Compute Library
 21.11
CLBoundingBoxTransformKernel.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  */
25 
32 #include "arm_compute/core/Utils.h"
33 #include "src/core/CL/CLValidate.h"
36 #include "support/StringSupport.h"
37 
38 namespace arm_compute
39 {
40 namespace
41 {
42 Status validate_arguments(const ITensorInfo *boxes, const ITensorInfo *pred_boxes, const ITensorInfo *deltas, const BoundingBoxTransformInfo &info)
43 {
44  ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(boxes, pred_boxes, deltas);
48  ARM_COMPUTE_RETURN_ERROR_ON(deltas->tensor_shape()[1] != boxes->tensor_shape()[1]);
49  ARM_COMPUTE_RETURN_ERROR_ON(deltas->tensor_shape()[0] % 4 != 0);
50  ARM_COMPUTE_RETURN_ERROR_ON(boxes->tensor_shape()[0] != 4);
51  ARM_COMPUTE_RETURN_ERROR_ON(deltas->num_dimensions() > 2);
52  ARM_COMPUTE_RETURN_ERROR_ON(boxes->num_dimensions() > 2);
53 
54  const bool is_qasymm16 = boxes->data_type() == DataType::QASYMM16;
55  if(is_qasymm16)
56  {
57  const UniformQuantizationInfo boxes_qinfo = boxes->quantization_info().uniform();
58  ARM_COMPUTE_RETURN_ERROR_ON(boxes_qinfo.scale != 0.125f);
59  ARM_COMPUTE_RETURN_ERROR_ON(boxes_qinfo.offset != 0);
61  }
62  else
63  {
65  }
66 
67  if(pred_boxes->total_size() > 0)
68  {
69  ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DIMENSIONS(pred_boxes->tensor_shape(), deltas->tensor_shape());
71  ARM_COMPUTE_RETURN_ERROR_ON(pred_boxes->num_dimensions() > 2);
72  if(is_qasymm16)
73  {
74  const UniformQuantizationInfo pred_boxes_qinfo = pred_boxes->quantization_info().uniform();
75  ARM_COMPUTE_RETURN_ERROR_ON(pred_boxes_qinfo.scale != 0.125f);
76  ARM_COMPUTE_RETURN_ERROR_ON(pred_boxes_qinfo.offset != 0);
77  }
78  }
79  ARM_COMPUTE_RETURN_ERROR_ON(info.scale() <= 0);
80 
81  return Status{};
82 }
83 } // namespace
84 
86  : _boxes(nullptr), _pred_boxes(nullptr), _deltas(nullptr)
87 {
89 }
90 
91 void CLBoundingBoxTransformKernel::configure(const ICLTensor *boxes, ICLTensor *pred_boxes, const ICLTensor *deltas, const BoundingBoxTransformInfo &info)
92 {
93  configure(CLKernelLibrary::get().get_compile_context(), boxes, pred_boxes, deltas, info);
94 }
95 
96 void CLBoundingBoxTransformKernel::configure(const CLCompileContext &compile_context, const ICLTensor *boxes, ICLTensor *pred_boxes, const ICLTensor *deltas, const BoundingBoxTransformInfo &info)
97 {
98  ARM_COMPUTE_ERROR_ON_NULLPTR(boxes, pred_boxes, deltas);
99  auto padding_info = get_padding_info({ boxes, pred_boxes, deltas });
100  auto_init_if_empty(*pred_boxes->info(), deltas->info()->clone()->set_data_type(boxes->info()->data_type()).set_quantization_info(boxes->info()->quantization_info()));
101 
102  ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(boxes->info(), pred_boxes->info(), deltas->info(), info));
103 
104  // Set instance variables
105  _boxes = boxes;
106  _pred_boxes = pred_boxes;
107  _deltas = deltas;
108 
109  // Get image height and widht (rescaled)
110  const int img_h = floor(info.img_height() / info.scale() + 0.5f);
111  const int img_w = floor(info.img_width() / info.scale() + 0.5f);
112 
113  const bool is_quantized = is_data_type_quantized(boxes->info()->data_type());
114 
115  // Set build options
116  CLBuildOptions build_opts;
117  build_opts.add_option("-DDATA_TYPE=" + get_cl_type_from_data_type(boxes->info()->data_type()));
118  build_opts.add_option("-DWEIGHT_X=" + float_to_string_with_full_precision(info.weights()[0]));
119  build_opts.add_option("-DWEIGHT_Y=" + float_to_string_with_full_precision(info.weights()[1]));
120  build_opts.add_option("-DWEIGHT_W=" + float_to_string_with_full_precision(info.weights()[2]));
121  build_opts.add_option("-DWEIGHT_H=" + float_to_string_with_full_precision(info.weights()[3]));
122  build_opts.add_option("-DBBOX_XFORM_CLIP=" + float_to_string_with_full_precision(info.bbox_xform_clip()));
123  build_opts.add_option("-DIMG_WIDTH=" + support::cpp11::to_string(img_w));
124  build_opts.add_option("-DIMG_HEIGHT=" + support::cpp11::to_string(img_h));
125  build_opts.add_option("-DBOX_FIELDS=" + support::cpp11::to_string(4));
126  build_opts.add_option("-DSCALE_BEFORE=" + float_to_string_with_full_precision(info.scale()));
127  build_opts.add_option_if(info.apply_scale(), "-DSCALE_AFTER=" + float_to_string_with_full_precision(info.scale()));
128  build_opts.add_option_if(info.correct_transform_coords(), "-DOFFSET=1");
129 
130  if(is_quantized)
131  {
132  build_opts.add_option("-DDATA_TYPE_DELTAS=" + get_cl_type_from_data_type(deltas->info()->data_type()));
133  const UniformQuantizationInfo boxes_qinfo = boxes->info()->quantization_info().uniform();
134  const UniformQuantizationInfo deltas_qinfo = deltas->info()->quantization_info().uniform();
135  const UniformQuantizationInfo pred_boxes_qinfo = pred_boxes->info()->quantization_info().uniform();
136  build_opts.add_option("-DOFFSET_BOXES=" + float_to_string_with_full_precision(boxes_qinfo.offset));
137  build_opts.add_option("-DSCALE_BOXES=" + float_to_string_with_full_precision(boxes_qinfo.scale));
138  build_opts.add_option("-DOFFSET_DELTAS=" + float_to_string_with_full_precision(deltas_qinfo.offset));
139  build_opts.add_option("-DSCALE_DELTAS=" + float_to_string_with_full_precision(deltas_qinfo.scale));
140  build_opts.add_option("-DOFFSET_PRED_BOXES=" + float_to_string_with_full_precision(pred_boxes_qinfo.offset));
141  build_opts.add_option("-DSCALE_PRED_BOXES=" + float_to_string_with_full_precision(pred_boxes_qinfo.scale));
142  }
143 
144  // Create kernel
145  const std::string kernel_name = (is_quantized) ? "bounding_box_transform_quantized" : "bounding_box_transform";
146  _kernel = create_kernel(compile_context, kernel_name, build_opts.options());
147 
148  // Since the number of columns is a multiple of 4 by definition, we don't need to pad the tensor
149  const unsigned int num_elems_processed_per_iteration = 4;
150  Window win = calculate_max_window(*deltas->info(), Steps(num_elems_processed_per_iteration));
151  ICLKernel::configure_internal(win);
153 }
154 
156 {
157  ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments(boxes, pred_boxes, deltas, info));
158  return Status{};
159 }
160 
161 void CLBoundingBoxTransformKernel::run(const Window &window, cl::CommandQueue &queue)
162 {
165 
167 
168  // Set arguments
169  unsigned int idx = 0;
170  add_1D_tensor_argument(idx, _boxes, slice);
171  add_2D_tensor_argument(idx, _pred_boxes, slice);
172  add_2D_tensor_argument(idx, _deltas, slice);
173 
174  // Note that we don't need to loop over the slices, as we are sure that we are dealing with all 2D tensors
175  enqueue(queue, *this, slice, lws_hint());
176 }
177 } // namespace arm_compute
bool is_data_type_quantized(DataType dt)
Check if a given data type is of quantized type.
Definition: Utils.h:981
Window first_slice_window_2D() const
First 2D slice of the window.
Definition: Window.h:283
Window calculate_max_window(const ValidRegion &valid_region, const Steps &steps, bool skip_border, BorderSize border_size)
#define ARM_COMPUTE_RETURN_ERROR_ON_F16_UNSUPPORTED(tensor)
Definition: CLValidate.h:35
const Window & window() const
The maximum window the kernel can be executed on.
Definition: IKernel.cpp:28
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
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.
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
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
Quantization info when assuming per layer quantization.
quantized, asymmetric fixed-point 16-bit number
Manages all the OpenCL kernels compilation and caching, provides accessors for the OpenCL Context...
Status class.
Definition: Error.h:52
#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.
1 channel, 1 F16 per channel
#define ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(...)
Definition: Validate.h:159
void add_option(std::string option)
Adds option to the existing build option list.
void configure(const ICLTensor *boxes, ICLTensor *pred_boxes, const ICLTensor *deltas, const BoundingBoxTransformInfo &info)
Set the input and output tensors.
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
std::string float_to_string_with_full_precision(float val)
Create a string with the float in full precision.
Definition: Utils.h:1075
quantized, asymmetric fixed-point 8-bit number unsigned
Class to describe a number of elements in each dimension.
Definition: Steps.h:40
unsigned int num_elems_processed_per_iteration
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:39
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 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&#39;s metadata.
Bounding Box Transform information class.
Definition: Types.h:1442
Elementeise CL kernel type.
Definition: CLTypes.h:84
virtual QuantizationInfo quantization_info() const =0
Get the quantization settings (scale and offset) of the tensor.
#define ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL(k)
Definition: Validate.h:915
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.
static Status validate(const ITensorInfo *boxes, const ITensorInfo *pred_boxes, const ITensorInfo *deltas, const BoundingBoxTransformInfo &info)
Static function to check if given info will lead to a valid configuration of CLBoundingBoxTransform.
void run(const Window &window, cl::CommandQueue &queue) override
Enqueue the OpenCL kernel to process the given window on the passed OpenCL command queue...
void add_2D_tensor_argument(unsigned int &idx, const ICLTensor *tensor, const Window &window)
Add the passed 2D tensor&#39;s parameters to the object&#39;s kernel&#39;s arguments starting from the index idx...
Definition: ICLKernel.h:190
ScaleKernelInfo info(interpolation_policy, default_border_mode, PixelValue(), sampling_policy, false)
Interface for OpenCL tensor.
Definition: ICLTensor.h:42
#define ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(...)
Definition: Validate.h:541
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.
std::array< float, 4 > weights() const
Definition: Types.h:1463
#define ARM_COMPUTE_ERROR_ON_NULLPTR(...)
Definition: Validate.h:157
#define ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_NOT_IN(t,...)
Definition: Validate.h:690
void add_1D_tensor_argument(unsigned int &idx, const ICLTensor *tensor, const Window &window)
Add the passed 1D tensor&#39;s parameters to the object&#39;s kernel&#39;s arguments starting from the index idx...
Definition: ICLKernel.h:166
std::string kernel_name
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)