21#include "GPUChainTrackingGetters.inc"
45#include "GPUDefParametersRuntime.h"
64GPUChainTracking::GPUChainTracking(
GPUReconstruction*
rec, uint32_t maxTPCHits, uint32_t maxTRDTracklets) :
GPUChain(
rec), mIOPtrs(processors()->ioPtrs), mInputsHost(new
GPUTrackingInputProvider), mInputsShadow(new
GPUTrackingInputProvider), mClusterNativeAccess(new
ClusterNativeAccess), mTriggerBuffer(new
GPUTriggerOutputs), mMaxTPCHits(maxTPCHits), mMaxTRDTracklets(maxTRDTracklets), mDebugFile(new
std::ostream(nullptr))
118#ifdef GPUCA_KERNEL_DEBUGGER_OUTPUT
163#ifdef GPUCA_KERNEL_DEBUGGER_OUTPUT
170 gpuMem = constants::GPU_DEFAULT_MEMORY_SIZE;
171 pageLockedHostMem = constants::GPU_DEFAULT_HOST_MEMORY_SIZE;
177 GPUError(
"Invalid Reconstruction Step Setting: dEdx requires TPC Merger to be active");
181 GPUError(
"Invalid GPU Reconstruction Step Setting: dEdx requires TPC Merger to be active");
185 GPUError(
"Invalid Reconstruction Step Setting: Tracking requires TPC Conversion to be active");
189 GPUError(
"Invalid input, TPC Clusterizer needs TPC raw input");
193 GPUError(
"Invalid input / output / step, merger cannot read/store sectors tracks and needs TPC conversion");
198 GPUError(
"Invalid Inputs for track merging, TPC Clusters required");
203 GPUError(
"Can not run TPC GPU Cluster Finding with Run 2 Data");
208 GPUError(
"Missing input for TPC Cluster conversion / sector tracking / compression / dEdx: TPC Clusters required");
212 GPUError(
"Input for TPC merger missing");
216 GPUError(
"Input for TPC compressor missing");
220 GPUError(
"Input for TRD Tracker missing");
224 GPUError(
"TRD Reco Parameters are missing");
228 GPUError(
"TPC Raw / TPC Clusters / TRD Tracklets cannot be output");
232 GPUError(
"No TPC Merged Track Output available");
236 GPUError(
"No TPC Compression Output available");
240 GPUError(
"No TRD Tracker Output available");
244 GPUError(
"Cannot run dE/dx without dE/dx calibration container object");
248 GPUError(
"Cannot run gain calibration without calibration object");
252 GPUError(
"Cannot run TPC ZS Decoder without mapping object. (tpczslinkmapping.dump missing?)");
261 if ((
param().
rec.tpc.nWays & 1) == 0) {
262 GPUError(
"nWay setting musst be odd number!");
266 GPUError(
"Cannot do error interpolation with NWays < 3!");
269 if (
param().continuousMaxTimeBin > (
int32_t)GPUSettings::TPC_MAX_TF_TIME_BIN) {
270 GPUError(
"configured max time bin exceeds 256 orbits");
278 GPUError(
"noGPUMemoryRegistration only possible with gather mode 3 (set to %d / %d)",
mRec->
GetProcessingSettings().tpcCompressionGatherMode, gatherMode);
282 GPUError(
"Clusterizer and merger Sanity checks only supported when not running on GPU");
286 GPUError(
"tpcWriteClustersAfterRejection requires compressionTypeMask = 0, no GPU usage, and compression enabled");
291 GPUError(
"Cannot use double pipeline with tpcFreeAllocatedMemoryAfterProcessing");
302 GPUError(
"Must use external output for double pipeline mode");
305 if (gatherMode == 1) {
306 GPUError(
"Double pipeline incompatible to compression mode 1");
310 GPUError(
"Invalid reconstruction settings for double pipeline: Needs compression and cluster finding");
315 GPUError(
"Invalid tpcCompressionGatherMode for compression on CPU");
319 GPUError(
"tpcApplyClusterFilterOnCPU cannot be used with GPU clusterization or with MC labels");
324 GPUError(
"TRD tracking can only run on O2 TPC tracks if createO2Output is enabled (%d), and matBudLUT is available (0x%p)", (
int32_t)
GetProcessingSettings().createO2Output, (
void*)
GetMatLUT());
328 GPUError(
"TRD tracking can only run on GPU TPC tracks if the createO2Output setting does not suppress them");
332 GPUError(
"Cannot use TRD tracking or Refit on GPU without GPU polynomial field map (%d) or matlut table (%p)", (
int32_t)
GetProcessingSettings().o2PropagatorUseGPUField, (
void*)
GetMatLUT());
344 for (
uint32_t i = 0;
i <
sizeof(gpudatatypes::RECO_STEP_NAMES) /
sizeof(gpudatatypes::RECO_STEP_NAMES[0]);
i++) {
346 printf(
" - %s", gpudatatypes::RECO_STEP_NAMES[
i]);
371 qa.reset(
new GPUQA(
this));
377 throw std::runtime_error(
"Error loading event display");
397 std::streambuf* backing =
nullptr;
427 if (
processors()->calibObjects.matLUT && (ptrMask ==
nullptr || ptrMask->
matLUT)) {
468 throw std::runtime_error(
"Forwading zero-suppressed hits not supported");
478 qa.reset(
new GPUQA(
this));
480 if (!
GetQA()->IsInitialized()) {
508 char* fastTransformBase = (
char*)mem;
511 char* podBuf =
nullptr;
614 std::unique_ptr<GPUSettingsGRP> grp;
615 const GPUSettingsProcessing* p =
nullptr;
616 std::lock_guard lk(mMutexUpdateCalib);
640 GPUFatal(
"GPU magnetic field for propagator requested, but received an O2 propagator without GPU field");
665 memcpy(oldFlatPtrsDevice.data(), (
void*)&
mFlatObjectsDevice, oldFlatPtrsDevice.size());
667 bool ptrsChanged = memcmp(oldFlatPtrs.data(), (
void*)&
mFlatObjectsShadow, oldFlatPtrs.size()) || memcmp(oldFlatPtrsDevice.data(), (
void*)&
mFlatObjectsDevice, oldFlatPtrsDevice.size());
669 GPUInfo(
"Updating all calib objects since pointers changed");
688 GPUFatal(
"Cannot run TRD tracking or refit with o2 track model without o2 propagator");
714 }
catch (
const std::bad_alloc&
e) {
715 GPUError(
"Memory Allocation Error");
765 GPUInfo(
"Preempting tpcZS input of foreign chain");
768 mPipelineFinalizationCtx->rec = this->
mRec;
769 foreignChain->mPipelineNotifyCtx = mPipelineFinalizationCtx.get();
804 int32_t retVal2 = RunChainFinalize();
805 return retVal2 ? retVal2 :
retVal;
808int32_t GPUChainTracking::RunChainFinalize()
840 GPUInfo(
"Starting Event Display...");
842 GPUError(
"Error starting Event Display");
856 GPUInfo(
"Press key for next event!");
862 iKey = kbhit() ? getch() : 0;
863 if (iKey == 27 && kbhit()) {
874 }
else if (iKey ==
'n') {
888 GPUInfo(
"Loading next event...");
898 if (mPipelineFinalizationCtx) {
900 std::unique_lock<std::mutex> lock(mPipelineFinalizationCtx->mutex);
901 auto* ctx = mPipelineFinalizationCtx.get();
902 mPipelineFinalizationCtx->cond.wait(lock, [ctx]() {
return ctx->ready; });
904 mPipelineFinalizationCtx.reset();
906 return RunChainFinalize();
912 bool hasDebugError =
false;
924 static int32_t errorsShown = 0;
925 static bool quiet =
false;
926 static std::chrono::time_point<std::chrono::steady_clock> silenceFrom;
928 silenceFrom = std::chrono::steady_clock::now();
931 auto currentTime = std::chrono::steady_clock::now();
932 std::chrono::duration<double> elapsed_seconds = currentTime - silenceFrom;
933 if (elapsed_seconds.count() > 60 * 10) {
940 GPUWarning(
"GPUReconstruction suffered from an error in the %s part",
i ?
"GPU" :
"CPU");
942 GPUError(
"GPUReconstruction suffered from an error in the %s part",
i ?
"GPU" :
"CPU");
951 fillErrors->emplace_back(std::array<uint32_t, 4>{pErrors[4 *
j], pErrors[4 *
j + 1], pErrors[4 *
j + 2], pErrors[4 *
j + 3]});
956 hasDebugError =
true;
962 hasDebugError =
true;
989 std::lock_guard lk(mMutexUpdateCalib);
991 void*
const* pSrc = (
void*
const*)&obj;
993 for (
uint32_t i = 0;
i <
sizeof(*mNewCalibObjects) /
sizeof(
void*);
i++) {
1018 GPUFatal(
"GPU magnetic field for propagator requested, but received an O2 propagator without GPU field");
1025 rec.useMatLUT =
false;
1027 if (
proc.rtc.optSpecialCode == -1) {
1028 proc.rtc.optSpecialCode = syncMode;
1030 if (dEdxMode != -2) {
Definition of container class for dE/dx corrections.
Timeframe settings Average distance in cm of events merged into Randomize distance around average distance of merged events Simulate correct bunch interactions instead of placing only the average number of events A value[n] sets TFlen for[n] collisions in Number of bunches per trainsort Number of bunch trains Instantaneous interaction rate Do not simulate border Total number of collisions to be placed in the interior of all time Do not select random but walk over array of events in stride steps h Event generator settings h configStandalone Use CPU for GPU Force usage of the specified GPU device no CPU fallback Continue processing after an error First event to Number of events to Number of iterations to Number of iterations to Number of initial iterations excluded from const char Directory with events to Events directory is and not inside events Show standalone event Run event generator maximum time bin of continuous for triggered for automatic continuous for automatic continuous triggered uint64_t
Timeframe settings Average distance in cm of events merged into Randomize distance around average distance of merged events Simulate correct bunch interactions instead of placing only the average number of events A value[n] sets TFlen for[n] collisions in Number of bunches per trainsort Number of bunch trains Instantaneous interaction rate Do not simulate border Total number of collisions to be placed in the interior of all time Do not select random but walk over array of events in stride steps h Event generator settings h configStandalone Use CPU for GPU Force usage of the specified GPU device no CPU fallback Continue processing after an error First event to Number of events to Number of iterations to Number of iterations to Number of initial iterations excluded from const char e
configStandalone Reconstruction rec_tpc multiplier for the combined cluster track error during track following area in cm for the search of for z only used if decrease removal by when above this local occupancy rowx Shift of TPC revert cluster flag edge bit distance to edge exceeds this ZOffset for secondary multiple of radius max P of electron K BB bands crossing max P of electron p BB bands crossing max P of electron d BB bands crossing max P of electron t BB bands crossing min accepted dEdxMaxTPC of the track Additive extra cluster error for Y2 if edge flag set Additive extra cluster error for Y2 if single set Additive extra cluster error for Y2 if or single set Additive extra cluster error for Z2 if or single set Extra errors to add to track when crossing depending on addErrorsCECrossing Inner portion of y range in sector that is not used in searching for extrapolated track candidates factor2YS for track merging factor2K for track merging Margin in cm of Y position when rejecting edge clusters based on uncorrected track Y Max fraction of shared clusters for tracklet Except pad from cluster finding if total number of charges in a fragment is above this Except pad from cluster finding if number of consecutive charges in a fragment is above this Threshold where a timebin is considered disabling the noisy pad check for that pad Treat signal above this minimum as saturated Smoothing factor for the exponential Highly Ionising Particle tail filter Number of timebins to use for the averaging Add additional custom track errors when crossing relative as fraction of Do N fit passes in final fit of< 0:reject at max -n hits") AddOptionRTC(rejectIFCLowRadiusCluster, uint8_t, 1, "", 0, "Reject clusters that get the IFC mask error during refit") AddOptionRTC(dEdxTruncLow, uint8_t, 2, "", 0, "Low truncation threshold, fraction of 128") AddOptionRTC(dEdxTruncHigh, uint8_t, 77, "", 0, "High truncation threshold, fraction of 128") AddOptionRTC(extrapolationTracking, int8_t, 1, "", 0, "Enable Extrapolation Tracking(prolong tracks to adjacent sectors to find short segments)") AddOptionRTC(disableRefitAttachment, uint8_t, 0, "", 0, "Bitmask to disable certain attachment steps during refit(1:attachment, 2:propagation, 4:loop following, 8:mirroring)") AddOptionRTC(rejectionStrategy, uint8_t, o2::gpu::GPUSettings::RejectionStrategyA, "", 0, "Enable rejection of TPC clusters for compression(0=no, 1=strategy A, 2=strategy B)") AddOptionRTC(mergeLoopersAfterburner, uint8_t, 1, "", 0, "Run afterburner for additional looper merging") AddOptionRTC(compressionTypeMask, uint8_t, o2::gpu::GPUSettings::CompressionFull, "", 0, "TPC Compression mode bits(1=truncate charge/width LSB, 2=differences, 4=track-model)") AddOptionRTC(compressionSortOrder, uint8_t, o2::gpu::GPUSettings::SortTime, "", 0, "Sort order of TPC compression(0=time, 1=pad, 2=Z-time-pad, 3=Z-pad-time, 4=no sorting(use incoming order))") AddOptionRTC(sigBitsCharge, uint8_t, 4, "", 0, "Number of significant bits for TPC cluster charge in compression mode 1") AddOptionRTC(sigBitsWidth, uint8_t, 3, "", 0, "Number of significant bits for TPC cluster width in compression mode 1") AddOptionRTC(dropLoopers, uint8_t, 0, "", 0, "Drop looping tracks starting from second loop") AddOptionRTC(mergerInterpolateErrors, uint8_t, 1, "", 0, "Use interpolation instead of extrapolation for chi2 based cluster rejection") AddOptionRTC(mergerInterpolateRejectAlsoOnCurrentPosition, uint8_t, 1, "", 0, "When using mergerInterpolateErrors, reject based on chi2 twice computed with interpolated and current track position starting from NDF > mergerNonInterpolateRejectMinNDF Merge tracks accross the central electrode Enable PID response Add sub threshold clusters in TPC dEdx computation OR mask of TPC flags that will reject the cluster in alternative dEdx Reject edge clusters based on uncorrected track Y during track fit Increase decrease removal by when on this number of pads from the edge h configStandalone Reconstruction rec_trd Max chi2 for TRD tracklets to be matched to a track Chi2 cut for strict matching mode Number of sigmas for ITS TPC track time error estimate Addition to search road around track prolongation along Y in cm Optimal tracklet rphi resolution in maximum chi2 per attached tracklet for TRD tracks take into account the full likelihood of the slope including exponential tails Tracks with less attached tracklets are discarded after the tracking Post trigger Pile up integration time in BCs fits all iterations but the last one in projections Use material lookup table for TPC refit Apply workarounds to avoid cyclic merge should not be needed tpc dyn configStandalone Processing proc_rtc Replace constant variables static by constexpr expressions Compile RTC in deterministic with NO_FAST_MATH flags and GPUCA_DETERMINISTIC_MODE define Use RTC to optimize GPU code h configStandalone Processing proc_rtctech Use a file lock to serialize access to the cache folder int32_t
configStandalone Reconstruction rec_tpc multiplier for the combined cluster track error during track following area in cm for the search of for z only used if decrease removal by when above this local occupancy rowx Shift of TPC revert cluster flag edge bit distance to edge exceeds this ZOffset for secondary multiple of radius max P of electron K BB bands crossing max P of electron p BB bands crossing max P of electron d BB bands crossing max P of electron t BB bands crossing min accepted dEdxMaxTPC of the track Additive extra cluster error for Y2 if edge flag set Additive extra cluster error for Y2 if single set Additive extra cluster error for Y2 if or single set Additive extra cluster error for Z2 if or single set Extra errors to add to track when crossing depending on addErrorsCECrossing Inner portion of y range in sector that is not used in searching for extrapolated track candidates factor2YS for track merging factor2K for track merging Margin in cm of Y position when rejecting edge clusters based on uncorrected track Y Max fraction of shared clusters for tracklet Except pad from cluster finding if total number of charges in a fragment is above this Except pad from cluster finding if number of consecutive charges in a fragment is above this Threshold where a timebin is considered disabling the noisy pad check for that pad Treat signal above this minimum as saturated Smoothing factor for the exponential Highly Ionising Particle tail filter Number of timebins to use for the averaging Add additional custom track errors when crossing relative as fraction of Do N fit passes in final fit of< 0:reject at max -n hits") AddOptionRTC(rejectIFCLowRadiusCluster, uint8_t, 1, "", 0, "Reject clusters that get the IFC mask error during refit") AddOptionRTC(dEdxTruncLow, uint8_t, 2, "", 0, "Low truncation threshold, fraction of 128") AddOptionRTC(dEdxTruncHigh, uint8_t, 77, "", 0, "High truncation threshold, fraction of 128") AddOptionRTC(extrapolationTracking, int8_t, 1, "", 0, "Enable Extrapolation Tracking(prolong tracks to adjacent sectors to find short segments)") AddOptionRTC(disableRefitAttachment, uint8_t, 0, "", 0, "Bitmask to disable certain attachment steps during refit(1:attachment, 2:propagation, 4:loop following, 8:mirroring)") AddOptionRTC(rejectionStrategy, uint8_t, o2::gpu::GPUSettings::RejectionStrategyA, "", 0, "Enable rejection of TPC clusters for compression(0=no, 1=strategy A, 2=strategy B)") AddOptionRTC(mergeLoopersAfterburner, uint8_t, 1, "", 0, "Run afterburner for additional looper merging") AddOptionRTC(compressionTypeMask, uint8_t, o2::gpu::GPUSettings::CompressionFull, "", 0, "TPC Compression mode bits(1=truncate charge/width LSB, 2=differences, 4=track-model)") AddOptionRTC(compressionSortOrder, uint8_t, o2::gpu::GPUSettings::SortTime, "", 0, "Sort order of TPC compression(0=time, 1=pad, 2=Z-time-pad, 3=Z-pad-time, 4=no sorting(use incoming order))") AddOptionRTC(sigBitsCharge, uint8_t, 4, "", 0, "Number of significant bits for TPC cluster charge in compression mode 1") AddOptionRTC(sigBitsWidth, uint8_t, 3, "", 0, "Number of significant bits for TPC cluster width in compression mode 1") AddOptionRTC(dropLoopers, uint8_t, 0, "", 0, "Drop looping tracks starting from second loop") AddOptionRTC(mergerInterpolateErrors, uint8_t, 1, "", 0, "Use interpolation instead of extrapolation for chi2 based cluster rejection") AddOptionRTC(mergerInterpolateRejectAlsoOnCurrentPosition, uint8_t, 1, "", 0, "When using mergerInterpolateErrors, reject based on chi2 twice computed with interpolated and current track position starting from NDF > mergerNonInterpolateRejectMinNDF Merge tracks accross the central electrode Enable PID response Add sub threshold clusters in TPC dEdx computation OR mask of TPC flags that will reject the cluster in alternative dEdx Reject edge clusters based on uncorrected track Y during track fit Increase decrease removal by when on this number of pads from the edge h configStandalone Reconstruction rec_trd Max chi2 for TRD tracklets to be matched to a track Chi2 cut for strict matching mode Number of sigmas for ITS TPC track time error estimate Addition to search road around track prolongation along Y in cm Optimal tracklet rphi resolution in maximum chi2 per attached tracklet for TRD tracks take into account the full likelihood of the slope including exponential tails Tracks with less attached tracklets are discarded after the tracking Post trigger Pile up integration time in BCs fits all iterations but the last one in projections Use material lookup table for TPC refit Apply workarounds to avoid cyclic merge should not be needed tpc dyn configStandalone proc
Online TRD tracker based on extrapolated TPC tracks.
Used for storing the MC labels for the TRD tracklets.
TRD Tracklet word for GPU tracker - 32bit tracklet info + half chamber ID + index.
Definition of a container to keep Monte Carlo truth external to simulation objects.
bitfield & setBits(const bitfield v, bool w)
bool isSet(const bitfield &v) const
void SetMatLUT(std::unique_ptr< o2::base::MatLayerCylSet > &&lut)
~GPUChainTracking() override
static void ApplySyncSettings(GPUSettingsProcessing &proc, GPUSettingsRec &rec, gpudatatypes::RecoStepField &steps, bool syncMode, int32_t dEdxMode=-2)
std::unique_ptr< o2::base::MatLayerCylSet > mMatLUTU
void PrintKernelDebugOutput()
int32_t RunTPCCompression()
const o2::base::Propagator * GetDeviceO2Propagator()
std::unique_ptr< o2::tpc::ClusterNativeAccess > mClusterNativeAccess
std::unique_ptr< GPUTRDRecoParam > mTRDRecoParamU
void SetTRDGeometry(std::unique_ptr< o2::trd::GeometryFlat > &&geo)
GPUChainTracking * mQAFromForeignChain
void SetO2Propagator(const o2::base::Propagator *prop)
std::unique_ptr< GPUQA > mQA
GPUTrackingFlatObjects mFlatObjectsDevice
int32_t RunTPCClusterizer(bool synchronizeOutput=true)
int32_t RunTPCTrackingSectors()
void UpdateGPUCalibObjectsPtrs(int32_t stream)
int32_t RunTPCTrackingMerger(bool synchronizeOutput=true)
std::unique_ptr< GPUTrackingInputProvider > mInputsHost
HashStreamBuf mDebugFileStream
std::unique_ptr< GPUTPCClusterStatistics > mCompressionStatistics
const o2::base::MatLayerCylSet * GetMatLUT() const
void SetTRDRecoParam(std::unique_ptr< GPUTRDRecoParam > &&par)
std::unique_ptr< std::ostream > mDebugFile
std::array< GPUOutputControl *, GPUTrackingOutputs::count()> mSubOutputControls
void SetTPCFastTransform(aligned_unique_buffer_ptr< TPCFastTransformPOD > &&tpcFastTransform)
void SetUpdateCalibObjects(const GPUCalibObjectsConst &obj, const GPUNewCalibValues &vals)
bool mUpdateNewCalibObjects
std::unique_ptr< GPUCalibObjectsConst > mNewCalibObjects
std::unique_ptr< o2::trd::GeometryFlat > mTRDGeometryU
int32_t Finalize() override
GPUTrackingFlatObjects mFlatObjectsShadow
void RegisterPermanentMemoryAndProcessors() override
void RegisterGPUProcessors() override
void ClearErrorCodes(bool cpuOnly=false)
const GPUQA * GetQA() const
void MemorySize(size_t &gpuMem, size_t &pageLockedHostMem) override
GPUChainTracking(GPUReconstruction *rec, uint32_t maxTPCHits=constants::GPU_MEM_MAX_TPC_CLUSTERS, uint32_t maxTRDTracklets=constants::GPU_MEM_MAX_TRD_TRACKLETS)
int32_t RunTPCDecompression()
int32_t DoQueuedUpdates(int32_t stream, bool updateSlave=true)
std::unique_ptr< GPUNewCalibValues > mNewCalibValues
int32_t CheckErrorCodes(bool cpuOnly=false, bool forceShowErrors=false, std::vector< std::array< uint32_t, 4 > > *fillErrors=nullptr) override
bool mFractionalQAEnabled
void PrepareKernelDebugOutput()
int32_t PrepareEvent() override
int32_t RunChain() override
void UpdateGPUCalibObjects(int32_t stream, const GPUCalibObjectsConst *ptrMask=nullptr)
std::filebuf mDebugFileBuffer
int32_t ConvertNativeToClusterData()
GPUTrackingInOutPointers & mIOPtrs
struct o2::gpu::GPUChainTracking::InOutMemory mIOMem
const o2::tpc::CalibdEdxContainer * GetdEdxCalibContainer() const
std::unique_ptr< GPUTrackingInputProvider > mInputsShadow
const GPUSettingsQA * mConfigQA
int32_t FinalizePipelinedProcessing() override
bool mTPCSectorScratchOnStack
std::unique_ptr< GPUDisplayInterface > mEventDisplay
std::unique_ptr< o2::tpc::ClusterNativeAccess > mClusterNativeAccessReduced
aligned_unique_buffer_ptr< TPCFastTransformPOD > mTPCFastTransformU
void TransferMemoryResourceLinkToGPU(RecoStep step, int16_t res, int32_t stream=-1, deviceEvent *ev=nullptr, deviceEvent *evList=nullptr, int32_t nEvents=1)
GPUConstantMem * processorsDevice()
GPUReconstruction::RecoStepField GetRecoStepsGPU() const
GPUReconstruction::RecoStepField GetRecoSteps() const
void WriteToConstantMemory(RecoStep step, size_t offset, const void *src, size_t size, int32_t stream=-1, deviceEvent *ev=nullptr)
GPUChain * GetNextChainInQueue()
GPUReconstruction::InOutTypeField GetRecoStepsOutputs() const
virtual std::unique_ptr< GPUReconstructionProcessing::threadContext > GetThreadContext()
GPUConstantMem * processors()
const GPUSettingsProcessing & GetProcessingSettings() const
void SynchronizeStream(int32_t stream)
GPUReconstructionCPU * mRec
GPUConstantMem * processorsShadow()
GPUReconstruction::InOutTypeField GetRecoStepsInputs() const
static constexpr int32_t NSECTORS
void TransferMemoryResourceLinkToHost(RecoStep step, int16_t res, int32_t stream=-1, deviceEvent *ev=nullptr, deviceEvent *evList=nullptr, int32_t nEvents=1)
void AllocateIOMemoryHelper(uint32_t n, const T *&ptr, std::unique_ptr< T[]> &u)
int32_t runRecoStep(RecoStep step, S T::*func, Args... args)
GPUReconstruction * rec()
static GPUDisplayInterface * getDisplay(GPUDisplayFrontendInterface *frontend, GPUChainTracking *chain, GPUQA *qa, const GPUParam *param=nullptr, const GPUCalibObjectsConst *calib=nullptr, const GPUSettingsDisplay *config=nullptr, const GPUSettingsProcessing *proc=nullptr)
const uint32_t * getErrorPtr() const
void setMemory(GPUglobalref() uint32_t *m)
bool printErrors(bool silent=false, uint64_t mask=0)
uint32_t getNErrors() const
static void computePointerWithAlignment(T *&basePtr, S *&objPtr, size_t nEntries=1)
void InitGPUProcessor(GPUReconstruction *rec, ProcessorType type=PROCESSOR_TYPE_CPU, GPUProcessor *slaveProcessor=nullptr)
static bool QAAvailable()
int32_t DrawQAHistograms()
void UpdateChain(GPUChainTracking *chain)
static bool IsInitialized()
int32_t InitQA(int32_t tasks=0)
void RunQA(bool matchOnly=false)
HighResTimer & getGeneralStepTimer(GeneralStep step)
void SetNActiveThreads(int32_t n)
const GPUDefParameters & getGPUParameters(bool doGPU) const override
void AddGPUEvents(T *&events)
std::vector< std::array< uint32_t, 4 > > * getErrorCodeOutput()
RecoStepField GetRecoStepsGPU() const
void RegisterGPUDeviceProcessor(GPUProcessor *proc, GPUProcessor *slaveProcessor)
void RegisterGPUProcessor(T *proc, bool deviceSlave)
void setDebugDumpCallback(std::function< void()> &&callback=std::function< void()>(nullptr))
void ResetRegisteredMemoryPointers(GPUProcessor *proc)
int16_t RegisterMemoryAllocation(T *proc, void *(T::*setPtr)(void *), int32_t type, const char *name="", const GPUMemoryReuse &re=GPUMemoryReuse())
uint32_t getNEventsProcessed()
void PopNonPersistentMemory(RecoStep step, uint64_t tag, const GPUProcessor *proc=nullptr)
void UpdateSettings(const GPUSettingsGRP *g, const GPUSettingsProcessing *p=nullptr, const GPUSettingsRecDynamic *d=nullptr)
uint32_t NStreams() const
void PushNonPersistentMemory(uint64_t tag)
GPUMemorySizeScalers * MemoryScalers()
const GPUSettingsProcessing & GetProcessingSettings() const
const GPUSettingsGRP & GetGRPSettings() const
GPUOutputControl & OutputControl()
GPUCalibObjectsTemplate< ConstPtr > GPUCalibObjectsConst
Global TPC definitions and constants.
std::string to_string(gsl::span< T, Size > span)
std::unique_ptr< GPUDisplayFrontendInterface > eventDisplay
constexpr T qStr2Tag(const char(&str)[N])
S< o2::trd::GeometryFlat >::type * trdGeometry
S< o2::tpc::CalibdEdxContainer >::type * dEdxCalibContainer
S< TPCZSLinkMapping >::type * tpcZSLinkMapping
S< TPCFastTransformPOD >::type * fastTransform
S< GPUTRDRecoParam >::type * trdRecoParam
S< TPCPadGainCalib >::type * tpcPadGain
S< o2::base::PropagatorImpl< float > >::type * o2Propagator
S< o2::base::MatLayerCylSet >::type * matLUT
void * SetPointersFlatObjects(void *mem)
GPUCalibObjects mCalibObjects
char * mdEdxSplinesBuffer
GPUChainTracking * mChainTracking
std::unique_ptr< GPUTPCMCInfo[]> mcInfosTPC
std::unique_ptr< GPUTRDTrackletWord[]> trdTracklets
std::unique_ptr< GPUTPCMCInfoCol[]> mcInfosTPCCol
std::unique_ptr< GPUTPCGMMergedTrackHit[]> mergedTrackHits
std::unique_ptr< int32_t[]> trdTrackletIdxFirst
std::unique_ptr< GPUTPCGMMergedTrack[]> mergedTracks
std::unique_ptr< AliHLTTPCClusterMCLabel[]> mcLabelsTPC
std::unique_ptr< GPUTPCClusterData[]> clusterData[NSECTORS]
std::unique_ptr< o2::tpc::ClusterNative[]> clustersNative
std::unique_ptr< float[]> trdTriggerTimes
std::unique_ptr< GPUTPCTrack[]> sectorTracks[NSECTORS]
std::unique_ptr< GPUTRDSpacePoint[]> trdSpacePoints
std::unique_ptr< uint8_t[]> trdTrigRecMask
std::unique_ptr< o2::tpc::ClusterNativeAccess > clusterNativeAccess
std::unique_ptr< AliHLTTPCRawCluster[]> rawClusters[NSECTORS]
std::unique_ptr< GPUTPCHitId[]> sectorClusters[NSECTORS]
std::unique_ptr< GPUTRDTrackGPU[]> trdTracks
GPUTRDTrackerGPU trdTrackerGPU
GPUCalibObjectsConst calibObjects
GPUTRDTracker trdTrackerO2
const int32_t * trdTrackletIdxFirst
const GPUTPCHitId * sectorClusters[NSECTORS]
const o2::tpc::ClusterNativeAccess * clustersNative
const GPUTPCMCInfo * mcInfosTPC
uint32_t nClusterData[NSECTORS]
uint32_t nRawClusters[NSECTORS]
const o2::tpc::CompressedClustersFlat * tpcCompressedClusters
const AliHLTTPCClusterMCLabel * mcLabelsTPC
uint32_t nSectorClusters[NSECTORS]
const GPUTRDSpacePoint * trdSpacePoints
const GPUTPCTrack * sectorTracks[NSECTORS]
const GPUTRDTrackGPU * trdTracks
const GPUTRDTrackletWord * trdTracklets
const GPUTrackingInOutZS * tpcZS
const AliHLTTPCRawCluster * rawClusters[NSECTORS]
const GPUTPCClusterData * clusterData[NSECTORS]
uint32_t nSectorTracks[NSECTORS]
uint32_t nMergedTrackHits
const float * trdTriggerTimes
const uint8_t * trdTrigRecMask
const GPUTPCGMMergedTrackHit * mergedTrackHits
const GPUTrackingInOutDigits * tpcPackedDigits
uint32_t nTRDTriggerRecords
const GPUTPCMCInfoCol * mcInfosTPCCol
const GPUTPCGMMergedTrack * mergedTracks
GPUOutputControl clustersNative
size_t getIndex(const GPUOutputControl &v)
GPUOutputControl sharedClusterMap
GPUOutputControl compressedClusters
GPUOutputControl tpcTracks
unsigned int nClustersTotal