arm_compute::CLReduceOperationKernel Class Reference

Class to define interface for the reduce operation kernel. More...

#include <CLReduceOperationKernel.h>

Collaboration diagram for arm_compute::CLReduceOperationKernel:

Public Member Functions
	CLReduceOperationKernel ()
	Default constructor.
	CLReduceOperationKernel (const CLReduceOperationKernel &)=delete
	Prevent instances of this class from being copied (As this class contains pointers)
CLReduceOperationKernel &	operator= (const CLReduceOperationKernel &)=delete
	Prevent instances of this class from being copied (As this class contains pointers)
	CLReduceOperationKernel (CLReduceOperationKernel &&)=default
	Allow instances of this class to be moved.
CLReduceOperationKernel &	operator= (CLReduceOperationKernel &&)=default
	Allow instances of this class to be moved.
	~CLReduceOperationKernel ()=default
	Default destructor.
void	configure (const ICLTensor *input, ICLTensor *output, const uint32_t axis, ReductionOperation op)
	Set the input and output tensors.
void	run (const Window &window, cl::CommandQueue &queue) override

Static Public Member Functions
static Status	validate (const ITensorInfo *input, const ITensorInfo *output, const uint32_t axis, ReductionOperation op)
	Static function to check if given info will lead to a valid configuration of CLReduceOperationKernel.

Detailed Description

Class to define interface for the reduce operation kernel.

Definition at line 60 of file CLReduceOperationKernel.h.

Constructor & Destructor Documentation

CLReduceOperationKernel() [1/3]

CLReduceOperationKernel::CLReduceOperationKernel

(

)

Default constructor.

Definition at line 101 of file CLReduceOperationKernel.cpp.

: _input(nullptr), _output(nullptr), _axis() {}

CLReduceOperationKernel() [2/3]

arm_compute::CLReduceOperationKernel::CLReduceOperationKernel ( const CLReduceOperationKernel &  )

delete

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

CLReduceOperationKernel() [3/3]

arm_compute::CLReduceOperationKernel::CLReduceOperationKernel ( CLReduceOperationKernel &&  )

default

Allow instances of this class to be moved.

~CLReduceOperationKernel()

arm_compute::CLReduceOperationKernel::~CLReduceOperationKernel ( )

default

Default destructor.

References validate().

Member Function Documentation

configure()

void CLReduceOperationKernel::configure	(	const ICLTensor *	input,
		ICLTensor *	output,
		const uint32_t	axis,
		ReductionOperation	op
	)

Set the input and output tensors.

Parameters

[in]	input	Source tensor. Data types supported: U8/S32/F32.
[out]	output	Destination tensor. Data types supported: Same as input. Output will have the same number of dimensions as input.
[in]	axis	Axis along which to reduce.
[in]	op	Reduce operation to perform.

Returns

N/A

Definition at line 103 of file CLReduceOperationKernel.cpp.

{
  ARM_COMPUTE_ERROR_ON_NULLPTR(input, output);
 
  ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(input->info(), output->info(), axis, op));
 
  _input = input;
  _output = output;
  _axis = axis;
 
  std::unique_ptr<ITensorInfo> output_info = output->info()->clone();
  output_info->set_tensor_shape(inferOutputShape(input->info()->tensor_shape(), axis));
 
  // Construct kernel name
  std::string kernel_name;
  int op_code = 0;
  if (op == ReductionOperation::MAX)
  {
    kernel_name = "reduce_min_max";
    op_code = 1;
  }
  else if (op == ReductionOperation::MIN)
  {
    kernel_name = "reduce_min_max";
    op_code = 2;
  }
  else if (op == ReductionOperation::SUM)
  {
    kernel_name = "reduce_sum_mean";
    op_code = 3;
  }
  else if (op == ReductionOperation::MEAN_SUM)
  {
    kernel_name = "reduce_sum_mean";
    op_code = 4;
  }
  else
    throw std::runtime_error("Operation not supported, yet");
 
  // Set kernel build options
  std::set<std::string> build_opts;
  build_opts.emplace("-DDATA_TYPE=" + get_cl_type_from_data_type(output_info->data_type()));
  build_opts.emplace("-DDEPTH_OUT=" + support::cpp11::to_string(output_info->dimension(2)));
  build_opts.emplace("-DOP_CODE=" + support::cpp11::to_string(op_code));
 
  // Create kernel
  _kernel =
    static_cast<cl::Kernel>(CLKernelLibraryEx::get().create_kernel(kernel_name, build_opts));
 
  // Configure  kernel window
  Window win = calculate_max_window(*output_info, Steps());
 
  Coordinates coord;
  coord.set_num_dimensions(output_info->num_dimensions());
  output->info()->set_valid_region(ValidRegion(coord, output_info->tensor_shape()));
 
  ICLKernel::configure_internal(win);
}

References arm_compute::CLKernelLibraryEx::create_kernel(), and arm_compute::CLKernelLibraryEx::get().

operator=() [1/2]

CLReduceOperationKernel & arm_compute::CLReduceOperationKernel::operator= ( CLReduceOperationKernel &&  )

default

Allow instances of this class to be moved.

operator=() [2/2]

CLReduceOperationKernel & arm_compute::CLReduceOperationKernel::operator= ( const CLReduceOperationKernel &  )

delete

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

run()

void CLReduceOperationKernel::run ( const Window &  window, cl::CommandQueue &  queue  )

override

Definition at line 172 of file CLReduceOperationKernel.cpp.

{
  ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL(this);
  ARM_COMPUTE_ERROR_ON_INVALID_SUBWINDOW(ICLKernel::window(), window);
 
  const TensorShape &shape_in = _input->info()->tensor_shape();
 
  unsigned int idx = 2 * num_arguments_per_4D_tensor(); // Skip the input and output parameters
 
  _kernel.setArg<cl_int>(idx++, _axis);
  _kernel.setArg<cl_int>(idx++, shape_in[_axis]);
 
  // Support dimensions up to 4
  Window slice_out = window.collapse(ICLKernel::window(), 2, 4);
 
  // Setup input slice
  Window slice_in(slice_out);
  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));
  slice_in.set(3, Window::Dimension(0, 0, 0));
 
  // Copy output's shape in order to use for recovering at end of this method
  // TODO Remove changing and recovering output's shape if it is guaranteed that the axis positions
  // of input and output are the same
  const TensorShape shape_out = _output->info()->tensor_shape();
  _output->info()->set_tensor_shape(inferOutputShape(shape_in, _axis));
 
  idx = 0;
  add_4D_tensor_argument(idx, _input, slice_in);
  add_4D_tensor_argument(idx, _output, slice_out);
  enqueue(queue, *this, slice_out, lws_hint());
 
  // Recover output's shape of output tensor
  _output->info()->set_tensor_shape(shape_out);
}

Referenced by package.infer.session::inference().

validate()

Status CLReduceOperationKernel::validate ( const ITensorInfo *  input, const ITensorInfo *  output, const uint32_t  axis, ReductionOperation  op  )

static

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

Parameters

[in]	input	Source tensor info. Data types supported: U8/S32/F32.
[in]	output	Destination tensor info. Data types supported: Same as input. Output will have the same number of dimensions as input.
[in]	axis	Axis along which to reduce.
[in]	op	Reduce operation to perform.

Returns

a status

Definition at line 163 of file CLReduceOperationKernel.cpp.

{
  ARM_COMPUTE_ERROR_ON_NULLPTR(input, output);
  ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments(input, output, axis, op));
 
  return Status{};
}

Referenced by arm_compute::CLReduceOperation::validate(), and ~CLReduceOperationKernel().

---

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

• compute/ARMComputeEx/arm_compute/core/CL/kernels/CLReduceOperationKernel.h
• compute/ARMComputeEx/src/core/CL/kernels/CLReduceOperationKernel.cpp