Symbolic scan. More...

#include <operation.h>

Inheritance diagram for tvm::te::ScanOpNode:

Collaboration diagram for tvm::te::ScanOpNode:

Public Member Functions
	ScanOpNode ()
	constructor More...

int	num_outputs () const final

Array< IterVar >	root_iter_vars () const final

DataType	output_dtype (size_t i) const final
	Get data type. i-th output tensor. More...

Array< PrimExpr >	output_shape (size_t i) const final
	Get shape of i-th output tensor. More...

Array< Tensor >	InputTensors () const final
	List all the input Tensors. More...

Operation	ReplaceInputs (const Operation &self, const std::unordered_map< Tensor, Tensor > &rmap) const final
	Replace the input of the operation by pattern specified by rmap. More...

void	PropBoundToInputs (const Operation &self, arith::Analyzer analyzer, const std::unordered_map< const VarNode , IntSet > &dom_map, std::unordered_map< Tensor, TensorDom > *out_dom_map) const final
	Propagate the bounds to inputs. More...

void	GatherBound (const Operation &self, const std::unordered_map< Tensor, TensorDom > &tensor_dom, std::unordered_map< IterVar, Range > *out_dom_map) const final
	Gather the bound from output tensor. Set the range of each root_iter_vars in the op to out_dom_map. More...

Stmt	BuildRealize (const Stage &stage, const std::unordered_map< IterVar, Range > &realize_map, const Stmt &body, String storage_scope="") const final
	Build the Realize statement that realizes the op's output tensors. More...

Stmt	BuildProvide (const Stage &stage, const std::unordered_map< IterVar, Range > &dom_map, bool debug_keep_trivial_loop) const final
	Build the statement that provide the output tensors. More...

void	VisitAttrs (AttrVisitor *v)

	TVM_DECLARE_FINAL_OBJECT_INFO (ScanOpNode, OperationNode)

Public Member Functions inherited from tvm::te::OperationNode
virtual	~OperationNode ()

	TVM_DECLARE_BASE_OBJECT_INFO (OperationNode, Object)

Public Member Functions inherited from tvm::runtime::Object
uint32_t	type_index () const

std::string	GetTypeKey () const

size_t	GetTypeKeyHash () const

template<typename TargetType >
bool	IsInstance () const

bool	unique () const

	Object ()

	Object (const Object &other)

	Object (Object &&other)

Object &	operator= (const Object &other)

Object &	operator= (Object &&other)

Public Attributes
IterVar	scan_axis
	IterVar to scan over. More...

Array< Tensor >	init
	the initialization tensors More...

Array< Tensor >	update
	the update function represented by tensor More...

Array< Tensor >	state_placeholder
	The placeholder to refer as states in update. More...

Array< Tensor >	inputs
	the inputs to the scan, these are optionally provided But they can be helpful to provide hints to speedup get of scan body. More...

Array< IterVar >	spatial_axis_
	Spatial axis to indicate spatial dimension of each output. They corresponds to flattened spatial axis of the outputs. More...

Public Attributes inherited from tvm::te::OperationNode
std::string	name
	optional name of the operation More...

std::string	tag
	optional tag of the operation More...

Map< String, ObjectRef >	attrs
	additional attributes of the operation More...

Static Public Attributes
static constexpr const char *	_type_key = "ScanOp"

Static Public Attributes inherited from tvm::te::OperationNode
static constexpr const char *	_type_key = "Operation"

Static Public Attributes inherited from tvm::runtime::Object
static constexpr const char *	_type_key = "runtime.Object"

static constexpr bool	_type_final = false

static constexpr uint32_t	_type_child_slots = 0

static constexpr bool	_type_child_slots_can_overflow = true

static constexpr bool	_type_has_method_visit_attrs = true

static constexpr bool	_type_has_method_sequal_reduce = false

static constexpr bool	_type_has_method_shash_reduce = false

static constexpr uint32_t	_type_index = TypeIndex::kDynamic

Additional Inherited Members
Public Types inherited from tvm::runtime::Object
typedef void(*	FDeleter) (Object *self)
	Object deleter. More...

using	RefCounterType = std::atomic< int32_t >

Static Public Member Functions inherited from tvm::runtime::Object
static std::string	TypeIndex2Key (uint32_t tindex)
	Get the type key of the corresponding index from runtime. More...

static size_t	TypeIndex2KeyHash (uint32_t tindex)
	Get the type key hash of the corresponding index from runtime. More...

static uint32_t	TypeKey2Index (const std::string &key)
	Get the type index of the corresponding key from runtime. More...

static uint32_t	_GetOrAllocRuntimeTypeIndex ()

static uint32_t	RuntimeTypeIndex ()

Protected Member Functions inherited from tvm::runtime::Object
void	IncRef ()
	developer function, increases reference counter. More...

void	DecRef ()
	developer function, decrease reference counter. More...

Static Protected Member Functions inherited from tvm::runtime::Object
static uint32_t	GetOrAllocRuntimeTypeIndex (const std::string &key, uint32_t static_tindex, uint32_t parent_tindex, uint32_t type_child_slots, bool type_child_slots_can_overflow)
	Get the type index using type key. More...

Protected Attributes inherited from tvm::runtime::Object
uint32_t	type_index_ {0}
	Type index(tag) that indicates the type of the object. More...

RefCounterType	ref_counter_ {0}
	The internal reference counter. More...

FDeleter	deleter_ = nullptr
	deleter of this object to enable customized allocation. If the deleter is nullptr, no deletion will be performed. The creator of the object must always set the deleter field properly. More...

Detailed Description

Symbolic scan.

Constructor & Destructor Documentation

◆ ScanOpNode()

tvm::te::ScanOpNode::ScanOpNode ( )

inline

constructor

Member Function Documentation

◆ BuildProvide()

Stmt tvm::te::ScanOpNode::BuildProvide	(	const Stage &	stage,
		const std::unordered_map< IterVar, Range > &	dom_map,
		bool	debug_keep_trivial_loop
	)		const

finalvirtual

Build the statement that provide the output tensors.

Parameters

stage	The schedule stage of the op.
dom_map	The domain map of all iteration domains.
debug_keep_trivial_loop	Whether keep trivial loops with extent of 1

Returns: A statement that add production and wraps consumer.

Implements tvm::te::OperationNode.

◆ BuildRealize()

Stmt tvm::te::ScanOpNode::BuildRealize	(	const Stage &	stage,
		const std::unordered_map< IterVar, Range > &	realize_map,
		const Stmt &	body,
		String	storage_scope = `""`
	)		const

finalvirtual

Build the Realize statement that realizes the op's output tensors.

Parameters

stage	the op's stage.
realize_map	The realization domain map of the operators.
body	The body that is going to get
storage_scope	The storage scope associated with this realization

Returns: A realization statement that wraps body.

Implements tvm::te::OperationNode.

◆ GatherBound()

void tvm::te::ScanOpNode::GatherBound	(	const Operation &	self,
		const std::unordered_map< Tensor, TensorDom > &	tensor_dom,
		std::unordered_map< IterVar, Range > *	out_dom_map
	)		const

finalvirtual

Gather the bound from output tensor. Set the range of each root_iter_vars in the op to out_dom_map.

Parameters

self	The reference to self.
tensor_dom	Domain map of Tensor->access set of each dimension.
out_dom_map	The output domain map of each IterVar to be setted.

Implements tvm::te::OperationNode.

◆ InputTensors()

Array<Tensor> tvm::te::ScanOpNode::InputTensors ( ) const

finalvirtual

List all the input Tensors.

Returns: List of input tensors.

Implements tvm::te::OperationNode.

◆ num_outputs()

int tvm::te::ScanOpNode::num_outputs ( ) const

finalvirtual

Returns: number of outputs

Implements tvm::te::OperationNode.

◆ output_dtype()

DataType tvm::te::ScanOpNode::output_dtype ( size_t i ) const

finalvirtual

Get data type. i-th output tensor.

Parameters

i	The output index.

Returns: type of i-th output.

Implements tvm::te::OperationNode.

◆ output_shape()

Array<PrimExpr> tvm::te::ScanOpNode::output_shape ( size_t i ) const

finalvirtual

Get shape of i-th output tensor.

Parameters

i	The output index.

Returns: shape of i-th output.

Implements tvm::te::OperationNode.

◆ PropBoundToInputs()

void tvm::te::ScanOpNode::PropBoundToInputs	(	const Operation &	self,
		arith::Analyzer *	analyzer,
		const std::unordered_map< const VarNode *, IntSet > &	dom_map,
		std::unordered_map< Tensor, TensorDom > *	out_dom_map
	)		const

finalvirtual

Propagate the bounds to inputs.

Parameters

self	The reference to self.
analyzer	The analyzer to be used in the function.
dom_map	the domain map of Variables(corresponds to root_iter_vars)
out_dom_map	The output domain. The function is only asked to fill the bounds for Tensors that is already in the out_dom_map

Implements tvm::te::OperationNode.

◆ ReplaceInputs()

Operation tvm::te::ScanOpNode::ReplaceInputs	(	const Operation &	self,
		const std::unordered_map< Tensor, Tensor > &	rmap
	)		const

finalvirtual

Replace the input of the operation by pattern specified by rmap.

Parameters

self	The reference to self.
rmap	The replacement map.

Returns: self if nothing is replaced, otherwise return replaced op.

Implements tvm::te::OperationNode.

◆ root_iter_vars()

Array<IterVar> tvm::te::ScanOpNode::root_iter_vars ( ) const

finalvirtual

Returns: The list of iteration variable at root

Note: root_iter_vars decides the shape of the outputs.

Implements tvm::te::OperationNode.

◆ TVM_DECLARE_FINAL_OBJECT_INFO()

tvm::te::ScanOpNode::TVM_DECLARE_FINAL_OBJECT_INFO	(	ScanOpNode	,
		OperationNode
	)

◆ VisitAttrs()

void tvm::te::ScanOpNode::VisitAttrs ( AttrVisitor * v )

inline

Member Data Documentation

◆ _type_key

constexpr const char* tvm::te::ScanOpNode::_type_key = "ScanOp"

staticconstexpr

◆ init

Array<Tensor> tvm::te::ScanOpNode::init

the initialization tensors

◆ inputs

Array<Tensor> tvm::te::ScanOpNode::inputs

the inputs to the scan, these are optionally provided But they can be helpful to provide hints to speedup get of scan body.

◆ scan_axis

IterVar tvm::te::ScanOpNode::scan_axis

IterVar to scan over.

◆ spatial_axis_

Array<IterVar> tvm::te::ScanOpNode::spatial_axis_

Spatial axis to indicate spatial dimension of each output. They corresponds to flattened spatial axis of the outputs.

[output[0].axis[1], output[0].axis[2]... output[k].axis[j]...] These are auxiliary data structure for storing result of bound inference. They do not corresponds to splittable iterations, thus the name comes with underscore.

◆ state_placeholder

Array<Tensor> tvm::te::ScanOpNode::state_placeholder

The placeholder to refer as states in update.

◆ update

Array<Tensor> tvm::te::ScanOpNode::update

the update function represented by tensor

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

include/tvm/te/operation.h

Public Member Functions

Public Attributes

Static Public Attributes

Additional Inherited Members

Detailed Description

Constructor & Destructor Documentation

◆ ScanOpNode()

Member Function Documentation

◆ BuildProvide()

◆ BuildRealize()

◆ GatherBound()

◆ InputTensors()

◆ num_outputs()

◆ output_dtype()

◆ output_shape()

◆ PropBoundToInputs()

◆ ReplaceInputs()

◆ root_iter_vars()

◆ TVM_DECLARE_FINAL_OBJECT_INFO()

◆ VisitAttrs()

Member Data Documentation

◆ _type_key

◆ init

◆ inputs

◆ scan_axis

◆ spatial_axis_

◆ state_placeholder

◆ update