Compute Library
 21.02
GCIm2ColKernel Class Reference

Interface for the im2col reshape kernel. More...

#include <GCIm2ColKernel.h>

Collaboration diagram for GCIm2ColKernel:
[legend]

Public Member Functions

 GCIm2ColKernel ()
 Default constructor. More...
 
 GCIm2ColKernel (const GCIm2ColKernel &)=delete
 Prevent instances of this class from being copied (As this class contains pointers) More...
 
GCIm2ColKerneloperator= (const GCIm2ColKernel &)=delete
 Prevent instances of this class from being copied (As this class contains pointers) More...
 
 GCIm2ColKernel (GCIm2ColKernel &&)=default
 Allow instances of this class to be moved. More...
 
GCIm2ColKerneloperator= (GCIm2ColKernel &&)=default
 Allow instances of this class to be moved. More...
 
void configure (const IGCTensor *input, IGCTensor *output, const Size2D &kernel_dims, const PadStrideInfo &conv_info, bool has_bias, const Size2D &dilation=Size2D(1U, 1U))
 Set the input and output of the kernel. More...
 
void run (const Window &window) override
 Enqueue the OpenGL ES shader to process the given window. More...
 
- Public Member Functions inherited from IGCKernel
 IGCKernel ()
 Constructor. More...
 
GCKernelkernel ()
 Returns a reference to the GLES kernel of this object. More...
 
void add_1D_tensor_argument (unsigned int &idx, const IGCTensor *tensor, const unsigned int binding_point, const Window &window)
 Add the passed 1D tensor's parameters to the object's kernel's arguments starting from the index idx. More...
 
void add_2D_tensor_argument (unsigned int &idx, const IGCTensor *tensor, const unsigned int binding_point, const Window &window)
 Add the passed 2D tensor's parameters to the object's kernel's arguments starting from the index idx. More...
 
void add_3D_tensor_argument (unsigned int &idx, const IGCTensor *tensor, const unsigned int binding_point, const Window &window)
 Add the passed 3D tensor's parameters to the object's kernel's arguments starting from the index idx. More...
 
unsigned int num_arguments_per_1D_tensor () const
 Returns the number of arguments enqueued per 1D tensor object. More...
 
unsigned int num_arguments_per_2D_tensor () const
 Returns the number of arguments enqueued per 2D tensor object. More...
 
unsigned int num_arguments_per_3D_tensor () const
 Returns the number of arguments enqueued per 3D tensor object. More...
 
void set_lws_hint (gles::NDRange &lws_hint)
 Set the Local-Workgroup-Size hint. More...
 
void set_target (GPUTarget target)
 Set the targeted GPU architecture. More...
 
GPUTarget get_target () const
 Get the targeted GPU architecture. More...
 
- Public Member Functions inherited from IKernel
 IKernel ()
 Constructor. More...
 
virtual ~IKernel ()=default
 Destructor. More...
 
virtual bool is_parallelisable () const
 Indicates whether or not the kernel is parallelisable. More...
 
virtual BorderSize border_size () const
 The size of the border for that kernel. More...
 
const Windowwindow () const
 The maximum window the kernel can be executed on. More...
 

Static Public Member Functions

static Status validate (const ITensorInfo *input, const ITensorInfo *output, const Size2D &kernel_dims, const PadStrideInfo &conv_info, bool has_bias, const Size2D &dilation=Size2D(1U, 1U))
 Static function to check if given info will lead to a valid configuration of CLIm2ColKernel. More...
 

Detailed Description

Interface for the im2col reshape kernel.

Rearranges image blocks into columns. It is used to strip out each convolution block to a single column. It is used to transform a convolution to a plain matrix multiplication.

For example taking into account the image below and assuming 3x3 image blocks with stride of 1 we have:

\[ \left( \begin{array}{cccc} a00 & a01 & a02 & a03 \\ a10 & a11 & a12 & a13 \\ a20 & a21 & a22 & a23 \\ a30 & a31 & a32 & a33 \\ \end{array} \right) = \left( \begin{array}{ccccccccc} a00 & a01 & a02 & a10 & a11 & a12 & a20 & a21 & a22 \\ a01 & a02 & a03 & a11 & a12 & a13 & a21 & a22 & a23 \\ a10 & a11 & a12 & a20 & a21 & a22 & a30 & a31 & a32 \\ a11 & a12 & a13 & a21 & a22 & a23 & a31 & a32 & a33 \\ \end{array} \right) \]

Definition at line 57 of file GCIm2ColKernel.h.

Constructor & Destructor Documentation

◆ GCIm2ColKernel() [1/3]

Default constructor.

Definition at line 64 of file GCIm2ColKernel.cpp.

65  : _input(nullptr), _output(nullptr), _convolved_dims(), _kernel_dims(), _num_elems_processed_per_iteration(1), _run_func(nullptr)
66 {
67 }

◆ GCIm2ColKernel() [2/3]

GCIm2ColKernel ( const GCIm2ColKernel )
delete

Prevent instances of this class from being copied (As this class contains pointers)

◆ GCIm2ColKernel() [3/3]

GCIm2ColKernel ( GCIm2ColKernel &&  )
default

Allow instances of this class to be moved.

Member Function Documentation

◆ configure()

void configure ( const IGCTensor input,
IGCTensor output,
const Size2D kernel_dims,
const PadStrideInfo conv_info,
bool  has_bias,
const Size2D dilation = Size2D(1U, 1U) 
)

Set the input and output of the kernel.

Parameters
[in]inputThe input tensor to convert. 3 lower dimensions represent a single input [width, height, IFM], while every optional dimension from 4 and above represent a batch of inputs. Data types supported: F16/F32
[out]outputThe output tensor. First 2 lower dimensions represent a transform of each 3D input, while every dimension above represents a batch. Data types supported: Same as input
[in]kernel_dimsThe kernel dimensions (width and height).
[in]conv_infoContains padding and stride information described in PadStrideInfo.
[in]has_biasIn case biases are provided expands the matrix with 1.
[in]dilation(Optional) Dilation, in elements, across x and y. Defaults to (1, 1).

Definition at line 69 of file GCIm2ColKernel.cpp.

References ARM_COMPUTE_ERROR_ON_NULLPTR, ARM_COMPUTE_ERROR_THROW_ON, arm_compute::calculate_max_window(), Dimensions< T >::cbegin(), arm_compute::ceil_to_multiple(), Dimensions< T >::cend(), arm_compute::test::validation::conv_info, GCKernelLibrary::create_kernel(), ITensorInfo::data_type(), ITensorInfo::dimension(), Window::DimZ, ITensorInfo::element_size(), arm_compute::mlgo::parser::end(), arm_compute::F16, arm_compute::F32, GCKernelLibrary::get(), PadStrideInfo::has_padding(), Size2D::height, ITensor::info(), arm_compute::test::validation::input, input_height, input_width, kernel_name, BorderSize::left, Dimensions< size_t >::num_max_dimensions, PadStrideInfo::pad_bottom(), PadStrideInfo::pad_left(), PadStrideInfo::pad_right(), PadStrideInfo::pad_top(), ITensorInfo::padding(), BorderSize::right, arm_compute::scaled_dimensions(), Window::set_dimension_step(), ITensorInfo::set_valid_region(), enable_tracing::start, PadStrideInfo::stride(), ITensorInfo::tensor_shape(), arm_compute::support::cpp11::to_string(), arm_compute::U, arm_compute::update_window_and_padding(), arm_compute::validate_arguments(), Size2D::width, Size2D::x(), and Size2D::y().

Referenced by GCConvolutionLayer::configure().

70 {
71  ARM_COMPUTE_ERROR_ON_NULLPTR(input, output);
72 
73  // Perform validation step
75 
76  _input = input;
77  _output = output;
78 
79  // Create kernel
80  std::set<std::string> build_opts;
81  std::string dt_name = (input->info()->data_type() == DataType::F32) ? "DATA_TYPE_FP32" : "DATA_TYPE_FP16";
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  build_opts.insert("#define " + dt_name);
86 
87  if(has_bias)
88  {
89  build_opts.emplace("#define HAS_BIAS");
90  }
91 
92  int stride_x = 0;
93  int stride_y = 0;
94 
95  std::tie(stride_x, stride_y) = conv_info.stride();
96  _kernel_dims = std::make_pair(kernel_dims.width, kernel_dims.height);
97 
98  const bool run_img2col_reduced = (output->info()->dimension(0) == (input->info()->dimension(0) * input->info()->dimension(1) * input->info()->dimension(2))) && (TensorShape::num_max_dimensions >= 4)
99  && (std::equal(input->info()->tensor_shape().cbegin() + 3,
100  input->info()->tensor_shape().cend(),
101  output->info()->tensor_shape().cbegin() + 1))
102  && ((stride_x == 1) && (stride_y == 1) && !conv_info.has_padding())
103  && (dilation == Size2D(1U, 1U));
104 
105  std::string kernel_name = "im2col_generic";
106  if(!run_img2col_reduced)
107  {
108  if(input->info()->data_type() == DataType::F16 && _kernel_dims == std::pair<unsigned int, unsigned int>(1, 1))
109  {
110  build_opts.emplace("#define KERNEL_1x1");
111  }
112 
113  build_opts.emplace("#define IM2COL_GENERIC");
114  _convolved_dims = scaled_dimensions(input->info()->dimension(0), input->info()->dimension(1),
115  kernel_dims.width, kernel_dims.height,
116  conv_info, dilation);
117  _num_elems_processed_per_iteration = (input->info()->data_type() == DataType::F32) ? 1 : 2;
118 
119  build_opts.emplace("#define KERNEL_WIDTH " + support::cpp11::to_string(kernel_dims.width));
120  build_opts.emplace("#define KERNEL_HEIGHT " + support::cpp11::to_string(kernel_dims.height));
121  build_opts.emplace("#define KERNEL_DEPTH " + support::cpp11::to_string(input->info()->dimension(2)));
122  build_opts.emplace("#define CONVOLVED_WIDTH " + support::cpp11::to_string(_convolved_dims.first));
123  build_opts.emplace("#define CONVOLVED_HEIGHT " + support::cpp11::to_string(_convolved_dims.second));
124  build_opts.emplace("#define STRIDE_X " + support::cpp11::to_string(conv_info.stride().first));
125  build_opts.emplace("#define STRIDE_Y " + support::cpp11::to_string(conv_info.stride().second));
126  build_opts.emplace("#define PAD_LEFT " + support::cpp11::to_string(conv_info.pad_left()));
127  build_opts.emplace("#define PAD_TOP " + support::cpp11::to_string(conv_info.pad_top()));
128  build_opts.emplace("#define PAD_RIGHT " + support::cpp11::to_string(conv_info.pad_right()));
129  build_opts.emplace("#define PAD_BOTTOM " + support::cpp11::to_string(conv_info.pad_bottom()));
130  build_opts.emplace("#define SRC_WIDTH " + support::cpp11::to_string(input->info()->dimension(0)));
131  build_opts.emplace("#define SRC_HEIGHT " + support::cpp11::to_string(input->info()->dimension(1)));
132  build_opts.emplace("#define DILATION_X " + support::cpp11::to_string(dilation.x()));
133  build_opts.emplace("#define DILATION_Y " + support::cpp11::to_string(dilation.y()));
134 
135  _run_func = &GCIm2ColKernel::run_generic;
136  }
137  else
138  {
139  build_opts.emplace("#define IM2COL_REDUCED");
140  kernel_name = "im2col_reduced";
141 
142  if(input->info()->data_type() == DataType::F32)
143  {
144  _num_elems_processed_per_iteration = 4 / input->info()->element_size();
145  }
146  else if(input->info()->data_type() == DataType::F16)
147  {
148  int input_width = input->info()->dimension(0);
149  int input_height = input->info()->dimension(1);
150 
151  build_opts.emplace("#define IMAGE_SIZE " + support::cpp11::to_string(input_width * input_height));
152  if(input_width % 8 == 0)
153  {
154  _num_elems_processed_per_iteration = 8;
155  build_opts.emplace("#define IM2COL_REDUCED_8X");
156  }
157  else if(input_width % 4 == 0)
158  {
159  _num_elems_processed_per_iteration = 4;
160  build_opts.emplace("#define IM2COL_REDUCED_4X");
161  }
162  else if(input_width % 2 == 0)
163  {
164  _num_elems_processed_per_iteration = 2;
165  build_opts.emplace("#define IM2COL_REDUCED_2X");
166  }
167  else
168  {
169  _num_elems_processed_per_iteration = 2;
170  build_opts.emplace("#define IM2COL_REDUCED_GENERIC");
171  }
172  }
173 
174  _run_func = &GCIm2ColKernel::run_reduced;
175  }
176 
177  // Create kernel
178  _kernel = static_cast<GCKernel>(GCKernelLibrary::get().create_kernel(kernel_name, build_opts));
179 
180  // Configure kernel window
181  Window win = calculate_max_window(*input->info(), Steps(_num_elems_processed_per_iteration));
182 
183  if(input->info()->data_type() == DataType::F16)
184  {
185  // Calculate input right and bottom border
186  const int input_width = input->info()->dimension(0);
187  const int input_height = input->info()->dimension(1);
188  int input_total_width = input->info()->padding().left + input_width + input->info()->padding().right;
189  int input_padding_right = ceil_to_multiple(input_total_width, _num_elems_processed_per_iteration) - input_total_width;
190  input_total_width = input_width + input_padding_right + input->info()->padding().right;
191  AccessWindowStatic input_access(input->info(), 0, 0, input_total_width, input_height);
192 
193  // Calculate output right and bottom border
194  const int output_width = output->info()->dimension(0);
195  const int output_height = output->info()->dimension(1);
196  const int output_padding_right = ceil_to_multiple(output_width, _num_elems_processed_per_iteration) - output_width;
197  AccessWindowStatic output_access(output->info(), 0, 0, output_width + output_padding_right, output_height);
198 
199  update_window_and_padding(win, input_access, output_access);
200  }
201 
202  output->info()->set_valid_region(ValidRegion(Coordinates(), output->info()->tensor_shape()));
203 
204  if(!run_img2col_reduced)
205  {
206  // set the Z dimension's step same size as the whole dimension so that one can't split across the Z dimension
208  }
209 
210  IGCKernel::configure(win);
211 }
Window calculate_max_window(const ValidRegion &valid_region, const Steps &steps, bool skip_border, BorderSize border_size)
virtual size_t dimension(size_t index) const =0
Return the size of the requested dimension.
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_THROW_ON(status)
Definition: Error.h:455
GCKernel class.
size_t x() const
Semantic accessor for width as x.
Definition: Size2D.h:74
unsigned int pad_top() const
Get the top padding.
Definition: Types.h:806
virtual void set_valid_region(const ValidRegion &valid_region)=0
Set the valid region of the tensor.
size_t height
Height of the image region or rectangle.
Definition: Size2D.h:90
1 channel, 1 F16 per channel
std::pair< unsigned int, unsigned int > scaled_dimensions(int width, int height, int kernel_width, int kernel_height, const PadStrideInfo &pad_stride_info, const Size2D &dilation=Size2D(1U, 1U))
Returns expected width and height of output scaled tensor depending on dimensions rounding mode...
Definition: Utils.cpp:419
Implementation of a static rectangular access pattern.
bool update_window_and_padding(Window &win, Ts &&... patterns)
Update window and padding size for each of the access patterns.
Definition: WindowHelpers.h:46
virtual const TensorShape & tensor_shape() const =0
Size for each dimension of the tensor.
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
Class to describe a number of elements in each dimension.
Definition: Steps.h:40
Coordinates of an item.
Definition: Coordinates.h:37
std::pair< unsigned int, unsigned int > stride() const
Get the stride.
Definition: Types.h:770
std::string kernel_name
virtual ITensorInfo * info() const =0
Interface to be implemented by the child class to return the tensor&#39;s metadata.
unsigned int pad_right() const
Get the right padding.
Definition: Types.h:801
virtual size_t element_size() const =0
Element size in bytes calculated as data_size() * num_channels()
void end(TokenStream &in, bool &valid)
Definition: MLGOParser.cpp:290
virtual PaddingSize padding() const =0
Padding of tensor.
unsigned int left
left of the border
Definition: Types.h:378
unsigned int right
right of the border
Definition: Types.h:376
size_t y() const
Semantic accessor for height as y.
Definition: Size2D.h:83
std::array< T, num_max_dimensions >::const_iterator cend() const
Returns a read-only (constant) iterator that points one past the last element in the dimension array...
Definition: Dimensions.h:255
static GCKernelLibrary & get()
Get the static instance of GCKernelLibrary.
void set_dimension_step(size_t dimension, int step)
Set the step of a given dimension.
Definition: Window.inl:167
std::array< T, num_max_dimensions >::const_iterator cbegin() const
Returns a read-only (constant) iterator that points to the first element in the dimension array...
Definition: Dimensions.h:231
size_t width
Width of the image region or rectangle.
Definition: Size2D.h:89
static constexpr size_t DimZ
Alias for dimension 2 also known as Z dimension.
Definition: Window.h:47
Class for specifying the size of an image or rectangle.
Definition: Size2D.h:34
Status validate_arguments(const ITensorInfo *input, const ITensorInfo *bias, const ITensorInfo *output, const GEMMLowpOutputStageInfo *output_stage)
GCKernel create_kernel(const std::string &shader_name, const StringSet &build_options_set={}) const
Creates a kernel from the kernel library.
#define ARM_COMPUTE_ERROR_ON_NULLPTR(...)
Definition: Validate.h:161
Container for valid region of a window.
Definition: Types.h:188
unsigned int pad_bottom() const
Get the bottom padding.
Definition: Types.h:811
static constexpr size_t num_max_dimensions
Number of dimensions the tensor has.
Definition: Dimensions.h:46
unsigned int pad_left() const
Get the left padding.
Definition: Types.h:796
Describe a multidimensional execution window.
Definition: Window.h:39
bool has_padding() const
Check whether this has any padding.
Definition: Types.h:823

◆ operator=() [1/2]

GCIm2ColKernel& operator= ( const GCIm2ColKernel )
delete

Prevent instances of this class from being copied (As this class contains pointers)

◆ operator=() [2/2]

GCIm2ColKernel& operator= ( GCIm2ColKernel &&  )
default

Allow instances of this class to be moved.

◆ run()

◆ validate()

Status validate ( const ITensorInfo input,
const ITensorInfo output,
const Size2D kernel_dims,
const PadStrideInfo conv_info,
bool  has_bias,
const Size2D dilation = Size2D(1U, 1U) 
)
static

Static function to check if given info will lead to a valid configuration of CLIm2ColKernel.

Parameters
[in]inputThe input tensor to convert. 3 lower dimensions represent a single input [width, height, IFM], while every optional dimension from 4 and above represent a batch of inputs. Data types supported: F16/F32
[in]outputThe output tensor. First 2 lower dimensions represent a transform of each 3D input, while every dimension above represents a batch. Data types supported: Same as input
[in]kernel_dimsThe kernel dimensions (width and height).
[in]conv_infoContains padding and stride information described in PadStrideInfo.
[in]has_biasIn case biases are provided expands the matrix with 1.
[in]dilation(Optional) Dilation, in elements, across x and y. Defaults to (1, 1).
Returns
a status

Definition at line 213 of file GCIm2ColKernel.cpp.

References ARM_COMPUTE_RETURN_ON_ERROR, ARM_COMPUTE_UNUSED, and arm_compute::validate_arguments().

214 {
215  ARM_COMPUTE_UNUSED(kernel_dims);
216  ARM_COMPUTE_UNUSED(conv_info);
218  ARM_COMPUTE_UNUSED(dilation);
220  return Status{};
221 }
#define ARM_COMPUTE_RETURN_ON_ERROR(status)
Checks if a status contains an error and returns it.
Definition: Error.h:204
Status class.
Definition: Error.h:52
#define ARM_COMPUTE_UNUSED(...)
To avoid unused variables warnings.
Definition: Error.h:152
Status validate_arguments(const ITensorInfo *input, const ITensorInfo *bias, const ITensorInfo *output, const GEMMLowpOutputStageInfo *output_stage)

The documentation for this class was generated from the following files: