Compute Library
 21.11
CLFillBorderKernel.cpp
Go to the documentation of this file.
1 /*
2  * Copyright (c) 2016-2021 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 
31 #include "arm_compute/core/Utils.h"
34 #include "support/Cast.h"
35 #include "support/StringSupport.h"
36 
37 namespace arm_compute
38 {
40  : ICLKernel(), _tensor(nullptr)
41 {
43 }
44 
46 {
47  return false;
48 }
49 
50 template <class T>
51 void CLFillBorderKernel::set_constant_border(unsigned int idx, const PixelValue &constant_border_value)
52 {
53  T value;
54  constant_border_value.get(value);
55  ICLKernel::add_argument<T>(idx, static_cast<T>(value));
56 }
57 
58 void CLFillBorderKernel::configure(ICLTensor *tensor, BorderSize border_size, BorderMode border_mode, const PixelValue &constant_border_value)
59 {
60  configure(CLKernelLibrary::get().get_compile_context(), tensor, border_size, border_mode, constant_border_value);
61 }
62 
63 void CLFillBorderKernel::configure(const CLCompileContext &compile_context, ICLTensor *tensor, BorderSize border_size, BorderMode border_mode, const PixelValue &constant_border_value)
64 {
65  _tensor = tensor;
66  configure(compile_context, tensor->info(), border_size, border_mode, constant_border_value);
67 }
68 
69 void CLFillBorderKernel::configure(const CLCompileContext &compile_context, ITensorInfo *tensor, BorderSize border_size, BorderMode border_mode, const PixelValue &constant_border_value)
70 {
71  ARM_COMPUTE_ERROR_ON(tensor == nullptr);
72  ARM_COMPUTE_ERROR_ON(tensor->num_channels() != 1);
73  auto padding_info = get_padding_info({ tensor });
74 
75  border_size.limit(tensor->padding());
76 
77  // If there is no border: early exit
78  if(border_size.empty() || border_mode == BorderMode::UNDEFINED)
79  {
80  return;
81  }
82 
83  // Select appropriate kernel
84  std::string kernel_name = "fill_image_borders_" + lower_string(string_from_border_mode(border_mode));
85 
86  const DataType dt = tensor->data_type();
87 
88  // Define build options
89  CLBuildOptions build_opts;
90  build_opts.add_option("-DDATA_TYPE=" + get_cl_type_from_data_type(dt));
91  build_opts.add_option("-DBORDER_SIZE_TOP=" + support::cpp11::to_string(border_size.top));
92  build_opts.add_option("-DBORDER_SIZE_BOTTOM=" + support::cpp11::to_string(border_size.bottom));
93  build_opts.add_option("-DBORDER_SIZE_LEFT=" + support::cpp11::to_string(border_size.left));
94  build_opts.add_option("-DBORDER_SIZE_RIGHT=" + support::cpp11::to_string(border_size.right));
95 
96  // Create kernel
97  _kernel = create_kernel(compile_context, kernel_name, build_opts.options());
98 
99  // Create static kernel arguments
100  const unsigned int valid_width = tensor->valid_region().shape[0];
101  const unsigned int valid_height = tensor->valid_region().shape[1];
102  const cl_int2 valid_region_coords =
103  {
104  {
105  static_cast<cl_int>(tensor->valid_region().anchor[0]),
106  static_cast<cl_int>(tensor->valid_region().anchor[1]),
107  }
108  };
109  const unsigned int total_valid_width = border_size.left + valid_width + border_size.right;
110 
111  // Set static kernel arguments
112  unsigned int idx = num_arguments_per_3D_tensor(); //Skip the tensor parameters
113  ICLKernel::add_argument<cl_uint>(idx, valid_width);
114  ICLKernel::add_argument<cl_uint>(idx, valid_height);
115  ICLKernel::add_argument<cl_int2>(idx, valid_region_coords);
116  if(BorderMode::CONSTANT == border_mode)
117  {
118  switch(dt)
119  {
120  case DataType::U8:
121  case DataType::QASYMM8:
122  set_constant_border<uint8_t>(idx, constant_border_value);
123  break;
124  case DataType::S8:
126  set_constant_border<int8_t>(idx, constant_border_value);
127  break;
128  case DataType::U16:
129  set_constant_border<uint16_t>(idx, constant_border_value);
130  break;
131  case DataType::S16:
132  set_constant_border<int16_t>(idx, constant_border_value);
133  break;
134  case DataType::U32:
135  set_constant_border<uint32_t>(idx, constant_border_value);
136  break;
137  case DataType::S32:
138  set_constant_border<int32_t>(idx, constant_border_value);
139  break;
140  case DataType::F32:
141  static_assert(sizeof(float) == 4, "Float must be 32 bit");
142  set_constant_border<float>(idx, constant_border_value);
143  break;
144  case DataType::F16:
145  static_assert(sizeof(cl_half) == sizeof(half), "Half must be same size as cl_half");
146  static_assert(sizeof(cl_half) == 2, "Half must be 16 bit");
147  set_constant_border<half>(idx, constant_border_value);
148  break;
149  default:
150  ARM_COMPUTE_ERROR("Not handled");
151  }
152  }
153 
154  // Configure kernel window
155  Window win;
156  win.set(Window::DimX, Window::Dimension(0, total_valid_width + valid_height));
157  win.set(Window::DimY, Window::Dimension(0, 1, 1));
159  ICLKernel::configure_internal(win);
160 
161  // Set config_id for enabling LWS tuning
162  _config_id = kernel_name;
163  _config_id += "_";
164  _config_id += lower_string(string_from_data_type(dt));
165  _config_id += "_";
166  _config_id += support::cpp11::to_string(tensor->dimension(0));
167  _config_id += "_";
168  _config_id += support::cpp11::to_string(tensor->dimension(1));
169  _config_id += "_";
170  _config_id += lower_string(string_from_border_mode(border_mode));
172 }
173 
174 void CLFillBorderKernel::run_op(ITensorPack &tensors, const Window &window, cl::CommandQueue &queue)
175 {
176  // Border mode undefined or border width == 0
177  if(_kernel() == nullptr)
178  {
179  return;
180  }
181 
182  const auto tensor = utils::cast::polymorphic_downcast<const ICLTensor *>(tensors.get_const_tensor(TensorType::ACL_SRC));
183 
186 
188  Window slice = collapsed.first_slice_window_3D();
189 
190  do
191  {
192  unsigned int idx = 0;
193  add_3D_tensor_argument(idx, tensor, slice);
194  enqueue(queue, *this, slice, lws_hint());
195  }
196  while(collapsed.slide_window_slice_3D(slice));
197 }
198 
199 void CLFillBorderKernel::run(const Window &window, cl::CommandQueue &queue)
200 {
201  // Border mode undefined or border width == 0
202  if(_kernel() == nullptr)
203  {
204  return;
205  }
206 
209 
211  Window slice = collapsed.first_slice_window_3D();
212 
213  do
214  {
215  unsigned int idx = 0;
216  add_3D_tensor_argument(idx, _tensor, slice);
217  enqueue(queue, *this, slice, lws_hint());
218  }
219  while(collapsed.slide_window_slice_3D(slice));
220 }
221 } // namespace arm_compute
BorderMode
Methods available to handle borders.
Definition: Types.h:261
void run_op(ITensorPack &tensors, const Window &window, cl::CommandQueue &queue) override
Enqueue the OpenCL kernel to process the given window on the passed OpenCL command queue...
unsigned int top
top of the border
Definition: Types.h:377
Class describing the value of a pixel for any image format.
Definition: PixelValue.h:34
void configure(const CLCompileContext &compile_context, ICLTensor *tensor, BorderSize border_size, BorderMode border_mode, const PixelValue &constant_border_value=PixelValue())
Initialise the kernel&#39;s input, output and border mode.
const Window & window() const
The maximum window the kernel can be executed on.
Definition: IKernel.cpp:28
TensorShape shape
Shape of the valid region.
Definition: Types.h:257
virtual size_t dimension(size_t index) const =0
Return the size of the requested dimension.
Container for 2D border size.
Definition: Types.h:269
void enqueue(cl::CommandQueue &queue, ICLKernel &kernel, const Window &window, const cl::NDRange &lws_hint=CLKernelLibrary::get().default_ndrange(), bool use_dummy_work_items=false)
Add the kernel to the command queue with the given window.
Definition: ICLKernel.cpp:32
const StringSet & options() const
Gets the current options list set.
#define ARM_COMPUTE_ERROR(msg)
Print the given message then throw an std::runtime_error.
Definition: Error.h:352
cl::NDRange lws_hint() const
Return the Local-Workgroup-Size hint.
Definition: ICLKernel.h:318
1 channel, 1 U8 per channel
void get(uint8_t &v) const
Interpret the pixel value as a U8.
Definition: PixelValue.h:244
std::string to_string(T &&value)
Convert integer and float values to string.
virtual DataType data_type() const =0
Data type used for each element of the tensor.
half_float::half half
16-bit floating point type
Definition: Types.h:48
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
bool is_parallelisable() const override
Indicates whether or not the kernel is parallelisable.
static CLKernelLibrary & get()
Access the KernelLibrary singleton.
Store the tensor&#39;s metadata.
Definition: ITensorInfo.h:40
Describe one of the image&#39;s dimensions with a start, end and step.
Definition: Window.h:77
1 channel, 1 U16 per channel
unsigned int bottom
bottom of the border
Definition: Types.h:379
Manages all the OpenCL kernels compilation and caching, provides accessors for the OpenCL Context...
std::string lower_string(const std::string &val)
Lower a given string.
Definition: Utils.cpp:326
Common interface for all the OpenCL kernels.
Definition: ICLKernel.h:68
void set_constant_border(unsigned int idx, const PixelValue &constant_border_value)
Function to set the constant value on fill border kernel depending on type.
void add_3D_tensor_argument(unsigned int &idx, const ICLTensor *tensor, const Window &window)
Add the passed 3D tensor&#39;s parameters to the object&#39;s kernel&#39;s arguments starting from the index idx...
Definition: ICLKernel.h:214
void use_tensor_dimensions(const TensorShape &shape, size_t first_dimension=Window::DimX)
Use the tensor&#39;s dimensions to fill the window dimensions.
Definition: Window.inl:276
Copyright (c) 2017-2021 Arm Limited.
1 channel, 1 F16 per channel
virtual ValidRegion valid_region() const =0
Valid region of the tensor.
const std::string & string_from_border_mode(BorderMode border_mode)
Translates a given border mode policy to a string.
Definition: Utils.cpp:199
1 channel, 1 S32 per channel
void add_option(std::string option)
Adds option to the existing build option list.
const ITensor * get_const_tensor(int id) const
Get constant tensor of a given id.
Definition: ITensorPack.cpp:54
cl::Kernel create_kernel(const CLCompileContext &ctx, const std::string &kernel_name, const std::set< std::string > &build_opts=std::set< std::string >())
Creates an opencl kernel using a compile context.
Definition: CLHelpers.cpp:391
const std::string & string_from_data_type(DataType dt)
Convert a data type identity into a string.
Definition: Utils.cpp:135
static constexpr size_t DimX
Alias for dimension 0 also known as X dimension.
Definition: Window.h:43
static constexpr unsigned int num_arguments_per_3D_tensor()
Returns the number of arguments enqueued per 3D tensor object.
Definition: ICLKernel.h:256
#define ARM_COMPUTE_ERROR_ON_MISMATCHING_WINDOWS(f, w)
Definition: Validate.h:179
Window collapse_if_possible(const Window &full_window, size_t first, size_t last, bool *has_collapsed=nullptr) const
Collapse the dimensions between first and last if possible.
Definition: Window.inl:68
1 channel, 1 U32 per channel
virtual const TensorShape & tensor_shape() const =0
Size for each dimension of the tensor.
quantized, asymmetric fixed-point 8-bit number unsigned
std::string get_cl_type_from_data_type(const DataType &dt)
Translates a tensor data type to the appropriate OpenCL type.
Definition: CLHelpers.cpp:39
CLFillBorderKernel()
Default constructor.
virtual ITensorInfo * info() const =0
Interface to be implemented by the child class to return the tensor&#39;s metadata.
virtual BorderSize border_size() const
The size of the border for that kernel.
Definition: IKernel.cpp:46
void set(size_t dimension, const Dimension &dim)
Set the values of a given dimension.
Definition: Window.inl:49
virtual PaddingSize padding() const =0
Padding of tensor.
unsigned int left
left of the border
Definition: Types.h:380
Elementeise CL kernel type.
Definition: CLTypes.h:84
bool slide_window_slice_3D(Window &slice) const
Slide the passed 3D window slice.
Definition: Window.h:335
unsigned int right
right of the border
Definition: Types.h:378
#define ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL(k)
Definition: Validate.h:915
1 channel, 1 S16 per channel
bool has_padding_changed(const std::unordered_map< const ITensorInfo *, PaddingSize > &padding_map)
Check if the previously stored padding info has changed after configuring a kernel.
Definition: Utils.cpp:533
CLCompileContext class.
void run(const Window &window, cl::CommandQueue &queue) override
Enqueue the OpenCL kernel to process the given window on the passed OpenCL command queue...
static constexpr size_t DimY
Alias for dimension 1 also known as Y dimension.
Definition: Window.h:45
Interface for OpenCL tensor.
Definition: ICLTensor.h:42
constexpr bool empty() const
Check if the entire border is zero.
Definition: Types.h:302
static constexpr size_t DimZ
Alias for dimension 2 also known as Z dimension.
Definition: Window.h:47
Borders are left undefined.
std::unordered_map< const ITensorInfo *, PaddingSize > get_padding_info(std::initializer_list< const ITensorInfo *> infos)
Stores padding information before configuring a kernel.
Definition: Utils.cpp:518
Wrapper to configure the Khronos OpenCL C++ header.
Tensor packing service.
Definition: ITensorPack.h:39
quantized, asymmetric fixed-point 8-bit number signed
Window first_slice_window_3D() const
First 3D slice of the window.
Definition: Window.h:291
std::string kernel_name
DataType
Available data types.
Definition: Types.h:79
signed 8-bit number
Describe a multidimensional execution window.
Definition: Window.h:39
virtual size_t num_channels() const =0
The number of channels for each tensor element.
void limit(const BorderSize &limit)
Limit this border size.
Definition: Types.h:369
Coordinates anchor
Anchor for the start of the valid region.
Definition: Types.h:256
SimpleTensor< T > slice(const SimpleTensor< T > &src, Coordinates starts, Coordinates ends)