Compute Library
 20.08
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"
30 
31 #include <cmath>
32 #include <tuple>
33 
34 namespace arm_compute
35 {
36 NEDirectConvolutionLayer::NEDirectConvolutionLayer(std::shared_ptr<IMemoryManager> memory_manager)
37  : _memory_group(std::move(memory_manager)), _output_stage_kernel(), _conv_kernel(), _input_border_handler(), _activationlayer_function(), _accumulator(), _has_bias(false),
38  _is_activationlayer_enabled(false), _dim_split(Window::DimZ)
39 {
40 }
41 
43 {
44  ARM_COMPUTE_ERROR_ON(input->info()->data_layout() == DataLayout::UNKNOWN);
45 
46  // Free accumulator
47  if(_accumulator.buffer() != nullptr)
48  {
49  _accumulator.allocator()->free();
50  }
51 
52  _dim_split = input->info()->data_layout() == DataLayout::NCHW ? Window::DimZ : Window::DimY;
53 
54  // Check if bias should be added in the convolution result
55  _has_bias = (bias != nullptr);
56 
57  _conv_kernel.configure(input, weights, output, conv_info);
58  if(_has_bias)
59  {
60  _output_stage_kernel.configure(output, bias);
61  }
62 
63  // Add zero padding XY
64  _input_border_handler.configure(input, _conv_kernel.border_size(), BorderMode::CONSTANT, PixelValue(static_cast<float>(0.f)));
65 
66  //Configure Activation Layer
67  _is_activationlayer_enabled = act_info.enabled();
68  if(_is_activationlayer_enabled)
69  {
70  _activationlayer_function.configure(output, nullptr, act_info);
71  }
72 }
73 
75  const ActivationLayerInfo &act_info)
76 {
78 
79  // output might not be initialized since it can be an intermediate tensor of another layer
80  DataType data_type = input->data_type();
81  TensorInfo accumulator(output->clone()->set_is_resizable(true).reset_padding().set_data_type(data_type));
82 
83  // Validate Convolution kernel
85 
86  if(bias != nullptr)
87  {
89  ARM_COMPUTE_RETURN_ERROR_ON_MSG(bias->dimension(0) != weights->dimension(3),
90  "Biases size and number of input feature maps should match");
91  ARM_COMPUTE_RETURN_ERROR_ON_MSG(bias->num_dimensions() > 1, "Biases should be one dimensional");
92  }
93 
94  // Validate bias kernel
96 
97  if(act_info.enabled())
98  {
99  ARM_COMPUTE_RETURN_ON_ERROR(NEActivationLayer::validate(output, nullptr, act_info));
100  }
101 
102  return Status{};
103 }
104 
106 {
107  NEScheduler::get().schedule(&_input_border_handler, Window::DimZ);
108 
109  MemoryGroupResourceScope scope_mg(_memory_group);
110 
111  NEScheduler::get().schedule(&_conv_kernel, _dim_split);
112  if(_has_bias)
113  {
114  NEScheduler::get().schedule(&_output_stage_kernel, Window::DimY);
115  }
116 
117  if(_is_activationlayer_enabled)
118  {
119  _activationlayer_function.run();
120  }
121 }
122 } // namespace arm_compute
Class describing the value of a pixel for any image format.
Definition: PixelValue.h:34
bool enabled() const
Check if initialised.
Definition: Types.h:1567
void configure(const ITensor *input, const ITensor *weights, ITensor *output, const PadStrideInfo &conv_info)
Set the input, weights, and output tensors.
#define ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(...)
Definition: Validate.h:545
#define ARM_COMPUTE_RETURN_ON_ERROR(status)
Checks if a status contains an error and returns it.
Definition: Error.h:204
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's metadata.
Definition: ITensorInfo.h:40
void configure(ITensor *tensor, BorderSize border_size, BorderMode border_mode, const PixelValue &constant_border_value=PixelValue())
Initialise the function.
Status class.
Definition: Error.h:52
Activation Layer Information class.
Definition: Types.h:1517
Interface for NEON tensor.
Definition: ITensor.h:36
Copyright (c) 2017-2020 Arm Limited.
TensorAllocator * allocator()
Return a pointer to the tensor's allocator.
Definition: Tensor.cpp:48
NEDirectConvolutionLayer(std::shared_ptr< IMemoryManager > memory_manager=nullptr)
Constructor.
void configure(ITensor *input, const ITensor *bias=nullptr, ITensor *output=nullptr, const DirectConvolutionLayerOutputStageKernelInfo &info=DirectConvolutionLayerOutputStageKernelInfo())
Set the accumulate buffer and the biases of the kernel.
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.
Padding and stride information class.
Definition: Types.h:689
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.
#define ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(...)
Definition: Validate.h:163
static constexpr size_t DimY
Alias for dimension 1 also known as Y dimension.
Definition: Window.h:45
BorderSize border_size() const override
The size of the border for that kernel.
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
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'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
static IScheduler & get()
Access the scheduler singleton.
Definition: Scheduler.cpp:95