#include <GPUReconstructionProcessing.h>

Inherited by o2::gpu::GPUReconstructionCPUBackend.

Classes
struct	RecoStepTimerMeta

struct	timerMeta

Public Member Functions
	~GPUReconstructionProcessing () override=default

int32_t	getNKernelHostThreads (bool splitCores)

uint32_t	getNActiveThreadsOuterLoop () const

void	SetNActiveThreadsOuterLoop (uint32_t f)

uint32_t	SetAndGetNActiveThreadsOuterLoop (bool condition, uint32_t max)

void	runParallelOuterLoop (bool doGPU, uint32_t nThreads, std::function< void(uint32_t)> lambda)

void	SetNActiveThreads (int32_t n)

auto &	getRecoStepTimer (RecoStep step)

HighResTimer &	getGeneralStepTimer (GeneralStep step)

template<class T >
void	AddGPUEvents (T *&events)

virtual std::unique_ptr< gpu_reconstruction_kernels::threadContext >	GetThreadContext () override

Public Member Functions inherited from o2::gpu::GPUReconstruction
virtual	~GPUReconstruction ()

	GPUReconstruction (const GPUReconstruction &)=delete

GPUReconstruction &	operator= (const GPUReconstruction &)=delete

template<class T , typename... Args>
T *	AddChain (Args... args)

int32_t	Init ()

int32_t	Finalize ()

int32_t	Exit ()

void	DumpSettings (const char *dir="")

int32_t	ReadSettings (const char *dir="")

void	PrepareEvent ()

virtual int32_t	RunChains ()=0

uint32_t	getNEventsProcessed ()

uint32_t	getNEventsProcessedInStat ()

int32_t	registerMemoryForGPU (const void *ptr, size_t size)

int32_t	unregisterMemoryForGPU (const void *ptr)

virtual void *	getGPUPointer (void *ptr)

virtual void	startGPUProfiling ()

virtual void	endGPUProfiling ()

int32_t	CheckErrorCodes (bool cpuOnly=false, bool forceShowErrors=false, std::vector< std::array< uint32_t, 4 > > *fillErrors=nullptr)

void	RunPipelineWorker ()

void	TerminatePipelineWorker ()

GPUMemoryResource &	Res (int16_t num)

template<class T >
int16_t	RegisterMemoryAllocation (T proc, void (T::setPtr)(void ), int32_t type, const char *name="", const GPUMemoryReuse &re=GPUMemoryReuse())

size_t	AllocateMemoryResources ()

size_t	AllocateRegisteredMemory (GPUProcessor *proc, bool resetCustom=false)

size_t	AllocateRegisteredMemory (int16_t res, GPUOutputControl *control=nullptr)

void	AllocateRegisteredForeignMemory (int16_t res, GPUReconstruction rec, GPUOutputControl control=nullptr)

void *	AllocateUnmanagedMemory (size_t size, int32_t type)

void *	AllocateVolatileDeviceMemory (size_t size)

void *	AllocateVolatileMemory (size_t size, bool device)

void	FreeRegisteredMemory (GPUProcessor *proc, bool freeCustom=false, bool freePermanent=false)

void	FreeRegisteredMemory (int16_t res)

void	ClearAllocatedMemory (bool clearOutputs=true)

void	ReturnVolatileDeviceMemory ()

void	ReturnVolatileMemory ()

void	PushNonPersistentMemory (uint64_t tag)

void	PopNonPersistentMemory (RecoStep step, uint64_t tag)

void	BlockStackedMemory (GPUReconstruction *rec)

void	UnblockStackedMemory ()

void	ResetRegisteredMemoryPointers (GPUProcessor *proc)

void	ResetRegisteredMemoryPointers (int16_t res)

void	ComputeReuseMax (GPUProcessor *proc)

void	PrintMemoryStatistics ()

void	PrintMemoryOverview ()

void	PrintMemoryMax ()

void	SetMemoryExternalInput (int16_t res, void *ptr)

GPUMemorySizeScalers *	MemoryScalers ()

virtual void	GetITSTraits (std::unique_ptr< o2::its::TrackerTraits > trackerTraits, std::unique_ptr< o2::its::VertexerTraits > vertexerTraits, std::unique_ptr< o2::its::TimeFrame > *timeFrame)

bool	slavesExist ()

DeviceType	GetDeviceType () const

bool	IsGPU () const

const GPUParam &	GetParam () const

const GPUConstantMem &	GetConstantMem () const

const GPUSettingsGRP &	GetGRPSettings () const

const GPUSettingsDeviceBackend &	GetDeviceBackendSettings ()

const GPUSettingsProcessing &	GetProcessingSettings () const

bool	IsInitialized () const

void	SetSettings (float solenoidBzNominalGPU, const GPURecoStepConfiguration *workflow=nullptr)

void	SetSettings (const GPUSettingsGRP grp, const GPUSettingsRec rec=nullptr, const GPUSettingsProcessing proc=nullptr, const GPURecoStepConfiguration workflow=nullptr)

void	SetResetTimers (bool reset)

void	SetDebugLevelTmp (int32_t level)

void	UpdateSettings (const GPUSettingsGRP g, const GPUSettingsProcessing p=nullptr, const GPUSettingsRecDynamic *d=nullptr)

void	UpdateDynamicSettings (const GPUSettingsRecDynamic *d)

void	SetOutputControl (const GPUOutputControl &v)

void	SetOutputControl (void *ptr, size_t size)

void	SetInputControl (void *ptr, size_t size)

GPUOutputControl &	OutputControl ()

int32_t	NStreams () const

const void *	DeviceMemoryBase () const

RecoStepField	GetRecoSteps () const

RecoStepField	GetRecoStepsGPU () const

InOutTypeField	GetRecoStepsInputs () const

InOutTypeField	GetRecoStepsOutputs () const

int32_t	getRecoStepNum (RecoStep step, bool validCheck=true)

int32_t	getGeneralStepNum (GeneralStep step, bool validCheck=true)

void	setErrorCodeOutput (std::vector< std::array< uint32_t, 4 > > *v)

std::vector< std::array< uint32_t, 4 > > *	getErrorCodeOutput ()

template<class T >
void	RegisterGPUProcessor (T *proc, bool deviceSlave)

template<class T >
void	SetupGPUProcessor (T *proc, bool allocate)

void	RegisterGPUDeviceProcessor (GPUProcessor proc, GPUProcessor slaveProcessor)

void	ConstructGPUProcessor (GPUProcessor *proc)

virtual void	PrintKernelOccupancies ()

double	GetStatKernelTime ()

double	GetStatWallTime ()

int32_t	GetMaxBackendThreads () const

Static Public Member Functions
template<class T , int32_t I>
static constexpr const char *	GetKernelName ()

Static Public Member Functions inherited from o2::gpu::GPUReconstruction
static DeviceType	GetDeviceType (const char *type)

static uint32_t	getNIOTypeMultiplicity (InOutPointerType type)

static GPUReconstruction *	CreateInstance (const GPUSettingsDeviceBackend &cfg)

static GPUReconstruction *	CreateInstance (DeviceType type=DeviceType::CPU, bool forceType=true, GPUReconstruction *master=nullptr)

static GPUReconstruction *	CreateInstance (int32_t type, bool forceType, GPUReconstruction *master=nullptr)

static GPUReconstruction *	CreateInstance (const char type, bool forceType, GPUReconstruction master=nullptr)

static bool	CheckInstanceAvailable (DeviceType type, bool verbose)

static int32_t	getHostThreadIndex ()

Protected Types
using	deviceEvent = gpu_reconstruction_kernels::deviceEvent

Protected Member Functions
	GPUReconstructionProcessing (const GPUSettingsDeviceBackend &cfg)

template<class T , int32_t I = 0>
HighResTimer &	getKernelTimer (RecoStep step, int32_t num=0, size_t addMemorySize=0, bool increment=true)

template<class T , int32_t J = -1>
HighResTimer &	getTimer (const char *name, int32_t num=-1)

Protected Member Functions inherited from o2::gpu::GPUReconstruction
void	AllocateRegisteredMemoryInternal (GPUMemoryResource res, GPUOutputControl control, GPUReconstruction *recPool)

void	FreeRegisteredMemory (GPUMemoryResource *res)

	GPUReconstruction (const GPUSettingsDeviceBackend &cfg)

int32_t	InitPhaseBeforeDevice ()

virtual void	UpdateAutomaticProcessingSettings ()

virtual int32_t	InitDevice ()=0

int32_t	InitPhasePermanentMemory ()

int32_t	InitPhaseAfterDevice ()

void	WriteConstantParams ()

virtual int32_t	ExitDevice ()=0

virtual size_t	WriteToConstantMemory (size_t offset, const void src, size_t size, int32_t stream=-1, gpu_reconstruction_kernels::deviceEvent ev=nullptr)=0

void	UpdateMaxMemoryUsed ()

int32_t	EnqueuePipeline (bool terminate=false)

GPUChain *	GetNextChainInQueue ()

virtual int32_t	registerMemoryForGPU_internal (const void *ptr, size_t size)=0

virtual int32_t	unregisterMemoryForGPU_internal (const void *ptr)=0

size_t	AllocateRegisteredMemoryHelper (GPUMemoryResource res, void &ptr, void &memorypool, void memorybase, size_t memorysize, void (GPUMemoryResource::SetPointers)(void ), void &memorypoolend, const char *device)

size_t	AllocateRegisteredPermanentMemory ()

template<class T , class S >
uint32_t	DumpData (FILE fp, const T const entries, const S num, InOutPointerType type)

template<class T , class S >
size_t	ReadData (FILE fp, const T entries, S num, std::unique_ptr< T[]> mem, InOutPointerType type, T *nonConstPtrs=nullptr)

template<class T >
T *	AllocateIOMemoryHelper (size_t n, const T *&ptr, std::unique_ptr< T[]> &u)

template<class T >
void	DumpFlatObjectToFile (const T obj, const char file)

template<class T >
std::unique_ptr< T >	ReadFlatObjectFromFile (const char *file)

template<class T >
void	DumpStructToFile (const T obj, const char file)

template<class T >
std::unique_ptr< T >	ReadStructFromFile (const char *file)

template<class T >
int32_t	ReadStructFromFile (const char file, T obj)

virtual RecoStepField	AvailableGPURecoSteps ()

virtual bool	CanQueryMaxMemory ()

GPUConstantMem *	processors ()

const GPUConstantMem *	processors () const

GPUParam &	param ()

Protected Attributes
int32_t	mActiveHostKernelThreads = 0

uint32_t	mNActiveThreadsOuterLoop = 1

std::vector< std::vector< deviceEvent > >	mEvents

HighResTimer	mTimersGeneralSteps [GPUDataTypes::N_GENERAL_STEPS]

std::vector< std::unique_ptr< timerMeta > >	mTimers

RecoStepTimerMeta	mTimersRecoSteps [GPUDataTypes::N_RECO_STEPS]

HighResTimer	mTimerTotal

Protected Attributes inherited from o2::gpu::GPUReconstruction
std::shared_ptr< LibraryLoader >	mMyLib = nullptr

std::vector< GPUMemoryResource >	mMemoryResources

std::vector< std::unique_ptr< char[]> >	mUnmanagedChunks

std::vector< std::unique_ptr< char[]> >	mVolatileChunks

std::vector< std::unique_ptr< GPUChain > >	mChains

std::unique_ptr< GPUConstantMem >	mHostConstantMem

GPUConstantMem *	mDeviceConstantMem = nullptr

GPUSettingsGRP	mGRPSettings

GPUSettingsDeviceBackend	mDeviceBackendSettings

GPUSettingsProcessing	mProcessingSettings

GPUOutputControl	mOutputControl

GPUOutputControl	mInputControl

std::unique_ptr< GPUMemorySizeScalers >	mMemoryScalers

GPURecoStepConfiguration	mRecoSteps

std::string	mDeviceName = "CPU"

void *	mHostMemoryBase = nullptr

void *	mHostMemoryPermanent = nullptr

void *	mHostMemoryPool = nullptr

void *	mHostMemoryPoolEnd = nullptr

void *	mHostMemoryPoolBlocked = nullptr

size_t	mHostMemorySize = 0

size_t	mHostMemoryUsedMax = 0

void *	mDeviceMemoryBase = nullptr

void *	mDeviceMemoryPermanent = nullptr

void *	mDeviceMemoryPool = nullptr

void *	mDeviceMemoryPoolEnd = nullptr

void *	mDeviceMemoryPoolBlocked = nullptr

size_t	mDeviceMemorySize = 0

void *	mVolatileMemoryStart = nullptr

size_t	mDeviceMemoryUsedMax = 0

std::unordered_set< const void * >	mRegisteredMemoryPtrs

GPUReconstruction *	mMaster = nullptr

std::vector< GPUReconstruction * >	mSlaves

bool	mInitialized = false

uint32_t	mStatNEvents = 0

uint32_t	mNEventsProcessed = 0

double	mStatKernelTime = 0.

double	mStatWallTime = 0.

double	mStatCPUTime = 0.

std::shared_ptr< GPUROOTDumpCore >	mROOTDump

std::vector< std::array< uint32_t, 4 > > *	mOutputErrorCodes = nullptr

int32_t	mMaxBackendThreads = 0

int32_t	mGPUStuck = 0

int32_t	mNStreams = 1

int32_t	mMaxHostThreads = 0

std::vector< ProcessorData >	mProcessors

std::unordered_map< GPUMemoryReuse::ID, MemoryReuseMeta >	mMemoryReuse1to1

std::vector< std::tuple< void , void , size_t, uint64_t > >	mNonPersistentMemoryStack

std::vector< GPUMemoryResource * >	mNonPersistentIndividualAllocations

std::unique_ptr< GPUReconstructionPipelineContext >	mPipelineContext

Additional Inherited Members
Public Types inherited from o2::gpu::GPUReconstruction
enum	InOutPointerType : uint32_t { CLUSTER_DATA = 0 , SECTOR_OUT_TRACK = 1 , SECTOR_OUT_CLUSTER = 2 , MC_LABEL_TPC = 3 , MC_INFO_TPC = 4 , MERGED_TRACK = 5 , MERGED_TRACK_HIT = 6 , TRD_TRACK = 7 , TRD_TRACKLET = 8 , RAW_CLUSTERS = 9 , CLUSTERS_NATIVE = 10 , TRD_TRACKLET_MC = 11 , TPC_COMPRESSED_CL = 12 , TPC_DIGIT = 13 , TPC_ZS = 14 , CLUSTER_NATIVE_MC = 15 , TPC_DIGIT_MC = 16 , TRD_SPACEPOINT = 17 , TRD_TRIGGERRECORDS = 18 , TF_SETTINGS = 19 }

enum class	krnlDeviceType : int32_t { CPU = 0 , Device = 1 , Auto = -1 }

using	GeometryType = GPUDataTypes::GeometryType

using	DeviceType = GPUDataTypes::DeviceType

using	RecoStep = GPUDataTypes::RecoStep

using	GeneralStep = GPUDataTypes::GeneralStep

using	RecoStepField = GPUDataTypes::RecoStepField

using	InOutTypeField = GPUDataTypes::InOutTypeField

Public Attributes inherited from o2::gpu::GPUReconstruction
std::shared_ptr< GPUReconstructionThreading >	mThreading

Static Public Attributes inherited from o2::gpu::GPUReconstruction
static constexpr uint32_t	NSECTORS = GPUCA_NSECTORS

static constexpr const char *const	GEOMETRY_TYPE_NAMES [] = {"INVALID", "ALIROOT", "O2"}

static constexpr GeometryType	geometryType = GeometryType::ALIROOT

static constexpr const char *const	IOTYPENAMES []

Static Protected Member Functions inherited from o2::gpu::GPUReconstruction
static std::shared_ptr< LibraryLoader > *	GetLibraryInstance (DeviceType type, bool verbose)

static GPUReconstruction *	GPUReconstruction_Create_CPU (const GPUSettingsDeviceBackend &cfg)

Static Protected Attributes inherited from o2::gpu::GPUReconstruction
static std::shared_ptr< LibraryLoader >	sLibCUDA

static std::shared_ptr< LibraryLoader >	sLibHIP

static std::shared_ptr< LibraryLoader >	sLibOCL

Detailed Description

Definition at line 62 of file GPUReconstructionProcessing.h.

Member Typedef Documentation

◆ deviceEvent

using o2::gpu::GPUReconstructionProcessing::deviceEvent = gpu_reconstruction_kernels::deviceEvent

protected

Definition at line 101 of file GPUReconstructionProcessing.h.

Constructor & Destructor Documentation

◆ ~GPUReconstructionProcessing()

o2::gpu::GPUReconstructionProcessing::~GPUReconstructionProcessing ( )

overridedefault

◆ GPUReconstructionProcessing()

o2::gpu::GPUReconstructionProcessing::GPUReconstructionProcessing ( const GPUSettingsDeviceBackend & cfg )

inlineprotected

Definition at line 100 of file GPUReconstructionProcessing.h.

Member Function Documentation

◆ AddGPUEvents()

template<class T >

void o2::gpu::GPUReconstructionProcessing::AddGPUEvents ( T *& events )

inline

Definition at line 136 of file GPUReconstructionProcessing.h.

◆ getGeneralStepTimer()

HighResTimer & o2::gpu::GPUReconstructionProcessing::getGeneralStepTimer ( GeneralStep step )

inline

Definition at line 81 of file GPUReconstructionProcessing.h.

◆ GetKernelName()

template<class T , int32_t I>

static constexpr const char * o2::gpu::GPUReconstructionProcessing::GetKernelName ( )

staticconstexpr

◆ getKernelTimer()

template<class T , int32_t I>

HighResTimer & o2::gpu::GPUReconstructionProcessing::getKernelTimer	(	RecoStep	step,
		int32_t	num = `0`,
		size_t	addMemorySize = `0`,
		bool	increment = `true`
	)

protected

Definition at line 143 of file GPUReconstructionProcessing.h.

◆ getNActiveThreadsOuterLoop()

uint32_t o2::gpu::GPUReconstructionProcessing::getNActiveThreadsOuterLoop ( ) const

inline

Definition at line 69 of file GPUReconstructionProcessing.h.

◆ getNKernelHostThreads()

int32_t GPUReconstructionProcessing::getNKernelHostThreads ( bool splitCores )

Definition at line 20 of file GPUReconstructionProcessing.cxx.

◆ getRecoStepTimer()

auto & o2::gpu::GPUReconstructionProcessing::getRecoStepTimer ( RecoStep step )

inline

Definition at line 80 of file GPUReconstructionProcessing.h.

◆ GetThreadContext()

std::unique_ptr< gpu_reconstruction_kernels::threadContext > GPUReconstructionProcessing::GetThreadContext ( )

overridevirtual

Implements o2::gpu::GPUReconstruction.

Reimplemented in o2::gpu::GPUReconstructionCUDA.

Definition at line 113 of file GPUReconstructionProcessing.cxx.

◆ getTimer()

template<class T , int32_t J>

HighResTimer & o2::gpu::GPUReconstructionProcessing::getTimer	(	const char *	name,
		int32_t	num = `-1`
	)

protected

Definition at line 160 of file GPUReconstructionProcessing.h.

◆ runParallelOuterLoop()

void GPUReconstructionProcessing::runParallelOuterLoop	(	bool	doGPU,
		uint32_t	nThreads,
		std::function< void(uint32_t)>	lambda
	)

Definition at line 46 of file GPUReconstructionProcessing.cxx.

◆ SetAndGetNActiveThreadsOuterLoop()

uint32_t GPUReconstructionProcessing::SetAndGetNActiveThreadsOuterLoop	(	bool	condition,
		uint32_t	max
	)

Definition at line 100 of file GPUReconstructionProcessing.cxx.

◆ SetNActiveThreads()

void GPUReconstructionProcessing::SetNActiveThreads ( int32_t n )

Definition at line 37 of file GPUReconstructionProcessing.cxx.

◆ SetNActiveThreadsOuterLoop()

void o2::gpu::GPUReconstructionProcessing::SetNActiveThreadsOuterLoop ( uint32_t f )

inline

Definition at line 70 of file GPUReconstructionProcessing.h.

Member Data Documentation

◆ mActiveHostKernelThreads

int32_t o2::gpu::GPUReconstructionProcessing::mActiveHostKernelThreads = 0

protected

Definition at line 103 of file GPUReconstructionProcessing.h.

◆ mEvents

std::vector<std::vector<deviceEvent> > o2::gpu::GPUReconstructionProcessing::mEvents

protected

Definition at line 106 of file GPUReconstructionProcessing.h.

◆ mNActiveThreadsOuterLoop

uint32_t o2::gpu::GPUReconstructionProcessing::mNActiveThreadsOuterLoop = 1

protected

Definition at line 104 of file GPUReconstructionProcessing.h.

◆ mTimers

std::vector<std::unique_ptr<timerMeta> > o2::gpu::GPUReconstructionProcessing::mTimers

protected

Definition at line 121 of file GPUReconstructionProcessing.h.

◆ mTimersGeneralSteps

HighResTimer o2::gpu::GPUReconstructionProcessing::mTimersGeneralSteps[GPUDataTypes::N_GENERAL_STEPS]

protected

Definition at line 119 of file GPUReconstructionProcessing.h.

◆ mTimersRecoSteps

RecoStepTimerMeta o2::gpu::GPUReconstructionProcessing::mTimersRecoSteps[GPUDataTypes::N_RECO_STEPS]

protected

Definition at line 122 of file GPUReconstructionProcessing.h.

◆ mTimerTotal

HighResTimer o2::gpu::GPUReconstructionProcessing::mTimerTotal

protected

Definition at line 123 of file GPUReconstructionProcessing.h.

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

/home/runner/work/AliceO2/AliceO2/GPU/GPUTracking/Base/GPUReconstructionProcessing.h
/home/runner/work/AliceO2/AliceO2/GPU/GPUTracking/Base/GPUReconstructionProcessing.cxx

Classes

Public Member Functions

Static Public Member Functions

Protected Types

Protected Member Functions

Protected Attributes

Additional Inherited Members

Detailed Description

Member Typedef Documentation

◆ deviceEvent

Constructor & Destructor Documentation

◆ ~GPUReconstructionProcessing()

◆ GPUReconstructionProcessing()

Member Function Documentation

◆ AddGPUEvents()

◆ getGeneralStepTimer()

◆ GetKernelName()

◆ getKernelTimer()

◆ getNActiveThreadsOuterLoop()

◆ getNKernelHostThreads()

◆ getRecoStepTimer()

◆ GetThreadContext()

◆ getTimer()

◆ runParallelOuterLoop()

◆ SetAndGetNActiveThreadsOuterLoop()

◆ SetNActiveThreads()

◆ SetNActiveThreadsOuterLoop()

Member Data Documentation

◆ mActiveHostKernelThreads

◆ mEvents

◆ mNActiveThreadsOuterLoop

◆ mTimers

◆ mTimersGeneralSteps

◆ mTimersRecoSteps

◆ mTimerTotal