Compute Library
 22.11
ClElementwiseKernel.cpp
Go to the documentation of this file.
1 /*
2  * Copyright (c) 2018-2021 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 "src/core/CL/CLValidate.h"
32 #include "support/Cast.h"
33 #include "support/StringSupport.h"
34 #include <map>
35 
36 namespace arm_compute
37 {
38 namespace opencl
39 {
40 namespace kernels
41 {
42 namespace
43 {
44 constexpr unsigned int vector_size_byte_opencl = 16;
45 
46 std::map<ArithmeticOperation, std::string> supported_arithmetic_ops =
47 {
48  { ArithmeticOperation::ADD, "ADD" },
49  { ArithmeticOperation::SUB, "SUB" },
50  { ArithmeticOperation::DIV, "DIV" },
51  { ArithmeticOperation::SQUARED_DIFF, "SQUARED_DIFF" },
52  { ArithmeticOperation::MIN, "MIN" },
53  { ArithmeticOperation::MAX, "MAX" },
54  { ArithmeticOperation::POWER, "POWER" },
55  { ArithmeticOperation::PRELU, "PRELU" },
56 };
57 
58 std::map<ArithmeticOperation, std::string> supported_sat_arithmetic_ops =
59 {
60  { ArithmeticOperation::ADD, "ADD" },
61  { ArithmeticOperation::SUB, "SUB" },
62 };
63 
64 std::string generate_id_for_tuning_common(const std::string &kernel_name, const ITensorInfo &src1, const ITensorInfo &dst)
65 {
66  std::string config_id;
67  // Set config_id for enabling LWS tuning
68  config_id = kernel_name;
69  config_id += "_";
70  config_id += lower_string(string_from_data_type(src1.data_type()));
71  config_id += "_";
72  config_id += support::cpp11::to_string(dst.dimension(0));
73  config_id += "_";
74  config_id += support::cpp11::to_string(dst.dimension(1));
75  return config_id;
76 }
77 
78 Status validate_in_place_output_shape(const bool in_place, const bool src1_in_place, const ITensorInfo &src1, const ITensorInfo &src2, const ITensorInfo &dst, const TensorShape &out_shape)
79 {
80  if(in_place)
81  {
82  ARM_COMPUTE_RETURN_ERROR_ON_MSG(detail::have_different_dimensions(out_shape, src1_in_place ? src1.tensor_shape() : src2.tensor_shape(), 0),
83  "Wrong shape for dst, cannot do in_place calculation");
84  }
85  else
86  {
88  "Wrong shape for dst");
89  }
90  return Status{};
91 }
92 
93 Status validate_arguments_with_float_only_supported_rules(const ITensorInfo &src1, const ITensorInfo &src2, const ITensorInfo &dst)
94 {
95  ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(&src1, &src2, &dst);
99 
100  // Check whether it is in_place calculation
101  const bool in_place = (&src1 == &dst) || (&src2 == &dst);
102  const bool src1_in_place = in_place && (&src1 == &dst);
103 
104  const TensorShape out_shape = TensorShape::broadcast_shape(src1.tensor_shape(), src2.tensor_shape());
105 
106  ARM_COMPUTE_RETURN_ERROR_ON_MSG(out_shape.total_size() == 0, "Inputs are not broadcast compatible");
107 
108  // Validate in case of configured dst
109  if(dst.total_size() > 0)
110  {
113  ARM_COMPUTE_RETURN_ON_ERROR(validate_in_place_output_shape(in_place, src1_in_place, src1, src2, dst, out_shape));
114  }
115 
116  return Status{};
117 }
118 
119 Status validate_arguments_divide_operation(const ITensorInfo *src1, const ITensorInfo *src2, const ITensorInfo *dst)
120 {
121  ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(src1, src2, dst);
125 
126  // Check whether it is in_place calculation
127  const bool in_place = (src1 == dst) || (src2 == dst);
128  const bool src1_in_place = in_place && (src1 == dst);
129 
130  const TensorShape out_shape = TensorShape::broadcast_shape(src1->tensor_shape(), src2->tensor_shape());
131 
132  ARM_COMPUTE_RETURN_ERROR_ON_MSG(out_shape.total_size() == 0, "Inputs are not broadcast compatible");
133 
134  // Validate in case of configured dst
135  if(dst->total_size() > 0)
136  {
139  ARM_COMPUTE_RETURN_ON_ERROR(validate_in_place_output_shape(in_place, src1_in_place, *src1, *src2, *dst, out_shape));
140  }
141 
142  return Status{};
143 }
144 
145 Status validate_arguments_with_arithmetic_rules(const ITensorInfo &src1, const ITensorInfo &src2, const ITensorInfo &dst)
146 {
152 
153  if(is_data_type_quantized_symmetric(src1.data_type()))
154  {
155  const int32_t in1_offset = src1.quantization_info().uniform().offset;
156  const int32_t in2_offset = src2.quantization_info().uniform().offset;
157  ARM_COMPUTE_RETURN_ERROR_ON_MSG(in1_offset != 0, "For quantized symmetric, offset must be zero");
158  ARM_COMPUTE_RETURN_ERROR_ON_MSG(in2_offset != 0, "For quantized symmetric, offset must be zero");
159  }
160 
161  // Check whether it is in_place calculation
162  const bool in_place = (&src1 == &dst) || (&src2 == &dst);
163  const bool src1_in_place = in_place && (&src1 == &dst);
164 
165  const TensorShape out_shape = TensorShape::broadcast_shape(src1.tensor_shape(), src2.tensor_shape());
166  ARM_COMPUTE_RETURN_ERROR_ON_MSG(out_shape.total_size() == 0, "Inputs are not broadcast compatible");
167 
168  // Validate in case of configured dst
169  if(dst.total_size() > 0)
170  {
172  ARM_COMPUTE_RETURN_ERROR_ON_MSG(detail::have_different_dimensions(out_shape, dst.tensor_shape(), 0), "Wrong shape for dst");
173  ARM_COMPUTE_RETURN_ON_ERROR(validate_in_place_output_shape(in_place, src1_in_place, src1, src2, dst, out_shape));
174 
175  if(is_data_type_quantized_symmetric(dst.data_type()))
176  {
177  const int32_t offset = dst.quantization_info().uniform().offset;
178  ARM_COMPUTE_RETURN_ERROR_ON_MSG(offset != 0, "For quantized symmetric, offset must be zero");
179  }
180  }
181  return Status{};
182 }
183 
184 CLBuildOptions generate_build_options_with_arithmetic_rules(const ITensorInfo &src1, const ITensorInfo &src2, const ITensorInfo &dst, const std::string &operation_string)
185 {
186  CLBuildOptions build_opts;
187 
188  const unsigned int num_elems_processed_per_iteration = adjust_vec_size(vector_size_byte_opencl / dst.element_size(), dst.dimension(0));
189 
190  build_opts.add_option("-DDATA_TYPE=" + get_cl_type_from_data_type(src1.data_type()));
191  build_opts.add_option("-DVEC_SIZE_IN1=" + support::cpp11::to_string(src1.dimension(0) == 1 ? 1 : num_elems_processed_per_iteration));
192  build_opts.add_option("-DVEC_SIZE_IN2=" + support::cpp11::to_string(src2.dimension(0) == 1 ? 1 : num_elems_processed_per_iteration));
193  build_opts.add_option("-DVEC_SIZE_OUT=" + support::cpp11::to_string(num_elems_processed_per_iteration));
194  build_opts.add_option("-DVEC_SIZE_LEFTOVER=" + support::cpp11::to_string(dst.dimension(0) % num_elems_processed_per_iteration));
195  build_opts.add_option("-DOP=" + operation_string);
196  if(is_data_type_quantized(src1.data_type()))
197  {
198  const UniformQuantizationInfo iq1info = src1.quantization_info().uniform();
199  const UniformQuantizationInfo iq2info = src2.quantization_info().uniform();
200  const UniformQuantizationInfo oqinfo = dst.quantization_info().uniform();
201 
202  build_opts.add_option("-DOFFSET_IN1=" + support::cpp11::to_string(iq1info.offset));
203  build_opts.add_option("-DOFFSET_IN2=" + support::cpp11::to_string(iq2info.offset));
204  build_opts.add_option("-DOFFSET_OUT=" + support::cpp11::to_string(oqinfo.offset));
205  build_opts.add_option("-DSCALE_IN1=" + float_to_string_with_full_precision(iq1info.scale));
206  build_opts.add_option("-DSCALE_IN2=" + float_to_string_with_full_precision(iq2info.scale));
207  build_opts.add_option("-DSCALE_OUT=" + float_to_string_with_full_precision(oqinfo.scale));
208  }
209  build_opts.add_option_if(src1.data_type() == DataType::S32, "-DS32");
210 
211  // Check whether it is in_place calculation
212  const bool in_place = (&src1 == &dst) || (&src2 == &dst);
213  const bool src1_in_place = in_place && (&src1 == &dst);
214  build_opts.add_option_if(in_place, "-DIN_PLACE");
215  build_opts.add_option_if(src1_in_place, "-DSRC1_IN_PLACE");
216 
217  return build_opts;
218 }
219 
220 std::pair<Status, Window> configure_window_arithmetic_common(ITensorInfo &dst)
221 {
222  const unsigned int num_elems_processed_per_iteration = adjust_vec_size(vector_size_byte_opencl / dst.element_size(), dst.dimension(0));
223  Window win = calculate_max_window(dst, Steps(num_elems_processed_per_iteration));
224  return std::make_pair(Status{}, win);
225 }
226 
227 std::pair<Status, Window> validate_and_configure_window_for_arithmetic_operators(ITensorInfo &src1, ITensorInfo &src2, ITensorInfo &dst)
228 {
229  const std::pair<TensorShape, ValidRegion> broadcast_pair = ITensorInfo::broadcast_shape_and_valid_region(src1, src2);
230  const TensorShape &out_shape = broadcast_pair.first;
231 
232  auto_init_if_empty(dst, out_shape, 1, src1.data_type());
233 
234  return configure_window_arithmetic_common(dst);
235 }
236 
237 std::pair<Status, Window> validate_and_configure_window_for_logical_binary_operators(ITensorInfo &src1, ITensorInfo &src2, ITensorInfo &dst)
238 {
239  const std::pair<TensorShape, ValidRegion> broadcast_pair = ITensorInfo::broadcast_shape_and_valid_region(src1, src2);
240  const TensorShape &out_shape = broadcast_pair.first;
241 
242  set_shape_if_empty(dst, out_shape);
244 
245  return configure_window_arithmetic_common(dst);
246 }
247 
248 std::pair<Status, Window> validate_and_configure_window_for_division(ITensorInfo &src1, ITensorInfo &src2, ITensorInfo &dst)
249 {
250  const std::pair<TensorShape, ValidRegion> broadcast_pair = ITensorInfo::broadcast_shape_and_valid_region(src1, src2);
251  const TensorShape &out_shape = broadcast_pair.first;
252 
253  auto_init_if_empty(dst, out_shape, 1, src1.data_type());
254 
255  return configure_window_arithmetic_common(dst);
256 }
257 } // namespace
258 
260 {
262 }
263 
264 void ClElementwiseKernel::configure_common(const ClCompileContext &compile_context, ITensorInfo *src1, ITensorInfo *src2, ITensorInfo *dst)
265 {
266  // Configure kernel window
267  auto win_config = validate_and_configure_window(*src1, *src2, *dst);
268  ARM_COMPUTE_ERROR_THROW_ON(win_config.first);
269 
270  std::string kernel_name = "elementwise_operation_" + name();
271  if(is_data_type_quantized(src1->data_type()))
272  {
273  kernel_name += "_quantized";
274  }
275 
276  // Set kernel build options
277  CLBuildOptions build_opts = generate_build_options(*src1, *src2, *dst);
278  if(_act_info.enabled())
279  {
280  build_opts.add_option("-DACTIVATION_TYPE=" + lower_string(string_from_activation_func(_act_info.activation())));
281  build_opts.add_option("-DA_VAL=" + float_to_string_with_full_precision(_act_info.a()));
282  build_opts.add_option("-DB_VAL=" + float_to_string_with_full_precision(_act_info.b()));
283  }
284 
285  // Create kernel
286  _kernel = create_kernel(compile_context, kernel_name, build_opts.options());
287 
288  ICLKernel::configure_internal(win_config.second);
289 
290  _config_id = generate_id_for_tuning(kernel_name, *src1, *dst);
291 }
292 
293 void ClElementwiseKernel::run_op(ITensorPack &tensors, const Window &window, ::cl::CommandQueue &queue)
294 {
297 
298  const auto src_0 = utils::cast::polymorphic_downcast<const ICLTensor *>(tensors.get_const_tensor(TensorType::ACL_SRC_0));
299  const auto src_1 = utils::cast::polymorphic_downcast<const ICLTensor *>(tensors.get_const_tensor(TensorType::ACL_SRC_1));
300  auto dst = utils::cast::polymorphic_downcast<ICLTensor *>(tensors.get_tensor(TensorType::ACL_DST));
301 
302  ARM_COMPUTE_ERROR_ON_NULLPTR(src_0, src_1, dst);
303 
304  const TensorShape &in_shape1 = src_0->info()->tensor_shape();
305  const TensorShape &in_shape2 = src_1->info()->tensor_shape();
306  const TensorShape &out_shape = dst->info()->tensor_shape();
307 
308  bool can_collapse = true;
309  const bool is_vector = in_shape1.num_dimensions() == 1 || in_shape2.num_dimensions() == 1;
310  if(std::min(in_shape1.total_size(), in_shape2.total_size()) > 1 && !is_vector)
311  {
312  can_collapse = (std::min(in_shape1.num_dimensions(), in_shape2.num_dimensions()) > Window::DimZ);
313  for(size_t d = Window::DimZ; can_collapse && (d < out_shape.num_dimensions()); d++)
314  {
315  can_collapse = (in_shape1[d] == in_shape2[d]);
316  }
317  }
318 
319  bool has_collapsed = false;
320  Window collapsed = can_collapse ? window.collapse_if_possible(ICLKernel::window(), Window::DimZ, &has_collapsed) : window;
321 
322  const TensorShape &in_shape1_collapsed = has_collapsed ? in_shape1.collapsed_from(Window::DimZ) : in_shape1;
323  const TensorShape &in_shape2_collapsed = has_collapsed ? in_shape2.collapsed_from(Window::DimZ) : in_shape2;
324 
325  Window slice = collapsed.first_slice_window_3D();
326  Window slice_src1 = slice.broadcast_if_dimension_le_one(in_shape1_collapsed);
327  Window slice_src2 = slice.broadcast_if_dimension_le_one(in_shape2_collapsed);
328 
329  // Check whether it is in_place calculation
330  const bool in_place = (src_0 == dst) || (src_1 == dst);
331  do
332  {
333  unsigned int idx = 0;
334  add_3D_tensor_argument(idx, src_0, slice_src1);
335  add_3D_tensor_argument(idx, src_1, slice_src2);
336  if(!in_place)
337  {
338  add_3D_tensor_argument(idx, dst, slice);
339  }
340 
341  enqueue(queue, *this, slice, lws_hint());
342  ARM_COMPUTE_UNUSED(collapsed.slide_window_slice_3D(slice_src1));
343  ARM_COMPUTE_UNUSED(collapsed.slide_window_slice_3D(slice_src2));
344  }
345  while(collapsed.slide_window_slice_3D(slice));
346 }
347 
348 /** Logical binary */
349 
351 {
352  ARM_COMPUTE_ERROR_ON_NULLPTR(src1, src2, dst);
354  _op = op;
355  configure_common(compile_context, src1, src2, dst);
356 }
357 
359 {
360  ARM_COMPUTE_UNUSED(op);
362  ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(src1, src2, dst);
363 
366 
367  ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments_with_arithmetic_rules(*src1, *src2, *dst));
368  ARM_COMPUTE_RETURN_ON_ERROR(validate_and_configure_window_for_logical_binary_operators(*src1->clone(), *src2->clone(), *dst->clone()).first);
369 
370  return Status{};
371 }
372 
373 std::string ClLogicalBinaryKernel::name()
374 {
375  switch(_op)
376  {
378  return "AND";
380  return "OR";
382  /* fall through */
383  default:
384  ARM_COMPUTE_ASSERT(true);
385  }
386  return "";
387 }
388 
389 std::pair<Status, Window> ClLogicalBinaryKernel::validate_and_configure_window(ITensorInfo &src1, ITensorInfo &src2, ITensorInfo &dst)
390 {
391  return validate_and_configure_window_for_logical_binary_operators(src1, src2, dst);
392 }
393 
394 CLBuildOptions ClLogicalBinaryKernel::generate_build_options(const ITensorInfo &src1, const ITensorInfo &src2, const ITensorInfo &dst)
395 {
396  // The arithmetic utility functions can be share
397  return generate_build_options_with_arithmetic_rules(src1, src2, dst, name());
398 }
399 
400 std::string ClLogicalBinaryKernel::generate_id_for_tuning(const std::string &kernel_name, const ITensorInfo &src1, const ITensorInfo &dst)
401 {
402  return generate_id_for_tuning_common(kernel_name, src1, dst);
403 }
404 
405 /** Arithmetic operations with saturation*/
407  const ConvertPolicy &policy,
408  const ActivationLayerInfo &act_info)
409 {
410  ARM_COMPUTE_ERROR_ON_NULLPTR(input1, input2, output);
411  ARM_COMPUTE_ERROR_THROW_ON(ClSaturatedArithmeticKernel::validate(op, input1, input2, output, policy, act_info));
412  auto padding_info = get_padding_info({ input1, input2, output });
413 
414  _policy = policy;
415  _op = op;
416  _act_info = act_info;
417  configure_common(compile_context, input1, input2, output);
419 }
420 
422  const ActivationLayerInfo &act_info)
423 {
424  ARM_COMPUTE_UNUSED(op, policy);
425  ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(input1, input2, output);
426  ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments_with_arithmetic_rules(*input1, *input2, *output));
427  ARM_COMPUTE_RETURN_ON_ERROR(validate_and_configure_window_for_arithmetic_operators(*input1->clone(), *input2->clone(), *output->clone()).first);
429 
430  return Status{};
431 }
432 
433 std::pair<Status, Window> ClSaturatedArithmeticKernel::validate_and_configure_window(ITensorInfo &input1, ITensorInfo &input2, ITensorInfo &output)
434 {
435  return validate_and_configure_window_for_arithmetic_operators(input1, input2, output);
436 }
437 
438 CLBuildOptions ClSaturatedArithmeticKernel::generate_build_options(const ITensorInfo &input1, const ITensorInfo &input2, const ITensorInfo &output)
439 {
440  const bool has_float_out = is_data_type_float(output.data_type());
441  auto build_options = generate_build_options_with_arithmetic_rules(input1, input2, output, name());
442  build_options.add_option((_policy == ConvertPolicy::WRAP || has_float_out) ? "-DWRAP" : "-DSATURATE");
443  return build_options;
444 }
445 
446 std::string ClSaturatedArithmeticKernel::generate_id_for_tuning(const std::string &kernel_name, const ITensorInfo &input1, const ITensorInfo &output)
447 {
448  auto config_id = generate_id_for_tuning_common(kernel_name, input1, output);
449  config_id += (_policy == ConvertPolicy::WRAP) ? "_wrap_" : "_saturate_";
450  config_id += lower_string(string_from_data_layout(input1.data_layout()));
451  return config_id;
452 }
453 
454 std::string ClSaturatedArithmeticKernel::name()
455 {
456  return supported_sat_arithmetic_ops[_op];
457 }
458 
459 /** Arithmetic operations*/
461  const ActivationLayerInfo &act_info)
462 {
463  ARM_COMPUTE_ERROR_ON_NULLPTR(src1, src2, dst);
464  ARM_COMPUTE_ERROR_THROW_ON(ClArithmeticKernel::validate(op, src1, src2, dst, act_info));
465  auto padding_info = get_padding_info({ src1, src2, dst });
466 
467  _op = op;
468  _act_info = act_info;
469  configure_common(compile_context, src1, src2, dst);
471 }
472 
474 {
475  ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(src1, src2, dst);
476  if(op == ArithmeticOperation::DIV)
477  {
478  // Partial integer support S32/F32/F16
479  ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments_divide_operation(src1, src2, dst));
480  ARM_COMPUTE_RETURN_ON_ERROR(validate_and_configure_window_for_division(*src1->clone(), *src2->clone(), *dst->clone()).first);
481  }
482  else if(op == ArithmeticOperation::POWER)
483  {
484  // Power operators doesn't support integer arithmetic
485  ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments_with_float_only_supported_rules(*src1, *src2, *dst));
486  ARM_COMPUTE_RETURN_ON_ERROR(validate_and_configure_window_for_division(*src1->clone(), *src2->clone(), *dst->clone()).first);
487  }
488  else
489  {
490  ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments_with_arithmetic_rules(*src1, *src2, *dst));
491  ARM_COMPUTE_RETURN_ON_ERROR(validate_and_configure_window_for_arithmetic_operators(*src1->clone(), *src2->clone(), *dst->clone()).first);
492  }
494 
495  return Status{};
496 }
497 std::pair<Status, Window> ClArithmeticKernel::validate_and_configure_window(ITensorInfo &src1, ITensorInfo &src2, ITensorInfo &dst)
498 {
500  {
501  // Division and Power operators don't support integer arithmetic
502  return validate_and_configure_window_for_division(src1, src2, dst);
503  }
504  else
505  {
506  return validate_and_configure_window_for_arithmetic_operators(src1, src2, dst);
507  }
508 }
509 
510 CLBuildOptions ClArithmeticKernel::generate_build_options(const ITensorInfo &src1, const ITensorInfo &src2, const ITensorInfo &dst)
511 {
512  return generate_build_options_with_arithmetic_rules(src1, src2, dst, name());
513 }
514 std::string ClArithmeticKernel::generate_id_for_tuning(const std::string &kernel_name, const ITensorInfo &src1, const ITensorInfo &dst)
515 {
516  return generate_id_for_tuning_common(kernel_name, src1, dst);
517 }
518 
519 std::string ClArithmeticKernel::name()
520 {
521  return supported_arithmetic_ops[_op];
522 }
523 } // namespace kernels
524 } // namespace opencl
525 } // namespace arm_compute
static Status validate(LogicalOperation op, const ITensorInfo *src1, const ITensorInfo *src2, const ITensorInfo *dst)
Static function to check if given info will lead to a valid configuration.
bool is_data_type_quantized(DataType dt)
Check if a given data type is of quantized type.
Definition: Utils.h:1030
static Status validate(ArithmeticOperation op, const ITensorInfo *src1, const ITensorInfo *src2, const ITensorInfo *dst, const ActivationLayerInfo &act_info=ActivationLayerInfo())
Static function to check if given info will lead to a valid configuration.
__global uchar * offset(const Image *img, int x, int y)
Get the pointer position of a Image.
Definition: helpers.h:1084
Window calculate_max_window(const ValidRegion &valid_region, const Steps &steps, bool skip_border, BorderSize border_size)
#define ARM_COMPUTE_ASSERT(cond)
Definition: Validate.h:37
ArithmeticOperation
Available element-wise operations.
Definition: Types.h:489
#define ARM_COMPUTE_RETURN_ERROR_ON_F16_UNSUPPORTED(tensor)
Definition: CLValidate.h:35
const Window & window() const
The maximum window the kernel can be executed on.
Definition: IKernel.cpp:28
Shape of a tensor.
Definition: TensorShape.h:39
quantized, symmetric fixed-point 16-bit number
bool set_data_type_if_unknown(ITensorInfo &info, DataType data_type)
Set the data type and number of channels to the specified value if the current data type is unknown...
bool enabled() const
Check if initialised.
Definition: Types.h:1694
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:32
const StringSet & options() const
Gets the current options list set.
cl::NDRange lws_hint() const
Return the Local-Workgroup-Size hint.
Definition: ICLKernel.h:383
TensorShape collapsed_from(size_t start) const
Return a copy with collapsed dimensions starting from a given point.
Definition: TensorShape.h:161
1 channel, 1 U8 per channel
float a() const
Get the alpha value.
Definition: Types.h:1684
#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.
1 channel, 1 F32 per channel
static TensorShape broadcast_shape(const Shapes &... shapes)
If shapes are broadcast compatible, return the broadcasted shape.
Definition: TensorShape.h:211
#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
Store the tensor&#39;s metadata.
Definition: ITensorInfo.h:40
#define ARM_COMPUTE_ERROR_THROW_ON(status)
Definition: Error.h:455
Status class.
Definition: Error.h:52
std::string lower_string(const std::string &val)
Lower a given string.
Definition: Utils.cpp:353
bool is_data_type_quantized_symmetric(DataType dt)
Check if a given data type is of symmetric quantized type.
Definition: Utils.h:1088
#define ARM_COMPUTE_RETURN_ERROR_ON(cond)
If the condition is true, an error is returned.
Definition: Error.h:296
static std::pair< TensorShape, ValidRegion > broadcast_shape_and_valid_region(const Infos &... infos)
If infos are broadcast compatible tensor info&#39;s, return the broadcasted shape and the intersection of...
Definition: ITensorInfo.h:317
Activation Layer Information class.
Definition: Types.h:1639
std::set< std::string > build_options
void add_3D_tensor_argument(unsigned int &idx, const ICLTensor *tensor, const Window &window)
Add the passed 3D tensor&#39;s parameters to the object&#39;s kernel&#39;s arguments starting from the index idx...
Definition: ICLKernel.h:226
void configure(const ClCompileContext &compile_context, ArithmeticOperation op, ITensorInfo *input1, ITensorInfo *input2, ITensorInfo *output, const ConvertPolicy &policy, const ActivationLayerInfo &act_info=ActivationLayerInfo())
Static function to check if given info will lead to a valid configuration of ClSaturatedArithmeticKer...
Copyright (c) 2017-2022 Arm Limited.
1 channel, 1 F16 per channel
#define ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(...)
Definition: Validate.h:159
1 channel, 1 S32 per channel
void add_option(std::string option)
Adds option to the existing build option list.
const std::string & config_id() const
Get the configuration ID.
Definition: ICLKernel.h:419
const ITensor * get_const_tensor(int id) const
Get constant tensor of a given id.
Definition: ITensorPack.cpp:54
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:404
const std::string & string_from_data_type(DataType dt)
Convert a data type identity into a string.
Definition: Utils.cpp:135
#define ARM_COMPUTE_UNUSED(...)
To avoid unused variables warnings.
Definition: Error.h:152
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
std::string float_to_string_with_full_precision(float val)
Create a string with the float in full precision.
Definition: Utils.h:1124
void run_op(ITensorPack &tensors, const Window &window, ::cl::CommandQueue &queue) override
quantized, asymmetric fixed-point 8-bit number unsigned
static Status validate(ArithmeticOperation op, const ITensorInfo *input1, const ITensorInfo *input2, const ITensorInfo *output, const ConvertPolicy &policy, const ActivationLayerInfo &act_info=ActivationLayerInfo())
Static function to check if given info will lead to a valid configuration.
size_t total_size() const
Collapses all dimensions to a single linear total size.
Definition: TensorShape.h:172
unsigned int num_elems_processed_per_iteration
std::string get_cl_type_from_data_type(const DataType &dt)
Translates a tensor data type to the appropriate OpenCL type.
Definition: CLHelpers.cpp:39
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...
virtual std::unique_ptr< T > clone() const =0
Provide a clone of the current object of class T.
bool set_shape_if_empty(ITensorInfo &info, const TensorShape &shape)
Set the shape to the specified value if the current assignment is empty.
bool have_different_dimensions(const Dimensions< T > &dim1, const Dimensions< T > &dim2, unsigned int upper_dim)
Definition: Validate.h:47
Window broadcast_if_dimension_le_one(const TensorShape &shape) const
Don&#39;t advance in the dimension where shape is less equal to 1.
Definition: Window.inl:120
Elementwise CL kernel type.
Definition: CLTypes.h:85
bool slide_window_slice_3D(Window &slice) const
Slide the passed 3D window slice.
Definition: Window.h:349
y*x if x < 0, x otherwise
#define ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL(k)
Definition: Validate.h:915
1 channel, 1 S16 per channel
bool has_padding_changed(const std::unordered_map< const ITensorInfo *, PaddingSize > &padding_map)
Check if the previously stored padding info has changed after configuring a kernel.
Definition: Utils.cpp:603
CLCompileContext class.
ITensor * get_tensor(int id)
Get tensor of a given id from the pac.
Definition: ITensorPack.cpp:64
const std::string & string_from_data_layout(DataLayout dl)
Convert a data layout identity into a string.
Definition: Utils.cpp:123
void configure(const ClCompileContext &compile_context, LogicalOperation op, ITensorInfo *src1, ITensorInfo *src2, ITensorInfo *dst)
Function to configure kernel.
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:143
#define ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(...)
Definition: Validate.h:541
#define ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(t, c,...)
Definition: Validate.h:788
std::unordered_map< const ITensorInfo *, PaddingSize > get_padding_info(std::initializer_list< const ITensorInfo *> infos)
Stores padding information before configuring a kernel.
Definition: Utils.cpp:588
#define ARM_COMPUTE_RETURN_ERROR_ON_MSG(cond, msg)
If the condition is true, an error is returned.
Definition: Error.h:244
Tensor packing service.
Definition: ITensorPack.h:39
#define ARM_COMPUTE_ERROR_ON_NULLPTR(...)
Definition: Validate.h:157
unsigned int adjust_vec_size(unsigned int vec_size, size_t dim0)
Returns the adjusted vector size in case it is less than the input&#39;s first dimension, getting rounded down to its closest valid vector size.
Definition: Utils.h:1222
ActivationFunction activation() const
Get the type of activation function.
Definition: Types.h:1679
float b() const
Get the beta value.
Definition: Types.h:1689
quantized, asymmetric fixed-point 8-bit number signed
void configure(const ClCompileContext &compile_context, ArithmeticOperation op, ITensorInfo *src1, ITensorInfo *src2, ITensorInfo *dst, const ActivationLayerInfo &act_info=ActivationLayerInfo())
Static function to check if given info will lead to a valid configuration of ClArithmeticKernel.
Window first_slice_window_3D() const
First 3D slice of the window.
Definition: Window.h:305
std::string kernel_name
LogicalOperation
List of supported logical operations.
Definition: KernelTypes.h:30
Describe a multidimensional execution window.
Definition: Window.h:39
ConvertPolicy
Policy to handle integer overflow.
Definition: Types.h:404
bool is_data_type_float(DataType dt)
Check if a given data type is of floating point type.
Definition: Utils.h:1010
#define ARM_COMPUTE_ERROR_ON_INVALID_SUBWINDOW(f, s)
Definition: Validate.h:201
SimpleTensor< T > slice(const SimpleTensor< T > &src, Coordinates starts, Coordinates ends)
virtual DataLayout data_layout() const =0
Get the data layout of the tensor.