#include <ginkgo/core/log/performance_hint.hpp>

Inheritance diagram for gko::log::PerformanceHint:

Public Member Functions
void	on_allocation_completed (const Executor *exec, const size_type &num_bytes, const uintptr &location) const override

void	on_free_completed (const Executor *exec, const uintptr &location) const override

void	on_copy_completed (const Executor from, const Executor to, const uintptr &location_from, const uintptr &location_to, const size_type &num_bytes) const override

void	print_status () const

Public Member Functions inherited from gko::log::Logger
template<size_type Event, typename... Params>
std::enable_if_t< Event==0 &&(0< event_count_max)>	on (Params &&... params) const

template<size_type Event, typename... Params>
std::enable_if_t< Event==1 &&(1< event_count_max)>	on (Params &&... params) const

template<size_type Event, typename... Params>
std::enable_if_t< Event==2 &&(2< event_count_max)>	on (Params &&... params) const

template<size_type Event, typename... Params>
std::enable_if_t< Event==3 &&(3< event_count_max)>	on (Params &&... params) const

template<size_type Event, typename... Params>
std::enable_if_t< Event==4 &&(4< event_count_max)>	on (Params &&... params) const

template<size_type Event, typename... Params>
std::enable_if_t< Event==5 &&(5< event_count_max)>	on (Params &&... params) const

template<size_type Event, typename... Params>
std::enable_if_t< Event==6 &&(6< event_count_max)>	on (Params &&... params) const

template<size_type Event, typename... Params>
std::enable_if_t< Event==7 &&(7< event_count_max)>	on (Params &&... params) const

template<size_type Event, typename... Params>
std::enable_if_t< Event==8 &&(8< event_count_max)>	on (Params &&... params) const

template<size_type Event, typename... Params>
std::enable_if_t< Event==9 &&(9< event_count_max)>	on (Params &&... params) const

template<size_type Event, typename... Params>
std::enable_if_t< Event==10 &&(10< event_count_max)>	on (Params &&... params) const

template<size_type Event, typename... Params>
std::enable_if_t< Event==11 &&(11< event_count_max)>	on (Params &&... params) const

template<size_type Event, typename... Params>
std::enable_if_t< Event==12 &&(12< event_count_max)>	on (Params &&... params) const

template<size_type Event, typename... Params>
std::enable_if_t< Event==13 &&(13< event_count_max)>	on (Params &&... params) const

template<size_type Event, typename... Params>
std::enable_if_t< Event==14 &&(14< event_count_max)>	on (Params &&... params) const

template<size_type Event, typename... Params>
std::enable_if_t< Event==15 &&(15< event_count_max)>	on (Params &&... params) const

template<size_type Event, typename... Params>
std::enable_if_t< Event==16 &&(16< event_count_max)>	on (Params &&... params) const

template<size_type Event, typename... Params>
std::enable_if_t< Event==17 &&(17< event_count_max)>	on (Params &&... params) const

template<size_type Event, typename... Params>
std::enable_if_t< Event==18 &&(18< event_count_max)>	on (Params &&... params) const

template<size_type Event, typename... Params>
std::enable_if_t< Event==19 &&(19< event_count_max)>	on (Params &&... params) const

template<size_type Event, typename... Params>
std::enable_if_t< Event==20 &&(20< event_count_max)>	on (Params &&... params) const

template<size_type Event, typename... Params>
std::enable_if_t< Event==21 &&(21< event_count_max)>	on (Params &&... params) const

template<size_type Event, typename... Params>
std::enable_if_t< Event==22 &&(22< event_count_max)>	on (Params &&... params) const

template<size_type Event, typename... Params>
std::enable_if_t< Event==23 &&(23< event_count_max)>	on (Params &&... params) const

template<size_type Event, typename... Params>
std::enable_if_t< Event==24 &&(24< event_count_max)>	on (Params &&... params) const

template<size_type Event, typename... Params>
std::enable_if_t< Event==25 &&(25< event_count_max)>	on (Params &&... params) const

template<size_type Event, typename... Params>
std::enable_if_t< Event==26 &&(26< event_count_max)>	on (Params &&... params) const

virtual bool	needs_propagation () const

Static Public Member Functions
static std::unique_ptr< PerformanceHint >	create (std::ostream &os=std::cerr, size_type allocation_size_limit=16, size_type copy_size_limit=16, size_type histogram_max_size=1024)

Protected Member Functions
	PerformanceHint (std::ostream &os, size_type allocation_size_limit, size_type copy_size_limit, size_type histogram_max_size)

Protected Member Functions inherited from gko::log::Logger
virtual void	on_allocation_started (const Executor *exec, const size_type &num_bytes) const

virtual void	on_free_started (const Executor *exec, const uintptr &location) const

virtual void	on_copy_started (const Executor exec_from, const Executor exec_to, const uintptr &loc_from, const uintptr &loc_to, const size_type &num_bytes) const

virtual void	on_operation_launched (const Executor exec, const Operation op) const

virtual void	on_operation_completed (const Executor exec, const Operation op) const

virtual void	on_polymorphic_object_create_started (const Executor exec, const PolymorphicObject po) const

virtual void	on_polymorphic_object_create_completed (const Executor exec, const PolymorphicObject input, const PolymorphicObject *output) const

virtual void	on_polymorphic_object_copy_started (const Executor exec, const PolymorphicObject input, const PolymorphicObject *output) const

virtual void	on_polymorphic_object_copy_completed (const Executor exec, const PolymorphicObject input, const PolymorphicObject *output) const

virtual void	on_polymorphic_object_deleted (const Executor exec, const PolymorphicObject po) const

virtual void	on_linop_apply_started (const LinOp A, const LinOp b, const LinOp *x) const

virtual void	on_linop_apply_completed (const LinOp A, const LinOp b, const LinOp *x) const

virtual void	on_linop_advanced_apply_started (const LinOp A, const LinOp alpha, const LinOp b, const LinOp beta, const LinOp *x) const

virtual void	on_linop_advanced_apply_completed (const LinOp A, const LinOp alpha, const LinOp b, const LinOp beta, const LinOp *x) const

virtual void	on_linop_factory_generate_started (const LinOpFactory factory, const LinOp input) const

virtual void	on_linop_factory_generate_completed (const LinOpFactory factory, const LinOp input, const LinOp *output) const

virtual void	on_criterion_check_started (const stop::Criterion criterion, const size_type &it, const LinOp r, const LinOp tau, const LinOp x, const uint8 &stopping_id, const bool &set_finalized) const

virtual void	on_criterion_check_completed (const stop::Criterion criterion, const size_type &it, const LinOp r, const LinOp tau, const LinOp x, const uint8 &stopping_id, const bool &set_finalized, const array< stopping_status > *status, const bool &one_changed, const bool &all_converged) const

virtual void	on_criterion_check_completed (const stop::Criterion criterion, const size_type &it, const LinOp r, const LinOp tau, const LinOp implicit_tau_sq, const LinOp x, const uint8 &stopping_id, const bool &set_finalized, const array< stopping_status > status, const bool &one_changed, const bool &all_converged) const

virtual void	on_iteration_complete (const LinOp solver, const size_type &it, const LinOp r, const LinOp x=nullptr, const LinOp tau=nullptr) const

virtual void	on_iteration_complete (const LinOp solver, const size_type &it, const LinOp r, const LinOp x, const LinOp tau, const LinOp *implicit_tau_sq) const

virtual void	on_iteration_complete (const LinOp solver, const LinOp b, const LinOp x, const size_type &it, const LinOp r, const LinOp tau, const LinOp implicit_tau_sq, const array< stopping_status > *status, bool stopped) const

virtual void	on_polymorphic_object_move_started (const Executor exec, const PolymorphicObject input, const PolymorphicObject *output) const

virtual void	on_polymorphic_object_move_completed (const Executor exec, const PolymorphicObject input, const PolymorphicObject *output) const

virtual void	on_batch_linop_factory_generate_started (const batch::BatchLinOpFactory factory, const batch::BatchLinOp input) const

virtual void	on_batch_linop_factory_generate_completed (const batch::BatchLinOpFactory factory, const batch::BatchLinOp input, const batch::BatchLinOp *output) const

virtual void	on_batch_solver_completed (const array< int > &iters, const array< double > &residual_norms) const

virtual void	on_batch_solver_completed (const array< int > &iters, const array< float > &residual_norms) const

virtual void	on_batch_solver_completed (const array< int > &iters, const array< gko::half > &residual_norms) const

	Logger (std::shared_ptr< const gko::Executor > exec, const mask_type &enabled_events=all_events_mask)

	Logger (const mask_type &enabled_events=all_events_mask)

Additional Inherited Members
Public Types inherited from gko::log::Logger
using	mask_type = gko::uint64

Static Public Attributes inherited from gko::log::Logger
static constexpr size_type	event_count_max = sizeof(mask_type) * byte_size

static constexpr mask_type	all_events_mask = ~mask_type{0}

static constexpr size_type	allocation_started { 0 }

static constexpr mask_type	allocation_started_mask {mask_type{1} << 0 }

static constexpr size_type	allocation_completed { 1 }

static constexpr mask_type	allocation_completed_mask {mask_type{1} << 1 }

static constexpr size_type	free_started { 2 }

static constexpr mask_type	free_started_mask {mask_type{1} << 2 }

static constexpr size_type	free_completed { 3 }

static constexpr mask_type	free_completed_mask {mask_type{1} << 3 }

static constexpr size_type	copy_started { 4 }

static constexpr mask_type	copy_started_mask {mask_type{1} << 4 }

static constexpr size_type	copy_completed { 5 }

static constexpr mask_type	copy_completed_mask {mask_type{1} << 5 }

static constexpr size_type	operation_launched { 6 }

static constexpr mask_type	operation_launched_mask {mask_type{1} << 6 }

static constexpr size_type	operation_completed { 7 }

static constexpr mask_type	operation_completed_mask {mask_type{1} << 7 }

static constexpr size_type	polymorphic_object_create_started { 8 }

static constexpr mask_type	polymorphic_object_create_started_mask {mask_type{1} << 8 }

static constexpr size_type	polymorphic_object_create_completed { 9 }

static constexpr mask_type	polymorphic_object_create_completed_mask {mask_type{1} << 9 }

static constexpr size_type	polymorphic_object_copy_started { 10 }

static constexpr mask_type	polymorphic_object_copy_started_mask {mask_type{1} << 10 }

static constexpr size_type	polymorphic_object_copy_completed { 11 }

static constexpr mask_type	polymorphic_object_copy_completed_mask {mask_type{1} << 11 }

static constexpr size_type	polymorphic_object_deleted { 12 }

static constexpr mask_type	polymorphic_object_deleted_mask {mask_type{1} << 12 }

static constexpr size_type	linop_apply_started { 13 }

static constexpr mask_type	linop_apply_started_mask {mask_type{1} << 13 }

static constexpr size_type	linop_apply_completed { 14 }

static constexpr mask_type	linop_apply_completed_mask {mask_type{1} << 14 }

static constexpr size_type	linop_advanced_apply_started { 15 }

static constexpr mask_type	linop_advanced_apply_started_mask {mask_type{1} << 15 }

static constexpr size_type	linop_advanced_apply_completed { 16 }

static constexpr mask_type	linop_advanced_apply_completed_mask {mask_type{1} << 16 }

static constexpr size_type	linop_factory_generate_started { 17 }

static constexpr mask_type	linop_factory_generate_started_mask {mask_type{1} << 17 }

static constexpr size_type	linop_factory_generate_completed { 18 }

static constexpr mask_type	linop_factory_generate_completed_mask {mask_type{1} << 18 }

static constexpr size_type	criterion_check_started { 19 }

static constexpr mask_type	criterion_check_started_mask {mask_type{1} << 19 }

static constexpr size_type	criterion_check_completed { 20 }

static constexpr mask_type	criterion_check_completed_mask {mask_type{1} << 20 }

static constexpr size_type	iteration_complete {21}

static constexpr mask_type	iteration_complete_mask {mask_type{1} << 21}

static constexpr size_type	polymorphic_object_move_started { 22 }

static constexpr mask_type	polymorphic_object_move_started_mask {mask_type{1} << 22 }

static constexpr size_type	polymorphic_object_move_completed { 23 }

static constexpr mask_type	polymorphic_object_move_completed_mask {mask_type{1} << 23 }

static constexpr size_type	batch_linop_factory_generate_started { 24 }

static constexpr mask_type	batch_linop_factory_generate_started_mask {mask_type{1} << 24 }

static constexpr size_type	batch_linop_factory_generate_completed { 25 }

static constexpr mask_type	batch_linop_factory_generate_completed_mask {mask_type{1} << 25 }

static constexpr size_type	batch_solver_completed {26}

static constexpr mask_type	batch_solver_completed_mask {mask_type{1} << 26}

static constexpr mask_type	executor_events_mask

static constexpr mask_type	operation_events_mask

static constexpr mask_type	polymorphic_object_events_mask

static constexpr mask_type	linop_events_mask

static constexpr mask_type	linop_factory_events_mask

static constexpr mask_type	batch_linop_factory_events_mask

static constexpr mask_type	criterion_events_mask

Detailed Description

PerformanceHint is a Logger which analyzes the performance of the application and outputs hints for unnecessary copies and allocations. The specific patterns it checks for are:

repeated cross-executor copies from or to the same pointer
repeated allocation/free pairs of the same size

Member Function Documentation

◆ create()

static std::unique_ptr< PerformanceHint > gko::log::PerformanceHint::create	(	std::ostream &	os = `std::cerr`,
		size_type	allocation_size_limit = `16`,
		size_type	copy_size_limit = `16`,
		size_type	histogram_max_size = `1024`
	)

inlinestatic

Creates a PerformanceHint logger. This dynamically allocates the memory, constructs the object and returns an std::unique_ptr to this object.

Parameters

os	the stream used for this logger
allocation_size_limit	ignore allocations below this limit (bytes)
copy_size_limit	ignore copies below this limit (bytes)
histogram_max_size	how many allocation sizes and/or pointers to keep track of at most?

Returns: an std::unique_ptr to the the constructed object

◆ on_allocation_completed()

void gko::log::PerformanceHint::on_allocation_completed	(	const Executor *	exec,
		const size_type &	num_bytes,
		const uintptr &	location
	)		const

overridevirtual

Executor's allocation completed event.

Parameters

exec	the executor used
num_bytes	the number of bytes allocated
location	the address at which the data was allocated

Reimplemented from gko::log::Logger.

◆ on_copy_completed()

void gko::log::PerformanceHint::on_copy_completed	(	const Executor *	exec_from,
		const Executor *	exec_to,
		const uintptr &	loc_from,
		const uintptr &	loc_to,
		const size_type &	num_bytes
	)		const

overridevirtual

Executor's copy completed event.

Parameters

exec_from	the executor copied from
exec_to	the executor copied to
loc_from	the address at which the data was copied from
loc_to	the address at which the data was copied to
num_bytes	the number of bytes copied

Reimplemented from gko::log::Logger.

◆ on_free_completed()

void gko::log::PerformanceHint::on_free_completed	(	const Executor *	exec,
		const uintptr &	location
	)		const

overridevirtual

Executor's free completed event.

Parameters

exec	the executor used
location	the address at which the data was freed

Reimplemented from gko::log::Logger.

◆ print_status()

void gko::log::PerformanceHint::print_status ( ) const

Writes out the cross-executor writes and allocations that have been stored so far.

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

ginkgo/core/log/performance_hint.hpp

PerformanceHint Class Reference

PerformanceHint Class Reference#

Public Member Functions

Static Public Member Functions

Protected Member Functions

Additional Inherited Members

Detailed Description

Member Function Documentation

◆ create()

◆ on_allocation_completed()

◆ on_copy_completed()

◆ on_free_completed()

◆ print_status()