#include <State.h>

Classes
struct	FeatureLoopNestMutator

Public Member Functions
	State ()=default

	State (const State &)=delete

	State (State &&)=delete

void	operator= (const State &)=delete

void	operator= (State &&)=delete

uint64_t	structural_hash (int depth) const

void	compute_featurization (const FunctionDAG &dag, const Adams2019Params &params, StageMap< ScheduleFeatures > *features, const CachingOptions &cache_options)

void	save_featurization (const FunctionDAG &dag, const Adams2019Params &params, const CachingOptions &cache_options, std::ostream &out)

bool	calculate_cost (const FunctionDAG &dag, const Adams2019Params &params, CostModel *cost_model, const CachingOptions &cache_options, int verbosity=99)

IntrusivePtr< State >	make_child () const

void	generate_children (const FunctionDAG &dag, const Adams2019Params &params, CostModel cost_model, std::function< void(IntrusivePtr< State > &&)> &accept_child, Cache cache) const

void	dump (std::ostream &os) const

void	apply_schedule (const FunctionDAG &dag, const Adams2019Params &params)

	State ()=default

	State (const State &)=delete

	State (State &&)=delete

void	operator= (const State &)=delete

void	operator= (State &&)=delete

uint64_t	structural_hash (int depth) const

void	compute_loop_nest_parents (LoopNestMap &p, const LoopNest *here, int depth) const

const LoopNest *	deepest_common_ancestor (const LoopNestMap &parent, const LoopNest a, const LoopNest b) const

template<typename PostCreateMutator >
LoopNest *	create_feature_root (const PostCreateMutator &post_create_mutator) const

bool	has_loop_nest_without_thread_loops () const

bool	has_compute_root_loops_without_blocks () const

IntrusivePtr< const LoopNest >	get_root_for_features (const Anderson2021Params &params, const Target &target) const

void	set_gpu_store_site (const LoopNestMap &parent, const LoopNest *loop, LoopNest::Sites &site) const

bool	compute_featurization (const FunctionDAG &dag, const Anderson2021Params &params, const Target &target, StageMap< ScheduleFeatures > *features, Statistics &stats, bool verbose=false) const

void	save_featurization (const FunctionDAG &dag, const Anderson2021Params &params, const Target &target, std::ostream &out) const

bool	contains_store_at (const set< const FunctionDAG::Node * > &outermost_store_at, const IntrusivePtr< const LoopNest > &parent) const

bool	contains_store_at_further_in_than_outermost () const

bool	has_dynamic_allocation_inside_thread () const

bool	exceeds_serial_extents_limit (const Target &target) const

int64_t	get_shared_mem_alloc_size (const LoopNest block, const LoopNest loop) const

bool	exceeds_shared_memory_limit (const Anderson2021Params &params, const Target &target) const

bool	exceeds_local_memory_limit (const Anderson2021Params &params, const Target &target) const

bool	calculate_cost (const FunctionDAG &dag, const Anderson2021Params &params, const Target &target, CostModel *cost_model, Statistics &stats, bool verbose=false)

IntrusivePtr< State >	make_child () const

void	dump () const

void	print_compute_locations () const

void	fuse_gpu_blocks (LoopNest::StageScheduleState *state, Stage &stage, const vector< VarOrRVar > &parallel_vars, const vector< int64_t > &parallel_extents, const vector< int > &constant_extents) const

void	mark_gpu_blocks (LoopNest::StageScheduleState *state, Stage &stage, const vector< VarOrRVar > &parallel_vars, const vector< int64_t > &parallel_extents) const

bool	mark_gpu_threads (LoopNest::StageScheduleState *state, Stage &stage, std::unordered_set< std::string > &new_serial_vars, std::ostringstream &staged_funcs_schedule_source) const

bool	can_fuse_gpu (const vector< int64_t > &parallel_extents) const

void	apply_schedule (const FunctionDAG &dag, const Anderson2021Params &params, const Target &target)

bool	should_always_consider_inline (const FunctionDAG::Node *node) const

void	add_to_always_consider_inline_options (const FunctionDAG::Node *node)

void	update_always_consider_inline_options (const FunctionDAG::Node *node)

const LoopNest *	deepest_valid_compute_location (const Anderson2021Params &params, const LoopNestMap &parent, const FunctionDAG::Node &node, const LoopNest loop, const LoopNest root, StageMap< int64_t > &total_shared_mem_alloc_sizes) const

int64_t	total_loop_extents_of_ancestors (const LoopNestMap &parent, const LoopNest *loop) const

Public Attributes
RefCount	ref_count

IntrusivePtr< const LoopNest >	root

IntrusivePtr< const State >	parent

double	cost = 0

int	num_decisions_made = 0

bool	penalized = false

string	schedule_source

std::vector< double >	cost_per_stage

NodeMap< bool >	always_consider_inline

Static Public Attributes
static int	cost_calculations

Detailed Description

Definition at line 75 of file State.h.

Constructor & Destructor Documentation

◆ State() [1/6]

Halide::Internal::Autoscheduler::State::State ( )

default

◆ State() [2/6]

Halide::Internal::Autoscheduler::State::State ( const State & )

delete

◆ State() [3/6]

Halide::Internal::Autoscheduler::State::State ( State && )

delete

◆ State() [4/6]

Halide::Internal::Autoscheduler::State::State ( )

default

◆ State() [5/6]

Halide::Internal::Autoscheduler::State::State ( const State & )

delete

◆ State() [6/6]

Halide::Internal::Autoscheduler::State::State ( State && )

delete

Member Function Documentation

◆ operator=() [1/4]

void Halide::Internal::Autoscheduler::State::operator= ( const State & )

delete

◆ operator=() [2/4]

void Halide::Internal::Autoscheduler::State::operator= ( State && )

delete

◆ structural_hash() [1/2]

uint64_t Halide::Internal::Autoscheduler::State::structural_hash ( int depth ) const

◆ compute_featurization() [1/2]

void Halide::Internal::Autoscheduler::State::compute_featurization	(	const FunctionDAG &	dag,
		const Adams2019Params &	params,
		StageMap< ScheduleFeatures > *	features,
		const CachingOptions &	cache_options )

◆ save_featurization() [1/2]

void Halide::Internal::Autoscheduler::State::save_featurization	(	const FunctionDAG &	dag,
		const Adams2019Params &	params,
		const CachingOptions &	cache_options,
		std::ostream &	out )

◆ calculate_cost() [1/2]

bool Halide::Internal::Autoscheduler::State::calculate_cost	(	const FunctionDAG &	dag,
		const Adams2019Params &	params,
		CostModel *	cost_model,
		const CachingOptions &	cache_options,
		int	verbosity = 99 )

◆ make_child() [1/2]

IntrusivePtr< State > Halide::Internal::Autoscheduler::State::make_child ( ) const

◆ generate_children()

void Halide::Internal::Autoscheduler::State::generate_children	(	const FunctionDAG &	dag,
		const Adams2019Params &	params,
		CostModel *	cost_model,
		std::function< void(IntrusivePtr< State > &&)> &	accept_child,
		Cache *	cache ) const

◆ dump() [1/2]

void Halide::Internal::Autoscheduler::State::dump ( std::ostream & os ) const

◆ apply_schedule() [1/2]

void Halide::Internal::Autoscheduler::State::apply_schedule	(	const FunctionDAG &	dag,
		const Adams2019Params &	params )

◆ operator=() [3/4]

void Halide::Internal::Autoscheduler::State::operator= ( const State & )

delete

◆ operator=() [4/4]

void Halide::Internal::Autoscheduler::State::operator= ( State && )

delete

◆ structural_hash() [2/2]

uint64_t Halide::Internal::Autoscheduler::State::structural_hash ( int depth ) const

◆ compute_loop_nest_parents()

void Halide::Internal::Autoscheduler::State::compute_loop_nest_parents	(	LoopNestMap &	p,
		const LoopNest *	here,
		int	depth ) const

◆ deepest_common_ancestor()

const LoopNest * Halide::Internal::Autoscheduler::State::deepest_common_ancestor	(	const LoopNestMap &	parent,
		const LoopNest *	a,
		const LoopNest *	b ) const

◆ create_feature_root()

template<typename PostCreateMutator >

LoopNest * Halide::Internal::Autoscheduler::State::create_feature_root ( const PostCreateMutator & post_create_mutator ) const

inline

Definition at line 106 of file State.h.

References Halide::Internal::Autoscheduler::deep_copy_loop_nest(), and root.

◆ has_loop_nest_without_thread_loops()

bool Halide::Internal::Autoscheduler::State::has_loop_nest_without_thread_loops ( ) const

◆ has_compute_root_loops_without_blocks()

bool Halide::Internal::Autoscheduler::State::has_compute_root_loops_without_blocks ( ) const

◆ get_root_for_features()

IntrusivePtr< const LoopNest > Halide::Internal::Autoscheduler::State::get_root_for_features	(	const Anderson2021Params &	params,
		const Target &	target ) const

◆ set_gpu_store_site()

void Halide::Internal::Autoscheduler::State::set_gpu_store_site	(	const LoopNestMap &	parent,
		const LoopNest *	loop,
		LoopNest::Sites &	site ) const

◆ compute_featurization() [2/2]

bool Halide::Internal::Autoscheduler::State::compute_featurization	(	const FunctionDAG &	dag,
		const Anderson2021Params &	params,
		const Target &	target,
		StageMap< ScheduleFeatures > *	features,
		Statistics &	stats,
		bool	verbose = false ) const

◆ save_featurization() [2/2]

void Halide::Internal::Autoscheduler::State::save_featurization	(	const FunctionDAG &	dag,
		const Anderson2021Params &	params,
		const Target &	target,
		std::ostream &	out ) const

◆ contains_store_at()

bool Halide::Internal::Autoscheduler::State::contains_store_at	(	const set< const FunctionDAG::Node * > &	outermost_store_at,
		const IntrusivePtr< const LoopNest > &	parent ) const

◆ contains_store_at_further_in_than_outermost()

bool Halide::Internal::Autoscheduler::State::contains_store_at_further_in_than_outermost ( ) const

◆ has_dynamic_allocation_inside_thread()

bool Halide::Internal::Autoscheduler::State::has_dynamic_allocation_inside_thread ( ) const

◆ exceeds_serial_extents_limit()

bool Halide::Internal::Autoscheduler::State::exceeds_serial_extents_limit ( const Target & target ) const

◆ get_shared_mem_alloc_size()

int64_t Halide::Internal::Autoscheduler::State::get_shared_mem_alloc_size	(	const LoopNest *	block,
		const LoopNest *	loop ) const

◆ exceeds_shared_memory_limit()

bool Halide::Internal::Autoscheduler::State::exceeds_shared_memory_limit	(	const Anderson2021Params &	params,
		const Target &	target ) const

◆ exceeds_local_memory_limit()

bool Halide::Internal::Autoscheduler::State::exceeds_local_memory_limit	(	const Anderson2021Params &	params,
		const Target &	target ) const

◆ calculate_cost() [2/2]

bool Halide::Internal::Autoscheduler::State::calculate_cost	(	const FunctionDAG &	dag,
		const Anderson2021Params &	params,
		const Target &	target,
		CostModel *	cost_model,
		Statistics &	stats,
		bool	verbose = false )

◆ make_child() [2/2]

IntrusivePtr< State > Halide::Internal::Autoscheduler::State::make_child ( ) const

◆ dump() [2/2]

void Halide::Internal::Autoscheduler::State::dump ( ) const

◆ print_compute_locations()

void Halide::Internal::Autoscheduler::State::print_compute_locations ( ) const

◆ fuse_gpu_blocks()

void Halide::Internal::Autoscheduler::State::fuse_gpu_blocks	(	LoopNest::StageScheduleState *	state,
		Stage &	stage,
		const vector< VarOrRVar > &	parallel_vars,
		const vector< int64_t > &	parallel_extents,
		const vector< int > &	constant_extents ) const

◆ mark_gpu_blocks()

void Halide::Internal::Autoscheduler::State::mark_gpu_blocks	(	LoopNest::StageScheduleState *	state,
		Stage &	stage,
		const vector< VarOrRVar > &	parallel_vars,
		const vector< int64_t > &	parallel_extents ) const

◆ mark_gpu_threads()

bool Halide::Internal::Autoscheduler::State::mark_gpu_threads	(	LoopNest::StageScheduleState *	state,
		Stage &	stage,
		std::unordered_set< std::string > &	new_serial_vars,
		std::ostringstream &	staged_funcs_schedule_source ) const

◆ can_fuse_gpu()

bool Halide::Internal::Autoscheduler::State::can_fuse_gpu ( const vector< int64_t > & parallel_extents ) const

◆ apply_schedule() [2/2]

void Halide::Internal::Autoscheduler::State::apply_schedule	(	const FunctionDAG &	dag,
		const Anderson2021Params &	params,
		const Target &	target )

◆ should_always_consider_inline()

bool Halide::Internal::Autoscheduler::State::should_always_consider_inline ( const FunctionDAG::Node * node ) const

◆ add_to_always_consider_inline_options()

void Halide::Internal::Autoscheduler::State::add_to_always_consider_inline_options ( const FunctionDAG::Node * node )

◆ update_always_consider_inline_options()

void Halide::Internal::Autoscheduler::State::update_always_consider_inline_options ( const FunctionDAG::Node * node )

◆ deepest_valid_compute_location()

const LoopNest * Halide::Internal::Autoscheduler::State::deepest_valid_compute_location	(	const Anderson2021Params &	params,
		const LoopNestMap &	parent,
		const FunctionDAG::Node &	node,
		const LoopNest *	loop,
		const LoopNest *	root,
		StageMap< int64_t > &	total_shared_mem_alloc_sizes ) const

◆ total_loop_extents_of_ancestors()

int64_t Halide::Internal::Autoscheduler::State::total_loop_extents_of_ancestors	(	const LoopNestMap &	parent,
		const LoopNest *	loop ) const

Member Data Documentation

◆ ref_count

RefCount Halide::Internal::Autoscheduler::State::ref_count

mutable

Definition at line 22 of file State.h.

◆ root

IntrusivePtr< const LoopNest > Halide::Internal::Autoscheduler::State::root

Definition at line 24 of file State.h.

Referenced by create_feature_root().

◆ parent

IntrusivePtr< const State > Halide::Internal::Autoscheduler::State::parent

Definition at line 26 of file State.h.

◆ cost

double Halide::Internal::Autoscheduler::State::cost = 0

Definition at line 28 of file State.h.

◆ num_decisions_made

int Halide::Internal::Autoscheduler::State::num_decisions_made = 0

Definition at line 30 of file State.h.

◆ penalized

bool Halide::Internal::Autoscheduler::State::penalized = false

Definition at line 32 of file State.h.

◆ schedule_source

string Halide::Internal::Autoscheduler::State::schedule_source

Definition at line 36 of file State.h.

◆ cost_calculations

int Halide::Internal::Autoscheduler::State::cost_calculations

static

Definition at line 40 of file State.h.

◆ cost_per_stage

std::vector<double> Halide::Internal::Autoscheduler::State::cost_per_stage

Definition at line 80 of file State.h.

◆ always_consider_inline

NodeMap<bool> Halide::Internal::Autoscheduler::State::always_consider_inline

Definition at line 81 of file State.h.

The documentation for this struct was generated from the following files:

src/autoschedulers/adams2019/State.h
src/autoschedulers/anderson2021/State.h

Classes

Public Member Functions

Public Attributes

Static Public Attributes

Detailed Description

Constructor & Destructor Documentation

◆ State() [1/6]

◆ State() [2/6]

◆ State() [3/6]

◆ State() [4/6]

◆ State() [5/6]

◆ State() [6/6]

Member Function Documentation

◆ operator=() [1/4]

◆ operator=() [2/4]

◆ structural_hash() [1/2]

◆ compute_featurization() [1/2]

◆ save_featurization() [1/2]

◆ calculate_cost() [1/2]

◆ make_child() [1/2]

◆ generate_children()

◆ dump() [1/2]

◆ apply_schedule() [1/2]

◆ operator=() [3/4]

◆ operator=() [4/4]

◆ structural_hash() [2/2]

◆ compute_loop_nest_parents()

◆ deepest_common_ancestor()

◆ create_feature_root()

◆ has_loop_nest_without_thread_loops()

◆ has_compute_root_loops_without_blocks()

◆ get_root_for_features()

◆ set_gpu_store_site()

◆ compute_featurization() [2/2]

◆ save_featurization() [2/2]

◆ contains_store_at()

◆ contains_store_at_further_in_than_outermost()

◆ has_dynamic_allocation_inside_thread()

◆ exceeds_serial_extents_limit()

◆ get_shared_mem_alloc_size()

◆ exceeds_shared_memory_limit()

◆ exceeds_local_memory_limit()

◆ calculate_cost() [2/2]

◆ make_child() [2/2]

◆ dump() [2/2]

◆ print_compute_locations()

◆ fuse_gpu_blocks()

◆ mark_gpu_blocks()

◆ mark_gpu_threads()

◆ can_fuse_gpu()

◆ apply_schedule() [2/2]

◆ should_always_consider_inline()

◆ add_to_always_consider_inline_options()

◆ update_always_consider_inline_options()

◆ deepest_valid_compute_location()

◆ total_loop_extents_of_ancestors()

Member Data Documentation

◆ ref_count

◆ root

◆ parent

◆ cost

◆ num_decisions_made

◆ penalized

◆ schedule_source

◆ cost_calculations

◆ cost_per_stage

◆ always_consider_inline