Compute Library
 21.05
CLFillBorderKernel.cpp
Go to the documentation of this file.
1 /*
2  * Copyright (c) 2016-2020 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 
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 {
42 }
43 
45 {
46  return false;
47 }
48 
49 template <class T>
50 void CLFillBorderKernel::set_constant_border(unsigned int idx, const PixelValue &constant_border_value)
51 {
52  T value;
53  constant_border_value.get(value);
54  ICLKernel::add_argument<T>(idx, static_cast<T>(value));
55 }
56 
57 void CLFillBorderKernel::configure(ICLTensor *tensor, BorderSize border_size, BorderMode border_mode, const PixelValue &constant_border_value)
58 {
59  configure(CLKernelLibrary::get().get_compile_context(), tensor, border_size, border_mode, constant_border_value);
60 }
61 
62 void CLFillBorderKernel::configure(const CLCompileContext &compile_context, ICLTensor *tensor, BorderSize border_size, BorderMode border_mode, const PixelValue &constant_border_value)
63 {
64  _tensor = tensor;
65  configure(compile_context, tensor->info(), border_size, border_mode, constant_border_value);
66 }
67 
68 void CLFillBorderKernel::configure(const CLCompileContext &compile_context, ITensorInfo *tensor, BorderSize border_size, BorderMode border_mode, const PixelValue &constant_border_value)
69 {
70  ARM_COMPUTE_ERROR_ON(tensor == nullptr);
71  ARM_COMPUTE_ERROR_ON(tensor->num_channels() != 1);
72  auto padding_info = get_padding_info({ tensor });
73 
74  border_size.limit(tensor->padding());
75 
76  // If there is no border: early exit
77  if(border_size.empty() || border_mode == BorderMode::UNDEFINED)
78  {
79  return;
80  }
81 
82  // Select appropriate kernel
83  std::string kernel_name = "fill_image_borders_" + lower_string(string_from_border_mode(border_mode));
84 
85  const DataType dt = tensor->data_type();
86 
87  // Define build options
88  CLBuildOptions build_opts;
89  build_opts.add_option("-DDATA_TYPE=" + get_cl_type_from_data_type(dt));
90  build_opts.add_option("-DBORDER_SIZE_TOP=" + support::cpp11::to_string(border_size.top));
91  build_opts.add_option("-DBORDER_SIZE_BOTTOM=" + support::cpp11::to_string(border_size.bottom));
92  build_opts.add_option("-DBORDER_SIZE_LEFT=" + support::cpp11::to_string(border_size.left));
93  build_opts.add_option("-DBORDER_SIZE_RIGHT=" + support::cpp11::to_string(border_size.right));
94 
95  // Create kernel
96  _kernel = create_kernel(compile_context, kernel_name, build_opts.options());
97 
98  // Create static kernel arguments
99  const unsigned int valid_width = tensor->valid_region().shape[0];
100  const unsigned int valid_height = tensor->valid_region().shape[1];
101  const cl_int2 valid_region_coords =
102  {
103  {
104  static_cast<cl_int>(tensor->valid_region().anchor[0]),
105  static_cast<cl_int>(tensor->valid_region().anchor[1]),
106  }
107  };
108  const unsigned int total_valid_width = border_size.left + valid_width + border_size.right;
109 
110  // Set static kernel arguments
111  unsigned int idx = num_arguments_per_3D_tensor(); //Skip the tensor parameters
112  ICLKernel::add_argument<cl_uint>(idx, valid_width);
113  ICLKernel::add_argument<cl_uint>(idx, valid_height);
114  ICLKernel::add_argument<cl_int2>(idx, valid_region_coords);
115  if(BorderMode::CONSTANT == border_mode)
116  {
117  switch(dt)
118  {
119  case DataType::U8:
120  case DataType::QASYMM8:
121  set_constant_border<uint8_t>(idx, constant_border_value);
122  break;
123  case DataType::S8:
125  set_constant_border<int8_t>(idx, constant_border_value);
126  break;
127  case DataType::U16:
128  set_constant_border<uint16_t>(idx, constant_border_value);
129  break;
130  case DataType::S16:
131  set_constant_border<int16_t>(idx, constant_border_value);
132  break;
133  case DataType::U32:
134  set_constant_border<uint32_t>(idx, constant_border_value);
135  break;
136  case DataType::S32:
137  set_constant_border<int32_t>(idx, constant_border_value);
138  break;
139  case DataType::F32:
140  static_assert(sizeof(float) == 4, "Float must be 32 bit");
141  set_constant_border<float>(idx, constant_border_value);
142  break;
143  case DataType::F16:
144  static_assert(sizeof(cl_half) == sizeof(half), "Half must be same size as cl_half");
145  static_assert(sizeof(cl_half) == 2, "Half must be 16 bit");
146  set_constant_border<half>(idx, constant_border_value);
147  break;
148  default:
149  ARM_COMPUTE_ERROR("Not handled");
150  }
151  }
152 
153  // Configure kernel window
154  Window win;
155  win.set(Window::DimX, Window::Dimension(0, total_valid_width + valid_height));
156  win.set(Window::DimY, Window::Dimension(0, 1, 1));
158  ICLKernel::configure_internal(win);
159 
160  // Set config_id for enabling LWS tuning
161  _config_id = kernel_name;
162  _config_id += "_";
163  _config_id += lower_string(string_from_data_type(dt));
164  _config_id += "_";
165  _config_id += support::cpp11::to_string(tensor->dimension(0));
166  _config_id += "_";
167  _config_id += support::cpp11::to_string(tensor->dimension(1));
168  _config_id += "_";
169  _config_id += lower_string(string_from_border_mode(border_mode));
171 }
172 
173 void CLFillBorderKernel::run_op(ITensorPack &tensors, const Window &window, cl::CommandQueue &queue)
174 {
175  // Border mode undefined or border width == 0
176  if(_kernel() == nullptr)
177  {
178  return;
179  }
180 
181  const auto tensor = utils::cast::polymorphic_downcast<const ICLTensor *>(tensors.get_const_tensor(TensorType::ACL_SRC));
182 
185 
187  Window slice = collapsed.first_slice_window_3D();
188 
189  do
190  {
191  unsigned int idx = 0;
192  add_3D_tensor_argument(idx, tensor, slice);
193  enqueue(queue, *this, slice, lws_hint());
194  }
195  while(collapsed.slide_window_slice_3D(slice));
196 }
197 
198 void CLFillBorderKernel::run(const Window &window, cl::CommandQueue &queue)
199 {
200  // Border mode undefined or border width == 0
201  if(_kernel() == nullptr)
202  {
203  return;
204  }
205 
208 
210  Window slice = collapsed.first_slice_window_3D();
211 
212  do
213  {
214  unsigned int idx = 0;
215  add_3D_tensor_argument(idx, _tensor, slice);
216  enqueue(queue, *this, slice, lws_hint());
217  }
218  while(collapsed.slide_window_slice_3D(slice));
219 }
220 } // namespace arm_compute
BorderMode
Methods available to handle borders.
Definition: Types.h:259
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:375
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's input, output and border mode.
const Window & window() const
The maximum window the kernel can be executed on.
Definition: IKernel.cpp:28
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:489
TensorShape shape
Shape of the valid region.
Definition: Types.h:255
virtual size_t dimension(size_t index) const =0
Return the size of the requested dimension.
Container for 2D border size.
Definition: Types.h:267
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:276
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:46
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's metadata.
Definition: ITensorInfo.h:40
Describe one of the image'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:377
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:46
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's parameters to the object's kernel's arguments starting from the index idx.
Definition: ICLKernel.h:172
void use_tensor_dimensions(const TensorShape &shape, size_t first_dimension=Window::DimX)
Use the tensor'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.
DataType dt
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:45
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:403
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:214
#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 kernel_name
std::string get_cl_type_from_data_type(const DataType &dt)
Translates a tensor data type to the appropriate OpenCL type.
Definition: CLHelpers.cpp:37
CLFillBorderKernel()
Default constructor.
virtual ITensorInfo * info() const =0
Interface to be implemented by the child class to return the tensor'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:378
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:376
#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:504
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:300
static constexpr size_t DimZ
Alias for dimension 2 also known as Z dimension.
Definition: Window.h:47
Borders are left undefined.
Manages all the OpenCL kernels compilation and caching, provides accessors for the OpenCL Context.
Wrapper to configure the Khronos OpenCL C++ header.
Tensor packing service.
Definition: ITensorPack.h:37
quantized, asymmetric fixed-point 8-bit number signed
Window first_slice_window_3D() const
First 3D slice of the window.
Definition: Window.h:291
DataType
Available data types.
Definition: Types.h:77
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:367
Coordinates anchor
Anchor for the start of the valid region.
Definition: Types.h:254
SimpleTensor< T > slice(const SimpleTensor< T > &src, Coordinates starts, Coordinates ends)