Compute Library
 21.11
CLTileKernel.cpp
Go to the documentation of this file.
1 /*
2  * Copyright (c) 2018-2021 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  */
29 #include "support/StringSupport.h"
30 
31 namespace arm_compute
32 {
33 namespace
34 {
35 Status validate_arguments(const ITensorInfo *input, const ITensorInfo *output, const Multiples &multiples)
36 {
38  ARM_COMPUTE_RETURN_ERROR_ON(input->data_type() == DataType::UNKNOWN);
39  ARM_COMPUTE_RETURN_ERROR_ON(multiples.size() > 4);
40  ARM_COMPUTE_RETURN_ERROR_ON(multiples.empty());
41  ARM_COMPUTE_RETURN_ERROR_ON(std::any_of(multiples.begin(), multiples.end(), [](uint32_t e)
42  {
43  return e == 0;
44  }));
45 
46  // Validate output if initialized
47  if(output->total_size() != 0)
48  {
49  ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DIMENSIONS(misc::shape_calculator::compute_tiled_shape(input->tensor_shape(), multiples), output->tensor_shape());
51  }
52 
53  return Status{};
54 }
55 } // namespace
56 
58  : _input(nullptr), _output(nullptr)
59 {
61 }
62 
63 void CLTileKernel::configure(const ICLTensor *input, ICLTensor *output, const Multiples &multiples)
64 {
65  configure(CLKernelLibrary::get().get_compile_context(), input, output, multiples);
66 }
67 
68 void CLTileKernel::configure(const CLCompileContext &compile_context, const ICLTensor *input, ICLTensor *output, const Multiples &multiples)
69 {
70  ARM_COMPUTE_ERROR_ON_NULLPTR(input, output);
71 
72  // Auto initialize output
73  TensorShape tiled_shape = misc::shape_calculator::compute_tiled_shape(input->info()->tensor_shape(), multiples);
74  auto_init_if_empty(*output->info(), tiled_shape, 1, input->info()->data_type());
75 
76  // Validate
77  ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(input->info(), output->info(), multiples));
78 
79  _input = input;
80  _output = output;
81 
82  const DataType data_type = input->info()->data_type();
83  const int vec_size_x = 16 / input->info()->element_size();
84  const int input_width_x = input->info()->tensor_shape().x();
85  const unsigned int offset = ceil_to_multiple(input_width_x, vec_size_x) - input_width_x;
86  const bool multi_access_x = (input_width_x / vec_size_x > 0);
87 
88  // Create kernel
89  CLBuildOptions build_opts;
90  build_opts.add_option("-DDATA_TYPE=" + get_cl_unsigned_type_from_element_size(data_size_from_type(data_type)));
91  build_opts.add_option("-DSRC_WIDTH=" + support::cpp11::to_string(input_width_x));
92  build_opts.add_option("-DSRC_HEIGHT=" + support::cpp11::to_string(input->info()->dimension(1)));
93  build_opts.add_option("-DSRC_DEPTH=" + support::cpp11::to_string(input->info()->dimension(2)));
94  build_opts.add_option("-DSRC_BATCHES=" + support::cpp11::to_string(input->info()->dimension(3)));
95  build_opts.add_option("-DDST_DEPTH=" + support::cpp11::to_string(output->info()->dimension(2)));
96  build_opts.add_option_if(multi_access_x, "-DOFFSET=" + support::cpp11::to_string(offset));
97  build_opts.add_option_if(multi_access_x, "-DVEC_SIZE=" + support::cpp11::to_string(vec_size_x));
98  _kernel = create_kernel(compile_context, "tile", build_opts.options());
99 
100  // Configure window without padding
101  Window win = calculate_max_window(*output->info());
102 
103  if(multi_access_x)
104  {
105  // If multi-access is enabled, no thread should cross the tile boundaries. This means we need
106  // as many threads as those to cover a single tile times multiples[0]. Note that if threads
107  // do not cross the boundaries of the tiles, they won't cross the boundaries of the last tile, and
108  // we don't need to pad the output
109  const unsigned int size_win_x = ceil_to_multiple(input->info()->dimension(0), vec_size_x) * multiples[0];
110  win.set(Window::DimX,
111  Window::Dimension(win.x().start(), size_win_x, vec_size_x));
112  }
113 
114  ICLKernel::configure_internal(win);
115 
116  // Set config_id for enabling LWS tuning
117  _config_id = "tile";
118  _config_id += "_";
119  _config_id += lower_string(string_from_data_type(input->info()->data_type()));
120  for(unsigned int i = 0; i < multiples.size(); ++i)
121  {
122  _config_id += "_";
123  _config_id += support::cpp11::to_string(input->info()->dimension(i));
124  _config_id += "_";
125  _config_id += support::cpp11::to_string(multiples[i]);
126  }
127 }
128 
129 Status CLTileKernel::validate(const ITensorInfo *input, const ITensorInfo *output, const Multiples &multiples)
130 {
131  ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments(input, output, multiples));
132  return Status{};
133 }
134 
135 void CLTileKernel::run(const Window &window, cl::CommandQueue &queue)
136 {
139 
141  Window slice = collapsed.first_slice_window_4D();
142 
143  do
144  {
145  unsigned int idx = 0;
146  add_4D_tensor_argument(idx, _input, slice);
147  add_4D_tensor_argument(idx, _output, slice);
148  enqueue(queue, *this, slice, lws_hint());
149  }
150  while(collapsed.slide_window_slice_4D(slice));
151 }
152 } // namespace arm_compute
__global uchar * offset(const Image *img, int x, int y)
Get the pointer position of a Image.
Definition: helpers.h:1069
CLTileKernel()
Default constructor.
Window calculate_max_window(const ValidRegion &valid_region, const Steps &steps, bool skip_border, BorderSize border_size)
const Window & window() const
The maximum window the kernel can be executed on.
Definition: IKernel.cpp:28
Shape of a tensor.
Definition: TensorShape.h:39
virtual size_t dimension(size_t index) const =0
Return the size of the requested dimension.
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.
cl::NDRange lws_hint() const
Return the Local-Workgroup-Size hint.
Definition: ICLKernel.h:318
#define ARM_COMPUTE_RETURN_ON_ERROR(status)
Checks if a status contains an error and returns it.
Definition: Error.h:204
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.
static CLKernelLibrary & get()
Access the KernelLibrary singleton.
Store the tensor&#39;s metadata.
Definition: ITensorInfo.h:40
#define ARM_COMPUTE_ERROR_THROW_ON(status)
Definition: Error.h:455
Describe one of the image&#39;s dimensions with a start, end and step.
Definition: Window.h:77
Status class.
Definition: Error.h:52
std::string lower_string(const std::string &val)
Lower a given string.
Definition: Utils.cpp:326
#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
Copyright (c) 2017-2021 Arm Limited.
#define ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(...)
Definition: Validate.h:159
void configure(const ICLTensor *input, ICLTensor *output, const Multiples &multiples)
Set the source, destination of the kernel.
void add_option(std::string option)
Adds option to the existing build option list.
T x() const
Alias to access the size of the first dimension.
Definition: Dimensions.h:87
const DataType data_type
Definition: Im2Col.cpp:150
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
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 DimX
Alias for dimension 0 also known as X dimension.
Definition: Window.h:43
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
virtual const TensorShape & tensor_shape() const =0
Size for each dimension of the tensor.
auto ceil_to_multiple(S value, T divisor) -> decltype(((value+divisor - 1)/divisor) *divisor)
Computes the smallest number larger or equal to value that is a multiple of divisor.
Definition: Utils.h:71
TensorShape compute_tiled_shape(const TensorShape &input_shape, const Multiples &multiples)
Calculate the tiled shape of a tensor.
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 ITensorInfo * info() const =0
Interface to be implemented by the child class to return the tensor&#39;s metadata.
size_t data_size_from_type(DataType data_type)
The size in bytes of the data type.
Definition: Utils.h:106
void add_option_if(bool cond, std::string option)
Adds option if a given condition is true;.
virtual size_t element_size() const =0
Element size in bytes calculated as data_size() * num_channels()
Elementeise CL kernel type.
Definition: CLTypes.h:84
#define ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL(k)
Definition: Validate.h:915
CLCompileContext class.
Interface for OpenCL tensor.
Definition: ICLTensor.h:42
static constexpr size_t DimZ
Alias for dimension 2 also known as Z dimension.
Definition: Window.h:47
#define ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(...)
Definition: Validate.h:541
std::string get_cl_unsigned_type_from_element_size(size_t element_size)
Translates the element size to an unsigned integer data type.
Definition: CLHelpers.cpp:105
Window first_slice_window_4D() const
First 4D slice of the window.
Definition: Window.h:299
bool slide_window_slice_4D(Window &slice) const
Slide the passed 4D window slice.
Definition: Window.h:347
#define ARM_COMPUTE_ERROR_ON_NULLPTR(...)
Definition: Validate.h:157
static Status validate(const ITensorInfo *input, const ITensorInfo *output, const Multiples &multiples)
Static function to check if given info will lead to a valid configuration of CLTileKernel.
DataType
Available data types.
Definition: Types.h:79
void add_4D_tensor_argument(unsigned int &idx, const ICLTensor *tensor, const Window &window)
Add the passed 4D tensor&#39;s parameters to the object&#39;s kernel&#39;s arguments starting from the index idx...
Definition: ICLKernel.h:224
std::vector< uint32_t > Multiples
Information to produce a tiled version of a Tensor.
Definition: Types.h:443
Describe a multidimensional execution window.
Definition: Window.h:39
#define ARM_COMPUTE_ERROR_ON_INVALID_SUBWINDOW(f, s)
Definition: Validate.h:201
SimpleTensor< T > slice(const SimpleTensor< T > &src, Coordinates starts, Coordinates ends)