Compute Library
 21.02
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 
28 #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_arguments_with_float_only_supported_rules(const ITensorInfo &src1, const ITensorInfo &src2, const ITensorInfo &dst)
79 {
80  ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(&src1, &src2, &dst);
84 
85  const TensorShape out_shape = TensorShape::broadcast_shape(src1.tensor_shape(), src2.tensor_shape());
86 
87  ARM_COMPUTE_RETURN_ERROR_ON_MSG(out_shape.total_size() == 0, "Inputs are not broadcast compatible");
88 
89  // Validate in case of configured dst
90  if(dst.total_size() > 0)
91  {
95  "Wrong shape for dst");
96  }
97 
98  return Status{};
99 }
100 
101 Status validate_arguments_with_arithmetic_rules(const ITensorInfo &src1, const ITensorInfo &src2, const ITensorInfo &dst)
102 {
111 
112  const bool is_quantized = is_data_type_quantized(src1.data_type()) || is_data_type_quantized(src2.data_type());
113  if(is_quantized)
114  {
116 
117  if(is_data_type_quantized_symmetric(src1.data_type()))
118  {
119  const int32_t in1_offset = src1.quantization_info().uniform().offset;
120  const int32_t in2_offset = src2.quantization_info().uniform().offset;
121  ARM_COMPUTE_RETURN_ERROR_ON_MSG(in1_offset != 0, "For quantized symmetric, offset must be zero");
122  ARM_COMPUTE_RETURN_ERROR_ON_MSG(in2_offset != 0, "For quantized symmetric, offset must be zero");
123  }
124  }
125 
126  const TensorShape out_shape = TensorShape::broadcast_shape(src1.tensor_shape(), src2.tensor_shape());
127 
128  ARM_COMPUTE_RETURN_ERROR_ON_MSG(out_shape.total_size() == 0, "Inputs are not broadcast compatible");
129 
130  // Validate in case of configured dst
131  if(dst.total_size() > 0)
132  {
137  ARM_COMPUTE_RETURN_ERROR_ON_MSG((dst.data_type() == DataType::U8) && ((src1.data_type() != DataType::U8) || (src2.data_type() != DataType::U8)),
138  "dst can only be U8 if both inputs are U8");
139  ARM_COMPUTE_RETURN_ERROR_ON_MSG(detail::have_different_dimensions(out_shape, dst.tensor_shape(), 0),
140  "Wrong shape for dst");
141 
142  if(is_quantized)
143  {
145 
146  if(is_data_type_quantized_symmetric(dst.data_type()))
147  {
148  const int32_t offset = dst.quantization_info().uniform().offset;
149  ARM_COMPUTE_RETURN_ERROR_ON_MSG(offset != 0, "For quantized symmetric, offset must be zero");
150  }
151  }
152  }
153  return Status{};
154 }
155 
156 CLBuildOptions generate_build_options_with_arithmetic_rules(const ITensorInfo &src1, const ITensorInfo &src2, const ITensorInfo &dst, const std::string &operation_string)
157 {
158  CLBuildOptions build_opts;
159 
160  const unsigned int num_elems_processed_per_iteration = adjust_vec_size(vector_size_byte_opencl / dst.element_size(), dst.dimension(0));
161 
162  build_opts.add_option("-DDATA_TYPE_IN1=" + get_cl_type_from_data_type(src1.data_type()));
163  build_opts.add_option("-DDATA_TYPE_IN2=" + get_cl_type_from_data_type(src2.data_type()));
164  build_opts.add_option("-DDATA_TYPE_OUT=" + get_cl_type_from_data_type(dst.data_type()));
165  build_opts.add_option("-DVEC_SIZE_IN1=" + support::cpp11::to_string(src1.dimension(0) == 1 ? 1 : num_elems_processed_per_iteration));
166  build_opts.add_option("-DVEC_SIZE_IN2=" + support::cpp11::to_string(src2.dimension(0) == 1 ? 1 : num_elems_processed_per_iteration));
167  build_opts.add_option("-DVEC_SIZE_OUT=" + support::cpp11::to_string(num_elems_processed_per_iteration));
168  build_opts.add_option("-DVEC_SIZE_LEFTOVER=" + support::cpp11::to_string(dst.dimension(0) % num_elems_processed_per_iteration));
169  build_opts.add_option("-DOP=" + operation_string);
170  if(is_data_type_quantized(src1.data_type()))
171  {
172  const UniformQuantizationInfo iq1info = src1.quantization_info().uniform();
173  const UniformQuantizationInfo iq2info = src2.quantization_info().uniform();
174  const UniformQuantizationInfo oqinfo = dst.quantization_info().uniform();
175 
176  build_opts.add_option("-DOFFSET_IN1=" + support::cpp11::to_string(iq1info.offset));
177  build_opts.add_option("-DOFFSET_IN2=" + support::cpp11::to_string(iq2info.offset));
178  build_opts.add_option("-DOFFSET_OUT=" + support::cpp11::to_string(oqinfo.offset));
179  build_opts.add_option("-DSCALE_IN1=" + float_to_string_with_full_precision(iq1info.scale));
180  build_opts.add_option("-DSCALE_IN2=" + float_to_string_with_full_precision(iq2info.scale));
181  build_opts.add_option("-DSCALE_OUT=" + float_to_string_with_full_precision(oqinfo.scale));
182  }
183  return build_opts;
184 }
185 
186 std::pair<Status, Window> configure_window_arithmetic_common(ITensorInfo &dst)
187 {
188  const unsigned int num_elems_processed_per_iteration = adjust_vec_size(vector_size_byte_opencl / dst.element_size(), dst.dimension(0));
189  Window win = calculate_max_window(dst, Steps(num_elems_processed_per_iteration));
190  return std::make_pair(Status{}, win);
191 }
192 
193 std::pair<Status, Window> validate_and_configure_window_for_arithmetic_operators(ITensorInfo &src1, ITensorInfo &src2, ITensorInfo &dst)
194 {
195  const std::pair<TensorShape, ValidRegion> broadcast_pair = ITensorInfo::broadcast_shape_and_valid_region(src1, src2);
196  const TensorShape &out_shape = broadcast_pair.first;
197 
198  set_shape_if_empty(dst, out_shape);
199 
200  if(src1.data_type() == DataType::S16 || src2.data_type() == DataType::S16)
201  {
203  }
204  else if(src1.data_type() == DataType::F16 || src2.data_type() == DataType::F16)
205  {
207  }
208  else if(src1.data_type() == DataType::F32 || src2.data_type() == DataType::F32)
209  {
211  }
212  else if(src1.data_type() == DataType::QASYMM8 || src2.data_type() == DataType::QASYMM8)
213  {
215  }
216  else if(src1.data_type() == DataType::QASYMM8_SIGNED || src2.data_type() == DataType::QASYMM8_SIGNED)
217  {
219  }
220  else if(src1.data_type() == DataType::QSYMM16 || src2.data_type() == DataType::QSYMM16)
221  {
223  }
224 
225  return configure_window_arithmetic_common(dst);
226 }
227 
228 std::pair<Status, Window> validate_and_configure_window_for_logical_binary_operators(ITensorInfo &src1, ITensorInfo &src2, ITensorInfo &dst)
229 {
230  const std::pair<TensorShape, ValidRegion> broadcast_pair = ITensorInfo::broadcast_shape_and_valid_region(src1, src2);
231  const TensorShape &out_shape = broadcast_pair.first;
232 
233  set_shape_if_empty(dst, out_shape);
235 
236  // The arithmetic utility functions can be share
237  return configure_window_arithmetic_common(dst);
238 }
239 
240 std::pair<Status, Window> validate_and_configure_window_for_division(ITensorInfo &src1, ITensorInfo &src2, ITensorInfo &dst)
241 {
242  const std::pair<TensorShape, ValidRegion> broadcast_pair = ITensorInfo::broadcast_shape_and_valid_region(src1, src2);
243  const TensorShape &out_shape = broadcast_pair.first;
244  auto_init_if_empty(dst, out_shape, 1, src1.data_type());
245  return configure_window_arithmetic_common(dst);
246 }
247 } // namespace
248 
249 void ClElementwiseKernel::configure_common(ITensorInfo *src1, ITensorInfo *src2, ITensorInfo *dst)
250 {
251  configure_common(CLKernelLibrary::get().get_compile_context(), src1, src2, dst);
252 }
253 
254 void ClElementwiseKernel::configure_common(const ClCompileContext &compile_context, ITensorInfo *src1, ITensorInfo *src2, ITensorInfo *dst)
255 {
256  // Configure kernel window
257  auto win_config = validate_and_configure_window(*src1, *src2, *dst);
258  ARM_COMPUTE_ERROR_THROW_ON(win_config.first);
259 
260  _src1 = src1;
261  _src2 = src2;
262  _dst = dst;
263 
264  std::string kernel_name = "elementwise_operation_" + name();
265  if(is_data_type_quantized(src1->data_type()))
266  {
267  kernel_name += "_quantized";
268  }
269 
270  // Set kernel build options
271  CLBuildOptions build_opts = generate_build_options(*src1, *src2, *dst);
272  if(_act_info.enabled())
273  {
274  build_opts.add_option("-DACTIVATION_TYPE=" + lower_string(string_from_activation_func(_act_info.activation())));
275  build_opts.add_option("-DA_VAL=" + float_to_string_with_full_precision(_act_info.a()));
276  build_opts.add_option("-DB_VAL=" + float_to_string_with_full_precision(_act_info.b()));
277  }
278 
279  // Create kernel
280  _kernel = create_kernel(compile_context, kernel_name, build_opts.options());
281 
282  ICLKernel::configure_internal(win_config.second);
283 
284  _config_id = generate_id_for_tuning(kernel_name, *src1, *dst);
285 }
286 
287 void ClElementwiseKernel::run_op(ITensorPack &tensors, const Window &window, ::cl::CommandQueue &queue)
288 {
291 
292  const auto src_0 = utils::cast::polymorphic_downcast<const ICLTensor *>(tensors.get_const_tensor(TensorType::ACL_SRC_0));
293  const auto src_1 = utils::cast::polymorphic_downcast<const ICLTensor *>(tensors.get_const_tensor(TensorType::ACL_SRC_1));
294  auto dst = utils::cast::polymorphic_downcast<ICLTensor *>(tensors.get_tensor(TensorType::ACL_DST));
295 
296  const TensorShape &in_shape1 = src_0->info()->tensor_shape();
297  const TensorShape &in_shape2 = src_1->info()->tensor_shape();
298  const TensorShape &out_shape = dst->info()->tensor_shape();
299 
300  bool can_collapse = true;
301  const bool is_vector = in_shape1.num_dimensions() == 1 || in_shape2.num_dimensions() == 1;
302  if(std::min(in_shape1.total_size(), in_shape2.total_size()) > 1 && !is_vector)
303  {
304  can_collapse = (std::min(in_shape1.num_dimensions(), in_shape2.num_dimensions()) > Window::DimZ);
305  for(size_t d = Window::DimZ; can_collapse && (d < out_shape.num_dimensions()); d++)
306  {
307  can_collapse = (in_shape1[d] == in_shape2[d]);
308  }
309  }
310 
311  bool has_collapsed = false;
312  Window collapsed = can_collapse ? window.collapse_if_possible(ICLKernel::window(), Window::DimZ, &has_collapsed) : window;
313 
314  const TensorShape &in_shape1_collapsed = has_collapsed ? in_shape1.collapsed_from(Window::DimZ) : in_shape1;
315  const TensorShape &in_shape2_collapsed = has_collapsed ? in_shape2.collapsed_from(Window::DimZ) : in_shape2;
316 
317  Window slice = collapsed.first_slice_window_3D();
318  Window slice_src1 = slice.broadcast_if_dimension_le_one(in_shape1_collapsed);
319  Window slice_src2 = slice.broadcast_if_dimension_le_one(in_shape2_collapsed);
320  do
321  {
322  unsigned int idx = 0;
323  add_3D_tensor_argument(idx, src_0, slice_src1);
324  add_3D_tensor_argument(idx, src_1, slice_src2);
325  add_3D_tensor_argument(idx, dst, slice);
326 
327  enqueue(queue, *this, slice, lws_hint());
328  ARM_COMPUTE_UNUSED(collapsed.slide_window_slice_3D(slice_src1));
329  ARM_COMPUTE_UNUSED(collapsed.slide_window_slice_3D(slice_src2));
330  }
331  while(collapsed.slide_window_slice_3D(slice));
332 }
333 
334 /** Logical binary */
335 
337 {
338  ARM_COMPUTE_ERROR_ON_NULLPTR(src1, src2, dst);
340  _op = op;
341  configure_common(compile_context, src1, src2, dst);
342 }
343 
345 {
346  ARM_COMPUTE_UNUSED(op);
348  ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(src1, src2, dst);
349 
352 
353  ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments_with_arithmetic_rules(*src1, *src2, *dst));
354  ARM_COMPUTE_RETURN_ON_ERROR(validate_and_configure_window_for_logical_binary_operators(*src1->clone(), *src2->clone(), *dst->clone()).first);
355 
356  return Status{};
357 }
358 
359 std::string ClLogicalBinaryKernel::name()
360 {
361  switch(_op)
362  {
364  return "AND";
366  return "OR";
368  /* fall through */
369  default:
370  ARM_COMPUTE_ASSERT(true);
371  }
372  return "";
373 }
374 
375 std::pair<Status, Window> ClLogicalBinaryKernel::validate_and_configure_window(ITensorInfo &src1, ITensorInfo &src2, ITensorInfo &dst)
376 {
377  return validate_and_configure_window_for_logical_binary_operators(src1, src2, dst);
378 }
379 
380 CLBuildOptions ClLogicalBinaryKernel::generate_build_options(const ITensorInfo &src1, const ITensorInfo &src2, const ITensorInfo &dst)
381 {
382  // The arithmetic utility functions can be share
383  return generate_build_options_with_arithmetic_rules(src1, src2, dst, name());
384 }
385 
386 std::string ClLogicalBinaryKernel::generate_id_for_tuning(const std::string &kernel_name, const ITensorInfo &src1, const ITensorInfo &dst)
387 {
388  return generate_id_for_tuning_common(kernel_name, src1, dst);
389 }
390 
391 /** Arithmetic operations with saturation*/
393  const ConvertPolicy &policy,
394  const ActivationLayerInfo &act_info)
395 {
396  ARM_COMPUTE_ERROR_ON_NULLPTR(input1, input2, output);
397  ARM_COMPUTE_ERROR_THROW_ON(ClSaturatedArithmeticKernel::validate(op, input1, input2, output, policy, act_info));
398  auto padding_info = get_padding_info({ input1, input2, output });
399 
400  _policy = policy;
401  _op = op;
402  _act_info = act_info;
403  configure_common(compile_context, input1, input2, output);
405 }
406 
408  const ActivationLayerInfo &act_info)
409 {
410  ARM_COMPUTE_UNUSED(op, policy);
411  ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(input1, input2, output);
412  ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments_with_arithmetic_rules(*input1, *input2, *output));
413  ARM_COMPUTE_RETURN_ON_ERROR(validate_and_configure_window_for_arithmetic_operators(*input1->clone(), *input2->clone(), *output->clone()).first);
415 
416  return Status{};
417 }
418 
419 std::pair<Status, Window> ClSaturatedArithmeticKernel::validate_and_configure_window(ITensorInfo &input1, ITensorInfo &input2, ITensorInfo &output)
420 {
421  return validate_and_configure_window_for_arithmetic_operators(input1, input2, output);
422 }
423 
424 CLBuildOptions ClSaturatedArithmeticKernel::generate_build_options(const ITensorInfo &input1, const ITensorInfo &input2, const ITensorInfo &output)
425 {
426  const bool has_float_out = is_data_type_float(output.data_type());
427  auto build_options = generate_build_options_with_arithmetic_rules(input1, input2, output, name());
428  build_options.add_option((_policy == ConvertPolicy::WRAP || has_float_out) ? "-DWRAP" : "-DSATURATE");
429  return build_options;
430 }
431 
432 std::string ClSaturatedArithmeticKernel::generate_id_for_tuning(const std::string &kernel_name, const ITensorInfo &input1, const ITensorInfo &output)
433 {
434  auto config_id = generate_id_for_tuning_common(kernel_name, input1, output);
435  config_id += (_policy == ConvertPolicy::WRAP) ? "_wrap_" : "_saturate_";
436  config_id += lower_string(string_from_data_layout(input1.data_layout()));
437  return config_id;
438 }
439 
440 std::string ClSaturatedArithmeticKernel::name()
441 {
442  return supported_sat_arithmetic_ops[_op];
443 }
444 
445 /** Arithmetic operations*/
447  const ActivationLayerInfo &act_info)
448 {
449  ARM_COMPUTE_ERROR_ON_NULLPTR(src1, src2, dst);
450  ARM_COMPUTE_ERROR_THROW_ON(ClArithmeticKernel::validate(op, src1, src2, dst, act_info));
451  auto padding_info = get_padding_info({ src1, src2, dst });
452 
453  _op = op;
454  _act_info = act_info;
455  configure_common(compile_context, src1, src2, dst);
457 }
458 
460 {
461  ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(src1, src2, dst);
463  {
464  // Division and Power operators don't support integer arithmetic
465  ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments_with_float_only_supported_rules(*src1, *src2, *dst));
466  ARM_COMPUTE_RETURN_ON_ERROR(validate_and_configure_window_for_division(*src1->clone(), *src2->clone(), *dst->clone()).first);
467  }
468  else
469  {
470  ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments_with_arithmetic_rules(*src1, *src2, *dst));
471  ARM_COMPUTE_RETURN_ON_ERROR(validate_and_configure_window_for_arithmetic_operators(*src1->clone(), *src2->clone(), *dst->clone()).first);
472  }
474 
475  return Status{};
476 }
477 std::pair<Status, Window> ClArithmeticKernel::validate_and_configure_window(ITensorInfo &src1, ITensorInfo &src2, ITensorInfo &dst)
478 {
480  {
481  // Division and Power operators don't support integer arithmetic
482  return validate_and_configure_window_for_division(src1, src2, dst);
483  }
484  else
485  {
486  return validate_and_configure_window_for_arithmetic_operators(src1, src2, dst);
487  }
488 }
489 
490 CLBuildOptions ClArithmeticKernel::generate_build_options(const ITensorInfo &src1, const ITensorInfo &src2, const ITensorInfo &dst)
491 {
492  return generate_build_options_with_arithmetic_rules(src1, src2, dst, name());
493 }
494 std::string ClArithmeticKernel::generate_id_for_tuning(const std::string &kernel_name, const ITensorInfo &src1, const ITensorInfo &dst)
495 {
496  return generate_id_for_tuning_common(kernel_name, src1, dst);
497 }
498 
499 std::string ClArithmeticKernel::name()
500 {
501  return supported_arithmetic_ops[_op];
502 }
503 } // namespace kernels
504 } // namespace opencl
505 } // namespace arm_compute
bool set_format_if_unknown(ITensorInfo &info, Format format)
Set the format, data type and number of channels to the specified value if the current data type is u...
static Status validate(LogicalOperation op, const ITensorInfo *src1, const ITensorInfo *src2, const ITensorInfo *dst)
Static function to check if the given configuration is valid for this kernel.
bool is_data_type_quantized(DataType dt)
Check if a given data type is of quantized type.
Definition: Utils.h:1168
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 of ClArithmeticKernel.
__global uchar * offset(const Image *img, int x, int y)
Get the pointer position of a Image.
Definition: helpers.h:846
Window calculate_max_window(const ValidRegion &valid_region, const Steps &steps, bool skip_border, BorderSize border_size)
ArithmeticOperation
Available element-wise operations.
Definition: Types.h:534
#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...
void enqueue(IGCKernel &kernel, const Window &window, const gles::NDRange &lws=gles::NDRange(1U, 1U, 1U))
Add the kernel to the command queue with the given window.
Definition: IGCKernel.cpp:41
bool enabled() const
Check if initialised.
Definition: Types.h:1600
cl::NDRange lws_hint() const
Return the Local-Workgroup-Size hint.
Definition: ICLKernel.h:276
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:1590
#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
static CLKernelLibrary & get()
Access the KernelLibrary singleton.
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:350
bool is_data_type_quantized_symmetric(DataType dt)
Check if a given data type is of symmetric quantized type.
Definition: Utils.h:1226
#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:271
Activation Layer Information class.
Definition: Types.h:1550
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:172
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) 2017-2021 Arm Limited.
1 channel, 1 F16 per channel
#define ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(...)
Definition: Validate.h:163
1 channel, 1 S32 per channel
const std::string & config_id() const
Get the configuration ID.
Definition: ICLKernel.h:312
const ITensor * get_const_tensor(int id) const
Get constant tensor of a given id.
Definition: ITensorPack.cpp:40
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
#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:1262
void run_op(ITensorPack &tensors, const Window &window, ::cl::CommandQueue &queue) override
quantized, asymmetric fixed-point 8-bit number unsigned
arm_compute::CLCompileContext ClCompileContext
std::set< std::string > build_options
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 of ClSaturatedArithmeticKer...
std::string kernel_name
size_t total_size() const
Collapses all dimensions to a single linear total size.
Definition: TensorShape.h:172
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
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:51
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
bool slide_window_slice_3D(Window &slice) const
Slide the passed 3D window slice.
Definition: Window.h:335
y*x if x < 0, x otherwise
#define ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL(k)
Definition: Validate.h:941
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:528
CLCompileContext class.
ITensor * get_tensor(int id)
Get tensor of a given id from the pac.
Definition: ITensorPack.cpp:50
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:545
#define ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(t, c,...)
Definition: Validate.h:792
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:513
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
Tensor packing service.
Definition: ITensorPack.h:37
#define ARM_COMPUTE_ERROR_ON_NULLPTR(...)
Definition: Validate.h:161
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:1358
#define ARM_COMPUTE_ASSERT(cond)
Definition: Validate.h:37
ActivationFunction activation() const
Get the type of activation function.
Definition: Types.h:1585
float b() const
Get the beta value.
Definition: Types.h:1595
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:291
LogicalOperation
List of supported logical operations.
Definition: KernelTypes.h:30
Describe a multidimensional execution window.
Definition: Window.h:39
ConvertPolicy
Policy to handle overflow.
Definition: Types.h:385
bool is_data_type_float(DataType dt)
Check if a given data type is of floating point type.
Definition: Utils.h:1148
#define ARM_COMPUTE_ERROR_ON_INVALID_SUBWINDOW(f, s)
Definition: Validate.h:205
SimpleTensor< T > slice(const SimpleTensor< T > &src, Coordinates starts, Coordinates ends)
virtual DataLayout data_layout() const =0
Get the data layout of the tensor.