Compute Library
 20.02.1
CLComparisonKernel.cpp
Go to the documentation of this file.
1 /*
2  * Copyright (c) 2018-2020 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 
29 
30 #include <map>
31 
32 namespace arm_compute
33 {
34 namespace
35 {
36 // Create supported comparisons map
37 const std::map<ComparisonOperation, std::string> supported_comparison_ops =
38 {
39  { ComparisonOperation::Equal, "EQUAL" },
40  { ComparisonOperation::NotEqual, "NOTEQUAL" },
41  { ComparisonOperation::Greater, "GREATER" },
42  { ComparisonOperation::GreaterEqual, "GREATEREQUAL" },
43  { ComparisonOperation::Less, "LESS" },
44  { ComparisonOperation::LessEqual, "LESSEQUAL" },
45 };
46 
47 int calculate_num_elems_processed_per_iteration(const ITensorInfo &input)
48 {
49  return 16 / input.element_size();
50 }
51 
52 Status validate_arguments(const ITensorInfo &input1, const ITensorInfo &input2, const ITensorInfo &output, ComparisonOperation operation)
53 {
55  ARM_COMPUTE_RETURN_ERROR_ON(input1.data_type() == DataType::UNKNOWN);
57  ARM_COMPUTE_RETURN_ERROR_ON(supported_comparison_ops.count(operation) == 0);
58 
59  const TensorShape out_shape = TensorShape::broadcast_shape(input1.tensor_shape(), input2.tensor_shape());
60  ARM_COMPUTE_RETURN_ERROR_ON_MSG(out_shape.total_size() == 0, "Inputs are not broadcast compatible");
61 
62  // Validate in case of configured output
63  if(output.total_size() > 0)
64  {
66  ARM_COMPUTE_RETURN_ERROR_ON_MSG(detail::have_different_dimensions(out_shape, output.tensor_shape(), 0),
67  "Wrong shape for output");
68  }
69 
70  return Status{};
71 }
72 
73 std::pair<Status, Window> validate_and_configure_window(ITensorInfo &input1, ITensorInfo &input2, ITensorInfo &output)
74 {
75  const std::pair<TensorShape, ValidRegion> broadcast_pair = ITensorInfo::broadcast_shape_and_valid_region(input1, input2);
76  const TensorShape &out_shape = broadcast_pair.first;
77  const ValidRegion &valid_region = broadcast_pair.second;
78 
79  const unsigned int num_elems_processed_per_iteration = calculate_num_elems_processed_per_iteration(input1);
80 
81  // Auto initialize output if not initialized
82  auto_init_if_empty(output, out_shape, 1, DataType::U8, QuantizationInfo());
83 
85  Window win_input1 = win.broadcast_if_dimension_le_one(input1);
86  Window win_input2 = win.broadcast_if_dimension_le_one(input2);
87 
88  AccessWindowHorizontal input1_access(&input1, 0, num_elems_processed_per_iteration);
89  AccessWindowHorizontal input2_access(&input2, 0, num_elems_processed_per_iteration);
90  AccessWindowHorizontal output_access(&output, 0, num_elems_processed_per_iteration);
91 
92  bool window_changed = update_window_and_padding(win_input1, input1_access)
93  || update_window_and_padding(win_input2, input2_access)
94  || update_window_and_padding(win, output_access);
95 
96  output_access.set_valid_region(win, valid_region);
97 
98  Status err = (window_changed) ? ARM_COMPUTE_CREATE_ERROR(ErrorCode::RUNTIME_ERROR, "Insufficient Padding!") : Status{};
99  return std::make_pair(err, win);
100 }
101 } // namespace
102 
104  : _input1(nullptr), _input2(nullptr), _output(nullptr)
105 {
106 }
107 
108 void CLComparisonKernel::configure(const ICLTensor *input1, const ICLTensor *input2, ICLTensor *output, ComparisonOperation operation)
109 {
110  ARM_COMPUTE_ERROR_ON_NULLPTR(input1, input2, output);
111  ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(*input1->info(), *input2->info(), *output->info(), operation));
112 
113  // Configure kernel window
114  auto win_config = validate_and_configure_window(*input1->info(), *input2->info(), *output->info());
115  ARM_COMPUTE_ERROR_THROW_ON(win_config.first);
116 
117  _input1 = input1;
118  _input2 = input2;
119  _output = output;
120 
121  const std::string &operation_name = supported_comparison_ops.at(operation);
122  std::string kernel_name = "compare_" + lower_string(operation_name);
123 
124  // Set kernel build options
125  std::set<std::string> build_opts;
126  build_opts.emplace("-DDATA_TYPE=" + get_cl_type_from_data_type(input1->info()->data_type()));
127  build_opts.emplace("-DVEC_SIZE=" + support::cpp11::to_string(calculate_num_elems_processed_per_iteration(*input1->info())));
128  build_opts.emplace("-DOP=" + operation_name);
129  build_opts.emplace("-DOP_NAME=" + lower_string(operation_name));
130  if(is_data_type_quantized(input1->info()->data_type()))
131  {
132  const UniformQuantizationInfo iq1_info = input1->info()->quantization_info().uniform();
133  const UniformQuantizationInfo iq2_info = input2->info()->quantization_info().uniform();
134 
135  build_opts.emplace("-DOFFSET_IN1=" + support::cpp11::to_string(iq1_info.offset));
136  build_opts.emplace("-DOFFSET_IN2=" + support::cpp11::to_string(iq2_info.offset));
137  build_opts.emplace("-DSCALE_IN1=" + float_to_string_with_full_precision(iq1_info.scale));
138  build_opts.emplace("-DSCALE_IN2=" + float_to_string_with_full_precision(iq2_info.scale));
139  kernel_name += "_quantized";
140  }
141 
142  // Create kernel
143  _kernel = static_cast<cl::Kernel>(CLKernelLibrary::get().create_kernel(kernel_name, build_opts));
144 
145  ICLKernel::configure_internal(win_config.second);
146 
147  // Set config_id for enabling LWS tuning
148  _config_id = kernel_name;
149  _config_id += "_";
150  _config_id += lower_string(string_from_data_type(input1->info()->data_type()));
151  _config_id += "_";
152  _config_id += support::cpp11::to_string(output->info()->dimension(0));
153  _config_id += "_";
154  _config_id += support::cpp11::to_string(output->info()->dimension(1));
155  _config_id += lower_string(string_from_data_layout(input1->info()->data_layout()));
156 }
157 
158 Status CLComparisonKernel::validate(const ITensorInfo *input1, const ITensorInfo *input2, const ITensorInfo *output, ComparisonOperation operation)
159 {
160  ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(input1, input2, output);
161 
162  ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments(*input1, *input2, *output, operation));
163  ARM_COMPUTE_RETURN_ON_ERROR(validate_and_configure_window(*input1->clone(), *input2->clone(), *output->clone()).first);
164 
165  return Status{};
166 }
167 
168 void CLComparisonKernel::run(const Window &window, cl::CommandQueue &queue)
169 {
172 
173  const TensorShape &in_shape1 = _input1->info()->tensor_shape();
174  const TensorShape &in_shape2 = _input2->info()->tensor_shape();
175  const TensorShape &out_shape = _output->info()->tensor_shape();
176 
177  bool can_collapse = true;
178  const bool is_vector = in_shape1.num_dimensions() == 1 || in_shape2.num_dimensions() == 1;
179  if(std::min(in_shape1.total_size(), in_shape2.total_size()) > 1 && !is_vector)
180  {
181  can_collapse = (std::min(in_shape1.num_dimensions(), in_shape2.num_dimensions()) > Window::DimZ);
182  for(size_t d = Window::DimZ; can_collapse && (d < out_shape.num_dimensions()); d++)
183  {
184  can_collapse = (in_shape1[d] == in_shape2[d]);
185  }
186  }
187 
188  bool has_collapsed = false;
189  Window collapsed = can_collapse ? window.collapse_if_possible(ICLKernel::window(), Window::DimZ, &has_collapsed) : window;
190 
191  const TensorShape &in_shape1_collapsed = has_collapsed ? in_shape1.collapsed_from(Window::DimZ) : in_shape1;
192  const TensorShape &in_shape2_collapsed = has_collapsed ? in_shape2.collapsed_from(Window::DimZ) : in_shape2;
193 
194  Window slice = collapsed.first_slice_window_3D();
195  Window slice_input1 = slice.broadcast_if_dimension_le_one(in_shape1_collapsed);
196  Window slice_input2 = slice.broadcast_if_dimension_le_one(in_shape2_collapsed);
197 
198  do
199  {
200  unsigned int idx = 0;
201 
202  add_3D_tensor_argument(idx, _input1, slice_input1);
203  add_3D_tensor_argument(idx, _input2, slice_input2);
204  add_3D_tensor_argument(idx, _output, slice);
205 
206  enqueue(queue, *this, slice, lws_hint());
207 
208  ARM_COMPUTE_UNUSED(collapsed.slide_window_slice_3D(slice_input1));
209  ARM_COMPUTE_UNUSED(collapsed.slide_window_slice_3D(slice_input2));
210  }
211  while(collapsed.slide_window_slice_3D(slice));
212 }
213 
215 {
216  const int num_elems_processed_per_iteration = calculate_num_elems_processed_per_iteration(*_input1->info());
217 
218  const unsigned int replicateSize = _output->info()->dimension(0) - std::min(_input1->info()->dimension(0), _input2->info()->dimension(0));
219  const unsigned int border = std::min<unsigned int>(num_elems_processed_per_iteration - 1U, replicateSize);
220  return BorderSize{ 0, border, 0, 0 };
221 }
222 } // namespace arm_compute
bool is_data_type_quantized(DataType dt)
Check if a given data type is of quantized type.
Definition: Utils.h:1117
#define ARM_COMPUTE_RETURN_ERROR_ON_F16_UNSUPPORTED(tensor)
Definition: CLValidate.h:34
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.
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:39
#define ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(...)
Definition: Validate.h:545
cl::NDRange lws_hint() const
Return the Local-Workgroup-Size hint.
Definition: ICLKernel.h:247
TensorShape collapsed_from(size_t start) const
Return a copy with collapsed dimensions starting from a given point.
Definition: TensorShape.h:160
1 channel, 1 U8 per channel
#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.
#define ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(t, c,...)
Definition: Validate.h:792
static TensorShape broadcast_shape(const Shapes &... shapes)
If shapes are broadcast compatible, return the broadcasted shape.
Definition: TensorShape.h:210
void configure(const ICLTensor *input1, const ICLTensor *input2, ICLTensor *output, ComparisonOperation operation)
Set the inputs and output tensors.
static CLKernelLibrary & get()
Access the KernelLibrary singleton.
Store the tensor's metadata.
Definition: ITensorInfo.h:40
#define ARM_COMPUTE_ERROR_THROW_ON(status)
Definition: Error.h:455
Quantization info when assuming per layer quantization.
Status class.
Definition: Error.h:52
std::string lower_string(const std::string &val)
Lower a given string.
Definition: Utils.cpp:333
#define ARM_COMPUTE_RETURN_ERROR_ON(cond)
If the condition is true, an error is returned.
Definition: Error.h:296
static std::pair< TensorShape, ValidRegion > broadcast_shape_and_valid_region(const Infos &... infos)
If infos are broadcast compatible tensor info's, return the broadcasted shape and the intersection of...
Definition: ITensorInfo.h:259
Window calculate_max_window(const ValidRegion &valid_region, const Steps &steps=Steps(), bool skip_border=false, BorderSize border_size=BorderSize())
Calculate the maximum window for a given tensor shape and border setting.
Definition: Helpers.cpp:28
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:158
Copyright (c) 2017-2020 ARM Limited.
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: Helpers.inl:202
const std::string & string_from_data_type(DataType dt)
Convert a data type identity into a string.
Definition: Utils.cpp:144
bool update_window_and_padding(Window &win, Ts &&... patterns)
Update window and padding size for each of the access patterns.
Definition: Helpers.h:402
#define ARM_COMPUTE_UNUSED(...)
To avoid unused variables warnings.
Definition: Error.h:152
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
std::string float_to_string_with_full_precision(float val)
Create a string with the float in full precision.
Definition: Utils.h:1211
virtual const TensorShape & tensor_shape() const =0
Size for each dimension of the tensor.
CLComparisonKernel()
Default constructor.
std::string kernel_name
size_t total_size() const
Collapses all dimensions to a single linear total size.
Definition: TensorShape.h:171
UniformQuantizationInfo uniform() const
Return per layer quantization info.
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
virtual std::unique_ptr< T > clone() const =0
Provide a clone of the current object of class T.
virtual ITensorInfo * info() const =0
Interface to be implemented by the child class to return the tensor's metadata.
bool have_different_dimensions(const Dimensions< T > &dim1, const Dimensions< T > &dim2, unsigned int upper_dim)
Definition: Validate.h:51
ComparisonOperation
Supported comparison operations.
Definition: Types.h:173
std::unique_ptr< Kernel > create_kernel()
Helper function to create and return a unique_ptr pointed to a CL/GLES kernel object.
Definition: Helpers.h:86
void run(const Window &window, cl::CommandQueue &queue) override
Enqueue the OpenCL kernel to process the given window on the passed OpenCL command queue.
bool slide_window_slice_3D(Window &slice) const
Slide the passed 3D window slice.
Definition: Window.h:333
virtual QuantizationInfo quantization_info() const =0
Get the quantization settings (scale and offset) of the tensor.
#define ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(...)
Definition: Validate.h:163
#define ARM_COMPUTE_ERROR_ON_NULLPTR(...)
Definition: Validate.h:161
Interface for OpenCL tensor.
Definition: ICLTensor.h:42
const std::string & string_from_data_layout(DataLayout dl)
Convert a data layout identity into a string.
Definition: Utils.cpp:132
#define ARM_COMPUTE_CREATE_ERROR(error_code, msg)
Creates an error with a given message.
Definition: Error.h:159
static constexpr size_t DimZ
Alias for dimension 2 also known as Z dimension.
Definition: Window.h:47
unsigned int num_dimensions() const
Returns the effective dimensionality of the tensor.
Definition: Dimensions.h:122
static Status validate(const ITensorInfo *input1, const ITensorInfo *input2, const ITensorInfo *output, ComparisonOperation operation)
Static function to check if given info will lead to a valid configuration of CLComparisonKernel.
unsigned int num_elems_processed_per_iteration
#define ARM_COMPUTE_RETURN_ERROR_ON_MSG(cond, msg)
If the condition is true, an error is returned.
Definition: Error.h:244
Window first_slice_window_3D() const
First 3D slice of the window.
Definition: Window.h:289
#define ARM_COMPUTE_ERROR_ON_INVALID_SUBWINDOW(f, s)
Definition: Validate.h:205
BorderSize border_size() const override
The size of the border for that kernel.
Describe a multidimensional execution window.
Definition: Window.h:39
#define ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL(k)
Definition: Validate.h:941
SimpleTensor< T > slice(const SimpleTensor< T > &src, Coordinates starts, Coordinates ends)
virtual DataLayout data_layout() const =0
Get the data layout of the tensor.