Deeploy.Targets.Generic.Parsers.Conv2DParser

class Deeploy.Targets.Generic.Parsers.Conv2DParser(noBiasHoisting=True)

Bases: ConvParser

Methods

__init__(noBiasHoisting=True)

__init__([noBiasHoisting])

parse(ctxt, node[, default_channels_first, ...])

DONT OVERRIDE - Uses other NodeParser functions to implement a full parsing passing of the node

parseInputs(ctxt, node)

DONT OVERRIDE - Takes care of hoisting IO tensors into the NetworkContext.

parseNode(node)

Parser-specific method to-be-implemented.

parseNodeCtxt(ctxt, node[, channels_first])

Parses the node's input and output tensors, and adds them to its operatorRepresentation.

parseOutputs(ctxt, node)

DONT OVERRIDE - registers the output tensor of the operator

Attributes

operatorRepresentation

The internal representation of the operator this parser has analyzed that describes all relevant attributes of the node to be used by code generation
parseNode(node: Node) → bool

Parser-specific method to-be-implemented. Given a graphsurgeon node, this method returns whether its attributes are mappable by this parser.

Parameters:

node (gs.Node) – Graphsurgeon node to be evaluated

Returns:

False if any attribute in the node cannot be mapped correctly.

Return type:

bool

parseNodeCtxt(ctxt: NetworkContext, node: Node, channels_first: bool = True) → Tuple[NetworkContext, bool]

Parses the node’s input and output tensors, and adds them to its operatorRepresentation. May also be used to assert certain input- and output-level characteristics like correct dimensions.

Parameters:

  • ctxt (NetworkContext) – Current NetworkContext

  • node (gs.Node) – Node to be analyzed

  • channels_first (bool) – Flag to indicate whether tensor dimensions are expected to be in CxHxW layout (true) or HxWxC layout (false)

Returns:

Tuple of the updated NetworkContext and return boolean to indicate whether the node, including it’s IO tensors can be mapped.

Return type:

Tuple[NetworkContext, bool]

parse(ctxt: NetworkContext, node: Node, default_channels_first: bool = True, ioParse: bool = True) → Tuple[NetworkContext, bool]

DONT OVERRIDE - Uses other NodeParser functions to implement a full parsing passing of the node

Parameters:

  • ctxt (NetworkContext) – Current NetworkContext

  • node (gs.Node) – Node to be parsed

  • default_channels_first (bool) – The default channels_first value if none is provided by the node’s attributes

  • ioParse (bool) – Flag to indicate whether to go through IO parsing or not

Returns:

Returns updated NetworkContext and flag to indicate success

Return type:

Tuple[NetworkContext, bool]

classmethod parseInputs(ctxt: NetworkContext, node: Node) → NetworkContext

DONT OVERRIDE - Takes care of hoisting IO tensors into the NetworkContext. Also verifies that all inputs have been registered and the output has not been registered.

Parameters:

  • ctxt (NetworkContext) – Current NetworkContext

  • node (gs.Node) – Node whose IO tensors should be hoisted

Returns:

Updated NetworkContext with hoisted IO tensors

Return type:

NetworkContext

classmethod parseOutputs(ctxt: NetworkContext, node: Node) → NetworkContext

DONT OVERRIDE - registers the output tensor of the operator

Parameters:

  • ctxt (NetworkContext) – Current NetworkContext

  • node (gs.Node) – Operator whose outputs should be parsed

Returns:

Updated NetworkContext

Return type:

NetworkContext

operatorRepresentation: OperatorRepresentation

The internal representation of the operator this parser has analyzed that describes all relevant attributes of the node to be used by code generation

Type:

Dict[str, Any]