Combines hardware counters and timers and executes the empirical roofline toolkit during application termination to estimate the peak possible performance for the machine. More...

#include "timemory/components/roofline/cpu_roofline.hpp"

Collaboration diagram for tim::component::cpu_roofline< Types >:

Public Types
enum class	MODE { OP , AI }

using	size_type = std::size_t

using	event_type = std::vector< int >

using	array_type = std::vector< long long >

using	data_type = long long *

using	value_type = std::pair< array_type, double >

using	this_type = cpu_roofline< Types... >

using	base_type = base< this_type, value_type >

using	storage_type = typename base_type::storage_type

using	record_type = std::function< value_type()>

using	unit_type = typename trait::units< this_type >::type

using	display_unit_type = typename trait::units< this_type >::display_type

using	device_t = device::cpu

using	count_type = wall_clock

using	ratio_t = typename count_type::ratio_t

using	types_tuple = std::tuple< Types... >

using	ert_data_t = ert::exec_data< count_type >

using	ert_data_ptr_t = std::shared_ptr< ert_data_t >

template<typename Tp >
using	ert_config_type = ert::configuration< device_t, Tp, count_type >

template<typename Tp >
using	ert_counter_type = ert::counter< device_t, Tp, count_type >

template<typename Tp >
using	ert_executor_type = ert::executor< device_t, Tp, count_type >

template<typename Tp >
using	ert_callback_type = ert::callback< ert_executor_type< Tp > >

using	ert_config_t = std::tuple< ert_config_type< Types >... >

using	ert_counter_t = std::tuple< ert_counter_type< Types >... >

using	ert_executor_t = std::tuple< ert_executor_type< Types >... >

using	ert_callback_t = std::tuple< ert_callback_type< Types >... >

using	iterator = typename array_type::iterator

using	const_iterator = typename array_type::const_iterator

using	strvec_t = std::vector< std::string >

using	intvec_t = std::vector< int >

using	events_callback_t = std::function< intvec_t(const MODE &)>

Public Member Functions
display_unit_type	display_unit ()

unit_type	get_unit ()

display_unit_type	get_display_unit ()

value_type	record ()

	cpu_roofline ()

	~cpu_roofline ()=default

	cpu_roofline (const cpu_roofline &rhs)=default

	cpu_roofline (cpu_roofline &&rhs) noexcept=default

cpu_roofline &	operator= (const cpu_roofline &)=default

cpu_roofline &	operator= (cpu_roofline &&) noexcept=default

std::vector< double >	get () const

void	start ()

void	stop ()

this_type &	operator+= (const this_type &rhs)

this_type &	operator-= (const this_type &rhs)

std::vector< double >	get_display () const

template<typename Archive >
void	load (Archive &ar, const unsigned int)

template<typename Archive >
void	save (Archive &ar, const unsigned int) const

strvec_t	label_array () const

strvec_t	description_array () const

strvec_t	display_unit_array () const

std::vector< int64_t >	unit_array () const

Static Public Member Functions
static events_callback_t &	get_events_callback ()
	replace this callback to add in custom HW counters More...

static bool &	use_predefined_enums ()
	set to false to suppress adding predefined enumerations More...

static MODE &	event_mode ()

static ert_config_t &	get_finalizer ()

static ert_data_ptr_t	get_ert_data ()

static event_type	get_events ()

static void	configure ()

static void	global_init ()

static void	thread_init ()

static void	thread_finalize ()

template<typename Tp , typename FuncT >
static void	set_executor_callback (FuncT &&f)

static void	global_finalize (storage_type *_store)

template<typename Archive >
static void	extra_serialization (Archive &ar)

static std::string	get_mode_string ()

static std::string	get_type_string ()

static unit_type	unit ()

static std::string	label ()

static std::string	description ()

static void	cleanup ()

Static Public Attributes
static const short	precision = 3

static const short	width = 8

Friends
struct	operation::record< this_type >

struct	operation::start< this_type >

struct	operation::stop< this_type >

struct	operation::set_started< this_type >

struct	operation::set_stopped< this_type >

class	impl::storage< this_type, trait::uses_value_storage< this_type, value_type >::value >

std::ostream &	operator<< (std::ostream &os, const this_type &obj)

Detailed Description

template<typename... Types>
struct tim::component::cpu_roofline< Types >

Combines hardware counters and timers and executes the empirical roofline toolkit during application termination to estimate the peak possible performance for the machine.

Template Parameters

Types Variadic list of data types for roofline analysis

Definition at line 72 of file cpu_roofline.hpp.

Member Typedef Documentation

◆ array_type

template<typename... Types>

using tim::component::cpu_roofline< Types >::array_type = std::vector<long long>

Definition at line 80 of file cpu_roofline.hpp.

◆ base_type

template<typename... Types>

using tim::component::cpu_roofline< Types >::base_type = base<this_type, value_type>

Definition at line 84 of file cpu_roofline.hpp.

◆ const_iterator

template<typename... Types>

using tim::component::cpu_roofline< Types >::const_iterator = typename array_type::const_iterator

Definition at line 126 of file cpu_roofline.hpp.

◆ count_type

template<typename... Types>

using tim::component::cpu_roofline< Types >::count_type = wall_clock

Definition at line 98 of file cpu_roofline.hpp.

◆ data_type

template<typename... Types>

using tim::component::cpu_roofline< Types >::data_type = long long*

Definition at line 81 of file cpu_roofline.hpp.

◆ device_t

template<typename... Types>

using tim::component::cpu_roofline< Types >::device_t = device::cpu

Definition at line 97 of file cpu_roofline.hpp.

◆ display_unit_type

template<typename... Types>

using tim::component::cpu_roofline< Types >::display_unit_type = typename trait::units<this_type>::display_type

Definition at line 95 of file cpu_roofline.hpp.

◆ ert_callback_t

template<typename... Types>

using tim::component::cpu_roofline< Types >::ert_callback_t = std::tuple<ert_callback_type<Types>...>

Definition at line 119 of file cpu_roofline.hpp.

◆ ert_callback_type

template<typename... Types>

template<typename Tp >

using tim::component::cpu_roofline< Types >::ert_callback_type = ert::callback<ert_executor_type<Tp> >

Definition at line 113 of file cpu_roofline.hpp.

◆ ert_config_t

template<typename... Types>

using tim::component::cpu_roofline< Types >::ert_config_t = std::tuple<ert_config_type<Types>...>

Definition at line 116 of file cpu_roofline.hpp.

◆ ert_config_type

template<typename... Types>

template<typename Tp >

using tim::component::cpu_roofline< Types >::ert_config_type = ert::configuration<device_t, Tp, count_type>

Definition at line 107 of file cpu_roofline.hpp.

◆ ert_counter_t

template<typename... Types>

using tim::component::cpu_roofline< Types >::ert_counter_t = std::tuple<ert_counter_type<Types>...>

Definition at line 117 of file cpu_roofline.hpp.

◆ ert_counter_type

template<typename... Types>

template<typename Tp >

using tim::component::cpu_roofline< Types >::ert_counter_type = ert::counter<device_t, Tp, count_type>

Definition at line 109 of file cpu_roofline.hpp.

◆ ert_data_ptr_t

template<typename... Types>

using tim::component::cpu_roofline< Types >::ert_data_ptr_t = std::shared_ptr<ert_data_t>

Definition at line 103 of file cpu_roofline.hpp.

◆ ert_data_t

template<typename... Types>

using tim::component::cpu_roofline< Types >::ert_data_t = ert::exec_data<count_type>

Definition at line 102 of file cpu_roofline.hpp.

◆ ert_executor_t

template<typename... Types>

using tim::component::cpu_roofline< Types >::ert_executor_t = std::tuple<ert_executor_type<Types>...>

Definition at line 118 of file cpu_roofline.hpp.

◆ ert_executor_type

template<typename... Types>

template<typename Tp >

using tim::component::cpu_roofline< Types >::ert_executor_type = ert::executor<device_t, Tp, count_type>

Definition at line 111 of file cpu_roofline.hpp.

◆ event_type

template<typename... Types>

using tim::component::cpu_roofline< Types >::event_type = std::vector<int>

Definition at line 79 of file cpu_roofline.hpp.

◆ events_callback_t

template<typename... Types>

using tim::component::cpu_roofline< Types >::events_callback_t = std::function<intvec_t(const MODE&)>

Definition at line 144 of file cpu_roofline.hpp.

◆ intvec_t

template<typename... Types>

using tim::component::cpu_roofline< Types >::intvec_t = std::vector<int>

Definition at line 143 of file cpu_roofline.hpp.

◆ iterator

template<typename... Types>

using tim::component::cpu_roofline< Types >::iterator = typename array_type::iterator

Definition at line 125 of file cpu_roofline.hpp.

◆ ratio_t

template<typename... Types>

using tim::component::cpu_roofline< Types >::ratio_t = typename count_type::ratio_t

Definition at line 99 of file cpu_roofline.hpp.

◆ record_type

template<typename... Types>

using tim::component::cpu_roofline< Types >::record_type = std::function<value_type()>

Definition at line 86 of file cpu_roofline.hpp.

◆ size_type

template<typename... Types>

using tim::component::cpu_roofline< Types >::size_type = std::size_t

Definition at line 78 of file cpu_roofline.hpp.

◆ storage_type

template<typename... Types>

using tim::component::cpu_roofline< Types >::storage_type = typename base_type::storage_type

Definition at line 85 of file cpu_roofline.hpp.

◆ strvec_t

template<typename... Types>

using tim::component::cpu_roofline< Types >::strvec_t = std::vector<std::string>

Definition at line 142 of file cpu_roofline.hpp.

◆ this_type

template<typename... Types>

using tim::component::cpu_roofline< Types >::this_type = cpu_roofline<Types...>

Definition at line 83 of file cpu_roofline.hpp.

◆ types_tuple

template<typename... Types>

using tim::component::cpu_roofline< Types >::types_tuple = std::tuple<Types...>

Definition at line 100 of file cpu_roofline.hpp.

◆ unit_type

template<typename... Types>

using tim::component::cpu_roofline< Types >::unit_type = typename trait::units<this_type>::type

Definition at line 94 of file cpu_roofline.hpp.

◆ value_type

template<typename... Types>

using tim::component::cpu_roofline< Types >::value_type = std::pair<array_type, double>

Definition at line 82 of file cpu_roofline.hpp.

Member Enumeration Documentation

◆ MODE

template<typename... Types>

enum tim::component::cpu_roofline::MODE

strong

Enumerator
OP
AI

Definition at line 134 of file cpu_roofline.hpp.

     {
         OP,
         AI
     };

Constructor & Destructor Documentation

◆ cpu_roofline() [1/3]

template<typename... Types>

tim::component::cpu_roofline< Types >::cpu_roofline ( )

inline

Definition at line 414 of file cpu_roofline.hpp.

     : base_type()
     {
         configure();
         m_papi_vector                        = std::make_shared<papi_vector>();
         m_wall_clock                         = std::make_shared<wall_clock>();
         std::tie(value.second, accum.second) = std::make_pair(0, 0);
     }

References tim::component::cpu_roofline< Types >::configure().

◆ ~cpu_roofline()

template<typename... Types>

tim::component::cpu_roofline< Types >::~cpu_roofline ( )

default

◆ cpu_roofline() [2/3]

template<typename... Types>

tim::component::cpu_roofline< Types >::cpu_roofline ( const cpu_roofline< Types > & rhs )

default

◆ cpu_roofline() [3/3]

template<typename... Types>

tim::component::cpu_roofline< Types >::cpu_roofline ( cpu_roofline< Types > && rhs )

defaultnoexcept

Member Function Documentation

◆ cleanup()

template<typename... Types>

static void tim::component::cpu_roofline< Types >::cleanup ( )

inlinestatic

Definition at line 664 of file cpu_roofline.hpp.

664 {}

◆ configure()

template<typename... Types>

static void tim::component::cpu_roofline< Types >::configure ( )

inlinestatic

Definition at line 269 of file cpu_roofline.hpp.

     {
         if(!is_configured())
         {
             if(settings::debug() || settings::verbose() > 1)
                 PRINT_HERE("%s", "configuring cpu_roofline");
  
             // do this BEFORE setting is_configured to true to ensure mode is updated
             // properly
             auto _events    = get_events();
             is_configured() = true;
             for(auto&& itr : _events)
                 papi_vector::add_event(itr);
             papi_vector::configure();
         }
     }

References tim::component::papi_common::add_event(), tim::component::papi_vector::configure(), tim::debug, tim::component::cpu_roofline< Types >::get_events(), PRINT_HERE, and tim::verbose.

Referenced by tim::component::cpu_roofline< Types >::cpu_roofline(), tim::component::cpu_roofline< Types >::global_init(), and tim::component::cpu_roofline< Types >::thread_init().

◆ description()

template<typename... Types>

static std::string tim::component::cpu_roofline< Types >::description ( )

inlinestatic

Definition at line 398 of file cpu_roofline.hpp.

     {
         return "Model used to provide performance relative to the peak possible "
                "performance on a CPU architecture.";
     }

◆ description_array()

template<typename... Types>

strvec_t tim::component::cpu_roofline< Types >::description_array ( ) const

inline

Definition at line 628 of file cpu_roofline.hpp.

     {
         strvec_t arr = m_papi_vector->description_array();
         arr.push_back("Runtime");
         return arr;
     }

◆ display_unit()

template<typename... Types>

display_unit_type tim::component::cpu_roofline< Types >::display_unit ( )

inline

Definition at line 367 of file cpu_roofline.hpp.

     {
         auto _units = m_papi_vector->display_unit_array();
         _units.push_back(m_wall_clock->display_unit());
         return _units;
     }

Referenced by tim::component::cpu_roofline< Types >::get_display_unit().

◆ display_unit_array()

template<typename... Types>

strvec_t tim::component::cpu_roofline< Types >::display_unit_array ( ) const

inline

Definition at line 637 of file cpu_roofline.hpp.

     {
         strvec_t arr = m_papi_vector->display_unit_array();
         arr.push_back(count_type::get_display_unit());
         return arr;
     }

References tim::component::base< wall_clock, int64_t >::get_display_unit().

◆ event_mode()

template<typename... Types>

static MODE& tim::component::cpu_roofline< Types >::event_mode ( )

inlinestatic

Definition at line 164 of file cpu_roofline.hpp.

     {
         auto&& _get = []() {
             auto&& aslc = [](std::string str) {
                 for(auto& itr : str)
                     itr = tolower(itr);
                 return str;
             };
             // check the standard variable
             std::string _env = aslc(settings::cpu_roofline_mode());
             if(_env.empty())
                 _env = aslc(settings::roofline_mode());
             auto _val =
                 (_env == "op" || _env == "hw" || _env == "counters")
                     ? MODE::OP
                     : ((_env == "ai" || _env == "ac" || _env == "activity") ? MODE::AI
                                                                             : MODE::OP);
             if(settings::verbose() > 1 || settings::debug())
             {
                 std::cerr << "[" << demangle<this_type>()
                           << "]> roofline mode: " << ((_val == MODE::OP) ? "op" : "ai")
                           << std::endl;
             }
             return _val;
         };
  
         static MODE _instance = _get();
         if(!is_configured())
             _instance = _get();
         return _instance;
     }

References tim::component::cpu_roofline< Types >::AI, tim::cpu_roofline_mode, tim::debug, tim::component::cpu_roofline< Types >::OP, tim::roofline_mode, and tim::verbose.

Referenced by tim::component::cpu_roofline< Types >::get_events(), tim::component::cpu_roofline< Types >::get_mode_string(), and tim::component::cpu_roofline< Types >::unit().

◆ extra_serialization()

template<typename... Types>

template<typename Archive >

static void tim::component::cpu_roofline< Types >::extra_serialization ( Archive & ar )

inlinestatic

Definition at line 336 of file cpu_roofline.hpp.

     {
         auto _ert_data = get_ert_data();
         if(!_ert_data)  // for input
             _ert_data = std::make_shared<ert_data_t>();
         ar(cereal::make_nvp("roofline", *_ert_data));
     }

References tim::component::cpu_roofline< Types >::get_ert_data().

◆ get()

template<typename... Types>

std::vector<double> tim::component::cpu_roofline< Types >::get ( ) const

inline

Definition at line 431 of file cpu_roofline.hpp.

     {
         auto _data = m_papi_vector->get();
         _data.push_back(m_wall_clock->get());
         return _data;
     }

Referenced by tim::component::cpu_roofline< Types >::get_display(), and tim::component::cpu_roofline< Types >::save().

◆ get_display()

template<typename... Types>

std::vector<double> tim::component::cpu_roofline< Types >::get_display ( ) const

inline

Definition at line 503 of file cpu_roofline.hpp.

503 { return get(); }

tim::component::cpu_roofline::get

std::vector< double > get() const

Definition: cpu_roofline.hpp:431

References tim::component::cpu_roofline< Types >::get().

Referenced by tim::component::cpu_roofline< Types >::load(), and tim::component::cpu_roofline< Types >::save().

◆ get_display_unit()

template<typename... Types>

display_unit_type tim::component::cpu_roofline< Types >::get_display_unit ( )

inline

Definition at line 380 of file cpu_roofline.hpp.

380 { return display_unit(); }

tim::component::cpu_roofline::display_unit

display_unit_type display_unit()

Definition: cpu_roofline.hpp:367

References tim::component::cpu_roofline< Types >::display_unit().

◆ get_ert_data()

template<typename... Types>

static ert_data_ptr_t tim::component::cpu_roofline< Types >::get_ert_data ( )

inlinestatic

Definition at line 206 of file cpu_roofline.hpp.

     {
         static ert_data_ptr_t _instance = std::make_shared<ert_data_t>();
         return _instance;
     }

Referenced by tim::component::cpu_roofline< Types >::extra_serialization(), and tim::component::cpu_roofline< Types >::global_finalize().

◆ get_events()

template<typename... Types>

static event_type tim::component::cpu_roofline< Types >::get_events ( )

inlinestatic

Definition at line 214 of file cpu_roofline.hpp.

     {
         static auto _instance = []() {
             event_type _events;
             auto       _mode = event_mode();
             if(_mode == MODE::OP)
             {
                 //
                 // add in user callback events BEFORE presets based on type so that
                 // the user can override the counters being used
                 //
                 auto _extra_events = get_events_callback()(_mode);
                 for(const auto& itr : _extra_events)
                     _events.push_back(itr);
  
                 //
                 //  add some presets based on data types
                 //
                 if(use_predefined_enums())
                 {
                     if(is_one_of<float, types_tuple>::value)
                         _events.push_back(PAPI_SP_OPS);
                     if(is_one_of<double, types_tuple>::value)
                         _events.push_back(PAPI_DP_OPS);
                 }
             }
             else if(_mode == MODE::AI)
             {
                 //
                 //  add the load/store hardware counter
                 //
                 if(use_predefined_enums())
                 {
                     _events.push_back(PAPI_LD_INS);
                     _events.push_back(PAPI_SR_INS);
                     _events.push_back(PAPI_LST_INS);
                     _events.push_back(PAPI_TOT_INS);
                 }
                 //
                 // add in user callback events AFTER load/store so that load/store
                 // instructions are always counted
                 //
                 auto _extra_events = get_events_callback()(_mode);
                 for(const auto& itr : _extra_events)
                     _events.push_back(itr);
             }
  
             return _events;
         }();
  
         return _instance;
     }

References tim::component::cpu_roofline< Types >::AI, tim::component::cpu_roofline< Types >::event_mode(), tim::component::cpu_roofline< Types >::get_events_callback(), tim::component::cpu_roofline< Types >::OP, and tim::component::cpu_roofline< Types >::use_predefined_enums().

Referenced by tim::component::cpu_roofline< Types >::configure().

◆ get_events_callback()

template<typename... Types>

static events_callback_t& tim::component::cpu_roofline< Types >::get_events_callback ( )

inlinestatic

replace this callback to add in custom HW counters

Definition at line 148 of file cpu_roofline.hpp.

     {
         static events_callback_t _instance = [](const MODE&) { return intvec_t{}; };
         return _instance;
     }

Referenced by tim::component::cpu_roofline< Types >::get_events().

◆ get_finalizer()

template<typename... Types>

static ert_config_t& tim::component::cpu_roofline< Types >::get_finalizer ( )

inlinestatic

Definition at line 198 of file cpu_roofline.hpp.

     {
         static ert_config_t _instance;
         return _instance;
     }

Referenced by tim::component::cpu_roofline< Types >::global_finalize().

◆ get_mode_string()

template<typename... Types>

static std::string tim::component::cpu_roofline< Types >::get_mode_string ( )

inlinestatic

Definition at line 346 of file cpu_roofline.hpp.

     {
         return (event_mode() == MODE::OP) ? "op" : "ai";
     }

References tim::component::cpu_roofline< Types >::event_mode(), and tim::component::cpu_roofline< Types >::OP.

Referenced by tim::component::cpu_roofline< Types >::label(), and tim::component::cpu_roofline< Types >::save().

◆ get_type_string()

template<typename... Types>

static std::string tim::component::cpu_roofline< Types >::get_type_string ( )

inlinestatic

Definition at line 353 of file cpu_roofline.hpp.

     {
         return mpl::apply<std::string>::join('_', demangle(typeid(Types).name())...);
     }

References tim::demangle(), and tim::mpl::apply< Ret >::join().

Referenced by tim::component::cpu_roofline< Types >::label(), and tim::component::cpu_roofline< Types >::save().

◆ get_unit()

template<typename... Types>

unit_type tim::component::cpu_roofline< Types >::get_unit ( )

inline

Definition at line 376 of file cpu_roofline.hpp.

376 { return unit(); }

tim::component::cpu_roofline::unit

static unit_type unit()

Definition: cpu_roofline.hpp:360

References tim::component::cpu_roofline< Types >::unit().

◆ global_finalize()

template<typename... Types>

static void tim::component::cpu_roofline< Types >::global_finalize ( storage_type * _store )

inlinestatic

Definition at line 318 of file cpu_roofline.hpp.

     {
         // query environment for whether this is part of CI test
         // auto ci = get_env<bool>("CONTINUOUS_INTEGRATION", false);
         if(_store && _store->size() > 0)
         {
             // run roofline peak generation
             auto ert_config = get_finalizer();
             auto ert_data   = get_ert_data();
             mpl::apply<void>::access<ert_executor_t>(ert_config, ert_data);
             if(ert_data && (settings::verbose() > 1 || settings::debug()))
                 std::cout << *(ert_data) << std::endl;
         }
     }

References tim::debug, tim::component::cpu_roofline< Types >::get_ert_data(), tim::component::cpu_roofline< Types >::get_finalizer(), and tim::verbose.

◆ global_init()

template<typename... Types>

static void tim::component::cpu_roofline< Types >::global_init ( )

inlinestatic

Definition at line 288 of file cpu_roofline.hpp.

     {
         if(settings::debug() || settings::verbose() > 2)
             PRINT_HERE("%s", "global initialization of cpu_roofline");
         configure();
     }

References tim::component::cpu_roofline< Types >::configure(), tim::debug, PRINT_HERE, and tim::verbose.

◆ label()

template<typename... Types>

static std::string tim::component::cpu_roofline< Types >::label ( )

inlinestatic

Definition at line 384 of file cpu_roofline.hpp.

     {
         if(settings::roofline_type_labels_cpu() || settings::roofline_type_labels())
         {
             return std::string("cpu_roofline_") + get_type_string() + "_" +
                    get_mode_string();
         }
         {
             return std::string("cpu_roofline_") + get_mode_string();
         }
     }

References tim::component::cpu_roofline< Types >::get_mode_string(), tim::component::cpu_roofline< Types >::get_type_string(), tim::roofline_type_labels, and tim::roofline_type_labels_cpu.

◆ label_array()

template<typename... Types>

strvec_t tim::component::cpu_roofline< Types >::label_array ( ) const

inline

Definition at line 618 of file cpu_roofline.hpp.

     {
         strvec_t arr = m_papi_vector->label_array();
         arr.push_back("Runtime");
         return arr;
     }

Referenced by tim::component::cpu_roofline< Types >::load(), and tim::component::cpu_roofline< Types >::save().

◆ load()

template<typename... Types>

template<typename Archive >

void tim::component::cpu_roofline< Types >::load	(	Archive &	ar,
		const unsigned int
	)

inline

Definition at line 575 of file cpu_roofline.hpp.

     {
         auto _disp  = get_display();
         auto labels = label_array();
  
         ar(cereal::make_nvp("laps", laps), cereal::make_nvp("labels", labels),
            cereal::make_nvp("papi_vector", m_papi_vector));
         ar(cereal::make_nvp("value", value));
         ar(cereal::make_nvp("accum", accum));
     }

References tim::component::cpu_roofline< Types >::get_display(), tim::component::cpu_roofline< Types >::label_array(), and tim::component::base< cpu_roofline< Types... >, std::pair< std::vector< long long >, double > >::laps.

◆ operator+=()

template<typename... Types>

this_type& tim::component::cpu_roofline< Types >::operator+= ( const this_type & rhs )

inline

Definition at line 459 of file cpu_roofline.hpp.

     {
         if(rhs.value.first.size() > value.first.size())
             value.first.resize(rhs.value.first.size());
         if(rhs.accum.first.size() > accum.first.size())
             accum.first.resize(rhs.accum.first.size());
         value += rhs.value;
         accum += rhs.accum;
         return *this;
     }

◆ operator-=()

template<typename... Types>

this_type& tim::component::cpu_roofline< Types >::operator-= ( const this_type & rhs )

inline

Definition at line 472 of file cpu_roofline.hpp.

     {
         if(rhs.value.first.size() > value.first.size())
             value.first.resize(rhs.value.first.size());
         if(rhs.accum.first.size() > accum.first.size())
             accum.first.resize(rhs.accum.first.size());
         value -= rhs.value;
         accum -= rhs.accum;
         return *this;
     }

◆ operator=() [1/2]

template<typename... Types>

cpu_roofline& tim::component::cpu_roofline< Types >::operator= ( const cpu_roofline< Types > & )

default

◆ operator=() [2/2]

template<typename... Types>

cpu_roofline& tim::component::cpu_roofline< Types >::operator= ( cpu_roofline< Types > && )

defaultnoexcept

◆ record()

template<typename... Types>

value_type tim::component::cpu_roofline< Types >::record ( )

inline

Definition at line 406 of file cpu_roofline.hpp.

     {
         auto hwcount  = m_papi_vector->record();
         auto duration = m_wall_clock->record();
         return value_type(hwcount, duration);
     }

◆ save()

template<typename... Types>

template<typename Archive >

void tim::component::cpu_roofline< Types >::save	(	Archive &	ar,
		const unsigned int
	)		const

inline

Definition at line 589 of file cpu_roofline.hpp.

     {
         auto _disp  = get_display();
         auto labels = label_array();
  
         ar(cereal::make_nvp("laps", laps), cereal::make_nvp("display", _disp),
            cereal::make_nvp("mode", get_mode_string()),
            cereal::make_nvp("type", get_type_string()),
            cereal::make_nvp("labels", labels),
            cereal::make_nvp("papi_vector", m_papi_vector));
  
         auto data = get();
         ar.setNextName("repr_data");
         ar.startNode();
         auto litr = labels.begin();
         auto ditr = data.begin();
         for(; litr != labels.end() && ditr != data.end(); ++litr, ++ditr)
             ar(cereal::make_nvp(*litr, double(*ditr)));
         ar.finishNode();
  
         ar(cereal::make_nvp("value", value));
         ar(cereal::make_nvp("accum", accum));
         // ar(cereal::make_nvp("units", unit_array()));
         // ar(cereal::make_nvp("display_units", display_unit_array()));
     }

References tim::component::cpu_roofline< Types >::get(), tim::component::cpu_roofline< Types >::get_display(), tim::component::cpu_roofline< Types >::get_mode_string(), tim::component::cpu_roofline< Types >::get_type_string(), tim::component::cpu_roofline< Types >::label_array(), and tim::component::base< cpu_roofline< Types... >, std::pair< std::vector< long long >, double > >::laps.

◆ set_executor_callback()

template<typename... Types>

template<typename Tp , typename FuncT >

static void tim::component::cpu_roofline< Types >::set_executor_callback ( FuncT && f )

inlinestatic

Definition at line 311 of file cpu_roofline.hpp.

     {
         ert_executor_type<Tp>::get_callback() = std::forward<FuncT>(f);
     }

References tim::ert::executor< DeviceT, Tp, CounterT >::get_callback().

◆ start()

template<typename... Types>

void tim::component::cpu_roofline< Types >::start ( )

inline

Definition at line 440 of file cpu_roofline.hpp.

     {
         m_wall_clock->start();
         m_papi_vector->start();
         value = value_type{ m_papi_vector->get_value(), m_wall_clock->get_value() };
     }

◆ stop()

template<typename... Types>

void tim::component::cpu_roofline< Types >::stop ( )

inline

Definition at line 449 of file cpu_roofline.hpp.

     {
         m_papi_vector->stop();
         m_wall_clock->stop();
         value = value_type{ m_papi_vector->get_value(), m_wall_clock->get_value() };
         accum += value_type{ m_papi_vector->get_accum(), m_wall_clock->get_accum() };
     }

◆ thread_finalize()

template<typename... Types>

static void tim::component::cpu_roofline< Types >::thread_finalize ( )

inlinestatic

Definition at line 306 of file cpu_roofline.hpp.

306 {}

◆ thread_init()

template<typename... Types>

static void tim::component::cpu_roofline< Types >::thread_init ( )

inlinestatic

Definition at line 297 of file cpu_roofline.hpp.

     {
         if(settings::debug() || settings::verbose() > 2)
             PRINT_HERE("%s", "thread initialization of cpu_roofline");
         configure();
     }

References tim::component::cpu_roofline< Types >::configure(), tim::debug, PRINT_HERE, and tim::verbose.

◆ unit()

template<typename... Types>

static unit_type tim::component::cpu_roofline< Types >::unit ( )

inlinestatic

Definition at line 360 of file cpu_roofline.hpp.

     {
         return (event_mode() == MODE::OP) ? (1.0 / count_type::unit()) : 1.0;
     }

References tim::component::cpu_roofline< Types >::event_mode(), tim::component::cpu_roofline< Types >::OP, and tim::component::base< wall_clock, int64_t >::unit().

Referenced by tim::component::cpu_roofline< Types >::get_unit().

◆ unit_array()

template<typename... Types>

std::vector<int64_t> tim::component::cpu_roofline< Types >::unit_array ( ) const

inline

Definition at line 647 of file cpu_roofline.hpp.

     {
         auto arr = m_papi_vector->unit_array();
         arr.push_back(count_type::get_unit());
         return arr;
     }

References tim::component::base< wall_clock, int64_t >::get_unit().

◆ use_predefined_enums()

template<typename... Types>

static bool& tim::component::cpu_roofline< Types >::use_predefined_enums ( )

inlinestatic

set to false to suppress adding predefined enumerations

Definition at line 156 of file cpu_roofline.hpp.

     {
         static bool _instance = true;
         return _instance;
     }

Referenced by tim::component::cpu_roofline< Types >::get_events().

Friends And Related Function Documentation

◆ impl::storage< this_type, trait::uses_value_storage< this_type, value_type >::value >

template<typename... Types>

friend class impl::storage< this_type, trait::uses_value_storage< this_type, value_type >::value >

friend

Definition at line 472 of file cpu_roofline.hpp.

◆ operation::record< this_type >

template<typename... Types>

friend struct operation::record< this_type >

friend

Definition at line 75 of file cpu_roofline.hpp.

◆ operation::set_started< this_type >

template<typename... Types>

friend struct operation::set_started< this_type >

friend

Definition at line 75 of file cpu_roofline.hpp.

◆ operation::set_stopped< this_type >

template<typename... Types>

friend struct operation::set_stopped< this_type >

friend

Definition at line 75 of file cpu_roofline.hpp.

◆ operation::start< this_type >

template<typename... Types>

friend struct operation::start< this_type >

friend

Definition at line 75 of file cpu_roofline.hpp.

◆ operation::stop< this_type >

template<typename... Types>

friend struct operation::stop< this_type >

friend

Definition at line 75 of file cpu_roofline.hpp.

◆ operator<<

template<typename... Types>

std::ostream& operator<<	(	std::ostream &	os,
		const this_type &	obj
	)

friend

Definition at line 507 of file cpu_roofline.hpp.

     {
         using namespace tim::stl::ostream;
  
         // output the time
         auto&             _obj = obj.load();
         std::stringstream sst;
         auto              t_value = _obj.second;
         auto              t_label = count_type::get_label();
         auto              t_disp  = count_type::get_display_unit();
         auto              t_prec  = count_type::get_precision();
         auto              t_width = count_type::get_width();
         auto              t_flags = count_type::get_format_flags();
  
         sst.setf(t_flags);
         sst << std::setw(t_width) << std::setprecision(t_prec) << t_value;
         if(!t_disp.empty())
             sst << " " << t_disp;
         if(!t_label.empty())
             sst << " " << t_label;
         sst << ", ";
  
         auto _prec  = count_type::get_precision();
         auto _width = this_type::get_width();
         auto _flags = count_type::get_format_flags();
  
         // output the roofline metric
         auto _value = obj.get();
         auto _label = obj.label_array();
         auto _disp  = obj.display_unit_array();
  
 #if defined(DEBUG)
         if(settings::debug())
         {
             std::cout << "value: " << _value << std::endl;
             std::cout << "label: " << _label << std::endl;
             std::cout << "displ: " << _disp << std::endl;
         }
 #endif
         assert(_value.size() <= _label.size());
         assert(_value.size() <= _disp.size());
  
         auto n = _label.size();
         for(size_t i = 0; i < n; ++i)
         {
             std::stringstream ss_value;
             std::stringstream ss_extra;
             ss_value.setf(_flags);
             ss_value << std::setw(_width) << std::setprecision(_prec) << _value.at(i);
             if(!_disp.at(i).empty())
             {
                 ss_extra << " " << _disp.at(i);
             }
             else if(!_label.at(i).empty())
             {
                 ss_extra << " " << _label.at(i);
             }
             os << sst.str() << ss_value.str() << ss_extra.str();
             if(i + 1 < n)
                 os << ", ";
         }
  
         return os;
     }

Member Data Documentation

◆ precision

template<typename... Types>

const short tim::component::cpu_roofline< Types >::precision = 3

static

Definition at line 128 of file cpu_roofline.hpp.

◆ width

template<typename... Types>

const short tim::component::cpu_roofline< Types >::width = 8

static

Definition at line 129 of file cpu_roofline.hpp.

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

timemory/components/roofline/cpu_roofline.hpp

Public Types

Public Member Functions

Static Public Member Functions

Static Public Attributes

Friends

Detailed Description

template<typename... Types> struct tim::component::cpu_roofline< Types >

Member Typedef Documentation

◆ array_type

◆ base_type

◆ const_iterator

◆ count_type

◆ data_type

◆ device_t

◆ display_unit_type

◆ ert_callback_t

◆ ert_callback_type

◆ ert_config_t

◆ ert_config_type

◆ ert_counter_t

◆ ert_counter_type

◆ ert_data_ptr_t

◆ ert_data_t

◆ ert_executor_t

◆ ert_executor_type

◆ event_type

◆ events_callback_t

◆ intvec_t

◆ iterator

◆ ratio_t

◆ record_type

◆ size_type

◆ storage_type

◆ strvec_t

◆ this_type

◆ types_tuple

◆ unit_type

◆ value_type

Member Enumeration Documentation

◆ MODE

Constructor & Destructor Documentation

◆ cpu_roofline() [1/3]

◆ ~cpu_roofline()

◆ cpu_roofline() [2/3]

◆ cpu_roofline() [3/3]

Member Function Documentation

◆ cleanup()

◆ configure()

◆ description()

◆ description_array()

◆ display_unit()

◆ display_unit_array()

◆ event_mode()

◆ extra_serialization()

◆ get()

◆ get_display()

◆ get_display_unit()

◆ get_ert_data()

◆ get_events()

◆ get_events_callback()

◆ get_finalizer()

◆ get_mode_string()

◆ get_type_string()

◆ get_unit()

◆ global_finalize()

◆ global_init()

◆ label()

◆ label_array()

◆ load()

◆ operator+=()

◆ operator-=()

◆ operator=() [1/2]

◆ operator=() [2/2]

◆ record()

◆ save()

◆ set_executor_callback()

◆ start()

◆ stop()

◆ thread_finalize()

◆ thread_init()

template<typename... Types>
struct tim::component::cpu_roofline< Types >