tvm::meta_schedule Namespace Reference

Classes
class	ArgInfoNode
	The argument information. More...
class	ArgInfo
	Managed reference to ArgInfoNode. More...
class	TensorInfoNode
	The tensor argument information. More...
class	TensorInfo
	Managed reference to TensorInfoNode. More...
class	BuilderInputNode
	The builder's input, containing an IRModule and the target. More...
class	BuilderInput
	Managed reference to BuilderInputNode. More...
class	BuilderResultNode
	The builder's output, containing the artifact path or error message if any. More...
class	BuilderResult
	Managed reference to BuilderResultNode. More...
class	BuilderNode
	The abstract builder interface. More...
class	Builder
	Managed reference to BuilderNode. More...
class	PyBuilderNode
	An abstract builder with customized build method on the python-side. More...
class	CostModelNode
	Cost model. More...
class	PyCostModelNode
	The cost model with customized methods on the python-side. More...
class	CostModel
	Managed reference to CostModelNode. More...
class	WorkloadNode
	A workload, i.e. an IRModule and its structural hash. More...
class	Workload
	Managed reference to WorkloadNode. More...
struct	WorkloadHash
	The hash method for Workload. More...
struct	WorkloadEqual
	The equality check for Workload. More...
class	TuningRecordNode
	The class of tuning records. More...
class	TuningRecord
	The managed reference of TuningRecordNode. More...
class	DatabaseNode
class	PyDatabaseNode
	The database with customized methods on the python-side. More...
class	Database
	Managed reference to DatabaseNode. More...
class	ExtractedTaskNode
	A tuning task extracted from the high-level IR. More...
class	ExtractedTask
	Managed reference to ExtractedTaskNode. More...
class	FeatureExtractorNode
	Extractor for features from measure candidates for use in cost model. More...
class	PyFeatureExtractorNode
	The feature extractor with customized methods on the python-side. More...
class	FeatureExtractor
	Managed reference to FeatureExtractorNode. More...
class	MeasureCallbackNode
	Rules to apply after measure results is available. More...
class	PyMeasureCallbackNode
	The measure callback with customized methods on the python-side. More...
class	MeasureCallback
	Managed reference to MeasureCallbackNode. More...
class	MeasureCandidateNode
	The schedule (with input shapes) to be measured. More...
class	MeasureCandidate
	Managed reference to MeasureCandidateNode. More...
class	MutatorNode
	Mutator is designed to mutate the trace to explore the design space. More...
class	Mutator
	Managed reference to MutatorNode. More...
class	PyMutatorNode
	The mutator with customized methods on the python-side. More...
class	PostprocNode
	Rules to apply a postprocessor to a schedule. More...
class	Postproc
	Managed reference to PostprocNode. More...
class	PyPostprocNode
	The postprocessor with customized methods on the python-side. More...
class	ScopedTimer
class	ProfilerNode
	A generic profiler. More...
class	Profiler
	Managed reference to ProfilerNode. More...
class	RunnerInputNode
	Runner's input containing path of artifact, type of device and argument info. More...
class	RunnerInput
	Managed reference to RunnerInputNode. More...
class	RunnerResultNode
	Runner's output containing measurement result of MeasureCandidate or error msg if any. More...
class	RunnerResult
	Managed reference to RunnerResultNode. More...
class	RunnerFutureNode
	A class to asynchronously fetch runner's output. More...
class	RunnerFuture
	Managed reference to RunnerFutureNode. More...
class	RunnerNode
	The abstract runner interface. More...
class	Runner
	Managed reference to RunnerNode. More...
class	PyRunnerNode
	An abstract runner with customized build method on the python-side. More...
class	ScheduleRuleNode
	Rules to modify a block in a schedule. More...
class	ScheduleRule
	Managed reference to ScheduleRuleNode. More...
class	PyScheduleRuleNode
	The schedule rule with customized methods on the python-side. More...
class	SearchStrategyNode
	The search strategy for measure candidates generation. More...
class	SearchStrategy
	Managed reference to SearchStrategyNode. More...
class	PySearchStrategyNode
	The python side customizable class for measure candidate generation. More...
class	SpaceGeneratorNode
	The abstract class for design space generation. More...
class	SpaceGenerator
	Managed reference to SpaceGeneratorNode. More...
class	PySpaceGeneratorNode
	The design space generator with customized methods on the python-side. More...
class	TaskRecordNode
class	TaskRecord
	Managed reference to TaskRecordNode. More...
class	TaskSchedulerNode
	The abstract interface of task schedulers. More...
class	PyTaskSchedulerNode
	The task scheduler with customized methods on the python-side. More...
class	TaskScheduler
	Managed reference to TaskSchedulerNode. More...
class	TuneContextNode
	The auto tuning context. More...
class	TuneContext
	Managed reference to TuneContextNode. More...

Functions
std::function< tir::ExprRV(int64_t)>	MakeFactorSampler (tir::Schedule sch, Array< Integer > thread_extents)
	Given candidates of thread_extents, make a sampler that use sch->SampleCategorical to return a random thread extent. More...
Array< tir::LoopRV >	BindSpatialLoop (tir::Schedule sch, tir::LoopRV loop, int64_t max_threadblocks, int64_t max_threads_per_block, std::function< tir::ExprRV(int64_t)> get_factor=nullptr)
	Bind blockIdx.x and threadIdx.x to the given loop. More...
void	BindBlockThreadIdx (tir::Schedule sch, tir::BlockRV block, int64_t max_threadblocks, int64_t max_threads_per_block, std::function< tir::ExprRV(int64_t max_extent)> get_factor=nullptr)
	Bind the given block if it is not bound to blockIdx or threadIdx. More...
tir::BlockRV	GetWinogradProducerAndInlineConst (tir::Schedule sch, tir::BlockRV block)
	Get the producer block of a given block. If there is a constant winograd transform matrix, inline it. More...

Function Documentation

BindBlockThreadIdx()

void tvm::meta_schedule::BindBlockThreadIdx	(	tir::Schedule	sch,
		tir::BlockRV	block,
		int64_t	max_threadblocks,
		int64_t	max_threads_per_block,
		std::function< tir::ExprRV(int64_t max_extent)>	get_factor = nullptr
	)

Bind the given block if it is not bound to blockIdx or threadIdx.

Parameters

sch	The schedule.
block	The block to be bound.
max_threadblocks	The maximum number of threadblocks allowed.
max_threads_per_block	The maximum number of threads allowed.
get_factor	A function that returns the tiling factor.

BindSpatialLoop()

Array<tir::LoopRV> tvm::meta_schedule::BindSpatialLoop	(	tir::Schedule	sch,
		tir::LoopRV	loop,
		int64_t	max_threadblocks,
		int64_t	max_threads_per_block,
		std::function< tir::ExprRV(int64_t)>	get_factor = nullptr
	)

Bind blockIdx.x and threadIdx.x to the given loop.

Parameters

sch	The schedule.
loop	The loop to be bound.
max_threadblocks	The maximum number of threadblocks allowed.
max_threads_per_block	The maximum number of threads allowed.
get_factor	A function that returns the tiling factor.

Returns

The binded loops in the order of blockIdx.x, threadIdx.x, and the rest.

GetWinogradProducerAndInlineConst()

tir::BlockRV tvm::meta_schedule::GetWinogradProducerAndInlineConst	(	tir::Schedule	sch,
		tir::BlockRV	block
	)

Get the producer block of a given block. If there is a constant winograd transform matrix, inline it.

Returns

The only producer block.

MakeFactorSampler()

std::function<tir::ExprRV(int64_t)> tvm::meta_schedule::MakeFactorSampler	(	tir::Schedule	sch,
		Array< Integer >	thread_extents
	)

Given candidates of thread_extents, make a sampler that use sch->SampleCategorical to return a random thread extent.

Parameters

sch	The schedule
thread_extents	The candidate thread extents.

Returns

A sampler that returns a random thread extent.