Compute Library
 20.08
CLGEMMLowpOffsetContributionOutputStageKernel.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 
29 #include "arm_compute/core/Error.h"
32 #include "arm_compute/core/Types.h"
33 #include "arm_compute/core/Utils.h"
36 #include "support/StringSupport.h"
37 
38 #include <cstddef>
39 #include <cstdint>
40 
41 namespace arm_compute
42 {
43 namespace
44 {
45 Status validate_arguments(const ITensorInfo *mm_result, const ITensorInfo *vector_sum_col, const ITensorInfo *vector_sum_row, const ITensorInfo *bias, const ITensorInfo *output,
46  int32_t a_offset, int32_t b_offset, const GEMMLowpOutputStageInfo &output_stage, const ITensorInfo *output_multipliers, const ITensorInfo *output_shifts)
47 {
49 
50  if(bias != nullptr)
51  {
53  ARM_COMPUTE_RETURN_ERROR_ON(bias->num_dimensions() > 1);
54  ARM_COMPUTE_RETURN_ERROR_ON(mm_result->dimension(0) != bias->dimension(0));
55  }
56 
58  ARM_COMPUTE_RETURN_ERROR_ON(output_multipliers->num_dimensions() > 1);
60  ARM_COMPUTE_RETURN_ERROR_ON(output_shifts->num_dimensions() > 1);
61  if(output_stage.is_quantized_per_channel)
62  {
63  ARM_COMPUTE_RETURN_ERROR_ON(mm_result->dimension(0) != output_shifts->dimension(0));
64  ARM_COMPUTE_RETURN_ERROR_ON(mm_result->dimension(0) != output_multipliers->dimension(0));
65  }
66 
67  // If a_offset == 0, vector_sum_col can be a nullptr
68  if(a_offset != 0)
69  {
71  ARM_COMPUTE_RETURN_ERROR_ON(vector_sum_col->dimension(0) != mm_result->dimension(0));
72  }
73 
74  // If b_offset == 0, vector_sum_row can be a nullptr
75  if(b_offset != 0)
76  {
78 
79  // Check if input is a 3D reinterpretation
80  const bool reinterpret_as_3d = mm_result->num_dimensions() > 1 && mm_result->tensor_shape().y() != vector_sum_row->tensor_shape().x();
81 
82  // Validate input
83  ARM_COMPUTE_RETURN_ERROR_ON(reinterpret_as_3d && vector_sum_row->dimension(0) != (mm_result->dimension(1) * mm_result->dimension(2)));
84  ARM_COMPUTE_RETURN_ERROR_ON(!reinterpret_as_3d && vector_sum_row->dimension(0) != mm_result->dimension(1));
85 
86  TensorShape output_shape = mm_result->tensor_shape();
88  {
89  const unsigned int output_batch_idx = reinterpret_as_3d ? 3 : 2;
90 
91  TensorShape vector_sum_row_shape = vector_sum_row->tensor_shape();
92  vector_sum_row_shape.collapse_from(1);
93  output_shape.collapse_from(output_batch_idx);
94 
95  ARM_COMPUTE_RETURN_ERROR_ON_MSG(vector_sum_row_shape[1] != output_shape[output_batch_idx],
96  "mm_result tensor must have the same number of batches of output tensor");
97 
98  if(a_offset != 0)
99  {
100  TensorShape vector_sum_col_shape = vector_sum_col->tensor_shape();
101  vector_sum_col_shape.collapse_from(1);
102 
103  ARM_COMPUTE_RETURN_ERROR_ON_MSG(vector_sum_col_shape[1] != 1 && vector_sum_col_shape[1] != vector_sum_row_shape[1],
104  "vector_sum_col tensor must have the same number of batches of vector_sum_row_shape or the number of batches must be set to 1");
105  }
106  }
107  }
108 
110  // Checks performed when output is configured
111  if((output != nullptr) && (output->total_size() != 0))
112  {
113  ARM_COMPUTE_RETURN_ERROR_ON(output_stage.output_data_type != output->data_type());
116  }
117 
118  ARM_COMPUTE_RETURN_ERROR_ON(output_stage.gemmlowp_min_bound > output_stage.gemmlowp_max_bound);
119  ARM_COMPUTE_RETURN_ERROR_ON_MSG(output_stage.gemmlowp_multipliers.size() != output_stage.gemmlowp_shifts.size(), "per channel quantization info is incorrect");
120 
121  return Status{};
122 }
123 
124 std::pair<Status, Window> validate_and_configure_window(ITensorInfo *mm_result, ITensorInfo *vector_sum_col, ITensorInfo *vector_sum_row, ITensorInfo *bias, ITensorInfo *output,
125  int32_t a_offset, int32_t b_offset, const GEMMLowpOutputStageInfo &output_stage, ITensorInfo *output_multipliers, ITensorInfo *output_shifts)
126 {
127  constexpr unsigned int num_elems_processed_per_iteration = 4;
128  bool window_changed = false;
129 
130  // Auto initialize the output
131  auto_init_if_empty(*output, mm_result->clone()->set_data_type(output_stage.output_data_type));
132 
133  // Configure kernel window
134  Window win = calculate_max_window(*mm_result, Steps(num_elems_processed_per_iteration));
135 
136  AccessWindowHorizontal mm_result_access(mm_result, 0, num_elems_processed_per_iteration);
137  window_changed = window_changed || update_window_and_padding(win, mm_result_access);
138 
139  AccessWindowHorizontal output_access(output, 0, num_elems_processed_per_iteration);
140  window_changed = window_changed || update_window_and_padding(win, output_access);
141 
142  if(a_offset != 0)
143  {
144  AccessWindowHorizontal vector_sum_col_access(vector_sum_col, 0, num_elems_processed_per_iteration);
145  window_changed = window_changed || update_window_and_padding(win, vector_sum_col_access);
146  }
147  if(b_offset != 0)
148  {
149  AccessWindowStatic vector_sum_row_access(vector_sum_row, 0, 0, vector_sum_row->dimension(0), 0); // NOLINT
150  window_changed = window_changed || update_window_and_padding(win, vector_sum_row_access);
151  }
152 
153  if(bias != nullptr)
154  {
155  AccessWindowStatic bias_access(bias, 0, 0, ceil_to_multiple(bias->dimension(0), num_elems_processed_per_iteration), bias->tensor_shape()[1]);
156  window_changed = window_changed || update_window_and_padding(win, bias_access);
157  }
158 
159  if(output_multipliers->dimension(0) > 1)
160  {
161  AccessWindowHorizontal output_multipliers_access(output_multipliers, 0, num_elems_processed_per_iteration);
162  AccessWindowHorizontal output_shifts_access(output_shifts, 0, num_elems_processed_per_iteration);
163  window_changed = window_changed || update_window_and_padding(win, output_multipliers_access, output_shifts_access);
164  }
165 
166  Status err = (window_changed) ? ARM_COMPUTE_CREATE_ERROR(ErrorCode::RUNTIME_ERROR, "Insufficient Padding!") : Status{};
167  return std::make_pair(err, win);
168 }
169 } // namespace
170 
172  : _mm_result(nullptr),
173  _vector_sum_col(nullptr),
174  _vector_sum_row(nullptr),
175  _bias(nullptr),
176  _output(nullptr),
177  _output_multipliers(nullptr),
178  _output_shifts(nullptr),
179  _is_quantized_per_channel(false)
180 {
181 }
182 
183 void CLGEMMLowpOffsetContributionOutputStageKernel::configure(const ICLTensor *mm_result, const ICLTensor *vector_sum_col, const ICLTensor *vector_sum_row, const ICLTensor *bias, ICLTensor *output,
184  int32_t k, int32_t a_offset, int32_t b_offset, const GEMMLowpOutputStageInfo &output_stage,
185  const ICLTensor *output_multipliers, const ICLTensor *output_shifts)
186 {
187  configure(CLKernelLibrary::get().get_compile_context(), mm_result, vector_sum_col, vector_sum_row, bias, output, k, a_offset, b_offset, output_stage, output_multipliers, output_shifts);
188 }
189 
190 void CLGEMMLowpOffsetContributionOutputStageKernel::configure(const CLCompileContext &compile_context, const ICLTensor *mm_result, const ICLTensor *vector_sum_col, const ICLTensor *vector_sum_row,
191  const ICLTensor *bias, ICLTensor *output,
192  int32_t k, int32_t a_offset, int32_t b_offset, const GEMMLowpOutputStageInfo &output_stage,
193  const ICLTensor *output_multipliers, const ICLTensor *output_shifts)
194 {
195  // Perform validate step
196  ARM_COMPUTE_ERROR_ON_NULLPTR(mm_result, output, output_multipliers, output_shifts);
198  vector_sum_col != nullptr ? vector_sum_col->info() : nullptr,
199  vector_sum_row != nullptr ? vector_sum_row->info() : nullptr,
200  bias != nullptr ? bias->info() : nullptr,
201  output->info(),
202  a_offset, b_offset, output_stage,
203  output_multipliers->info(), output_shifts->info())); // NOLINT
204 
205  const int min = output_stage.gemmlowp_min_bound;
206  const int max = output_stage.gemmlowp_max_bound;
207 
208  _vector_sum_col = vector_sum_col;
209  _vector_sum_row = vector_sum_row;
210  _mm_result = mm_result;
211  _bias = bias;
212  _output = output;
213  _output_multipliers = output_multipliers;
214  _output_shifts = output_shifts;
215  _is_quantized_per_channel = output_stage.is_quantized_per_channel;
216 
217  // Check if input is a 3D reinterpretation
218  const bool reinterpret_as_3d = vector_sum_row != nullptr
219  && mm_result->info()->num_dimensions() > 1
220  && mm_result->info()->tensor_shape().y() != vector_sum_row->info()->tensor_shape().x();
221 
222  // Set the arguments to pass at compile time
223  CLBuildOptions build_opts;
224 
225  // If a_offset == 0, vector_sum_col can be a nullptr
226  if(a_offset != 0)
227  {
228  build_opts.add_option("-DA_OFFSET=" + support::cpp11::to_string(a_offset));
229  build_opts.add_option_if(vector_sum_col->info()->tensor_shape().num_dimensions() > 1, "-DSUM_COL_HAS_BATCHES");
230  }
231  // If b_offset == 0, vector_sum_row can be a nullptr
232  build_opts.add_option_if(b_offset != 0, "-DB_OFFSET=" + support::cpp11::to_string(b_offset));
233  build_opts.add_option("-DK_OFFSET=" + support::cpp11::to_string(a_offset * b_offset * k));
234  build_opts.add_option_if(reinterpret_as_3d, "-DHEIGHT_INPUT3D=" + support::cpp11::to_string(mm_result->info()->dimension(1)));
235  build_opts.add_option_if(reinterpret_as_3d, "-DDEPTH_INPUT3D=" + support::cpp11::to_string(mm_result->info()->dimension(2)));
236  build_opts.add_option_if(bias != nullptr, "-DADD_BIAS");
237  build_opts.add_option("-DRESULT_OFFSET=" + support::cpp11::to_string(output_stage.gemmlowp_offset));
238  build_opts.add_option("-DRESULT_MULTIPLIER=" + support::cpp11::to_string(output_stage.gemmlowp_multipliers[0]));
239  build_opts.add_option("-DRESULT_SHIFT=" + support::cpp11::to_string(output_stage.gemmlowp_shifts[0]));
240  build_opts.add_option_if(_is_quantized_per_channel, "-DPER_CHANNEL_QUANTIZATION");
241  build_opts.add_option("-DOUTPUT_DATA_TYPE=" + get_cl_type_from_data_type(output->info()->data_type()));
242 
243  PixelValue min_val{};
244  PixelValue max_val{};
245  std::tie(min_val, max_val) = get_min_max(output->info()->data_type());
246  build_opts.add_option_if((min > min_val.get<int32_t>()), "-DMIN_BOUND=" + support::cpp11::to_string(min));
247  build_opts.add_option_if((max < max_val.get<int32_t>()), "-DMAX_BOUND=" + support::cpp11::to_string(max));
248 
249  std::string kernel_name("gemmlowp_offset_contribution");
250  kernel_name += "_" + string_from_gemmlowp_output_stage(output_stage.type);
251 
252  // Create kernel
253  _kernel = create_kernel(compile_context, kernel_name, build_opts.options());
254 
255  // Configure kernel window
256  auto win_config = validate_and_configure_window(mm_result->info(),
257  vector_sum_col != nullptr ? vector_sum_col->info() : nullptr,
258  vector_sum_row != nullptr ? vector_sum_row->info() : nullptr,
259  bias != nullptr ? bias->info() : nullptr,
260  output->info(),
261  a_offset, b_offset, output_stage,
262  output_multipliers->info(), output_shifts->info()); // NOLINT
263  ARM_COMPUTE_ERROR_THROW_ON(win_config.first);
264  ICLKernel::configure_internal(win_config.second);
265 
266  // Set config_id for enabling LWS tuning
267  _config_id = kernel_name + "_";
268  _config_id += support::cpp11::to_string(mm_result->info()->dimension(0));
269  _config_id += "_";
270  _config_id += support::cpp11::to_string(mm_result->info()->dimension(1));
271  _config_id += "_";
272  _config_id += support::cpp11::to_string(mm_result->info()->dimension(2));
273 }
274 
275 Status CLGEMMLowpOffsetContributionOutputStageKernel::validate(const ITensorInfo *mm_result, const ITensorInfo *vector_sum_col, const ITensorInfo *vector_sum_row, const ITensorInfo *bias,
276  const ITensorInfo *output, int32_t a_offset, int32_t b_offset, const GEMMLowpOutputStageInfo &output_stage,
277  const ITensorInfo *output_multipliers, const ITensorInfo *output_shifts)
278 {
279  ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments(mm_result, vector_sum_col, vector_sum_row, bias, output, a_offset, b_offset, output_stage, output_multipliers, output_shifts));
280  ARM_COMPUTE_RETURN_ON_ERROR(validate_and_configure_window(mm_result->clone().get(),
281  vector_sum_col != nullptr ? vector_sum_col->clone().get() : nullptr,
282  vector_sum_row != nullptr ? vector_sum_row->clone().get() : nullptr,
283  bias != nullptr ? bias->clone().get() : nullptr,
284  output->clone().get(),
285  a_offset, b_offset, output_stage,
286  output_multipliers->clone().get(), output_shifts->clone().get())
287  .first); // NOLINT
288 
289  return Status{};
290 }
291 
292 void CLGEMMLowpOffsetContributionOutputStageKernel::run(const Window &window, cl::CommandQueue &queue)
293 {
296 
298  Window slice = collapsed.first_slice_window_3D();
299 
300  // Set window for vector_sum_col
301  Window win_vector_sum_col = slice;
302  win_vector_sum_col.set(Window::DimY, Window::Dimension(0, 0, 0));
303  win_vector_sum_col.set(Window::DimZ, Window::Dimension(0, 0, 0));
304 
305  // Set window for vector_sum_row
306  Window win_vector_sum_row = slice;
307  win_vector_sum_row.set(Window::DimX, Window::Dimension(0, 0, 0));
308  win_vector_sum_row.set(Window::DimY, Window::Dimension(0, 0, 0));
309  win_vector_sum_col.set(Window::DimZ, Window::Dimension(0, 0, 0));
310 
311  Window biases_slice = slice;
312  biases_slice.set(Window::DimY, Window::Dimension(0, 1, 1));
313  biases_slice.set(Window::DimZ, Window::Dimension(0, 1, 1));
314 
315  do
316  {
317  unsigned int idx = 0;
318  add_3D_tensor_argument(idx, _mm_result, slice);
319  add_2D_tensor_argument_if((_vector_sum_col != nullptr), idx, _vector_sum_col, win_vector_sum_col);
320  add_2D_tensor_argument_if((_vector_sum_row != nullptr), idx, _vector_sum_row, win_vector_sum_row);
321  add_1D_tensor_argument_if((_bias != nullptr), idx, _bias, biases_slice);
322  add_3D_tensor_argument(idx, _output, slice);
323  add_1D_tensor_argument_if(_is_quantized_per_channel, idx, _output_multipliers, biases_slice);
324  add_1D_tensor_argument_if(_is_quantized_per_channel, idx, _output_shifts, biases_slice);
325  enqueue(queue, *this, slice, lws_hint());
326  }
327  while(collapsed.slide_window_slice_3D(slice));
328 }
329 } // namespace arm_compute
void add_1D_tensor_argument_if(bool cond, 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:122
virtual size_t num_dimensions() const =0
The number of dimensions of the tensor (rank)
Class describing the value of a pixel for any image format.
Definition: PixelValue.h:34
const Window & window() const
The maximum window the kernel can be executed on.
Definition: IKernel.cpp:28
void add_2D_tensor_argument_if(bool cond, 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:146
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.
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.
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_DATA_TYPE_CHANNEL_NOT_IN(t, c,...)
Definition: Validate.h:792
void configure(const ICLTensor *mm_result, const ICLTensor *vector_sum_col, const ICLTensor *vector_sum_row, const ICLTensor *bias, ICLTensor *output, int32_t k, int32_t a_offset, int32_t b_offset, const GEMMLowpOutputStageInfo &output_stage, const ICLTensor *output_multipliers, const ICLTensor *output_shifts)
Initialise the kernel's input and output.
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
Describe one of the image's dimensions with a start, end and step.
Definition: Window.h:75
Status class.
Definition: Error.h:52
const std::string & string_from_gemmlowp_output_stage(GEMMLowpOutputStageType output_stage)
Translates a given GEMMLowp output stage to a string.
Definition: Utils.cpp:260
#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:159
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
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.
T x() const
Alias to access the size of the first dimension.
Definition: Dimensions.h:81
static Status validate(const ITensorInfo *mm_result, const ITensorInfo *vector_sum_col, const ITensorInfo *vector_sum_row, const ITensorInfo *bias, const ITensorInfo *output, int32_t a_offset, int32_t b_offset, const GEMMLowpOutputStageInfo &output_stage, const ITensorInfo *output_multipliers, const ITensorInfo *output_shifts)
Static function to check if given info will lead to a valid configuration of CLGEMMLowpOffsetContribu...
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
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
void collapse_from(size_t start)
Collapse dimensions starting from a given point.
Definition: Dimensions.h:162
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
#define ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_SHAPES(...)
Definition: Validate.h:443
virtual const TensorShape & tensor_shape() const =0
Size for each dimension of the tensor.
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
quantized, asymmetric fixed-point 8-bit number unsigned
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.
GEMMLowp output stage info.
Definition: Types.h:1881
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;.
void set(size_t dimension, const Dimension &dim)
Set the values of a given dimension.
Definition: Window.inl:49
bool slide_window_slice_3D(Window &slice) const
Slide the passed 3D window slice.
Definition: Window.h:333
CLCompileContext class.
static constexpr size_t DimY
Alias for dimension 1 also known as Y dimension.
Definition: Window.h:45
#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
unsigned int num_dimensions() const
Returns the effective dimensionality of the tensor.
Definition: Dimensions.h:122
Status validate_arguments(const ITensorInfo *input, const ITensorInfo *bias, const ITensorInfo *output, const GEMMLowpOutputStageInfo *output_stage)
unsigned int num_elems_processed_per_iteration
#define ARM_COMPUTE_RETURN_ERROR_ON_MSG(cond, msg)
If the condition is true, an error is returned.
Definition: Error.h:244
T y() const
Alias to access the size of the second dimension.
Definition: Dimensions.h:86
quantized, asymmetric fixed-point 8-bit number signed
Window first_slice_window_3D() const
First 3D slice of the window.
Definition: Window.h:289
#define ARM_COMPUTE_ERROR_ON_INVALID_SUBWINDOW(f, s)
Definition: Validate.h:205
std::tuple< PixelValue, PixelValue > get_min_max(DataType dt)
Compute the mininum and maximum values a data type can take.
Definition: Utils.h:560
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)