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/gpu_roofline.hpp"

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

Public Types
enum class	MODE { COUNTERS , ACTIVITY }

using	value_type = std::tuple< typename cupti_activity::value_type, typename cupti_counters::value_type >

using	this_type = gpu_roofline< Types... >

using	base_type = base< this_type, value_type >

using	storage_type = typename base_type::storage_type

using	size_type = std::size_t

using	counters_type = cupti_counters

using	activity_type = cupti_activity

using	device_t = device::gpu

using	result_type = std::vector< double >

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

using	count_type = wall_clock

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

using	events_callback_t = std::function< strvec_t()>

using	metrics_callback_t = events_callback_t

using	activity_value_type = typename cupti_activity::value_type

using	counters_value_type = typename cupti_counters::value_type

using	secondary_type = std::unordered_multimap< std::string, value_type >

Public Member Functions
	gpu_roofline ()

	~gpu_roofline ()=default

	gpu_roofline (const gpu_roofline &rhs)

gpu_roofline &	operator= (const gpu_roofline &rhs)

	gpu_roofline (gpu_roofline &&) noexcept=default

gpu_roofline &	operator= (gpu_roofline &&) noexcept=default

result_type	get () const

void	start ()

void	stop ()

this_type &	operator+= (const this_type &rhs)

this_type &	operator-= (const this_type &rhs)

this_type &	operator+= (const value_type &rhs)

secondary_type	get_secondary () const

string_t	get_display () const

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

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

Static Public Member Functions
static events_callback_t &	get_events_callback ()

static metrics_callback_t &	get_metrics_callback ()

static MODE &	event_mode ()

static void	configure (const MODE &_mode, int _device=0)

static void	configure ()

static std::string	get_mode_string ()

static std::string	get_type_string ()

static ert_config_t &	get_finalizer ()

static ert_data_ptr_t &	get_ert_data ()

static void	global_init ()

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

static void	global_finalize (storage_type *_store)

static void	thread_init ()

static void	thread_finalize ()

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

static int64_t	unit ()

static std::string	label ()

static std::string	description ()

static std::string	display_unit ()

static value_type	record ()

static label_type	label_array ()

static label_type	display_unit_array ()

template<typename... Args>
static opaque	get_opaque (Args &&...)

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::gpu_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 70 of file gpu_roofline.hpp.

Member Typedef Documentation

◆ activity_type

template<typename... Types>

using tim::component::gpu_roofline< Types >::activity_type = cupti_activity

Definition at line 82 of file gpu_roofline.hpp.

◆ activity_value_type

template<typename... Types>

using tim::component::gpu_roofline< Types >::activity_value_type = typename cupti_activity::value_type

Definition at line 580 of file gpu_roofline.hpp.

◆ base_type

template<typename... Types>

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

Definition at line 77 of file gpu_roofline.hpp.

◆ count_type

template<typename... Types>

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

Definition at line 86 of file gpu_roofline.hpp.

◆ counters_type

template<typename... Types>

using tim::component::gpu_roofline< Types >::counters_type = cupti_counters

Definition at line 81 of file gpu_roofline.hpp.

◆ counters_value_type

template<typename... Types>

using tim::component::gpu_roofline< Types >::counters_value_type = typename cupti_counters::value_type

Definition at line 581 of file gpu_roofline.hpp.

◆ device_t

template<typename... Types>

using tim::component::gpu_roofline< Types >::device_t = device::gpu

Definition at line 83 of file gpu_roofline.hpp.

◆ ert_callback_t

template<typename... Types>

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

Definition at line 112 of file gpu_roofline.hpp.

◆ ert_callback_type

template<typename... Types>

template<typename Tp >

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

Definition at line 106 of file gpu_roofline.hpp.

◆ ert_config_t

template<typename... Types>

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

Definition at line 109 of file gpu_roofline.hpp.

◆ ert_config_type

template<typename... Types>

template<typename Tp >

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

Definition at line 100 of file gpu_roofline.hpp.

◆ ert_counter_t

template<typename... Types>

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

Definition at line 110 of file gpu_roofline.hpp.

◆ ert_counter_type

template<typename... Types>

template<typename Tp >

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

Definition at line 102 of file gpu_roofline.hpp.

◆ ert_data_ptr_t

template<typename... Types>

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

Definition at line 96 of file gpu_roofline.hpp.

◆ ert_data_t

template<typename... Types>

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

Definition at line 95 of file gpu_roofline.hpp.

◆ ert_executor_t

template<typename... Types>

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

Definition at line 111 of file gpu_roofline.hpp.

◆ ert_executor_type

template<typename... Types>

template<typename Tp >

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

Definition at line 104 of file gpu_roofline.hpp.

◆ events_callback_t

template<typename... Types>

using tim::component::gpu_roofline< Types >::events_callback_t = std::function<strvec_t()>

Definition at line 132 of file gpu_roofline.hpp.

◆ label_type

template<typename... Types>

using tim::component::gpu_roofline< Types >::label_type = std::vector<std::string>

Definition at line 85 of file gpu_roofline.hpp.

◆ metrics_callback_t

template<typename... Types>

using tim::component::gpu_roofline< Types >::metrics_callback_t = events_callback_t

Definition at line 133 of file gpu_roofline.hpp.

◆ result_type

template<typename... Types>

using tim::component::gpu_roofline< Types >::result_type = std::vector<double>

Definition at line 84 of file gpu_roofline.hpp.

◆ secondary_type

template<typename... Types>

using tim::component::gpu_roofline< Types >::secondary_type = std::unordered_multimap<std::string, value_type>

Definition at line 582 of file gpu_roofline.hpp.

◆ size_type

template<typename... Types>

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

Definition at line 80 of file gpu_roofline.hpp.

◆ storage_type

template<typename... Types>

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

Definition at line 78 of file gpu_roofline.hpp.

◆ strvec_t

template<typename... Types>

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

Definition at line 131 of file gpu_roofline.hpp.

◆ this_type

template<typename... Types>

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

Definition at line 76 of file gpu_roofline.hpp.

◆ types_tuple

template<typename... Types>

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

Definition at line 87 of file gpu_roofline.hpp.

◆ value_type

template<typename... Types>

using tim::component::gpu_roofline< Types >::value_type = std::tuple<typename cupti_activity::value_type, typename cupti_counters::value_type>

Definition at line 74 of file gpu_roofline.hpp.

Member Enumeration Documentation

◆ MODE

template<typename... Types>

enum class tim::component::gpu_roofline::MODE

strong

Enumerator
COUNTERS
ACTIVITY

Definition at line 123 of file gpu_roofline.hpp.

    {
        COUNTERS,
        ACTIVITY
    };

Constructor & Destructor Documentation

◆ gpu_roofline() [1/3]

template<typename... Types>

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

inline

Definition at line 427 of file gpu_roofline.hpp.

427{ configure(); }

tim::component::gpu_roofline::configure

static void configure()

Definition: gpu_roofline.hpp:261

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

◆ ~gpu_roofline()

template<typename... Types>

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

default

◆ gpu_roofline() [2/3]

template<typename... Types>

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

inline

Definition at line 430 of file gpu_roofline.hpp.

    : base_type(rhs)
    , m_data(rhs.m_data)
    {}

◆ gpu_roofline() [3/3]

template<typename... Types>

tim::component::gpu_roofline< Types >::gpu_roofline ( gpu_roofline< Types > && )

defaultnoexcept

Member Function Documentation

◆ configure() [1/2]

template<typename... Types>

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

inlinestatic

Definition at line 261 of file gpu_roofline.hpp.

    {
        if(!is_configured())
            configure(event_mode());
    }

References tim::component::gpu_roofline< Types >::configure(), and tim::component::gpu_roofline< Types >::event_mode().

Referenced by tim::component::gpu_roofline< Types >::gpu_roofline(), and tim::component::gpu_roofline< Types >::configure().

◆ configure() [2/2]

template<typename... Types>

static void tim::component::gpu_roofline< Types >::configure	(	const MODE &	_mode,
		int	_device = `0`
	)

inlinestatic

Definition at line 180 of file gpu_roofline.hpp.

    {
        if(is_configured())
            return;
        is_configured() = true;
 
        event_mode() = _mode;
 
        if(event_mode() == MODE::ACTIVITY)
        {
            get_labels() = { std::string("runtime") };
        }
        else
        {
            strvec_t events  = { "global_load", "global_store" };
            strvec_t metrics = { "ldst_executed" };
#if defined(TIMEMORY_USE_CUDA_HALF)
            if(is_one_of<cuda::fp16_t, types_tuple>::value)
            {
                metrics.push_back("flop_count_hp");
            }
#endif
 
            if(is_one_of<float, types_tuple>::value)
            {
                metrics.push_back("flop_count_sp");
            }
 
            if(is_one_of<double, types_tuple>::value)
            {
                metrics.push_back("flop_count_dp");
            }
 
            // integer
            if(mpl::is_one_of_integral<types_tuple>::value ||
               settings::instruction_roofline())
            {
                for(const auto* itr :
                    { "ipc", "inst_executed", "inst_integer", "inst_fp_64", "inst_fp_32",
                      "inst_fp_16", "local_load_transactions_per_request",
                      "local_store_transactions_per_request",
                      "shared_load_transactions_per_request",
                      "shared_store_transactions_per_request",
                      "gld_transactions_per_request", "gst_transactions_per_request",
                      "inst_executed_global_reductions", "inst_executed_global_stores",
                      "inst_executed_global_loads", "inst_executed_local_loads",
                      "inst_executed_local_stores", "inst_executed_shared_loads",
                      "inst_executed_shared_stores" })
                    metrics.emplace_back(itr);
            }
 
            // add in extra events
            auto _extra_events = get_events_callback()();
            for(const auto& itr : _extra_events)
                events.push_back(itr);
 
            // add in extra metrics
            auto _extra_metrics = get_metrics_callback()();
            for(const auto& itr : _extra_metrics)
                metrics.push_back(itr);
 
            auto _get_unique = [](const strvec_t& _vec) {
                std::set<std::string> _set;
                for(const auto& itr : _vec)
                    _set.insert(itr);
                strvec_t _ret;
                for(const auto& itr : _set)
                    _ret.push_back(itr);
                return _ret;
            };
 
            metrics = _get_unique(metrics);
            events  = _get_unique(events);
 
            counters_type::configure(_device, events, metrics);
            get_labels() = counters_type::label_array();
        }
    }

References tim::component::gpu_roofline< Types >::ACTIVITY, tim::component::cupti_counters::configure(), tim::component::gpu_roofline< Types >::event_mode(), tim::component::gpu_roofline< Types >::get_events_callback(), tim::component::gpu_roofline< Types >::get_metrics_callback(), tim::instruction_roofline, and tim::component::cupti_counters::label_array().

◆ description()

template<typename... Types>

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

inlinestatic

Definition at line 390 of file gpu_roofline.hpp.

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

◆ display_unit()

template<typename... Types>

static std::string tim::component::gpu_roofline< Types >::display_unit ( )

inlinestatic

Definition at line 396 of file gpu_roofline.hpp.

    {
        if(event_mode() == MODE::ACTIVITY)
            return activity_type::display_unit();
        return counters_type::display_unit();
    }

References tim::component::gpu_roofline< Types >::ACTIVITY, tim::component::base< cupti_activity, intmax_t >::display_unit(), tim::component::cupti_counters::display_unit(), and tim::component::gpu_roofline< Types >::event_mode().

Referenced by tim::component::gpu_roofline< Types >::display_unit_array().

◆ display_unit_array()

template<typename... Types>

static label_type tim::component::gpu_roofline< Types >::display_unit_array ( )

inlinestatic

Definition at line 662 of file gpu_roofline.hpp.

    {
        const auto& _labels = get_labels();
        return label_type(_labels.size(), this_type::display_unit());
    }

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

◆ event_mode()

template<typename... Types>

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

inlinestatic

Definition at line 152 of file gpu_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::gpu_roofline_mode());
            if(_env.empty())
                _env = aslc(settings::roofline_mode());
            return (_env == "op" || _env == "hw" || _env == "counters")
                       ? MODE::COUNTERS
                       : ((_env == "ai" || _env == "ac" || _env == "activity")
                              ? MODE::ACTIVITY
                              : MODE::COUNTERS);
        };
 
        static MODE _instance = _get();
        if(!is_configured())
            _instance = _get();
        return _instance;
    }

References tim::component::gpu_roofline< Types >::ACTIVITY, tim::component::gpu_roofline< Types >::COUNTERS, tim::gpu_roofline_mode, and tim::roofline_mode.

◆ extra_serialization()

template<typename... Types>

template<typename Archive >

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

inlinestatic

Definition at line 358 of file gpu_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::gpu_roofline< Types >::get_ert_data().

◆ get()

template<typename... Types>

result_type tim::component::gpu_roofline< Types >::get ( ) const

inline

Definition at line 450 of file gpu_roofline.hpp.

    {
        switch(event_mode())
        {
            case MODE::ACTIVITY: return result_type({ m_data.activity->get() });
            case MODE::COUNTERS: return m_data.counters->get();
            default: break;
        }
        return result_type{};
    }

References tim::component::gpu_roofline< Types >::ACTIVITY, tim::component::gpu_roofline< Types >::COUNTERS, and tim::component::gpu_roofline< Types >::event_mode().

Referenced by tim::component::gpu_roofline< Types >::save().

◆ get_display()

template<typename... Types>

string_t tim::component::gpu_roofline< Types >::get_display ( ) const

inline

Definition at line 635 of file gpu_roofline.hpp.

    {
        std::stringstream ss;
        if(event_mode() == MODE::COUNTERS)
        {
            return m_data.counters->get_display();
        }
        {
            ss << m_data.activity->get_display();
        }
        return ss.str();
    }

References tim::component::gpu_roofline< Types >::COUNTERS, and tim::component::gpu_roofline< Types >::event_mode().

Referenced by tim::component::gpu_roofline< Types >::save().

◆ get_ert_data()

template<typename... Types>

static ert_data_ptr_t & tim::component::gpu_roofline< Types >::get_ert_data ( )

inlinestatic

Definition at line 291 of file gpu_roofline.hpp.

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

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

◆ get_events_callback()

template<typename... Types>

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

inlinestatic

Definition at line 137 of file gpu_roofline.hpp.

    {
        static events_callback_t _instance = []() { return strvec_t{}; };
        return _instance;
    }

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

◆ get_finalizer()

template<typename... Types>

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

inlinestatic

Definition at line 283 of file gpu_roofline.hpp.

    {
        static ert_config_t _instance;
        return _instance;
    }

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

◆ get_metrics_callback()

template<typename... Types>

static metrics_callback_t & tim::component::gpu_roofline< Types >::get_metrics_callback ( )

inlinestatic

Definition at line 143 of file gpu_roofline.hpp.

    {
        static metrics_callback_t _instance = []() { return strvec_t{}; };
        return _instance;
    }

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

◆ get_mode_string()

template<typename... Types>

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

inlinestatic

Definition at line 269 of file gpu_roofline.hpp.

    {
        return (event_mode() == MODE::COUNTERS) ? "counters" : "activity";
    }

References tim::component::gpu_roofline< Types >::COUNTERS, and tim::component::gpu_roofline< Types >::event_mode().

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

◆ get_opaque()

template<typename... Args>

static opaque tim::component::empty_base::get_opaque ( Args && ... )

inlinestaticinherited

Definition at line 72 of file declaration.hpp.

    {
        return opaque{};
    }

◆ get_secondary()

template<typename... Types>

secondary_type tim::component::gpu_roofline< Types >::get_secondary ( ) const

inline

Definition at line 586 of file gpu_roofline.hpp.

    {
        secondary_type ret;
        switch(event_mode())
        {
            case MODE::ACTIVITY:
            {
                auto&& _tmp = m_data.activity->get_secondary();
                for(auto&& itr : _tmp)
                {
                    ret.insert(
                        { itr.first, value_type{ itr.second, counters_value_type{} } });
                }
                break;
            }
            case MODE::COUNTERS:
            {
                auto&& _tmp = m_data.counters->get_secondary();
                for(auto&& itr : _tmp)
                {
                    ret.insert(
                        { itr.first, value_type{ activity_value_type{}, itr.second } });
                }
                break;
            }
        }
        return ret;
    }

References tim::component::gpu_roofline< Types >::ACTIVITY, tim::component::gpu_roofline< Types >::COUNTERS, and tim::component::gpu_roofline< Types >::event_mode().

◆ get_type_string()

template<typename... Types>

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

inlinestatic

Definition at line 276 of file gpu_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::gpu_roofline< Types >::label(), and tim::component::gpu_roofline< Types >::save().

◆ global_finalize()

template<typename... Types>

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

inlinestatic

Definition at line 323 of file gpu_roofline.hpp.

    {
        // disable the activity/counters before running ERT
        if(event_mode() == MODE::ACTIVITY)
        {
            operation::fini<activity_type>{}(
                operation::mode_constant<operation::fini_mode::global>{});
        }
        else
        {
            operation::fini<counters_type>{}(
                operation::mode_constant<operation::fini_mode::global>{});
        }
 
        // run ERT
        if(_store && _store->size() > 0)
        {
            assert(_store->is_finalizing());
            // 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() > 0 || settings::debug()))
                std::cout << *(ert_data) << std::endl;
        }
    }

References tim::component::gpu_roofline< Types >::ACTIVITY, tim::debug, tim::component::gpu_roofline< Types >::event_mode(), tim::component::gpu_roofline< Types >::get_ert_data(), tim::component::gpu_roofline< Types >::get_finalizer(), tim::component::empty_storage::size(), and tim::verbose.

◆ global_init()

template<typename... Types>

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

inlinestatic

Definition at line 299 of file gpu_roofline.hpp.

    {
        if(event_mode() == MODE::ACTIVITY)
        {
            operation::init<activity_type>{}(
                operation::mode_constant<operation::init_mode::global>{});
        }
        else
        {
            operation::init<counters_type>{}(
                operation::mode_constant<operation::init_mode::global>{});
        }
    }

References tim::component::gpu_roofline< Types >::ACTIVITY, and tim::component::gpu_roofline< Types >::event_mode().

◆ label()

template<typename... Types>

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

inlinestatic

Definition at line 375 of file gpu_roofline.hpp.

    {
        if(settings::roofline_type_labels_gpu() || settings::roofline_type_labels())
        {
            auto ret = std::string("gpu_roofline_") + get_type_string() + "_" +
                       get_mode_string();
            // erase consecutive underscores
            while(ret.find("__") != std::string::npos)
                ret.erase(ret.find("__"), 1);
            return ret;
        }
 
        return std::string("gpu_roofline_") + get_mode_string();
    }

References tim::component::gpu_roofline< Types >::get_mode_string(), tim::component::gpu_roofline< Types >::get_type_string(), tim::roofline_type_labels, and tim::roofline_type_labels_gpu.

◆ label_array()

template<typename... Types>

static label_type tim::component::gpu_roofline< Types >::label_array ( )

inlinestatic

Definition at line 658 of file gpu_roofline.hpp.

658{ return this_type::get_labels(); }

◆ load()

template<typename... Types>

template<typename Archive >

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

inline

Definition at line 844 of file gpu_roofline.hpp.

    {
        std::string              _disp;
        result_type              _data;
        std::vector<std::string> _labels;
        std::string              _mode_str;
        std::string              _type_str;
 
        ar(cereal::make_nvp("laps", laps), cereal::make_nvp("display", _disp),
           cereal::make_nvp("mode", _mode_str), cereal::make_nvp("type", _type_str),
           cereal::make_nvp("labels", _labels));
 
        if(_mode_str == "counters")
        {
            event_mode() = MODE::COUNTERS;
        }
        else if(_mode_str == "activity")
        {
            event_mode() = MODE::ACTIVITY;
        }
 
        _data.resize(_labels.size());
 
        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, *ditr));
        ar.finishNode();
 
        ar.setNextName("value");
        ar.startNode();
        if(event_mode() == MODE::ACTIVITY)
        {
            ar(std::get<0>(value));
        }
        else
        {
            ar(std::get<1>(value));
        }
        ar.finishNode();
 
        ar.setNextName("accum");
        ar.startNode();
        if(event_mode() == MODE::ACTIVITY)
        {
            ar(std::get<0>(accum));
        }
        else
        {
            ar(std::get<1>(accum));
        }
        ar.finishNode();
    }

References tim::component::gpu_roofline< Types >::ACTIVITY, tim::component::gpu_roofline< Types >::COUNTERS, tim::component::gpu_roofline< Types >::event_mode(), and tim::component::base< gpu_roofline< Types... >, std::tuple< cupti_activity::value_type, cupti_counters::value_type > >::laps.

◆ operator+=() [1/2]

template<typename... Types>

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

inline

Definition at line 508 of file gpu_roofline.hpp.

    {
        switch(event_mode())
        {
            case MODE::ACTIVITY:
            {
                *m_data.activity += *rhs.m_data.activity;
                std::get<0>(accum) = m_data.activity->get_accum();
                std::get<0>(value) = m_data.activity->get_value();
                break;
            }
            case MODE::COUNTERS:
            {
                *m_data.counters += *rhs.m_data.counters;
                std::get<1>(accum) = m_data.counters->get_accum();
                std::get<1>(value) = m_data.counters->get_value();
                break;
            }
        }
        return *this;
    }

References tim::component::gpu_roofline< Types >::ACTIVITY, tim::component::gpu_roofline< Types >::COUNTERS, and tim::component::gpu_roofline< Types >::event_mode().

◆ operator+=() [2/2]

template<typename... Types>

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

inline

Definition at line 556 of file gpu_roofline.hpp.

    {
        switch(event_mode())
        {
            case MODE::ACTIVITY:
            {
                *m_data.activity += std::get<0>(rhs);
                std::get<0>(accum) = m_data.activity->get_accum();
                std::get<0>(value) = m_data.activity->get_value();
                break;
            }
            case MODE::COUNTERS:
            {
                *m_data.counters += std::get<1>(rhs);
                std::get<1>(accum) = m_data.counters->get_accum();
                std::get<1>(value) = m_data.counters->get_value();
                break;
            }
        }
        return *this;
    }

References tim::component::gpu_roofline< Types >::ACTIVITY, tim::component::gpu_roofline< Types >::COUNTERS, and tim::component::gpu_roofline< Types >::event_mode().

◆ operator-=()

template<typename... Types>

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

inline

Definition at line 532 of file gpu_roofline.hpp.

    {
        switch(event_mode())
        {
            case MODE::ACTIVITY:
            {
                *m_data.activity -= *rhs.m_data.activity;
                std::get<0>(accum) = m_data.activity->get_accum();
                std::get<0>(value) = m_data.activity->get_value();
                break;
            }
            case MODE::COUNTERS:
            {
                *m_data.counters -= *rhs.m_data.counters;
                std::get<1>(accum) = m_data.counters->get_accum();
                std::get<1>(value) = m_data.counters->get_value();
                break;
            }
        }
        return *this;
    }

References tim::component::gpu_roofline< Types >::ACTIVITY, tim::component::gpu_roofline< Types >::COUNTERS, and tim::component::gpu_roofline< Types >::event_mode().

◆ operator=() [1/2]

template<typename... Types>

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

inline

Definition at line 435 of file gpu_roofline.hpp.

    {
        if(this != &rhs)
        {
            base_type::operator=(rhs);
            m_data             = rhs.m_data;
        }
        return *this;
    }

◆ operator=() [2/2]

template<typename... Types>

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

defaultnoexcept

◆ record()

template<typename... Types>

static value_type tim::component::gpu_roofline< Types >::record ( )

inlinestatic

Definition at line 405 of file gpu_roofline.hpp.

    {
        value_type tmp;
        switch(event_mode())
        {
            case MODE::ACTIVITY: std::get<0>(tmp) = activity_type::record(); break;
            case MODE::COUNTERS: std::get<1>(tmp) = counters_type::record(); break;
            default: break;
        }
        return tmp;
    }

References tim::component::gpu_roofline< Types >::ACTIVITY, tim::component::gpu_roofline< Types >::COUNTERS, tim::component::gpu_roofline< Types >::event_mode(), tim::component::cupti_activity::record(), and tim::component::cupti_counters::record().

◆ save()

template<typename... Types>

template<typename Archive >

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

inline

Definition at line 795 of file gpu_roofline.hpp.

    {
        auto _disp   = get_display();
        auto _data   = get();
        auto _labels = get_labels();
 
        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));
 
        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, *ditr));
        ar.finishNode();
 
        ar.setNextName("value");
        ar.startNode();
        ar.makeArray();
        if(event_mode() == MODE::ACTIVITY)
        {
            ar(std::get<0>(value));
        }
        else
        {
            ar(std::get<1>(value));
        }
        ar.finishNode();
 
        ar.setNextName("accum");
        ar.startNode();
        ar.makeArray();
        if(event_mode() == MODE::ACTIVITY)
        {
            ar(std::get<0>(accum));
        }
        else
        {
            ar(std::get<1>(accum));
        }
        ar.finishNode();
    }

References tim::component::gpu_roofline< Types >::ACTIVITY, tim::component::gpu_roofline< Types >::event_mode(), tim::component::gpu_roofline< Types >::get(), tim::component::gpu_roofline< Types >::get_display(), tim::component::gpu_roofline< Types >::get_mode_string(), tim::component::gpu_roofline< Types >::get_type_string(), and tim::component::base< gpu_roofline< Types... >, std::tuple< cupti_activity::value_type, cupti_counters::value_type > >::laps.

◆ set_executor_callback()

template<typename... Types>

template<typename Tp , typename FuncT >

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

inlinestatic

Definition at line 316 of file gpu_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::gpu_roofline< Types >::start ( )

inline

Definition at line 463 of file gpu_roofline.hpp.

    {
        switch(event_mode())
        {
            case MODE::ACTIVITY:
            {
                m_data.activity->start();
                std::get<0>(value) = m_data.activity->get_value();
                break;
            }
            case MODE::COUNTERS:
            {
                m_data.counters->start();
                std::get<1>(value) = m_data.counters->get_value();
                break;
            }
        }
    }

References tim::component::gpu_roofline< Types >::ACTIVITY, tim::component::gpu_roofline< Types >::COUNTERS, and tim::component::gpu_roofline< Types >::event_mode().

◆ stop()

template<typename... Types>

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

inline

Definition at line 484 of file gpu_roofline.hpp.

    {
        using namespace tim::component::operators;
        switch(event_mode())
        {
            case MODE::ACTIVITY:
            {
                m_data.activity->stop();
                std::get<0>(accum) = m_data.activity->get_accum();
                std::get<0>(value) = m_data.activity->get_value();
                break;
            }
            case MODE::COUNTERS:
            {
                m_data.counters->stop();
                std::get<1>(accum) = m_data.counters->get_accum();
                std::get<1>(value) = m_data.counters->get_value();
                break;
            }
        }
    }

References tim::component::gpu_roofline< Types >::ACTIVITY, tim::component::gpu_roofline< Types >::COUNTERS, and tim::component::gpu_roofline< Types >::event_mode().

◆ thread_finalize()

template<typename... Types>

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

inlinestatic

Definition at line 353 of file gpu_roofline.hpp.

353{}

◆ thread_init()

template<typename... Types>

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

inlinestatic

Definition at line 352 of file gpu_roofline.hpp.

352{}

◆ unit()

template<typename... Types>

static int64_t tim::component::gpu_roofline< Types >::unit ( )

inlinestatic

Definition at line 368 of file gpu_roofline.hpp.

    {
        if(event_mode() == MODE::ACTIVITY)
            return activity_type::unit();
        return counters_type::unit();
    }

References tim::component::gpu_roofline< Types >::ACTIVITY, tim::component::gpu_roofline< Types >::event_mode(), tim::component::base< cupti_activity, intmax_t >::unit(), and tim::component::cupti_counters::unit().

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 586 of file gpu_roofline.hpp.

◆ operation::record< this_type >

template<typename... Types>

friend struct operation::record< this_type >

friend

Definition at line 1 of file gpu_roofline.hpp.

◆ operation::set_started< this_type >

template<typename... Types>

friend struct operation::set_started< this_type >

friend

Definition at line 1 of file gpu_roofline.hpp.

◆ operation::set_stopped< this_type >

template<typename... Types>

friend struct operation::set_stopped< this_type >

friend

Definition at line 1 of file gpu_roofline.hpp.

◆ operation::start< this_type >

template<typename... Types>

friend struct operation::start< this_type >

friend

Definition at line 1 of file gpu_roofline.hpp.

◆ operation::stop< this_type >

template<typename... Types>

friend struct operation::stop< this_type >

friend

Definition at line 1 of file gpu_roofline.hpp.

◆ operator<<

template<typename... Types>

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

friend

Definition at line 650 of file gpu_roofline.hpp.

    {
        os << as_string(obj.get_display());
        return os;
    }

Member Data Documentation

◆ precision

template<typename... Types>

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

static

Definition at line 118 of file gpu_roofline.hpp.

◆ width

template<typename... Types>

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

static

Definition at line 119 of file gpu_roofline.hpp.

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

timemory/components/roofline/gpu_roofline.hpp

Public Types

Public Member Functions

Static Public Member Functions

Static Public Attributes

Friends

Detailed Description

Member Typedef Documentation

◆ activity_type

◆ activity_value_type

◆ base_type

◆ count_type

◆ counters_type

◆ counters_value_type

◆ device_t

◆ 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

◆ events_callback_t

◆ label_type

◆ metrics_callback_t

◆ result_type

◆ secondary_type

◆ size_type

◆ storage_type

◆ strvec_t

◆ this_type

◆ types_tuple

◆ value_type

Member Enumeration Documentation

◆ MODE

Constructor & Destructor Documentation

◆ gpu_roofline() [1/3]

◆ ~gpu_roofline()

◆ gpu_roofline() [2/3]

◆ gpu_roofline() [3/3]

Member Function Documentation

◆ configure() [1/2]

◆ configure() [2/2]

◆ description()

◆ display_unit()

◆ display_unit_array()

◆ event_mode()

◆ extra_serialization()

◆ get()

◆ get_display()

◆ get_ert_data()

◆ get_events_callback()

◆ get_finalizer()

◆ get_metrics_callback()

◆ get_mode_string()

◆ get_opaque()

◆ get_secondary()

◆ get_type_string()

◆ global_finalize()

◆ global_init()

◆ label()

◆ label_array()

◆ load()

◆ operator+=() [1/2]

◆ operator+=() [2/2]

◆ operator-=()

◆ operator=() [1/2]

◆ operator=() [2/2]

◆ record()

◆ save()

◆ set_executor_callback()

◆ start()

◆ stop()

◆ thread_finalize()

◆ thread_init()

◆ unit()

Friends And Related Function Documentation

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