Compute Library
 21.11
CLGenerateProposalsLayerKernel.cpp
Go to the documentation of this file.
1 /*
2  * Copyright (c) 2019-2021 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 *anchors, const ITensorInfo *all_anchors, const ComputeAnchorsInfo &info)
43 {
44  ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(anchors, all_anchors);
46  ARM_COMPUTE_RETURN_ERROR_ON(anchors->dimension(0) != info.values_per_roi());
48  ARM_COMPUTE_RETURN_ERROR_ON(anchors->num_dimensions() > 2);
49  if(all_anchors->total_size() > 0)
50  {
51  size_t feature_height = info.feat_height();
52  size_t feature_width = info.feat_width();
53  size_t num_anchors = anchors->dimension(1);
55  ARM_COMPUTE_RETURN_ERROR_ON(all_anchors->num_dimensions() > 2);
56  ARM_COMPUTE_RETURN_ERROR_ON(all_anchors->dimension(0) != info.values_per_roi());
57  ARM_COMPUTE_RETURN_ERROR_ON(all_anchors->dimension(1) != feature_height * feature_width * num_anchors);
58 
59  if(is_data_type_quantized(anchors->data_type()))
60  {
62  }
63  }
64  return Status{};
65 }
66 } // namespace
67 
69  : _anchors(nullptr), _all_anchors(nullptr)
70 {
72 }
73 
74 void CLComputeAllAnchorsKernel::configure(const ICLTensor *anchors, ICLTensor *all_anchors, const ComputeAnchorsInfo &info)
75 {
76  configure(CLKernelLibrary::get().get_compile_context(), anchors, all_anchors, info);
77 }
78 
79 void CLComputeAllAnchorsKernel::configure(const CLCompileContext &compile_context, const ICLTensor *anchors, ICLTensor *all_anchors, const ComputeAnchorsInfo &info)
80 {
81  ARM_COMPUTE_ERROR_ON_NULLPTR(anchors, all_anchors);
82  auto padding_info = get_padding_info({ anchors, all_anchors });
83  ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(anchors->info(), all_anchors->info(), info));
84 
85  // Metadata
86  const size_t num_anchors = anchors->info()->dimension(1);
87  const DataType data_type = anchors->info()->data_type();
88  const float width = info.feat_width();
89  const float height = info.feat_height();
90 
91  // Initialize the output if empty
92  const TensorShape output_shape(info.values_per_roi(), width * height * num_anchors);
93  auto_init_if_empty(*all_anchors->info(), TensorInfo(output_shape, 1, data_type, anchors->info()->quantization_info()));
94 
95  // Set instance variables
96  _anchors = anchors;
97  _all_anchors = all_anchors;
98 
99  const bool is_quantized = is_data_type_quantized(anchors->info()->data_type());
100 
101  // Set build options
102  CLBuildOptions build_opts;
103  build_opts.add_option("-DDATA_TYPE=" + get_cl_type_from_data_type(data_type));
104  build_opts.add_option("-DWIDTH=" + float_to_string_with_full_precision(width));
105  build_opts.add_option("-DHEIGHT=" + float_to_string_with_full_precision(height));
106  build_opts.add_option("-DSTRIDE=" + float_to_string_with_full_precision(1.f / info.spatial_scale()));
107  build_opts.add_option("-DNUM_ANCHORS=" + support::cpp11::to_string(num_anchors));
108  build_opts.add_option("-DNUM_ROI_FIELDS=" + support::cpp11::to_string(info.values_per_roi()));
109 
110  if(is_quantized)
111  {
113  build_opts.add_option("-DSCALE=" + float_to_string_with_full_precision(qinfo.scale));
114  build_opts.add_option("-DOFFSET=" + float_to_string_with_full_precision(qinfo.offset));
115  }
116 
117  // Create kernel
118  const std::string kernel_name = (is_quantized) ? "generate_proposals_compute_all_anchors_quantized" : "generate_proposals_compute_all_anchors";
119  _kernel = create_kernel(compile_context, kernel_name, build_opts.options());
120 
121  // The tensor all_anchors can be interpreted as an array of structs (each structs has values_per_roi fields).
122  // This means we don't need to pad on the X dimension, as we know in advance how many fields
123  // compose the struct.
124  Window win = calculate_max_window(*all_anchors->info(), Steps(info.values_per_roi()));
125  ICLKernel::configure_internal(win);
127 }
128 
130 {
131  ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments(anchors, all_anchors, info));
132  return Status{};
133 }
134 
135 void CLComputeAllAnchorsKernel::run(const Window &window, cl::CommandQueue &queue)
136 {
139 
140  // Collapse everything on the first dimension
141  Window collapsed = window.collapse(ICLKernel::window(), Window::DimX);
142 
143  // Set arguments
144  unsigned int idx = 0;
145  add_1D_tensor_argument(idx, _anchors, collapsed);
146  add_1D_tensor_argument(idx, _all_anchors, collapsed);
147 
148  // Note that we don't need to loop over the slices, as we are launching exactly
149  // as many threads as all the anchors generated
150  enqueue(queue, *this, collapsed, lws_hint());
151 }
152 } // 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 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
void configure(const ICLTensor *anchors, ICLTensor *all_anchors, const ComputeAnchorsInfo &info)
Set the input and output tensors.
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
quantized, symmetric fixed-point 16-bit number
virtual size_t dimension(size_t index) const =0
Return the size of the requested dimension.
#define ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_QUANTIZATION_INFO(...)
Definition: Validate.h:606
size_t values_per_roi() const
Definition: Types.h:1429
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
void run(const Window &window, cl::CommandQueue &queue) override
Enqueue the OpenCL kernel to process the given window on the passed OpenCL command queue...
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's metadata.
Definition: ITensorInfo.h:40
#define ARM_COMPUTE_ERROR_THROW_ON(status)
Definition: Error.h:455
Quantization info when assuming per layer quantization.
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
float spatial_scale() const
Definition: Types.h:1423
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.
const DataType data_type
Definition: Im2Col.cpp:150
Window collapse(const Window &full_window, size_t first, size_t last=Coordinates::num_max_dimensions) const
Collapse the dimensions between first and last.
Definition: Window.inl:111
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
static Status validate(const ITensorInfo *anchors, const ITensorInfo *all_anchors, const ComputeAnchorsInfo &info)
Static function to check if given info will lead to a valid configuration of CLComputeAllAnchorsKerne...
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
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.
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.
ScaleKernelInfo info(interpolation_policy, default_border_mode, PixelValue(), sampling_policy, false)
Interface for OpenCL tensor.
Definition: ICLTensor.h:42
const QuantizationInfo qinfo
Definition: Im2Col.cpp:155
#define ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(...)
Definition: Validate.h:541
ComputeAnchors information class.
Definition: Types.h:1392
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.
#define ARM_COMPUTE_ERROR_ON_NULLPTR(...)
Definition: Validate.h:157
Store the tensor&#39;s metadata.
Definition: TensorInfo.h:43
#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
DataType
Available data types.
Definition: Types.h:79
Describe a multidimensional execution window.
Definition: Window.h:39
#define ARM_COMPUTE_ERROR_ON_INVALID_SUBWINDOW(f, s)
Definition: Validate.h:201