void SliceMD::slice(typename MDEventWorkspace<MDE, nd>::sptr ws) { // Create the ouput workspace typename MDEventWorkspace<OMDE, ond>::sptr outWS( new MDEventWorkspace<OMDE, ond>()); for (size_t od = 0; od < m_binDimensions.size(); od++) { outWS->addDimension(m_binDimensions[od]); } outWS->setCoordinateSystem(ws->getSpecialCoordinateSystem()); outWS->initialize(); // Copy settings from the original box controller BoxController_sptr bc = ws->getBoxController(); // store wrute buffer size for the future // uint64_t writeBufSize = // bc->getFileIO()getDiskBuffer().getWriteBufferSize(); // and disable write buffer (if any) for input MD Events for this algorithm // purposes; // bc->setCacheParameters(1,0); BoxController_sptr obc = outWS->getBoxController(); // Use the "number of bins" as the "split into" parameter for (size_t od = 0; od < m_binDimensions.size(); od++) obc->setSplitInto(od, m_binDimensions[od]->getNBins()); obc->setSplitThreshold(bc->getSplitThreshold()); bool bTakeDepthFromInputWorkspace = getProperty("TakeMaxRecursionDepthFromInput"); int tempDepth = getProperty("MaxRecursionDepth"); size_t maxDepth = bTakeDepthFromInputWorkspace ? bc->getMaxDepth() : size_t(tempDepth); obc->setMaxDepth(maxDepth); // size_t outputSize = writeBufSize; // obc->setCacheParameters(sizeof(OMDE),outputSize); obc->resetNumBoxes(); // Perform the first box splitting outWS->splitBox(); size_t lastNumBoxes = obc->getTotalNumMDBoxes(); // --- File back end ? ---------------- std::string filename = getProperty("OutputFilename"); if (!filename.empty()) { // First save to the NXS file g_log.notice() << "Running SaveMD to create file back-end" << std::endl; IAlgorithm_sptr alg = createChildAlgorithm("SaveMD"); alg->setPropertyValue("Filename", filename); alg->setProperty("InputWorkspace", outWS); alg->setProperty("MakeFileBacked", true); alg->executeAsChildAlg(); if (!obc->isFileBacked()) throw std::runtime_error("SliceMD with file-backed output: Can not set " "up file-backed output workspace "); auto IOptr = obc->getFileIO(); size_t outBufSize = IOptr->getWriteBufferSize(); // the buffer size for resulting workspace; reasonable size is at least 10 // data chunk sizes (nice to verify) if (outBufSize < 10 * IOptr->getDataChunk()) { outBufSize = 10 * IOptr->getDataChunk(); IOptr->setWriteBufferSize(outBufSize); } } // Function defining which events (in the input dimensions) to place in the // output MDImplicitFunction *function = this->getImplicitFunctionForChunk(NULL, NULL); std::vector<API::IMDNode *> boxes; // Leaf-only; no depth limit; with the implicit function passed to it. ws->getBox()->getBoxes(boxes, 1000, true, function); // Sort boxes by file position IF file backed. This reduces seeking time, // hopefully. bool fileBackedWS = bc->isFileBacked(); if (fileBackedWS) API::IMDNode::sortObjByID(boxes); Progress *prog = new Progress(this, 0.0, 1.0, boxes.size()); // The root of the output workspace MDBoxBase<OMDE, ond> *outRootBox = outWS->getBox(); // if target workspace has events, we should count them as added uint64_t totalAdded = outWS->getNEvents(); uint64_t numSinceSplit = 0; // Go through every box for this chunk. // PARALLEL_FOR_IF( !bc->isFileBacked() ) for (int i = 0; i < int(boxes.size()); i++) { MDBox<MDE, nd> *box = dynamic_cast<MDBox<MDE, nd> *>(boxes[i]); // Perform the binning in this separate method. if (box) { // An array to hold the rotated/transformed coordinates coord_t outCenter[ond]; const std::vector<MDE> &events = box->getConstEvents(); typename std::vector<MDE>::const_iterator it = events.begin(); typename std::vector<MDE>::const_iterator it_end = events.end(); for (; it != it_end; it++) { // Cache the center of the event (again for speed) const coord_t *inCenter = it->getCenter(); if (function->isPointContained(inCenter)) { // Now transform to the output dimensions m_transformFromOriginal->apply(inCenter, outCenter); // Create the event OMDE newEvent(it->getSignal(), it->getErrorSquared(), outCenter); // Copy extra data, if any copyEvent(*it, newEvent); // Add it to the workspace outRootBox->addEvent(newEvent); numSinceSplit++; } } box->releaseEvents(); // Ask BC if one needs to split boxes if (obc->shouldSplitBoxes(totalAdded, numSinceSplit, lastNumBoxes)) // if (numSinceSplit > 20000000 || (i == int(boxes.size()-1))) { // This splits up all the boxes according to split thresholds and sizes. Kernel::ThreadScheduler *ts = new ThreadSchedulerFIFO(); ThreadPool tp(ts); outWS->splitAllIfNeeded(ts); tp.joinAll(); // Accumulate stats totalAdded += numSinceSplit; numSinceSplit = 0; lastNumBoxes = obc->getTotalNumMDBoxes(); // Progress reporting if (!fileBackedWS) prog->report(i); } if (fileBackedWS) { if (!(i % 10)) prog->report(i); } } // is box } // for each box in the vector prog->report(); outWS->splitAllIfNeeded(NULL); // Refresh all cache. outWS->refreshCache(); g_log.notice() << totalAdded << " " << OMDE::getTypeName() << "'s added to the output workspace." << std::endl; if (outWS->isFileBacked()) { // Update the file-back-end g_log.notice() << "Running SaveMD" << std::endl; IAlgorithm_sptr alg = createChildAlgorithm("SaveMD"); alg->setProperty("UpdateFileBackEnd", true); alg->setProperty("InputWorkspace", outWS); alg->executeAsChildAlg(); } // return the size of the input workspace write buffer to its initial value // bc->setCacheParameters(sizeof(MDE),writeBufSize); this->setProperty("OutputWorkspace", boost::dynamic_pointer_cast<IMDEventWorkspace>(outWS)); delete prog; }
void LoadMD::doLoad(typename MDEventWorkspace<MDE, nd>::sptr ws) { // Are we using the file back end? bool fileBackEnd = getProperty("FileBackEnd"); if (fileBackEnd && m_BoxStructureAndMethadata) throw std::invalid_argument("Combination of BoxStructureOnly or " "MetaDataOnly were set to TRUE with " "fileBackEnd " ": this is not possible."); CPUTimer tim; auto prog = new Progress(this, 0.0, 1.0, 100); prog->report("Opening file."); std::string title; try { m_file->getAttr("title", title); } catch (std::exception &) { // Leave the title blank if error on loading } ws->setTitle(title); // Load the WorkspaceHistory "process" if (this->getProperty("LoadHistory")) { ws->history().loadNexus(m_file.get()); } this->loadAffineMatricies(boost::dynamic_pointer_cast<IMDWorkspace>(ws)); m_file->closeGroup(); m_file->close(); // Add each of the dimension for (size_t d = 0; d < nd; d++) ws->addDimension(m_dims[d]); // Coordinate system ws->setCoordinateSystem(m_coordSystem); // ----------------------------------------- Box Structure // ------------------------------ prog->report("Reading box structure from HDD."); MDBoxFlatTree FlatBoxTree; int nDims = static_cast<int>(nd); // should be safe FlatBoxTree.loadBoxStructure(m_filename, nDims, MDE::getTypeName()); BoxController_sptr bc = ws->getBoxController(); bc->fromXMLString(FlatBoxTree.getBCXMLdescr()); prog->report("Restoring box structure and connectivity"); std::vector<API::IMDNode *> boxTree; FlatBoxTree.restoreBoxTree(boxTree, bc, fileBackEnd, m_BoxStructureAndMethadata); size_t numBoxes = boxTree.size(); // ---------------------------------------- DEAL WITH BOXES // ------------------------------------ if (fileBackEnd) { // TODO:: call to the file format factory auto loader = boost::shared_ptr<API::IBoxControllerIO>( new DataObjects::BoxControllerNeXusIO(bc.get())); loader->setDataType(sizeof(coord_t), MDE::getTypeName()); bc->setFileBacked(loader, m_filename); // boxes have been already made file-backed when restoring the boxTree; // How much memory for the cache? { // TODO: Clean up, only a write buffer now double mb = getProperty("Memory"); // Defaults have changed, default disk buffer size should be 10 data // chunks TODO: find optimal, 100 may be better. if (mb <= 0) mb = double(10 * loader->getDataChunk() * sizeof(MDE)) / double(1024 * 1024); // Express the cache memory in units of number of events. uint64_t cacheMemory = static_cast<uint64_t>((mb * 1024. * 1024.) / sizeof(MDE)) + 1; // Set these values in the diskMRU bc->getFileIO()->setWriteBufferSize(cacheMemory); g_log.information() << "Setting a DiskBuffer cache size of " << mb << " MB, or " << cacheMemory << " events.\n"; } } // Not file back end else if (!m_BoxStructureAndMethadata) { // ---------------------------------------- READ IN THE BOXES // ------------------------------------ // TODO:: call to the file format factory auto loader = file_holder_type(new DataObjects::BoxControllerNeXusIO(bc.get())); loader->setDataType(sizeof(coord_t), MDE::getTypeName()); loader->openFile(m_filename, "r"); const std::vector<uint64_t> &BoxEventIndex = FlatBoxTree.getEventIndex(); prog->setNumSteps(numBoxes); for (size_t i = 0; i < numBoxes; i++) { prog->report(); MDBox<MDE, nd> *box = dynamic_cast<MDBox<MDE, nd> *>(boxTree[i]); if (!box) continue; if (BoxEventIndex[2 * i + 1] > 0) // Load in memory NOT using the file as the back-end, { boxTree[i]->reserveMemoryForLoad(BoxEventIndex[2 * i + 1]); boxTree[i]->loadAndAddFrom( loader.get(), BoxEventIndex[2 * i], static_cast<size_t>(BoxEventIndex[2 * i + 1])); } } loader->closeFile(); } else // box structure and metadata only { } g_log.debug() << tim << " to create all the boxes and fill them with events.\n"; // Box of ID 0 is the head box. ws->setBox(boxTree[0]); // Make sure the max ID is ok for later ID generation bc->setMaxId(numBoxes); // end-of bMetaDataOnly // Refresh cache // TODO:if(!fileBackEnd)ws->refreshCache(); ws->refreshCache(); g_log.debug() << tim << " to refreshCache(). " << ws->getNPoints() << " points after refresh.\n"; g_log.debug() << tim << " to finish up.\n"; delete prog; }