Compute Library
 22.08
CpuDirectConv2dKernel.cpp
Go to the documentation of this file.
1 /*
2  * Copyright (c) 2017-2022 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  */
26 
29 #include "src/core/CPP/Validate.h"
32 
33 using namespace arm_compute::detail;
34 
35 namespace arm_compute
36 {
37 namespace cpu
38 {
39 namespace kernels
40 {
41 static const std::vector<CpuDirectConv2dKernel::DirectConv2dKernel> available_kernels =
42 {
43  {
44  "neon_fp32_nhwc_directconv2d",
45  [](const DataTypeDataLayoutISASelectorData & data) { return data.dt == DataType::F32 && data.dl == DataLayout::NHWC; },
47  },
48  {
49  "neon_fp32_nchw_directconv2d",
50  [](const DataTypeDataLayoutISASelectorData & data) { return data.dt == DataType::F32 && data.dl == DataLayout::NCHW; },
52  },
53  {
54  "neon_fp16_nchw_directconv2d",
55  [](const DataTypeDataLayoutISASelectorData & data) { return data.dt == DataType::F16 && data.dl == DataLayout::NCHW && data.isa.fp16; },
57  },
58 };
59 
61 {
62  ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(src, weights, dst);
67 
68  const DataLayout data_layout = src->data_layout();
69  const int width_idx = get_data_layout_dimension_index(data_layout, DataLayoutDimension::WIDTH);
70  const int height_idx = get_data_layout_dimension_index(data_layout, DataLayoutDimension::HEIGHT);
71  const int channel_idx = get_data_layout_dimension_index(data_layout, DataLayoutDimension::CHANNEL);
72 
73  ARM_COMPUTE_RETURN_ERROR_ON(weights->dimension(channel_idx) != src->dimension(channel_idx));
74  ARM_COMPUTE_RETURN_ERROR_ON(weights->dimension(width_idx) != weights->dimension(height_idx));
77  ARM_COMPUTE_UNUSED(width_idx);
78  // Checks performed when output is configured
79  if(dst->total_size() != 0)
80  {
82 
83  DataType data_type = src->data_type();
84 
87  }
88 
89  return Status{};
90 }
91 
93 {
95  ARM_COMPUTE_UNUSED(src);
96 
97  Window win{};
98  bool window_changed = false;
99 
100  // Configure window without any padding
101  win = calculate_max_window(*dst, Steps());
102 
103  Status err = (window_changed) ? ARM_COMPUTE_CREATE_ERROR(ErrorCode::RUNTIME_ERROR, "Insufficient Padding!") : Status{};
104  return std::make_pair(err, win);
105 }
106 
108 {
109  ARM_COMPUTE_ERROR_ON_NULLPTR(src, weights, dst);
110 
111  _conv_info = conv_info;
112  _data_layout = src->data_layout();
113  _kernel_size = weights->dimension(get_data_layout_dimension_index(_data_layout, DataLayoutDimension::WIDTH));
114 
115  // Get convolved dimensions
117 
118  DataType data_type = src->data_type();
119 
120  // Output auto inizialitation if not yet initialized
121  auto_init_if_empty(*dst, output_shape, 1, data_type);
122 
123  // Perform validation step
124  ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(src, weights, dst, conv_info));
125 
126  // Configure kernel window
127  auto win_config = validate_and_configure_window(src, dst);
128  ARM_COMPUTE_ERROR_THROW_ON(win_config.first);
129  ICpuKernel::configure(win_config.second);
130 }
131 
133 {
134  ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments(src, weights, dst, conv_info));
136  dst->clone().get())
137  .first);
138 
139  return Status{};
140 }
141 
142 void CpuDirectConv2dKernel::run_op(ITensorPack &tensors, const Window &window, const ThreadInfo &info)
143 {
144  ARM_COMPUTE_UNUSED(info);
146  ARM_COMPUTE_ERROR_ON_INVALID_SUBWINDOW(ICpuKernel::window(), window);
147 
149  auto weights = tensors.get_const_tensor(TensorType::ACL_SRC_1);
150  auto dst = tensors.get_tensor(TensorType::ACL_DST);
151 
152  const auto *uk = CpuDirectConv2dKernel::get_implementation(DataTypeDataLayoutISASelectorData{ src->info()->data_type(), _data_layout, CPUInfo::get().get_isa() });
153  ARM_COMPUTE_ERROR_ON(uk == nullptr);
154 
155  uk->ukernel(window, src, weights, dst, _conv_info);
156 }
157 const char *CpuDirectConv2dKernel::name() const
158 {
159  return "CpuDirectConvolutionLayerKernel";
160 }
161 
162 const std::vector<CpuDirectConv2dKernel::DirectConv2dKernel> &CpuDirectConv2dKernel::get_available_kernels()
163 {
164  return available_kernels;
165 }
166 
167 } // namespace kernels
168 } // namespace cpu
169 } // namespace arm_compute
virtual size_t num_dimensions() const =0
The number of dimensions of the tensor (rank)
Window calculate_max_window(const ValidRegion &valid_region, const Steps &steps, bool skip_border, BorderSize border_size)
Status validate(const OperatorGraph &op_graph)
Return the validity of op_graph, usually after performing an operation (e.g.
Shape of a tensor.
Definition: TensorShape.h:39
#define ARM_COMPUTE_RETURN_ERROR_ON_CPU_F16_UNSUPPORTED(tensor)
Definition: Validate.h:115
#define REGISTER_FP16_NEON(func_name)
Definition: Registrars.h:48
virtual size_t dimension(size_t index) const =0
Return the size of the requested dimension.
void neon_fp16_nchw_directconv2d(const Window &window, const ITensor *src, const ITensor *weights, ITensor *dst, const PadStrideInfo &conv_info)
#define REGISTER_FP32_NEON(func_name)
Definition: Registrars.h:74
#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
#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
#define ARM_COMPUTE_ERROR_THROW_ON(status)
Definition: Error.h:455
Status class.
Definition: Error.h:52
Status validate_arguments(const ITensorInfo *src, const ITensorInfo *weights, const ITensorInfo *dst, const PadStrideInfo &conv_info)
#define ARM_COMPUTE_RETURN_ERROR_ON(cond)
If the condition is true, an error is returned.
Definition: Error.h:296
#define ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DIMENSIONS(...)
Definition: Validate.h:284
SimpleTensor< float > src
Definition: DFT.cpp:155
Copyright (c) 2017-2022 Arm Limited.
1 channel, 1 F16 per channel
#define ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(...)
Definition: Validate.h:159
const ITensor * get_const_tensor(int id) const
Get constant tensor of a given id.
Definition: ITensorPack.cpp:54
#define ARM_COMPUTE_UNUSED(...)
To avoid unused variables warnings.
Definition: Error.h:152
void neon_fp32_nchw_directconv2d(const Window &window, const ITensor *src, const ITensor *weights, ITensor *dst, const PadStrideInfo &conv_info)
Definition: all.cpp:56
virtual const TensorShape & tensor_shape() const =0
Size for each dimension of the tensor.
Class to describe a number of elements in each dimension.
Definition: Steps.h:40
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.
Padding and stride information class.
Definition: Types.h:669
const char * name
#define ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL(k)
Definition: Validate.h:915
Num samples, channels, height, width.
ScaleKernelInfo info(interpolation_policy, default_border_mode, PixelValue(), sampling_policy, false)
std::pair< Status, Window > validate_and_configure_window(ITensorInfo *src, ITensorInfo *dst)
ITensor * get_tensor(int id)
Get tensor of a given id from the pac.
Definition: ITensorPack.cpp:64
Information about executing thread and CPU.
Definition: CPPTypes.h:179
virtual size_t total_size() const =0
Returns the total size of the tensor in bytes.
#define ARM_COMPUTE_CREATE_ERROR(error_code, msg)
Creates an error with a given message.
Definition: Error.h:159
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:193
#define ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(...)
Definition: Validate.h:541
Num samples, height, width, channels.
#define ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(t, c,...)
Definition: Validate.h:788
Tensor packing service.
Definition: ITensorPack.h:39
#define ARM_COMPUTE_ERROR_ON_NULLPTR(...)
Definition: Validate.h:157
void neon_fp32_nhwc_directconv2d(const Window &window, const ITensor *src, const ITensor *weights, ITensor *dst, const PadStrideInfo &conv_info)
Definition: fp32.cpp:33
static CPUInfo & get()
Access the KernelLibrary singleton.
Definition: CPPTypes.cpp:40
im2col_func configure(src_target.info(), dst_target.info(), spatial_kernel, conv_info, has_bias)
DataType
Available data types.
Definition: Types.h:79
DataLayout
[DataLayout enum definition]
Definition: Types.h:113
Describe a multidimensional execution window.
Definition: Window.h:39
TensorShape compute_deep_convolution_shape(const TensorShape &input_shape, DataLayout input_data_layout, const TensorShape &weights_shape, const PadStrideInfo &conv_info)
Calculate the deep convolution shape output shape of a tensor.
#define ARM_COMPUTE_ERROR_ON_INVALID_SUBWINDOW(f, s)
Definition: Validate.h:201
cpuinfo::CpuIsaInfo get_isa() const
Gets the current cpu&#39;s ISA information.
Definition: CPPTypes.cpp:124
virtual DataLayout data_layout() const =0
Get the data layout of the tensor.