Global schedule container For operations and all the operations they depend on. The schedule per Operation is named as stage. More...

#include <schedule.h>

Inheritance diagram for tvm::te::Schedule:

Collaboration diagram for tvm::te::Schedule:

Public Types
using	ContainerType = ScheduleNode

Public Types inherited from tvm::runtime::ObjectRef
using	ContainerType = Object
	type indicate the container type. More...

Public Member Functions
	Schedule ()

	Schedule (ObjectPtr< Object > n)

	Schedule (Array< Operation > ops)
	Create a schedule for array of ops(and their dependencies). More...

Schedule	copy () const
	Get a copy of current schedule. More...

Stage	operator[] (const Operation &op)
	Get the stage corresponds to the op. More...

Stage	operator[] (const Tensor &tensor)
	Short hand for getting the stage of tensor's operation. More...

Stage	create_group (const Array< Tensor > &outputs, const Array< Tensor > &inputs, bool include_inputs=false)
	Create a new stage group for all intermediate operations between inputs and outputs. More...

Tensor	cache_read (const Tensor &tensor, const std::string &scope, const Array< Operation > &readers)
	create a cache read of original tensor for readers. This will mutate the body of the readers. A new stage will be created for the tensor. More...

Array< Tensor >	cache_write (const Array< Tensor > &tensor, const std::string &scope)
	Create a cache write tensor for producing tensor. The tensor will take over body of original tensor op. More...

Tensor	cache_write (const Tensor &tensor, const std::string &scope)
	Create a cache write tensor for producing tensor. The tensor will take over body of original tensor op. More...

Array< Tensor >	rfactor (const Tensor &tensor, const IterVar &axis, int factor_axis=0)
	Factor a reduction axis in tensor's schedule to be an explicit axis. This will create a new stage that generated the new tensor with axis as the first dimension. The tensor's body will be rewritten as a reduction over the factored tensor. More...

Schedule	normalize ()
	Normalize the schedule. This is needed before bound inference. Insert necessary RebaseNode to make sure all leaf_iter_vars are in form [0, extent) More...

Schedule	normalize_for_feature_extraction ()
	Normalize the schedule for feature extraction in auto-scheduler. This is similar to `Schedule::normalize`, but we do aggressive simplification to the TE compute with const_matrix=True for faster compilation and feature extraction. The resulted schedule may be wrong, but it is good enough for feature extraction purposes. More...

const ScheduleNode *	operator-> () const
	access the internal node container More...

ScheduleNode *	operator-> ()
	access the internal node container More...

Public Member Functions inherited from tvm::runtime::ObjectRef
	ObjectRef ()=default
	default constructor More...

	ObjectRef (ObjectPtr< Object > data)
	Constructor from existing object ptr. More...

bool	same_as (const ObjectRef &other) const
	Comparator. More...

bool	operator== (const ObjectRef &other) const
	Comparator. More...

bool	operator!= (const ObjectRef &other) const
	Comparator. More...

bool	operator< (const ObjectRef &other) const
	Comparator. More...

bool	defined () const

const Object *	get () const

const Object *	operator-> () const

bool	unique () const

int	use_count () const

template<typename ObjectType , typename = std::enable_if_t<std::is_base_of_v<Object, ObjectType>>>
const ObjectType *	as () const
	Try to downcast the internal Object to a raw pointer of a corresponding type. More...

template<typename ObjectRefType , typename = std::enable_if_t<std::is_base_of_v<ObjectRef, ObjectRefType>>>
Optional< ObjectRefType >	as () const
	Try to downcast the ObjectRef to a Optional<T> of the requested type. More...

Additional Inherited Members
Static Public Attributes inherited from tvm::runtime::ObjectRef
static constexpr bool	_type_is_nullable = true

Protected Member Functions inherited from tvm::runtime::ObjectRef
Object *	get_mutable () const

Static Protected Member Functions inherited from tvm::runtime::ObjectRef
template<typename T >
static T	DowncastNoCheck (ObjectRef ref)
	Internal helper function downcast a ref without check. More...

static void	FFIClearAfterMove (ObjectRef *ref)
	Clear the object ref data field without DecRef after we successfully moved the field. More...

template<typename ObjectType >
static ObjectPtr< ObjectType >	GetDataPtr (const ObjectRef &ref)
	Internal helper function get data_ as ObjectPtr of ObjectType. More...

Protected Attributes inherited from tvm::runtime::ObjectRef
ObjectPtr< Object >	data_
	Internal pointer that backs the reference. More...

Detailed Description

Global schedule container For operations and all the operations they depend on. The schedule per Operation is named as stage.

Member Typedef Documentation

◆ ContainerType

using tvm::te::Schedule::ContainerType = ScheduleNode

Constructor & Destructor Documentation

◆ Schedule() [1/3]

tvm::te::Schedule::Schedule ( )

inline

◆ Schedule() [2/3]

tvm::te::Schedule::Schedule ( ObjectPtr< Object > n )

inlineexplicit

◆ Schedule() [3/3]

tvm::te::Schedule::Schedule ( Array< Operation > ops )

explicit

Create a schedule for array of ops(and their dependencies).

Parameters

ops	The ops to be scheduled.

Member Function Documentation

◆ cache_read()

Tensor tvm::te::Schedule::cache_read	(	const Tensor &	tensor,
		const std::string &	scope,
		const Array< Operation > &	readers
	)

create a cache read of original tensor for readers. This will mutate the body of the readers. A new stage will be created for the tensor.

Parameters

tensor	The tensor cached.
scope	The scope of the cache.
readers	The readers to redirect to the tensor.

Returns: The created tensor.

◆ cache_write() [1/2]

Array<Tensor> tvm::te::Schedule::cache_write	(	const Array< Tensor > &	tensor,
		const std::string &	scope
	)

Create a cache write tensor for producing tensor. The tensor will take over body of original tensor op.

This function can be used to do data layout transformation. If there is a split/fuse/reorder on the data parallel axis of tensor before cache_write is called. The intermediate cache stores the data in the layout as the iteration order of leave axis. The data will be transformed back to the original layout in the original tensor. User can further call compute_inline to inline the original layout and keep the data stored in the transformed layout.

Parameters

tensor	The tensors to be produced.
scope	The scope of the storage.

Returns: The created tensor.

◆ cache_write() [2/2]

Tensor tvm::te::Schedule::cache_write	(	const Tensor &	tensor,
		const std::string &	scope
	)

Create a cache write tensor for producing tensor. The tensor will take over body of original tensor op.

This function can be used to do data layout transformation. If there is a split/fuse/reorder on the data parallel axis of tensor before cache_write is called. The intermediate cache stores the data in the layout as the iteration order of leave axis. The data will be transformed back to the original layout in the original tensor. User can further call compute_inline to inline the original layout and keep the data stored in the transformed layout.

Parameters

tensor	The tensor to be produced.
scope	The scope of the storage.

Returns: The created tensor.

◆ copy()

Schedule tvm::te::Schedule::copy ( ) const

Get a copy of current schedule.

Returns: The copied schedule.

◆ create_group()

Stage tvm::te::Schedule::create_group	(	const Array< Tensor > &	outputs,
		const Array< Tensor > &	inputs,
		bool	include_inputs = `false`
	)

Create a new stage group for all intermediate operations between inputs and outputs.

Parameters

outputs	The output boundary of the group.
inputs	The input boundary of the group.
include_inputs	Whether include inputs if they are reachable from outputs.

Returns: The new grouped stage.

◆ normalize()

Schedule tvm::te::Schedule::normalize ( )

Normalize the schedule. This is needed before bound inference. Insert necessary RebaseNode to make sure all leaf_iter_vars are in form [0, extent)

Returns: A normalized schedule, can be same as current one.

◆ normalize_for_feature_extraction()

Schedule tvm::te::Schedule::normalize_for_feature_extraction ( )

Normalize the schedule for feature extraction in auto-scheduler. This is similar to Schedule::normalize, but we do aggressive simplification to the TE compute with const_matrix=True for faster compilation and feature extraction. The resulted schedule may be wrong, but it is good enough for feature extraction purposes.

Returns: A normalized schedule, can be same as current one.

◆ operator->() [1/2]

ScheduleNode * tvm::te::Schedule::operator-> ( )

inline

access the internal node container

Returns: the pointer to the internal node container

◆ operator->() [2/2]

const ScheduleNode * tvm::te::Schedule::operator-> ( ) const

inline

access the internal node container

Returns: the pointer to the internal node container

◆ operator[]() [1/2]

Stage tvm::te::Schedule::operator[] ( const Operation & op )

Get the stage corresponds to the op.

Parameters

op	The operation.

◆ operator[]() [2/2]

Stage tvm::te::Schedule::operator[] ( const Tensor & tensor )

inline

Short hand for getting the stage of tensor's operation.

Parameters

tensor The tensor

Returns: The stage corresponding to the tensor's op

◆ rfactor()

Array<Tensor> tvm::te::Schedule::rfactor	(	const Tensor &	tensor,
		const IterVar &	axis,
		int	factor_axis = `0`
	)

Factor a reduction axis in tensor's schedule to be an explicit axis. This will create a new stage that generated the new tensor with axis as the first dimension. The tensor's body will be rewritten as a reduction over the factored tensor.

P. Suriana, A. Adams and S. Kamil. Parallel associative reductions in halide. CGO'17

Parameters

tensor	The tensor to be factored.
axis	The reduction axis in tensor's schedule to be factored.
factor_axis	The position where the new axis is placed.

Returns: The created factored tensors.

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

include/tvm/te/schedule.h

Public Types

Public Member Functions

Additional Inherited Members

Detailed Description

Member Typedef Documentation

◆ ContainerType

Constructor & Destructor Documentation

◆ Schedule() [1/3]

◆ Schedule() [2/3]

◆ Schedule() [3/3]

Member Function Documentation

◆ cache_read()

◆ cache_write() [1/2]

◆ cache_write() [2/2]

◆ copy()

◆ create_group()

◆ normalize()

◆ normalize_for_feature_extraction()

◆ operator->() [1/2]

◆ operator->() [2/2]

◆ operator[]() [1/2]

◆ operator[]() [2/2]

◆ rfactor()