Compute Library
 23.08
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 {
67 
68  const DataLayout data_layout = src->data_layout();
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 
86  ARM_COMPUTE_RETURN_ERROR_ON(dst->data_type() != data_type);
87  }
88 
89  return Status{};
90 }
91 
93 {
94  ARM_COMPUTE_ERROR_ON(src->data_layout() == DataLayout::UNKNOWN);
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 {
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
122 
123  // Perform validation step
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 {
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 {
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
arm_compute::Steps
Class to describe a number of elements in each dimension.
Definition: Steps.h:40
arm_compute::DataLayout::NCHW
@ NCHW
Num samples, channels, height, width.
arm_compute::test::validation::configure
im2col_func configure(src_target.info(), dst_target.info(), spatial_kernel, conv_info, has_bias)
arm_compute::test::validation::src
SimpleTensor< float > src
Definition: DFT.cpp:155
arm_compute::DataLayout
DataLayout
[DataLayout enum definition]
Definition: CoreTypes.h:109
arm_compute::DataLayoutDimension::CHANNEL
@ CHANNEL
channel
arm_compute::calculate_max_window
Window calculate_max_window(const ValidRegion &valid_region, const Steps &steps, bool skip_border, BorderSize border_size)
Definition: WindowHelpers.cpp:28
arm_compute::test::validation::output_shape
const auto output_shape
Definition: ConvolutionLayer.cpp:411
arm_compute::DataLayout::NHWC
@ NHWC
Num samples, height, width, channels.
arm_compute::TensorShape
Shape of a tensor.
Definition: TensorShape.h:39
arm_compute::test::validation::dst
auto dst
Definition: DFT.cpp:170
arm_compute::cpu::kernels::validate_arguments
Status validate_arguments(const ITensorInfo *src, const ITensorInfo *weights, const ITensorInfo *dst, const PadStrideInfo &conv_info)
Definition: CpuDirectConv2dKernel.cpp:60
ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL
#define ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL(k)
Definition: Validate.h:1004
arm_compute::CPUInfo::get
static CPUInfo & get()
Access the KernelLibrary singleton.
Definition: CPPTypes.cpp:40
arm_compute::ITensorPack::get_tensor
ITensor * get_tensor(int id)
Get tensor of a given id from the pac.
Definition: ITensorPack.cpp:64
REGISTER_FP16_NEON
#define REGISTER_FP16_NEON(func_name)
Definition: Registrars.h:48
arm_compute::ACL_SRC_0
@ ACL_SRC_0
Definition: Types.h:45
CpuDirectConv2dKernel.h
arm_compute::ACL_SRC_1
@ ACL_SRC_1
Definition: Types.h:46
ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES
#define ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(...)
Definition: Validate.h:630
arm_compute::cpu::kernels::neon_fp16_nchw_directconv2d
void neon_fp16_nchw_directconv2d(const Window &window, const ITensor *src, const ITensor *weights, ITensor *dst, const PadStrideInfo &conv_info)
arm_compute::DataLayoutDimension::WIDTH
@ WIDTH
width
ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN
#define ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(t, c,...)
Definition: Validate.h:877
arm_compute::test::validation::data_layout
const auto data_layout
Definition: ConvolutionLayer.cpp:406
ARM_COMPUTE_RETURN_ON_ERROR
#define ARM_COMPUTE_RETURN_ON_ERROR(status)
Checks if a status contains an error and returns it.
Definition: Error.h:204
arm_compute::ITensorInfo::dimension
virtual size_t dimension(size_t index) const =0
Return the size of the requested dimension.
REGISTER_FP32_NEON
#define REGISTER_FP32_NEON(func_name)
Definition: Registrars.h:74
arm_compute::ErrorCode::RUNTIME_ERROR
@ RUNTIME_ERROR
Generic runtime error.
ARM_COMPUTE_ERROR_ON_NULLPTR
#define ARM_COMPUTE_ERROR_ON_NULLPTR(...)
Definition: Validate.h:161
ARM_COMPUTE_ERROR_ON
#define ARM_COMPUTE_ERROR_ON(cond)
If the condition is true then an error message is printed and an exception thrown.
Definition: Error.h:467
arm_compute::ITensorPack::get_const_tensor
const ITensor * get_const_tensor(int id) const
Get constant tensor of a given id.
Definition: ITensorPack.cpp:54
ARM_COMPUTE_ERROR_THROW_ON
#define ARM_COMPUTE_ERROR_THROW_ON(status)
Definition: Error.h:456
arm_compute::ITensorPack
Tensor packing service.
Definition: ITensorPack.h:39
arm_compute::DataLayoutDimension::HEIGHT
@ HEIGHT
height
ARM_COMPUTE_RETURN_ERROR_ON
#define ARM_COMPUTE_RETURN_ERROR_ON(cond)
If the condition is true, an error is returned.
Definition: Error.h:297
arm_compute::ACL_DST
@ ACL_DST
Definition: Types.h:55
ARM_COMPUTE_RETURN_ERROR_ON_CPU_F16_UNSUPPORTED
#define ARM_COMPUTE_RETURN_ERROR_ON_CPU_F16_UNSUPPORTED(tensor)
Definition: Validate.h:115
arm_compute::auto_init_if_empty
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...
Definition: AutoConfiguration.h:43
arm_compute::detail
Definition: Validate.h:40
arm_compute::Status
Status class.
Definition: Error.h:52
WindowHelpers.h
ARM_COMPUTE_ERROR_ON_INVALID_SUBWINDOW
#define ARM_COMPUTE_ERROR_ON_INVALID_SUBWINDOW(f, s)
Definition: Validate.h:205
name
const char * name
Definition: NEBatchNormalizationLayerKernel.cpp:60
arm_compute::cpu::kernels::validate_and_configure_window
std::pair< Status, Window > validate_and_configure_window(ITensorInfo *src, ITensorInfo *dst)
Definition: CpuDirectConv2dKernel.cpp:92
ARM_COMPUTE_UNUSED
#define ARM_COMPUTE_UNUSED(...)
To avoid unused variables warnings.
Definition: Error.h:152
arm_compute::cpu::kernels::neon_fp32_nhwc_directconv2d
void neon_fp32_nhwc_directconv2d(const Window &window, const ITensor *src, const ITensor *weights, ITensor *dst, const PadStrideInfo &conv_info)
Definition: fp32.cpp:33
arm_compute::PadStrideInfo
Definition: CoreTypes.h:138
AutoConfiguration.h
arm_compute::test::validation::data_type
data_type
Definition: Cast.cpp:223
arm_compute::ThreadInfo
Information about executing thread and CPU.
Definition: CPPTypes.h:180
arm_compute::get_data_layout_dimension_index
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:203
ShapeCalculator.h
arm_compute::cpu::kernels::neon_fp32_nchw_directconv2d
void neon_fp32_nchw_directconv2d(const Window &window, const ITensor *src, const ITensor *weights, ITensor *dst, const PadStrideInfo &conv_info)
Definition: all.cpp:56
arm_compute::Window
Describe a multidimensional execution window.
Definition: Window.h:39
Validate.h
arm_compute
Copyright (c) 2017-2023 Arm Limited.
Definition: introduction.dox:24
arm_compute::test::validation::conv_info
const auto conv_info
Definition: ConvolutionLayer.cpp:407
ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DIMENSIONS
#define ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DIMENSIONS(...)
Definition: Validate.h:288
arm_compute::DataType::F16
@ F16
16-bit floating-point number
ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR
#define ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(...)
Definition: Validate.h:163
arm_compute::validate
Status validate(const ITensorInfo *scores_in, const ITensorInfo *boxes_in, const ITensorInfo *batch_splits_in, const ITensorInfo *scores_out, const ITensorInfo *boxes_out, const ITensorInfo *classes, const ITensorInfo *batch_splits_out, const ITensorInfo *keeps, const ITensorInfo *keeps_size, const BoxNMSLimitInfo info)
Definition: CPPBoxWithNonMaximaSuppressionLimit.cpp:214
arm_compute::DataLayout::UNKNOWN
@ UNKNOWN
Unknown data layout.
arm_compute::ITensorInfo
Store the tensor's metadata.
Definition: ITensorInfo.h:43
arm_compute::DataType::F32
@ F32
32-bit floating-point number
ARM_COMPUTE_CREATE_ERROR
#define ARM_COMPUTE_CREATE_ERROR(error_code, msg)
Creates an error with a given message.
Definition: Error.h:159
arm_compute::test::validation::info
ScaleKernelInfo info(interpolation_policy, default_border_mode, PixelValue(), sampling_policy, false)
arm_compute::misc::shape_calculator::compute_deep_convolution_shape
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.
Definition: ShapeCalculator.h:715
arm_compute::cpu::kernels::DataTypeDataLayoutISASelectorData
Definition: CpuKernelSelectionTypes.h:43
Validate.h
arm_compute::DataType
DataType
Available data types.
Definition: CoreTypes.h:82
arm_compute::ITensorInfo::num_dimensions
virtual size_t num_dimensions() const =0
The number of dimensions of the tensor (rank)
list.h