.add_option_if(_input->info()->tensor_shape().total_size_upper(3) > 1, "-DDST_DEPTH=" + support::cpp11::to_string(static_cast<int>(_output->info()->dimension(2))));
225  build_opts.add_option("-DDATA_TYPE=" + get_cl_type_from_data_type(_input->info()->data_type()));
226  build_opts.add_option("-DACTIVATION_TYPE=" + lower_string(string_from_activation_func(dwc_info.activation_info.activation())));
227  build_opts.add_option("-DDEPTH_MULTIPLIER=" + support::cpp11::to_string(depth_multiplier));
228  build_opts.add_option("-DN0=" + support::cpp11::to_string(dwc_weights_info.n0));
229  build_opts.add_option("-DSRC_DIM1=" + support::cpp11::to_string(_input->info()->dimension(1)));
230  build_opts.add_option("-DSRC_DIM2=" + support::cpp11::to_string(_input->info()->dimension(2)));
231  build_opts.add_option("-DKERNEL_WIDTH=" + support::cpp11::to_string(weights_width));
232  build_opts.add_option("-DKERNEL_HEIGHT=" + support::cpp11::to_string(weights_height));
233  build_opts.add_option("-DCONV_PAD_TOP=" + support::cpp11::to_string(conv_info.pad_top()));
234  build_opts.add_option("-DCONV_PAD_LEFT=" + support::cpp11::to_string(conv_info.pad_left()));
235  build_opts.add_option("-DCONV_STRIDE_X=" + support::cpp11::to_string(conv_info.stride().first));
236  build_opts.add_option("-DCONV_STRIDE_Y=" + support::cpp11::to_string(conv_info.stride().second));
237  build_opts.add_option("-DDILATION_X=" + support::cpp11::to_string(dilation.x()));
238  build_opts.add_option("-DDILATION_Y=" + support::cpp11::to_string(dilation.y()));
239 
240  std::string kernel_name = (_is_quantized) ? "dwc_MxN_native_quantized8_nhwc" : "dwc_MxN_native_fp_nhwc";
241 
242  if(_is_quantized)
243  {
244  const UniformQuantizationInfo iq_info = _input->info()->quantization_info().uniform();
245  const UniformQuantizationInfo wq_info = _weights->info()->quantization_info().uniform();
246  const UniformQuantizationInfo oq_info = _output->info()->quantization_info().uniform();
247 
248  build_opts.add_option("-DINPUT_OFFSET=" + support::cpp11::to_string(-iq_info.offset));
249  build_opts.add_option("-DWEIGHTS_OFFSET=" + support::cpp11::to_string(-wq_info.offset));
250  build_opts.add_option("-DOUTPUT_OFFSET=" + support::cpp11::to_string(oq_info.offset));
251  build_opts.add_option_if(is_data_type_quantized_per_channel(weights->info()->data_type()), "-DPER_CHANNEL_QUANTIZATION");
252 
253  // Compute non-per-channel multiplier and shift anyway to make OpenCL kernel simpler
254  float multiplier = iq_info.scale * wq_info.scale / oq_info.scale;
255  int output_multiplier = 0;
256  int output_shift = 0;
257  quantization::calculate_quantized_multiplier(multiplier, &output_multiplier, &output_shift);
258  build_opts.add_option("-DOUTPUT_MULTIPLIER=" + support::cpp11::to_string(output_multiplier));
259  build_opts.add_option("-DOUTPUT_SHIFT=" + support::cpp11::to_string(output_shift));
260 
261  if(dwc_info.activation_info.enabled())
262  {
263  const int a_val = quantize_qasymm8(dwc_info.activation_info.a(), oq_info);
264  const int b_val = quantize_qasymm8(dwc_info.activation_info.b(), oq_info);
265  const int o1 = oq_info.offset;
266 
267  build_opts.add_option("-DA_VAL=" + support::cpp11::to_string(a_val));
268  build_opts.add_option("-DB_VAL=" + support::cpp11::to_string(b_val));
269  build_opts.add_option("-DCONST_0=" + support::cpp11::to_string(o1));
270 
271  const float s1 = iq_info.scale;
272  build_opts.add_option("-DS1_VAL=" + float_to_string_with_full_precision(s1));
273  build_opts.add_option("-DO1_VAL=" + support::cpp11::to_string(o1));
274  }
275 
276  build_opts.add_option("-DDATA_TYPE=" + get_cl_type_from_data_type(input->info()->data_type()));
277  build_opts.add_option("-DWEIGHTS_TYPE=" + get_cl_type_from_data_type(weights->info()->data_type()));
278  }
279  else
280  {
281  build_opts.add_option_if(dwc_info.activation_info.enabled(), "-DA_VAL=" + float_to_string_with_full_precision(dwc_info.activation_info.a()));
282  build_opts.add_option_if(dwc_info.activation_info.enabled(), "-DB_VAL=" + float_to_string_with_full_precision(dwc_info.activation_info.b()));
283  }
284 
285  ICLKernel::configure_internal(win_config.second);
286  _kernel = static_cast<cl::Kernel>(CLKernelLibrary::get().create_kernel(kernel_name, build_opts.options()));
287 
288  // Set config_id for enabling LWS tuning
289  _config_id = kernel_name;
290  _config_id += "_";
291  _config_id += support::cpp11::to_string(input->info()->dimension(0));
292  _config_id += "_";
293  _config_id += support::cpp11::to_string(input->info()->dimension(1));
294  _config_id += "_";
295  _config_id += support::cpp11::to_string(input->info()->dimension(2));
296  _config_id += "_";
297  _config_id += support::cpp11::to_string(output->info()->dimension(0));
298  _config_id += "_";
299  _config_id += support::cpp11::to_string(output->info()->dimension(1));
300  _config_id += "_";
301  _config_id += support::cpp11::to_string(output->info()->dimension(2));
302  _config_id += "_";
303  _config_id += string_from_data_type(input->info()->data_type());
304 }
305 
307  const DWCWeightsKernelInfo &dwc_weights_info, const DWCKernelInfo &dwc_info, const PadStrideInfo &conv_info,
308  unsigned int depth_multiplier, const Size2D &dilation, const ITensorInfo *output_multipliers, const ITensorInfo *output_shifts)
309 {
310  ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments(input, weights, biases, output, dwc_weights_info, dwc_info, conv_info, depth_multiplier, dilation, output_multipliers, output_shifts));
311  ARM_COMPUTE_RETURN_ON_ERROR(validate_and_configure_window(input->clone().get(), weights->clone().get(),
312  biases != nullptr ? biases->clone().get() : nullptr,
313  output->clone().get(), dwc_weights_info, dwc_info, conv_info, depth_multiplier, dilation,
314  output_multipliers != nullptr ? output_multipliers->clone().get() : nullptr,
315  output_shifts != nullptr ? output_shifts->clone().get() : nullptr)
316  .first);
317 
318  return Status{};
319 }
320 
321 void CLDepthwiseConvolutionLayerNativeKernel::run(const Window &window, cl::CommandQueue &queue)
322 {
325 
326  // Collapse window
327  Window window_collapsed = window.collapse(ICLKernel::window(), Window::DimZ);
328  Window slice_in = window.first_slice_window_4D();
329  Window slice_out = window_collapsed.first_slice_window_4D();
330 
331  if(_depth_multiplier != 1)
332  {
333  ARM_COMPUTE_ERROR_ON(slice_out.x().step() != 1);
334  slice_out.set(Window::DimX, Window::Dimension(0, _input->info()->tensor_shape()[0], 1));
335  }
336 
337  unsigned int idx = 2 * num_arguments_per_4D_tensor() + num_arguments_per_3D_tensor();
338 
339  // Set output multipliers in case of quantized data type
340  if(_is_quantized)
341  {
342  add_1D_tensor_argument(idx, _output_multipliers, slice_in);
343  add_1D_tensor_argument(idx, _output_shifts, slice_in);
344  }
345 
346  if(_biases != nullptr)
347  {
348  add_1D_tensor_argument(idx, _biases, slice_in);
349  }
350 
351  do
352  {
353  idx = 0;
354  add_4D_tensor_argument(idx, _input, slice_in);
355  add_4D_tensor_argument(idx, _output, slice_out);
356  add_3D_tensor_argument(idx, _weights, slice_out);
357  enqueue(queue, *this, slice_out, lws_hint());
358  }
359  while(window_collapsed.slide_window_slice_4D(slice_out) && window.slide_window_slice_4D(slice_in));
360 }
361 } // namespace arm_compute
bool is_data_type_quantized(DataType dt)
Check if a given data type is of quantized type.
Definition: Utils.h:1117
#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
TensorInfo * info() const override
Interface to be implemented by the child class to return the tensor's metadata.
Definition: CLTensor.cpp:41
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.
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.
unsigned int n0
Number of columns processed by each thread.
#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:247
#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.
size_t dimension(size_t index) const override
Return the size of the requested dimension.
Definition: TensorInfo.h:232
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:172
size_t total_size_upper(size_t dimension) const
Collapses given dimension and above.
Definition: TensorShape.h:181
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 class.
Definition: Error.h:52
std::string lower_string(const std::string &val)
Lower a given string.
Definition: Utils.cpp:333
#define ARM_COMPUTE_RETURN_ERROR_ON(cond)
If the condition is true, an error is returned.
Definition: Error.h:296
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:158
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:202
1 channel, 1 F16 per channel
DataType data_type() const override
Data type used for each element of the tensor.
Definition: TensorInfo.h:265
1 channel, 1 S32 per channel
void run(const Window &window, cl::CommandQueue &queue) override
Enqueue the OpenCL kernel to process the given window on the passed OpenCL command queue.
void add_option(std::string option)
Adds option to the existing build option list.
Window collapse(const Window &full_window, size_t first, size_t last=Coordinates::num_max_dimensions) const
Collapse the dimensions between first and last.
Definition: Window.inl:111
const std::string & string_from_data_type(DataType dt)
Convert a data type identity into a string.
Definition: Utils.cpp:144
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:402
static constexpr unsigned int num_arguments_per_3D_tensor()
Returns the number of arguments enqueued per 3D tensor object.
Definition: ICLKernel.h:200
#define ARM_COMPUTE_UNUSED(...)
To avoid unused variables warnings.
Definition: Error.h:152
bool is_data_type_quantized_per_channel(DataType dt)
Check if a given data type is of per channel type.
Definition: Utils.h:1194
std::string float_to_string_with_full_precision(float val)
Create a string with the float in full precision.
Definition: Utils.h:1211
virtual const TensorShape & tensor_shape() const =0
Size for each dimension of the tensor.
#define ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DIMENSIONS(...)
Definition: Validate.h:288
quantized, asymmetric fixed-point 8-bit number unsigned
std::string kernel_name
UniformQuantizationInfo uniform() const
Return per layer quantization info.
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
Descriptor used by the depthwise convolution kernels.
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:686
void set(size_t dimension, const Dimension &dim)
Set the values of a given dimension.
Definition: Window.inl:49
std::unique_ptr< Kernel > create_kernel()
Helper function to create and return a unique_ptr pointed to a CL/GLES kernel object.
Definition: Helpers.h:86
static constexpr unsigned int num_arguments_per_4D_tensor()
Returns the number of arguments enqueued per 4D tensor object.
Definition: ICLKernel.h:208
void configure(const ICLTensor *input, const ICLTensor *weights, const ICLTensor *biases, ICLTensor *output, const DWCWeightsKernelInfo &dwc_weights_info, const DWCKernelInfo &dwc_info, const PadStrideInfo &conv_info, unsigned int depth_multiplier=1, const Size2D &dilation=Size2D(1U, 1U), const ICLTensor *output_multipliers=nullptr, const ICLTensor *output_shifts=nullptr)
Initialize the function's source, destination and parameters.
Descriptor used by the depthwise convolution kernels to retrieve the number of output elements proces...
virtual QuantizationInfo quantization_info() const =0
Get the quantization settings (scale and offset) of the tensor.
quantized, symmetric per channel fixed-point 8-bit number
#define ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(...)
Definition: Validate.h:163
#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
Num samples, height, width, channels.
Status calculate_quantized_multiplier(float multiplier, int32_t *quant_multiplier, int32_t *shift)
Calculate quantized representation of multiplier.
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
quantized, asymmetric fixed-point 8-bit number signed
size_t get_data_layout_dimension_index(const DataLayout data_layout, const DataLayoutDimension data_layout_dimension)
Get the index of the given dimension.
Definition: Helpers.inl:327
#define ARM_COMPUTE_RETURN_ERROR_ON_DATA_LAYOUT_NOT_IN(t,...)
Definition: Validate.h:746
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:110
#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:168
Describe a multidimensional execution window.
Definition: Window.h:39
#define ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL(k)
Definition: Validate.h:941
static Status validate(const ITensorInfo *input, const ITensorInfo *weights, const ITensorInfo *biases, const ITensorInfo *output, const DWCWeightsKernelInfo &dwc_weights_info, const DWCKernelInfo &dwc_info, const PadStrideInfo &conv_info, unsigned int depth_multiplier=1, 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...
constexpr const Dimension & x() const
Alias to access the first dimension of the window.
Definition: Window.h:143