Performs forward spatial convolution with weight sharing. Also supports built-in Relu cldnn_activation_desc available by setting it in arguments. More...

#include <convolution.hpp>

Inheritance diagram for cldnn::convolution:

[legend]

Collaboration diagram for cldnn::convolution:

[legend]

Public Types
typedef cldnn_convolution_desc	dto

Public Member Functions
	convolution (const primitive_id &id, const primitive_id &input, const std::vector< primitive_id > &weights, const std::vector< primitive_id > &bias, tensor stride={ 1, 1, 1, 1 }, tensor input_offset={ 0, 0, 0, 0 }, tensor dilation={ 1, 1, 1, 1 }, bool with_activation=false, float activation_slp=0.0f, const padding &output_padding=padding())
	Constructs convolution primitive. More...

	convolution (const primitive_id &id, const primitive_id &input, const std::vector< primitive_id > &weights, tensor stride={ 1, 1, 1, 1 }, tensor input_offset={ 0, 0, 0, 0 }, tensor dilation={ 1, 1, 1, 1 }, bool with_activation=false, float activation_slp=0.0f, const padding &output_padding=padding())
	Constructs convolution primitive (w/o bias). More...

	convolution (const primitive_id &id, const primitive_id &input, const std::vector< primitive_id > &weights, const std::vector< primitive_id > &bias, tensor stride, tensor input_offset, tensor dilation, bool with_activation, float activation_slp, tensor output_size, const padding &output_padding=padding())
	Constructs convolution primitive (computes input paddings to match output size). More...

	convolution (const primitive_id &id, const primitive_id &input, const std::vector< primitive_id > &weights, tensor stride, tensor input_offset, tensor dilation, bool with_activation, float activation_slp, tensor output_size, const padding &output_padding=padding())
	Constructs convolution primitive (w/o bias; computes input paddings to match output size). More...

	convolution (const dto *dto)
	Constructs a copy from C API cldnn_convolution_desc.

int32_t	split () const
	On how many cards split the computation to.

Public Member Functions inherited from cldnn::primitive_base< convolution, cldnn_convolution_desc >
const cldnn_primitive_desc *	get_dto () const override
	Returns pointer to a C API primitive descriptor casted to cldnn_primitive_desc.

Public Member Functions inherited from cldnn::primitive
	primitive (const primitive_type_id &type, const primitive_id &id, const std::vector< primitive_id > &input, const padding &output_padding=padding())

	primitive (const cldnn_primitive_desc *dto)
	Constructs a copy from basic C API cldnn_primitive_desc.

std::vector< std::reference_wrapper< primitive_id > >	dependecies ()
	Returns references to all primitive ids on which this primitive depends - inputs, weights, biases, etc.

std::vector< primitive_id >	dependecies () const
	Returns copy of all primitive ids on which this primitive depends - inputs, weights, biases, etc.

	operator primitive_id () const
	Implicit conversion to primiitive id.

Static Public Member Functions
static primitive_type_id	type_id ()

static convolution	create_with_output_size (const primitive_id &id, const primitive_id &input, const std::vector< primitive_id > &weights, const std::vector< primitive_id > &bias, tensor output_size, tensor stride={ 1, 1, 1, 1 }, tensor input_offset={ 0, 0, 0, 0 }, tensor dilation={ 1, 1, 1, 1 }, bool with_activation=false, float activation_slp=0.0f, const padding &output_padding=padding())
	Constructs convolution primitive (computes input paddings to match output size). More...

static convolution	create_with_output_size (const primitive_id &id, const primitive_id &input, const std::vector< primitive_id > &weights, tensor output_size, tensor stride={ 1, 1, 1, 1 }, tensor input_offset={ 0, 0, 0, 0 }, tensor dilation={ 1, 1, 1, 1 }, bool with_activation=false, float activation_slp=0.0f, const padding &output_padding=padding())
	Constructs convolution primitive (w/o bias; computes input paddings to match output size). More...

Public Attributes
fixed_size_vector_ref	weights
	List of primitive ids containing weights data.

fixed_size_vector_ref	bias
	List of primitive ids containing bias data.

tensor	input_offset
	Defines a shift, relative to (0,0) position of the input buffer, where (0,0) point of the convolution window should start calculations.

tensor	stride
	Defines shift in input buffer between adjacent calculations of output values.

tensor	dilation
	Defines gaps in the input - dilation rate k=1 is normal convolution, k=2 means skipping one pixel per input, k=4 means skipping 3 pixels. As an example in one dimension, a filter w of size 3 would compute over input x the following: w[0]x[0] + w[1]x[1] + w[2]x[2] for dilation of 1. For dilation 2 the filter would instead compute w[0]x[0] + w[1]x[2] + w[2]x[4].

bool	with_activation
	Enable Relu activation.

float	activation_negative_slope
	Relu activation slope.

bool	with_output_size
	Indicates that the primitive has user-defined output size (non-zero value).

tensor	output_size
	User-defined output data size of the primitive (w/o padding).

Public Attributes inherited from cldnn::primitive
const primitive_type_id	type
	Primitive's type id.

const primitive_id	id
	Primitive's id.

fixed_size_vector_ref	input
	List of ids of input primitives.

padding	output_padding
	Requested output padding.

Protected Member Functions
std::vector< std::reference_wrapper< const primitive_id > >	get_dependencies () const override

void	update_dto (dto &dto) const override

Protected Member Functions inherited from cldnn::primitive_base< convolution, cldnn_convolution_desc >
	primitive_base (const primitive_id &id, const std::vector< primitive_id > &input, const padding &output_padding=padding())

	primitive_base (const cldnn_convolution_desc *dto)

Protected Attributes
primitive_id_arr	_weights

primitive_id_arr	_bias

Protected Attributes inherited from cldnn::primitive
primitive_id_arr	_input

Detailed Description

Performs forward spatial convolution with weight sharing. Also supports built-in Relu cldnn_activation_desc available by setting it in arguments.

Parameters are defined in context of "direct" convolution, but actual algorithm is not implied.

Definition at line 34 of file convolution.hpp.

Constructor & Destructor Documentation

◆ convolution() [1/4]

cldnn::convolution::convolution	(	const primitive_id &	id,
		const primitive_id &	input,
		const std::vector< primitive_id > &	weights,
		const std::vector< primitive_id > &	bias,
		tensor	stride = `{ 1, 1, 1, 1 }`,
		tensor	input_offset = `{ 0,0,0,0 }`,
		tensor	dilation = `{ 1, 1, 1, 1 }`,
		bool	with_activation = `false`,
		float	activation_slp = `0.0f`,
		const padding &	output_padding = `padding()`
	)

inline

Constructs convolution primitive.

Parameters

id	This primitive id.
input	Input primitive id.
weights	List of primitive ids containing weights data.
bias	List of primitive ids containing bias data.
input_offset	Input padding/offset.
stride	Defines shift in input buffer between adjacent calculations of output values.
dilation	Defines gaps in the input - dilation rate k=1 is normal convolution, k=2 means skipping one pixel per input, k=4 means skipping 3 pixels. As an example in one dimension, a filter w of size 3 would compute over input x the following: w[0]x[0] + w[1]x[1] + w[2]x[2] for dilation of 1. For dilation 2 the filter would instead compute w[0]x[0] + w[1]x[2] + w[2]x[4].
with_activation	Enable Relu activation.
activation_slp	Relu activation slope.

Definition at line 50 of file convolution.hpp.

◆ convolution() [2/4]

cldnn::convolution::convolution	(	const primitive_id &	id,
		const primitive_id &	input,
		const std::vector< primitive_id > &	weights,
		tensor	stride = `{ 1, 1, 1, 1 }`,
		tensor	input_offset = `{ 0,0,0,0 }`,
		tensor	dilation = `{ 1, 1, 1, 1 }`,
		bool	with_activation = `false`,
		float	activation_slp = `0.0f`,
		const padding &	output_padding = `padding()`
	)

inline

Constructs convolution primitive (w/o bias).

Parameters

id	This primitive id.
input	Input primitive id.
weights	List of primitive ids containing weights data.
input_offset	Input padding/offset.
stride	Defines shift in input buffer between adjacent calculations of output values.
dilation	Defines gaps in the input - dilation rate k=1 is normal convolution, k=2 means skipping one pixel per input, k=4 means skipping 3 pixels. As an example in one dimension, a filter w of size 3 would compute over input x the following: w[0]x[0] + w[1]x[1] + w[2]x[2] for dilation of 1. For dilation 2 the filter would instead compute w[0]x[0] + w[1]x[2] + w[2]x[4].
with_activation	Enable Relu activation.
activation_slp	Relu activation slope.

Definition at line 89 of file convolution.hpp.

◆ convolution() [3/4]

cldnn::convolution::convolution	(	const primitive_id &	id,
		const primitive_id &	input,
		const std::vector< primitive_id > &	weights,
		const std::vector< primitive_id > &	bias,
		tensor	stride,
		tensor	input_offset,
		tensor	dilation,
		bool	with_activation,
		float	activation_slp,
		tensor	output_size,
		const padding &	output_padding = `padding()`
	)

inline

Constructs convolution primitive (computes input paddings to match output size).

Parameters

id	This primitive id.
input	Input primitive id.
weights	List of primitive ids containing weights data.
bias	List of primitive ids containing bias data.
input_offset	Input padding/offset.
stride	Defines shift in input buffer between adjacent calculations of output values.
dilation	Defines gaps in the input - dilation rate k=1 is normal convolution, k=2 means skipping one pixel per input, k=4 means skipping 3 pixels. As an example in one dimension, a filter w of size 3 would compute over input x the following: w[0]x[0] + w[1]x[1] + w[2]x[2] for dilation of 1. For dilation 2 the filter would instead compute w[0]x[0] + w[1]x[2] + w[2]x[4].
with_activation	Enable Relu activation.
activation_slp	Relu activation slope.
output_size	User-defined output data size of the primitive (w/o padding).

Definition at line 127 of file convolution.hpp.

◆ convolution() [4/4]

cldnn::convolution::convolution	(	const primitive_id &	id,
		const primitive_id &	input,
		const std::vector< primitive_id > &	weights,
		tensor	stride,
		tensor	input_offset,
		tensor	dilation,
		bool	with_activation,
		float	activation_slp,
		tensor	output_size,
		const padding &	output_padding = `padding()`
	)

inline

Constructs convolution primitive (w/o bias; computes input paddings to match output size).

Parameters

id	This primitive id.
input	Input primitive id.
weights	List of primitive ids containing weights data.
input_offset	Input padding/offset.
stride	Defines shift in input buffer between adjacent calculations of output values.
dilation	Defines gaps in the input - dilation rate k=1 is normal convolution, k=2 means skipping one pixel per input, k=4 means skipping 3 pixels. As an example in one dimension, a filter w of size 3 would compute over input x the following: w[0]x[0] + w[1]x[1] + w[2]x[2] for dilation of 1. For dilation 2 the filter would instead compute w[0]x[0] + w[1]x[2] + w[2]x[4].
with_activation	Enable Relu activation.
activation_slp	Relu activation slope.
output_size	User-defined output data size of the primitive (w/o padding).

Definition at line 169 of file convolution.hpp.

Member Function Documentation

◆ create_with_output_size() [1/2]

static convolution cldnn::convolution::create_with_output_size	(	const primitive_id &	id,
		const primitive_id &	input,
		const std::vector< primitive_id > &	weights,
		const std::vector< primitive_id > &	bias,
		tensor	output_size,
		tensor	stride = `{ 1, 1, 1, 1 }`,
		tensor	input_offset = `{ 0,0,0,0 }`,
		tensor	dilation = `{ 1, 1, 1, 1 }`,
		bool	with_activation = `false`,
		float	activation_slp = `0.0f`,
		const padding &	output_padding = `padding()`
	)

inlinestatic

Constructs convolution primitive (computes input paddings to match output size).

Parameters

id	This primitive id.
input	Input primitive id.
weights	List of primitive ids containing weights data.
bias	List of primitive ids containing bias data.
input_offset	Input padding/offset.
stride	Defines shift in input buffer between adjacent calculations of output values.
dilation	Defines gaps in the input - dilation rate k=1 is normal convolution, k=2 means skipping one pixel per input, k=4 means skipping 3 pixels. As an example in one dimension, a filter w of size 3 would compute over input x the following: w[0]x[0] + w[1]x[1] + w[2]x[2] for dilation of 1. For dilation 2 the filter would instead compute w[0]x[0] + w[1]x[2] + w[2]x[4].
with_activation	Enable Relu activation.
activation_slp	Relu activation slope.
output_size	User-defined output data size of the primitive (w/o padding).

Returns: Convolution primitive with specified settings.

Definition at line 229 of file convolution.hpp.

◆ create_with_output_size() [2/2]

static convolution cldnn::convolution::create_with_output_size	(	const primitive_id &	id,
		const primitive_id &	input,
		const std::vector< primitive_id > &	weights,
		tensor	output_size,
		tensor	stride = `{ 1, 1, 1, 1 }`,
		tensor	input_offset = `{ 0,0,0,0 }`,
		tensor	dilation = `{ 1, 1, 1, 1 }`,
		bool	with_activation = `false`,
		float	activation_slp = `0.0f`,
		const padding &	output_padding = `padding()`
	)

inlinestatic

Constructs convolution primitive (w/o bias; computes input paddings to match output size).

Parameters

id	This primitive id.
input	Input primitive id.
weights	List of primitive ids containing weights data.
input_offset	Input padding/offset.
stride	Defines shift in input buffer between adjacent calculations of output values.
dilation	Defines gaps in the input - dilation rate k=1 is normal convolution, k=2 means skipping one pixel per input, k=4 means skipping 3 pixels. As an example in one dimension, a filter w of size 3 would compute over input x the following: w[0]x[0] + w[1]x[1] + w[2]x[2] for dilation of 1. For dilation 2 the filter would instead compute w[0]x[0] + w[1]x[2] + w[2]x[4].
with_activation	Enable Relu activation.
activation_slp	Relu activation slope.
output_size	User-defined output data size of the primitive (w/o padding).

Returns: Convolution primitive with specified settings.

Definition at line 260 of file convolution.hpp.

The documentation for this struct was generated from the following file:

C:/ML/OpenSource/clDNN/api/CPP/convolution.hpp

Public Types

Public Member Functions

Static Public Member Functions

Public Attributes

Protected Member Functions

Protected Attributes

Detailed Description

Constructor & Destructor Documentation

◆ convolution() [1/4]

◆ convolution() [2/4]

◆ convolution() [3/4]

◆ convolution() [4/4]

Member Function Documentation

◆ create_with_output_size() [1/2]

◆ create_with_output_size() [2/2]