DefaultEventLoader.cpp

Go to the documentation of this file.

// Mantid Repository : https://github.com/mantidproject/mantid
//
// Copyright © 2018 ISIS Rutherford Appleton Laboratory UKRI,
//   NScD Oak Ridge National Laboratory, European Spallation Source,
//   Institut Laue - Langevin & CSNS, Institute of High Energy Physics, CAS
// SPDX - License - Identifier: GPL - 3.0 +
#include "MantidDataHandling/DefaultEventLoader.h"
#include "MantidAPI/Progress.h"
#include "MantidDataHandling/LoadBankFromDiskTask.h"
#include "MantidDataHandling/LoadEventNexus.h"
#include "MantidKernel/ThreadPool.h"
#include "MantidKernel/ThreadSchedulerMutexes.h"
 
using namespace Mantid::Kernel;
 
namespace Mantid::DataHandling {
 
void DefaultEventLoader::load(LoadEventNexus *alg, EventWorkspaceCollection &ws, bool haveWeights,
                              bool event_id_is_spec, std::vector<std::string> bankNames,
                              const std::vector<int> &periodLog, const std::string &classType,
                              std::vector<std::size_t> bankNumEvents, const bool oldNeXusFileNames, const bool precount,
                              const int chunk, const int totalChunks) {
  DefaultEventLoader loader(alg, ws, haveWeights, event_id_is_spec, bankNames.size(), precount, chunk, totalChunks);
 
  auto bankRange = loader.setupChunking(bankNames, bankNumEvents);
 
  // Make the thread pool
  auto scheduler = new ThreadSchedulerMutexes;
  ThreadPool pool(scheduler);
  auto diskIOMutex = std::make_shared<std::mutex>();
 
  // set up progress bar for the rest of the (multi-threaded) process
  size_t numProg = bankNames.size() * (1 + 3); // 1 = disktask, 3 = proc task
  if (loader.splitProcessing)
    numProg += bankNames.size() * 3; // 3 = second proc task
  auto prog = std::make_unique<API::Progress>(loader.alg, 0.3, 1.0, numProg);
 
  for (size_t i = bankRange.first; i < bankRange.second; i++) {
    if (bankNumEvents[i] > 0)
      pool.schedule(std::make_shared<LoadBankFromDiskTask>(loader, bankNames[i], classType, bankNumEvents[i],
                                                           oldNeXusFileNames, prog.get(), diskIOMutex, *scheduler,
                                                           periodLog));
  }
  // Start and end all threads
  pool.joinAll();
  diskIOMutex.reset();
}
 
DefaultEventLoader::DefaultEventLoader(LoadEventNexus *alg, EventWorkspaceCollection &ws, bool haveWeights,
                                       bool event_id_is_spec, const size_t numBanks, const bool precount,
                                       const int chunk, const int totalChunks)
    : m_haveWeights(haveWeights), event_id_is_spec(event_id_is_spec), precount(precount), chunk(chunk),
      totalChunks(totalChunks), firstChunkForBank(1), eventsPerChunk(0), alg(alg), m_ws(ws) {
  // This map will be used to find the workspace index
  if (event_id_is_spec)
    pixelID_to_wi_vector = m_ws.getSpectrumToWorkspaceIndexVector(pixelID_to_wi_offset);
  else
    pixelID_to_wi_vector = m_ws.getDetectorIDToWorkspaceIndexVector(pixelID_to_wi_offset, true);
 
  // Cache a map for speed.
  if (!haveWeights) {
    makeMapToEventLists(eventVectors);
  } else {
    // Convert to weighted events
    for (size_t i = 0; i < m_ws.getNumberHistograms(); i++) {
      m_ws.getSpectrum(i).switchTo(API::WEIGHTED);
    }
    makeMapToEventLists(weightedEventVectors);
  }
 
  // split banks up if the number of cores is more than twice the number of
  // banks
  splitProcessing = bool(numBanks * 2 < ThreadPool::getNumPhysicalCores());
}
 
std::pair<size_t, size_t> DefaultEventLoader::setupChunking(std::vector<std::string> &bankNames,
                                                            std::vector<std::size_t> &bankNumEvents) {
  size_t bank0 = 0;
  size_t bankn = bankNames.size();
  if (chunk != EMPTY_INT()) // We are loading part - work out the bank number range
  {
    const size_t total_events = std::accumulate(bankNumEvents.cbegin(), bankNumEvents.cend(), static_cast<size_t>(0));
    eventsPerChunk = total_events / totalChunks;
    // Sort banks by size
    size_t tmp;
    std::string stmp;
    for (size_t i = 0; i < bankn; i++)
      for (size_t j = 0; j < bankn - 1; j++)
        if (bankNumEvents[j] < bankNumEvents[j + 1]) {
          tmp = bankNumEvents[j];
          bankNumEvents[j] = bankNumEvents[j + 1];
          bankNumEvents[j + 1] = tmp;
          stmp = bankNames[j];
          bankNames[j] = bankNames[j + 1];
          bankNames[j + 1] = stmp;
        }
    int bigBanks = 0;
    for (size_t i = 0; i < bankn; i++)
      if (bankNumEvents[i] > eventsPerChunk)
        bigBanks++;
    // Each chunk is part of bank or multiple whole banks
    // 0.5 for last chunk of a bank with multiple chunks
    // 0.1 for multiple whole banks not completely filled
    eventsPerChunk +=
        static_cast<size_t>((static_cast<double>(bigBanks) / static_cast<double>(totalChunks) * 0.5 + 0.05) *
                            static_cast<double>(eventsPerChunk));
    double partialChunk = 0.;
    firstChunkForBank = 1;
    for (int chunki = 1; chunki <= chunk; chunki++) {
      if (partialChunk > 1.) {
        partialChunk = 0.;
        firstChunkForBank = chunki;
        bank0 = bankn;
      }
      if (bankNumEvents[bank0] > 1) {
        partialChunk += static_cast<double>(eventsPerChunk) / static_cast<double>(bankNumEvents[bank0]);
      }
      if (chunki < totalChunks)
        bankn = bank0 + 1;
      else
        bankn = bankNames.size();
      if (chunki == firstChunkForBank && partialChunk > 1.0)
        bankn += static_cast<size_t>(partialChunk) - 1;
      if (bankn > bankNames.size())
        bankn = bankNames.size();
    }
    for (size_t i = bank0; i < bankn; i++) {
      size_t start_event = (chunk - firstChunkForBank) * eventsPerChunk;
      size_t stop_event = bankNumEvents[i];
      // Don't change stop_event for the final chunk
      if (start_event + eventsPerChunk < stop_event)
        stop_event = start_event + eventsPerChunk;
      bankNumEvents[i] = stop_event - start_event;
    }
  }
  return {bank0, bankn};
}
 
} // namespace Mantid::DataHandling