Compute Library
 21.02
ITensorInfo.h
Go to the documentation of this file.
1 /*
2  * Copyright (c) 2016-2021 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  */
24 #ifndef ARM_COMPUTE_ITENSORINFO_H
25 #define ARM_COMPUTE_ITENSORINFO_H
26 
30 #include "arm_compute/core/Types.h"
31 #include "arm_compute/core/Utils.h"
33 #include "support/ICloneable.h"
34 
35 #include <cstddef>
36 
37 namespace arm_compute
38 {
39 /** Store the tensor's metadata */
40 class ITensorInfo : public misc::ICloneable<ITensorInfo>
41 {
42 public:
44 
45 public:
46  /** Default virtual destructor */
47  virtual ~ITensorInfo() = default;
48  /** Set the data type to the specified value.
49  *
50  * @warning This resets the format to UNKNOWN.
51  *
52  * @param[in] data_type The new data type.
53  *
54  * @return Reference to this ITensorInfo object
55  */
57  /** Set the number of channels to the specified value.
58  *
59  * @warning This resets the format to UNKNOWN.
60  *
61  * @param[in] num_channels New number of channels.
62  *
63  * @return Reference to this ITensorInfo object
64  */
65  virtual ITensorInfo &set_num_channels(int num_channels) = 0;
66  /** Set the format of an already initialized tensor.
67  *
68  * @note If the data type has already been configured (i.e. not UNKNOWN) it
69  * must match the new format. If data type hasn't been configured it will
70  * be based on the format.
71  *
72  * @param[in] format Single-plane format of the tensor.
73  *
74  * @return Reference to this ITensorInfo object
75  */
76  virtual ITensorInfo &set_format(Format format) = 0;
77  /** Set the shape of an already initialized tensor.
78  *
79  * @warning Changing the shape requires to recompute the strides and is
80  * therefore only possible if the tensor hasn't been allocated yet.
81  *
82  * @param[in] shape New tensor shape.
83  *
84  * @return Reference to this ITensorInfo object
85  */
86  virtual ITensorInfo &set_tensor_shape(const TensorShape &shape) = 0;
87  /** Set the state for each dimension of the tensor
88  *
89  * This sets the state of each dimension of the shape in terms of dynamic behavior using -1 where appropriate.
90  * The index in the state is a 1 to 1 mapping with the shape dimension index.
91  * For example if you want to express [?, 3, 3] as a dynamic input then [-1, 3, 3] has to be set as a state
92  *
93  * @param[in] state Tensor dimensions state
94  *
95  * @return Reference to this ITensorInfo object
96  */
97  virtual ITensorInfo &set_tensor_dims_state(const TensorDimsState &state) = 0;
98  /** Set the quantization settings (scale and offset) of the tensor.
99  *
100  * @param[in] quantization_info QuantizationInfo containing the scale and offset
101  *
102  * @return Reference to this ITensorInfo object
103  */
105  /** Set the data layout of the tensor.
106  *
107  * @param[in] data_layout DataLayout containing the layout data information.
108  *
109  * @return Reference to this ITensorInfo object
110  */
111  virtual ITensorInfo &set_data_layout(const DataLayout &data_layout) = 0;
112  /** Resets the padding settings of the tensor.
113  *
114  * @return Reference to this ITensorInfo object
115  */
116  virtual ITensorInfo &reset_padding() = 0;
117  /** Update the offset to the first element and the strides to automatically computed values.
118  *
119  * @note The padding used by this method is really conservative so that the tensor can be used for most functions.
120  *
121  * @return True if the strides or the offset to the first element have changed.
122  */
123  virtual bool auto_padding() = 0;
124  /** Update the offset to the first element, the strides and the total size.
125  *
126  * @note This function can only increase the offset, strides and total size.
127  *
128  * @param[in] padding Padding around the XY plane in number of elements.
129  *
130  * @return True if the strides, offset and total size have changed.
131  */
132  virtual bool extend_padding(const PaddingSize &padding) = 0;
133  /** Return the size of the requested dimension
134  *
135  * @param[in] index Index of the dimension
136  *
137  * @return Dimension of the requested dimension
138  */
139  virtual size_t dimension(size_t index) const = 0;
140  /** Return the size of the requested data layout dimension
141  *
142  * @param[in] dimension DataLayoutDimension of the dimension
143  *
144  * @return Dimension of the requested dimension
145  */
146  virtual size_t dimension(DataLayoutDimension dimension) const = 0;
147  /** The strides in bytes for accessing each dimension of the tensor
148  *
149  * @return Strides in bytes for each tensor dimension
150  */
151  virtual const Strides &strides_in_bytes() const = 0;
152  /** The offset from the beginning of the memory allocation to the first element of the tensor.
153  * This can be used to access efficiently elements in a 2D tensor
154  *
155  * @return The offset in bytes to access the first element of the tensor.
156  */
157  virtual size_t offset_first_element_in_bytes() const = 0;
158  /** The offset in bytes from the beginning of the memory allocation to access the element at position (x, y, z ...)
159  *
160  * @param[in] pos Vector with the coordinates of the element to access.
161  * The size of this vector must be equal to the number of dimensions of the tensor
162  *
163  * @return Offset in bytes from the beginning of the memory allocation to access the element (x, y, z, ...)
164  */
165  virtual int32_t offset_element_in_bytes(const Coordinates &pos) const = 0;
166 
167  /** Element size in bytes calculated as data_size() * num_channels()
168  *
169  * @return The size of one element in bytes
170  */
171  virtual size_t element_size() const = 0;
172  /** The number of dimensions of the tensor (rank)
173  *
174  * @return The number of dimensions of the tensor (rank)
175  */
176  virtual size_t num_dimensions() const = 0;
177  /** The number of channels for each tensor element
178  *
179  * @return The number of channels for each tensor element
180  */
181  virtual size_t num_channels() const = 0;
182  /** Size for each dimension of the tensor
183  *
184  * @return A vector with the size for each dimension of the tensor
185  */
186  virtual const TensorShape &tensor_shape() const = 0;
187  /** State of each dimension of the tensor shape
188  *
189  * @return A vector with the state for each dimension of the tensor, where -1 specifies dynamic dimension
190  */
191  virtual const TensorDimsState &tensor_dims_state() const = 0;
192  /** Data type used for each element of the tensor
193  *
194  * @return Tensor data type
195  */
196  virtual DataType data_type() const = 0;
197  /** Colour format of the image
198  *
199  * @return Colour format of the image
200  */
201  virtual Format format() const = 0;
202  /** Returns the total size of the tensor in bytes.
203  *
204  * @return Total size of the tensor in bytes.
205  */
206  virtual size_t total_size() const = 0;
207  /** Padding of tensor.
208  *
209  * @return Padding.
210  */
211  virtual PaddingSize padding() const = 0;
212  /** Checks if the tensor has been allocated with padding or not.
213  *
214  * @return True if padding is allocated in the tensor, otherwise false.
215  */
216  virtual bool has_padding() const = 0;
217  /** Flag indicating whether the size of the tensor can be changed.
218  *
219  * @return True if the tensor size can be changed.
220  */
221  virtual bool is_resizable() const = 0;
222  /** Flag indicating whether the shape of the tensor is dynamic, meaning that it can change on kernel/function execution.
223  *
224  * @return True if its dynamic else false
225  */
226  virtual bool is_dynamic() const = 0;
227  /** Set the flag whether the tensor size can be changed.
228  *
229  * @param[in] is_resizable Flag that marks the tensor if it can be changed or not.
230  *
231  * @return Reference to this ITensorInfo object
232  */
233  virtual ITensorInfo &set_is_resizable(bool is_resizable) = 0;
234  /** Valid region of the tensor. All elements in the valid region have defined values, i.e. are not undefined.
235  *
236  * @return The valid region.
237  */
238  virtual ValidRegion valid_region() const = 0;
239  /** Set the valid region of the tensor.
240  *
241  * @param[in] valid_region Valid region to set.
242  */
243  virtual void set_valid_region(const ValidRegion &valid_region) = 0;
244 
245  /** Get the quantization settings (scale and offset) of the tensor.
246  *
247  * @return A QuantizationInfo containing the scale and offset.
248  */
249  virtual QuantizationInfo quantization_info() const = 0;
250  /** Get the data layout of the tensor.
251  *
252  * @return A DataLayout containing the layout data information.
253  */
254  virtual DataLayout data_layout() const = 0;
255 
256  /** If infos are broadcast compatible tensor info's, return the broadcasted shape and the intersection of
257  * the broadcasted valid regions of the tensors.
258  *
259  * Two tensor info's are broadcast compatible if their shapes are broadcast compatible.
260  *
261  * Two tensor shapes are broadcast compatible if for each dimension, they're equal or one of them is 1.
262  *
263  * If two shapes are compatible, each dimension in the broadcasted shape is the max of the original dimensions.
264  *
265  * @param[in] infos Tensor info's.
266  *
267  * @return The broadcasted shape and valid region, or an empty shape and valid region if the info's are
268  * not broadcast compatible.
269  */
270  template <typename... Infos>
271  static std::pair<TensorShape, ValidRegion> broadcast_shape_and_valid_region(const Infos &... infos)
272  {
273  TensorShape bc_shape = TensorShape::broadcast_shape(infos.tensor_shape()...);
274  ValidRegion bc_valid_region{ Coordinates(), bc_shape };
275 
276  auto broadcast_valid_region = [&bc_valid_region](const ITensorInfo & info)
277  {
278  if(info.num_dimensions() != 0)
279  {
280  for(size_t d = 0; d < bc_valid_region.shape.num_dimensions(); ++d)
281  {
282  const bool is_broadcast = (info.tensor_shape()[d] == 1);
283 
284  const int anchor_max = std::max(bc_valid_region.anchor[d], info.valid_region().anchor[d]);
285  const size_t valid_min = std::min(bc_valid_region.shape[d], info.valid_region().shape[d]);
286 
287  if(!is_broadcast || (valid_min == 0))
288  {
289  bc_valid_region.anchor.set(d, anchor_max);
290  bc_valid_region.shape.set(d, valid_min);
291  }
292  }
293  }
294  };
295 
296  utility::for_each(broadcast_valid_region, infos...);
297 
298  return std::pair<TensorShape, ValidRegion>(bc_shape, bc_valid_region);
299  }
300 };
301 } // namespace arm_compute
302 #endif /*ARM_COMPUTE_TENSORINFO_H */
virtual size_t num_dimensions() const =0
The number of dimensions of the tensor (rank)
virtual ITensorInfo & set_num_channels(int num_channels)=0
Set the number of channels to the specified value.
Shape of a tensor.
Definition: TensorShape.h:39
virtual int32_t offset_element_in_bytes(const Coordinates &pos) const =0
The offset in bytes from the beginning of the memory allocation to access the element at position (x...
virtual size_t dimension(size_t index) const =0
Return the size of the requested dimension.
Container for 2D border size.
Definition: Types.h:273
virtual ITensorInfo & set_tensor_shape(const TensorShape &shape)=0
Set the shape of an already initialized tensor.
DataLayoutDimension
[DataLayout enum definition]
Definition: Types.h:129
virtual DataType data_type() const =0
Data type used for each element of the tensor.
static TensorShape broadcast_shape(const Shapes &... shapes)
If shapes are broadcast compatible, return the broadcasted shape.
Definition: TensorShape.h:211
Store the tensor&#39;s metadata.
Definition: ITensorInfo.h:40
virtual bool is_dynamic() const =0
Flag indicating whether the shape of the tensor is dynamic, meaning that it can change on kernel/func...
virtual ITensorInfo & reset_padding()=0
Resets the padding settings of the tensor.
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
Copyright (c) 2017-2021 Arm Limited.
virtual void set_valid_region(const ValidRegion &valid_region)=0
Set the valid region of the tensor.
virtual ValidRegion valid_region() const =0
Valid region of the tensor.
virtual bool is_resizable() const =0
Flag indicating whether the size of the tensor can be changed.
Quantization information.
virtual Format format() const =0
Colour format of the image.
virtual bool auto_padding()=0
Update the offset to the first element and the strides to automatically computed values.
virtual const TensorShape & tensor_shape() const =0
Size for each dimension of the tensor.
virtual ITensorInfo & set_data_layout(const DataLayout &data_layout)=0
Set the data layout of the tensor.
Format
Image colour formats.
Definition: Types.h:54
Coordinates of an item.
Definition: Coordinates.h:37
virtual size_t element_size() const =0
Element size in bytes calculated as data_size() * num_channels()
virtual PaddingSize padding() const =0
Padding of tensor.
virtual ITensorInfo & set_quantization_info(const QuantizationInfo &quantization_info)=0
Set the quantization settings (scale and offset) of the tensor.
virtual ITensorInfo & set_data_type(DataType data_type)=0
Set the data type to the specified value.
virtual QuantizationInfo quantization_info() const =0
Get the quantization settings (scale and offset) of the tensor.
void for_each(F &&)
Base case of for_each.
Definition: Utility.h:108
Strides of an item in bytes.
Definition: Strides.h:37
virtual size_t offset_first_element_in_bytes() const =0
The offset from the beginning of the memory allocation to the first element of the tensor...
ScaleKernelInfo info(interpolation_policy, default_border_mode, PixelValue(), sampling_policy, false)
virtual size_t total_size() const =0
Returns the total size of the tensor in bytes.
virtual const Strides & strides_in_bytes() const =0
The strides in bytes for accessing each dimension of the tensor.
Clonable Interface.
Definition: ICloneable.h:35
virtual const TensorDimsState & tensor_dims_state() const =0
State of each dimension of the tensor shape.
Container for valid region of a window.
Definition: Types.h:188
virtual ITensorInfo & set_format(Format format)=0
Set the format of an already initialized tensor.
DataType
Available data types.
Definition: Types.h:77
virtual ~ITensorInfo()=default
Default virtual destructor.
DataLayout
[DataLayout enum definition]
Definition: Types.h:120
virtual bool extend_padding(const PaddingSize &padding)=0
Update the offset to the first element, the strides and the total size.
virtual size_t num_channels() const =0
The number of channels for each tensor element.
TensorShape & set(size_t dimension, size_t value, bool apply_dim_correction=true, bool increase_dim_unit=true)
Accessor to set the value of one of the dimensions.
Definition: TensorShape.h:79
virtual bool has_padding() const =0
Checks if the tensor has been allocated with padding or not.
virtual ITensorInfo & set_is_resizable(bool is_resizable)=0
Set the flag whether the tensor size can be changed.
virtual DataLayout data_layout() const =0
Get the data layout of the tensor.
virtual ITensorInfo & set_tensor_dims_state(const TensorDimsState &state)=0
Set the state for each dimension of the tensor.