Compute Library
 21.11
CLArgMinMaxLayer.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  */
24 
26 
27 #include "arm_compute/core/Error.h"
29 #include "arm_compute/core/Types.h"
32 #include "src/core/CL/CLValidate.h"
35 #include "src/runtime/Utils.h"
36 
37 #include "src/common/utils/Log.h"
38 
39 namespace arm_compute
40 {
41 CLArgMinMaxLayer::CLArgMinMaxLayer(std::shared_ptr<IMemoryManager> memory_manager)
42  : _memory_group(std::move(memory_manager)), _results_vector(), _not_reshaped_output(), _reduction_kernels_vector(), _reshape(), _num_of_stages(), _reduction_axis()
43 {
44 }
45 
47 
49 {
50  ARM_COMPUTE_ERROR_ON_NULLPTR(input, output);
54  ARM_COMPUTE_RETURN_ERROR_ON_MSG(axis >= static_cast<int>(TensorShape::num_max_dimensions), "Reduction axis greater than max number of dimensions");
55  ARM_COMPUTE_RETURN_ERROR_ON_MSG(axis > 3, "Unsupported reduction axis");
56  const unsigned int num_of_stages = utils::calculate_number_of_stages_only_x_axis(input->dimension(0), axis);
57 
58  DataType output_data_type = DataType::S32;
59  TensorInfo not_reshaped_output;
60  const auto input_num_channles = input->num_channels();
61  const auto input_qinfo = input->quantization_info();
62 
63  if(output->total_size() != 0)
64  {
65  output_data_type = output->data_type();
66  const TensorInfo expected_output_shape = output->clone()->set_tensor_shape(arm_compute::misc::shape_calculator::compute_reduced_shape(input->tensor_shape(), axis, false));
67  ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_SHAPES(&expected_output_shape, output);
68  }
69 
70  auto shape_before_reshape = input->tensor_shape();
71  shape_before_reshape.set(axis, 1);
72  auto initialize_tensorinfo = [](TensorInfo & ti, TensorShape shape, DataType data_type, int num_channels, QuantizationInfo qinfo)
73  {
75  };
76 
77  initialize_tensorinfo(not_reshaped_output, shape_before_reshape, output_data_type, input_num_channles, input_qinfo);
78 
79  if(num_of_stages == 1)
80  {
81  ARM_COMPUTE_RETURN_ON_ERROR(CLArgMinMaxLayerKernel::validate(input, nullptr, &not_reshaped_output, axis, op));
82  }
83  else
84  {
85  // Create temporary tensor infos
86  std::vector<TensorInfo> sums_vector(num_of_stages - 1);
87 
88  // Create intermediate tensor info
89  TensorShape shape{ input->tensor_shape() };
90 
91  for(unsigned int i = 0; i < num_of_stages - 1; i++)
92  {
93  shape.set(0, ceil(shape.x() / 128.f));
94  sums_vector[i].set_data_type(input->data_type());
95  sums_vector[i].set_tensor_shape(shape);
96  sums_vector[i].set_num_channels(input->num_channels());
97  }
98 
99  // Validate ReductionOperation only on first kernel
100  ARM_COMPUTE_RETURN_ON_ERROR(CLArgMinMaxLayerKernel::validate(input, nullptr, &sums_vector[0], axis, op));
101 
102  // Validate ReductionOperation on intermediate stages
103  for(unsigned int i = 1; i < num_of_stages - 1; ++i)
104  {
105  ARM_COMPUTE_RETURN_ON_ERROR(CLArgMinMaxLayerKernel::validate(input, &sums_vector[i - 1], &sums_vector[i], axis, op));
106  }
107 
108  // Validate ReductionOperation on the last stage
109  const unsigned int last_stage = num_of_stages - 1;
110  ARM_COMPUTE_RETURN_ON_ERROR(CLArgMinMaxLayerKernel::validate(input, &sums_vector[last_stage - 1], &not_reshaped_output, axis, op));
111  }
112  ARM_COMPUTE_RETURN_ON_ERROR(CLReshapeLayer::validate(&not_reshaped_output, output));
113  return Status{};
114 }
115 
116 void CLArgMinMaxLayer::configure(const ICLTensor *input, int axis, ICLTensor *output, const ReductionOperation &op)
117 {
118  configure(CLKernelLibrary::get().get_compile_context(), input, axis, output, op);
119 }
120 
121 void CLArgMinMaxLayer::configure(const CLCompileContext &compile_context, const ICLTensor *input, int axis, ICLTensor *output, const ReductionOperation &op)
122 {
123  ARM_COMPUTE_ERROR_ON_NULLPTR(input, output);
124  ARM_COMPUTE_LOG_PARAMS(input, axis, output, op);
125 
126  _num_of_stages = utils::calculate_number_of_stages_only_x_axis(input->info()->dimension(0), axis);
127  _reduction_axis = axis;
128 
130  DataType output_data_type = (output->info()->data_type() == DataType::UNKNOWN) ? DataType::S32 : output->info()->data_type();
131  auto_init_if_empty(*output->info(), input->info()->clone()->set_tensor_shape(output_shape).set_data_type(output_data_type).reset_padding().set_is_resizable(true));
132 
133  // Configure reduction operation kernels
134  _reduction_kernels_vector.reserve(_num_of_stages);
135 
136  auto add_reduction_kernel = [this, &compile_context, axis, op](const ICLTensor * input, const ICLTensor * prev_output, ICLTensor * output)
137  {
138  _reduction_kernels_vector.emplace_back(std::make_unique<CLArgMinMaxLayerKernel>());
139  _reduction_kernels_vector.back()->configure(compile_context, input, prev_output, output, axis, op);
140  };
141 
142  _memory_group.manage(&_not_reshaped_output);
143  // Create temporary tensors
144  if(_num_of_stages == 1)
145  {
146  add_reduction_kernel(input, nullptr, &_not_reshaped_output);
147  }
148  else
149  {
150  _results_vector.resize(_num_of_stages - 1);
151  TensorShape shape{ input->info()->tensor_shape() };
152  for(unsigned int i = 0; i < _num_of_stages - 1; i++)
153  {
154  shape.set(0, ceil(shape.x() / 128.f));
155  _results_vector[i].allocator()->init(input->info()->clone()->set_tensor_shape(shape).set_data_type(output_data_type));
156  }
157 
158  // Apply ReductionOperation only on first kernel
159  _memory_group.manage(&_results_vector[0]);
160  add_reduction_kernel(input, nullptr, &_results_vector[0]);
161 
162  // Apply ReductionOperation on intermediate stages
163  for(unsigned int i = 1; i < _num_of_stages - 1; ++i)
164  {
165  _memory_group.manage(&_results_vector[i]);
166  add_reduction_kernel(input, &_results_vector[i - 1], &_results_vector[i]);
167  _results_vector[i - 1].allocator()->allocate();
168  }
169 
170  // Apply ReductionOperation on the last stage
171  const unsigned int last_stage = _num_of_stages - 1;
172  add_reduction_kernel(input, &_results_vector[last_stage - 1], &_not_reshaped_output);
173  _results_vector[last_stage - 1].allocator()->allocate();
174  }
175  _reshape.configure(compile_context, &_not_reshaped_output, output);
176  _not_reshaped_output.allocator()->allocate();
177 }
178 
180 {
181  MemoryGroupResourceScope scope_mg(_memory_group);
182 
183  for(unsigned int i = 0; i < _num_of_stages; ++i)
184  {
185  CLScheduler::get().enqueue(*_reduction_kernels_vector[i], false);
186  }
187  _reshape.run();
188 }
189 } // namespace arm_compute
unsigned int calculate_number_of_stages_only_x_axis(size_t input_x_dimension, unsigned int axis)
Calculate number of stages for parallel implementations.
Definition: Utils.cpp:68
~CLArgMinMaxLayer()
Default destructor.
virtual ITensorInfo & set_num_channels(int num_channels)=0
Set the number of channels to the specified value.
static Status validate(const ITensorInfo *input, const ITensorInfo *output)
Static function to check if given info will lead to a valid configuration of CLReshapeLayer.
#define ARM_COMPUTE_RETURN_ERROR_ON_F16_UNSUPPORTED(tensor)
Definition: CLValidate.h:35
Shape of a tensor.
Definition: TensorShape.h:39
virtual size_t dimension(size_t index) const =0
Return the size of the requested dimension.
ReductionOperation
Available reduction operations.
Definition: Types.h:463
static CLScheduler & get()
Access the scheduler singleton.
void run() override
Run the kernels contained in the function.
virtual ITensorInfo & set_tensor_shape(const TensorShape &shape)=0
Set the shape of an already initialized tensor.
#define ARM_COMPUTE_RETURN_ON_ERROR(status)
Checks if a status contains an error and returns it.
Definition: Error.h:204
virtual DataType data_type() const =0
Data type used for each element of the tensor.
1 channel, 1 F32 per channel
ITensorInfo & set_data_type(DataType data_type) override
Set the data type to the specified value.
Definition: TensorInfo.cpp:287
static CLKernelLibrary & get()
Access the KernelLibrary singleton.
Store the tensor&#39;s metadata.
Definition: ITensorInfo.h:40
CLTensorAllocator * allocator()
Return a pointer to the tensor&#39;s allocator.
Definition: CLTensor.cpp:61
Status class.
Definition: Error.h:52
Copyright (c) 2017-2021 Arm Limited.
1 channel, 1 F16 per channel
void run() override
Run the kernels contained in the function.
static Status validate(const ITensorInfo *input, int axis, const ITensorInfo *output, const ReductionOperation &op)
Static function to check if given info will lead to a valid configuration of CLArgMinMaxLayer.
1 channel, 1 S32 per channel
void manage(IMemoryManageable *obj) override
Sets a object to be managed by the given memory group.
Definition: MemoryGroup.h:79
const DataType data_type
Definition: Im2Col.cpp:150
Quantization information.
CLArgMinMaxLayer(std::shared_ptr< IMemoryManager > memory_manager=nullptr)
Default Constructor.
virtual const TensorShape & tensor_shape() const =0
Size for each dimension of the tensor.
quantized, asymmetric fixed-point 8-bit number unsigned
TensorShape compute_reduced_shape(const TensorShape &input, unsigned int axis, bool keep_dims=true)
Calculate the reduced shape of a tensor given an axis.
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.
virtual ITensorInfo * info() const =0
Interface to be implemented by the child class to return the tensor&#39;s metadata.
static Status validate(const ITensorInfo *input, const ITensorInfo *prev_output, const ITensorInfo *output, unsigned int axis, ReductionOperation op)
Static function to check if given info will lead to a valid configuration of CLArgMinMaxLayerKernel.
virtual ITensorInfo & set_quantization_info(const QuantizationInfo &quantization_info)=0
Set the quantization settings (scale and offset) of the tensor.
virtual QuantizationInfo quantization_info() const =0
Get the quantization settings (scale and offset) of the tensor.
void enqueue(ICLKernel &kernel, bool flush=true)
Schedule the execution of the passed kernel if possible.
CLCompileContext class.
void allocate() override
Allocate size specified by TensorInfo of OpenCL memory.
Memory group resources scope handling class.
Definition: IMemoryGroup.h:82
Interface for OpenCL tensor.
Definition: ICLTensor.h:42
virtual size_t total_size() const =0
Returns the total size of the tensor in bytes.
#define ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_SHAPES(...)
Definition: Validate.h:439
void configure(const ICLTensor *input, int axis, ICLTensor *output, const ReductionOperation &op)
Set the input and output tensors.
const QuantizationInfo qinfo
Definition: Im2Col.cpp:155
#define ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(t, c,...)
Definition: Validate.h:788
#define ARM_COMPUTE_RETURN_ERROR_ON_MSG(cond, msg)
If the condition is true, an error is returned.
Definition: Error.h:244
#define ARM_COMPUTE_LOG_PARAMS(...)
#define ARM_COMPUTE_ERROR_ON_NULLPTR(...)
Definition: Validate.h:157
Store the tensor&#39;s metadata.
Definition: TensorInfo.h:43
void configure(const ICLTensor *input, ICLTensor *output)
Initialise the kernel&#39;s inputs and outputs.
quantized, asymmetric fixed-point 8-bit number signed
static constexpr size_t num_max_dimensions
Number of dimensions the tensor has.
Definition: Dimensions.h:46
DataType
Available data types.
Definition: Types.h:79
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