#include <iostream>
#include <boost/thread.hpp>
#include <boost/chrono.hpp>

#include "triskeleDealThreads.hpp"
#include "TestThread.hpp"

using namespace std;
using namespace boost::chrono;
using namespace triskele;
using namespace otb;
using namespace otb::triskele;

static string timeTypeLabels [TimeTypeCard] = {
  "directDeal",
  "lambdaDeal",
  "threadDeal",
  "inThreadDeal",


  "initStats",
  "seqReadStats",
  "parReadStats",
  "seqWriteStats",
  "parWriteStats",
  "seqRWStats",
  "parRWStats"
};

const unsigned int
TestThread::maxCoreCount = boost::thread::hardware_concurrency ();

// ========================================
inline string
TestThread::ns2string (double delta) {
  ostringstream oss;
  duration<double> ns (delta);
  oss.fill ('0');
  // typedef duration<int, ratio<86400> > days;
  // auto d = duration_cast<days>(ns);
  // ns -= d;
  auto h = duration_cast<hours> (ns);
  ns -= h;
  auto m = duration_cast<minutes> (ns);
  ns -= m;
  oss << setw (2) << h.count () << ":"
      << setw (2) << m.count () << ":"
      << setw (9) << fixed << setprecision (6) << ns.count ();
  return oss.str ();
}

// ========================================
template<typename DimImg, typename FunctId>
inline void
nodealThreadRange (const DimImg &maxId, const unsigned int &coreCount, const FunctId &functId/* functId (id) */) {
  nodealThread (maxId, coreCount, [&functId] (const unsigned int &threadId, const DimImg &minVal, const DimImg &maxVal) {
      for (DimImg id = minVal; id < maxVal; ++id)
	functId (id);
    });
}

// ----------------------------------------
template<typename DimImg, typename FunctThreadId>
inline void
nodealThreadThreadRange (const DimImg &maxId, const unsigned int &coreCount, const FunctThreadId &functThreadId/* functThreadId (threadId, id) */) {
  nodealThread (maxId, coreCount, [&functThreadId] (const unsigned int &threadId, const DimImg &minVal, const DimImg &maxVal) {
      for (DimImg id = minVal; id < maxVal; ++id)
	functThreadId (threadId, id);
    });
}

// ----------------------------------------
template<typename DimImg, typename FunctThreadMinMax>
inline void
nodealThread (const DimImg &maxId, unsigned int coreCount, const FunctThreadMinMax &functThreadMinMax/* functThreadMinMax (threadId, minVal, maxVal) */) {
  //DEF_LOG ("dealThreadBound", "coreCount:" << coreCount << " maxId:" << maxId);
  if (!maxId || !coreCount)
    return;
  if (DimImg (coreCount) > maxId)
    coreCount = (unsigned int) maxId;
  if (coreCount == 1) {
    functThreadMinMax (0, 0, maxId);
    return;
  }
    
  std::vector<DimImg> maxIds = getDealThreadBounds (maxId, coreCount);
  for (unsigned int idCopyValInThread = 0; idCopyValInThread < coreCount; ++idCopyValInThread) {
    functThreadMinMax (idCopyValInThread, maxIds[idCopyValInThread], maxIds[idCopyValInThread+1]);
  }
}

// ========================================
TestThread::TestThread (const unsigned int &coreCount)
  : coreCount (coreCount),
    global (nbItem*coreCount, 0)
{
}

ostream &
TestThread::print (ostream &out, const AlgoStat stats[]) {
  out << endl
      << setw (16) << left << "Time"  << "\t"
      << setw (15) << left << "Sum"   << "\t"
      << setw (3)  << left << "Count" << "\t"
      << setw (15) << left << "Mean"  << "\t"
      << setw (15) << left << "Min"  << "\t"
      << setw (15) << left << "Max"  << endl;
  for (unsigned int i = 0; i < TimeTypeCard; ++i) {
    if (!ba::count (stats[i]))
      continue;
    out << setw (16) << right << timeTypeLabels[i] << "\t"
	<< ns2string (ba::sum (stats[i]))  << "\t" << setw (3) << ba::count (stats[i]) << "\t"
	<< ns2string (ba::mean (stats[i])) << "\t"
	<< ns2string (ba::min (stats[i]))  << "\t"
	<< ns2string (ba::max (stats[i]))
	<< endl << flush;
  }
  return out;
}

template <typename T>
void
TestThread::fillVector (vector<T> &vect) {
  for (size_t i = 0; i < vect.size (); ++i)
    vect[i] = (T) std::rand ();
}

// show algin


void
TestThread::multiTest () {
  for (int i = 0; i < 100; ++i) {
    vector<long> sumSeq (coreCount, 0);
    vector<long> sumPar (coreCount, 0);

    auto start = high_resolution_clock::now ();
    fillVector (global);

    // lecture seq => faire somme
    auto startSeqRead = high_resolution_clock::now ();
    nodealThreadThreadRange (nbItem, coreCount, [this, &sumSeq] (const unsigned int &threadId, const DimImg &item) {
	sumSeq[threadId] += global[item];
      });

    // lecture // => faire somme
    auto startParRead = high_resolution_clock::now ();
    dealThreadThreadRange (nbItem, coreCount, [this, &sumPar] (const unsigned int &threadId, const DimImg &item) {
	sumPar[threadId] += global[item];
      });
    // XXX vérifier égalité de sumSeq sumPar

    // écriture seq => écrire idx
    // écriture // => écrire idx

    // lecture/écriture seq => écrire x/2
    // lecture/écriture // => écrire x/2

    auto end = high_resolution_clock::now ();
    addTime (initStats, duration_cast<duration<double> > (startSeqRead-start).count ());
    addTime (seqReadStats, duration_cast<duration<double> > (startParRead-startSeqRead).count ());
    addTime (parReadStats, duration_cast<duration<double> > (end-startParRead).count ());
  }
  print (cout, timeStats);
}

// ========================================
int
main (int argc, char** argv) {
  cout << "start test" << endl;
  srand (time (NULL));

  TestThread tt;
  //tt.multiTest ();
  for (int i = 0; i < 100; ++i)
    tt.testDeal ();
  tt.print (cout, tt.timeStats);

  return 0;
}

// namespace utils {

//   inline size_t alignSize(size_t size, size_t alignment) {
//     return (size+alignment-1)&~(alignment-1);
//   }

//   template<typename T>
//   inline T * alignPtr(T * ptr, uintptr_t alignment) {
//     union {
//       T *p;
//       uintptr_t u;
//     } u;
//     u.p = ptr;
//     u.u = (u.u+alignment-1)&~(alignment-1);
//     return u.p;
//   }

// }//namespace utils
// void init (Index c, Data const &d) {
//   if (count != c) {
//     kill ();
//     count = c;
//     size_t const alignment = 64;
//     size_t size = alignment-1
//       + utils::alignSize (count*sizeof (Index), alignment)
//       + utils::alignSize (count*sizeof (Rank ), alignment)
//       + utils::alignSize (count*sizeof (Data ), alignment);
//     //memory.reset (new char[size]);
//     delete [] memory;
//     memory = nullptr;
//     memory = new char[size];

//     char *ptr = utils::alignPtr (memory, alignment);
//     parents = reinterpret_cast<Index*> (ptr);
//     ptr += utils::alignSize (count*sizeof (Index), alignment);
//     ranks   = reinterpret_cast<Rank *> (ptr);
//     ptr += utils::alignSize (count*sizeof (Rank ), alignment);
//     datas   = reinterpret_cast<Data *> (ptr);
//   }
//   reset (d);
// }