Compute Library
 19.08
CLConvolutionLayer.cpp
Go to the documentation of this file.
1 /*
2  * Copyright (c) 2017-2019 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"
32 
33 #include <cmath>
34 #include <memory>
35 #include <tuple>
36 
37 using namespace arm_compute;
39 
40 CLConvolutionLayer::CLConvolutionLayer(std::shared_ptr<IMemoryManager> memory_manager)
41  : _memory_manager(std::move(memory_manager)), _function()
42 {
43 }
44 
46  const Size2D &dilation, const ActivationLayerInfo &act_info, bool enable_fast_math, unsigned int num_groups)
47 {
48  ARM_COMPUTE_ERROR_ON_NULLPTR(input, weights, output);
49  ARM_COMPUTE_ERROR_THROW_ON(CLConvolutionLayer::validate(input->info(), weights->info(), ((biases != nullptr) ? biases->info() : nullptr), output->info(), conv_info, weights_info, dilation, act_info,
50  enable_fast_math, num_groups));
51 
53  weights_info, act_info, CLScheduler::get().target(), dilation, enable_fast_math))
54  {
56  {
58  auto f = arm_compute::support::cpp14::make_unique<CLWinogradConvolutionLayer>(_memory_manager);
59  f->configure(input, weights, biases, output, conv_info, act_info, enable_fast_math);
60  _function = std::move(f);
61  break;
62  }
64  {
66  auto f = arm_compute::support::cpp14::make_unique<CLDirectConvolutionLayer>();
67  f->configure(input, weights, biases, output, conv_info, act_info);
68  _function = std::move(f);
69  break;
70  }
72  {
73  auto f = arm_compute::support::cpp14::make_unique<CLGEMMConvolutionLayer>(_memory_manager);
74  f->configure(input, weights, biases, output, conv_info, weights_info, dilation, act_info, num_groups);
75  _function = std::move(f);
76  break;
77  }
79  {
80  auto f = arm_compute::support::cpp14::make_unique<CLFFTConvolutionLayer>(_memory_manager);
81  f->configure(input, weights, biases, output, conv_info, act_info);
82  _function = std::move(f);
83  break;
84  }
85  default:
86  ARM_COMPUTE_ERROR("Not supported.");
87  break;
88  }
89 }
90 
92  const WeightsInfo &weights_info, const Size2D &dilation, const ActivationLayerInfo &act_info, bool enable_fast_math, unsigned int num_groups)
93 {
95  ARM_COMPUTE_RETURN_ERROR_ON_MSG((num_groups != 1) && (input->data_layout() != DataLayout::NCHW), "Grouping (num_groups != 1) with NHWC data layout is not supported");
96 
97  const GPUTarget gpu_target = CLScheduler::get().target();
98 
99  switch(CLConvolutionLayer::get_convolution_method(input, weights, output, conv_info, weights_info, act_info, gpu_target, dilation, enable_fast_math))
100  {
102  {
103  //Validate Winograd
104  ARM_COMPUTE_RETURN_ERROR_ON_MSG(num_groups != 1, "Grouping (num_groups != 1) with CLWinogradConvolutionLayer is not supported");
106  break;
107  }
109  {
110  // Validate direct convolution layer
111  ARM_COMPUTE_RETURN_ERROR_ON_MSG(num_groups != 1, "Grouping (num_groups != 1) with CLDirectConvolutionLayer is not supported");
113  break;
114  }
116  {
117  // Validate gemm-based convolution layer
119  break;
120  }
122  {
123  // Validate FFT-based convolution layer
125  break;
126  }
127  default:
128  ARM_COMPUTE_ERROR("Not supported.");
129  break;
130  }
131 
132  return Status{};
133 }
134 
136  const WeightsInfo &weights_info, const ActivationLayerInfo &act_info, const GPUTarget gpu_target, const Size2D &dilation, bool enable_fast_math)
137 {
142  ARM_COMPUTE_UNUSED(gpu_target);
143 
147 
148  /* Input spatial dims, kernel size, IFM/OFM, conv info*/
149  using ConvolutionConfiguration = std::tuple<Size2D, Size2D, Size2D, PadStrideInfo, DataLayout>;
150  using ConfigurationMethod = std::pair<ConvolutionConfiguration, ConvolutionMethod>;
151 
152  const std::vector<ConfigurationMethod> known_configs =
153  {
154  // Alexnet
155  ConfigurationMethod(ConvolutionConfiguration(Size2D(27U, 27U), Size2D(5U, 5U), Size2D(48U, 128U), PadStrideInfo(1U, 1U, 2U, 2U), DataLayout::NCHW), ConvolutionMethod::DIRECT),
156  // VGG16 / VGG19
157  ConfigurationMethod(ConvolutionConfiguration(Size2D(224U, 224U), Size2D(3U, 3U), Size2D(3U, 64U), PadStrideInfo(1U, 1U, 1U, 1U), DataLayout::NCHW), ConvolutionMethod::DIRECT),
158  // Mobilenet 224
159  ConfigurationMethod(ConvolutionConfiguration(Size2D(224U, 224U), Size2D(3U, 3U), Size2D(3U, 32U), PadStrideInfo(2U, 2U, 0U, 1U, 0U, 1U, DimensionRoundingType::FLOOR), DataLayout::NCHW), ConvolutionMethod::GEMM),
160  // Mobilenet 160
161  ConfigurationMethod(ConvolutionConfiguration(Size2D(160U, 160U), Size2D(3U, 3U), Size2D(3U, 24U), PadStrideInfo(2U, 2U, 0U, 1U, 0U, 1U, DimensionRoundingType::FLOOR), DataLayout::NCHW), ConvolutionMethod::GEMM),
162  // Mobilenet 224
163  ConfigurationMethod(ConvolutionConfiguration(Size2D(224U, 224U), Size2D(3U, 3U), Size2D(3U, 32U), PadStrideInfo(2U, 2U, 0U, 1U, 0U, 1U, DimensionRoundingType::FLOOR), DataLayout::NHWC), ConvolutionMethod::GEMM),
164  // Mobilenet 160
165  ConfigurationMethod(ConvolutionConfiguration(Size2D(160U, 160U), Size2D(3U, 3U), Size2D(3U, 24U), PadStrideInfo(2U, 2U, 0U, 1U, 0U, 1U, DimensionRoundingType::FLOOR), DataLayout::NHWC), ConvolutionMethod::GEMM),
166  };
167 
168  const auto find_config = [&](ConfigurationMethod c)
169  {
170  const ConvolutionConfiguration config = c.first;
171  const PadStrideInfo info = std::get<3>(config);
172  const DataLayout data_layout = std::get<4>(config);
173 
174  return std::get<0>(config) == Size2D(input->dimension(idx_w), input->dimension(idx_h)) && std::get<1>(config) == Size2D(weights->dimension(idx_w), weights->dimension(idx_h))
175  && std::get<2>(config) == Size2D(weights->dimension(idx_c), weights->dimension(3)) && info.pad_top() == conv_info.pad_top() && info.pad_right() == conv_info.pad_right()
176  && info.pad_bottom() == conv_info.pad_bottom() && info.pad_left() == conv_info.pad_left() && info.stride() == conv_info.stride() && (data_layout == input->data_layout());
177  };
178 
179  std::vector<ConfigurationMethod>::const_iterator found;
180  if((found = std::find_if(known_configs.begin(), known_configs.end(), find_config)) != known_configs.end())
181  {
182  return (*found).second;
183  }
184 
185  if(dilation != Size2D(1U, 1U))
186  {
188  }
189  else
190  {
191  // SRGAN
192  if((input->dimension(idx_h) > 720U) && (output->dimension(idx_h) > 720U) && (weights->dimension(idx_h) == 9) && (conv_info.pad_top() < 3)
193  && (CLDirectConvolutionLayer::validate(input, weights, nullptr, output, conv_info, act_info)))
194  {
196  }
197  if((weights->dimension(idx_h) > 7) && (input->dimension(idx_c) > output->dimension(idx_c)) && (CLFFTConvolutionLayer::validate(input, weights, nullptr, output, conv_info, act_info)))
198  {
199  return ConvolutionMethod::FFT;
200  }
201  if(input->dimension(idx_c) < 16)
202  {
204  }
205  return bool(CLWinogradConvolutionLayer::validate(input, weights, nullptr, output, conv_info, act_info, enable_fast_math)) ? ConvolutionMethod::WINOGRAD : ConvolutionMethod::GEMM;
206  }
207 }
208 
210 {
211  prepare();
212  _function->run();
213 }
214 
216 {
217  _function->prepare();
218 }
#define ARM_COMPUTE_ERROR(...)
Print the given message then throw an std::runtime_error.
Definition: Error.h:261
const DataLayout data_layout
Definition: Im2Col.cpp:146
void run() override
Run the kernels contained in the function.
TensorInfo * info() const override
Interface to be implemented by the child class to return the tensor's metadata.
Definition: CLTensor.cpp:35
CLConvolutionLayer(std::shared_ptr< IMemoryManager > memory_manager=nullptr)
Default constructor.
void configure(ICLTensor *input, const ICLTensor *weights, const ICLTensor *biases, ICLTensor *output, const PadStrideInfo &conv_info, const WeightsInfo &weights_info=WeightsInfo(), const Size2D &dilation=Size2D(1U, 1U), const ActivationLayerInfo &act_info=ActivationLayerInfo(), bool enable_fast_math=false, unsigned int num_groups=1)
Set the input and output tensors.
virtual size_t dimension(size_t index) const =0
Return the size of the requested dimension.
static CLScheduler & get()
Access the scheduler singleton.
Definition: CLScheduler.cpp:41
GPUTarget target() const
Get the target GPU.
Definition: CLScheduler.h:112
#define ARM_COMPUTE_RETURN_ON_ERROR(status)
Checks if a status contains an error and returns it.
Definition: Error.h:193
void prepare() override
Prepare the function for executing.
#define ARM_COMPUTE_ERROR_ON(cond)
If the condition is true then an error message is printed and an exception thrown.
Definition: Error.h:337
static Status validate(const ITensorInfo *input, const ITensorInfo *weights, const ITensorInfo *biases, const ITensorInfo *output, const PadStrideInfo &conv_info, const WeightsInfo &weights_info=WeightsInfo(), const Size2D &dilation=Size2D(1U, 1U), const ActivationLayerInfo &act_info=ActivationLayerInfo(), unsigned int num_groups=1)
Static function to check if given info will lead to a valid configuration of CLGEMMConvolutionLayer.
Store the tensor's metadata.
Definition: ITensorInfo.h:40
#define ARM_COMPUTE_ERROR_THROW_ON(status)
Definition: Error.h:327
static Status validate(const ITensorInfo *input, const ITensorInfo *weights, const ITensorInfo *biases, 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 CLDirectConvolutionLayer...
static Status validate(const ITensorInfo *input, const ITensorInfo *weights, const ITensorInfo *biases, const ITensorInfo *output, const PadStrideInfo &conv_info, const WeightsInfo &weights_info=WeightsInfo(), const Size2D &dilation=Size2D(1U, 1U), const ActivationLayerInfo &act_info=ActivationLayerInfo(), bool enable_fast_math=false, unsigned int num_groups=1)
Static function to check if given info will lead to a valid configuration of CLConvolutionLayer.
Status class.
Definition: Error.h:52
ConvolutionMethod
Available ConvolutionMethod.
Definition: Types.h:132
Activation Layer Information class.
Definition: Types.h:1517
Copyright (c) 2017-2018 ARM Limited.
Convolution Layer Weights Information class.
Definition: Types.h:1658
#define ARM_COMPUTE_UNUSED(...)
To avoid unused variables warnings.
Definition: Error.h:160
const unsigned int num_groups
Definition: Im2Col.cpp:148
#define ARM_COMPUTE_RETURN_ERROR_ON_MSG(cond,...)
If the condition is true, an error is returned.
Definition: Error.h:214
virtual ITensorInfo * info() const =0
Interface to be implemented by the child class to return the tensor's metadata.
Padding and stride information class.
Definition: Types.h:676
static Status validate(const ITensorInfo *input, const ITensorInfo *weights, const ITensorInfo *biases, const ITensorInfo *output, const PadStrideInfo &conv_info, const ActivationLayerInfo &act_info=ActivationLayerInfo(), bool enable_fast_math=false)
Static function to check if given info will lead to a valid configuration of CLWinogradConvolutionLay...
Num samples, channels, height, width.
Convolution using Winograd.
#define ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(...)
Definition: Validate.h:163
#define ARM_COMPUTE_ERROR_ON_NULLPTR(...)
Definition: Validate.h:161
Interface for OpenCL tensor.
Definition: ICLTensor.h:42
GPUTarget
Available GPU Targets.
Definition: GPUTarget.h:34
Class for specifying the size of an image or rectangle.
Definition: Size2D.h:34
Num samples, height, width, channels.
static Status validate(const ITensorInfo *input, const ITensorInfo *weights, const ITensorInfo *biases, 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 CLFFTConvolutionLayer.
static ConvolutionMethod get_convolution_method(const ITensorInfo *input, const ITensorInfo *weights, const ITensorInfo *output, const PadStrideInfo &conv_info, const WeightsInfo &weights_info, const ActivationLayerInfo &act_info, const GPUTarget gpu_target, const Size2D &dilation=Size2D(1U, 1U), bool enable_fast_math=false)
Static function to check if given info will return the convolution called by CLConvolutionLayer.
size_t get_data_layout_dimension_index(const DataLayout data_layout, const DataLayoutDimension data_layout_dimension)
Get the index of the given dimension.
Definition: Helpers.inl:326
DataLayout
[DataLayout enum definition]
Definition: Types.h:114
virtual DataLayout data_layout() const =0
Get the data layout of the tensor.