Compute Library
 22.05
ClConcatenate.cpp
Go to the documentation of this file.
1 /*
2  * Copyright (c) 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 
35 
36 #include "arm_compute/core/Error.h"
38 #include "arm_compute/core/Types.h"
39 
40 #include "src/common/utils/Log.h"
42 
43 namespace arm_compute
44 {
45 namespace opencl
46 {
47 void ClConcatenate::configure(const CLCompileContext &compile_context, const std::vector<ITensorInfo *> &src_vector, ITensorInfo *dst, size_t axis)
48 {
49  ARM_COMPUTE_ERROR_ON(dst == nullptr);
50  ARM_COMPUTE_LOG_PARAMS(src_vector, dst, axis);
51  _axis = axis;
52  _num_inputs = src_vector.size();
53 
55  std::vector<const ITensorInfo *> const_src_vector(src_vector.size());
56  std::transform(src_vector.begin(), src_vector.end(), const_src_vector.begin(), [](ITensorInfo * t)
57  {
59  return t;
60  });
61 
62  // dst auto inizialitation if not yet initialized
63  auto_init_if_empty(*dst, dst_shape, 1, src_vector[0]->data_type());
64  ARM_COMPUTE_ERROR_THROW_ON(ClConcatenate::validate(const_src_vector, dst, axis));
65 
66  unsigned int offset = 0;
67  switch(_axis)
68  {
69  case Window::DimX:
70  {
71  switch(_num_inputs)
72  {
73  case 2:
74  {
75  // Configure WidthConcatenate2Tensors kernel
76  auto kernel = std::make_unique<kernels::ClWidthConcatenate2TensorsKernel>();
77  kernel->configure(compile_context, src_vector.at(0), src_vector.at(1), dst);
78  _concat_kernels.emplace_back(std::move(kernel));
79  break;
80  }
81  case 4:
82  {
83  // Configure WidthConcatenate4Tensors kernel
84  auto kernel = std::make_unique<kernels::ClWidthConcatenate4TensorsKernel>();
85  kernel->configure(compile_context, src_vector.at(0), src_vector.at(1), src_vector.at(2), src_vector.at(3), dst);
86  _concat_kernels.emplace_back(std::move(kernel));
87  break;
88  }
89  default:
90  {
91  // Configure generic case WidthConcatenate kernels
92  for(unsigned int i = 0; i < _num_inputs; ++i)
93  {
94  auto kernel = std::make_unique<kernels::ClWidthConcatenateKernel>();
95  kernel->configure(compile_context, src_vector.at(i), offset, dst);
96  offset += src_vector.at(i)->dimension(_axis);
97  _concat_kernels.emplace_back(std::move(kernel));
98  }
99  break;
100  }
101  }
102  break;
103  }
104  case Window::DimY:
105  {
106  for(unsigned int i = 0; i < _num_inputs; ++i)
107  {
108  auto kernel = std::make_unique<kernels::ClHeightConcatenateKernel>();
109  kernel->configure(compile_context, src_vector.at(i), offset, dst);
110  offset += src_vector.at(i)->dimension(_axis);
111  _concat_kernels.emplace_back(std::move(kernel));
112  }
113  break;
114  }
115  case Window::DimZ:
116  {
117  for(unsigned int i = 0; i < _num_inputs; ++i)
118  {
119  auto kernel = std::make_unique<kernels::ClDepthConcatenateKernel>();
120  kernel->configure(compile_context, src_vector.at(i), offset, dst);
121  offset += src_vector.at(i)->dimension(_axis);
122  _concat_kernels.emplace_back(std::move(kernel));
123  }
124  break;
125  }
126  case 3:
127  {
128  for(unsigned int i = 0; i < _num_inputs; ++i)
129  {
130  auto kernel = std::make_unique<kernels::ClBatchConcatenateKernel>();
131  kernel->configure(compile_context, src_vector.at(i), offset, dst);
132  offset += src_vector.at(i)->dimension(_axis);
133  _concat_kernels.emplace_back(std::move(kernel));
134  }
135  break;
136  }
137  default:
138  ARM_COMPUTE_ERROR("Axis not supported");
139  }
140 }
141 
142 Status ClConcatenate::validate(const std::vector<const ITensorInfo *> &src_vector, const ITensorInfo *dst, size_t axis)
143 {
144  ARM_COMPUTE_RETURN_ERROR_ON(dst == nullptr);
145  const unsigned int num_inputs = src_vector.size();
146 
148  ARM_COMPUTE_RETURN_ERROR_ON(num_inputs < 2);
149 
150  unsigned int offset = 0;
151  switch(axis)
152  {
153  case Window::DimX:
154  {
155  switch(num_inputs)
156  {
157  case 2:
158  // Validate WidthConcatenate2Tensors kernels if there are 2 inputs
159  ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(src_vector[0], src_vector[1]);
161  break;
162  case 4:
163  // Validate WidthConcatenate4Tensors kernels if there are 4 inputs
164  ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(src_vector[0], src_vector[1], src_vector[2], src_vector[3]);
165  ARM_COMPUTE_RETURN_ON_ERROR(kernels::ClWidthConcatenate4TensorsKernel::validate(src_vector[0], src_vector[1], src_vector[2], src_vector[3], dst));
166  break;
167  default:
168  // Validate generic case of WidthConcatenate kernel
169  for(const auto &src : src_vector)
170  {
173  offset += src->dimension(axis);
174  }
175  break;
176  }
177  break;
178  }
179  case Window::DimY:
180  {
181  for(const auto &src : src_vector)
182  {
184  offset += src->dimension(axis);
185  }
186  break;
187  }
188  case Window::DimZ:
189  {
190  for(const auto &src : src_vector)
191  {
193  offset += src->dimension(axis);
194  }
195  break;
196  }
197  case 3:
198  {
199  for(const auto &src : src_vector)
200  {
202  offset += src->dimension(axis);
203  }
204  break;
205  }
206  default:
207  ARM_COMPUTE_ERROR("Axis not supported");
208  }
209 
210  if(dst->total_size() != 0)
211  {
214  }
215 
216  return Status{};
217 }
218 
220 {
221  if(tensors.empty())
222  {
223  ARM_COMPUTE_ERROR("No inputs provided");
224  }
225 
226  if(static_cast<int>(tensors.size()) - 1 != static_cast<int>(_num_inputs))
227  {
228  ARM_COMPUTE_ERROR("Configured with different number of inputs");
229  }
230 
231  if(_axis == Window::DimX && (_num_inputs == 2 || _num_inputs == 4))
232  {
233  ARM_COMPUTE_ERROR_ON(_concat_kernels.empty());
234  CLScheduler::get().enqueue_op(*_concat_kernels.at(0), tensors, true);
235  }
236  else
237  {
238  int i = 0;
239  for(auto &k : _concat_kernels)
240  {
244  CLScheduler::get().enqueue_op(*k, pack, true);
245  ++i;
246  }
247  }
248 }
249 } // namespace opencl
250 } // namespace arm_compute
__global uchar * offset(const Image *img, int x, int y)
Get the pointer position of a Image.
Definition: helpers.h:1083
Shape of a tensor.
Definition: TensorShape.h:39
static Status validate(const ITensorInfo *src, unsigned int depth_offset, const ITensorInfo *dst)
Static function to check if given info will lead to a valid configuration.
static CLScheduler & get()
Access the scheduler singleton.
bool empty() const
Checks if pack is empty.
Definition: ITensorPack.cpp:80
#define ARM_COMPUTE_ERROR(msg)
Print the given message then throw an std::runtime_error.
Definition: Error.h:352
#define ARM_COMPUTE_RETURN_ON_ERROR(status)
Checks if a status contains an error and returns it.
Definition: Error.h:204
#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
Store the tensor&#39;s metadata.
Definition: ITensorInfo.h:40
static Status validate(const ITensorInfo *src, unsigned int batch_offset, const ITensorInfo *dst)
Static function to check if given info will lead to a valid configuration.
#define ARM_COMPUTE_ERROR_THROW_ON(status)
Definition: Error.h:455
Status class.
Definition: Error.h:52
static Status validate(const ITensorInfo *src, unsigned int width_offset, const ITensorInfo *dst)
Static function to check if given info will lead to a valid configuration.
#define ARM_COMPUTE_RETURN_ERROR_ON(cond)
If the condition is true, an error is returned.
Definition: Error.h:296
SimpleTensor< float > src
Definition: DFT.cpp:155
Copyright (c) 2017-2022 Arm Limited.
#define ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(...)
Definition: Validate.h:159
Interface to enqueue OpenCL kernels and get/set the OpenCL CommandQueue and ICLTuner.
const ITensor * get_const_tensor(int id) const
Get constant tensor of a given id.
Definition: ITensorPack.cpp:54
static constexpr size_t DimX
Alias for dimension 0 also known as X dimension.
Definition: Window.h:43
static Status validate(const ITensorInfo *src1, const ITensorInfo *src2, const ITensorInfo *src3, const ITensorInfo *src4, const ITensorInfo *dst)
Static function to check if given info will lead to a valid configuration.
void run(ITensorPack &tensors) override
Run the kernels contained in the function.
virtual const TensorShape & tensor_shape() const =0
Size for each dimension of the tensor.
void enqueue_op(ICLKernel &kernel, ITensorPack &tensors, bool flush=true)
Schedule the execution of the passed kernel if possible.
size_t total_size() const
Collapses all dimensions to a single linear total size.
Definition: TensorShape.h:172
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...
size_t size() const
Pack size accessor.
Definition: ITensorPack.cpp:75
CLCompileContext class.
static constexpr size_t DimY
Alias for dimension 1 also known as Y dimension.
Definition: Window.h:45
ITensor * get_tensor(int id)
Get tensor of a given id from the pac.
Definition: ITensorPack.cpp:64
virtual size_t total_size() const =0
Returns the total size of the tensor in bytes.
static constexpr size_t DimZ
Alias for dimension 2 also known as Z dimension.
Definition: Window.h:47
static Status validate(const std::vector< const ITensorInfo *> &src_vector, const ITensorInfo *dst, size_t axis)
Static function to check if given info will lead to a valid configuration.
static Status validate(const ITensorInfo *src, unsigned int height_offset, const ITensorInfo *dst)
Static function to check if given info will lead to a valid configuration.
Tensor packing service.
Definition: ITensorPack.h:39
#define ARM_COMPUTE_LOG_PARAMS(...)
#define ARM_COMPUTE_ERROR_ON_NULLPTR(...)
Definition: Validate.h:157
static Status validate(const ITensorInfo *src1, const ITensorInfo *src2, const ITensorInfo *dst)
Static function to check if given info will lead to a valid configuration.
TensorShape calculate_concatenate_shape(const std::vector< T *> &input, size_t axis)
Calculate the concatenate output shape of the concatenate operation along a single axis...
void configure(const ClCompileContext &compile_context, const std::vector< ITensorInfo *> &src_vector, ITensorInfo *dst, size_t axis)
Initialise the kernel&#39;s inputs vector and dst.
void add_tensor(int id, ITensor *tensor)
Add tensor to the pack.
Definition: ITensorPack.cpp:39