Compute Library
 21.08
CLROIAlignLayerKernel.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"
34 #include "src/core/CL/CLValidate.h"
37 #include "support/StringSupport.h"
38 
40 
41 namespace arm_compute
42 {
43 namespace
44 {
45 Status validate_arguments(const ITensorInfo *input, const ITensorInfo *rois, ITensorInfo *output, const ROIPoolingLayerInfo &pool_info)
46 {
47  ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(input, rois, output);
48  ARM_COMPUTE_RETURN_ERROR_ON(rois->dimension(0) != 5);
49  ARM_COMPUTE_RETURN_ERROR_ON(rois->num_dimensions() > 2);
53  ARM_COMPUTE_RETURN_ERROR_ON((pool_info.pooled_width() == 0) || (pool_info.pooled_height() == 0));
54 
55  if(output->total_size() != 0)
56  {
59  ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DIMENSIONS(compute_roi_align_shape(*input, *rois, pool_info), output->tensor_shape());
60  }
61 
62  if(is_data_type_quantized_asymmetric(input->data_type()))
63  {
65 
66  const UniformQuantizationInfo rois_qinfo = rois->quantization_info().uniform();
67  ARM_COMPUTE_RETURN_ERROR_ON(rois_qinfo.scale != 0.125f);
68  ARM_COMPUTE_RETURN_ERROR_ON(rois_qinfo.offset != 0);
69  }
70  else
71  {
73  }
74  return Status{};
75 }
76 
77 } // namespace
78 
80  : _input(nullptr), _output(nullptr), _rois(nullptr), _pool_info(0, 0, 0.f)
81 {
83 }
84 
85 void CLROIAlignLayerKernel::configure(const ICLTensor *input, const ICLTensor *rois, ICLTensor *output, const ROIPoolingLayerInfo &pool_info)
86 {
87  configure(CLKernelLibrary::get().get_compile_context(), input, rois, output, pool_info);
88 }
89 
90 void CLROIAlignLayerKernel::configure(const CLCompileContext &compile_context, const ICLTensor *input, const ICLTensor *rois, ICLTensor *output, const ROIPoolingLayerInfo &pool_info)
91 {
92  ARM_COMPUTE_ERROR_ON_NULLPTR(input, output, rois);
93  ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(input->info(), rois->info(), output->info(), pool_info));
94 
95  // Output auto inizialitation if not yet initialized
96  const TensorShape output_shape = compute_roi_align_shape(*input->info(), *rois->info(), pool_info);
97  auto_init_if_empty(*output->info(), output_shape, 1, input->info()->data_type());
98  output->info()->set_data_layout(input->info()->data_layout());
99 
100  auto padding_info = get_padding_info({ input, rois, output });
101 
102  _input = input;
103  _output = output;
104  _rois = rois;
105  _pool_info = pool_info;
106 
107  const DataType data_type = input->info()->data_type();
108  const bool is_qasymm = is_data_type_quantized_asymmetric(data_type);
109 
110  // Set build options
111  CLBuildOptions build_opts;
112  build_opts.add_option("-DDATA_TYPE=" + get_cl_type_from_data_type(data_type));
113  build_opts.add_option("-DDATA_SIZE=" + get_data_size_from_data_type(input->info()->data_type()));
114  build_opts.add_option("-DMAX_DIM_X=" + support::cpp11::to_string(_input->info()->dimension(get_data_layout_dimension_index(input->info()->data_layout(), DataLayoutDimension::WIDTH))));
115  build_opts.add_option("-DMAX_DIM_Y=" + support::cpp11::to_string(_input->info()->dimension(get_data_layout_dimension_index(input->info()->data_layout(), DataLayoutDimension::HEIGHT))));
116  build_opts.add_option("-DMAX_DIM_Z=" + support::cpp11::to_string(_input->info()->dimension(get_data_layout_dimension_index(input->info()->data_layout(), DataLayoutDimension::CHANNEL))));
117  build_opts.add_option("-DPOOLED_DIM_X=" + support::cpp11::to_string(pool_info.pooled_width()));
118  build_opts.add_option("-DPOOLED_DIM_Y=" + support::cpp11::to_string(pool_info.pooled_height()));
119  build_opts.add_option("-DSPATIAL_SCALE=" + float_to_string_with_full_precision(pool_info.spatial_scale()));
120  build_opts.add_option_if(input->info()->data_layout() == DataLayout::NHWC, "-DNHWC");
121  build_opts.add_option_if(pool_info.sampling_ratio() > 0, "-DSAMPLING_RATIO=" + support::cpp11::to_string(pool_info.sampling_ratio()));
122 
123  if(is_qasymm)
124  {
125  const UniformQuantizationInfo iq_info = input->info()->quantization_info().uniform();
126  const UniformQuantizationInfo roisq_info = rois->info()->quantization_info().uniform();
127  const UniformQuantizationInfo oq_info = output->info()->quantization_info().uniform();
128 
129  build_opts.add_option("-DOFFSET_IN=" + float_to_string_with_full_precision(iq_info.offset));
130  build_opts.add_option("-DSCALE_IN=" + float_to_string_with_full_precision(iq_info.scale));
131  build_opts.add_option("-DOFFSET_ROIS=" + float_to_string_with_full_precision(roisq_info.offset));
132  build_opts.add_option("-DSCALE_ROIS=" + float_to_string_with_full_precision(roisq_info.scale));
133  build_opts.add_option("-DOFFSET_OUT=" + float_to_string_with_full_precision(oq_info.offset));
134  build_opts.add_option("-DSCALE_OUT=" + float_to_string_with_full_precision(oq_info.scale));
135  }
136 
137  // Create kernel
138  const std::string kernel_name = (is_qasymm) ? "roi_align_layer_quantized" : "roi_align_layer";
139  _kernel = create_kernel(compile_context, kernel_name, build_opts.options());
140 
141  // Configure kernel window
142  Window win = calculate_max_window(*output->info(), Steps());
143  ICLKernel::configure_internal(win);
145 }
146 
148 {
149  ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments(input, rois, output, pool_info));
150  return Status{};
151 }
152 
153 void CLROIAlignLayerKernel::run(const Window &window, cl::CommandQueue &queue)
154 {
157 
159  Window slice_rois = slice;
160  // Parallelize spatially and across the fourth dimension of the output tensor (also across ROITensor)
161  slice_rois.set_dimension_step(Window::DimX, _rois->info()->dimension(0));
163 
164  // Set arguments
165  unsigned int idx = 0;
166  add_3D_tensor_argument(idx, _input, slice);
167  add_2D_tensor_argument(idx, _rois, slice_rois);
168  add_3D_tensor_argument(idx, _output, slice);
169  add_argument<cl_uint>(idx, _input->info()->strides_in_bytes()[3]);
170  add_argument<cl_uint>(idx, _output->info()->strides_in_bytes()[3]);
171 
172  enqueue(queue, *this, slice, lws_hint());
173 }
174 } // namespace arm_compute
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
Shape of a tensor.
Definition: TensorShape.h:39
#define ARM_COMPUTE_RETURN_ERROR_ON_DATA_LAYOUT_NOT_IN(t,...)
Definition: Validate.h:742
#define ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_LAYOUT(...)
Definition: Validate.h:490
virtual size_t dimension(size_t index) const =0
Return the size of the requested dimension.
TensorShape compute_roi_align_shape(const ITensorInfo &input, const ITensorInfo &rois, ROIPoolingLayerInfo pool_info)
Calculate the output roi align shape of a tensor.
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.
1 channel, 1 F32 per channel
void run(const Window &window, cl::CommandQueue &queue)
Enqueue the OpenCL kernel to process the given window on the passed OpenCL command queue...
#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...
unsigned int pooled_width() const
Get the pooled width of the layer.
Definition: Types.h:1249
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
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
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 *input, const ICLTensor *rois, ICLTensor *output, const ROIPoolingLayerInfo &pool_info)
Set the input and output tensors.
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
std::string get_data_size_from_data_type(const DataType &dt)
Get the size of a data type in number of bits.
Definition: CLHelpers.cpp:193
static constexpr size_t DimX
Alias for dimension 0 also known as X dimension.
Definition: Window.h:43
std::string float_to_string_with_full_precision(float val)
Create a string with the float in full precision.
Definition: Utils.h:1075
virtual ITensorInfo & set_data_layout(const DataLayout &data_layout)=0
Set the data layout of the tensor.
quantized, asymmetric fixed-point 8-bit number unsigned
Class to describe a number of elements in each dimension.
Definition: Steps.h:40
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 ITensorInfo * info() const =0
Interface to be implemented by the child class to return the tensor&#39;s metadata.
void add_option_if(bool cond, std::string option)
Adds option if a given condition is true;.
void set(size_t dimension, const Dimension &dim)
Set the values of a given dimension.
Definition: Window.inl:49
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
unsigned int sampling_ratio() const
Get sampling ratio.
Definition: Types.h:1264
Num samples, channels, height, width.
CLCompileContext class.
bool is_data_type_quantized_asymmetric(DataType dt)
Check if a given data type is of asymmetric quantized type.
Definition: Utils.h:1003
void set_dimension_step(size_t dimension, int step)
Set the step of a given dimension.
Definition: Window.inl:167
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
Interface for OpenCL tensor.
Definition: ICLTensor.h:42
unsigned int pooled_height() const
Get the pooled height of the layer.
Definition: Types.h:1254
ROI Pooling Layer Information class.
Definition: Types.h:1234
#define ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(...)
Definition: Validate.h:541
Num samples, height, width, channels.
#define ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(t, c,...)
Definition: Validate.h:788
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.
float spatial_scale() const
Get the spatial scale.
Definition: Types.h:1259
#define ARM_COMPUTE_ERROR_ON_NULLPTR(...)
Definition: Validate.h:157
quantized, asymmetric fixed-point 8-bit number signed
virtual const Strides & strides_in_bytes() const =0
The strides in bytes for accessing each dimension of the tensor.
size_t get_data_layout_dimension_index(const DataLayout data_layout, const DataLayoutDimension data_layout_dimension)
Get the index of the given dimension.
Definition: Helpers.inl:193
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:77
static Status validate(const ITensorInfo *input, const ITensorInfo *rois, ITensorInfo *output, const ROIPoolingLayerInfo &pool_info)
Static function to check if given info will lead to a valid configuration of CLROIAlignLayerKernel.
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)
virtual DataLayout data_layout() const =0
Get the data layout of the tensor.