Main CPU counters header. More...

#include "version.h"
#include "types.h"
#include "msr.h"
#include "pci.h"
#include "bw.h"
#include "width_extender.h"
#include "exceptions/unsupported_processor_exception.hpp"
#include <vector>
#include <array>
#include <limits>
#include <string>
#include <memory>
#include <map>
#include <unordered_map>
#include <string.h>
#include <assert.h>
#include <semaphore.h>
#include <sys/types.h>
#include <sys/stat.h>
#include <fcntl.h>
#include <sys/syscall.h>
#include <unistd.h>
#include "resctrl.h"

Include dependency graph for cpucounters.h:

This graph shows which files directly or indirectly include this file:

Go to the source code of this file.

Classes
struct	pcm::TopologyEntry

class	pcm::HWRegister

class	pcm::PCICFGRegister64

class	pcm::PCICFGRegister32

class	pcm::MMIORegister64

class	pcm::MMIORegister32

class	pcm::MSRRegister

class	pcm::CounterWidthExtenderRegister

class	pcm::UncorePMU

class	pcm::ServerPCICFGUncore
	Object to access uncore counters in a socket/processor with microarchitecture codename SandyBridge-EP (Jaketown) or Ivytown-EP or Ivytown-EX. More...

class	pcm::SimpleCounterState

class	pcm::PCM
	CPU Performance Monitor. More...

struct	pcm::PCM::SimplePCIeDevInfo

struct	pcm::PCM::CustomCoreEventDescription
	Custom Core event description. More...

struct	pcm::PCM::ExtendedCustomCoreEventDescription
	Extended custom core event description. More...

struct	pcm::PCM::CustomIIOEventDescription

struct	pcm::PCM::RawPMUConfig

class	pcm::BasicCounterState
	Basic core counter state. More...

class	pcm::UncoreCounterState
	Basic uncore counter state. More...

class	pcm::ServerUncoreCounterState
	Server uncore power counter state. More...

class	pcm::CoreCounterState
	(Logical) core-wide counter state More...

class	pcm::SocketCounterState
	Socket-wide counter state. More...

class	pcm::SystemCounterState
	System-wide counter state. More...

Macros
#define	PCM_API

#define	NOMINMAX

#define	PCM_SKL_PATH_CASES

#define	PCM_GENERATE_METRIC_AVAILABLE_FUNCTION(m) bool is##m() const { return m; }

Typedefs
typedef SimpleCounterState	pcm::PCIeCounterState

typedef SimpleCounterState	pcm::IIOCounterState

typedef std::vector< uint64 >	pcm::eventGroup_t

Enumerations
enum	ServerUncoreMemoryMetrics { PartialWrites, Pmem, PmemMemoryMode, PmemMixedMode }

Functions
uint64	pcm::RDTSC ()

uint64	pcm::RDTSCP ()

template<class CounterStateType >
int32	pcm::getThermalHeadroom (const CounterStateType &, const CounterStateType &after)

template<class CounterStateType >
double	pcm::getNormalizedQPIL0pTxCycles (uint32 port, const CounterStateType &before, const CounterStateType &after)
	Returns the ratio of QPI cycles in power saving half-lane mode. More...

template<class CounterStateType >
double	pcm::getNormalizedQPIL1Cycles (uint32 port, const CounterStateType &before, const CounterStateType &after)
	Returns the ratio of QPI cycles in power saving shutdown mode. More...

template<class CounterStateType >
uint64	pcm::getDRAMClocks (uint32 channel, const CounterStateType &before, const CounterStateType &after)
	Returns DRAM clock ticks. More...

template<class CounterStateType >
uint64	pcm::getMCDRAMClocks (uint32 channel, const CounterStateType &before, const CounterStateType &after)
	Returns MCDRAM clock ticks. More...

template<class CounterStateType >
uint64	pcm::getMCCounter (uint32 channel, uint32 counter, const CounterStateType &before, const CounterStateType &after)
	Direct read of memory controller PMU counter (counter meaning depends on the programming: power/performance/etc) More...

template<class CounterStateType >
uint64	pcm::getM3UPICounter (uint32 port, uint32 counter, const CounterStateType &before, const CounterStateType &after)
	Direct read of M3UPI PMU counter (counter meaning depends on the programming: power/performance/etc) More...

template<class CounterStateType >
uint64	pcm::getCBOCounter (uint32 cbo, uint32 counter, const CounterStateType &before, const CounterStateType &after)
	Direct read of CHA or CBO PMU counter (counter meaning depends on the programming: power/performance/etc) More...

template<class CounterStateType >
uint64	pcm::getUBOXCounter (uint32 counter, const CounterStateType &before, const CounterStateType &after)
	Direct read of UBOX PMU counter (counter meaning depends on the programming: power/performance/etc) More...

template<class CounterStateType >
uint64	pcm::getIIOCounter (uint32 stack, uint32 counter, const CounterStateType &before, const CounterStateType &after)
	Direct read of IIO PMU counter (counter meaning depends on the programming: power/performance/etc) More...

template<class CounterStateType >
uint64	pcm::getIRPCounter (uint32 stack, uint32 counter, const CounterStateType &before, const CounterStateType &after)
	Direct read of IRP PMU counter (counter meaning depends on the programming: power/performance/etc) More...

template<class CounterStateType >
uint64	pcm::getXPICounter (uint32 port, uint32 counter, const CounterStateType &before, const CounterStateType &after)
	Direct read of UPI or QPI PMU counter (counter meaning depends on the programming: power/performance/etc) More...

template<class CounterStateType >
uint64	pcm::getM2MCounter (uint32 controller, uint32 counter, const CounterStateType &before, const CounterStateType &after)
	Direct read of Memory2Mesh controller PMU counter (counter meaning depends on the programming: power/performance/etc) More...

template<class CounterStateType >
uint64	pcm::getEDCCounter (uint32 channel, uint32 counter, const CounterStateType &before, const CounterStateType &after)
	Direct read of embedded DRAM memory controller counter (counter meaning depends on the programming: power/performance/etc) More...

template<class CounterStateType >
uint64	pcm::getPCUCounter (uint32 counter, const CounterStateType &before, const CounterStateType &after)
	Direct read of power control unit PMU counter (counter meaning depends on the programming: power/performance/etc) More...

template<class CounterStateType >
uint64	pcm::getPCUClocks (const CounterStateType &before, const CounterStateType &after)
	Returns clock ticks of power control unit. More...

template<class CounterStateType >
uint64	pcm::getConsumedEnergy (const CounterStateType &before, const CounterStateType &after)
	Returns energy consumed by processor, excluding DRAM (measured in internal units) More...

template<class CounterStateType >
uint64	pcm::getDRAMConsumedEnergy (const CounterStateType &before, const CounterStateType &after)
	Returns energy consumed by DRAM (measured in internal units) More...

template<class CounterStateType >
int64	pcm::getFreeRunningCounter (const typename CounterStateType::FreeRunningCounterID &counter, const CounterStateType &before, const CounterStateType &after)
	Returns free running counter if it exists, -1 otherwise. More...

template<class CounterStateType >
uint64	pcm::getUncoreClocks (const CounterStateType &before, const CounterStateType &after)
	Returns uncore clock ticks. More...

template<class CounterStateType >
double	pcm::getConsumedJoules (const CounterStateType &before, const CounterStateType &after)
	Returns Joules consumed by processor (excluding DRAM) More...

template<class CounterStateType >
double	pcm::getDRAMConsumedJoules (const CounterStateType &before, const CounterStateType &after)
	Returns Joules consumed by DRAM. More...

template<class CounterStateType >
uint64	pcm::getQPIClocks (uint32 port, const CounterStateType &before, const CounterStateType &after)
	Returns QPI LL clock ticks. More...

template<class CounterStateType >
uint64	pcm::getQPIL0pTxCycles (uint32 port, const CounterStateType &before, const CounterStateType &after)
	Returns the number of QPI cycles in power saving half-lane mode. More...

template<class CounterStateType >
uint64	pcm::getQPIL1Cycles (uint32 port, const CounterStateType &before, const CounterStateType &after)
	Returns the number of QPI cycles in power saving shutdown mode. More...

SystemCounterState	pcm::getSystemCounterState ()
	Reads the counter state of the system. More...

SocketCounterState	pcm::getSocketCounterState (uint32 socket)
	Reads the counter state of a socket. More...

CoreCounterState	pcm::getCoreCounterState (uint32 core)
	Reads the counter state of a (logical) core. More...

template<class CounterStateType >
double	pcm::getIPC (const CounterStateType &before, const CounterStateType &after)
	Computes average number of retired instructions per core cycle (IPC) More...

template<class CounterStateType >
uint64	pcm::getInstructionsRetired (const CounterStateType &before, const CounterStateType &after)
	Computes the number of retired instructions. More...

template<class CounterStateType >
double	pcm::getExecUsage (const CounterStateType &before, const CounterStateType &after)
	Computes average number of retired instructions per time interval. More...

template<class CounterStateType >
uint64	pcm::getInstructionsRetired (const CounterStateType &now)
	Computes the number of retired instructions. More...

template<class CounterStateType >
uint64	pcm::getCycles (const CounterStateType &before, const CounterStateType &after)
	Computes the number core clock cycles when signal on a specific core is running (not halted) More...

template<class CounterStateType >
uint64	pcm::getRefCycles (const CounterStateType &before, const CounterStateType &after)
	Computes the number of reference clock cycles while clock signal on the core is running. More...

template<class CounterStateType >
uint64	pcm::getCycles (const CounterStateType &now)
	Computes the number executed core clock cycles. More...

template<class CounterStateType >
double	pcm::getCoreIPC (const CounterStateType &before, const CounterStateType &after)
	Computes average number of retired instructions per core cycle for the entire system combining instruction counts from logical cores to corresponding physical cores. More...

template<class CounterStateType >
double	pcm::getTotalExecUsage (const CounterStateType &before, const CounterStateType &after)
	Computes average number of retired instructions per time interval for the entire system combining instruction counts from logical cores to corresponding physical cores. More...

template<class StateType >
double	pcm::getAverageFrequencyFromClocks (const int64 clocks, const StateType &before, const StateType &after)

template<class CounterStateType >
double	pcm::getAverageFrequency (const CounterStateType &before, const CounterStateType &after)
	Computes average core frequency also taking Intel Turbo Boost technology into account. More...

template<class UncoreStateType >
double	pcm::getAverageUncoreFrequency (const UncoreStateType &before, const UncoreStateType &after)
	Computes average uncore frequency. More...

template<class CounterStateType >
double	pcm::getActiveAverageFrequency (const CounterStateType &before, const CounterStateType &after)
	Computes average core frequency when not in powersaving C0-state (also taking Intel Turbo Boost technology into account) More...

template<class CounterStateType >
double	pcm::getRelativeFrequency (const CounterStateType &before, const CounterStateType &after)
	Computes average core frequency also taking Intel Turbo Boost technology into account. More...

template<class CounterStateType >
double	pcm::getActiveRelativeFrequency (const CounterStateType &before, const CounterStateType &after)
	Computes average core frequency when not in powersaving C0-state (also taking Intel Turbo Boost technology into account) More...

template<class CounterStateType >
double	pcm::getL2CacheHitRatio (const CounterStateType &before, const CounterStateType &after)
	Computes L2 cache hit ratio. More...

template<class CounterStateType >
double	pcm::getL3CacheHitRatio (const CounterStateType &before, const CounterStateType &after)
	Computes L3 cache hit ratio. More...

template<class CounterStateType >
uint64	pcm::getL3CacheMisses (const CounterStateType &before, const CounterStateType &after)
	Computes number of L3 cache misses. More...

template<class CounterStateType >
uint64	pcm::getL2CacheMisses (const CounterStateType &before, const CounterStateType &after)
	Computes number of L2 cache misses. More...

template<class CounterStateType >
uint64	pcm::getL2CacheHits (const CounterStateType &before, const CounterStateType &after)
	Computes number of L2 cache hits. More...

template<class CounterStateType >
uint64	pcm::getL3CacheOccupancy (const CounterStateType &now)
	Computes L3 Cache Occupancy. More...

template<class CounterStateType >
uint64	pcm::getLocalMemoryBW (const CounterStateType &before, const CounterStateType &after)
	Computes Local Memory Bandwidth. More...

template<class CounterStateType >
uint64	pcm::getRemoteMemoryBW (const CounterStateType &before, const CounterStateType &after)
	Computes Remote Memory Bandwidth. More...

template<class CounterStateType >
uint64	pcm::getL3CacheHitsNoSnoop (const CounterStateType &before, const CounterStateType &after)
	Computes number of L3 cache hits where no snooping in sibling L2 caches had to be done. More...

template<class CounterStateType >
uint64	pcm::getL3CacheHitsSnoop (const CounterStateType &before, const CounterStateType &after)
	Computes number of L3 cache hits where snooping in sibling L2 caches had to be done. More...

template<class CounterStateType >
uint64	pcm::getL3CacheHits (const CounterStateType &before, const CounterStateType &after)
	Computes total number of L3 cache hits. More...

template<class CounterStateType >
uint64	pcm::getInvariantTSC (const CounterStateType &before, const CounterStateType &after)
	Computes number of invariant time stamp counter ticks. More...

template<class CounterStateType >
double	pcm::getCoreCStateResidency (int state, const CounterStateType &before, const CounterStateType &after)
	Computes residency in the core C-state. More...

template<class CounterStateType >
uint64	pcm::getCoreCStateResidency (int state, const CounterStateType &now)
	Reads raw residency counter for the core C-state. More...

template<class CounterStateType >
double	pcm::getPackageCStateResidency (int state, const CounterStateType &before, const CounterStateType &after)
	Computes residency in the package C-state. More...

template<class CounterStateType >
uint64	pcm::getPackageCStateResidency (int state, const CounterStateType &now)
	Reads raw residency counter for the package C-state. More...

template<class CounterStateType >
uint64	pcm::getBytesReadFromMC (const CounterStateType &before, const CounterStateType &after)
	Computes number of bytes read from DRAM memory controllers. More...

template<class CounterStateType >
uint64	pcm::getBytesWrittenToMC (const CounterStateType &before, const CounterStateType &after)
	Computes number of bytes written to DRAM memory controllers. More...

template<class CounterStateType >
uint64	pcm::getBytesReadFromPMM (const CounterStateType &before, const CounterStateType &after)
	Computes number of bytes read from PMM memory. More...

template<class CounterStateType >
uint64	pcm::getBytesWrittenToPMM (const CounterStateType &before, const CounterStateType &after)
	Computes number of bytes written to PMM memory. More...

template<class CounterStateType >
uint64	pcm::getBytesReadFromEDC (const CounterStateType &before, const CounterStateType &after)
	Computes number of bytes read from MCDRAM memory controllers. More...

template<class CounterStateType >
uint64	pcm::getBytesWrittenToEDC (const CounterStateType &before, const CounterStateType &after)
	Computes number of bytes written to MCDRAM memory controllers. More...

template<class CounterStateType >
uint64	pcm::getGTRequestBytesFromMC (const CounterStateType &before, const CounterStateType &after)
	Computes number of bytes of read/write requests from GT engine. More...

template<class CounterStateType >
uint64	pcm::getIARequestBytesFromMC (const CounterStateType &before, const CounterStateType &after)
	Computes number of bytes of read/write requests from all IA. More...

template<class CounterStateType >
uint64	pcm::getIORequestBytesFromMC (const CounterStateType &before, const CounterStateType &after)
	Computes number of bytes of read/write requests from all IO sources. More...

template<class CounterStateType >
uint64	pcm::getSMICount (const CounterStateType &before, const CounterStateType &after)
	Returns the number of occurred system management interrupts. More...

template<class CounterStateType >
uint64	pcm::getNumberOfCustomEvents (int32 eventCounterNr, const CounterStateType &before, const CounterStateType &after)
	Returns the number of occurred custom core events. More...

uint64	pcm::getIncomingQPILinkBytes (uint32 socketNr, uint32 linkNr, const SystemCounterState &before, const SystemCounterState &after)
	Get estimation of QPI data traffic per incoming QPI link. More...

double	pcm::getIncomingQPILinkUtilization (uint32 socketNr, uint32 linkNr, const SystemCounterState &before, const SystemCounterState &after)
	Get data utilization of incoming QPI link (0..1) More...

double	pcm::getOutgoingQPILinkUtilization (uint32 socketNr, uint32 linkNr, const SystemCounterState &before, const SystemCounterState &after)
	Get utilization of outgoing QPI link (0..1) More...

uint64	pcm::getOutgoingQPILinkBytes (uint32 socketNr, uint32 linkNr, const SystemCounterState &before, const SystemCounterState &after)
	Get estimation of QPI (data+nondata) traffic per outgoing QPI link. More...

uint64	pcm::getAllIncomingQPILinkBytes (const SystemCounterState &before, const SystemCounterState &after)
	Get estimation of total QPI data traffic. More...

uint64	pcm::getAllOutgoingQPILinkBytes (const SystemCounterState &before, const SystemCounterState &after)
	Get estimation of total QPI data+nondata traffic. More...

uint64	pcm::getIncomingQPILinkBytes (uint32 socketNr, uint32 linkNr, const SystemCounterState &now)
	Return current value of the counter of QPI data traffic per incoming QPI link. More...

uint64	pcm::getSocketIncomingQPILinkBytes (uint32 socketNr, const SystemCounterState &now)
	Get estimation of total QPI data traffic for this socket. More...

uint64	pcm::getAllIncomingQPILinkBytes (const SystemCounterState &now)
	Get estimation of Socket QPI data traffic. More...

double	pcm::getQPItoMCTrafficRatio (const SystemCounterState &before, const SystemCounterState &after)
	Get QPI data to Memory Controller traffic ratio. More...

template<class CounterStateType >
double	pcm::getLocalMemoryRequestRatio (const CounterStateType &before, const CounterStateType &after)
	Get local memory access ration measured in home agent. More...

template<class CounterType >
uint64	pcm::getNumberOfEvents (const CounterType &before, const CounterType &after)
	Returns the raw count of events. More...

template<class CounterStateType >
double	pcm::getLLCReadMissLatency (const CounterStateType &before, const CounterStateType &after)
	Returns average last level cache read+prefetch miss latency in ns.

template<class CounterStateType >
uint64	pcm::getAllSlots (const CounterStateType &before, const CounterStateType &after)

template<class CounterStateType >
uint64	pcm::getAllSlotsRaw (const CounterStateType &before, const CounterStateType &after)

template<class CounterStateType >
double	pcm::getBackendBound (const CounterStateType &before, const CounterStateType &after)
	Returns unutilized pipeline slots where no uop was delivered due to lack of back-end resources as range 0..1.

template<class CounterStateType >
double	pcm::getFrontendBound (const CounterStateType &before, const CounterStateType &after)
	Returns unutilized pipeline slots where Front-end did not deliver a uop while back-end is ready as range 0..1.

template<class CounterStateType >
double	pcm::getBadSpeculation (const CounterStateType &before, const CounterStateType &after)
	Returns wasted pipeline slots due to incorrect speculation, covering whole penalty: Utilized by uops that do not retire, or Recovery Bubbles (unutilized slots) as range 0..1.

template<class CounterStateType >
double	pcm::getRetiring (const CounterStateType &before, const CounterStateType &after)
	Returns pipeline slots utilized by uops that eventually retire (commit)

template<class CounterStateType >
uint64	pcm::getMSREvent (const uint64 &index, const PCM::MSRType &type, const CounterStateType &before, const CounterStateType &after)

Detailed Description

Main CPU counters header.

Include this header file if you want to access CPU counters (core and uncore - including memory controller chips and QPI)

Macro Definition Documentation

◆ PCM_SKL_PATH_CASES

#define PCM_SKL_PATH_CASES

Value:

case PCM::SKL_UY:  \
        case PCM::KBL:     \
        case PCM::KBL_1:   \
        case PCM::CML:     \
        case PCM::ICL:     \
        case PCM::RKL:     \
        case PCM::TGL:     \
        case PCM::SKL:

Function Documentation

◆ getActiveAverageFrequency()

template<class CounterStateType >

double pcm::getActiveAverageFrequency	(	const CounterStateType &	before,
		const CounterStateType &	after
	)

Computes average core frequency when not in powersaving C0-state (also taking Intel Turbo Boost technology into account)

Parameters

before	CPU counter state before the experiment
after	CPU counter state after the experiment

Returns: frequency in Hz

References pcm::getActiveAverageFrequency(), pcm::PCM::getInstance(), and pcm::PCM::getNominalFrequency().

Referenced by pcm::getActiveAverageFrequency().

◆ getActiveRelativeFrequency()

template<class CounterStateType >

double pcm::getActiveRelativeFrequency	(	const CounterStateType &	before,
		const CounterStateType &	after
	)

Computes average core frequency when not in powersaving C0-state (also taking Intel Turbo Boost technology into account)

Parameters

before	CPU counter state before the experiment
after	CPU counter state after the experiment

Returns: Fraction of nominal frequency (if >1.0 then Turbo was working during the measurement)

References pcm::getActiveRelativeFrequency(), and pcm::PCM::getInstance().

Referenced by pcm::getActiveRelativeFrequency().

◆ getAllIncomingQPILinkBytes() [1/2]

uint64 pcm::getAllIncomingQPILinkBytes	(	const SystemCounterState &	before,
		const SystemCounterState &	after
	)

inline

Get estimation of total QPI data traffic.

Returns an estimation of number of data bytes transferred to all sockets over all Intel(r) Quick Path Interconnect links

Parameters

before	System CPU counter state before the experiment
after	System CPU counter state after the experiment

Returns: Number of bytes

References pcm::getAllIncomingQPILinkBytes(), pcm::PCM::getInstance(), and pcm::PCM::getNumSockets().

Referenced by pcm::getAllIncomingQPILinkBytes(), and pcm::getQPItoMCTrafficRatio().

◆ getAllIncomingQPILinkBytes() [2/2]

uint64 pcm::getAllIncomingQPILinkBytes ( const SystemCounterState & now )

inline

Get estimation of Socket QPI data traffic.

Returns an estimation of number of data bytes transferred to all sockets over all Intel(r) Quick Path Interconnect links

Parameters

now	System CPU counter state

Returns: Number of bytes

References pcm::getAllIncomingQPILinkBytes(), pcm::PCM::getInstance(), pcm::PCM::getNumSockets(), and pcm::getSocketIncomingQPILinkBytes().

◆ getAllOutgoingQPILinkBytes()

uint64 pcm::getAllOutgoingQPILinkBytes	(	const SystemCounterState &	before,
		const SystemCounterState &	after
	)

inline

Get estimation of total QPI data+nondata traffic.

Returns an estimation of number of data and non-data bytes transferred from all sockets over all Intel(r) Quick Path Interconnect links

Parameters

before	System CPU counter state before the experiment
after	System CPU counter state after the experiment

Returns: Number of bytes

References pcm::getAllOutgoingQPILinkBytes(), pcm::PCM::getInstance(), and pcm::PCM::getNumSockets().

Referenced by pcm::getAllOutgoingQPILinkBytes().

◆ getAverageFrequency()

template<class CounterStateType >

double pcm::getAverageFrequency	(	const CounterStateType &	before,
		const CounterStateType &	after
	)

Computes average core frequency also taking Intel Turbo Boost technology into account.

Parameters

before	CPU counter state before the experiment
after	CPU counter state after the experiment

Returns: frequency in Hz

References pcm::getAverageFrequency().

Referenced by pcm::getAverageFrequency().

◆ getAverageUncoreFrequency()

template<class UncoreStateType >

double pcm::getAverageUncoreFrequency	(	const UncoreStateType &	before,
		const UncoreStateType &	after
	)

Computes average uncore frequency.

Parameters

before	CPU counter state before the experiment
after	CPU counter state after the experiment

Returns: frequency in Hz

References pcm::getAverageUncoreFrequency(), and pcm::PCM::getInstance().

Referenced by pcm::getAverageUncoreFrequency().

◆ getBytesReadFromEDC()

template<class CounterStateType >

uint64 pcm::getBytesReadFromEDC	(	const CounterStateType &	before,
		const CounterStateType &	after
	)

Computes number of bytes read from MCDRAM memory controllers.

Parameters

before	CPU counter state before the experiment
after	CPU counter state after the experiment

Returns: Number of bytes

References pcm::getBytesReadFromEDC(), and pcm::PCM::getInstance().

Referenced by pcm::getBytesReadFromEDC().

◆ getBytesReadFromMC()

template<class CounterStateType >

uint64 pcm::getBytesReadFromMC	(	const CounterStateType &	before,
		const CounterStateType &	after
	)

Computes number of bytes read from DRAM memory controllers.

Parameters

before	CPU counter state before the experiment
after	CPU counter state after the experiment

Returns: Number of bytes

References pcm::getBytesReadFromMC(), and pcm::PCM::getInstance().

Referenced by pcm::getBytesReadFromMC(), and pcm::getQPItoMCTrafficRatio().

◆ getBytesReadFromPMM()

template<class CounterStateType >

uint64 pcm::getBytesReadFromPMM	(	const CounterStateType &	before,
		const CounterStateType &	after
	)

Computes number of bytes read from PMM memory.

Parameters

before	CPU counter state before the experiment
after	CPU counter state after the experiment

Returns: Number of bytes

References pcm::getBytesReadFromPMM(), and pcm::PCM::getInstance().

Referenced by pcm::getBytesReadFromPMM(), and pcm::getQPItoMCTrafficRatio().

◆ getBytesWrittenToEDC()

template<class CounterStateType >

uint64 pcm::getBytesWrittenToEDC	(	const CounterStateType &	before,
		const CounterStateType &	after
	)

Computes number of bytes written to MCDRAM memory controllers.

Parameters

before	CPU counter state before the experiment
after	CPU counter state after the experiment

Returns: Number of bytes

References pcm::getBytesWrittenToEDC(), and pcm::PCM::getInstance().

Referenced by pcm::getBytesWrittenToEDC().

◆ getBytesWrittenToMC()

template<class CounterStateType >

uint64 pcm::getBytesWrittenToMC	(	const CounterStateType &	before,
		const CounterStateType &	after
	)

Computes number of bytes written to DRAM memory controllers.

Parameters

before	CPU counter state before the experiment
after	CPU counter state after the experiment

Returns: Number of bytes

References pcm::getBytesWrittenToMC(), and pcm::PCM::getInstance().

Referenced by pcm::getBytesWrittenToMC(), and pcm::getQPItoMCTrafficRatio().

◆ getBytesWrittenToPMM()

template<class CounterStateType >

uint64 pcm::getBytesWrittenToPMM	(	const CounterStateType &	before,
		const CounterStateType &	after
	)

Computes number of bytes written to PMM memory.

Parameters

before	CPU counter state before the experiment
after	CPU counter state after the experiment

Returns: Number of bytes

References pcm::getBytesWrittenToPMM(), and pcm::PCM::getInstance().

Referenced by pcm::getBytesWrittenToPMM(), and pcm::getQPItoMCTrafficRatio().

◆ getCBOCounter()

template<class CounterStateType >

uint64 pcm::getCBOCounter	(	uint32	cbo,
		uint32	counter,
		const CounterStateType &	before,
		const CounterStateType &	after
	)

Direct read of CHA or CBO PMU counter (counter meaning depends on the programming: power/performance/etc)

Parameters

counter	counter number
cbo	cbo or cha number
before	CPU counter state before the experiment
after	CPU counter state after the experiment

References pcm::getCBOCounter().

Referenced by pcm::getCBOCounter().

◆ getConsumedEnergy()

template<class CounterStateType >

uint64 pcm::getConsumedEnergy	(	const CounterStateType &	before,
		const CounterStateType &	after
	)

Returns energy consumed by processor, excluding DRAM (measured in internal units)

Parameters

before	CPU counter state before the experiment
after	CPU counter state after the experiment

References pcm::getConsumedEnergy().

Referenced by pcm::getConsumedEnergy(), and pcm::getConsumedJoules().

◆ getConsumedJoules()

template<class CounterStateType >

double pcm::getConsumedJoules	(	const CounterStateType &	before,
		const CounterStateType &	after
	)

Returns Joules consumed by processor (excluding DRAM)

Parameters

before	CPU counter state before the experiment
after	CPU counter state after the experiment

References pcm::getConsumedEnergy(), pcm::getConsumedJoules(), pcm::PCM::getInstance(), and pcm::PCM::getJoulesPerEnergyUnit().

Referenced by pcm::getConsumedJoules().

◆ getCoreCounterState()

PCM_API CoreCounterState pcm::getCoreCounterState ( uint32 core )

Reads the counter state of a (logical) core.

Helper function. Uses PCM object to access counters.

Parameters

core core id

Returns: State of counters in the core

References pcm::PCM::getCoreCounterState(), and pcm::PCM::getInstance().

◆ getCoreCStateResidency() [1/2]

template<class CounterStateType >

double pcm::getCoreCStateResidency	(	int	state,
		const CounterStateType &	before,
		const CounterStateType &	after
	)

inline

Computes residency in the core C-state.

Parameters

state	C-state
before	CPU counter state before the experiment
after	CPU counter state after the experiment

Returns: residence ratio (0..1): 0 - 0%, 1.0 - 100%

References pcm::getCoreCStateResidency(), pcm::PCM::getInstance(), pcm::getInvariantTSC(), pcm::getRefCycles(), and pcm::PCM::isCoreCStateResidencySupported().

Referenced by pcm::getCoreCStateResidency().

◆ getCoreCStateResidency() [2/2]

template<class CounterStateType >

uint64 pcm::getCoreCStateResidency	(	int	state,
		const CounterStateType &	now
	)

inline

Reads raw residency counter for the core C-state.

Parameters

state	C-state #
now	CPU counter state

Returns: raw residency value

References pcm::getCoreCStateResidency().

◆ getCoreIPC()

template<class CounterStateType >

double pcm::getCoreIPC	(	const CounterStateType &	before,
		const CounterStateType &	after
	)

inline

Computes average number of retired instructions per core cycle for the entire system combining instruction counts from logical cores to corresponding physical cores.

Use this metric to evaluate IPC improvement between SMT(Hyperthreading) on and SMT off.

Parameters

before	CPU counter state before the experiment
after	CPU counter state after the experiment

Returns: IPC

References pcm::getCoreIPC(), pcm::PCM::getInstance(), pcm::getIPC(), pcm::PCM::getNumCores(), pcm::PCM::getNumOnlineCores(), and pcm::PCM::getThreadsPerCore().

Referenced by pcm::getCoreIPC().

◆ getCycles() [1/2]

template<class CounterStateType >

uint64 pcm::getCycles	(	const CounterStateType &	before,
		const CounterStateType &	after
	)

Computes the number core clock cycles when signal on a specific core is running (not halted)

Returns number of used cycles (halted cyles are not counted). The counter does not advance in the following conditions:

an ACPI C-state is other than C0 for normal operation
HLT
STPCLK+ pin is asserted
being throttled by TM1
during the frequency switching phase of a performance state transition

The performance counter for this event counts across performance state transitions using different core clock frequencies

Parameters

before	CPU counter state before the experiment
after	CPU counter state after the experiment

Returns: number core clock cycles

References pcm::getCycles().

Referenced by pcm::getCycles().

◆ getCycles() [2/2]

template<class CounterStateType >

uint64 pcm::getCycles ( const CounterStateType & now )

Computes the number executed core clock cycles.

Returns number of used cycles (halted cyles are not counted).

Parameters

now	Current CPU counter state

Returns: number core clock cycles

References pcm::getCycles().

◆ getDRAMClocks()

template<class CounterStateType >

uint64 pcm::getDRAMClocks	(	uint32	channel,
		const CounterStateType &	before,
		const CounterStateType &	after
	)

Returns DRAM clock ticks.

Parameters

channel	DRAM channel number
before	CPU counter state before the experiment
after	CPU counter state after the experiment

References pcm::PCM::getCPUModel(), pcm::getDRAMClocks(), and pcm::PCM::getInstance().

Referenced by pcm::getDRAMClocks().

◆ getDRAMConsumedEnergy()

template<class CounterStateType >

uint64 pcm::getDRAMConsumedEnergy	(	const CounterStateType &	before,
		const CounterStateType &	after
	)

Returns energy consumed by DRAM (measured in internal units)

Parameters

before	CPU counter state before the experiment
after	CPU counter state after the experiment

References pcm::getDRAMConsumedEnergy().

Referenced by pcm::getDRAMConsumedEnergy(), and pcm::getDRAMConsumedJoules().

◆ getDRAMConsumedJoules()

template<class CounterStateType >

double pcm::getDRAMConsumedJoules	(	const CounterStateType &	before,
		const CounterStateType &	after
	)

Returns Joules consumed by DRAM.

Parameters

before	CPU counter state before the experiment
after	CPU counter state after the experiment

References pcm::PCM::getCPUModel(), pcm::getDRAMConsumedEnergy(), pcm::getDRAMConsumedJoules(), pcm::PCM::getInstance(), and pcm::PCM::getJoulesPerEnergyUnit().

Referenced by pcm::getDRAMConsumedJoules().

◆ getEDCCounter()

template<class CounterStateType >

uint64 pcm::getEDCCounter	(	uint32	channel,
		uint32	counter,
		const CounterStateType &	before,
		const CounterStateType &	after
	)

Direct read of embedded DRAM memory controller counter (counter meaning depends on the programming: power/performance/etc)

Parameters

counter	counter number
channel	channel number
before	CPU counter state before the experiment
after	CPU counter state after the experiment

References pcm::getEDCCounter(), and pcm::PCM::getInstance().

Referenced by pcm::getEDCCounter().

◆ getExecUsage()

template<class CounterStateType >

double pcm::getExecUsage	(	const CounterStateType &	before,
		const CounterStateType &	after
	)

Computes average number of retired instructions per time interval.

Parameters

before	CPU counter state before the experiment
after	CPU counter state after the experiment

Returns: usage

References pcm::getExecUsage().

Referenced by pcm::getExecUsage(), and pcm::getTotalExecUsage().

◆ getFreeRunningCounter()

template<class CounterStateType >

int64 pcm::getFreeRunningCounter	(	const typename CounterStateType::FreeRunningCounterID &	counter,
		const CounterStateType &	before,
		const CounterStateType &	after
	)

Returns free running counter if it exists, -1 otherwise.

Parameters

counter	name of the counter
before	CPU counter state before the experiment
after	CPU counter state after the experiment

References pcm::getFreeRunningCounter().

Referenced by pcm::getFreeRunningCounter().

◆ getGTRequestBytesFromMC()

template<class CounterStateType >

uint64 pcm::getGTRequestBytesFromMC	(	const CounterStateType &	before,
		const CounterStateType &	after
	)

Computes number of bytes of read/write requests from GT engine.

Parameters

before	CPU counter state before the experiment
after	CPU counter state after the experiment

Returns: Number of bytes

References pcm::getGTRequestBytesFromMC(), and pcm::PCM::getInstance().

Referenced by pcm::getGTRequestBytesFromMC().

◆ getIARequestBytesFromMC()

template<class CounterStateType >

uint64 pcm::getIARequestBytesFromMC	(	const CounterStateType &	before,
		const CounterStateType &	after
	)

Computes number of bytes of read/write requests from all IA.

Parameters

before	CPU counter state before the experiment
after	CPU counter state after the experiment

Returns: Number of bytes

References pcm::getIARequestBytesFromMC(), and pcm::PCM::getInstance().

Referenced by pcm::getIARequestBytesFromMC().

◆ getIIOCounter()

template<class CounterStateType >

uint64 pcm::getIIOCounter	(	uint32	stack,
		uint32	counter,
		const CounterStateType &	before,
		const CounterStateType &	after
	)

Direct read of IIO PMU counter (counter meaning depends on the programming: power/performance/etc)

Parameters

counter	counter number
stack	IIO stack number
before	CPU counter state before the experiment
after	CPU counter state after the experiment

References pcm::getIIOCounter().

Referenced by pcm::getIIOCounter().

◆ getIncomingQPILinkBytes() [1/2]

uint64 pcm::getIncomingQPILinkBytes	(	uint32	socketNr,
		uint32	linkNr,
		const SystemCounterState &	before,
		const SystemCounterState &	after
	)

inline

Get estimation of QPI data traffic per incoming QPI link.

Returns an estimation of number of data bytes transferred to a socket over Intel(r) Quick Path Interconnect

Parameters

socketNr	socket identifier
linkNr	linkNr
before	System CPU counter state before the experiment
after	System CPU counter state after the experiment

Returns: Number of bytes

References pcm::getIncomingQPILinkBytes(), and pcm::PCM::getInstance().

Referenced by pcm::getIncomingQPILinkBytes(), and pcm::getIncomingQPILinkUtilization().

◆ getIncomingQPILinkBytes() [2/2]

uint64 pcm::getIncomingQPILinkBytes	(	uint32	socketNr,
		uint32	linkNr,
		const SystemCounterState &	now
	)

inline

Return current value of the counter of QPI data traffic per incoming QPI link.

Returns the number of incoming data bytes to a socket over Intel(r) Quick Path Interconnect

Parameters

socketNr	socket identifier
linkNr	linkNr
now	Current System CPU counter state

Returns: Number of bytes

References pcm::getIncomingQPILinkBytes(), and pcm::PCM::getInstance().

◆ getIncomingQPILinkUtilization()

double pcm::getIncomingQPILinkUtilization	(	uint32	socketNr,
		uint32	linkNr,
		const SystemCounterState &	before,
		const SystemCounterState &	after
	)

inline

Get data utilization of incoming QPI link (0..1)

Returns an estimation of utilization of QPI link by data traffic transferred to a socket over Intel(r) Quick Path Interconnect

Parameters

socketNr	socket identifier
linkNr	linkNr
before	System CPU counter state before the experiment
after	System CPU counter state after the experiment

Returns: utilization (0..1)

References pcm::getIncomingQPILinkBytes(), pcm::getIncomingQPILinkUtilization(), pcm::PCM::getInstance(), pcm::getInvariantTSC(), pcm::PCM::getNominalFrequency(), pcm::PCM::getNumOnlineCores(), and pcm::PCM::getQPILinkSpeed().

Referenced by pcm::getIncomingQPILinkUtilization().

◆ getInstructionsRetired() [1/2]

template<class CounterStateType >

uint64 pcm::getInstructionsRetired	(	const CounterStateType &	before,
		const CounterStateType &	after
	)

Computes the number of retired instructions.

Parameters

before	CPU counter state before the experiment
after	CPU counter state after the experiment

Returns: number of retired instructions

References pcm::getInstructionsRetired().

Referenced by pcm::getInstructionsRetired().

◆ getInstructionsRetired() [2/2]

template<class CounterStateType >

uint64 pcm::getInstructionsRetired ( const CounterStateType & now )

Computes the number of retired instructions.

Parameters

now	Current CPU counter state

Returns: number of retired instructions

References pcm::getInstructionsRetired().

◆ getInvariantTSC()

template<class CounterStateType >

uint64 pcm::getInvariantTSC	(	const CounterStateType &	before,
		const CounterStateType &	after
	)

Computes number of invariant time stamp counter ticks.

This counter counts irrespectively of C-, P- or T-states

Parameters

before	CPU counter state before the experiment
after	CPU counter state after the experiment

Returns: number of time stamp counter ticks

References pcm::getInvariantTSC().

Referenced by pcm::getCoreCStateResidency(), pcm::getIncomingQPILinkUtilization(), pcm::getInvariantTSC(), pcm::getLLCReadMissLatency(), pcm::getOutgoingQPILinkBytes(), pcm::getOutgoingQPILinkUtilization(), and pcm::getPackageCStateResidency().

◆ getIORequestBytesFromMC()

template<class CounterStateType >

uint64 pcm::getIORequestBytesFromMC	(	const CounterStateType &	before,
		const CounterStateType &	after
	)

Computes number of bytes of read/write requests from all IO sources.

Parameters

before	CPU counter state before the experiment
after	CPU counter state after the experiment

Returns: Number of bytes

References pcm::PCM::getInstance(), and pcm::getIORequestBytesFromMC().

Referenced by pcm::getIORequestBytesFromMC().

◆ getIPC()

template<class CounterStateType >

double pcm::getIPC	(	const CounterStateType &	before,
		const CounterStateType &	after
	)

Computes average number of retired instructions per core cycle (IPC)

Parameters

before	CPU counter state before the experiment
after	CPU counter state after the experiment

Returns: IPC

References pcm::getIPC().

Referenced by pcm::getCoreIPC(), and pcm::getIPC().

◆ getIRPCounter()

template<class CounterStateType >

uint64 pcm::getIRPCounter	(	uint32	stack,
		uint32	counter,
		const CounterStateType &	before,
		const CounterStateType &	after
	)

Direct read of IRP PMU counter (counter meaning depends on the programming: power/performance/etc)

Parameters

counter	counter number
stack	IIO stack number
before	CPU counter state before the experiment
after	CPU counter state after the experiment

References pcm::getIRPCounter().

Referenced by pcm::getIRPCounter().

◆ getL2CacheHitRatio()

template<class CounterStateType >

double pcm::getL2CacheHitRatio	(	const CounterStateType &	before,
		const CounterStateType &	after
	)

Computes L2 cache hit ratio.

Parameters

before	CPU counter state before the experiment
after	CPU counter state after the experiment

Warning: Works only in the DEFAULT_EVENTS programming mode (see program() method)

Returns: value between 0 and 1

References pcm::PCM::getInstance(), pcm::getL2CacheHitRatio(), pcm::getL2CacheHits(), and pcm::getL2CacheMisses().

Referenced by pcm::getL2CacheHitRatio().

◆ getL2CacheHits()

template<class CounterStateType >

uint64 pcm::getL2CacheHits	(	const CounterStateType &	before,
		const CounterStateType &	after
	)

Computes number of L2 cache hits.

Parameters

before	CPU counter state before the experiment
after	CPU counter state after the experiment

Warning: Works only in the DEFAULT_EVENTS programming mode (see program() method)

Returns: number of hits

References pcm::PCM::getInstance(), and pcm::getL2CacheHits().

Referenced by pcm::getL2CacheHitRatio(), and pcm::getL2CacheHits().

◆ getL2CacheMisses()

template<class CounterStateType >

uint64 pcm::getL2CacheMisses	(	const CounterStateType &	before,
		const CounterStateType &	after
	)

Computes number of L2 cache misses.

Parameters

before	CPU counter state before the experiment
after	CPU counter state after the experiment

Warning: Works only in the DEFAULT_EVENTS programming mode (see program() method)

Returns: number of misses

References pcm::PCM::getInstance(), and pcm::getL2CacheMisses().

Referenced by pcm::getL2CacheHitRatio(), and pcm::getL2CacheMisses().

◆ getL3CacheHitRatio()

template<class CounterStateType >

double pcm::getL3CacheHitRatio	(	const CounterStateType &	before,
		const CounterStateType &	after
	)

Computes L3 cache hit ratio.

Parameters

before	CPU counter state before the experiment
after	CPU counter state after the experiment

Warning: Works only in the DEFAULT_EVENTS programming mode (see program() method)

Returns: value between 0 and 1

References pcm::PCM::getInstance(), pcm::getL3CacheHitRatio(), pcm::getL3CacheHits(), and pcm::getL3CacheMisses().

Referenced by pcm::getL3CacheHitRatio().

◆ getL3CacheHits()

template<class CounterStateType >

uint64 pcm::getL3CacheHits	(	const CounterStateType &	before,
		const CounterStateType &	after
	)

Computes total number of L3 cache hits.

Parameters

before	CPU counter state before the experiment
after	CPU counter state after the experiment

Warning: Works only in the DEFAULT_EVENTS programming mode (see program() method)

Returns: number of hits

References pcm::PCM::getInstance(), pcm::getL3CacheHits(), pcm::getL3CacheHitsNoSnoop(), and pcm::getL3CacheHitsSnoop().

Referenced by pcm::getL3CacheHitRatio(), and pcm::getL3CacheHits().

◆ getL3CacheHitsNoSnoop()

template<class CounterStateType >

uint64 pcm::getL3CacheHitsNoSnoop	(	const CounterStateType &	before,
		const CounterStateType &	after
	)

Computes number of L3 cache hits where no snooping in sibling L2 caches had to be done.

Parameters

before	CPU counter state before the experiment
after	CPU counter state after the experiment

Warning: Works only in the DEFAULT_EVENTS programming mode (see program() method)

Returns: number of hits

References pcm::PCM::getInstance(), and pcm::getL3CacheHitsNoSnoop().

Referenced by pcm::getL3CacheHits(), and pcm::getL3CacheHitsNoSnoop().

◆ getL3CacheHitsSnoop()

template<class CounterStateType >

uint64 pcm::getL3CacheHitsSnoop	(	const CounterStateType &	before,
		const CounterStateType &	after
	)

Computes number of L3 cache hits where snooping in sibling L2 caches had to be done.

Parameters

before	CPU counter state before the experiment
after	CPU counter state after the experiment

Warning: Works only in the DEFAULT_EVENTS programming mode (see program() method)

Returns: number of hits

References pcm::PCM::getInstance(), pcm::getL3CacheHitsSnoop(), and pcm::getL3CacheMisses().

Referenced by pcm::getL3CacheHits(), and pcm::getL3CacheHitsSnoop().

◆ getL3CacheMisses()

template<class CounterStateType >

uint64 pcm::getL3CacheMisses	(	const CounterStateType &	before,
		const CounterStateType &	after
	)

Computes number of L3 cache misses.

Parameters

before	CPU counter state before the experiment
after	CPU counter state after the experiment

Warning: Works only in the DEFAULT_EVENTS programming mode (see program() method)

Returns: number of misses

References pcm::PCM::getInstance(), and pcm::getL3CacheMisses().

Referenced by pcm::getL3CacheHitRatio(), pcm::getL3CacheHitsSnoop(), and pcm::getL3CacheMisses().

◆ getL3CacheOccupancy()

template<class CounterStateType >

uint64 pcm::getL3CacheOccupancy ( const CounterStateType & now )

Computes L3 Cache Occupancy.

References pcm::PCM::getInstance(), and pcm::getL3CacheOccupancy().

Referenced by pcm::getL3CacheOccupancy().

◆ getLocalMemoryBW()

template<class CounterStateType >

uint64 pcm::getLocalMemoryBW	(	const CounterStateType &	before,
		const CounterStateType &	after
	)

Computes Local Memory Bandwidth.

References pcm::PCM::getInstance(), and pcm::getLocalMemoryBW().

Referenced by pcm::getLocalMemoryBW(), and pcm::getRemoteMemoryBW().

◆ getLocalMemoryRequestRatio()

template<class CounterStateType >

double pcm::getLocalMemoryRequestRatio	(	const CounterStateType &	before,
		const CounterStateType &	after
	)

inline

Get local memory access ration measured in home agent.

Parameters

before	System CPU counter state before the experiment
after	System CPU counter state after the experiment

Returns: Ratio

References pcm::PCM::getInstance(), and pcm::getLocalMemoryRequestRatio().

Referenced by pcm::getLocalMemoryRequestRatio().

◆ getM2MCounter()

template<class CounterStateType >

uint64 pcm::getM2MCounter	(	uint32	controller,
		uint32	counter,
		const CounterStateType &	before,
		const CounterStateType &	after
	)

Direct read of Memory2Mesh controller PMU counter (counter meaning depends on the programming: power/performance/etc)

Parameters

counter	counter number
controller	controller number
before	CPU counter state before the experiment
after	CPU counter state after the experiment

References pcm::getM2MCounter().

Referenced by pcm::getM2MCounter().

◆ getM3UPICounter()

template<class CounterStateType >

uint64 pcm::getM3UPICounter	(	uint32	port,
		uint32	counter,
		const CounterStateType &	before,
		const CounterStateType &	after
	)

Direct read of M3UPI PMU counter (counter meaning depends on the programming: power/performance/etc)

Parameters

counter	counter number
port	UPI port number
before	CPU counter state before the experiment
after	CPU counter state after the experiment

References pcm::getM3UPICounter().

Referenced by pcm::getM3UPICounter().

◆ getMCCounter()

template<class CounterStateType >

uint64 pcm::getMCCounter	(	uint32	channel,
		uint32	counter,
		const CounterStateType &	before,
		const CounterStateType &	after
	)

Direct read of memory controller PMU counter (counter meaning depends on the programming: power/performance/etc)

Parameters

counter	counter number
channel	channel number
before	CPU counter state before the experiment
after	CPU counter state after the experiment

References pcm::getMCCounter().

Referenced by pcm::getMCCounter().

◆ getMCDRAMClocks()

template<class CounterStateType >

uint64 pcm::getMCDRAMClocks	(	uint32	channel,
		const CounterStateType &	before,
		const CounterStateType &	after
	)

Returns MCDRAM clock ticks.

Parameters

channel	MCDRAM channel number
before	CPU counter state before the experiment
after	CPU counter state after the experiment

References pcm::getMCDRAMClocks().

Referenced by pcm::getMCDRAMClocks().

◆ getNormalizedQPIL0pTxCycles()

template<class CounterStateType >

double pcm::getNormalizedQPIL0pTxCycles	(	uint32	port,
		const CounterStateType &	before,
		const CounterStateType &	after
	)

Returns the ratio of QPI cycles in power saving half-lane mode.

Parameters

port	QPI port number
before	CPU counter state before the experiment
after	CPU counter state after the experiment

Returns: 0..1 - ratio of QPI cycles in power saving half-lane mode

References pcm::getNormalizedQPIL0pTxCycles(), pcm::getQPIClocks(), and pcm::getQPIL0pTxCycles().

Referenced by pcm::getNormalizedQPIL0pTxCycles().

◆ getNormalizedQPIL1Cycles()

template<class CounterStateType >

double pcm::getNormalizedQPIL1Cycles	(	uint32	port,
		const CounterStateType &	before,
		const CounterStateType &	after
	)

Returns the ratio of QPI cycles in power saving shutdown mode.

Parameters

port	QPI port number
before	CPU counter state before the experiment
after	CPU counter state after the experiment

Returns: 0..1 - ratio of QPI cycles in power saving shutdown mode

References pcm::getNormalizedQPIL1Cycles(), pcm::getQPIClocks(), and pcm::getQPIL1Cycles().

Referenced by pcm::getNormalizedQPIL1Cycles().

◆ getNumberOfCustomEvents()

template<class CounterStateType >

uint64 pcm::getNumberOfCustomEvents	(	int32	eventCounterNr,
		const CounterStateType &	before,
		const CounterStateType &	after
	)

Returns the number of occurred custom core events.

Read number of events programmed with the CUSTOM_CORE_EVENTS

Parameters

eventCounterNr	Event/counter number (value from 0 to 3)
before	CPU counter state before the experiment
after	CPU counter state after the experiment

Returns: Number of bytes

References pcm::getNumberOfCustomEvents().

Referenced by pcm::getNumberOfCustomEvents().

◆ getNumberOfEvents()

template<class CounterType >

uint64 pcm::getNumberOfEvents	(	const CounterType &	before,
		const CounterType &	after
	)

inline

Returns the raw count of events.

Parameters

before	counter state before the experiment
after	counter state after the experiment

References pcm::getNumberOfEvents().

Referenced by pcm::getNumberOfEvents().

◆ getOutgoingQPILinkBytes()

uint64 pcm::getOutgoingQPILinkBytes	(	uint32	socketNr,
		uint32	linkNr,
		const SystemCounterState &	before,
		const SystemCounterState &	after
	)

inline

Get estimation of QPI (data+nondata) traffic per outgoing QPI link.

Returns an estimation of number of data bytes transferred from a socket over Intel(r) Quick Path Interconnect

Parameters

socketNr	socket identifier
linkNr	linkNr
before	System CPU counter state before the experiment
after	System CPU counter state after the experiment

Returns: Number of bytes

References pcm::PCM::getInstance(), pcm::getInvariantTSC(), pcm::PCM::getNominalFrequency(), pcm::PCM::getNumOnlineCores(), pcm::getOutgoingQPILinkBytes(), pcm::getOutgoingQPILinkUtilization(), and pcm::PCM::getQPILinkSpeed().

Referenced by pcm::getOutgoingQPILinkBytes().

◆ getOutgoingQPILinkUtilization()

double pcm::getOutgoingQPILinkUtilization	(	uint32	socketNr,
		uint32	linkNr,
		const SystemCounterState &	before,
		const SystemCounterState &	after
	)

inline

Get utilization of outgoing QPI link (0..1)

Returns an estimation of utilization of QPI link by (data+nondata) traffic transferred from a socket over Intel(r) Quick Path Interconnect

Parameters

socketNr	socket identifier
linkNr	linkNr
before	System CPU counter state before the experiment
after	System CPU counter state after the experiment

Returns: utilization (0..1)

References pcm::PCM::getInstance(), pcm::getInvariantTSC(), pcm::PCM::getNominalFrequency(), pcm::PCM::getNumOnlineCores(), pcm::getOutgoingQPILinkUtilization(), and pcm::PCM::getQPILinkSpeed().

Referenced by pcm::getOutgoingQPILinkBytes(), and pcm::getOutgoingQPILinkUtilization().

◆ getPackageCStateResidency() [1/2]

template<class CounterStateType >

double pcm::getPackageCStateResidency	(	int	state,
		const CounterStateType &	before,
		const CounterStateType &	after
	)

inline

Computes residency in the package C-state.

Parameters

state	C-state
before	CPU counter state before the experiment
after	CPU counter state after the experiment

Returns: residence ratio (0..1): 0 - 0%, 1.0 - 100%

References pcm::PCM::getInstance(), pcm::getInvariantTSC(), pcm::getPackageCStateResidency(), and pcm::PCM::isPackageCStateResidencySupported().

Referenced by pcm::getPackageCStateResidency().

◆ getPackageCStateResidency() [2/2]

template<class CounterStateType >

uint64 pcm::getPackageCStateResidency	(	int	state,
		const CounterStateType &	now
	)

inline

Reads raw residency counter for the package C-state.

Parameters

state	C-state #
now	CPU counter state

Returns: raw residency value

References pcm::getPackageCStateResidency().

◆ getPCUClocks()

template<class CounterStateType >

uint64 pcm::getPCUClocks	(	const CounterStateType &	before,
		const CounterStateType &	after
	)

Returns clock ticks of power control unit.

Parameters

before	CPU counter state before the experiment
after	CPU counter state after the experiment

References pcm::getPCUClocks(), and pcm::getPCUCounter().

Referenced by pcm::getPCUClocks().

◆ getPCUCounter()

template<class CounterStateType >

uint64 pcm::getPCUCounter	(	uint32	counter,
		const CounterStateType &	before,
		const CounterStateType &	after
	)

Direct read of power control unit PMU counter (counter meaning depends on the programming: power/performance/etc)

Parameters

counter	counter number
before	CPU counter state before the experiment
after	CPU counter state after the experiment

References pcm::getPCUCounter().

Referenced by pcm::getPCUClocks(), and pcm::getPCUCounter().

◆ getQPIClocks()

template<class CounterStateType >

uint64 pcm::getQPIClocks	(	uint32	port,
		const CounterStateType &	before,
		const CounterStateType &	after
	)

Returns QPI LL clock ticks.

Parameters

port	QPI port number
before	CPU counter state before the experiment
after	CPU counter state after the experiment

References pcm::getQPIClocks(), and pcm::getXPICounter().

Referenced by pcm::getNormalizedQPIL0pTxCycles(), pcm::getNormalizedQPIL1Cycles(), and pcm::getQPIClocks().

◆ getQPIL0pTxCycles()

template<class CounterStateType >

uint64 pcm::getQPIL0pTxCycles	(	uint32	port,
		const CounterStateType &	before,
		const CounterStateType &	after
	)

Returns the number of QPI cycles in power saving half-lane mode.

Parameters

port	QPI port number
before	CPU counter state before the experiment
after	CPU counter state after the experiment

References pcm::getQPIL0pTxCycles(), and pcm::getXPICounter().

Referenced by pcm::getNormalizedQPIL0pTxCycles(), and pcm::getQPIL0pTxCycles().

◆ getQPIL1Cycles()

template<class CounterStateType >

uint64 pcm::getQPIL1Cycles	(	uint32	port,
		const CounterStateType &	before,
		const CounterStateType &	after
	)

Returns the number of QPI cycles in power saving shutdown mode.

Parameters

port	QPI port number
before	CPU counter state before the experiment
after	CPU counter state after the experiment

References pcm::getQPIL1Cycles(), and pcm::getXPICounter().

Referenced by pcm::getNormalizedQPIL1Cycles(), and pcm::getQPIL1Cycles().

◆ getQPItoMCTrafficRatio()

double pcm::getQPItoMCTrafficRatio	(	const SystemCounterState &	before,
		const SystemCounterState &	after
	)

inline

Get QPI data to Memory Controller traffic ratio.

Ideally for NUMA-optmized programs the ratio should be close to 0.

Parameters

before	System CPU counter state before the experiment
after	System CPU counter state after the experiment

Returns: Ratio

References pcm::getAllIncomingQPILinkBytes(), pcm::getBytesReadFromMC(), pcm::getBytesReadFromPMM(), pcm::getBytesWrittenToMC(), pcm::getBytesWrittenToPMM(), pcm::PCM::getInstance(), and pcm::getQPItoMCTrafficRatio().

Referenced by pcm::getQPItoMCTrafficRatio().

◆ getRefCycles()

template<class CounterStateType >

uint64 pcm::getRefCycles	(	const CounterStateType &	before,
		const CounterStateType &	after
	)

Computes the number of reference clock cycles while clock signal on the core is running.

The reference clock operates at a fixed frequency, irrespective of core frequency changes due to performance state transitions. See Intel(r) Software Developer's Manual for more details

Parameters

before	CPU counter state before the experiment
after	CPU counter state after the experiment

Returns: number core clock cycles

References pcm::getRefCycles().

Referenced by pcm::getCoreCStateResidency(), and pcm::getRefCycles().

◆ getRelativeFrequency()

template<class CounterStateType >

double pcm::getRelativeFrequency	(	const CounterStateType &	before,
		const CounterStateType &	after
	)

Computes average core frequency also taking Intel Turbo Boost technology into account.

Parameters

before	CPU counter state before the experiment
after	CPU counter state after the experiment

Returns: Fraction of nominal frequency

References pcm::getRelativeFrequency().

Referenced by pcm::getRelativeFrequency().

◆ getRemoteMemoryBW()

template<class CounterStateType >

uint64 pcm::getRemoteMemoryBW	(	const CounterStateType &	before,
		const CounterStateType &	after
	)

Computes Remote Memory Bandwidth.

References pcm::PCM::getInstance(), pcm::getLocalMemoryBW(), and pcm::getRemoteMemoryBW().

Referenced by pcm::getRemoteMemoryBW().

◆ getSMICount()

template<class CounterStateType >

uint64 pcm::getSMICount	(	const CounterStateType &	before,
		const CounterStateType &	after
	)

Returns the number of occurred system management interrupts.

Parameters

before	CPU counter state before the experiment
after	CPU counter state after the experiment

Returns: Number of SMIs (system manegement interrupts)

References pcm::getSMICount().

Referenced by pcm::getSMICount().

◆ getSocketCounterState()

PCM_API SocketCounterState pcm::getSocketCounterState ( uint32 socket )

Reads the counter state of a socket.

Helper function. Uses PCM object to access counters.

Parameters

socket socket id

Returns: State of counters in the socket

References pcm::PCM::getInstance(), and pcm::PCM::getSocketCounterState().

◆ getSocketIncomingQPILinkBytes()

uint64 pcm::getSocketIncomingQPILinkBytes	(	uint32	socketNr,
		const SystemCounterState &	now
	)

inline

Get estimation of total QPI data traffic for this socket.

Returns an estimation of number of bytes transferred to this sockets over all Intel(r) Quick Path Interconnect links on this socket

Parameters

before	System CPU counter state before the experiment
after	System CPU counter state after the experiment

Returns: Number of bytes

References pcm::PCM::getInstance(), and pcm::getSocketIncomingQPILinkBytes().

Referenced by pcm::getAllIncomingQPILinkBytes(), and pcm::getSocketIncomingQPILinkBytes().

◆ getSystemCounterState()

PCM_API SystemCounterState pcm::getSystemCounterState ( )

Reads the counter state of the system.

Helper function. Uses PCM object to access counters.

System consists of several sockets (CPUs). Socket has a CPU in it. Socket (CPU) consists of several (logical) cores.

Returns: State of counters in the entire system

References pcm::PCM::getInstance(), and pcm::PCM::getSystemCounterState().

◆ getTotalExecUsage()

template<class CounterStateType >

double pcm::getTotalExecUsage	(	const CounterStateType &	before,
		const CounterStateType &	after
	)

inline

Computes average number of retired instructions per time interval for the entire system combining instruction counts from logical cores to corresponding physical cores.

Use this metric to evaluate cores utilization improvement between SMT(Hyperthreading) on and SMT off.

Parameters

before	CPU counter state before the experiment
after	CPU counter state after the experiment

Returns: usage

References pcm::getExecUsage(), pcm::PCM::getInstance(), pcm::PCM::getNumCores(), pcm::PCM::getNumOnlineCores(), pcm::PCM::getThreadsPerCore(), and pcm::getTotalExecUsage().

Referenced by pcm::getTotalExecUsage().

◆ getUBOXCounter()

template<class CounterStateType >

uint64 pcm::getUBOXCounter	(	uint32	counter,
		const CounterStateType &	before,
		const CounterStateType &	after
	)

Direct read of UBOX PMU counter (counter meaning depends on the programming: power/performance/etc)

Parameters

counter	counter number
before	CPU counter state before the experiment
after	CPU counter state after the experiment

References pcm::getUBOXCounter().

Referenced by pcm::getUBOXCounter().

◆ getUncoreClocks()

template<class CounterStateType >

uint64 pcm::getUncoreClocks	(	const CounterStateType &	before,
		const CounterStateType &	after
	)

Returns uncore clock ticks.

Parameters

before	CPU counter state before the experiment
after	CPU counter state after the experiment

References pcm::getUncoreClocks().

Referenced by pcm::getUncoreClocks(), and pcm::PCM::isClientCPU().

◆ getXPICounter()

template<class CounterStateType >

uint64 pcm::getXPICounter	(	uint32	port,
		uint32	counter,
		const CounterStateType &	before,
		const CounterStateType &	after
	)

Direct read of UPI or QPI PMU counter (counter meaning depends on the programming: power/performance/etc)

Parameters

counter	counter number
port	UPI/QPI port number
before	CPU counter state before the experiment
after	CPU counter state after the experiment

References pcm::getXPICounter().

Referenced by pcm::getQPIClocks(), pcm::getQPIL0pTxCycles(), pcm::getQPIL1Cycles(), and pcm::getXPICounter().

Classes

Macros

Typedefs

Enumerations

Functions

Detailed Description

Macro Definition Documentation

◆ PCM_SKL_PATH_CASES

Function Documentation

◆ getActiveAverageFrequency()

◆ getActiveRelativeFrequency()

◆ getAllIncomingQPILinkBytes() [1/2]

◆ getAllIncomingQPILinkBytes() [2/2]

◆ getAllOutgoingQPILinkBytes()

◆ getAverageFrequency()

◆ getAverageUncoreFrequency()

◆ getBytesReadFromEDC()

◆ getBytesReadFromMC()

◆ getBytesReadFromPMM()

◆ getBytesWrittenToEDC()

◆ getBytesWrittenToMC()

◆ getBytesWrittenToPMM()

◆ getCBOCounter()

◆ getConsumedEnergy()

◆ getConsumedJoules()

◆ getCoreCounterState()

◆ getCoreCStateResidency() [1/2]

◆ getCoreCStateResidency() [2/2]

◆ getCoreIPC()

◆ getCycles() [1/2]

◆ getCycles() [2/2]

◆ getDRAMClocks()

◆ getDRAMConsumedEnergy()

◆ getDRAMConsumedJoules()

◆ getEDCCounter()

◆ getExecUsage()

◆ getFreeRunningCounter()

◆ getGTRequestBytesFromMC()

◆ getIARequestBytesFromMC()

◆ getIIOCounter()

◆ getIncomingQPILinkBytes() [1/2]

◆ getIncomingQPILinkBytes() [2/2]

◆ getIncomingQPILinkUtilization()

◆ getInstructionsRetired() [1/2]

◆ getInstructionsRetired() [2/2]

◆ getInvariantTSC()

◆ getIORequestBytesFromMC()

◆ getIPC()

◆ getIRPCounter()

◆ getL2CacheHitRatio()

◆ getL2CacheHits()

◆ getL2CacheMisses()

◆ getL3CacheHitRatio()

◆ getL3CacheHits()

◆ getL3CacheHitsNoSnoop()

◆ getL3CacheHitsSnoop()

◆ getL3CacheMisses()

◆ getL3CacheOccupancy()

◆ getLocalMemoryBW()

◆ getLocalMemoryRequestRatio()

◆ getM2MCounter()

◆ getM3UPICounter()

◆ getMCCounter()

◆ getMCDRAMClocks()

◆ getNormalizedQPIL0pTxCycles()

◆ getNormalizedQPIL1Cycles()

◆ getNumberOfCustomEvents()

◆ getNumberOfEvents()

◆ getOutgoingQPILinkBytes()

◆ getOutgoingQPILinkUtilization()

◆ getPackageCStateResidency() [1/2]

◆ getPackageCStateResidency() [2/2]

◆ getPCUClocks()

◆ getPCUCounter()

◆ getQPIClocks()

◆ getQPIL0pTxCycles()

◆ getQPIL1Cycles()

◆ getQPItoMCTrafficRatio()

◆ getRefCycles()

◆ getRelativeFrequency()