Compute Library
 21.05
CLArgMinMaxLayer.cpp
Go to the documentation of this file.
1 /*
2  * Copyright (c) 2018-2020 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 namespace arm_compute
38 {
39 CLArgMinMaxLayer::CLArgMinMaxLayer(std::shared_ptr<IMemoryManager> memory_manager)
40  : _memory_group(std::move(memory_manager)), _results_vector(), _not_reshaped_output(), _reduction_kernels_vector(), _reshape(), _num_of_stages(), _reduction_axis()
41 {
42 }
43 
45 
47 {
52  ARM_COMPUTE_RETURN_ERROR_ON_MSG(axis >= static_cast<int>(TensorShape::num_max_dimensions), "Reduction axis greater than max number of dimensions");
53  ARM_COMPUTE_RETURN_ERROR_ON_MSG(axis > 3, "Unsupported reduction axis");
54  const unsigned int num_of_stages = utils::calculate_number_of_stages_only_x_axis(input->dimension(0), axis);
55 
56  DataType output_data_type = DataType::S32;
57  TensorInfo not_reshaped_output;
58  const auto input_num_channles = input->num_channels();
59  const auto input_qinfo = input->quantization_info();
60 
61  if(output->total_size() != 0)
62  {
63  output_data_type = output->data_type();
64  const TensorInfo expected_output_shape = output->clone()->set_tensor_shape(arm_compute::misc::shape_calculator::compute_reduced_shape(input->tensor_shape(), axis, false));
65  ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_SHAPES(&expected_output_shape, output);
66  }
67 
68  auto shape_before_reshape = input->tensor_shape();
69  shape_before_reshape.set(axis, 1);
70  auto initialize_tensorinfo = [](TensorInfo & ti, TensorShape shape, DataType data_type, int num_channels, QuantizationInfo qinfo)
71  {
73  };
74 
75  initialize_tensorinfo(not_reshaped_output, shape_before_reshape, output_data_type, input_num_channles, input_qinfo);
76 
77  if(num_of_stages == 1)
78  {
79  ARM_COMPUTE_RETURN_ON_ERROR(CLArgMinMaxLayerKernel::validate(input, nullptr, &not_reshaped_output, axis, op));
80  }
81  else
82  {
83  // Create temporary tensor infos
84  std::vector<TensorInfo> sums_vector(num_of_stages - 1);
85 
86  // Create intermediate tensor info
87  TensorShape shape{ input->tensor_shape() };
88 
89  for(unsigned int i = 0; i < num_of_stages - 1; i++)
90  {
91  shape.set(0, ceil(shape.x() / 128.f));
92  sums_vector[i].set_data_type(input->data_type());
93  sums_vector[i].set_tensor_shape(shape);
94  sums_vector[i].set_num_channels(input->num_channels());
95  }
96 
97  // Validate ReductionOperation only on first kernel
98  ARM_COMPUTE_RETURN_ON_ERROR(CLArgMinMaxLayerKernel::validate(input, nullptr, &sums_vector[0], axis, op));
99 
100  // Validate ReductionOperation on intermediate stages
101  for(unsigned int i = 1; i < num_of_stages - 1; ++i)
102  {
103  ARM_COMPUTE_RETURN_ON_ERROR(CLArgMinMaxLayerKernel::validate(input, &sums_vector[i - 1], &sums_vector[i], axis, op));
104  }
105 
106  // Validate ReductionOperation on the last stage
107  const unsigned int last_stage = num_of_stages - 1;
108  ARM_COMPUTE_RETURN_ON_ERROR(CLArgMinMaxLayerKernel::validate(input, &sums_vector[last_stage - 1], &not_reshaped_output, axis, op));
109  }
110  ARM_COMPUTE_RETURN_ON_ERROR(CLReshapeLayer::validate(&not_reshaped_output, output));
111  return Status{};
112 }
113 
114 void CLArgMinMaxLayer::configure(const ICLTensor *input, int axis, ICLTensor *output, const ReductionOperation &op)
115 {
116  configure(CLKernelLibrary::get().get_compile_context(), input, axis, output, op);
117 }
118 
119 void CLArgMinMaxLayer::configure(const CLCompileContext &compile_context, const ICLTensor *input, int axis, ICLTensor *output, const ReductionOperation &op)
120 {
122  _num_of_stages = utils::calculate_number_of_stages_only_x_axis(input->info()->dimension(0), axis);
123  _reduction_axis = axis;
124 
126  DataType output_data_type = (output->info()->data_type() == DataType::UNKNOWN) ? DataType::S32 : output->info()->data_type();
127  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));
128 
129  // Configure reduction operation kernels
130  _reduction_kernels_vector.reserve(_num_of_stages);
131 
132  auto add_reduction_kernel = [this, &compile_context, axis, op](const ICLTensor * input, const ICLTensor * prev_output, ICLTensor * output)
133  {
134  _reduction_kernels_vector.emplace_back(std::make_unique<CLArgMinMaxLayerKernel>());
135  _reduction_kernels_vector.back()->configure(compile_context, input, prev_output, output, axis, op);
136  };
137 
138  _memory_group.manage(&_not_reshaped_output);
139  // Create temporary tensors
140  if(_num_of_stages == 1)
141  {
142  add_reduction_kernel(input, nullptr, &_not_reshaped_output);
143  }
144  else
145  {
146  _results_vector.resize(_num_of_stages - 1);
147  TensorShape shape{ input->info()->tensor_shape() };
148  for(unsigned int i = 0; i < _num_of_stages - 1; i++)
149  {
150  shape.set(0, ceil(shape.x() / 128.f));
151  _results_vector[i].allocator()->init(input->info()->clone()->set_tensor_shape(shape).set_data_type(output_data_type));
152  }
153 
154  // Apply ReductionOperation only on first kernel
155  _memory_group.manage(&_results_vector[0]);
156  add_reduction_kernel(input, nullptr, &_results_vector[0]);
157 
158  // Apply ReductionOperation on intermediate stages
159  for(unsigned int i = 1; i < _num_of_stages - 1; ++i)
160  {
161  _memory_group.manage(&_results_vector[i]);
162  add_reduction_kernel(input, &_results_vector[i - 1], &_results_vector[i]);
163  _results_vector[i - 1].allocator()->allocate();
164  }
165 
166  // Apply ReductionOperation on the last stage
167  const unsigned int last_stage = _num_of_stages - 1;
168  add_reduction_kernel(input, &_results_vector[last_stage - 1], &_not_reshaped_output);
169  _results_vector[last_stage - 1].allocator()->allocate();
170  }
171  _reshape.configure(compile_context, &_not_reshaped_output, output);
172  _not_reshaped_output.allocator()->allocate();
173 }
174 
176 {
177  MemoryGroupResourceScope scope_mg(_memory_group);
178 
179  for(unsigned int i = 0; i < _num_of_stages; ++i)
180  {
181  CLScheduler::get().enqueue(*_reduction_kernels_vector[i], false);
182  }
183  _reshape.run();
184 }
185 } // 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
ReductionOperation
Available reduction operations.
Definition: Types.h:457
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:286
static CLKernelLibrary & get()
Access the KernelLibrary singleton.
Store the tensor's metadata.
Definition: ITensorInfo.h:40
CLTensorAllocator * allocator()
Return a pointer to the tensor'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.
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'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.
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_ERROR_ON_NULLPTR(...)
Definition: Validate.h:157
Store the tensor's metadata.
Definition: TensorInfo.h:43
void configure(const ICLTensor *input, ICLTensor *output)
Initialise the kernel'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:77