tvm::meta_schedule::SearchStrategy Class Reference

Managed reference to SearchStrategyNode. More...

#include <search_strategy.h>

Inheritance diagram for tvm::meta_schedule::SearchStrategy:

Collaboration diagram for tvm::meta_schedule::SearchStrategy:

Public Types
using	FInitializeWithTuneContext = ffi::TypedFunction< void(const TuneContext &)>
	The function type of InitializeWithTuneContext method. More...
using	FPreTuning = ffi::TypedFunction< void(int max_trials, int num_trials_per_iter, const Array< tir::Schedule > &, const Optional< Database > &, const Optional< CostModel > &)>
	The function type of PreTuning method. More...
using	FPostTuning = ffi::TypedFunction< void()>
	The function type of PostTuning method. More...
using	FGenerateMeasureCandidates = ffi::TypedFunction< Optional< Array< MeasureCandidate > >()>
	The function type of GenerateMeasureCandidates method. More...
using	FNotifyRunnerResults = ffi::TypedFunction< void(const Array< MeasureCandidate > &, const Array< RunnerResult > &)>
	The function type of NotifyRunnerResults method. More...
using	FClone = ffi::TypedFunction< SearchStrategy()>
	The function type of Clone method. More...

Public Member Functions
	TVM_DEFINE_MUTABLE_OBJECT_REF_METHODS (SearchStrategy, ObjectRef, SearchStrategyNode)

Static Public Member Functions
static SearchStrategy	PySearchStrategy (FInitializeWithTuneContext f_initialize_with_tune_context, FPreTuning f_pre_tuning, FPostTuning f_post_tuning, FGenerateMeasureCandidates f_generate_measure_candidates, FNotifyRunnerResults f_notify_runner_results, FClone f_clone)
	Create a search strategy with customized methods on the python-side. More...
static SearchStrategy	ReplayTrace (int max_fail_count)
	Constructor of replay trace search strategy. More...
static SearchStrategy	ReplayFunc ()
	Constructor of replay func search strategy. More...
static SearchStrategy	EvolutionarySearch (int population_size, double init_measured_ratio, int init_min_unmeasured, int max_fail_count, int genetic_num_iters, double genetic_mutate_prob, int genetic_max_fail_count, double eps_greedy)
	Constructor of evolutionary search strategy. More...

Detailed Description

Managed reference to SearchStrategyNode.

See also

SearchStrategyNode

Member Typedef Documentation

FClone

using tvm::meta_schedule::SearchStrategy::FClone = ffi::TypedFunction<SearchStrategy()>

The function type of Clone method.

Returns

The cloned search strategy.

FGenerateMeasureCandidates

using tvm::meta_schedule::SearchStrategy::FGenerateMeasureCandidates = ffi::TypedFunction<Optional<Array<MeasureCandidate> >()>

The function type of GenerateMeasureCandidates method.

Returns

The measure candidates generated, nullptr if finished.

FInitializeWithTuneContext

using tvm::meta_schedule::SearchStrategy::FInitializeWithTuneContext = ffi::TypedFunction<void(const TuneContext&)>

The function type of InitializeWithTuneContext method.

Parameters

context

The tuning context for initialization.

FNotifyRunnerResults

using tvm::meta_schedule::SearchStrategy::FNotifyRunnerResults = ffi::TypedFunction<void(const Array<MeasureCandidate>&, const Array<RunnerResult>&)>

The function type of NotifyRunnerResults method.

Parameters

results

The measurement results from the runner.

FPostTuning

using tvm::meta_schedule::SearchStrategy::FPostTuning = ffi::TypedFunction<void()>

The function type of PostTuning method.

FPreTuning

using tvm::meta_schedule::SearchStrategy::FPreTuning = ffi::TypedFunction<void(int max_trials, int num_trials_per_iter, const Array<tir::Schedule>&, const Optional<Database>&, const Optional<CostModel>&)>

The function type of PreTuning method.

Member Function Documentation

EvolutionarySearch()

static SearchStrategy tvm::meta_schedule::SearchStrategy::EvolutionarySearch ( int  population_size, double  init_measured_ratio, int  init_min_unmeasured, int  max_fail_count, int  genetic_num_iters, double  genetic_mutate_prob, int  genetic_max_fail_count, double  eps_greedy  )

static

Constructor of evolutionary search strategy.

Parameters

population_size	The initial sample population.
init_measured_ratio	The ratio of measures samples in initial population.
init_min_unmeasured	The minimal size of unmeasured population in the initial sampling.
max_fail_count	The max number of failure during initial sampling.
genetic_num_iters	The iterations to run the genetic algorithm.
genetic_mutate_prob	The probability of mutation.
genetic_max_fail_count	The maximum number to try evolving the given trace.
eps_greedy	The ratio to select samples in a greedy fashion via their predicted score.

PySearchStrategy()

static SearchStrategy tvm::meta_schedule::SearchStrategy::PySearchStrategy ( FInitializeWithTuneContext  f_initialize_with_tune_context, FPreTuning  f_pre_tuning, FPostTuning  f_post_tuning, FGenerateMeasureCandidates  f_generate_measure_candidates, FNotifyRunnerResults  f_notify_runner_results, FClone  f_clone  )

static

Create a search strategy with customized methods on the python-side.

Parameters

f_initialize_with_tune_context	The packed function of InitializeWithTuneContext.
f_pre_tuning	The packed function of PreTuning.
f_post_tuning	The packed function of PostTuning.
f_generate_measure_candidates	The packed function of GenerateMeasureCandidates.
f_notify_runner_results	The packed function of NotifyRunnerResults.
f_clone	The packed function of Clone.

Returns

The search strategy created.

ReplayFunc()

static SearchStrategy tvm::meta_schedule::SearchStrategy::ReplayFunc ( )

static

Constructor of replay func search strategy.

ReplayTrace()

static SearchStrategy tvm::meta_schedule::SearchStrategy::ReplayTrace ( int  max_fail_count )

static

Constructor of replay trace search strategy.

Parameters

max_fail_count

The max number of failures during trace replaying.

TVM_DEFINE_MUTABLE_OBJECT_REF_METHODS()

tvm::meta_schedule::SearchStrategy::TVM_DEFINE_MUTABLE_OBJECT_REF_METHODS	(	SearchStrategy	,
		ObjectRef	,
		SearchStrategyNode
	)

---

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

• include/tvm/meta_schedule/search_strategy.h