Merge branch 'master' into master

Author: dajones2

ALICE3/DataModel/OTFMulticharm.h

@@ -40,6 +40,8 @@ DECLARE_SOA_COLUMN(XiccMass, xiccMass, float);
 // kine vars
 DECLARE_SOA_COLUMN(XiccPt, xiccPt, float);
 DECLARE_SOA_COLUMN(XiccEta, xiccEta, float);
+DECLARE_SOA_COLUMN(XicPt, xicPt, float);
+DECLARE_SOA_COLUMN(XicEta, xicEta, float);
 
 // topo vars
 DECLARE_SOA_COLUMN(XiDCAz, xiDCAz, float);
@@ -122,12 +124,15 @@ DECLARE_SOA_TABLE(MCharmIndices, "AOD", "MCharmIndices",
                   otfmulticharm::XiCCPionId);
 
 DECLARE_SOA_TABLE(MCharmCores, "AOD", "MCharmCores",
-                  otfmulticharm::XicDauDCA,
-                  otfmulticharm::XiccDauDCA,
-                  otfmulticharm::XicMass,
                   otfmulticharm::XiccMass,
                   otfmulticharm::XiccPt,
                   otfmulticharm::XiccEta,
+                  otfmulticharm::XiccDauDCA,
+
+                  otfmulticharm::XicMass,
+                  otfmulticharm::XicPt,
+                  otfmulticharm::XicEta,
+                  otfmulticharm::XicDauDCA,
 
                   otfmulticharm::XiDCAxy,
                   otfmulticharm::XiDCAz,

ALICE3/TableProducer/alice3-multicharmTable.cxx

@@ -734,9 +734,10 @@ struct alice3multicharmTable {
                 picc.globalIndex());
 
               multiCharmCore(
-                thisXiCcandidate.dca, thisXiCCcandidate.dca,
-                thisXiCcandidate.mass, thisXiCCcandidate.mass,
-                thisXiCCcandidate.pt, thisXiCCcandidate.eta,
+                thisXiCCcandidate.mass, thisXiCCcandidate.pt,
+                thisXiCCcandidate.eta, thisXiCCcandidate.dca,
+                thisXiCcandidate.mass, thisXiCcandidate.pt,
+                thisXiCcandidate.eta, thisXiCcandidate.dca,
                 xi.dcaXY(), xi.dcaZ(),
                 xicdcaXY, xicdcaZ,
                 xiccdcaXY, xiccdcaZ,
@@ -756,12 +757,10 @@ struct alice3multicharmTable {
                 pi1cTOFDiffOuter, pi1c.nSigmaPionOuterTOF(),
                 pi1c.hasSigPi(), pi1c.nSigmaPionRich(),
                 getPdgCodeForTrack(pi1c),
-
                 pi2cTOFDiffInner, pi2c.nSigmaPionInnerTOF(),
                 pi2cTOFDiffOuter, pi2c.nSigmaPionOuterTOF(),
                 pi2c.hasSigPi(), pi2c.nSigmaPionRich(),
                 getPdgCodeForTrack(pi2c),
-
                 piccTOFDiffInner, picc.nSigmaPionInnerTOF(),
                 piccTOFDiffOuter, picc.nSigmaPionOuterTOF(),
                 picc.hasSigPi(), picc.nSigmaPionRich(),
@@ -774,9 +773,7 @@ struct alice3multicharmTable {
                 prFromLa.dcaXY(), prFromLa.dcaZ(),
                 piFromLa.pt(), piFromLa.eta(),
                 piFromLa.dcaXY(), piFromLa.dcaZ(),
-                pi1c.eta(),
-                pi2c.eta(),
-                picc.eta());
+                pi1c.eta(), pi2c.eta(), picc.eta());
             }
           }
           histos.fill(HIST("hCombinationsXiCC"), nCombinationsCC);

ALICE3/Tasks/CMakeLists.txt

@@ -61,7 +61,7 @@ o2physics_add_dpl_workflow(alice3-taskcorrelationddbar
 
 o2physics_add_dpl_workflow(alice3-multicharm
                     SOURCES alice3-multicharm.cxx
-                    PUBLIC_LINK_LIBRARIES O2Physics::AnalysisCore
+                    PUBLIC_LINK_LIBRARIES O2Physics::AnalysisCore O2Physics::MLCore
                     COMPONENT_NAME Analysis)
 
 o2physics_add_dpl_workflow(alice3-efficiency

ALICE3/Tasks/alice3-multicharm.cxx

@@ -58,7 +58,7 @@
 /PWGMM/Lumi @alibuild @aalkin @jgcn
 /PWGMM/UE   @alibuild @aalkin @aortizve @jgcn
 
-/PWGUD @alibuild @pbuehler @abylinkin @rolavick
+/PWGUD @alibuild @pbuehler @nystrand @rolavick
 /PWGJE @alibuild @lhavener @maoyx @nzardosh @fjonasALICE @mfasDa @mhemmer-cern
 /Tools/PIDML @alibuild @saganatt
 /Tools/ML @alibuild @fcatalan92 @fmazzasc

CODEOWNERS

@@ -125,6 +125,7 @@ DECLARE_SOA_TABLE(MFTMults, "AOD", "MFTMULT", //! Multiplicity with MFT
                   mult::MFTNalltracks, mult::MFTNtracks);
 using BarrelMults = soa::Join<TrackletMults, TPCMults, PVMults>;
 using Mults = soa::Join<BarrelMults, FV0Mults, FT0Mults, FDDMults, ZDCMults>;
+using MultsRun3 = soa::Join<TPCMults, PVMults, FV0Mults, FT0Mults, FDDMults, ZDCMults>;
 using FT0Mult = FT0Mults::iterator;
 using MFTMult = MFTMults::iterator;
 using Mult = Mults::iterator;

Common/DataModel/Multiplicity.h

@@ -150,9 +150,11 @@ struct tpcPid {
   Configurable<int> useNetworkHe{"useNetworkHe", 1, {"Switch for applying neural network on the helium3 mass hypothesis (if network enabled) (set to 0 to disable)"}};
   Configurable<int> useNetworkAl{"useNetworkAl", 1, {"Switch for applying neural network on the alpha mass hypothesis (if network enabled) (set to 0 to disable)"}};
   Configurable<float> networkBetaGammaCutoff{"networkBetaGammaCutoff", 0.45, {"Lower value of beta-gamma to override the NN application"}};
+  Configurable<float> networkInputBatchedMode{"networkInputBatchedMode", -1, {"-1: Takes all tracks, >0: Takes networkInputBatchedMode number of tracks at once"}};
 
   // Parametrization configuration
   bool useCCDBParam = false;
+  std::vector<float> track_properties;
 
   void init(o2::framework::InitContext& initContext)
   {
@@ -298,8 +300,6 @@ struct tpcPid {
   std::vector<float> createNetworkPrediction(C const& collisions, T const& tracks, B const& bcs, const size_t size)
   {
 
-    std::vector<float> network_prediction;
-
     auto start_network_total = std::chrono::high_resolution_clock::now();
     if (autofetchNetworks) {
       const auto& bc = bcs.begin();
@@ -345,20 +345,24 @@ struct tpcPid {
     // Defining some network parameters
     int input_dimensions = network.getNumInputNodes();
     int output_dimensions = network.getNumOutputNodes();
-    const uint64_t track_prop_size = input_dimensions * size;
     const uint64_t prediction_size = output_dimensions * size;
+    const uint8_t numSpecies = 9;
+    const uint64_t total_eval_size = size * numSpecies; // 9 species
 
-    network_prediction = std::vector<float>(prediction_size * 9); // For each mass hypotheses
     const float nNclNormalization = response->GetNClNormalization();
     float duration_network = 0;
 
-    std::vector<float> track_properties(track_prop_size);
-    uint64_t counter_track_props = 0;
-    int loop_counter = 0;
+    uint64_t counter_track_props = 0, exec_counter = 0, in_batch_counter = 0, total_input_count = 0;
+    uint64_t track_prop_size = networkInputBatchedMode.value;
+    if (networkInputBatchedMode.value <= 0) {
+      track_prop_size = size; // If the networkInputBatchedMode is not set, we use all tracks at once
+    }
+    track_properties.resize(track_prop_size * input_dimensions);         // If the networkInputBatchedMode is set, we use the number of tracks specified in the config
+    std::vector<float> network_prediction(prediction_size * numSpecies); // For each mass hypotheses
 
     // Filling a std::vector<float> to be evaluated by the network
     // Evaluation on single tracks brings huge overhead: Thus evaluation is done on one large vector
-    for (int i = 0; i < 9; i++) { // Loop over particle number for which network correction is used
+    for (int species = 0; species < numSpecies; species++) { // Loop over particle number for which network correction is used
       for (auto const& trk : tracks) {
         if (!trk.hasTPC()) {
           continue;
@@ -368,30 +372,38 @@ struct tpcPid {
             continue;
           }
         }
-        track_properties[counter_track_props] = trk.tpcInnerParam();
+
+        if ((in_batch_counter == track_prop_size) || (total_input_count == total_eval_size)) { // If the batch size is reached, reset the counter
+          int32_t fill_shift = (exec_counter * track_prop_size - ((total_input_count == total_eval_size) ? (total_input_count % track_prop_size) : 0)) * output_dimensions;
+          auto start_network_eval = std::chrono::high_resolution_clock::now();
+          float* output_network = network.evalModel(track_properties);
+          auto stop_network_eval = std::chrono::high_resolution_clock::now();
+          duration_network += std::chrono::duration<float, std::ratio<1, 1000000000>>(stop_network_eval - start_network_eval).count();
+
+          for (uint64_t i = 0; i < (in_batch_counter * output_dimensions); i += output_dimensions) {
+            for (int j = 0; j < output_dimensions; j++) {
+              network_prediction[i + j + fill_shift] = output_network[i + j];
+            }
+          }
+          counter_track_props = 0;
+          in_batch_counter = 0;
+          exec_counter++;
+        }
+
+        // LOG(info) << "counter_tracks_props: " << counter_track_props << "; in_batch_counter: " << in_batch_counter << "; total_input_count: " << total_input_count << "; exec_counter: " << exec_counter << "; track_prop_size: " << track_prop_size << "; size: " << size << "; track_properties.size(): " << track_properties.size();
+        track_properties[counter_track_props] = trk.tpcInnerParam(); // (tracks.asArrowTable()->GetColumn<float>("tpcInnerParam")).GetData();
         track_properties[counter_track_props + 1] = trk.tgl();
         track_properties[counter_track_props + 2] = trk.signed1Pt();
-        track_properties[counter_track_props + 3] = o2::track::pid_constants::sMasses[i];
+        track_properties[counter_track_props + 3] = o2::track::pid_constants::sMasses[species];
         track_properties[counter_track_props + 4] = trk.has_collision() ? collisions.iteratorAt(trk.collisionId()).multTPC() / 11000. : 1.;
         track_properties[counter_track_props + 5] = std::sqrt(nNclNormalization / trk.tpcNClsFound());
         if (input_dimensions == 7 && networkVersion == "2") {
           track_properties[counter_track_props + 6] = trk.has_collision() ? collisions.iteratorAt(trk.collisionId()).ft0cOccupancyInTimeRange() / 60000. : 1.;
         }
         counter_track_props += input_dimensions;
+        in_batch_counter++;
+        total_input_count++;
       }
-
-      auto start_network_eval = std::chrono::high_resolution_clock::now();
-      float* output_network = network.evalModel(track_properties);
-      auto stop_network_eval = std::chrono::high_resolution_clock::now();
-      duration_network += std::chrono::duration<float, std::ratio<1, 1000000000>>(stop_network_eval - start_network_eval).count();
-      for (uint64_t i = 0; i < prediction_size; i += output_dimensions) {
-        for (int j = 0; j < output_dimensions; j++) {
-          network_prediction[i + j + prediction_size * loop_counter] = output_network[i + j];
-        }
-      }
-
-      counter_track_props = 0;
-      loop_counter += 1;
     }
     track_properties.clear();