Operator APIs - Atlas Intelligent Edge Solution V100R020C00 IR Model Building Guide 01

Atlas Intelligent Edge Solution V100R020C00 IR Model Building Guide 01

Huawei uses machine translation combined with human proofreading to translate this document to different languages in order to help you better understand the content of this document. Note: Even the most advanced machine translation cannot match the quality of professional translators. Huawei shall not bear any responsibility for translation accuracy and it is recommended that you refer to the English document (a link for which has been provided).

Operator APIs

Class Operator
Class Tensor
Class TensorDesc
Class Shape
Class AttrValue

Class Operator

Constructor and Destructor

Function

Operator constructor and destructor

Syntax

Operator();

explicit Operator(const string& type);

explicit Operator(const string& name, const string& type);

virtual ~Operator()

Parameter Description

Parameter	Input/Output	Description
type	Input	Operator type.
name	Input	Operator name.

Return Value

The Operator constructor returns an object of the Operator type.

Exception Handling

None

Restriction

None

GetName

Function

Obtains the name of an operator.

Syntax

string GetName() const

Parameter Description

None

Return Value

Parameter	Type	Description
-	string	Operator name.

Exception Handling

None

Restriction

None

GetOpType

Function

Obtains the operator type.

Syntax

string GetOpType() const

Parameter Description

None

Return Value

Parameter	Type	Description
-	string	Operator type. If this parameter is not set, the return value is "".

Exception Handling

None

Restriction

None

SetInput

Function

Specifies the input source of the operator, that is, which operator's output is used as the input of the current operator.

There are two SetInput methods:

If output 0 of srcOprt is specified as the input of the current operator, use the first syntax to set the input of the current operator, and the output name of srcOprt does not need to be specified. If another output of srcOprt is specified as the input of the current operator, use the second syntax to set the input of the current operator, and the output name of srcOprt needs to be specified.

Syntax

Operator& SetInput(const string& dst_name, const Operator& src_oprt);

Operator& SetInput(const string& dst_name, const Operator& src_oprt, const string &name)

Parameter Description

Parameter	Input/Output	Description
dst_name	Input	Input name of the operator.
src_oprt	Input	Input operator object whose input name is dstName.
name	Input	Output name of srcOprt.

Return Value

Parameter	Type	Description
-	Operator&	Scheduler itself.

Exception Handling

None

Restriction

Operator& srcOprt can have only one output.

AddControlInput

Function

Adds a control edge of the operator. Currently, the control edge controls only the execution sequence of the operator.

Syntax

Operator& AddControlInput(const Operator& src_oprt)

Parameter Description

Parameter	Input/Output	Description
src_oprt	Input	Source operator corresponding to the control edge.

Return Value

Parameter	Type	Description
-	Operator&	Operator object itself.

Exception Handling

None

Restriction

None

IsEmpty

Function

Checks whether the operator object is empty. If yes, the object is unavailable.

Syntax

bool IsEmpty() const

Parameter Description

None

Return Value

Parameter	Type	Description
-	const	True: not empty False: empty

Exception Handling

None

Restriction

None

GetInputDesc

Function

Obtains the input TensorDesc of the operator based on its input name or input index.

Syntax

TensorDesc GetInputDesc(const string& name) const;

TensorDesc GetInputDesc(uint32_t index) const

Parameter Description

Parameter	Input/Output	Description
name	Input	Input name of the operator. If no operator input name is available, the default object constructed by TensorDesc is returned. Set DataType to DT_FLOAT (indicating the float type) and set Format to FORMAT_NCHW (indicating the NCHW format).
index	Input	Input index of the operator. If no operator input index is available, the default object constructed by TensorDesc is returned. Set DataType to DT_FLOAT (indicating the float type) and set Format to FORMAT_NCHW (indicating the NCHW format).

Parameter

Input/Output

Description

name

Input

Input name of the operator.

If no operator input name is available, the default object constructed by TensorDesc is returned. Set DataType to DT_FLOAT (indicating the float type) and set Format to FORMAT_NCHW (indicating the NCHW format).

index

Input

Input index of the operator.

If no operator input index is available, the default object constructed by TensorDesc is returned. Set DataType to DT_FLOAT (indicating the float type) and set Format to FORMAT_NCHW (indicating the NCHW format).

Return Value

Parameter	Type	Description
-	TensorDesc	TensorDesc of the operator input.

Exception Handling

None

Restriction

None

GetOutputDesc

Function

Obtains the output TensorDesc of the operator based on the operator output name or output index.

Syntax

TensorDesc GetOutputDesc(const string& name) const;

TensorDesc GetOutputDesc(uint32_t index) const

Parameter Description

Parameter	Input/Output	Description
name	Input	Output name of the operator. If no operator output name is available, the default object constructed by TensorDesc is returned. Set DataType to DT_FLOAT (indicating the float type) and set Format to FORMAT_NCHW (indicating the NCHW format).
index	Input	Output index of the operator. If no operator output index is available, the default object constructed by TensorDesc is returned. Set DataType to DT_FLOAT (indicating the float type) and set Format to FORMAT_NCHW (indicating the NCHW format).

Parameter

Input/Output

Description

name

Input

Output name of the operator.

If no operator output name is available, the default object constructed by TensorDesc is returned. Set DataType to DT_FLOAT (indicating the float type) and set Format to FORMAT_NCHW (indicating the NCHW format).

index

Input

Output index of the operator.

If no operator output index is available, the default object constructed by TensorDesc is returned. Set DataType to DT_FLOAT (indicating the float type) and set Format to FORMAT_NCHW (indicating the NCHW format).

Return Value

Parameter	Type	Description
-	TensorDesc	TensorDesc of the operator output.

Exception Handling

None

Restriction

None

TryGetInputDesc

Function

Obtains the input TensorDesc of the operator based on its input name or input index.

Syntax

graphStatus TryGetInputDesc(const string& name, TensorDesc& tensor_desc) const

Parameter Description

Parameter	Input/Output	Description
name	Input	Input name of the operator.
tensor_desc	Output	Returns the current format of the operator port, which is a TensorDesc object.

Return Value

Parameter	Type	Description
-	graphStatus	True: The port is available, and TensorDesc is successfully obtained. False: The port is unavailable, the output parameter is empty, TensorDesc fails to be obtained.

Parameter

Type

Description

graphStatus

True: The port is available, and TensorDesc is successfully obtained.

False: The port is unavailable, the output parameter is empty, TensorDesc fails to be obtained.

Exception Handling

Exception Scenario	Description
No input of the corresponding name.	False is returned.

Restriction

None

GetInputConstData

Function

Obtains the data of the Const node if it corresponds to the input of the specified operator.

Syntax

graphStatus GetInputConstData(const string& dst_name, Tensor& data) const

Parameter Description

Parameter	Input/Output	Description
dst_name	Input	Input name.
data	Output	Returns the data Tensor of the Const node.

Return Value

Parameter	Type	Description
-	graphStatus	If the node corresponding to the input of the specified operator is a Const node and data is obtained successfully, GRAPH_SUCCESS is returned. Otherwise, GRAPH_FAILED is returned.

Exception Handling

None

Restriction

None

GetDynamicInputNum

Function

Obtains the actual number of dynamic inputs of an operator.

Syntax

int GetDynamicInputNum(const string& name) const

Parameter Description

Parameter	Input/Output	Description
name	Input	Name of the dynamic inputs of an operator.

Return Value

Parameter	Type	Description
-	int	Actual number of dynamic inputs. If name is invalid or the operator has no dynamic input, –1 is returned.

Parameter

Type

Description

int

Actual number of dynamic inputs.

If name is invalid or the operator has no dynamic input, –1 is returned.

Restriction

None

GetDynamicOutputNum

Function

Obtains the actual number of dynamic outputs of an operator.

Syntax

int GetDynamicOutputNum(const string& name) const

Parameter Description

Parameter	Input/Output	Description
name	Input	Name of the dynamic outputs of an operator.

Return Value

Parameter	Type	Description
-	int	Actual number of dynamic outputs. If name is invalid or the operator has no dynamic output, 0 is returned.

Parameter

Type

Description

int

Actual number of dynamic outputs.

If name is invalid or the operator has no dynamic output, 0 is returned.

Restriction

None

GetDynamicInputDesc

Function

Obtains the dynamic input TensorDesc of an operator based on the combination of its input name and index.

Syntax

TensorDesc GetDynamicInputDesc(const string& name,uint32_t index) const

Parameter Description

Parameter	Input/Output	Description
name	Input	Dynamic input name of the operator.
index	Input	Dynamic input index of the operator, which starts from 1.

Return Value

Parameter	Type	Description
-	TensorDesc	If TensorDesc is obtained successfully, the required TensorDesc object is returned. Otherwise, the default TensorDesc object is returned. Set DataType to DT_FLOAT (indicating the float type) and Format to FORMAT_NCHW (indicating the NCHW format).

Exception Handling

None

Restriction

None

GetDynamicOutputDesc

Function

Obtains the dynamic output TensorDesc of an operator based on the combination of its input name and index.

Syntax

TensorDesc GetDynamicOutputDesc (const string& name, uint32_t index) const

Parameter Description

Parameter	Input/Output	Description
name	Input	Dynamic output name of the operator.
index	Input	Dynamic output index of the operator, which starts from 1.

Return Value

Parameter	Type	Description
-	TensorDesc	If TensorDesc is obtained successfully, the required TensorDesc object is returned. Otherwise, the default TensorDesc object is returned. Set DataType to DT_FLOAT (indicating the float type) and Format to FORMAT_NCHW (indicating the NCHW format).

Exception Handling

None

Restriction

None

UpdateInputDesc

Function

Updates the input TensorDesc of an operator based on its input name.

Syntax

graphStatus UpdateInputDesc(const string& name, const TensorDesc& tensor_desc)

Parameter Description

Parameter	Input/Output	Description
name	Input	Operator input name.
tensor_desc	Input	TensorDesc object.

Return Value

Parameter	Type	Description
-	graphStatus	If TensorDesc is updated successfully, GRAPH_SUCCESS is returned. Otherwise, GRAPH_FAILED is returned.

Exception Handling

Exception	Description
No input corresponding to the name.	The function ends in advance, and GRAPH_FAILED is returned.

Restriction

None

UpdateOutputDesc

Function

Updates the output TensorDesc of an operator based on its output name.

Syntax

graphStatus UpdateOutputDesc (const string& name, const TensorDesc& tensor_desc)

Parameter Description

Parameter	Input/Output	Description
name	Input	Output name of the operator.
tensor_desc	Input	TensorDesc object.

Return Value

Parameter	Type	Description
-	graphStatus	If TensorDesc is updated successfully, GRAPH_SUCCESS is returned. Otherwise, GRAPH_FAILED is returned.

Exception Handling

None

Restriction

None

UpdateDynamicInputDesc

Function

Updates the dynamic input TensorDesc of an operator based on the combination of its output name and index.

Syntax

graphStatus UpdateDynamicInputDesc(const string& name, uint32_t index, const TensorDesc& tensor_desc)

Parameter Description

Parameter	Input/Output	Description
name	Input	Dynamic input name of the operator.
index	Input	Dynamic input index of the operator, which starts from 1.
tensor_desc	Input	TensorDesc object.

Return Value

Parameter	Type	Description
-	graphStatus	If the dynamic input TensorDesc is updated successfully, GRAPH_SUCCESS is returned. Otherwise, GRAPH_FAILED is returned.

Exception Handling

None

Restriction

None

UpdateDynamicOutputDesc

Function

Updates the dynamic output TensorDesc of an operator based on the combination of its output name and index.

Syntax

graphStatus UpdateDynamicOutputDesc (const string& name, uint32_t index, const TensorDesc& tensor_desc)

Parameter Description

Parameter	Input/Output	Description
name	Input	Dynamic output name of the operator.
index	Input	Dynamic output index of the operator.
tensor_desc	Input	TensorDesc object.

Return Value

Parameter	Type	Description
-	graphStatus	If the dynamic output TensorDesc is updated successfully, GRAPH_SUCCESS is returned. Otherwise, GRAPH_FAILED is returned.

Exception Handling

None

Restriction

None

InferShapeAndType

Function

Infers the shape and data type of the output of the operator.

For details about the definition of data types, see DataType.

Syntax

graphStatus InferShapeAndType()

Parameter Description

None

Return Value

Parameter	Type	Description
-	graphStatus	If the inference is successful, GRAPH_SUCCESS is returned. Otherwise, GRAPH_FAILED is returned.

Exception Handling

None

Restriction

None

SetInferenceContext

Function

Sets the shape and data type of the predecessor operator of the unknowShape operator.

Syntax

void SetInferenceContext(const InferenceContextPtr &inference_context)

Parameter Description

Parameter	Input/Output	Description
inference_context	Input	Inference context of the current operator. InferenceContextPtr is the alias of the pointer to class InferenceContext. using InferenceContextPtr = std::shared_ptr<InferenceContext>;

Parameter

Input/Output

Description

inference_context

Input

Inference context of the current operator.

InferenceContextPtr is the alias of the pointer to class InferenceContext.

using InferenceContextPtr = std::shared_ptr<InferenceContext>;

Return Value

None

Exception Handling

None

Restriction

None

GetInferenceContext

Function

Obtains the shape and data type of the predecessor operator of operator unknowShape.

Syntax

InferenceContextPtr GetInferenceContext() const

Parameter Description

None

Return Value

Parameter	Type	Description
-	InferenceContextPtr	Returns the inference context of the current operator. InferenceContextPtr is the alias of the pointer to class InferenceContext. using InferenceContextPtr = std::shared_ptr<InferenceContext>;

Parameter

Type

Description

InferenceContextPtr

Returns the inference context of the current operator.

InferenceContextPtr is the alias of the pointer to class InferenceContext.

using InferenceContextPtr = std::shared_ptr<InferenceContext>;

Exception Handling

None

Restriction

None

VerifyAllAttr

Function

Verifies whether the attributes, inputs, and outputs of an operator are valid based on the value of disableCommonVerifier.

Syntax

graphStatus VerifyAllAttr(bool disable_common_verifier = false)

Parameter Description

Parameter	Input/Output	Description
disable_common_verifier	Input	false: only the attribute validity is verified. true: validity of the attribute, input, and output of the operator is verified. The default value is false.

Parameter

Input/Output

Description

disable_common_verifier

Input

false: only the attribute validity is verified. true: validity of the attribute, input, and output of the operator is verified.

The default value is false.

Return Value

Parameter	Type	Description
-	graphStatus	If the inference is successful, GRAPH_SUCCESS is returned. Otherwise, GRAPH_FAILED is returned.

Exception Handling

None

Restriction

None

GetInputsSize

Function

Obtains the number of inputs of the current operator.

Syntax

size_t GetInputsSize() const

Parameter Description

None

Return Value

Parameter	Type	Description
-	size_t	Returns the number of inputs of the current operator.

Exception Handling

None

Restriction

None

GetOutputsSize

Function

Obtains the number of all outputs of an operator.

Syntax

size_t GetOutputsSize() const

Parameter Description

None

Return Value

Parameter	Type	Description
-	size_t	Returns the number of outputs of the current operator.

Restriction

None

GetAllAttrNamesAndTypes

Function

Obtains the names and types of all attributes of the operator.

Syntax

const std::map<std::string, std::string> GetAllAttrNamesAndTypes() const

Parameter Description

None

Return Value

Parameter	Type	Description
-	map	All attribute names and types.

Exception Handling

None

Restriction

None

SetAttr

Description

Sets the attribute values of an operator.

An operator may have multiple attributes. After an attribute value is set for the first time, the data type of the attribute is fixed. Must be one of the following data types:

Integer: int64_t, uint32_t, and int32_t
Use SetAttr(const string& name, int64_t attrValue) to set the attribute value, and use GetAttr(const string& name, int32_t& attrValue) or GetAttr(const string& name, uint32_t& attrValue) to obtain the value. Ensure that the integer data is not truncated. It is the same for the types int32_t and uint32_t.
List of integers: std::vector<int64_t>, std::vector<int32_t>, std::vector<uint32_t>, and std::initializer_list<int64_t>&&
Floating-point number: float
List of floating-point numbers: std::vector<float>
String: string
List of strings: std::vector<string>
Bool: bool
List of bools: std::vector<bool>
Tensor: Tensor
List of Tensors: std::vector<Tensor>
Bytes: byte arrays represented by OpBytes (vector<uint8_t>) and const uint8_t* data, size_t size
Quantized data: UsrQuantizeFactorParams
AttrValue type: AttrValue
2D list of integers: std::vector<std::vector<int64_t>
List of DataTypes: vector<ge::DataType>
Data type: DataType
NamedAttrs type: NamedAttrs
List of NamedAttrs: vector<<NamedAttrs>>

Prototype

Operator& SetAttr(const string& name, int64_t attrValue);

Operator& SetAttr(const string& name, int32_t attrValue);

Operator& SetAttr(const string& name, uint32_t attrValue);

Operator& SetAttr(const string& name, const std::vector<int64_t>& attrValue);

Operator& SetAttr(const string& name, const std::vector<int32_t>& attrValue);

Operator& SetAttr(const string& name, const std::vector<uint32_t>& attrValue);

Operator& SetAttr(const string& name, std::initializer_list<int64_t>&& attrValue);

Operator& SetAttr(const string& name, float attrValue);

Operator& SetAttr(const string& name, const std::vector<float>& attrValue);

Operator& SetAttr(const string& name, AttrValue&& attr_value);

Operator& SetAttr(const string& name, const string& attrValue);

Operator& SetAttr(const string& name, const std::vector<string>& attrValue);

Operator& SetAttr(const string& name, bool attrValue);

Operator& SetAttr(const string& name, const std::vector<bool>& attrValue);

Operator& SetAttr(const string& name, const Tensor& attrValue);

Operator& SetAttr(const string& name, const std::vector<Tensor>& attrValue);

Operator& SetAttr(const string& name, const OpBytes& attrValue);

Operator& SetAttr(const string& name, const std::vector<std::vector<int64_t>>& attr_value);

Operator& SetAttr(const string& name, const std::vector<ge::DataType>& attr_value);

Operator& SetAttr(const string& name, const ge::DataType& attr_value)

Operator &SetAttr(const string &name, const ge::NamedAttrs &attr_value);

Operator &SetAttr(const string &name, const std::vector<ge::NamedAttrs> &attr_value);

Parameters

Parameter	Input/Output	Description
name	Input	Attribute name
attrValue	Input	Attribute value of type int64_t
attrValue	Input	Attribute value of type int32_t
attrValue	Input	Attribute value of type uint32_t
attrValue	Input	Attribute value of type vector<int64_t>
attrValue	Input	Attribute value of type vector<int32_t>
attrValue	Input	Attribute value of type vector<uint32_t>
attrValue	Input	Attribute value of type std::initializer_list<int64_t>&&
attrValue	Input	Attribute value of type float
attrValue	Input	Attribute value of type std::vector<float>
attrValue	Input	Attribute value of type bool
attrValue	Input	Attribute value of type std::vector<bool>
attrValue	Input	Attribute value of type AttrValue
attrValue	Input	Attribute value of type string
attrValue	Input	Attribute value of type std::vector<string>
attrValue	Input	Attribute value of type Tensor
attrValue	Input	Attribute value of type std::vector<Tensor>
attrValue	Input	Attribute value of type Bytes. OpBytes indicates vector<uint_8>.
data	Input	Attribute value of type Bytes, specifying the start address of a byte stream
size	Input	Attribute value of type Bytes, specifying the length of a byte stream
attrValue	Input	Attribute value of type UsrQuantizeFactorParams
attrValue	Input	Attribute value of type vector<vector<int64_t>>
attrValue	Input	Attribute value of type vector<ge::DataType>
attrValue	Input	Attribute value of type DataType
attrValue	Input	Attribute value of type NamedAttrs
attrValue	Input	Attribute value of type vector<ge::NamedAttrs>

Returns

Parameter	Type	Description
-	Operator&	Object itself

Exception Handling

None

Restrictions

None

GetAttr

Description

Obtains the attribute value based on the attribute name.

Prototype

graphStatus GetAttr(const string& name, int64_t& attrValue) const;

graphStatus GetAttr(const string& name, int32_t& attrValue) const;

graphStatus GetAttr(const string& name, uint32_t& attrValue) const;

graphStatus GetAttr(const string& name, std::vector<int64_t>& attrValue) const;

graphStatus GetAttr(const string& name, std::vector<int32_t>& attrValue) const;

graphStatus GetAttr(const string& name, std::vector<uint32_t>& attrValue) const;

graphStatus GetAttr(const string& name, float& attrValue) const;

graphStatus GetAttr(const string& name, std::vector<float>& attrValue) const;

graphStatus GetAttr(const string& name, AttrValue& attr_value) const;

graphStatus GetAttr(const string& name, string& attrValue) const;

graphStatus GetAttr(const string& name, std::vector<string>& attrValue) const;

graphStatus GetAttr(const string& name, bool& attrValue) const;

graphStatus GetAttr(const string& name, std::vector<bool>& attrValue) const;

graphStatus GetAttr(const string& name, Tensor& attrValue) const;

graphStatus GetAttr(const string& name, std::vector<Tensor>& attrValue) const;

graphStatus GetAttr(const string& name, OpBytes& attrValue) const;

graphStatus GetAttr(const string& name, std::vector<std::vector<int64_t>>& attr_value) const;

graphStatus GetAttr(const string& name, std::vector<ge::DataType>& attr_value) const;

graphStatus GetAttr(const string& name, ge::DataType& attr_value) const

graphStatus GetAttr(const string& name, std::vector<ge::NamedAttrs>& attr_value) const;

graphStatus GetAttr(const string& name, ge::NamedAttrs& attr_value) const

Parameters

Parameter	Input/Output	Description
name	Output	Attribute name.
attrValue	Output	Returned attribute value of type int64_t.
attrValue	Output	Returned attribute value of type int32_t.
attrValue	Output	Returned attribute value of type uint32_t.
attrValue	Output	Returned attribute value of type vector<int64_t>, which is an integer list.
attrValue	Output	Returned attribute value of type vector<int32_t>, which is an integer list.
attrValue	Output	Returned attribute value of type vector<uint32_t>, which is an integer list.
attrValue	Output	Returned attribute value of the floating point type.
attrValue	Output	Returned attribute value of the floating point list type.
attrValue	Output	Returned attribute value of type AttrValue.
attrValue	Output	Returned attribute value of type bool.
attrValue	Output	Returned attribute value of the bool list type.
attrValue	Output	Returned attribute value of type string.
attrValue	Output	Returned attribute value of the string list type.
attrValue	Output	Returned attribute value of type Tensor.
attrValue	Output	Returned attribute value of the tensor list type.
attrValue	Output	Returned attribute value of type Bytes. OpBytes indicates vector<uint_8>.
attrValue	Output	Returned attribute value of the quantized data type.
attrValue	Output	Returned attribute value of type vector<vector<int64_t>>, which is an integer two-dimensional list type.
attrValue	Output	Returned attribute value of type vector<ge::DataType>, which is a DataType list type.
attrValue	Output	Returned attribute value of type DataType.
attrValue	Output	Returned attribute value of type vector<ge::NamedAttrs>.
attrValue	Output	Returned attribute value of type NamedAttrs.

Returns

Parameter	Type	Description
-	graphStatus	GRAPH_SUCCESS if the corresponding name is found; else, GRAPH_FAILED

Exception Handling

None

Restrictions

None

BreakConnect

Function

Deletes all connections between the current operator and the previous operator and between the current operator and the next operator.

Syntax

void BreakConnect() const

Parameter Description

None

Return Value

None

Exception Handling

None

Restriction

None

GetSubgraphNamesCount

Description

Obtains the number of subgraphs of an operator.

Prototype

size_t Operator::GetSubgraphNamesCount() const

Parameters

None

Returns

Parameter	Type	Description
-	size_t	Number of subgraphs of the operator

Exception Handling

None

Restrictions

None

GetSubgraphNames

Description

Obtains the subgraph names of an operator.

Prototype

std::vector<std::string> GetSubgraphNames() const

Parameters

None

Returns

std::vector<std::string>

Exception Handling

None

Restrictions

None

GetSubgraphBuilder

Description

Obtains the SubgraphBuilder object of an operator subgraph.

Prototype

SubgraphBuilder GetSubgraphBuilder(const string &name) const

Parameters

Parameter	Type	Description
name	string	Subgraph name

Returns

The SubgraphBuilder object

Exception Handling

None

Restrictions

None

GetSubgraph

Description

Obtains the subgraph of an operator.

Prototype

Graph GetSubgraph(const string &name) const

Parameters

Parameter	Type	Description
name	string	Subgraph name

Returns

The Graph object

Exception Handling

None

Restrictions

None

GetDynamicSubgraphBuilder

Description

Obtains the SubgraphBuilder object of a dynamic-input operator subgraph.

Prototype

SubgraphBuilder GetDynamicSubgraphBuilder(const string &name, uint32_t index) const

Parameters

Parameter	Type	Description
name	string	Subgraph name
index	uint32_t	Subgraph index

Returns

The SubgraphBuilder object

Exception Handling

None

Restrictions

None

GetDynamicSubgraph

Description

Obtains a dynamic-input operator subgraph.

Prototype

Graph GetDynamicSubgraph(const string &name, uint32_t index) const

Parameters

Parameter	Type	Description
name	string	Subgraph name
index	uint32_t	Subgraph index

Returns

The Graph object

Exception Handling

None

Restrictions

None

Class Tensor

The copy constructor and assignment operations of class Tensor share the tensor information. The Clone method is used to copy tensors.

Constructor and Destructor

Description

Constructs or destructs a Tensor.

Prototype

Tensor();

explicit Tensor(const TensorDesc &tensorDesc);

Tensor(const TensorDesc &tensorDesc, const std::vector<uint8_t> &data);

Tensor(const TensorDesc &tensorDesc, const uint8_t *data, size_t size);

Tensor(TensorDesc &&tensorDesc, std::vector<uint8_t> &&data);

~Tensor();

Parameters

Parameter	Input/Output	Description
tensorDesc	Input	TensorDesc object
data	Input	Data to be set
size	Input	Data length, in bytes

Returns

The Tensor object

Exception Handling

None

Restrictions

None

SetTensorDesc

Function

Sets the descriptor of a tensor: TensorDesc.

Syntax

graphStatus SetTensorDesc(const TensorDesc &tensorDesc)

Parameter Description

Parameter	Input/Output	Description
tensorDesc	Input	Tensor descriptor to be set.

Return Value

Parameter	Type	Description
-	graphStatus	If the operation is successful, GRAPH_SUCCESS is returned. Otherwise, GRAPH_FAILED is returned.

Exception Handling

None

Restriction

None

GetTensorDesc

Function

Obtains the descriptor of a tensor: TensorDesc.

Syntax

TensorDesc GetTensorDesc() const

Parameter Description

None

Return Value

Parameter	Type	Description
-	TensorDesc	Tensor descriptor.

Exception Handling

None

Restriction

This API modifies the returned TensorDesc, which does not affect the existing TensorDesc information in the Tensor object.

SetData

Function

Sets data to a tensor.

Syntax

graphStatus SetData(std::vector<uint8_t> &&data);

graphStatus SetData(const std::vector<uint8_t> &data);

graphStatus SetData(const uint8_t *data, size_t size)

Parameter Description

Parameter	Input/Output	Description
data	Input	Data to be set.
size	Input	Data length, in bytes.

Return Value

Parameter	Type	Description
-	graphStatus	If the operation is successful, GRAPH_SUCCESS is returned. Otherwise, GRAPH_FAILED is returned.

Exception Handling

None

Restriction

None

GetData

Function

Obtains the data of a tensor.

Data returned by const uint8_t* GetData() const cannot be modified, while data returned by uint8_t* GetData() can be modified.

Syntax

const uint8_t* GetData() const;

uint8_t* GetData()

Parameter Description

None

Return Value

Parameter	Type	Description
-	const Buffer	Tensor data.

Exception Handling

None

Restriction

None

GetSize

Function

Obtains the size of data in a tensor.

Syntax

size_t GetSize() const

Parameter Description

None

Return Value

Parameter	Type	Description
-	size_t	Size of data stored in a tensor, in bytes.

Exception Handling

None

Restriction

None

IsValid

Function

Checks whether a Tensor object is valid.

If the actual size of the tensor data is the same as that of the tensor data described by TensorDesc, the object is valid.

Syntax

graphStatus IsValid()

Parameter Description

None

Return Value

Parameter	Type	Description
-	graphStatus	If the Tensor object is valid, GRAPH_SUCCESS is returned. Otherwise, GRAPH_FAILED is returned.

Exception Handling

None

Restriction

None

Clone

Function

Clones a tensor.

Syntax

Tensor Clone() const

Parameter Description

None

Return Value

Parameter	Type	Description
-	Tensor	Cloned tensor object.

Exception Handling

None

Restriction

None

Class TensorDesc

The copy constructor and assignment operations of class Tensor are value copy (deep copy) and do not share TensorDesc information.

When the Move constructor and Move assignment of class TensorDesc will move the original TensorDesc information to the new TensorDesc object.

Constructor and Destructor

Function

TensorDesc constructor and destructor

Syntax

TensorDesc();

explicit TensorDesc(Shape shape, Format format = FORMAT_ND, DataType dt = DT_FLOAT);

TensorDesc(const TensorDesc& desc);

TensorDesc(TensorDesc&& desc);

~TensorDesc()

Parameter Description

Parameter

Input/Output

Description

shape

Input

Shape object.

format

Input

Format object. The default value is FORMAT_ND.

For the definition of the Format data type, see Format.

Input

DataType object. The default value is DT_FLOAT.

For details about the definition of data types, see DataType.

Return Value

The TensorDesc constructor returns an object of the TensorDesc type.

Exception Handling

None

Restriction

None

Update

Function

Updates the shape, format, and data type of a TensorDesc object.

Syntax

void Update(Shape shape, Format format = FORMAT_ND, DataType dt = DT_FLOAT)

Parameter Description

Parameter

Input/Output

Description

shape

Input

Shape object to be updated.

format

Input

Format object to be updated. The default value is FORMAT_ND.

Input

DataType object to be updated. The default value is DT_FLOAT.

Return Value

None

Exception Handling

None

Restriction

None

SetShape

Function

Sets the shape of a tensor into TensorDesc.

It refers to the shape of the model adapted to the Ascend AI Processor.

Syntax

void SetShape(Shape shape)

Parameter Description

Parameter	Input/Output	Description
shape	Input	shape object to be set into TensorDesc.

Return Value

None

Exception Handling

None

Restriction

None

GetShape

Function

Obtains the shape of a tensor from TensorDesc.

Syntax

Shape GetShape() const

Parameter Description

None

Return Value

Parameter	Type	Description
-	Shape	shape described by TensorDesc

Exception Handling

None

Restriction

The returned shape is obtained through value copy (deep copy). This API modifies the returned shape, which does not affect the existing shape in TensorDesc.

SetUnknownDimNumShape

Description

Sets the shape of a tensor to {-2} to indicate that it is an unknown tensor.

Prototype

graphStatus SetUnknownDimNumShape()

Parameters

Parameter	Input/Output	Description
NA	-	-

Returns

Parameter	Type	Description
-	graphStatus	Execution result: GRAPH_SUCCESS (0) on success; else, failure

Exception Handling

None

Restrictions

None

SetShapeRange

Description

Sets the shape range.

Prototype

graphStatus SetShapeRange(const std::vector<std::pair<int64_t,int64_t>> &range)

Parameters

Parameter

Input/Output

Description

range

Input

Shape range. Each element in a vector is a pair. The first value in each pair specifies the minimum dimension size, and the second value specifies the maximum dimension size. For example:

If the shape of the tensor is {1, 1, –1, 2}, and the maximum size of the third axis is 100, range can be set to

{{1,1},{1,1},{1,100},{2,2}}

Returns

Parameter	Type	Description
-	graphStatus	Execution result: GRAPH_SUCCESS (0) on success; else, failure

Exception Handling

None

Restrictions

None

GetShapeRange

Description

Obtains the shape range.

Prototype

graphStatus GetShapeRange(std::vector<std::pair<int64_t,int64_t>> &range) const

Parameters

Parameter	Input/Output	Description
range	Output	Shape range

Returns

Parameter	Type	Description
-	graphStatus	Execution result: GRAPH_SUCCESS (0) on success; else, failure

Exception Handling

None

Restrictions

None

SetFormat

Function

Sets the tensor format into TensorDesc.

The format refers to the format of the model adapted to the Ascend AI Processor.

Syntax

void SetFormat(Format format)

Parameter Description

Parameter

Input/Output

Description

format

Input

format information to be set.

For details about the format type, see Format.

Return Value

None

Exception Handling

None

Restriction

None

GetFormat

Function

Obtains the format information of a tensor described by TensorDesc.

Syntax

Format GetFormat() const

Parameter Description

None

Return Value

Parameter	Type	Description
-	Format	format information of the tensor described by TensorDesc.

Exception Handling

None

Restriction

The returned format is obtained through value copy (deep copy). This API is used to modify the returned format, which does not affect the existing format information in TensorDesc.

SetDataType

Function

Sets the data type of a tensor into TensorDesc.

The data type refers to that of the model adapted to the Ascend AI Processor.

Syntax

void SetDataType(DataType dt)

Parameter Description

Parameter

Input/Output

Description

Input

Data type of the tensor to be set described by TensorDesc.

For details about data types, see DataType.

Return Value

None

Exception Handling

None

Restriction

None

GetDataType

Function

Obtains the data type of a tensor described by TensorDesc.

Syntax

DataType GetDataType() const

Parameter Description

None

Return Value

Parameter	Type	Description
-	DataType	Data type of the tensor described by TensorDesc.

Exception Handling

None

Restriction

The returned data type is obtained through value copy (deep copy). This API is used to modify the returned data type, which does not affect the existing data type information in TensorDesc.

SetSize

Function

Sets the data size of a tensor into TensorDesc.

Syntax

void SetSize(int64_t size)

Parameter Description

Parameter	Input/Output	Description
size	Input	Data size of the tensor to be set.

Return Value

None

Exception Handling

None

Restriction

None

GetSize

Function

Obtains the data size of a tensor from TensorDesc.

Syntax

int64_t GetSize() const

Parameter Description

None

Return Value

Parameter	Type	Description
-	int64_t	Data size of the tensor described by TensorDesc.

Exception Handling

None

Restriction

None

SetName

Function

Sets the name of a tensor into TensorDesc.

Syntax

void SetName(const std::string &name)

Parameter Description

Parameter	Input/Output	Description
name	Input	Tensor name to be set.

Return Value

None

Exception Handling

None

Restriction

None

GetName

Function

Obtains the name of a tensor from TensorDesc.

Syntax

std::string GetName() const

Parameter Description

None

Return Value

Parameter	Type	Description
-	string	Tensor name.

Exception Handling

None

Restriction

None

SetRealDimCnt

Function

Sets the actual number of dimensions of a tensor into TensorDesc.

The number of shape dimensions returned by the GetShape API may be 1 more than the actual number due to padding. This API is used to set the actual number of shape dimensions.

Syntax

void SetRealDimCnt(const int64_t realDimCnt)

Parameter Description

Parameter	Input/Output	Description
realDimCnt	Input	Actual number of data dimensions of TensorDesc to be set.

Return Value

None

Exception Handling

None

Restriction

None

GetRealDimCnt

Function

Obtains the actual number of dimensions of a tensor described by TensorDesc.

Syntax

int64_t GetRealDimCnt() const

Parameter Description

None

Return Value

Parameter	Type	Description
-	int64_t	Number of actual dimensions described by TensorDesc.

Exception Handling

None

Restriction

None

SetOriginShape

Function

Sets the original shape of a tensor into TensorDesc.

The shape refers to that of the original network model.

Syntax

void SetOriginShape(Shape originShape)

Parameter Description

Parameter	Input/Output	Description
originShape	Input	Original shape set into TensorDesc.

Return Value

None

Exception Handling

None

Restriction

None

GetOriginShape

Function

Obtains the original shape of a tensor described by TensorDesc.

Syntax

Shape GetOriginShape() const

Parameter Description

None

Return Value

Parameter	Type	Description
-	Shape	originShape described by TensorDesc.

Exception Handling

None

Restriction

None

SetOriginFormat

Function

Sets the original format of a tensor into TensorDesc.

The format refers to that of the original network model.

Syntax

void SetOriginFormat(Format originFormat)

Parameter Description

Parameter

Input/Output

Description

originFormat

Input

Original format to be set.

For the definition of the format data type, see Format.

Return Value

None

Exception Handling

None

Restriction

None

GetOriginFormat

Function

Obtains the original format of a tensor described by TensorDesc.

Syntax

Format GetOriginFormat() const

Parameter Description

None

Return Value

Parameter

Type

Description

Format

Original format information of the tensor described by TensorDesc.

For the definition of the format data type, see Format.

Exception Handling

None

Restriction

None

Class Shape

Constructor and Destructor

Function

Shape constructor and destructor

Syntax

Shape();

~Shape();

explicit Shape(const std::vector<int64_t>& dims)

Parameter Description

Parameter

Input/Output

Description

dims

Input

Value of Shape.

Shape indicates the dimensions of a tensor. std::vector<int64_t> indicates each dimension.

Return Value

The Shape constructor returns an object of the Shape type.

Exception Handling

None

Restriction

None

GetDimNum

Function

Obtains the dimension count of Shape.

Syntax

size_t GetDimNum() const

Parameter Description

None

Return Value

Parameter	Type	Description
-	size_t	Shape dimension count of the tensor

Exception Handling

None

Restriction

None

SetDim

Function

Sets dimension idx in Shape to value.

Syntax

graphStatus SetDim(size_t idx, int64_t value)

Parameter Description

Parameter	Input/Output	Description
idx	Input	Dimension index of Shape, which starts from 0.
value	Input	Value to be set.

Return Value

Parameter	Type	Description
-	graphStatus	If the operation is successful, GRAPH_SUCCESS is returned. Otherwise, GRAPH_FAILED is returned.

Exception Handling

None

Restriction

None

GetDim

Function

Obtains the size of dimension idx of Shape.

Syntax

int64_t GetDim(size_t idx) const

Parameter Description

Parameter	Input/Output	Description
idx	Input	Dimension index, which starts from 0.

Return Value

Parameter	Type	Description
-	int64_t	Size of dimension idx.

Exception Handling

None

Restriction

None

GetDims

Function

Obtains the vector composed of dimensions in Shape.

Syntax

std::vector<int64_t> GetDims() const

Parameter Description

None

Return Value

Parameter	Type	Description
-	std::vector<int64_t>	Vector composed of dimensions of Shape.

Exception Handling

None

Restriction

None

GetShapeSize

Function

Obtains the product of all dimensions in Shape .

Syntax

int64_t GetShapeSize() const

Parameter Description

None

Return Value

Parameter	Type	Description
-	int64_t	Multiplication product of all dimensions.

Exception Handling

None

Restriction

None

Class AttrValue

Constructor and Destructor

Function

AttrValue constructor and destructor.

Syntax

AttrValue();

~AttrValue()

Parameter Description

None

Return Value

The AttrValue constructor returns an object of the AttrValue type.

Exception Handling

None

Restriction

None

GetValue

Function

Obtains the value of value in the key-value pair and converts the obtained value from the T type to the DT type.

converting INT to int64_t
converting FLOAT to float
converting STR to std::string

Syntax

template<typename T, typename DT>

graphStatus GetValue(DT& val) const

Parameter Description

Parameter	Input/Output	Description
val	Output	Parameter of the DT type.

Return Value

Parameter	Type	Description
-	graphStatus	If the data type conversion succeeds, GRAPH_SUCCESS is returned. Otherwise, GRAPH_FAILED is returned.

Exception Handling

None

Restriction

None

CreateFrom

Function

Converts DT-type data types (int64_t, float, and std::string) to T-type data types (INT, FLOAT, and STR).

converting int64_t type to INT
converting float to FLOAT
converting std::string type to STR

Syntax

template<typename T, typename DT>

static T CreateFrom(DT&& val)

Parameter Description

Parameter	Input/Output	Description
val	Input	Parameter of the DT type.

Return Value

Parameter	Type	Description
-	INT FLOAT STR	Parameter of the T type.

Exception Handling

None

Restriction

None

Class Operator
Class Tensor
Class TensorDesc
Class Shape
Class AttrValue

Translation

简体中文

Download

Updated: 2020-07-25

Document ID: EDOC1100150950

Downloads: 7

Average rating:

This Document Applies to these Products