arm_compute::CLQuantizationSymmetricKernel Class Reference

#include <CLQuantizationSymmetricKernel.h>

Collaboration diagram for arm_compute::CLQuantizationSymmetricKernel:

Public Member Functions
	CLQuantizationSymmetricKernel ()
	CLQuantizationSymmetricKernel (const CLQuantizationSymmetricKernel &)=delete
CLQuantizationSymmetricKernel &	operator= (const CLQuantizationSymmetricKernel &)=delete
	CLQuantizationSymmetricKernel (CLQuantizationSymmetricKernel &&)=default
CLQuantizationSymmetricKernel &	operator= (CLQuantizationSymmetricKernel &&)=default
	~CLQuantizationSymmetricKernel ()=default
void	configure (const ICLTensor *input, const ICLTensor *scale_factor, ICLTensor *output)
void	run (const Window &window, cl::CommandQueue &queue) override

Static Public Member Functions
static Status	validate (const ITensorInfo *input, const ITensorInfo *scale_factor, const ITensorInfo *output)

Detailed Description

Interface for the quantization layer kernel.

Note

The implementation supports only 2D input tensors.

Definition at line 54 of file CLQuantizationSymmetricKernel.h.

Constructor & Destructor Documentation

CLQuantizationSymmetricKernel() [1/3]

arm_compute::CLQuantizationSymmetricKernel::CLQuantizationSymmetricKernel

(

)

Default constructor

Definition at line 107 of file CLQuantizationSymmetricKernel.cpp.

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

CLQuantizationSymmetricKernel() [2/3]

arm_compute::CLQuantizationSymmetricKernel::CLQuantizationSymmetricKernel ( const CLQuantizationSymmetricKernel &  )

delete

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

CLQuantizationSymmetricKernel() [3/3]

arm_compute::CLQuantizationSymmetricKernel::CLQuantizationSymmetricKernel ( CLQuantizationSymmetricKernel &&  )

default

Default Move Constructor.

~CLQuantizationSymmetricKernel()

arm_compute::CLQuantizationSymmetricKernel::~CLQuantizationSymmetricKernel ( )

default

Default destructor

References validate().

Member Function Documentation

configure()

void arm_compute::CLQuantizationSymmetricKernel::configure	(	const ICLTensor *	input,
		const ICLTensor *	scale_factor,
		ICLTensor *	output
	)

Set the input, output.

Parameters

[in]	input	Source tensor. Data types supported: F32/F16.
[in]	scale_factor	Scale tensor of output. Data type supported: Same as input.
[out]	output	Destination tensor with the same dimensions of input. Data types supported: S8.

Note

Output auto initialization is not supported by this kernel

Definition at line 112 of file CLQuantizationSymmetricKernel.cpp.

{
  ARM_COMPUTE_ERROR_ON_NULLPTR(input, scale_factor, output);
  ARM_COMPUTE_ERROR_THROW_ON(
    validate_arguments(input->info(), scale_factor->info(), output->info()));
 
  _input = input;
  _scale_factor = scale_factor;
  _output = output;
 
  const int vec_size_x = 16 / input->info()->element_size();
  const int input_width_x = input->info()->tensor_shape().x();
  const bool multi_access_x = (input_width_x / vec_size_x > 0);
 
  // Configure kernel window
  auto [error, window] = validate_and_configure_window(input->info(), output->info());
  ARM_COMPUTE_ERROR_THROW_ON(error);
  ICLKernel::configure_internal(window);
 
  // Create kernel
  CLBuildOptions build_opts;
  build_opts.add_option("-DVEC_SIZE=" + support::cpp11::to_string(vec_size_x));
  build_opts.add_option("-DDATA_TYPE_IN=" + get_cl_type_from_data_type(input->info()->data_type()));
  build_opts.add_option("-DDATA_TYPE_OUT=" +
                        get_cl_type_from_data_type(output->info()->data_type()));
  build_opts.add_option_if(
    multi_access_x,
    "-DLAST_ACCESSED_X=" + support::cpp11::to_string(std::max<int>(input_width_x - vec_size_x, 0)));
 
  _kernel = static_cast<cl::Kernel>(
    CLKernelLibraryEx::get().create_kernel("quantization_symm8", build_opts.options()));
}

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

Referenced by arm_compute::CLFullyConnectedHybridLayer::configure().

operator=() [1/2]

CLQuantizationSymmetricKernel & arm_compute::CLQuantizationSymmetricKernel::operator= ( CLQuantizationSymmetricKernel &&  )

default

Default move assignment operator

operator=() [2/2]

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

delete

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

run()

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

override

Definition at line 157 of file CLQuantizationSymmetricKernel.cpp.

{
  ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL(this);
  ARM_COMPUTE_ERROR_ON_INVALID_SUBWINDOW(ICLKernel::window(), window);
 
  // Support only 2D
  Window window_collapsed = window.collapse_if_possible(ICLKernel::window(), Window::DimZ);
  Window slice = window_collapsed.first_slice_window_2D();
 
  do
  {
    Window scale_slice = slice.shift_dimensions(1);
 
    unsigned int idx = 0;
    add_2D_tensor_argument(idx, _input, slice);
    add_1D_tensor_argument(idx, _scale_factor, scale_slice);
    add_2D_tensor_argument(idx, _output, slice);
    enqueue(queue, *this, slice, lws_hint());
  } while (window_collapsed.slide_window_slice_2D(slice));
}

validate()

Status arm_compute::CLQuantizationSymmetricKernel::validate ( const ITensorInfo *  input, const ITensorInfo *  scale_factor, const ITensorInfo *  output  )

static

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

Parameters

[in]	input	Input tensor info. Data types supported: F32/F16.
[in]	scale_factor	Scale tensor of output. Data type supported: Same as input.
[in]	output	Destination tensor info with the same dimensions of input. Data types supported: S8.

Returns

a status

Definition at line 146 of file CLQuantizationSymmetricKernel.cpp.

{
  ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments(input, scale_factor, output));
  ARM_COMPUTE_RETURN_ON_ERROR(
    validate_and_configure_window(input->clone().get(), output->clone().get()).first);
 
  return Status{};
}

Referenced by arm_compute::CLFullyConnectedHybridLayer::validate(), and ~CLQuantizationSymmetricKernel().

---

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

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