23#include <condition_variable>
45#ifndef GPUCA_STANDALONE
55struct GPUReconstructionPipelineQueue {
59 std::condition_variable c;
68 std::condition_variable
cond;
80static ptrdiff_t ptrDiff(
void*
a,
void*
b) {
return (
char*)
a - (
char*)
b; }
86 throw std::invalid_argument(
"device type of master and slave GPUReconstruction does not match");
89 throw std::invalid_argument(
"Cannot be slave to a slave");
109 GPUError(
"GPU Reconstruction not properly deinitialized!");
118 if (vertexerTraits) {
128 return std::max<int32_t>(0, tbb::this_task_arena::current_thread_index());
134 throw std::runtime_error(
"Must not call init on slave!");
148 GPUError(
"Error initialization slave (before deviceinit)");
175 GPUError(
"Error initialization slave (deviceinit)");
179 GPUError(
"Error initialization slave (permanent memory)");
195 GPUError(
"Error initialization slave (after device init)");
208 const char* tbbEnv = getenv(
"TBB_NUM_THREADS");
209 uint32_t tbbNum = tbbEnv ? atoi(tbbEnv) : 0;
213 const char* ompEnv = getenv(
"OMP_NUM_THREADS");
214 uint32_t ompNum = ompEnv ? atoi(ompEnv) : 0;
218 return tbb::info::default_concurrency();
226 printf(
"\nConfig Dump %s\n",
mMaster ?
"Slave" :
"Master");
269 bool detMode =
false;
270#ifdef GPUCA_DETERMINISTIC_MODE
278 GPUError(
"WARNING, deterministicGPUReconstruction needs GPUCA_DETERMINISTIC_MODE for being fully deterministic, without only most indeterminism by concurrency is removed, but floating point effects remain!");
280 GPUInfo(
"GPU Deterministic Reconstruction is enabled");
283 GPUError(
"WARNING: debugMask %d - debug output might not be deterministic with intermediate steps missing",
mProcessingSettings->debugMask);
292 GPUError(
"WARNING, compiled with GPUCA_DETERMINISTIC_MODE but deterministicGPUReconstruction not set, only compile-time determinism and deterministic math enforced, not fully deterministic!");
325 GPUFatal(
"Must not use both nHostThreads and ompThreads at the same time!");
328 GPUWarning(
"You are using the deprecated ompThreads option, please switch to nHostThreads!");
338 mThreading = std::make_shared<GPUReconstructionThreading>();
339 mThreading->control = std::make_unique<tbb::global_control>(tbb::global_control::max_allowed_parallelism,
mMaxHostThreads);
365 GPUError(
"Must use double pipeline mode only with exactly one chain that must support it");
373 GPUError(
"Not allowed to create optimized RTC code with more than one GPUReconstruction instances");
382 mChains[
i]->RegisterPermanentMemoryAndProcessors();
383 size_t memPrimary, memPageLocked;
384 mChains[
i]->MemorySize(memPrimary, memPageLocked);
386 memPageLocked = memPrimary;
414 mChains[
i]->RegisterGPUProcessors();
464 GPUError(
"Error exiting slave");
494 auto& re = it->second;
495 if (
proc ==
nullptr || re.proc ==
proc) {
497 resMain.mOverrideSize = 0;
500 resMain.mOverrideSize = std::max<size_t>(resMain.mOverrideSize, ptrDiff(
res.SetPointers((
void*)1), (
char*)1));
520 throw std::bad_alloc();
529 it->second.res.emplace_back(
retVal);
538 GPUInfo(
"Allocating memory %p", (
void*)
proc);
551 GPUInfo(
"Allocating memory done");
559 GPUInfo(
"Allocating Permanent Memory");
562 GPUError(
"Must not allocate permanent memory while volatile chunks are allocated");
563 throw std::bad_alloc();
574 GPUInfo(
"Permanent Memory Done");
581 if (
res->mReuse >= 0) {
583 if (
ptr ==
nullptr) {
584 GPUError(
"Invalid reuse ptr (%s)",
res->mName);
585 throw std::bad_alloc();
590 throw std::bad_alloc();
593 std::cout <<
"Reused (" <<
device <<
") " <<
res->mName <<
": " <<
retVal <<
"\n";
597 if (memorypool ==
nullptr) {
598 GPUError(
"Cannot allocate memory from uninitialized pool");
599 throw std::bad_alloc();
604 retVal = ptrDiff((
res->*setPtr)((
char*)1), (
char*)(1));
605 memorypoolend = (
void*)((
char*)memorypoolend - GPUProcessor::getAlignmentMod<constants::GPU_MEMALIGN>(memorypoolend));
606 if (retVal < res->mOverrideSize) {
609 retVal += GPUProcessor::getAlignment<constants::GPU_MEMALIGN>(
retVal);
610 memorypoolend = (
char*)memorypoolend -
retVal;
615 memorypool = (
char*)((
res->*setPtr)(
ptr));
617 if (retVal < res->mOverrideSize) {
619 memorypool = (
char*)
ptr +
res->mOverrideSize;
621 memorypool = (
void*)((
char*)memorypool + GPUProcessor::getAlignment<constants::GPU_MEMALIGN>(memorypool));
623 if (memorypoolend ? (memorypool > memorypoolend) : ((size_t)ptrDiff(memorypool, memorybase) > memorysize)) {
624 std::cerr <<
"Memory pool size exceeded (" <<
device <<
") (" <<
res->mName <<
": " << (memorypoolend ? (memorysize + ptrDiff(memorypool, memorypoolend)) : ptrDiff(memorypool, memorybase)) <<
" > " << memorysize <<
"\n";
625 throw std::bad_alloc();
628 std::cout <<
"Allocated (" <<
device <<
") " <<
res->mName <<
": " <<
retVal <<
" - available: " << (memorypoolend ? ptrDiff(memorypoolend, memorypool) : (memorysize - ptrDiff(memorypool, memorybase))) <<
"\n";
637 if (
res->mPtrDevice &&
res->mReuse < 0) {
640 res->mSize = std::max((
size_t)
res->SetPointers((
void*)1) - 1,
res->mOverrideSize);
641 if (
res->mReuse >= 0) {
643 GPUError(
"Invalid reuse, insufficient size: %ld < %ld", (int64_t)
mMemoryResources[
res->mReuse].mSize, (int64_t)
res->mSize);
644 throw std::bad_alloc();
648 res->mPtrDevice = alignedDefaultBufferAllocator<char>(
res->mSize + constants::GPU_BUFFER_ALIGNMENT);
650 res->mPtr = GPUProcessor::alignPointer<constants::GPU_BUFFER_ALIGNMENT>(
res->mPtrDevice);
651 res->SetPointers(
res->mPtr);
659 if ((
size_t)
res->mPtr % constants::GPU_BUFFER_ALIGNMENT) {
660 GPUError(
"Got buffer with insufficient alignment");
661 throw std::bad_alloc();
665 if (
res->mPtr !=
nullptr) {
666 GPUError(
"Double allocation! (%s)",
res->mName);
667 throw std::bad_alloc();
669 if (
IsGPU() &&
res->mOverrideSize < constants::GPU_BUFFER_ALIGNMENT) {
670 res->mOverrideSize = constants::GPU_BUFFER_ALIGNMENT;
675 res->mSize = std::max((
size_t)
res->SetPointers((
void*)1) - 1,
res->mOverrideSize);
676 res->mPtr = control->
allocator(CAMath::nextMultipleOf<constants::GPU_BUFFER_ALIGNMENT>(
res->mSize));
677 res->mSize = std::max<size_t>(ptrDiff(
res->SetPointers(
res->mPtr),
res->mPtr),
res->mOverrideSize);
679 std::cout <<
"Allocated (from callback) " <<
res->mName <<
": " <<
res->mSize <<
"\n";
682 void* dummy =
nullptr;
688 if ((
size_t)
res->mPtr % constants::GPU_BUFFER_ALIGNMENT) {
689 GPUError(
"Got buffer with insufficient alignment");
690 throw std::bad_alloc();
694 if (
res->mProcessor->mLinkedProcessor ==
nullptr) {
695 GPUError(
"Device Processor not set (%s)",
res->mName);
696 throw std::bad_alloc();
699 GPUError(
"Must not allocate non-stacked device memory while volatile chunks are allocated");
700 throw std::bad_alloc();
706 }
else if (
size !=
res->mSize) {
707 GPUError(
"Inconsistent device memory allocation (%s: device %lu vs %lu)",
res->mName,
size,
res->mSize);
708 throw std::bad_alloc();
710 if ((
size_t)
res->mPtrDevice % constants::GPU_BUFFER_ALIGNMENT) {
711 GPUError(
"Got buffer with insufficient alignment");
712 throw std::bad_alloc();
721 rec->AllocateRegisteredMemoryInternal(&
rec->mMemoryResources[ires], control,
this);
732 return res->mReuse >= 0 ? 0 :
res->mSize;
739 char*
retVal = alignedDefaultBufferAllocator<char>(
size);
749 throw std::runtime_error(
"Requested invalid memory typo for direct allocation");
752 GPUError(
"Must not allocate direct memory while volatile chunks are allocated");
753 throw std::bad_alloc();
760 poolend = (
char*)poolend -
size;
761 poolend = (
char*)poolend - GPUProcessor::getAlignmentMod<constants::GPU_MEMALIGN>(poolend);
766 if (pool > poolend) {
767 GPUError(
"Insufficient unmanaged memory: missing %ld bytes", ptrDiff(pool, poolend));
768 throw std::bad_alloc();
790 throw std::bad_alloc();
804 char*
retVal = alignedDefaultBufferAllocator<char>(
size);
848 size_t size = ptrDiff(
res->SetPointers(basePtr), basePtr);
849 if (basePtr &&
size > std::max(
res->mSize,
res->mOverrideSize)) {
850 std::cerr <<
"Updated pointers exceed available memory size: " <<
size <<
" > " << std::max(
res->mSize,
res->mOverrideSize) <<
" - host - " <<
res->mName <<
"\n";
851 throw std::bad_alloc();
856 size_t size = ptrDiff(
res->SetDevicePointers(basePtr), basePtr);
857 if (basePtr &&
size > std::max(
res->mSize,
res->mOverrideSize)) {
858 std::cerr <<
"Updated pointers exceed available memory size: " <<
size <<
" > " << std::max(
res->mSize,
res->mOverrideSize) <<
" - GPU - " <<
res->mName <<
"\n";
859 throw std::bad_alloc();
881 std::cout <<
"Freeing " <<
res->mName <<
": size " <<
res->mSize <<
" (reused " <<
res->mReuse <<
")\n";
887 res->mPtrDevice =
nullptr;
898 GPUFatal(
"Processor-depending memory-free works only with allocation strategy ALLOCATION_INDIVIDUAL");
904 GPUFatal(
"Trying to pop memory state from empty stack");
921 std::cout <<
"Freeing NonPersistent " <<
res->mName <<
": size " <<
res->mSize <<
" (reused " <<
res->mReuse <<
")\n";
923 if (
res->mReuse < 0) {
927 res->mPtrDevice =
nullptr;
942 throw std::runtime_error(
"temporary memory stack already blocked");
951 throw std::runtime_error(
"cannot unblock while there is stacked memory");
1001 printf(
"Memory Allocation: Host %'13zd / %'13zu (Permanent %'13zd, Data %'13zd, Scratch %'13zd), Device %'13zd / %'13zu, (Permanent %'13zd, Data %'13zd, Scratch %'13zd) %zu chunks\n",
1002 ptrDiff(
mHostMemoryPool,
mHostMemoryBase) + ptrDiff((
char*)
mHostMemoryBase +
mHostMemorySize,
mHostMemoryPoolEnd),
mHostMemorySize, ptrDiff(
mHostMemoryPermanent,
mHostMemoryBase), ptrDiff(
mHostMemoryPool,
mHostMemoryPermanent), ptrDiff((
char*)
mHostMemoryBase +
mHostMemorySize,
mHostMemoryPoolEnd),
1003 ptrDiff(
mDeviceMemoryPool,
mDeviceMemoryBase) + ptrDiff((
char*)
mDeviceMemoryBase +
mDeviceMemorySize,
mDeviceMemoryPoolEnd),
mDeviceMemorySize, ptrDiff(
mDeviceMemoryPermanent,
mDeviceMemoryBase), ptrDiff(
mDeviceMemoryPool,
mDeviceMemoryPermanent), ptrDiff((
char*)
mDeviceMemoryBase +
mDeviceMemorySize,
mDeviceMemoryPoolEnd),
1010 std::map<std::string, std::array<size_t, 3>>
sizes;
1013 if (
res.mReuse >= 0) {
1020 if (
res.mPtrDevice) {
1027 printf(
"%59s CPU / %9s GPU\n",
"",
"");
1028 for (
auto it =
sizes.begin(); it !=
sizes.end(); it++) {
1029 printf(
"Allocation %50s %s: Size %'14zu / %'14zu\n", it->first.c_str(), it->second[2] ?
"P" :
" ", it->second[0], it->second[1]);
1033 mChains[
i]->PrintMemoryStatistics();
1067constexpr static inline int32_t getStepNum(T step,
bool validCheck, int32_t N,
const char* err =
"Invalid step num")
1069 static_assert(
sizeof(step) ==
sizeof(uint32_t),
"Invalid step enum size");
1071 if ((uint32_t)step == 0 ||
retVal >= N) {
1075 throw std::runtime_error(
"Invalid General Step");
1088 throw std::invalid_argument(
"Cannot start double pipeline mode");
1091 GPUInfo(
"Pipeline worker started");
1093 bool terminate =
false;
1094 while (!terminate) {
1100 GPUReconstructionPipelineQueue* q;
1109 q->retVal = q->chain->RunChain();
1112 std::lock_guard<std::mutex> lk(q->m);
1120 GPUInfo(
"Pipeline worker ended");
1139 std::unique_ptr<GPUReconstructionPipelineQueue> qu(
new GPUReconstructionPipelineQueue);
1140 GPUReconstructionPipelineQueue* q = qu.get();
1141 q->chain = terminate ? nullptr :
mChains[0].get();
1142 q->op = terminate ? 1 : 0;
1143 std::unique_lock<std::mutex> lkdone(q->m);
1145 std::lock_guard<std::mutex> lkpipe(
rec->mPipelineContext->mutex);
1146 if (
rec->mPipelineContext->terminate) {
1147 throw std::runtime_error(
"Must not enqueue work after termination request");
1149 rec->mPipelineContext->pipelineQueue.push(q);
1150 rec->mPipelineContext->terminate = terminate;
1151 rec->mPipelineContext->cond.notify_one();
1153 q->c.wait(lkdone, [&q]() {
return q->done; });
1154 if (terminate || (q->retVal && (q->retVal != 3 || !
GetProcessingSettings().ignoreNonFatalGPUErrors))) {
1158 return retVal2 ? retVal2 : q->retVal;
1164 std::lock_guard<std::mutex> lk(
rec->mPipelineContext->mutex);
1165 return rec->mPipelineContext->pipelineQueue.size() &&
rec->mPipelineContext->pipelineQueue.front()->op == 0 ?
rec->mPipelineContext->pipelineQueue.front()->chain :
nullptr;
1208 throw std::runtime_error(
"GPU Backend Failure");
1217 f +=
"settings.dump";
1252 f +=
"settings.dump";
1268#ifdef GPUCA_STANDALONE
1283 GPUError(
"Cannot update settings while initialized");
1284 throw std::runtime_error(
"Settings updated while initialized");
1317 mAlloc = [&
r](
size_t n) {
return (
char*)
r->AllocateVolatileDeviceMemory(
n); };
unsigned short int uint16_t
configStandalone Light OpenGL display display_light Apply smoothing to points Enable Z buffer Highlight links Highlight seeds Highlight sector tracks Highlight final tracks Show individual Show related Highlight grid Show only individual collision Color clusters belonging to track objects Set line Enable drawing TPC data Enable drawing TOF data Enable drawing magnetic field Set field line step count Invert colors configStandalone Heavy OpenGL display display_heavy Show tracks with ITS contribution Show tracks with TOF contribution Enable propagation of loopers Show clusters at nominal x of pad row for early transformed data Split CE tracks when they cross the central Mark clusters Mark adjacent clusters Hide rejected tracks Apply filter on tracks to be displayed Separate increase X h configStandalone Camera window renderer OpenGL display display_camera Orient the camera such that the y axis is always upwards Full Screen MultiSample Anti Downsampling Anti Enable Vertical Sync Target frame or if the backend cannot limit to the refresh Use OpenGL indirect draws to reduce number of draw calls Display h configStandalone Vulkan display display_vulkan Have uniform buffers in host accessible device memory h configStandalone Internal display display_internal Number of updates the Qt GUI may send ahead of the last one the display has before it waits for the display Opacity of the console background Number of console lines the history is shortened to this when it reaches twice the size Time in seconds an info message fades out after setInfoTime h OpenGL display display Font size Do not use Freetype for font Renderer for event Frontend to use for event ROOT macros used as track filter Console script file
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 d
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 rtc
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
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 Print launch bounds used for RTC code as debugging for exit after Allow RTC with slave GPUReconstruction instances with optConstexpr and optSpecialcode Maximum number of RTC cache files per backend in the cache the oldest are Prepend RTC compilation commands by this string Load a parameter object containing the launch bounds from a file h configStandalone Processing proc_param TPC Cluster Z Error Parameterization h configStandalone Processing settings for neural network proc_nn std::string Specify inference device(cpu(default), rocm, cuda)") AddOption(nnInferenceDeviceId
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 rtctech
char error[CURL_ERROR_SIZE]
bool isSet(const bitfield &v) const
const GPUSettingsDisplay * GetEventDisplayConfig() const
const GPUSettingsQA * GetQAConfig() const
static void dumpConfig(const GPUSettingsRec *rec, const GPUSettingsProcessing *proc, const GPUSettingsQA *qa, const GPUSettingsDisplay *display, const GPUSettingsDeviceBackend *device, const GPURecoStepConfiguration *workflow)
void * SetDevicePointers(void *ptr) const
void * SetPointers(void *ptr) const
static void computePointerWithAlignment(T *&basePtr, S *&objPtr, size_t nEntries=1)
GPURecoStepConfiguration mRecoSteps
int32_t InitPhaseBeforeDevice()
std::unordered_set< const void * > mRegisteredMemoryPtrs
int32_t InitPhasePermanentMemory()
int16_t RegisterMemoryAllocationHelper(GPUProcessor *proc, void *(GPUProcessor::*setPtr)(void *), int32_t type, const char *name, const GPUMemoryReuse &re)
std::vector< std::unique_ptr< GPUChain > > mChains
GPUReconstruction * mMaster
void * AllocateVolatileMemory(size_t size, bool device)
ThrustVolatileAllocator getThrustVolatileDeviceAllocator()
std::unique_ptr< GPUMemorySizeScalers > mMemoryScalers
void AllocateRegisteredForeignMemory(int16_t res, GPUReconstruction *rec, GPUOutputControl *control=nullptr)
std::unique_ptr< T > ReadStructFromFile(const char *file, T *obj=nullptr, bool *errorOnMissing=nullptr, bool allowSmaller=false)
void SetInputControl(void *ptr, size_t size)
GPUConstantMem * mDeviceConstantMem
void ConstructGPUProcessor(GPUProcessor *proc)
void TerminatePipelineWorker()
std::shared_ptr< GPUROOTDumpCore > mROOTDump
GPUConstantMem * processors()
void ReturnVolatileMemory()
const GPUSettingsDeviceBackend & GetDeviceBackendSettings() const
void ComputeReuseMax(GPUProcessor *proc)
void SetMemoryExternalInput(int16_t res, void *ptr)
int32_t getGeneralStepNum(GeneralStep step, bool validCheck=true)
static constexpr uint32_t NSECTORS
void MakeFutureDeviceMemoryAllocationsVolatile()
GPUOutputControl mInputControl
RecoStepField GetRecoStepsGPU() const
void SetResetTimers(bool reset)
void RegisterGPUDeviceProcessor(GPUProcessor *proc, GPUProcessor *slaveProcessor)
std::vector< GPUReconstruction * > mSlaves
std::vector< std::tuple< void *, void *, size_t, size_t, uint64_t > > mNonPersistentMemoryStack
alignedDeleter< char, constants::GPU_BUFFER_ALIGNMENT > alignedDefaultBufferDeleter
void UpdateDynamicSettings(const GPUSettingsRecDynamic *d)
std::unique_ptr< GPUSettingsDeviceBackend > mDeviceBackendSettings
std::vector< GPUMemoryResource > mMemoryResources
std::unique_ptr< GPUReconstructionPipelineContext > mPipelineContext
std::unique_ptr< GPUConstantMem > mHostConstantMem
size_t AllocateRegisteredPermanentMemory()
void ResetRegisteredMemoryPointers(GPUProcessor *proc)
void DumpStructToFile(const T *obj, const char *file)
void AllocateRegisteredMemoryInternal(GPUMemoryResource *res, GPUOutputControl *control, GPUReconstruction *recPool)
std::vector< std::unique_ptr< char[], alignedDefaultBufferDeleter > > mVolatileChunks
virtual int32_t GPUChkErrInternal(const int64_t retval, const char *file, int32_t line) const
virtual int32_t registerMemoryForGPU_internal(const void *ptr, size_t size)=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
std::unordered_map< GPUMemoryReuse::ID, MemoryReuseMeta > mMemoryReuse1to1
size_t mDeviceMemoryUsedMax
std::vector< ProcessorData > mProcessors
void ReturnVolatileDeviceMemory()
void * AllocateVolatileDeviceMemory(size_t size)
bool mDeviceMemoryAsVolatile
virtual int32_t InitDevice()=0
void SetSettings(float solenoidBzNominalGPU, const GPURecoStepConfiguration *workflow=nullptr)
virtual ~GPUReconstruction()
int32_t mMaxBackendThreads
const GPUCalibObjectsConst & GetCalib() const
const GPUTrackingInOutPointers GetIOPtrs() const
virtual std::unique_ptr< gpu_reconstruction_kernels::threadContext > GetThreadContext()=0
void UnblockStackedMemory()
GPUReconstruction(const GPUReconstruction &)=delete
static constexpr GeometryType geometryType
void WriteConstantParams(int32_t stream=-1)
std::atomic_flag mMemoryMutex
void FreeRegisteredMemory(GPUProcessor *proc, bool freeCustom=false, bool freePermanent=false)
void UpdateMaxMemoryUsed()
virtual RecoStepField AvailableGPURecoSteps()
static constexpr const char *const IOTYPENAMES[]
void PopNonPersistentMemory(RecoStep step, uint64_t tag, const GPUProcessor *proc=nullptr)
void UpdateSettings(const GPUSettingsGRP *g, const GPUSettingsProcessing *p=nullptr, const GPUSettingsRecDynamic *d=nullptr)
DeviceType GetDeviceType() const
int32_t CheckErrorCodes(bool cpuOnly=false, bool forceShowErrors=false, std::vector< std::array< uint32_t, 4 > > *fillErrors=nullptr)
const GPUParam & GetParam() const
void ClearAllocatedMemory(bool clearOutputs=true)
static constexpr const char *const GEOMETRY_TYPE_NAMES[]
GPUOutputControl mOutputControl
size_t mHostMemoryUsedMax
void * mDeviceMemoryPoolEnd
virtual int32_t ExitDevice()=0
void PrintMemoryOverview()
std::unique_ptr< GPUSettingsGRP > mGRPSettings
std::unique_ptr< GPUSettingsProcessing > mProcessingSettings
virtual bool CanQueryMaxMemory()
void PrintMemoryStatistics()
void PushNonPersistentMemory(uint64_t tag)
int32_t getRecoStepNum(RecoStep step, bool validCheck=true)
virtual int32_t unregisterMemoryForGPU_internal(const void *ptr)=0
int32_t InitPhaseAfterDevice()
static int32_t getHostThreadIndex()
void * mDeviceMemoryPermanent
void BlockStackedMemory(GPUReconstruction *rec)
const GPUSettingsProcessing & GetProcessingSettings() const
void DumpSettings(const char *dir="")
std::vector< std::unique_ptr< char[], alignedDefaultBufferDeleter > > mNonPersistentIndividualDirectAllocations
void * AllocateDirectMemory(size_t size, int32_t type)
void * mHostMemoryPoolBlocked
virtual void GetITSTraits(std::unique_ptr< o2::its::TrackerTraits< 7 > > *trackerTraits, std::unique_ptr< o2::its::VertexerTraits< 7 > > *vertexerTraits, std::unique_ptr< o2::its::TimeFrame< 7 > > *timeFrame)
int32_t unregisterMemoryForGPU(const void *ptr)
int32_t registerMemoryForGPU(const void *ptr, size_t size)
void SetDebugLevelTmp(int32_t level)
size_t AllocateRegisteredMemoryHelper(GPUMemoryResource *res, void *&ptr, void *&memorypool, void *memorybase, size_t memorysize, void *(GPUMemoryResource::*SetPointers)(void *) const, void *&memorypoolend, const char *device)
std::vector< std::unique_ptr< char[], alignedDefaultBufferDeleter > > mDirectMemoryChunks
int32_t EnqueuePipeline(bool terminate=false)
std::shared_ptr< GPUReconstructionThreading > mThreading
std::vector< GPUMemoryResource * > mNonPersistentIndividualAllocations
void * mHostMemoryPoolEnd
void * mDeviceMemoryPoolBlocked
void * mVolatileMemoryStart
GPUChain * GetNextChainInQueue()
void * mHostMemoryPermanent
int32_t GPUChkErrA(const int64_t error, const char *file, int32_t line, bool failOnError)
size_t AllocateRegisteredMemory(GPUProcessor *proc, bool resetCustom=false)
int32_t ReadSettings(const char *dir="")
void SetOutputControl(const GPUOutputControl &v)
void SetSector(int32_t iSector)
#define TPC_MAX_FRAGMENT_LEN_GPU
#define TPC_MAX_FRAGMENT_LEN_HOST
GLuint GLsizei const GLuint const GLintptr const GLsizeiptr * sizes
GLuint const GLchar * name
GLboolean GLboolean GLboolean b
GLint GLint GLsizei GLint GLenum GLenum type
typedef void(APIENTRYP PFNGLCULLFACEPROC)(GLenum mode)
GLboolean GLboolean GLboolean GLboolean a
GLubyte GLubyte GLubyte GLubyte w
std::unique_ptr< GPUDisplayFrontendInterface > eventDisplay
std::string qTag2Str(const T tag)
GPUTPCTracker tpcTrackers[GPUTPCGeometry::NSECTORS]
GPUCalibObjectsConst calibObjects
GPUTPCClusterFinder tpcClusterer[GPUTPCGeometry::NSECTORS]
GPUSettingsProcessing configProcessing
GPUSettingsO2 ReadConfigurableParam()
GPUSettingsRec configReconstruction
void set(void *p, size_t s)
std::function< void *(size_t)> allocator
void SetDefaults(float solenoidBz, bool assumeConstantBz=false)
void UpdateSettings(const GPUSettingsGRP *g, const GPUSettingsProcessing *p=nullptr, const GPURecoStepConfiguration *w=nullptr, const GPUSettingsRecDynamic *d=nullptr)
gpudatatypes::RecoStepField steps
gpudatatypes::InOutTypeField inputs
gpudatatypes::RecoStepField stepsGPUMask
gpudatatypes::InOutTypeField outputs
std::condition_variable cond
std::queue< GPUReconstructionPipelineQueue * > pipelineQueue
GPUReconstruction * master
float solenoidBzNominalGPU