#include <GPUReconstructionCUDA.h>

Inherits o2::gpu::GPUReconstructionProcessing::KernelInterface< GPUReconstructionCUDA, GPUReconstructionDeviceBase >.

Classes
struct	kernelBackendMacro

Public Member Functions
	GPUReconstructionCUDA (const GPUSettingsDeviceBackend &cfg)

	~GPUReconstructionCUDA () override

void	PrintKernelOccupancies () override

virtual int32_t	GPUChkErrInternal (const int64_t error, const char *file, int32_t line) const override

template<class T , int32_t I = 0, typename... Args>
void	runKernelBackend (const krnlSetupTime &_xyz, const Args &... args)

template<class T , int32_t I = 0, typename... Args>
void	runKernelBackendTimed (const krnlSetupTime &_xyz, const Args &... args)

Public Member Functions inherited from o2::gpu::GPUReconstructionProcessing::KernelInterface< GPUReconstructionCUDA, GPUReconstructionDeviceBase >
	KernelInterface (const Args &... args)

Public Member Functions inherited from o2::gpu::GPUReconstructionDeviceBase
	~GPUReconstructionDeviceBase () override

const GPUParam *	DeviceParam () const

Public Member Functions inherited from o2::gpu::GPUReconstructionCPU
	~GPUReconstructionCPU () override

template<class S , int32_t I = 0>
krnlProperties	getKernelProperties (int gpu=-1)

template<class T , int32_t I = 0, typename... Args>
void	runKernelBackend (const krnlSetupTime &_xyz, const Args &... args)

int32_t	GPUStuck ()

void	ResetDeviceProcessorTypes ()

int32_t	RunChains () override

void	UpdateParamOccupancyMap (const uint32_t mapHost, const uint32_t mapGPU, uint32_t occupancyTotal, int32_t stream=-1)

template<>
void	runKernelBackend (const krnlSetupTime &_xyz, void *const &ptr, uint64_t const &size)

template<class S , int32_t I>
GPUReconstructionProcessing::krnlProperties	getKernelProperties (int gpu)

Public Member Functions inherited from o2::gpu::GPUReconstructionProcessing::KernelInterface< GPUReconstructionCPU, GPUReconstructionProcessing >
	KernelInterface (const Args &... args)

Public Member Functions inherited from o2::gpu::GPUReconstructionProcessing
	~GPUReconstructionProcessing () override

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)

const std::string &	GetKernelName (int32_t i) const

auto &	getRecoStepTimer (RecoStep step)

HighResTimer &	getGeneralStepTimer (GeneralStep step)

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

const GPUDefParameters &	getGPUParameters (bool doGPU) const 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 ()

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)

int32_t	GPUChkErrA (const int64_t error, const char *file, int32_t line, bool failOnError)

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 *	AllocateDirectMemory (size_t size, int32_t type)

void *	AllocateVolatileDeviceMemory (size_t size)

void *	AllocateVolatileMemory (size_t size, bool device)

void	MakeFutureDeviceMemoryAllocationsVolatile ()

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

void	FreeRegisteredMemory (int16_t res)

void	ClearAllocatedMemory (bool clearOutputs=true)

void	ReturnVolatileDeviceMemory ()

void	ReturnVolatileMemory ()

ThrustVolatileAllocator	getThrustVolatileDeviceAllocator ()

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 ()

bool	slavesExist ()

DeviceType	GetDeviceType () const

bool	IsGPU () const

const GPUParam &	GetParam () const

const GPUConstantMem &	GetConstantMem () const

const GPUTrackingInOutPointers	GetIOPtrs () const

const GPUSettingsGRP &	GetGRPSettings () const

const GPUSettingsDeviceBackend &	GetDeviceBackendSettings () const

const GPUSettingsProcessing &	GetProcessingSettings () const

const GPUCalibObjectsConst &	GetCalib () 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 ()

uint32_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)

double	GetStatKernelTime ()

double	GetStatWallTime ()

int32_t	GetMaxBackendThreads () const

Protected Member Functions
int32_t	InitDevice_Runtime () override

int32_t	ExitDevice_Runtime () override

std::unique_ptr< threadContext >	GetThreadContext () override

void	SynchronizeGPU () override

int32_t	GPUDebug (const char *state="UNKNOWN", int32_t stream=-1, bool force=false) override

void	SynchronizeStream (int32_t stream) override

void	SynchronizeEvents (deviceEvent *evList, int32_t nEvents=1) override

void	StreamWaitForEvents (int32_t stream, deviceEvent *evList, int32_t nEvents=1) override

bool	IsEventDone (deviceEvent *evList, int32_t nEvents=1) override

int32_t	registerMemoryForGPU_internal (const void *ptr, size_t size) override

int32_t	unregisterMemoryForGPU_internal (const void *ptr) override

size_t	WriteToConstantMemory (size_t offset, const void src, size_t size, int32_t stream=-1, deviceEvent ev=nullptr) override

size_t	GPUMemCpy (void dst, const void src, size_t size, int32_t stream, int32_t toGPU, deviceEvent ev=nullptr, deviceEvent evList=nullptr, int32_t nEvents=1) override

void	ReleaseEvent (deviceEvent ev) override

void	RecordMarker (deviceEvent *ev, int32_t stream) override

void	SetONNXGPUStream (Ort::SessionOptions &session_options, int32_t stream, int32_t *deviceId) override

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

bool	CanQueryMaxMemory () override

void	startGPUProfiling () override

void	endGPUProfiling () override

Protected Member Functions inherited from o2::gpu::GPUReconstructionProcessing::KernelInterface< GPUReconstructionCUDA, GPUReconstructionDeviceBase >
virtual void	runKernelVirtual (const int num, const void *args)

Protected Member Functions inherited from o2::gpu::GPUReconstructionDeviceBase
	GPUReconstructionDeviceBase (const GPUSettingsDeviceBackend &cfg, size_t sizeCheck)

int32_t	InitDevice () override

int32_t	ExitDevice () override

int32_t	registerMemoryForGPU_internal (const void *ptr, size_t size) override

int32_t	unregisterMemoryForGPU_internal (const void *ptr) override

void	unregisterRemainingRegisteredMemory ()

size_t	TransferMemoryInternal (GPUMemoryResource res, int32_t stream, deviceEvent ev, deviceEvent evList, int32_t nEvents, bool toGPU, const void src, void *dst) override

size_t	GPUMemCpyAlways (bool onGpu, void dst, const void src, size_t size, int32_t stream, int32_t toGPU, deviceEvent ev=nullptr, deviceEvent evList=nullptr, int32_t nEvents=1) override

int32_t	GetGlobalLock (void *&pLock)

void	ReleaseGlobalLock (void *sem)

void	runConstantRegistrators ()

Protected Member Functions inherited from o2::gpu::GPUReconstructionCPU
	GPUReconstructionCPU (const GPUSettingsDeviceBackend &cfg)

size_t	TransferMemoryResourceToGPU (GPUMemoryResource res, int32_t stream=-1, deviceEvent ev=nullptr, deviceEvent *evList=nullptr, int32_t nEvents=1)

size_t	TransferMemoryResourceToHost (GPUMemoryResource res, int32_t stream=-1, deviceEvent ev=nullptr, deviceEvent *evList=nullptr, int32_t nEvents=1)

size_t	TransferMemoryResourcesToGPU (GPUProcessor *proc, int32_t stream=-1, bool all=false)

size_t	TransferMemoryResourcesToHost (GPUProcessor *proc, int32_t stream=-1, bool all=false)

size_t	TransferMemoryResourceLinkToGPU (int16_t res, int32_t stream=-1, deviceEvent ev=nullptr, deviceEvent evList=nullptr, int32_t nEvents=1)

size_t	TransferMemoryResourceLinkToHost (int16_t res, int32_t stream=-1, deviceEvent ev=nullptr, deviceEvent evList=nullptr, int32_t nEvents=1)

int32_t	GetThread ()

virtual int32_t	DoStuckProtection (int32_t stream, deviceEvent event)

Protected Member Functions inherited from o2::gpu::GPUReconstructionProcessing
	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 ()

int32_t	InitPhasePermanentMemory ()

int32_t	InitPhaseAfterDevice ()

void	WriteConstantParams ()

void	UpdateMaxMemoryUsed ()

int32_t	EnqueuePipeline (bool terminate=false)

GPUChain *	GetNextChainInQueue ()

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)

int16_t	RegisterMemoryAllocationHelper (GPUProcessor proc, void (GPUProcessor::setPtr)(void ), int32_t type, const char *name, const GPUMemoryReuse &re)

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 ()

GPUConstantMem *	processors ()

const GPUConstantMem *	processors () const

GPUParam &	param ()

Protected Attributes
template<class T , class S >
GPUReconstructionCUDAInternals *	mInternals

Protected Attributes inherited from o2::gpu::GPUReconstructionDeviceBase
int32_t	mDeviceId = -1

DebugEvents *	mDebugEvents = nullptr

std::vector< void * >	mDeviceConstantMemList

Protected Attributes inherited from o2::gpu::GPUReconstructionCPU
GPUProcessorProcessors	mProcShadow

GPUConstantMem *&	mProcessorsShadow = mProcShadow.mProcessorsProc

uint32_t	mBlockCount = 1

uint32_t	mThreadCount = 1

uint32_t	mWarpSize = 1

Protected Attributes inherited from o2::gpu::GPUReconstructionProcessing
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

GPUDefParameters *	mParCPU = nullptr

GPUDefParameters *	mParDevice = nullptr

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

std::vector< GPUMemoryResource >	mMemoryResources

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

std::unique_ptr< GPUConstantMem >	mHostConstantMem

GPUConstantMem *	mDeviceConstantMem = nullptr

std::unique_ptr< GPUSettingsGRP >	mGRPSettings

std::unique_ptr< GPUSettingsDeviceBackend >	mDeviceBackendSettings

std::unique_ptr< 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

size_t	mDeviceMemoryUsedMax = 0

void *	mVolatileMemoryStart = nullptr

bool	mDeviceMemoryAsVolatile = false

std::unordered_set< const void * >	mRegisteredMemoryPtrs

GPUReconstruction *	mMaster = nullptr

std::vector< GPUReconstruction * >	mSlaves

bool	mInitialized = false

bool	mInErrorHandling = 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, size_t, uint64_t > >	mNonPersistentMemoryStack

std::vector< GPUMemoryResource * >	mNonPersistentIndividualAllocations

std::vector< std::unique_ptr< char[], alignedDeleter > >	mNonPersistentIndividualDirectAllocations

std::vector< std::unique_ptr< char[], alignedDeleter > >	mDirectMemoryChunks

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

std::unique_ptr< GPUReconstructionPipelineContext >	mPipelineContext

Additional Inherited Members
Public Types inherited from o2::gpu::GPUReconstructionProcessing
using	deviceEvent = gpu_reconstruction_kernels::deviceEvent

using	threadContext = gpu_reconstruction_kernels::threadContext

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

Static Public Member Functions inherited from o2::gpu::GPUReconstructionProcessing
template<class T , int32_t I>
static const char *	GetKernelName ()

template<class T , int32_t I = 0>
static uint32_t	GetKernelNum ()

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 ()

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

Static Public Attributes inherited from o2::gpu::GPUReconstructionCPU
static constexpr krnlRunRange	krnlRunRangeNone {0}

static constexpr krnlEvent	krnlEventNone = krnlEvent{nullptr, nullptr, 0}

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::GPUReconstructionDeviceBase
static std::vector< void ()()> &	getDeviceConstantMemRegistratorsVector ()

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::GPUReconstructionProcessing
static const std::vector< std::string >	mKernelNames

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 38 of file GPUReconstructionCUDA.h.

Constructor & Destructor Documentation

◆ GPUReconstructionCUDA()

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

◆ ~GPUReconstructionCUDA()

o2::gpu::GPUReconstructionCUDA::~GPUReconstructionCUDA ( )

override

Member Function Documentation

◆ CanQueryMaxMemory()

bool o2::gpu::GPUReconstructionCUDA::CanQueryMaxMemory ( )

inlineoverrideprotectedvirtual

Reimplemented from o2::gpu::GPUReconstruction.

Definition at line 82 of file GPUReconstructionCUDA.h.

◆ endGPUProfiling()

void o2::gpu::GPUReconstructionCUDA::endGPUProfiling ( )

overrideprotectedvirtual

Reimplemented from o2::gpu::GPUReconstruction.

◆ ExitDevice_Runtime()

int32_t o2::gpu::GPUReconstructionCUDA::ExitDevice_Runtime ( )

overrideprotectedvirtual

Implements o2::gpu::GPUReconstructionDeviceBase.

◆ GetITSTraits()

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

overrideprotectedvirtual

Reimplemented from o2::gpu::GPUReconstruction.

◆ GetThreadContext()

std::unique_ptr< threadContext > o2::gpu::GPUReconstructionCUDA::GetThreadContext ( )

overrideprotectedvirtual

Reimplemented from o2::gpu::GPUReconstructionProcessing.

◆ GPUChkErrInternal()

virtual int32_t o2::gpu::GPUReconstructionCUDA::GPUChkErrInternal	(	const int64_t	error,
		const char *	file,
		int32_t	line
	)		const

overridevirtual

Implements o2::gpu::GPUReconstructionDeviceBase.

◆ GPUDebug()

int32_t o2::gpu::GPUReconstructionCUDA::GPUDebug	(	const char *	state = `"UNKNOWN"`,
		int32_t	stream = `-1`,
		bool	force = `false`
	)

overrideprotectedvirtual

Implements o2::gpu::GPUReconstructionDeviceBase.

◆ GPUMemCpy()

size_t o2::gpu::GPUReconstructionCUDA::GPUMemCpy	(	void *	dst,
		const void *	src,
		size_t	size,
		int32_t	stream,
		int32_t	toGPU,
		deviceEvent *	ev = `nullptr`,
		deviceEvent *	evList = `nullptr`,
		int32_t	nEvents = `1`
	)

overrideprotectedvirtual

Implements o2::gpu::GPUReconstructionDeviceBase.

◆ InitDevice_Runtime()

int32_t o2::gpu::GPUReconstructionCUDA::InitDevice_Runtime ( )

overrideprotectedvirtual

Implements o2::gpu::GPUReconstructionDeviceBase.

◆ IsEventDone()

bool o2::gpu::GPUReconstructionCUDA::IsEventDone	(	deviceEvent *	evList,
		int32_t	nEvents = `1`
	)

overrideprotectedvirtual

Reimplemented from o2::gpu::GPUReconstructionCPU.

◆ PrintKernelOccupancies()

void o2::gpu::GPUReconstructionCUDA::PrintKernelOccupancies ( )

overridevirtual

Reimplemented from o2::gpu::GPUReconstruction.

◆ RecordMarker()

void o2::gpu::GPUReconstructionCUDA::RecordMarker	(	deviceEvent *	ev,
		int32_t	stream
	)

overrideprotectedvirtual

Reimplemented from o2::gpu::GPUReconstructionCPU.

◆ registerMemoryForGPU_internal()

int32_t o2::gpu::GPUReconstructionCUDA::registerMemoryForGPU_internal	(	const void *	ptr,
		size_t	size
	)

overrideprotectedvirtual

Reimplemented from o2::gpu::GPUReconstructionCPU.

◆ ReleaseEvent()

void o2::gpu::GPUReconstructionCUDA::ReleaseEvent ( deviceEvent ev )

overrideprotectedvirtual

Reimplemented from o2::gpu::GPUReconstructionCPU.

◆ runKernelBackend()

template<class T , int32_t I = 0, typename... Args>

void o2::gpu::GPUReconstructionCUDA::runKernelBackend	(	const krnlSetupTime &	_xyz,
		const Args &...	args
	)

◆ runKernelBackendTimed()

template<class T , int32_t I = 0, typename... Args>

void o2::gpu::GPUReconstructionCUDA::runKernelBackendTimed	(	const krnlSetupTime &	_xyz,
		const Args &...	args
	)

◆ SetONNXGPUStream()

void o2::gpu::GPUReconstructionCUDA::SetONNXGPUStream	(	Ort::SessionOptions &	session_options,
		int32_t	stream,
		int32_t *	deviceId
	)

overrideprotectedvirtual

Reimplemented from o2::gpu::GPUReconstructionCPU.

◆ startGPUProfiling()

void o2::gpu::GPUReconstructionCUDA::startGPUProfiling ( )

overrideprotectedvirtual

Reimplemented from o2::gpu::GPUReconstruction.

◆ StreamWaitForEvents()

void o2::gpu::GPUReconstructionCUDA::StreamWaitForEvents	(	int32_t	stream,
		deviceEvent *	evList,
		int32_t	nEvents = `1`
	)

overrideprotectedvirtual

Reimplemented from o2::gpu::GPUReconstructionCPU.

◆ SynchronizeEvents()

void o2::gpu::GPUReconstructionCUDA::SynchronizeEvents	(	deviceEvent *	evList,
		int32_t	nEvents = `1`
	)

overrideprotectedvirtual

Reimplemented from o2::gpu::GPUReconstructionCPU.

◆ SynchronizeGPU()

void o2::gpu::GPUReconstructionCUDA::SynchronizeGPU ( )

overrideprotectedvirtual

Reimplemented from o2::gpu::GPUReconstructionCPU.

◆ SynchronizeStream()

void o2::gpu::GPUReconstructionCUDA::SynchronizeStream ( int32_t stream )

overrideprotectedvirtual

Reimplemented from o2::gpu::GPUReconstructionCPU.

◆ unregisterMemoryForGPU_internal()

int32_t o2::gpu::GPUReconstructionCUDA::unregisterMemoryForGPU_internal ( const void * ptr )

overrideprotectedvirtual

Reimplemented from o2::gpu::GPUReconstructionCPU.

◆ WriteToConstantMemory()

size_t o2::gpu::GPUReconstructionCUDA::WriteToConstantMemory	(	size_t	offset,
		const void *	src,
		size_t	size,
		int32_t	stream = `-1`,
		deviceEvent *	ev = `nullptr`
	)

overrideprotectedvirtual

Implements o2::gpu::GPUReconstructionDeviceBase.

Member Data Documentation

◆ mInternals

template<class T , class S >

GPUReconstructionCUDAInternals* o2::gpu::GPUReconstructionCUDA::mInternals

protected

Definition at line 58 of file GPUReconstructionCUDA.h.

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

/home/runner/work/AliceO2/AliceO2/GPU/GPUTracking/Base/cuda/GPUReconstructionCUDA.h
/home/runner/work/AliceO2/AliceO2/GPU/GPUTracking/Base/cuda/GPUReconstructionCUDAGenRTC.cxx

Classes

Public Member Functions

Protected Member Functions

Protected Attributes

Additional Inherited Members

Detailed Description

Constructor & Destructor Documentation

◆ GPUReconstructionCUDA()

◆ ~GPUReconstructionCUDA()

Member Function Documentation

◆ CanQueryMaxMemory()

◆ endGPUProfiling()

◆ ExitDevice_Runtime()

◆ GetITSTraits()

◆ GetThreadContext()

◆ GPUChkErrInternal()

◆ GPUDebug()

◆ GPUMemCpy()

◆ InitDevice_Runtime()

◆ IsEventDone()

◆ PrintKernelOccupancies()

◆ RecordMarker()

◆ registerMemoryForGPU_internal()

◆ ReleaseEvent()

◆ runKernelBackend()

◆ runKernelBackendTimed()

◆ SetONNXGPUStream()

◆ startGPUProfiling()

◆ StreamWaitForEvents()

◆ SynchronizeEvents()

◆ SynchronizeGPU()

◆ SynchronizeStream()

◆ unregisterMemoryForGPU_internal()

◆ WriteToConstantMemory()

Member Data Documentation

◆ mInternals