ReadStats.cxx

Go to the documentation of this file.

/*
  Copyright (C) 2002-2025 CERN for the benefit of the ATLAS collaboration
*/
 
// $Id: ReadStats.cxx 642099 2015-01-27 16:43:18Z krasznaa $
 
// System include(s):
#include <cstring>
#include <functional>
#include <algorithm>
 
// ROOT include(s):
#include <TCollection.h>
#include <TTree.h>
#include <TH1.h>
 
// EDM include(s):
#include "AthContainers/AuxTypeRegistry.h"
#include "AthContainers/normalizedTypeinfoName.h"
 
// Local include(s):
#include "xAODCore/tools/ReadStats.h"
#include "xAODCore/tools/Utils.h"
 
ClassImp( xAOD::BranchStats )
ClassImp( xAOD::ReadStats )
 
namespace {

   bool sortByEntries( const xAOD::BranchStats& b1,
                       const xAOD::BranchStats& b2 ) {
 
      if ( b1.readEntries() > b2.readEntries() ) {
        return true;
      }
      else if ( b1.readEntries() < b2.readEntries() ) {
        return false;
      }
 
      return strcmp (b1.GetName(), b2.GetName()) < 0;
   }

   bool sortByUnzippedBytes( const xAOD::BranchStats& b1,
                             const xAOD::BranchStats& b2 ) {
 
      return ( b1.unzippedBytesRead() > b2.unzippedBytesRead() );
   }

   class SelectByEntries
   {
 
   public:
      SelectByEntries( ::Long64_t entries ) : m_entries( entries ) {}
      SelectByEntries( const xAOD::ReadStats::Map_t& branches,
                       ::Double_t minEvFrac )
         : m_entries( 0 ) {
 
         ::Long64_t maxEntries = 0;
         xAOD::ReadStats::Vector_t::const_iterator vitr, vend;
         xAOD::ReadStats::Map_t::const_iterator itr = branches.begin();
         xAOD::ReadStats::Map_t::const_iterator end = branches.end();
         for( ; itr != end; ++itr ) {
            vitr = itr->second.begin();
            vend = itr->second.end();
            for( ; vitr != vend; ++vitr ) {
               if( ! *vitr ) continue;
               if( ( *vitr )->readEntries() > maxEntries ) {
                  maxEntries = ( *vitr )->readEntries();
               }
            }
         }
 
         m_entries = static_cast< ::Long64_t >( minEvFrac * maxEntries );
      }

      bool operator()( const xAOD::BranchStats& var ) const {
 
         return var.readEntries() >= m_entries;
      }
 
   private:
      ::Long64_t m_entries; 
 
   }; // class SelectByEntries

   class SelectByBytes
   {
 
   public:
      SelectByBytes( ::Long64_t bytes ) : m_bytes( bytes ) {}

      SelectByBytes( const xAOD::ReadStats::Map_t& branches,
                     ::Double_t minByteFrac )
         : m_bytes( 0 ) {
 
         ::Long64_t maxBytes = 0;
         xAOD::ReadStats::Vector_t::const_iterator vitr, vend;
         xAOD::ReadStats::Map_t::const_iterator itr = branches.begin();
         xAOD::ReadStats::Map_t::const_iterator end = branches.end();
         for( ; itr != end; ++itr ) {
            vitr = itr->second.begin();
            vend = itr->second.end();
            for( ; vitr != vend; ++vitr ) {
               if( ! *vitr ) continue;
               if( ( *vitr )->zippedBytesRead() > maxBytes ) {
                  maxBytes = ( *vitr )->zippedBytesRead();
               }
            }
         }
 
         m_bytes = static_cast< ::Long64_t >( minByteFrac * maxBytes );
      }

      bool operator()( const xAOD::BranchStats& var ) const {
 
         return var.zippedBytesRead() >= m_bytes;
      }
 
   private:
      ::Long64_t m_bytes; 
 
   }; // class SelectByBytes

   template< class SELECTOR >
   int addToCache( ::TTree* tree, const xAOD::ReadStats::Map_t& branches,
                   const SELECTOR& selector ) {
 
      int result = 0;
 
      xAOD::ReadStats::Map_t::const_iterator itr = branches.begin();
      xAOD::ReadStats::Map_t::const_iterator end = branches.end();
      for( ; itr != end; ++itr ) {
         for( size_t auxid = 0; auxid < itr->second.size(); ++auxid ) {
            if( ! itr->second[ auxid ] ) continue;
            if( ! selector( *( itr->second[ auxid ] ) ) ) continue;
            if( tree ) {
               const std::string brName = itr->first +
                  SG::AuxTypeRegistry::instance().getName( auxid );
               tree->AddBranchToCache( brName.c_str(), kTRUE );
            }
            ++result;
         }
      }
 
      return result;
   }


   template< class SELECTOR >
   std::vector< std::string >
   getBranches( const xAOD::ReadStats::Map_t& branches,
                const SELECTOR& selector ) {
 
      // The result object:
      std::vector< std::string > result;
 
      // Iterate over all the variables:
      xAOD::ReadStats::Map_t::const_iterator itr = branches.begin();
      xAOD::ReadStats::Map_t::const_iterator end = branches.end();
      for( ; itr != end; ++itr ) {
         for( size_t auxid = 0; auxid < itr->second.size(); ++auxid ) {
            // Check if the object exists:
            if( ! itr->second[ auxid ] ) continue;
            // Ignore the ones failing the selection:
            if( ! selector( *( itr->second[ auxid ] ) ) ) continue;
            // Add this variable to the result:
            const std::string brName = itr->first +
               SG::AuxTypeRegistry::instance().getName( auxid );
            result.push_back( std::move( brName ) );
         }
      }
 
      return result;
   }
 
} // private namespace
 
namespace xAOD {

   BranchStats::BranchStats( const char* name, const char* type,
                             ::Int_t nTreesAccessed, ::Long64_t nReadEntries,
                             ::Long64_t nUnzippedBytes, ::Long64_t nZippedBytes )
      : TNamed( name, type ), m_treesAccessed( nTreesAccessed ),
        m_readEntries( nReadEntries ), m_unzippedBytes( nUnzippedBytes ),
        m_zippedBytes( nZippedBytes ) {
 
   }

   BranchStats::BranchStats( const BranchStats& parent )
      : TNamed( parent ), m_treesAccessed( parent.m_treesAccessed ),
        m_readEntries( parent.m_readEntries ),
        m_unzippedBytes( parent.m_unzippedBytes ),
        m_zippedBytes( parent.m_zippedBytes ) {
 
   }

   BranchStats& BranchStats::operator= ( const BranchStats& parent ) {
 
      // Check for self-assignment:
      if( &parent == this ) {
         return *this;
      }
 
      // Set the properties of TNamed:
      SetName( parent.GetName() );
      SetTitle( parent.GetTitle() );
 
      // Set the properties of this class:
      m_treesAccessed = parent.m_treesAccessed;
      m_readEntries   = parent.m_readEntries;
      m_unzippedBytes = parent.m_unzippedBytes;
      m_zippedBytes   = parent.m_zippedBytes;
 
      return *this;
   }
 
   ::Int_t BranchStats::treesAccessed() const {
 
      return m_treesAccessed;
   }
 
   void BranchStats::setTreesAccessed( ::Int_t nTreesAccessed ) {
 
      m_treesAccessed = nTreesAccessed;
      return;
   }
 
   ::Long64_t BranchStats::readEntries() const {
 
      return m_readEntries;
   }
 
   void BranchStats::setReadEntries( ::Long64_t nReadEntries ) {
 
      m_readEntries = nReadEntries;
      return;
   }
 
   ::Long64_t BranchStats::unzippedBytesRead() const {
 
      return m_unzippedBytes;
   }
 
   void BranchStats::setUnzippedBytesRead( ::Long64_t nUnzippedBytes ) {
 
      m_unzippedBytes = nUnzippedBytes;
      return;
   }
 
   ::Long64_t BranchStats::zippedBytesRead() const {
 
      return m_zippedBytes;
   }
 
   void BranchStats::setZippedBytesRead( ::Long64_t nZippedBytes ) {
 
      m_zippedBytes = nZippedBytes;
      return;
   }
 
   BranchStats& BranchStats::add( const BranchStats& rh ) {
 
      // Set the properties of TNamed:
      SetName( rh.GetName() );
      SetTitle( rh.GetTitle() );
 
      // Accumulate the gathered statistics:
      m_treesAccessed += rh.m_treesAccessed;
      m_readEntries   += rh.m_readEntries;
      m_unzippedBytes += rh.m_unzippedBytes;
      m_zippedBytes   += rh.m_zippedBytes;
 
      return *this;
   }
 
   BranchStats& BranchStats::operator+= ( const BranchStats& rh ) {
 
      return add( rh );
   }

   ::Int_t BranchStats::Merge( ::TCollection* coll ) {
 
      // Some security checks:
      if( ! coll ) return 0;
      if( coll->IsEmpty() ) return 0;
 
      // Iterate over the elements of the collection:
      ::TIter next( coll );
      ::TObject* obj = nullptr;
      ::Int_t result = 0;
      while( ( obj = next() ) ) {
 
         // Check that the element is of the right type:
         BranchStats* vobj = dynamic_cast< BranchStats* >( obj );
         if( ! vobj ) {
            Error( "Merge", "Unknown object type encountered: %s",
                   obj->ClassName() );
            return 0;
         }
 
         // Add this element to this object:
         add( *vobj );
         ++result;
      }
 
      // Return how many objects were merged into this one:
      return result;
   }

   void BranchStats::Print( ::Option_t* option ) const {
 
      // Print the most basic info:
      Info( "Print", "Branch name \"%s\", type \"%s\", read %lli times",
            GetName(), GetTitle(), m_readEntries );
 
      // Print the access statistics only if requested:
      if( ! ::strcmp( option, "All" ) ) {
         Info( "Print", "  TTrees accessed for this branch: %i",
               m_treesAccessed );
         Info( "Print", "  Number of entries read         : %lli",
               m_readEntries );
         Info( "Print", "  Number of unzipped bytes read  : %lli",
               m_unzippedBytes );
         Info( "Print", "  Number of zipped bytes read    : %lli",
               m_zippedBytes );
      }
 
      return;
   }

   ReadStats::ReadStats( const char* name, const char* title )
     : ::TNamed( name, title ), m_branches(), m_containers(),
        m_bytesRead( 0 ), m_branchNum( 0 ),
        m_fileReads( 0 ), m_cacheSize( 0 ), m_readTime( 0.0 ),
        m_unzipTime( 0.0 ), m_processTime( 0.0 ),
        m_nEvents( 0 ), m_nEventsProcessed ( 0 ) {
 
   }

   ReadStats::ReadStats( const ReadStats& parent )
     : ::TNamed( parent ), m_branches( parent.m_branches ),
        m_containers(parent.m_containers),
        m_bytesRead( parent.m_bytesRead ), m_branchNum( parent.m_branchNum ),
        m_fileReads( parent.m_fileReads ), m_cacheSize( parent.m_cacheSize ),
        m_readTime( parent.m_readTime ),
        m_unzipTime( parent.m_unzipTime ),
        m_processTime( parent.m_processTime ), 
         m_nEvents( parent.m_nEvents ),
         m_nEventsProcessed( parent.m_nEventsProcessed ){
 
   }

   ReadStats::~ReadStats() {
 
      Map_t::iterator mitr = m_branches.begin();
      Map_t::iterator mend = m_branches.end();
      for( ; mitr != mend; ++mitr ) {
         Vector_t::iterator vitr = mitr->second.begin();
         Vector_t::iterator vend = mitr->second.end();
         for( ; vitr != vend; ++vitr ) {
            if( ! *vitr ) continue;
            delete *vitr;
         }
      }
   }

   ReadStats& ReadStats::operator= ( const ReadStats& parent ) {
 
      // Check for self-assignment:
      if( &parent == this ) {
         return *this;
      }
 
      // Set the properties of TNamed:
      SetName( parent.GetName() );
      SetTitle( parent.GetTitle() );
 
      // Set the properties of this class:
      m_branches         = parent.m_branches;
      m_containers       = parent.m_containers;
      m_bytesRead        = parent.m_bytesRead;
      m_branchNum        = parent.m_branchNum;
      m_fileReads        = parent.m_fileReads;
      m_cacheSize        = parent.m_cacheSize;
      m_readTime         = parent.m_readTime;
      m_unzipTime        = parent.m_unzipTime;
      m_processTime      = parent.m_processTime;
      m_nEvents          = parent.m_nEvents;
      m_nEventsProcessed = parent.m_nEventsProcessed;
 
      return *this;
   }
 
   void ReadStats::Clear( ::Option_t* ) {
 
      // Clear all accumulated statistics:
      m_branches.clear();
      m_containers.clear();
      m_bytesRead = 0;
      m_branchNum = 0;
      m_fileReads = 0;
      m_cacheSize = 0;
      m_readTime  = 0.0;
      m_unzipTime = 0.0;
      m_processTime = 0.0;
      m_nEvents = 0;
      m_nEventsProcessed = 0;
 
      return;
   }
 
   void ReadStats::setBytesRead( ::Long64_t num ) {
 
      m_bytesRead = num;
      return;
   }
 
   ::Long64_t ReadStats::bytesRead() const {
 
      return m_bytesRead;
   }
 
   void ReadStats::setBranchNum( ::Int_t num ) {
 
      m_branchNum = num;
      return;
   }
 
   ::Int_t ReadStats::branchNum() const {
 
      return m_branchNum;
   }
 
   void ReadStats::setFileReads( ::Int_t num ) {
 
      m_fileReads = num;
      return;
   }
 
   ::Int_t ReadStats::fileReads() const {
 
      return m_fileReads;
   }
 
   void ReadStats::setCacheSize( ::Int_t size ) {
 
      m_cacheSize = size;
      return;
   }
 
   ::Int_t ReadStats::cacheSize() const {
 
      return m_cacheSize;
   }
 
   void ReadStats::setReadTime( ::Double_t time ) {
 
      m_readTime = time;
      return;
   }
 
   ::Double_t ReadStats::readTime() const {
 
      return m_readTime;
   }
 
   void ReadStats::setUnzipTime( ::Double_t time ) {
 
      m_unzipTime = time;
      return;
   }
 
   ::Double_t ReadStats::unzipTime() const {
 
      return m_unzipTime;
   }
 
   void ReadStats::setProcessTime( ::Double_t time ) {
 
      m_processTime = time;
      return;
   }
 
   ::Double_t ReadStats::processTime() const {
 
      return m_processTime;
   }

   BranchStats* ReadStats::branch( const std::string& prefix,
                                   SG::auxid_t auxid ) {
 
      Vector_t& vec = m_branches[ prefix ];
      // Check if it's big enough:
      if( vec.size() <= auxid ) {
         vec.resize( auxid + 1 );
      }
      // Check if the object exists already:
      if( ! vec[ auxid ] ) {
         // Construct its name, mangling needed for dynamic variables:
         const std::string brName = prefix +
            SG::AuxTypeRegistry::instance().getName( auxid );
         const std::type_info* brType =
            SG::AuxTypeRegistry::instance().getType( auxid );
         if( ! brType ) {
            Error( "branch", "Coudln't find type_info for aux ID %i",
                   static_cast< int >( auxid ) );
            return nullptr;
         }
         const std::string brTypeName = SG::normalizedTypeinfoName( *brType );
 
         // Construct the new object:
         vec[ auxid ] = new BranchStats( brName.c_str(), brTypeName.c_str() );
      }
 
      // Return the object:
      return vec[ auxid ];
   }

   const BranchStats* ReadStats::branch( const std::string& prefix,
                                         SG::auxid_t auxid ) const {
 
      // Search for the variable:
      Map_t::const_iterator itr = m_branches.find( prefix );
      if( itr != m_branches.end() ) {
         const Vector_t& vec = itr->second;
         if( ( vec.size() > auxid ) && vec[ auxid ] ) {
            return vec[ auxid ];
         }
      }
 
      // Return a null pointer if the object was not found:
      return nullptr;
   }
 
   BranchStats* ReadStats::container( const std::string& name ) {
 
      // If it doesn't exist yet, create it now:
      auto ret = m_containers.try_emplace (name, name.c_str(), "CONTAINER");
      // Return a pointer to the object:
      return &ret.first->second;
   }
 
   const BranchStats* ReadStats::container( const std::string& name ) const {
 
      // Try to find it:
      MapC_t::const_iterator itr = m_containers.find( name );
      if( itr != m_containers.end() ) {
         return &( itr->second );
      }
 
      // We didn't find it:
      return nullptr;
   }
 
   const ReadStats::Map_t& ReadStats::branches() const {
 
      return m_branches;
   }
 
   const ReadStats::MapC_t& ReadStats::containers() const {
      
      return m_containers;
   }

   bool ReadStats::isCompatible( const ReadStats& rh ) const {
 
      return ( m_branchNum == rh.branchNum() ) &&
 
          ( m_cacheSize == rh.cacheSize() );
   }

   ReadStats& ReadStats::add( const ReadStats& rh ) {
 
      // Merge the variable statistics one by one:
      Map_t::const_iterator itr = rh.m_branches.begin();
      Map_t::const_iterator end = rh.m_branches.end();
      for( ; itr != end; ++itr ) {
         const Vector_t& rhvec = itr->second;
         Vector_t& vec = m_branches[ itr->first ];
         if( vec.size() < rhvec.size() ) {
            vec.resize( rhvec.size() );
         }
         for( size_t auxid = 0; auxid < rhvec.size(); ++auxid ) {
            if( ! rhvec[ auxid ] ) continue;
            if( ! vec[ auxid ] ) {
               vec[ auxid ] = new BranchStats();
            }
            vec[ auxid ]->add( *( rhvec[ auxid ] ) );
         }
      }
 
      // Sum up the simple statistics:
      m_bytesRead   += rh.m_bytesRead;
      m_fileReads   += rh.m_fileReads;
      m_readTime    += rh.m_readTime;
      m_unzipTime   += rh.m_unzipTime;
      m_processTime += rh.m_processTime;
 
      return *this;
   }
 
   ReadStats& ReadStats::operator+= ( const ReadStats& rh ) {
 
      return add( rh );
   }

   void ReadStats::addToTreeCache( ::TTree* tree ) const {
 
      // Add all the branches to the cache:
      const int counter = addToCache( tree, m_branches,
                                      SelectByEntries( 0 ) );
 
      // Let the user know what we did:
      Info( "addToTreeCache", "Added %i branches to the TTreeCache",
            counter );
 
      return;
   }

   void ReadStats::addToTreeCacheByEntries( ::TTree* tree,
                                            ::Long64_t minEntries ) const {
 
      // Add the selected branches to the cache:
      const int counter = addToCache( tree, m_branches,
                                      SelectByEntries( minEntries ) );
 
      // Let the user know what we did:
      Info( "addToTreeCacheByEntries", "Added %i branches to the TTreeCache",
            counter );
 
      return;
   }

   void ReadStats::addToTreeCacheByEntryFrac( ::TTree* tree,
                                              ::Double_t minEvFraction ) const {
 
      // Add the selected branches to the cache:
      const int counter = addToCache( tree, m_branches,
                                      SelectByEntries( m_branches,
                                                       minEvFraction ) );
 
      // Let the user know what we did:
      Info( "addToTreeCacheByEntries", "Added %i branches to the TTreeCache",
            counter );
 
      return;
   }

   void ReadStats::addToTreeCacheByBytes( ::TTree* tree,
                                          ::Long64_t minBytes ) const {
 
      // Add the selected branches to the cache:
      const int counter = addToCache( tree, m_branches,
                                      SelectByBytes( minBytes ) );
 
      // Let the user know what we did:
      Info( "addToTreeCacheByBytes", "Added %i branches to the TTreeCache",
            counter );
 
      return;
   }

   void
   ReadStats::addToTreeCacheByByteFrac( ::TTree* tree,
                                        ::Double_t minByteFraction ) const {
 
      // Add the selected branches to the cache:
      const int counter = addToCache( tree, m_branches,
                                      SelectByBytes( m_branches,
                                                     minByteFraction ) );
 
      // Let the user know what we did:
      Info( "addToTreeCacheByBytes", "Added %i branches to the TTreeCache",
            counter );
 
      return;
   }

   std::vector< std::string >
   ReadStats::branchesByEntries( ::Long64_t minEntries ) const {
 
      return getBranches( m_branches, SelectByEntries( minEntries ) );
   }

   std::vector< std::string >
   ReadStats::branchesByEntryFrac( ::Double_t minEvFrac ) const {
 
      return getBranches( m_branches,
                          SelectByEntries( m_branches, minEvFrac ) );
   }

   std::vector< std::string >
   ReadStats::branchesByBytes( ::Long64_t minBytes ) const {
 
      return getBranches( m_branches, SelectByBytes( minBytes ) );
   }

   std::vector< std::string >
   ReadStats::branchesByByteFrac( ::Double_t minByteFraction ) const {
 
      return getBranches( m_branches,
                          SelectByBytes( m_branches, minByteFraction ) );
   }

   ::TH1* ReadStats::treeAccessStat() const {
 
      // Find the branch(es) which accessed the most trees:
      ::Long64_t maxTrees = 0;
      Vector_t::const_iterator vitr, vend;
      Map_t::const_iterator itr = m_branches.begin();
      Map_t::const_iterator end = m_branches.end();
      for( ; itr != end; ++itr ) {
         vitr = itr->second.begin();
         vend = itr->second.end();
         for( ; vitr != vend; ++vitr ) {
            if( ! *vitr ) continue;
            if( ( *vitr )->treesAccessed() > maxTrees ) {
               maxTrees = ( *vitr )->treesAccessed();
            }
         }
      }
 
      // Create the histogram:
      ::TH1* result = new ::TH1D( "TreeAccessStat",
                                  "TTree access statistics;TTrees;Branches",
                                  100, 0.0, ( ( ::Double_t ) maxTrees ) * 1.1 );
 
      // Fill the histogram by looping over the variables once more:
      itr = m_branches.begin();
      end = m_branches.end();
      for( ; itr != end; ++itr ) {
         vitr = itr->second.begin();
         vend = itr->second.end();
         for( ; vitr != vend; ++vitr ) {
            if( ! *vitr ) continue;
            result->Fill( ( ::Double_t ) ( *vitr )->treesAccessed() );
         }
      }
 
      return result;
   }

   ::TH1* ReadStats::entryReadStat() const {
 
      // Find the branch(es) which accessed the most entries:
      ::Long64_t maxEntries = 0;
      Vector_t::const_iterator vitr, vend;
      Map_t::const_iterator itr = m_branches.begin();
      Map_t::const_iterator end = m_branches.end();
      for( ; itr != end; ++itr ) {
         vitr = itr->second.begin();
         vend = itr->second.end();
         for( ; vitr != vend; ++vitr ) {
            if( ! *vitr ) continue;
            if( ( *vitr )->readEntries() > maxEntries ) {
               maxEntries = ( *vitr )->readEntries();
            }
         }
      }
 
      // Create the histogram:
      ::TH1* result = new ::TH1D( "EntryAccessStat",
                                  "Entry access statistics;Entries;Branches",
                                  100, 0.0,
                                  ( ( ::Double_t ) maxEntries ) * 1.1 );
 
      // Fill the histogram by looping over the variables once more:
      itr = m_branches.begin();
      end = m_branches.end();
      for( ; itr != end; ++itr ) {
         vitr = itr->second.begin();
         vend = itr->second.end();
         for( ; vitr != vend; ++vitr ) {
            if( ! *vitr ) continue;
            result->Fill( ( ::Double_t ) ( *vitr )->readEntries() );
         }
      }
 
      return result;
   }

   ::TH1* ReadStats::zippedByteReadStat() const {
 
      // Find the branch(es) which accessed the most bytes:
      ::Long64_t maxBytes = 0;
      Vector_t::const_iterator vitr, vend;
      Map_t::const_iterator itr = m_branches.begin();
      Map_t::const_iterator end = m_branches.end();
      for( ; itr != end; ++itr ) {
         vitr = itr->second.begin();
         vend = itr->second.end();
         for( ; vitr != vend; ++vitr ) {
            if( ! *vitr ) continue;
            if( ( *vitr )->zippedBytesRead() > maxBytes ) {
               maxBytes = ( *vitr )->zippedBytesRead();
            }
         }
      }
 
      // Create the histogram:
      ::TH1* result = new ::TH1D( "ZippedByteAccessStat",
                                  "Zipped byte access statistics;Bytes;Branches",
                                  100, 0.0, ( ( ::Double_t ) maxBytes ) * 1.1 );
 
      // Fill the histogram by looping over the variables once more:
      itr = m_branches.begin();
      end = m_branches.end();
      for( ; itr != end; ++itr ) {
         vitr = itr->second.begin();
         vend = itr->second.end();
         for( ; vitr != vend; ++vitr ) {
            if( ! *vitr ) continue;
            result->Fill( ( ::Double_t ) ( *vitr )->zippedBytesRead() );
         }
      }
 
      return result;
   }

   ::TH1* ReadStats::unzippedByteReadStat() const {
 
      // Find the branch(es) which accessed the most bytes:
      ::Long64_t maxBytes = 0;
      Vector_t::const_iterator vitr, vend;
      Map_t::const_iterator itr = m_branches.begin();
      Map_t::const_iterator end = m_branches.end();
      for( ; itr != end; ++itr ) {
         vitr = itr->second.begin();
         vend = itr->second.end();
         for( ; vitr != vend; ++vitr ) {
            if( ! *vitr ) continue;
            if( ( *vitr )->unzippedBytesRead() > maxBytes ) {
               maxBytes = ( *vitr )->unzippedBytesRead();
            }
         }
      }
 
      // Create the histogram:
      ::TH1* result =
           new ::TH1D( "UnzippedByteAccessStat",
                       "Unzipped byte access statistics;Bytes;Branches",
                       100, 0.0, ( ( ::Double_t ) maxBytes ) * 1.1 );
 
      // Fill the histogram by looping over the variables once more:
      itr = m_branches.begin();
      end = m_branches.end();
      for( ; itr != end; ++itr ) {
         vitr = itr->second.begin();
         vend = itr->second.end();
         for( ; vitr != vend; ++vitr ) {
            if( ! *vitr ) continue;
            result->Fill( ( ::Double_t ) ( *vitr )->unzippedBytesRead() );
         }
      }
 
      return result;
   }

   ::Int_t ReadStats::Merge( ::TCollection* coll ) {
 
      // Some security checks:
      if( ! coll ) return 0;
      if( coll->IsEmpty() ) return 0;
 
      // Iterate over the elements of the collection:
      ::TIter next( coll );
      ::TObject* obj = nullptr;
      ::Int_t result = 0;
      while( ( obj = next() ) ) {
 
         // Check that the element is of the right type:
         ReadStats* dobj = dynamic_cast< ReadStats* >( obj );
         if( ! dobj ) {
            Error( "Merge", "Unknown object type encountered: %s",
                   obj->ClassName() );
            return 0;
         }
 
         // The compatibility of the objects is no longer checked. When
         // processing a large dataset, it's probable that the objects
         // created by the different PROOF workers will not be "compatible".
 
         // Add this element to this object:
         add( *dobj );
         ++result;
      }
 
      return result;
   }

   void ReadStats::Print( ::Option_t* option ) const {
 
      Info( "Print", "Printing xAOD I/O statistics" );
 
      // Create a temporary vector of the containers, so they can be ordered
      // if necessary:
      std::vector< BranchStats > conts;
      MapC_t::const_iterator cont_itr = m_containers.begin();
      MapC_t::const_iterator cont_end = m_containers.end();
      for( ; cont_itr != cont_end; ++cont_itr ) {
         conts.push_back( cont_itr->second );
      }
 
      // Create a temporary vector of the branches, so they can be ordered
      // if necessary:
      std::vector< BranchStats > vars;
      Vector_t::const_iterator vec_itr, vec_end;
      Map_t::const_iterator map_itr = m_branches.begin();
      Map_t::const_iterator map_end = m_branches.end();
      for( ; map_itr != map_end; ++map_itr ) {
         vec_itr = map_itr->second.begin();
         vec_end = map_itr->second.end();
         for( ; vec_itr != vec_end; ++vec_itr ) {
            if( ! *vec_itr ) continue;
            vars.push_back( **vec_itr );
         }
      }
 
      // Container and branch access summary
      if( ! ::strcmp( option, "SmartSlimming" ) ) {
 
         Info( "Print", " " );
         Info( "Print", "Smart Slimming Statistics" );
         Info( "Print", " " );
 
         const Double_t proFrac = ( ( Double_t ) m_nEventsProcessed /
                                    ( Double_t ) m_nEvents ) * 100.0;
         Info( "Print", "  Processed %lli events from %lli (%g%%)",
               m_nEventsProcessed, m_nEvents, proFrac );
 
         Info( "Print", "  Number of containers in on the input: %i",
               static_cast< int >( m_containers.size() ) );
         Info( "Print", "  Number of branches on the input: %i",
               static_cast< int >( vars.size() ) );
 
         // Sort the containers by number of accesses:
         std::sort( conts.begin(), conts.end(), sortByEntries );
 
         // Print which containers got accessed:
         Info( "Print", " " );
         Info( "Print", "Accessed containers:" );
         Info( "Print", " " );
         std::vector< BranchStats >::const_iterator itr = conts.begin();
         std::vector< BranchStats >::const_iterator end = conts.end();
         for( ; itr != end; ++itr ) {
            if( ! itr->readEntries() ) continue;
            itr->Print( option );
         }
 
         // Sort the branches by number of accesses:
         std::sort( vars.begin(), vars.end(), sortByEntries );
 
         // Print which branches got accessed:
         Info( "Print", " " );
         Info( "Print", "Accessed branches:" );
         Info( "Print", " " );
         itr = vars.begin();
         end = vars.end();
         for( ; itr != end; ++itr ) {
            if( ! itr->readEntries() ) continue;
            itr->Print( option );
         }
         Info( "Print", " " );
 
         // Let's exit at this point:
         return;
      }
 
      // Print the summary information:
      Info( "Print", "  Number of variables in the input xAOD : %i",
            m_branchNum );
      const Double_t proFrac = ( ( Double_t ) m_nEventsProcessed /
                                 ( Double_t ) m_nEvents ) * 100.0;
      Info( "Print", "  Processed events                      : "
            "%lli/%lli (%g%%)", m_nEventsProcessed, m_nEvents, proFrac );
      Info( "Print", "  TTreeCache size used                  : %s",
            Utils::sizeToString( m_cacheSize ).c_str() );
      Info( "Print", "  Total number of bytes read            : %s",
            Utils::sizeToString( m_bytesRead ).c_str() );
      Info( "Print", "  Data reading speed per process        : %s",
            ( std::abs( m_processTime ) > 0.0001 ?
              Utils::speedToString( m_bytesRead / m_processTime ).c_str() :
              "N/A" ) );
      Info( "Print", "  Total number of file read operations  : %i",
            m_fileReads );
      const ::Long64_t readInOneGo =
         ( ::Long64_t ) ( ( ::Double_t ) m_bytesRead /
                          ( ::Double_t ) m_fileReads );
      Info( "Print", "  Data read in one go (on average)      : %s",
            Utils::sizeToString( readInOneGo ).c_str() );
      Info( "Print", "  Cumulative time spent processing data : %s (%s/event)",
            Utils::timeToString( m_processTime ).c_str(),
            ( m_nEventsProcessed ?
              Utils::timeToString( m_processTime / m_nEventsProcessed ).c_str() :
              "N/A" ) );
      Info( "Print", "  Cumulative time spent reading data    : %s (%s/event)",
            Utils::timeToString( m_readTime ).c_str(),
            ( m_nEventsProcessed ?
              Utils::timeToString( m_readTime / m_nEventsProcessed ).c_str() :
              "N/A" ) );
      Info( "Print", "  Cumulative time spent unzipping data  : %s (%s/event)",
            Utils::timeToString( m_unzipTime ).c_str(),
            ( m_nEventsProcessed ?
              Utils::timeToString( m_unzipTime / m_nEventsProcessed ).c_str() :
              "N/A" ) );
 
      // If we just needed summary information, stop here:
      if( ! ::strcmp( option, "Summary" ) ) {
         return;
      }
 
      // Select the kind of ordering for the containers:
      if( ! ::strcmp( option, "ByEntries" ) ) {
         Info( "Print", "Containers, sorted by number of accesses:" );
         std::sort( conts.begin(), conts.end(), sortByEntries );
      } else if( ! ::strcmp( option, "ByBytes" ) ) {
         Info( "Print", "Containers, sorted by number of bytes read:" );
         std::sort( conts.begin(), conts.end(), sortByUnzippedBytes );
      } else {
         Info( "Print", "Containers, sorted by name:" );
      }
 
      // Print the statistics from each container:
      std::vector< BranchStats >::const_iterator itr = conts.begin();
      std::vector< BranchStats >::const_iterator end = conts.end();
      for( ; itr != end; ++itr ) {
         itr->Print( option );
      }
 
      // Select the kind of ordering for the variables:
      if( ! ::strcmp( option, "ByEntries" ) ) {
         Info( "Print", "Variables, sorted by number of accesses:" );
         std::sort( vars.begin(), vars.end(), sortByEntries );
      } else if( ! ::strcmp( option, "ByBytes" ) ) {
         Info( "Print", "Variables, sorted by number of bytes read:" );
         std::sort( vars.begin(), vars.end(), sortByUnzippedBytes );
      } else {
         Info( "Print", "Variables, sorted by name:" );
      }
 
      // Print the statistics from each variable:
      itr = vars.begin();
      end = vars.end();
      for( ; itr != end; ++itr ) {
         itr->Print( option );
      }
 
      return;
   }

   void ReadStats::printSmartSlimmingBranchList( bool autoIncludeLinks ) const {

      std::map< ::TString, ::TString > items;
 
      // Collect all the containers that were accessed during the job:
      MapC_t::const_iterator cont_itr = m_containers.begin();
      MapC_t::const_iterator cont_end = m_containers.end();
      for( ; cont_itr != cont_end; ++cont_itr ) {
         // Skip non-accessed containers:
         if( ! cont_itr->second.readEntries() ) continue;
         // Remember the container:
         items[ cont_itr->first ] = "";
      }
 
      // Now look for variables in all these containers:
      Map_t::const_iterator br_itr = m_branches.begin();
      Map_t::const_iterator br_end = m_branches.end();
      for( ; br_itr != br_end; ++br_itr ) {
         // First check if any variables got accessed. It has an effect on what
         // to do with the ElementLink types.
         bool containerAccessed = false;
         Vector_t::const_iterator itr = br_itr->second.begin();
         Vector_t::const_iterator end = br_itr->second.end();
         for( ; itr != end; ++itr ) {
            if( ( *itr ) && ( *itr )->readEntries() ) {
               containerAccessed = true;
               break;
            }
         }
         // If none were accessed, abandon this container:
         if( ! containerAccessed ) continue;
         // Now collect all the variables:
         for( ; itr != end; ++itr ) {
            // Check if it's an ElementLink type:
            const bool linkType =
               ( ( *itr ) &&
                 ::TString( ( *itr )->GetTitle() ).Contains( "ElementLink" ) );
            // Skip non-existent, non-accessed, and not requested link
            // variables:
            if( ( ( ! *itr ) || ( ! ( *itr )->readEntries() ) ) &&
                ( ! ( autoIncludeLinks && linkType ) ) ) continue;
            // Extract the name of the container and the variable from the
            // branch's name:
            const ::TString brname = ( *itr )->GetName();
            const ::Size_t dotPos = brname.First( '.' );
            if( dotPos == ::kNPOS ) {
               // Ignore the unknown containers:
               continue;
            }
            const ::TString cname = brname( 0, dotPos + 1 );
            const ::TString vname = brname( dotPos + 1,
                                            brname.Length() - dotPos );
            // Access the current variable list:
            ::TString& vars = items[ cname ];
            // Add a dot if there are already variables defined:
            if( vars.Length() ) {
               vars.Append( '.' );
            }
            // Add this variable:
            vars.Append( vname );
         }
      }
 
      // Print the collected information:
      Info( "printSmartSlimmingBranchList", "ItemList to use:" );
      if( autoIncludeLinks ) {
         Info( "printSmartSlimmingBranchList", "(All link variables added)" );
      }
      std::map< ::TString, ::TString >::const_iterator itr = items.begin();
      std::map< ::TString, ::TString >::const_iterator end = items.end();
      for( ; itr != end; ++itr ) {
         if( itr->first.EndsWith( "Aux." ) ) {
            Info( "printSmartSlimmingBranchList", "  %s%s",
                  itr->first.Data(),
                  ( itr->second.Length() ? itr->second.Data() : "-" ) );
         } else {
            Info( "printSmartSlimmingBranchList", "  %s",
                  itr->first.Data() );
         }
      }
 
      // Return gracefully:
      return;
   }
 
   void ReadStats::nextEvent(){
 
      // event counters
      ++m_nEventsProcessed;
      return;
   }
 
   ::Long64_t ReadStats::nEvents() const {
 
      return m_nEvents;
   }
 
   void ReadStats::setNEvents( ::Long64_t nevents ) {
 
      m_nEvents = nevents;
      return;
   }
 
   void ReadStats::readBranch( const std::string& prefix,
                               SG::auxid_t auxid ){
 
      // Access the branch:
      BranchStats* stat = branch( prefix, auxid );
      // Increment its access counter:
      stat->setReadEntries( stat->readEntries() + 1 );
      return;
   }
 
   void ReadStats::readContainer( const std::string& name ){
 
      // Access the branch:
      BranchStats* stat = container( name );
      // Increment its access counter:
      stat->setReadEntries( stat->readEntries() + 1 );
      return;
   }
 
} // namespace D3PDReader