Merge branch 'onnxruntime-cpu' into gpu_clusterizer

Author: ChSonnabend

CODEOWNERS

@@ -18,7 +18,7 @@
 /Common/Constants    @shahor02
 /Common/Field        @shahor02
 /Common/MathUtils    @shahor02
-/Common/SimConfig    @sawenzel @benedikt-voelkel
+/Common/SimConfig    @sawenzel
 #/Common/Topologies
 #/Common/Types
 /Common/Utils        @sawenzel
@@ -37,10 +37,10 @@
 /DataFormats/Detectors/ITSMFT                  @mcoquet642 @mconcas @shahor02
 /DataFormats/Detectors/MUON                    @AliceO2Group/muon-experts @shahor02
 /DataFormats/Detectors/PHOS                    @peressounko @kharlov
-/DataFormats/Detectors/Passive                 @sawenzel @benedikt-voelkel
+/DataFormats/Detectors/Passive                 @sawenzel
 /DataFormats/Detectors/TOF                     @noferini
 /DataFormats/Detectors/TPC                     @davidrohr @wiechula @shahor02
-/DataFormats/Detectors/TRD                     @f3sch @bazinski @martenole @wille10
+/DataFormats/Detectors/TRD                     @f3sch @bazinski @wille10
 /DataFormats/Detectors/Upgrades                @mconcas
 /DataFormats/Detectors/Upgrades/ITS3           @fgrosa @arossi81
 /DataFormats/Detectors/ZDC                     @coppedis
@@ -53,25 +53,25 @@
 /DataFormats/Reconstruction    @shahor02
 #/DataFormats/TimeFrame
 /DataFormats/common            @shahor02
-/DataFormats/simulation        @sawenzel @benedikt-voelkel
+/DataFormats/simulation        @sawenzel
 
-/Detectors/Base                    @sawenzel @shahor02 @benedikt-voelkel
+/Detectors/Base                    @sawenzel @shahor02
 /Detectors/Calibration             @chiarazampolli @shahor02
 /Detectors/CPV                     @peressounko @kharlov
 /Detectors/EMCAL                   @mfasDa @jokonig
 /Detectors/FIT                     @jotwinow @afurs @andreasmolander @arvindkhuntia @mslupeck
 /Detectors/FOCAL                   @maxrauch @mfasDa @iarsene @matthiasrichter
-/Detectors/Geometry                @sawenzel @shahor02 @benedikt-voelkel
+/Detectors/Geometry                @sawenzel @shahor02
 /Detectors/GlobalTracking          @shahor02
 /Detectors/GlobalTrackingWorkflow  @shahor02
 /Detectors/HMPID                   @gvolpe79
 /Detectors/ITSMFT                  @mcoquet642 @mconcas @shahor02
 /Detectors/MUON                    @AliceO2Group/muon-experts @shahor02
 /Detectors/PHOS                    @peressounko @kharlov
-/Detectors/Passive                 @sawenzel @benedikt-voelkel
+/Detectors/Passive                 @sawenzel
 /Detectors/TOF                     @noferini
 /Detectors/TPC                     @davidrohr @wiechula @shahor02
-/Detectors/TRD                     @f3sch @bazinski @martenole @wille10
+/Detectors/TRD                     @f3sch @bazinski @wille10
 /Detectors/Upgrades                @mconcas
 /Detectors/Upgrades/ITS3           @fgrosa @arossi81 @mconcas @f3sch
 /Detectors/ZDC                     @coppedis
@@ -110,13 +110,13 @@
 /GPU                          @davidrohr
 #/GPU/Common
 #/GPU/GPUTracking
-/GPU/GPUTracking/TRDTracking  @davidrohr @martenole
+/GPU/GPUTracking/TRDTracking  @davidrohr
 /GPU/TPCFastTransformation    @davidrohr @sgorbuno
 /GPU/TPCSpaceChargeBase       @davidrohr @ehellbar
 
-/Generators  @sawenzel @benedikt-voelkel
+/Generators  @sawenzel @jackal1-66
 
-/Steer @sawenzel @shahor02 @benedikt-voelkel
+/Steer @sawenzel @shahor02
 
 /Testing
 

Common/Constants/include/CommonConstants/PhysicsConstants.h

@@ -75,7 +75,8 @@ enum Pdg {
   kHyperTriton = 1010010030,
   kHyperHydrogen4 = 1010010040,
   kHyperHelium4 = 1010020040,
-  kHyperHelium5 = 1010020050
+  kHyperHelium5 = 1010020050,
+  kHyperHelium4Sigma = 1110020040
 };
 
 /// \brief Declarations of masses for additional particles
@@ -124,6 +125,7 @@ constexpr double MassHyperTriton = 2.99131;
 constexpr double MassHyperHydrogen4 = 3.9226;
 constexpr double MassHyperHelium4 = 3.9217;
 constexpr double MassHyperHelium5 = 4.841;
+constexpr double MassHyperHelium4Sigma = 3.995;
 
 /// \brief Declarations of masses for particles in ROOT PDG_t
 constexpr double MassDown = 0.00467;
@@ -194,6 +196,7 @@ constexpr double MassKaonNeutral = MassK0;
 constexpr double MassLambda = MassLambda0;
 constexpr double MassHyperhydrog4 = MassHyperHydrogen4;
 constexpr double MassHyperhelium4 = MassHyperHelium4;
+constexpr double MassHyperhelium4sigma = MassHyperHelium4Sigma;
 
 // Light speed
 constexpr float LightSpeedCm2S = 299792458.e2;            // C in cm/s

Common/Constants/include/CommonConstants/make_pdg_header.py

@@ -131,6 +131,7 @@ class Pdg(Enum):
     kHyperHydrogen4 = 1010010040
     kHyperHelium4 = 1010020040
     kHyperHelium5 = 1010020050
+    kHyperHelium4Sigma = 1110020040
 
 
 dbPdg = ROOT.o2.O2DatabasePDG

Common/ML/src/ort_interface.cxx

@@ -107,7 +107,7 @@ void OrtModel::reset(std::unordered_map<std::string, std::string> optionsMap)
   (pImplOrt->sessionOptions).SetLogSeverityLevel(OrtLoggingLevel(loggingLevel));
 
   pImplOrt->env = std::make_shared<Ort::Env>(OrtLoggingLevel(loggingLevel), (optionsMap["onnx-environment-name"].empty() ? "onnx_model_inference" : optionsMap["onnx-environment-name"].c_str()));
-  (pImplOrt->session).reset(new Ort::Session{*(pImplOrt->env), modelPath.c_str(), pImplOrt->sessionOptions});
+  pImplOrt->session = std::make_shared<Ort::Session>(*(pImplOrt->env), modelPath.c_str(), pImplOrt->sessionOptions);
 
   for (size_t i = 0; i < (pImplOrt->session)->GetInputCount(); ++i) {
     mInputNames.push_back((pImplOrt->session)->GetInputNameAllocated(i, pImplOrt->allocator).get());
@@ -145,7 +145,7 @@ void OrtModel::reset(std::unordered_map<std::string, std::string> optionsMap)
 
 void OrtModel::resetSession()
 {
-  (pImplOrt->session).reset(new Ort::Session{*(pImplOrt->env), modelPath.c_str(), pImplOrt->sessionOptions});
+  pImplOrt->session = std::make_shared<Ort::Session>(*(pImplOrt->env), modelPath.c_str(), pImplOrt->sessionOptions);
 }
 
 template <class I, class O>
@@ -156,8 +156,9 @@ std::vector<O> OrtModel::v2v(std::vector<I>& input, bool clearInput)
   } else {
     std::vector<O> output(input.size());
     std::transform(std::begin(input), std::end(input), std::begin(output), [](I f) { return O(f); });
-    if (clearInput)
+    if (clearInput) {
       input.clear();
+    }
     return output;
   }
 }
@@ -167,7 +168,7 @@ std::vector<O> OrtModel::inference(std::vector<I>& input)
 {
   std::vector<int64_t> inputShape{(int64_t)(input.size() / mInputShapes[0][1]), (int64_t)mInputShapes[0][1]};
   std::vector<Ort::Value> inputTensor;
-  inputTensor.emplace_back(Ort::Value::CreateTensor<O>(pImplOrt->memoryInfo, (reinterpret_cast<O*>(input)).data(), input.size(), inputShape.data(), inputShape.size()));
+  inputTensor.emplace_back(Ort::Value::CreateTensor<O>(pImplOrt->memoryInfo, reinterpret_cast<O*>(input.data()), input.size(), inputShape.data(), inputShape.size()));
   // input.clear();
   auto outputTensors = (pImplOrt->session)->Run(pImplOrt->runOptions, inputNamesChar.data(), inputTensor.data(), inputTensor.size(), outputNamesChar.data(), outputNamesChar.size());
   O* outputValues = reinterpret_cast<O*>(outputTensors[0].template GetTensorMutableData<O>());
@@ -182,7 +183,7 @@ std::vector<O> OrtModel::inference(std::vector<std::vector<I>>& input)
   std::vector<Ort::Value> inputTensor;
   for (auto i : input) {
     std::vector<int64_t> inputShape{(int64_t)(i.size() / mInputShapes[0][1]), (int64_t)mInputShapes[0][1]};
-    inputTensor.emplace_back(Ort::Value::CreateTensor<O>(pImplOrt->memoryInfo, (reinterpret_cast<O*>(i)).data(), i.size(), inputShape.data(), inputShape.size()));
+    inputTensor.emplace_back(Ort::Value::CreateTensor<O>(pImplOrt->memoryInfo, reinterpret_cast<O*>(i.data()), i.size(), inputShape.data(), inputShape.size()));
   }
   // input.clear();
   auto outputTensors = (pImplOrt->session)->Run(pImplOrt->runOptions, inputNamesChar.data(), inputTensor.data(), inputTensor.size(), outputNamesChar.data(), outputNamesChar.size());
@@ -195,8 +196,9 @@ std::vector<O> OrtModel::inference(std::vector<std::vector<I>>& input)
 std::string OrtModel::printShape(const std::vector<int64_t>& v)
 {
   std::stringstream ss("");
-  for (size_t i = 0; i < v.size() - 1; i++)
+  for (size_t i = 0; i < v.size() - 1; i++) {
     ss << v[i] << "x";
+  }
   ss << v[v.size() - 1];
   return ss.str();
 }

Common/MathUtils/include/MathUtils/SMatrixGPU.h

@@ -29,6 +29,7 @@
 #include "GPUCommonMath.h"
 #include "GPUCommonAlgorithm.h"
 #include "GPUCommonLogger.h"
+#include "GPUCommonTypeTraits.h"
 
 namespace o2::math_utils::detail
 {
@@ -446,6 +447,8 @@ class SMatrixGPU
   GPUdi() SMatrixGPU(SMatrixNoInit) {}
   GPUd() SMatrixGPU(SMatrixIdentity);
   GPUd() SMatrixGPU(const SMatrixGPU<T, D1, D2, R>& rhs);
+  template <class R2>
+  GPUd() SMatrixGPU(const SMatrixGPU<T, D1, D2, R2>& rhs);
   template <class A, class R2>
   GPUd() SMatrixGPU(const Expr<A, T, D1, D2, R2>& rhs);
   template <class M>
@@ -466,6 +469,9 @@ class SMatrixGPU
   GPUd() const T& operator()(unsigned int i, unsigned int j) const;
   GPUd() T& operator()(unsigned int i, unsigned int j);
 
+  template <typename Y, typename X>
+  GPUd() friend X& operator<<(Y& y, const SMatrixGPU&);
+
   class SMatrixRowGPU
   {
    public:
@@ -497,6 +503,11 @@ class SMatrixGPU
   GPUd() SMatrixRowGPU operator[](unsigned int i) { return SMatrixRowGPU(*this, i); }
   template <class R2>
   GPUd() SMatrixGPU<T, D1, D2, R>& operator+=(const SMatrixGPU<T, D1, D2, R2>& rhs);
+  GPUd() SMatrixGPU<T, D1, D2, R>& operator*=(const T& rhs);
+  template <class R2>
+  GPUd() SMatrixGPU<T, D1, D2, R>& operator*=(const SMatrixGPU<T, D1, D2, R2>& rhs);
+  template <class A, class R2>
+  GPUd() SMatrixGPU<T, D1, D2, R>& operator*=(const Expr<A, T, D1, D2, R2>& rhs);
 
   GPUd() bool Invert();
   GPUd() bool IsInUse(const T* p) const;
@@ -505,6 +516,15 @@ class SMatrixGPU
   R mRep;
 };
 
+#ifndef __OPENCL__ // TODO: current C++ for OpenCL 2021 is at C++17, so no concepts. But we don't need this trick for OpenCL anyway, so we can just hide it.
+template <class T, unsigned int D1, unsigned int D2, class R, typename Y, typename X = Y>
+  requires(sizeof(typename X::traits_type::pos_type) != 0) // do not provide a template to fair::Logger, etc... (pos_type is a member type of all std::ostream classes)
+GPUd() X& operator<<(Y& y, const SMatrixGPU<T, D1, D2, R>&)
+{
+  return y;
+}
+#endif
+
 template <class T, unsigned int D1, unsigned int D2, class R>
 GPUdi() SMatrixGPU<T, D1, D2, R>::SMatrixGPU(SMatrixIdentity)
 {
@@ -528,6 +548,13 @@ GPUdi() SMatrixGPU<T, D1, D2, R>::SMatrixGPU(const SMatrixGPU<T, D1, D2, R>& rhs
   mRep = rhs.mRep;
 }
 
+template <class T, unsigned int D1, unsigned int D2, class R>
+template <class R2>
+GPUd() SMatrixGPU<T, D1, D2, R>::SMatrixGPU(const SMatrixGPU<T, D1, D2, R2>& rhs)
+{
+  operator=(rhs);
+}
+
 template <class T, unsigned int D1, unsigned int D2, class R>
 GPUdi() T* SMatrixGPU<T, D1, D2, R>::begin()
 {
@@ -1387,6 +1414,29 @@ GPUdi() SMatrixGPU<T, D1, D2, R>& SMatrixGPU<T, D1, D2, R>::operator+=(const SMa
   return *this;
 }
 
+template <class T, unsigned int D1, unsigned int D2, class R>
+GPUdi() SMatrixGPU<T, D1, D2, R>& SMatrixGPU<T, D1, D2, R>::operator*=(const T & rhs)
+{
+  for (unsigned int i = 0; i < R::kSize; ++i) {
+    mRep.Array()[i] *= rhs;
+  }
+  return *this;
+}
+
+template <class T, unsigned int D1, unsigned int D2, class R>
+template <class R2>
+GPUdi() SMatrixGPU<T, D1, D2, R>& SMatrixGPU<T, D1, D2, R>::operator*=(const SMatrixGPU<T, D1, D2, R2>& rhs)
+{
+  return operator=(*this* rhs);
+}
+
+template <class T, unsigned int D1, unsigned int D2, class R>
+template <class A, class R2>
+GPUdi() SMatrixGPU<T, D1, D2, R>& SMatrixGPU<T, D1, D2, R>::operator*=(const Expr<A, T, D1, D2, R2>& rhs)
+{
+  return operator=(*this* rhs);
+}
+
 template <class T, unsigned int D1, unsigned int D2, class R>
 struct TranspPolicyGPU {
   enum {

Common/SimConfig/include/SimConfig/SimConfig.h

@@ -83,6 +83,7 @@ struct SimConfigData {
   bool mNoGeant = false;                              // if Geant transport should be turned off (when one is only interested in the generated events)
   bool mIsUpgrade = false;                            // true if the simulation is for Run 5
   std::string mFromCollisionContext = "";             // string denoting a collision context file; If given, this file will be used to determine number of events
+                                                      //
   bool mForwardKine = false;                          // true if tracks and event headers are to be published on a FairMQ channel (for reading by other consumers)
   bool mWriteToDisc = true;                           // whether we write simulation products (kine, hits) to disc
   VertexMode mVertexMode = VertexMode::kDiamondParam; // by default we should use die InteractionDiamond parameter
@@ -177,6 +178,10 @@ class SimConfig
   bool writeToDisc() const { return mConfigData.mWriteToDisc; }
   VertexMode getVertexMode() const { return mConfigData.mVertexMode; }
 
+  // returns the pair of collision context filename as well as event prefix encoded
+  // in the mFromCollisionContext string. Returns empty string if information is not available or set.
+  std::pair<std::string, std::string> getCollContextFilenameAndEventPrefix() const;
+
  private:
   SimConfigData mConfigData; //!
 

Common/SimConfig/src/SimConfig.cxx

@@ -76,7 +76,7 @@ void SimConfig::initOptions(boost::program_options::options_description& options
     "noGeant", bpo::bool_switch(), "prohibits any Geant transport/physics (by using tight cuts)")(
     "forwardKine", bpo::bool_switch(), "forward kinematics on a FairMQ channel")(
     "noDiscOutput", bpo::bool_switch(), "switch off writing sim results to disc (useful in combination with forwardKine)");
-  options.add_options()("fromCollContext", bpo::value<std::string>()->default_value(""), "Use a pregenerated collision context to infer number of events to simulate, how to embedd them, the vertex position etc. Takes precedence of other options such as \"--nEvents\".");
+  options.add_options()("fromCollContext", bpo::value<std::string>()->default_value(""), "Use a pregenerated collision context to infer number of events to simulate, how to embedd them, the vertex position etc. Takes precedence of other options such as \"--nEvents\". The format is COLLISIONCONTEXTFILE.root[:SIGNALNAME] where SIGNALNAME is the event part in the context which is relevant.");
 }
 
 void SimConfig::determineActiveModules(std::vector<std::string> const& inputargs, std::vector<std::string> const& skippedModules, std::vector<std::string>& activeModules, bool isUpgrade)
@@ -270,6 +270,21 @@ void SimConfig::determineReadoutDetectors(std::vector<std::string> const& active
   }
 }
 
+std::pair<std::string, std::string> SimConfig::getCollContextFilenameAndEventPrefix() const
+{
+  // we decompose the argument to fetch
+  // (a) collision contextfilename
+  // (b) sim prefix to use from the context
+  auto pos = mConfigData.mFromCollisionContext.find(':');
+  std::string collcontextfile{mConfigData.mFromCollisionContext};
+  std::string simprefix{mConfigData.mOutputPrefix};
+  if (pos != std::string::npos) {
+    collcontextfile = mConfigData.mFromCollisionContext.substr(0, pos);
+    simprefix = mConfigData.mFromCollisionContext.substr(pos + 1);
+  }
+  return std::make_pair(collcontextfile, simprefix);
+}
+
 bool SimConfig::resetFromParsedMap(boost::program_options::variables_map const& vm)
 {
   using o2::detectors::DetID;
@@ -333,17 +348,8 @@ bool SimConfig::resetFromParsedMap(boost::program_options::variables_map const&
     mConfigData.mFilterNoHitEvents = true;
   }
   mConfigData.mFromCollisionContext = vm["fromCollContext"].as<std::string>();
-  // we decompose the argument to fetch
-  // (a) collision contextfilename
-  // (b) sim prefix to use from the context
-  auto pos = mConfigData.mFromCollisionContext.find(':');
-  std::string collcontextfile{mConfigData.mFromCollisionContext};
-  std::string simprefix{mConfigData.mOutputPrefix};
-  if (pos != std::string::npos) {
-    collcontextfile = mConfigData.mFromCollisionContext.substr(0, pos);
-    simprefix = mConfigData.mFromCollisionContext.substr(pos + 1);
-  }
-  adjustFromCollContext(collcontextfile, simprefix);
+  auto collcontext_simprefix = getCollContextFilenameAndEventPrefix();
+  adjustFromCollContext(collcontext_simprefix.first, collcontext_simprefix.second);
 
   // analyse vertex options
   if (!parseVertexModeString(vm["vertexMode"].as<std::string>(), mConfigData.mVertexMode)) {

DataFormats/Detectors/TPC/include/DataFormatsTPC/CalibdEdxCorrection.h

@@ -91,8 +91,8 @@ class CalibdEdxCorrection
 
   void clear();
 
-  void writeToFile(std::string_view fileName) const;
-  void loadFromFile(std::string_view fileName);
+  void writeToFile(std::string_view fileName, std::string_view objName = "CalibdEdxCorrection") const;
+  void loadFromFile(std::string_view fileName, std::string_view objName = "CalibdEdxCorrection");
 
   /// \param outFileName name of the output file
   void dumpToTree(const char* outFileName = "calib_dedx.root") const;

DataFormats/Detectors/TPC/src/CalibdEdxCorrection.cxx

@@ -36,16 +36,16 @@ void CalibdEdxCorrection::clear()
   mDims = -1;
 }
 
-void CalibdEdxCorrection::writeToFile(std::string_view fileName) const
+void CalibdEdxCorrection::writeToFile(std::string_view fileName, std::string_view objName) const
 {
   std::unique_ptr<TFile> file(TFile::Open(fileName.data(), "recreate"));
-  file->WriteObject(this, "CalibdEdxCorrection");
+  file->WriteObject(this, objName.data());
 }
 
-void CalibdEdxCorrection::loadFromFile(std::string_view fileName)
+void CalibdEdxCorrection::loadFromFile(std::string_view fileName, std::string_view objName)
 {
   std::unique_ptr<TFile> file(TFile::Open(fileName.data()));
-  auto tmp = file->Get<CalibdEdxCorrection>("CalibdEdxCorrection");
+  auto tmp = file->Get<CalibdEdxCorrection>(objName.data());
   if (tmp != nullptr) {
     *this = *tmp;
   }

DataFormats/Detectors/TRD/src/Tracklet64.cxx

@@ -40,6 +40,16 @@ std::ostream& operator<<(std::ostream& stream, const Tracklet64& trg)
   trg.printStream(stream);
   return stream;
 }
+
+bool operator<(const Tracklet64& lhs, const Tracklet64& rhs)
+{
+  return (lhs.getDetector() < rhs.getDetector()) ||
+         (lhs.getDetector() == rhs.getDetector() && lhs.getROB() < rhs.getROB()) ||
+         (lhs.getDetector() == rhs.getDetector() && lhs.getROB() == rhs.getROB() && lhs.getMCM() < rhs.getMCM()) ||
+         (lhs.getDetector() == rhs.getDetector() && lhs.getROB() == rhs.getROB() && lhs.getMCM() == rhs.getMCM() && lhs.getPadRow() < rhs.getPadRow()) ||
+         (lhs.getDetector() == rhs.getDetector() && lhs.getROB() == rhs.getROB() && lhs.getMCM() == rhs.getMCM() && lhs.getPadRow() == rhs.getPadRow() && lhs.getPadCol() < rhs.getPadCol());
+}
+
 #endif // GPUCA_GPUCODE_DEVICE
 
 } // namespace trd