Compute Library
 21.02
NEDirectConvolutionLayer.cpp
Go to the documentation of this file.
1 /*
2  * Copyright (c) 2017-2020 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  */
25 
27 #include "arm_compute/core/Utils.h"
33 
34 namespace arm_compute
35 {
37 
38 NEDirectConvolutionLayer::NEDirectConvolutionLayer(std::shared_ptr<IMemoryManager> memory_manager)
39  : _memory_group(std::move(memory_manager)), _output_stage_kernel(), _conv_kernel(), _input_border_handler(), _activationlayer_function(), _accumulator(), _has_bias(false),
40  _is_activationlayer_enabled(false), _dim_split(Window::DimZ), _is_padding_required()
41 {
42 }
43 
44 void NEDirectConvolutionLayer::configure(ITensor *input, const ITensor *weights, const ITensor *bias, ITensor *output, const PadStrideInfo &conv_info, const ActivationLayerInfo &act_info)
45 {
47  _output_stage_kernel = std::make_unique<NEDirectConvolutionLayerOutputStageKernel>();
48  _conv_kernel = std::make_unique<NEDirectConvolutionLayerKernel>();
49  _input_border_handler = std::make_unique<NEFillBorderKernel>();
50 
51  // Free accumulator
52  if(_accumulator.buffer() != nullptr)
53  {
54  _accumulator.allocator()->free();
55  }
56 
57  _dim_split = input->info()->data_layout() == DataLayout::NCHW ? Window::DimZ : Window::DimY;
58 
59  // Check if bias should be added in the convolution result
60  _has_bias = (bias != nullptr);
61 
62  _conv_kernel->configure(input, weights, output, conv_info);
63  if(_has_bias)
64  {
65  _output_stage_kernel->configure(output, bias);
66  }
67  _is_padding_required = !_conv_kernel->border_size().empty();
68 
69  if(_is_padding_required)
70  {
71  // Add zero padding XY
72  _input_border_handler->configure(input, _conv_kernel->border_size(), BorderMode::CONSTANT, PixelValue(static_cast<float>(0.f)));
73  }
74 
75  //Configure Activation Layer
76  _is_activationlayer_enabled = act_info.enabled();
77  if(_is_activationlayer_enabled)
78  {
79  _activationlayer_function.configure(output, nullptr, act_info);
80  }
81 }
82 
84  const ActivationLayerInfo &act_info)
85 {
86  ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(input, weights, output);
87 
88  // output might not be initialized since it can be an intermediate tensor of another layer
89  DataType data_type = input->data_type();
90  TensorInfo accumulator(output->clone()->set_is_resizable(true).reset_padding().set_data_type(data_type));
91 
92  // Validate Convolution kernel
93  ARM_COMPUTE_RETURN_ON_ERROR(NEDirectConvolutionLayerKernel::validate(input, weights, &accumulator, conv_info));
94 
95  if(bias != nullptr)
96  {
98  ARM_COMPUTE_RETURN_ERROR_ON_MSG(bias->dimension(0) != weights->dimension(3),
99  "Biases size and number of input feature maps should match");
100  ARM_COMPUTE_RETURN_ERROR_ON_MSG(bias->num_dimensions() > 1, "Biases should be one dimensional");
101  }
102 
103  // Validate bias kernel
105 
106  if(act_info.enabled())
107  {
108  ARM_COMPUTE_RETURN_ON_ERROR(NEActivationLayer::validate(output, nullptr, act_info));
109  }
110 
111  return Status{};
112 }
113 
115 {
116  MemoryGroupResourceScope scope_mg(_memory_group);
117 
118  if(_is_padding_required)
119  {
120  NEScheduler::get().schedule(_input_border_handler.get(), Window::DimZ);
121  }
122  NEScheduler::get().schedule(_conv_kernel.get(), _dim_split);
123  if(_has_bias)
124  {
125  NEScheduler::get().schedule(_output_stage_kernel.get(), Window::DimY);
126  }
127 
128  if(_is_activationlayer_enabled)
129  {
130  _activationlayer_function.run();
131  }
132 }
133 } // namespace arm_compute
virtual size_t num_dimensions() const =0
The number of dimensions of the tensor (rank)
Class describing the value of a pixel for any image format.
Definition: PixelValue.h:34
~NEDirectConvolutionLayer()
Default destructor.
bool enabled() const
Check if initialised.
Definition: Types.h:1600
virtual size_t dimension(size_t index) const =0
Return the size of the requested dimension.
#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.
static Status validate(const ITensorInfo *input, const ITensorInfo *output, const ActivationLayerInfo &act_info)
[NEActivationLayer snippet]
void run() override
Run the kernels contained in the function.
#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
Status class.
Definition: Error.h:52
Activation Layer Information class.
Definition: Types.h:1550
Interface for Neon tensor.
Definition: ITensor.h:36
Copyright (c) 2017-2021 Arm Limited.
TensorAllocator * allocator()
Return a pointer to the tensor&#39;s allocator.
Definition: Tensor.cpp:48
#define ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(...)
Definition: Validate.h:163
NEDirectConvolutionLayer(std::shared_ptr< IMemoryManager > memory_manager=nullptr)
Constructor.
const DataType data_type
Definition: Im2Col.cpp:150
void run() override
Run the kernels contained in the function.
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.
Padding and stride information class.
Definition: Types.h:722
void free() override
Free allocated CPU memory.
Num samples, channels, height, width.
void configure(ITensor *input, const ITensor *weights, const ITensor *bias, ITensor *output, const PadStrideInfo &conv_info, const ActivationLayerInfo &act_info=ActivationLayerInfo())
Set the input, weights, biases and output tensors.
static constexpr size_t DimY
Alias for dimension 1 also known as Y dimension.
Definition: Window.h:45
static Status validate(const ITensorInfo *input, const ITensorInfo *bias=nullptr, const ITensorInfo *output=nullptr, const DirectConvolutionLayerOutputStageKernelInfo &info=DirectConvolutionLayerOutputStageKernelInfo())
Static function to check if given info will lead to a valid configuration of NEDirectConvolutionLayer...
Memory group resources scope handling class.
Definition: IMemoryGroup.h:82
virtual void schedule(ICPPKernel *kernel, const Hints &hints)=0
Runs the kernel in the same thread as the caller synchronously.
static constexpr size_t DimZ
Alias for dimension 2 also known as Z dimension.
Definition: Window.h:47
#define ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(...)
Definition: Validate.h:545
void configure(ITensor *input, ITensor *output, ActivationLayerInfo activation_info)
[NEActivationLayer snippet]
uint8_t * buffer() const override
Interface to be implemented by the child class to return a pointer to CPU memory. ...
Definition: Tensor.cpp:43
static Status validate(const ITensorInfo *input, const ITensorInfo *weights, const ITensorInfo *output, const PadStrideInfo &conv_info)
Static function to check if given info will lead to a valid configuration of NEDirectConvolutionLayer...
#define ARM_COMPUTE_RETURN_ERROR_ON_MSG(cond, msg)
If the condition is true, an error is returned.
Definition: Error.h:244
Store the tensor&#39;s metadata.
Definition: TensorInfo.h:45
DataType
Available data types.
Definition: Types.h:77
static Status validate(const ITensorInfo *input, const ITensorInfo *weights, const ITensorInfo *bias, const ITensorInfo *output, const PadStrideInfo &conv_info, const ActivationLayerInfo &act_info=ActivationLayerInfo())
Static function to check if given info will lead to a valid configuration of NEDirectConvolutionLayer...
Describe a multidimensional execution window.
Definition: Window.h:39
virtual DataLayout data_layout() const =0
Get the data layout of the tensor.
static IScheduler & get()
Access the scheduler singleton.
Definition: Scheduler.cpp:94