Compute Library
 21.11
CLROIPoolingLayerKernel.cpp
Go to the documentation of this file.
1 /*
2  * Copyright (c) 2017-2021 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 #include <cfloat>
39 #include <cmath>
40 #include <string>
41 
42 namespace arm_compute
43 {
45  : _input(nullptr), _rois(nullptr), _output(nullptr), _pool_info(0, 0, 0.f)
46 {
48 }
49 
51 {
52  ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(input, rois, output);
53 
54  //Validate arguments
55  ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(input, rois, output);
61  ARM_COMPUTE_RETURN_ERROR_ON((pool_info.pooled_width() == 0) || (pool_info.pooled_height() == 0));
62 
63  if(output->total_size() != 0)
64  {
66  ARM_COMPUTE_RETURN_ERROR_ON((output->dimension(0) != pool_info.pooled_width()) || (output->dimension(1) != pool_info.pooled_height()));
67  ARM_COMPUTE_RETURN_ERROR_ON(input->dimension(2) != output->dimension(2));
68  ARM_COMPUTE_RETURN_ERROR_ON(rois->dimension(1) != output->dimension(3));
69  }
70 
71  return Status{};
72 }
73 
74 void CLROIPoolingLayerKernel::configure(const ICLTensor *input, const ICLTensor *rois, ICLTensor *output, const ROIPoolingLayerInfo &pool_info)
75 {
76  configure(CLKernelLibrary::get().get_compile_context(), input, rois, output, pool_info);
77 }
78 
79 void CLROIPoolingLayerKernel::configure(const CLCompileContext &compile_context, const ICLTensor *input, const ICLTensor *rois, const ICLTensor *output, const ROIPoolingLayerInfo &pool_info)
80 {
81  ARM_COMPUTE_ERROR_THROW_ON(CLROIPoolingLayerKernel::validate(input->info(), rois->info(), output->info(), pool_info));
82 
83  auto padding_info = get_padding_info({ input, rois, output });
84 
85  // Output auto initialization if not yet initialized
86  TensorShape output_shape(pool_info.pooled_width(), pool_info.pooled_height(), input->info()->dimension(2), rois->info()->dimension(1));
87  auto_init_if_empty(*(output->info()), output_shape, 1, input->info()->data_type(), output->info()->quantization_info());
88 
89  // Set instance variables
90  _input = input;
91  _rois = rois;
92  _output = output;
93  _pool_info = pool_info;
94 
95  const DataType data_type = input->info()->data_type();
96  const bool is_qasymm = is_data_type_quantized_asymmetric(data_type);
97 
98  // Set build options
99  CLBuildOptions build_opts;
100  build_opts.add_option("-DDATA_TYPE=" + get_cl_type_from_data_type(data_type));
101  build_opts.add_option("-DDATA_SIZE=" + get_data_size_from_data_type(data_type));
102  build_opts.add_option("-DMAX_DIM_X=" + support::cpp11::to_string(_input->info()->dimension(Window::DimX)));
103  build_opts.add_option("-DMAX_DIM_Y=" + support::cpp11::to_string(_input->info()->dimension(Window::DimY)));
104  build_opts.add_option("-DMAX_DIM_Z=" + support::cpp11::to_string(_input->info()->dimension(Window::DimZ)));
105  build_opts.add_option("-DPOOLED_DIM_X=" + support::cpp11::to_string(pool_info.pooled_width()));
106  build_opts.add_option("-DPOOLED_DIM_Y=" + support::cpp11::to_string(pool_info.pooled_height()));
107  build_opts.add_option("-DSPATIAL_SCALE=" + support::cpp11::to_string(pool_info.spatial_scale()));
108 
109  if(is_qasymm)
110  {
111  // Determine quantization info scale, offset
113  uqinfo = compute_requantization_scale_offset(_input->info()->quantization_info().uniform(), _output->info()->quantization_info().uniform());
114  build_opts.add_option("-DOFFSET_OUT=" + float_to_string_with_full_precision(uqinfo.offset));
115  build_opts.add_option("-DSCALE_OUT=" + float_to_string_with_full_precision(uqinfo.scale));
116 
117  // Specify minimum possible value of datatype
118  build_opts.add_option("-DMIN_VALUE=" + support::cpp11::to_string(0));
119  }
120  else
121  {
122  // Specify min value of F32 datatype
123  build_opts.add_option("-DMIN_VALUE=" + support::cpp11::to_string(-FLT_MAX));
124  }
125 
126  Window win = calculate_max_window(*(output->info()), Steps());
127  ICLKernel::configure_internal(win);
128 
129  // Create kernel
130  std::string kernel_name = "roi_pooling_layer";
131  _kernel = create_kernel(compile_context, kernel_name, build_opts.options());
132 
134 }
135 
136 void CLROIPoolingLayerKernel::run(const Window &window, cl::CommandQueue &queue)
137 {
140 
142  Window slice_rois = slice;
143  // Parallelize spatially and across the fourth dimension of the output tensor (also across ROITensor)
144  slice_rois.set_dimension_step(Window::DimX, _rois->info()->dimension(0));
145  slice.set(Window::DimZ, window[3]);
146 
147  // Set arguments
148  unsigned int idx = 0;
149  add_3D_tensor_argument(idx, _input, slice);
150  add_2D_tensor_argument(idx, _rois, slice_rois);
151  add_3D_tensor_argument(idx, _output, slice);
152  add_argument<cl_uint>(idx, _input->info()->strides_in_bytes()[3]);
153  add_argument<cl_uint>(idx, _output->info()->strides_in_bytes()[3]);
154 
155  enqueue(queue, *this, slice, lws_hint());
156 }
157 } // namespace arm_compute
virtual size_t num_dimensions() const =0
The number of dimensions of the tensor (rank)
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
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.
static Status validate(const ITensorInfo *input, const ITensorInfo *rois, const ITensorInfo *output, const ROIPoolingLayerInfo &pool_info)
Static Validate function to check inputs will lead to valid configuration of CLROIPoolingLayer.
cl::NDRange lws_hint() const
Return the Local-Workgroup-Size hint.
Definition: ICLKernel.h:318
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.
1 channel, 1 U16 per channel
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:1283
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
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
void run(const Window &window, cl::CommandQueue &queue) override
Enqueue the OpenCL kernel to process the given window on the passed OpenCL command queue...
#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.
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
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.
UniformQuantizationInfo compute_requantization_scale_offset(const UniformQuantizationInfo &uqinfo_in, const UniformQuantizationInfo &uqinfo_out)
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
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
static constexpr size_t DimY
Alias for dimension 1 also known as Y dimension.
Definition: Window.h:45
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
virtual size_t total_size() const =0
Returns the total size of the tensor in bytes.
static constexpr size_t DimZ
Alias for dimension 2 also known as Z dimension.
Definition: Window.h:47
unsigned int pooled_height() const
Get the pooled height of the layer.
Definition: Types.h:1288
ROI Pooling Layer Information class.
Definition: Types.h:1268
#define ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(...)
Definition: Validate.h:541
#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:1293
virtual const Strides & strides_in_bytes() const =0
The strides in bytes for accessing each dimension of the tensor.
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:79
void configure(const ICLTensor *input, const ICLTensor *rois, ICLTensor *output, const ROIPoolingLayerInfo &pool_info)
Set the input and output tensors.
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)