Compute Library
 21.02
GCScaleKernel.cpp
Go to the documentation of this file.
1 /*
2  * Copyright (c) 2016-2020 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 
26 #include "arm_compute/core/Error.h"
38 #include "support/StringSupport.h"
39 
40 #include <set>
41 #include <string>
42 
43 using namespace arm_compute;
44 
46 {
47  return BorderSize(1);
48 }
49 
51 {
55  ARM_COMPUTE_ERROR_ON(output == input);
57 
58  _input = input;
59  _output = output;
60 
61  // Compute the ratio between source width/height and destination width/height
62  const auto wr = static_cast<float>(input->info()->dimension(0)) / static_cast<float>(output->info()->dimension(0));
63  const auto hr = static_cast<float>(input->info()->dimension(1)) / static_cast<float>(output->info()->dimension(1));
64 
65  // Compute actual border size
66  const bool border_undefined = info.border_mode == BorderMode::UNDEFINED;
67  BorderSize border = border_undefined ? BorderSize(0) : border_size();
68 
69  // Area interpolation behaves as Nearest Neighbour in case of up-sampling
70  auto interpolation_policy_to_use = info.interpolation_policy;
71  if(interpolation_policy_to_use == InterpolationPolicy::AREA && wr <= 1.f && hr <= 1.f)
72  {
73  interpolation_policy_to_use = InterpolationPolicy::NEAREST_NEIGHBOR;
74  }
75  else
76  {
77  ARM_COMPUTE_ERROR_ON(interpolation_policy_to_use == InterpolationPolicy::AREA);
78  }
79 
80  // Create kernel
81  std::set<std::string> build_opts;
82  build_opts.emplace("#define LOCAL_SIZE_X " + support::cpp11::to_string(1));
83  build_opts.emplace("#define LOCAL_SIZE_Y " + support::cpp11::to_string(1));
84  build_opts.emplace("#define LOCAL_SIZE_Z " + support::cpp11::to_string(1));
85 
86  build_opts.emplace("#define DATA_TYPE_FP16");
87  build_opts.emplace("#define BORDER_SIZE " + support::cpp11::to_string(border.right));
89  {
90  build_opts.emplace("#define SAMPLING_POLICY_TOP_LEFT");
91  }
92  else
93  {
94  build_opts.emplace("#define SAMPLING_POLICY_CENTER");
95  }
96 
97  // Configure kernel window
98  unsigned int num_elems_processed_per_iteration = 4;
99  unsigned int input_width_alignment = 2;
100 
101  // performance optimization for 2x upscaling with no border
102  if((fabs(wr - 0.5) < 1e-6) && (fabs(hr - 0.5) < 1e-6) && border_undefined)
103  {
104  num_elems_processed_per_iteration = 8;
105  input_width_alignment = 4;
106  build_opts.emplace("#define SCALE_NEAREST_8X");
107  }
108  else
109  {
110  build_opts.emplace("#define SCALE_NEAREST_GENERIC");
111  }
112 
113  std::string interpolation_name = string_from_interpolation_policy(interpolation_policy_to_use); // NOLINT
114  std::transform(interpolation_name.begin(), interpolation_name.end(), interpolation_name.begin(), ::tolower);
115  std::string kernel_name = "scale_" + interpolation_name;
116  _kernel = GCKernelLibrary::get().create_kernel(kernel_name, build_opts);
117 
118  Window win = calculate_max_window(*output->info(), Steps(num_elems_processed_per_iteration));
119 
120  const ValidRegion &input_valid_region = input->info()->valid_region();
121 
122  const int total_width = border.left + input_valid_region.anchor[0] + input_valid_region.shape[0] + border.right;
123  const int padding_right = ceil_to_multiple(total_width, input_width_alignment) - border.left - input_valid_region.anchor[0] - input_valid_region.shape[0];
124 
125  // Reads can occur within the valid region of the input
126  AccessWindowStatic input_access(input->info(),
127  input_valid_region.anchor[0] - border.left, input_valid_region.anchor[1] - border.top,
128  input_valid_region.anchor[0] + input_valid_region.shape[0] + padding_right,
129  input_valid_region.anchor[1] + input_valid_region.shape[1] + border.bottom);
130 
132 
133  update_window_and_padding(win, input_access, output_access);
134 
135  output_access.set_valid_region(win, calculate_valid_region_scale(*(input->info()),
136  output->info()->tensor_shape(),
138  info.sampling_policy,
139  border_undefined));
140 
141  IGCKernel::configure(win);
142 
143  unsigned int idx = 2 * num_arguments_per_3D_tensor(); //Skip the tensor parameters
144  _kernel.set_argument<float>(idx++, static_cast<float>(input->info()->dimension(0)));
145  _kernel.set_argument<float>(idx++, static_cast<float>(input->info()->dimension(1)));
146  _kernel.set_argument<float>(idx++, wr);
147  _kernel.set_argument<float>(idx++, hr);
148 }
149 
151 {
154 
155  _kernel.use();
156 
157  _output->set_needs_shifting(true);
158 
160  Window slice_in = window.first_slice_window_3D();
161 
162  slice.shift(Window::DimX, -(_output->info()->padding()).left);
163 
164  do
165  {
166  unsigned int idx = 0;
167  add_3D_tensor_argument(idx, _input, 1, slice_in);
168  add_3D_tensor_argument(idx, _output, 2, slice);
169  _kernel.update_shader_params();
170  enqueue(*this, slice);
171  }
172  while(window.slide_window_slice_3D(slice) && window.slide_window_slice_3D(slice_in));
173 }
BorderMode border_mode
Border mode policy.
unsigned int top
top of the border
Definition: Types.h:375
Window calculate_max_window(const ValidRegion &valid_region, const Steps &steps, bool skip_border, BorderSize border_size)
const Window & window() const
The maximum window the kernel can be executed on.
Definition: IKernel.cpp:28
void add_3D_tensor_argument(unsigned int &idx, const IGCTensor *tensor, const unsigned int binding_point, 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: IGCKernel.cpp:132
BorderSize border_size() const override
The size of the border for that kernel.
InterpolationPolicy interpolation_policy
Interpolation type to use.
void enqueue(IGCKernel &kernel, const Window &window, const gles::NDRange &lws=gles::NDRange(1U, 1U, 1U))
Add the kernel to the command queue with the given window.
Definition: IGCKernel.cpp:41
virtual size_t dimension(size_t index) const =0
Return the size of the requested dimension.
void shift(size_t dimension, int shift_value)
Shift the values of a given dimension by the given shift_value.
Definition: Window.inl:133
Container for 2D border size.
Definition: Types.h:273
std::string to_string(T &&value)
Convert integer and float values to string.
#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
Interface for GLES Compute tensor.
Definition: IGCTensor.h:35
unsigned int bottom
bottom of the border
Definition: Types.h:377
unsigned int num_arguments_per_3D_tensor() const
Returns the number of arguments enqueued per 3D tensor object.
Definition: IGCKernel.cpp:147
Output values are defined to match the source pixel whose center is nearest to the sample position...
Copyright (c) 2017-2021 Arm Limited.
1 channel, 1 F16 per channel
virtual ValidRegion valid_region() const =0
Valid region of the tensor.
ITensorInfo * info() const override
Interface to be implemented by the child class to return the tensor&#39;s metadata.
Definition: Tensor.cpp:33
Implementation of a static rectangular access pattern.
void configure(const IGCTensor *input, IGCTensor *output, const ScaleKernelInfo &info)
Initialise the kernel&#39;s inputs, output and interpolation policy.
std::string tolower(std::string string)
Convert string to lower case.
Definition: Utility.h:203
static constexpr size_t DimX
Alias for dimension 0 also known as X dimension.
Definition: Window.h:43
bool update_window_and_padding(Window &win, Ts &&... patterns)
Update window and padding size for each of the access patterns.
Definition: WindowHelpers.h:46
SamplingPolicy sampling_policy
Sampling policy used by the interpolation.
virtual const TensorShape & tensor_shape() const =0
Size for each dimension of the tensor.
#define ARM_COMPUTE_ERROR_ON_MISMATCHING_DATA_TYPES(...)
Definition: Validate.h:543
auto ceil_to_multiple(S value, T divisor) -> decltype(((value+divisor - 1)/divisor) *divisor)
Computes the smallest number larger or equal to value that is a multiple of divisor.
Definition: Utils.h:71
Manages all the GLES kernels compilation and caching, provides accessors for the GLES Context...
Class to describe a number of elements in each dimension.
Definition: Steps.h:40
Implementation of a row access pattern.
std::string kernel_name
const std::string & string_from_interpolation_policy(InterpolationPolicy policy)
Translates a given interpolation policy to a string.
Definition: Utils.cpp:212
virtual ITensorInfo * info() const =0
Interface to be implemented by the child class to return the tensor&#39;s metadata.
Samples are taken at pixel top left corner.
virtual PaddingSize padding() const =0
Padding of tensor.
unsigned int left
left of the border
Definition: Types.h:378
bool slide_window_slice_3D(Window &slice) const
Slide the passed 3D window slice.
Definition: Window.h:335
unsigned int right
right of the border
Definition: Types.h:376
#define ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL(k)
Definition: Validate.h:941
Output values are determined by averaging the source pixels whose areas fall under the area of the de...
#define ARM_COMPUTE_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(t, c,...)
Definition: Validate.h:790
ValidRegion calculate_valid_region_scale(const ITensorInfo &src_info, const TensorShape &dst_shape, InterpolationPolicy interpolate_policy, SamplingPolicy sampling_policy, bool border_undefined)
Helper function to calculate the Valid Region for Scale.
Definition: Helpers.cpp:28
Wrapper to configure the Khronos EGL and OpenGL ES C header.
static GCKernelLibrary & get()
Get the static instance of GCKernelLibrary.
ScaleKernelInfo info(interpolation_policy, default_border_mode, PixelValue(), sampling_policy, false)
Borders are left undefined.
void run(const Window &window) override
Enqueue the OpenGL ES shader to process the given window.
GCKernel create_kernel(const std::string &shader_name, const StringSet &build_options_set={}) const
Creates a kernel from the kernel library.
unsigned int num_elems_processed_per_iteration
#define ARM_COMPUTE_ERROR_ON_NULLPTR(...)
Definition: Validate.h:161
Container for valid region of a window.
Definition: Types.h:188
Window first_slice_window_3D() const
First 3D slice of the window.
Definition: Window.h:291
Describe a multidimensional execution window.
Definition: Window.h:39
#define ARM_COMPUTE_ERROR_ON_INVALID_SUBWINDOW(f, s)
Definition: Validate.h:205
SimpleTensor< T > slice(const SimpleTensor< T > &src, Coordinates starts, Coordinates ends)