Interface for the im2col reshape kernel. More...

#include <ClIm2ColKernel.h>

Collaboration diagram for ClIm2ColKernel:

Public Member Functions
	ClIm2ColKernel ()
	Default constructor. More...

	ARM_COMPUTE_DISALLOW_COPY_ALLOW_MOVE (ClIm2ColKernel)

void	configure (const ClCompileContext &compile_context, ITensorInfo src, ITensorInfo dst, const Size2D &kernel_dims, const PadStrideInfo &conv_info, bool has_bias, const Size2D &dilation=Size2D(1U, 1U), unsigned int num_groups=1)
	Set the input and output of the kernel. More...

void	run_op (ITensorPack &tensors, const Window &window, cl::CommandQueue &queue) override
	Enqueue the OpenCL kernel to process the given window on the passed OpenCL command queue. More...

Public Member Functions inherited from ICLKernel
	ICLKernel ()
	Constructor. More...

cl::Kernel &	kernel ()
	Returns a reference to the OpenCL kernel of this object. More...

CLKernelType	type () const
	Returns the CL kernel type. More...

template<typename T >
void	add_1D_array_argument (unsigned int &idx, const ICLArray< T > *array, const Strides &strides, unsigned int num_dimensions, const Window &window)
	Add the passed 1D array's parameters to the object's kernel's arguments starting from the index idx. More...

void	add_1D_tensor_argument (unsigned int &idx, const ICLTensor *tensor, 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_1D_tensor_argument_if (bool cond, unsigned int &idx, const ICLTensor *tensor, const Window &window)
	Add the passed 1D tensor's parameters to the object's kernel's arguments starting from the index idx if the condition is true. More...

void	add_2D_tensor_argument (unsigned int &idx, const ICLTensor *tensor, 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_2D_tensor_argument_if (bool cond, unsigned int &idx, const ICLTensor *tensor, const Window &window)
	Add the passed 2D tensor's parameters to the object's kernel's arguments starting from the index idx if the condition is true. More...

void	add_3D_tensor_argument (unsigned int &idx, const ICLTensor *tensor, const Window &window)
	Add the passed 3D tensor's parameters to the object's kernel's arguments starting from the index idx. More...

void	add_4D_tensor_argument (unsigned int &idx, const ICLTensor *tensor, const Window &window)
	Add the passed 4D tensor's parameters to the object's kernel's arguments starting from the index idx. More...

void	add_5D_tensor_argument (unsigned int &idx, const ICLTensor *tensor, const Window &window)
	Add the passed 5D tensor's parameters to the object's kernel's arguments starting from the index idx. More...

void	add_3d_tensor_nhw_argument (unsigned int &idx, const ICLTensor *tensor)
	Add the passed NHW 3D tensor's parameters to the object's kernel's arguments by passing strides, dimensions and the offset to the first valid element in bytes. More...

void	add_4d_tensor_nhwc_argument (unsigned int &idx, const ICLTensor *tensor)
	Add the passed NHWC 4D tensor's parameters to the object's kernel's arguments by passing strides, dimensions and the offset to the first valid element in bytes. More...

virtual void	run (const Window &window, cl::CommandQueue &queue)
	Enqueue the OpenCL kernel to process the given window on the passed OpenCL command queue. More...

template<typename T >
void	add_argument (unsigned int &idx, T value)
	Add the passed parameters to the object's kernel's arguments starting from the index idx. More...

void	set_lws_hint (const cl::NDRange &lws_hint)
	Set the Local-Workgroup-Size hint. More...

cl::NDRange	lws_hint () const
	Return the Local-Workgroup-Size hint. More...

void	set_wbsm_hint (const cl_int &wbsm_hint)
	Set the workgroup batch size modifier hint. More...

cl_int	wbsm_hint () const
	Return the workgroup batch size modifier hint. More...

const std::string &	config_id () const
	Get the configuration ID. More...

void	set_target (GPUTarget target)
	Set the targeted GPU architecture. More...

void	set_target (cl::Device &device)
	Set the targeted GPU architecture according to the CL device. More...

GPUTarget	get_target () const
	Get the targeted GPU architecture. More...

size_t	get_max_workgroup_size ()
	Get the maximum workgroup size for the device the CLKernelLibrary uses. More...

cl::NDRange	get_cached_gws () const
	Get the cached gws used to enqueue this kernel. More...

void	cache_gws (const cl::NDRange &gws)
	Cache the latest gws used to enqueue this kernel. More...

template<unsigned int dimension_size>
void	add_tensor_argument (unsigned &idx, const ICLTensor *tensor, const Window &window)

template<typename T , unsigned int dimension_size>
void	add_array_argument (unsigned &idx, const ICLArray< T > *array, const Strides &strides, unsigned int num_dimensions, const Window &window)
	Add the passed array's parameters to the object's kernel's arguments starting from the index idx. 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 Window &	window () const
	The maximum window the kernel can be executed on. More...

bool	is_window_configured () const
	Function to check if the embedded window of this kernel has been configured. 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), unsigned int num_groups=1)
	Static function to check if given info will lead to a valid configuration. More...

Static Public Member Functions inherited from ICLKernel
constexpr static unsigned int	num_arguments_per_3d_tensor_nhw ()
	Returns the number of arguments enqueued per NHW 3D Tensor object. More...

constexpr static unsigned int	num_arguments_per_4d_tensor_nhwc ()
	Returns the number of arguments enqueued per NHWC 4D Tensor object. More...

constexpr static unsigned int	num_arguments_per_1D_array ()
	Returns the number of arguments enqueued per 1D array object. More...

constexpr static unsigned int	num_arguments_per_1D_tensor ()
	Returns the number of arguments enqueued per 1D tensor object. More...

constexpr static unsigned int	num_arguments_per_2D_tensor ()
	Returns the number of arguments enqueued per 2D tensor object. More...

constexpr static unsigned int	num_arguments_per_3D_tensor ()
	Returns the number of arguments enqueued per 3D tensor object. More...

constexpr static unsigned int	num_arguments_per_4D_tensor ()
	Returns the number of arguments enqueued per 4D tensor object. More...

static cl::NDRange	gws_from_window (const Window &window, bool use_dummy_work_items)
	Get the global work size given an execution window. More...

Data Fields
DataLayout	_data_layout

std::pair< unsigned int, unsigned int >	_convolved_dims

unsigned int	_num_elems_processed_per_iteration

Size2D	_kernel_dims

PadStrideInfo	_conv_info

unsigned int	_num_groups

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 62 of file ClIm2ColKernel.h.

Constructor & Destructor Documentation

◆ ClIm2ColKernel()

ClIm2ColKernel ( )

Default constructor.

Definition at line 323 of file ClIm2ColKernel.cpp.

     : _data_layout(DataLayout::UNKNOWN),
       _convolved_dims(),
       _num_elems_processed_per_iteration(1),
       _kernel_dims(),
       _conv_info(),
       _num_groups()
 {
     _type = CLKernelType::ELEMENTWISE;
 }

References arm_compute::ELEMENTWISE, and arm_compute::UNKNOWN.

Member Function Documentation

◆ ARM_COMPUTE_DISALLOW_COPY_ALLOW_MOVE()

ARM_COMPUTE_DISALLOW_COPY_ALLOW_MOVE ( ClIm2ColKernel )

◆ configure()

void configure	(	const ClCompileContext &	compile_context,
		ITensorInfo *	src,
		ITensorInfo *	dst,
		const Size2D &	kernel_dims,
		const PadStrideInfo &	conv_info,
		bool	has_bias,
		const Size2D &	dilation = `Size2D(1U, 1U)`,
		unsigned int	num_groups = `1`
	)

Set the input and output of the kernel.

Parameters

[in]	compile_context	The compile context to be used.
[in]	src	The input tensor info 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: QASYMM8/QASYMM8_SIGNED/F16/F32
[out]	dst	The output tensor info. 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_dims	The kernel dimensions (width and height).
[in]	conv_info	Contains padding and stride information described in PadStrideInfo.
[in]	has_bias	In case biases are provided expands the matrix with 1.
[in]	dilation	(Optional) Dilation, in elements, across x and y. Defaults to (1, 1).
[in]	num_groups	(Optional) Number of groups when performing a grouped convolution. num_groups != 1 is only supported for NCHW data layout

Definition at line 334 of file ClIm2ColKernel.cpp.

 {
     ARM_COMPUTE_ERROR_ON_NULLPTR(src, dst);
     ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(src, dst, kernel_dims, conv_info, has_bias, dilation, num_groups));
  
     auto padding_info = get_padding_info({src, dst});
     _data_layout      = src->data_layout();
  
     const unsigned int width_idx    = get_data_layout_dimension_index(_data_layout, DataLayoutDimension::WIDTH);
     const unsigned int height_idx   = get_data_layout_dimension_index(_data_layout, DataLayoutDimension::HEIGHT);
     const unsigned int input_width  = src->dimension(width_idx);
     const unsigned int input_height = src->dimension(height_idx);
  
     // Select and configure the optimal OpenCL kernel to run.
     // This function returns the OpenCL kernel's name, the arguments to pass at compile time, the number of elements processed per iteration
     // and the padding requirement flag
     Im2ColConfiguration im2col_config =
         configure_opencl_kernel(src, kernel_dims, conv_info, has_bias, dilation, num_groups);
  
     // Create kernel
     _kernel = create_kernel(compile_context, im2col_config.kernel_name, im2col_config.build_options);
  
     _convolved_dims =
         scaled_dimensions(input_width, input_height, kernel_dims.width, kernel_dims.height, conv_info, dilation);
     _num_elems_processed_per_iteration = im2col_config.num_elems_processed_per_iteration;
     _kernel_dims                       = kernel_dims; // Only needed by the Tuner
     _conv_info                         = conv_info;   // Only needed by the Tuner
     _num_groups                        = num_groups;
  
     // Configure kernel window
     auto win_config = validate_and_configure_window(src, dst, kernel_dims, conv_info, has_bias, dilation,
                                                     im2col_config.num_elems_processed_per_iteration,
                                                     im2col_config.is_padding_required_nchw, num_groups);
     ARM_COMPUTE_ERROR_THROW_ON(win_config.first);
     IClKernel::configure_internal(win_config.second);
  
     // Set config_id for enabling LWS tuning
     _config_id = im2col_config.kernel_name;
     _config_id += "_";
     _config_id += lower_string(string_from_data_type(src->data_type()));
     _config_id += "_";
     _config_id += support::cpp11::to_string(num_groups);
     _config_id += "_";
     _config_id += support::cpp11::to_string(dst->dimension(0));
     _config_id += "_";
     _config_id += support::cpp11::to_string(dst->dimension(1));
     _config_id += "_";
     _config_id += lower_string(string_from_data_layout(_data_layout));
  
     ARM_COMPUTE_ERROR_ON(src->data_layout() == DataLayout::NHWC && has_padding_changed(padding_info));
 }

References ClIm2ColKernel::_conv_info, ClIm2ColKernel::_convolved_dims, ClIm2ColKernel::_data_layout, ClIm2ColKernel::_kernel_dims, ClIm2ColKernel::_num_elems_processed_per_iteration, ClIm2ColKernel::_num_groups, ARM_COMPUTE_ERROR_ON, ARM_COMPUTE_ERROR_ON_NULLPTR, ARM_COMPUTE_ERROR_THROW_ON, arm_compute::test::validation::conv_info, arm_compute::create_kernel(), arm_compute::test::validation::dst, arm_compute::get_data_layout_dimension_index(), arm_compute::get_padding_info(), arm_compute::test::validation::has_bias, arm_compute::has_padding_changed(), Size2D::height, arm_compute::HEIGHT, arm_compute::cpu::height_idx, arm_compute::lower_string(), arm_compute::NHWC, arm_compute::test::validation::num_groups, arm_compute::scaled_dimensions(), arm_compute::test::validation::src, arm_compute::string_from_data_layout(), arm_compute::string_from_data_type(), arm_compute::support::cpp11::to_string(), arm_compute::cpu::kernels::validate_and_configure_window(), arm_compute::cpu::kernels::validate_arguments(), Size2D::width, arm_compute::WIDTH, and arm_compute::cpu::width_idx.

◆ run_op()

void run_op	(	ITensorPack &	tensors,
		const Window &	window,
		cl::CommandQueue &	queue
	)

overridevirtual

Enqueue the OpenCL kernel to process the given window on the passed OpenCL command queue.

Note: The queue is not flushed by this method, and therefore the kernel will not have been executed by the time this method returns.

Parameters

[in]	tensors	A vector containing the tensors to operato on.
[in]	window	Region on which to execute the kernel. (Must be a valid region of the window returned by window()).
[in,out]	queue	Command queue on which to enqueue the kernel.

Reimplemented from ICLKernel.

Definition at line 412 of file ClIm2ColKernel.cpp.

 {
     ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL(this);
     ARM_COMPUTE_ERROR_ON_MISMATCHING_WINDOWS(IClKernel::window(), window);
     ARM_COMPUTE_ERROR_ON(tensors.empty());
  
     // Get initial windows
     // Collapse in order to have (SRC_DEPTH * BATCH_SIZE) on the 3rd dimension
     Window window_collapsed = window.collapse_if_possible(ICLKernel::window(), Window::DimZ);
     window_collapsed.set_dimension_step(Window::DimZ, 1);
  
     auto src = utils::cast::polymorphic_downcast<const ICLTensor *>(tensors.get_const_tensor(TensorType::ACL_SRC));
     auto dst = utils::cast::polymorphic_downcast<ICLTensor *>(tensors.get_tensor(TensorType::ACL_DST));
     ARM_COMPUTE_ERROR_ON_NULLPTR(src, dst);
  
     Window window_output;
     window_output.use_tensor_dimensions(dst->info()->tensor_shape());
  
     const Window first_slice_3d = window_collapsed.first_slice_window_3D();
  
     Window slice     = first_slice_3d;
     Window slice_in  = first_slice_3d;
     Window slice_out = window_output.first_slice_window_2D();
  
     if (_data_layout == DataLayout::NHWC)
     {
         const Window tmp_win     = window.collapse_if_possible(ICLKernel::window(), 3);
         const int    num_batches = tmp_win[3].end();
  
         slice.set(1, Window::Dimension(0, static_cast<int>(dst->info()->tensor_shape()[1]), 1));
         slice.set(2, Window::Dimension(0, static_cast<int>(num_batches), 1));
     }
     else
     {
         slice.set(0,
                   Window::Dimension(
                       0, static_cast<int>(ceil_to_multiple(_convolved_dims.first, _num_elems_processed_per_iteration)),
                       _num_elems_processed_per_iteration));
         slice.set(1, Window::Dimension(0, static_cast<int>(_convolved_dims.second), 1));
         // Note: In case of NCHW the 3rd dimension is already set collapsing the input window
     }
  
     // Setup input slice
     // The dimensions of the input are increased within the OpenCL kernel
     slice_in.set(Window::DimX, Window::Dimension(0, 0, 0));
     slice_in.set(Window::DimY, Window::Dimension(0, 0, 0));
     slice_in.set(Window::DimZ, Window::Dimension(0, 0, 0));
  
     // Setup output slice
     // The dimensions of the output are increased within the OpenCL kernel
     slice_out.set(Window::DimX, Window::Dimension(0, 0, 0));
     slice_out.set(Window::DimY, Window::Dimension(0, 0, 0));
  
     unsigned int idx = num_arguments_per_3D_tensor() +
                        (_num_groups == 1 ? num_arguments_per_2D_tensor() : num_arguments_per_3D_tensor());
     _kernel.setArg<cl_uint>(idx++, static_cast<unsigned int>(src->info()->strides_in_bytes()[3]));
     _kernel.setArg<cl_uint>(idx++,
                             static_cast<unsigned int>(dst->info()->strides_in_bytes()[((_num_groups == 1) ? 2 : 3)]));
     do
     {
         unsigned int idx = 0;
         add_3D_tensor_argument(idx, src, slice_in);
         if (_num_groups == 1)
         {
             add_2D_tensor_argument(idx, dst, slice_out);
         }
         else
         {
             add_3D_tensor_argument(idx, dst, slice_out);
         }
         enqueue(queue, *this, slice, lws_hint());
     } while (window_collapsed.slide_window_slice_3D(slice) && window_output.slide_window_slice_2D(slice_out) &&
              window_collapsed.slide_window_slice_3D(slice_in));
 }

References ClIm2ColKernel::_convolved_dims, ClIm2ColKernel::_data_layout, ClIm2ColKernel::_num_elems_processed_per_iteration, ClIm2ColKernel::_num_groups, arm_compute::ACL_DST, arm_compute::ACL_SRC, ICLKernel::add_2D_tensor_argument(), ICLKernel::add_3D_tensor_argument(), ARM_COMPUTE_ERROR_ON, ARM_COMPUTE_ERROR_ON_MISMATCHING_WINDOWS, ARM_COMPUTE_ERROR_ON_NULLPTR, ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL, arm_compute::ceil_to_multiple(), Window::collapse_if_possible(), Window::DimX, Window::DimY, Window::DimZ, arm_compute::test::validation::dst, ITensorPack::empty(), arm_compute::enqueue(), Window::first_slice_window_2D(), Window::first_slice_window_3D(), ITensorPack::get_const_tensor(), ITensorPack::get_tensor(), ICLKernel::lws_hint(), arm_compute::NHWC, ICLKernel::num_arguments_per_2D_tensor(), ICLKernel::num_arguments_per_3D_tensor(), Window::set(), Window::set_dimension_step(), arm_compute::test::validation::reference::slice(), Window::slide_window_slice_2D(), Window::slide_window_slice_3D(), arm_compute::test::validation::src, Window::use_tensor_dimensions(), and IKernel::window().

◆ 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)`,
		unsigned int	num_groups = `1`
	)

static

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

Similar to ClIm2ColKernel::configure()

Returns: a status

Definition at line 393 of file ClIm2ColKernel.cpp.

 {
     ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments(src, dst, kernel_dims, conv_info, has_bias, dilation, num_groups));
     Im2ColConfiguration im2col_config =
         configure_opencl_kernel(src, kernel_dims, conv_info, has_bias, dilation, num_groups);
     ARM_COMPUTE_RETURN_ON_ERROR(validate_and_configure_window(src->clone().get(), dst->clone().get(), kernel_dims,
                                                               conv_info, has_bias, dilation,
                                                               im2col_config.num_elems_processed_per_iteration,
                                                               im2col_config.is_padding_required_nchw, num_groups)
                                     .first);
     return Status{};
 }