Java Code Examples for org.apache.lucene.util.packed.PackedInts#bitsRequired()

private long getPageMemoryUsage(PackedLongValues values, float acceptableOverheadRatio, int pageSize, long pageMinOrdinal, long pageMaxOrdinal) {
    int bitsRequired;
    long pageMemorySize = 0;
    PackedInts.FormatAndBits formatAndBits;
    if (pageMaxOrdinal == Long.MIN_VALUE) {
        // empty page - will use the null reader which just stores size
        pageMemorySize += RamUsageEstimator.alignObjectSize(RamUsageEstimator.NUM_BYTES_OBJECT_HEADER + RamUsageEstimator.NUM_BYTES_INT);

    } else {
        long pageMinValue = values.get(pageMinOrdinal);
        long pageMaxValue = values.get(pageMaxOrdinal);
        long pageDelta = pageMaxValue - pageMinValue;
        if (pageDelta != 0) {
            bitsRequired = pageDelta < 0 ? 64 : PackedInts.bitsRequired(pageDelta);
            formatAndBits = PackedInts.fastestFormatAndBits(pageSize, bitsRequired, acceptableOverheadRatio);
            pageMemorySize += formatAndBits.format.longCount(PackedInts.VERSION_CURRENT, pageSize, formatAndBits.bitsPerValue) * RamUsageEstimator.NUM_BYTES_LONG;
            pageMemorySize += RamUsageEstimator.NUM_BYTES_LONG; // min value per page storage
        } else {
            // empty page
            pageMemorySize += RamUsageEstimator.alignObjectSize(RamUsageEstimator.NUM_BYTES_OBJECT_HEADER + RamUsageEstimator.NUM_BYTES_INT);
        }
    }
    return pageMemorySize;
}
 

@Override
void reset(byte[] bytes, int off, int len) {
  Objects.checkFromIndexSize(off, len, bytes.length);
  this.bytes = bytes;
  this.base = off;
  this.lastOff = off - 1;
  this.end = off + len;
  final int bitsPerOffset = PackedInts.bitsRequired(len - LAST_LITERALS);
  final int bitsPerOffsetLog = 32 - Integer.numberOfLeadingZeros(bitsPerOffset - 1);
  hashLog = MEMORY_USAGE + 3 - bitsPerOffsetLog;
  if (hashTable == null || hashTable.size() < 1 << hashLog || hashTable.getBitsPerValue() < bitsPerOffset) {
    hashTable = PackedInts.getMutable(1 << hashLog, bitsPerOffset, PackedInts.DEFAULT);
  } else {
    // Avoid calling hashTable.clear(), this makes it costly to compress many short sequences otherwise.
    // Instead, get() checks that references are less than the current offset.
    get(off); // this sets the hashTable for the first 4 bytes as a side-effect
  }
}
 

/**
 * Compute the required precision so that <code>count</code> distinct entries
 * would be counted with linear counting.
 */
public static int precisionFromThreshold(long count) {
    final long hashTableEntries = (long) Math.ceil(count / MAX_LOAD_FACTOR);
    int precision = PackedInts.bitsRequired(hashTableEntries * RamUsageEstimator.NUM_BYTES_INT);
    precision = Math.max(precision, MIN_PRECISION);
    precision = Math.min(precision, MAX_PRECISION);
    return precision;
}
 

public OrdinalsBuilder(long numTerms, int maxDoc, float acceptableOverheadRatio) throws IOException {
    this.maxDoc = maxDoc;
    int startBitsPerValue = 8;
    if (numTerms >= 0) {
        startBitsPerValue = PackedInts.bitsRequired(numTerms);
    }
    ordinals = new OrdinalsStore(maxDoc, startBitsPerValue, acceptableOverheadRatio);
    spare = new LongsRef();
}
 

/**
 * Return true if this impl is going to be smaller than {@link SinglePackedOrdinals} by at least 20%.
 */
public static boolean significantlySmallerThanSinglePackedOrdinals(int maxDoc, int numDocsWithValue, long numOrds, float acceptableOverheadRatio) {
    int bitsPerOrd = PackedInts.bitsRequired(numOrds);
    bitsPerOrd = PackedInts.fastestFormatAndBits(numDocsWithValue, bitsPerOrd, acceptableOverheadRatio).bitsPerValue;
    // Compute the worst-case number of bits per value for offsets in the worst case, eg. if no docs have a value at the
    // beginning of the block and all docs have one at the end of the block
    final float avgValuesPerDoc = (float) numDocsWithValue / maxDoc;
    final int maxDelta = (int) Math.ceil(OFFSETS_PAGE_SIZE * (1 - avgValuesPerDoc) * avgValuesPerDoc);
    int bitsPerOffset = PackedInts.bitsRequired(maxDelta) + 1; // +1 because of the sign
    bitsPerOffset = PackedInts.fastestFormatAndBits(maxDoc, bitsPerOffset, acceptableOverheadRatio).bitsPerValue;

    final long expectedMultiSizeInBytes = (long) numDocsWithValue * bitsPerOrd + (long) maxDoc * bitsPerOffset;
    final long expectedSingleSizeInBytes = (long) maxDoc * bitsPerOrd;
    return expectedMultiSizeInBytes < 0.8f * expectedSingleSizeInBytes;
}
 

protected DocValuesFieldUpdates(int maxDoc, long delGen, String field, DocValuesType type) {
  this.maxDoc = maxDoc;
  this.delGen = delGen;
  this.field = field;
  if (type == null) {
    throw new NullPointerException("DocValuesType must not be null");
  }
  this.type = type;
  bitsPerValue = PackedInts.bitsRequired(maxDoc - 1) + SHIFT;
  docs = new PagedMutable(1, PAGE_SIZE, bitsPerValue, PackedInts.DEFAULT);
}
 

private void rehash() throws IOException {
  final PagedGrowableWriter oldTable = table;

  table = new PagedGrowableWriter(2*oldTable.size(), 1<<30, PackedInts.bitsRequired(count), PackedInts.COMPACT);
  mask = table.size()-1;
  for(long idx=0;idx<oldTable.size();idx++) {
    final long address = oldTable.get(idx);
    if (address != 0) {
      addNew(address);
    }
  }
}
 

protected CommonSettings.MemoryStorageFormat chooseStorageFormat(LeafReader reader, PackedLongValues values, Ordinals build, RandomAccessOrds ordinals,
                                                                 long minValue, long maxValue, float acceptableOverheadRatio, int pageSize) {

    CommonSettings.MemoryStorageFormat format;

    // estimate memory usage for a single packed array
    long packedDelta = maxValue - minValue + 1; // allow for a missing value
    // valuesDelta can be negative if the difference between max and min values overflows the positive side of longs.
    int bitsRequired = packedDelta < 0 ? 64 : PackedInts.bitsRequired(packedDelta);
    PackedInts.FormatAndBits formatAndBits = PackedInts.fastestFormatAndBits(reader.maxDoc(), bitsRequired, acceptableOverheadRatio);
    final long singleValuesSize = formatAndBits.format.longCount(PackedInts.VERSION_CURRENT, reader.maxDoc(), formatAndBits.bitsPerValue) * 8L;

    // ordinal memory usage
    final long ordinalsSize = build.ramBytesUsed() + values.ramBytesUsed();

    // estimate the memory signature of paged packing
    long pagedSingleValuesSize = (reader.maxDoc() / pageSize + 1) * RamUsageEstimator.NUM_BYTES_OBJECT_REF; // array of pages
    int pageIndex = 0;
    long pageMinOrdinal = Long.MAX_VALUE;
    long pageMaxOrdinal = Long.MIN_VALUE;
    for (int i = 1; i < reader.maxDoc(); ++i, pageIndex = (pageIndex + 1) % pageSize) {
        ordinals.setDocument(i);
        if (ordinals.cardinality() > 0) {
            long ordinal = ordinals.ordAt(0);
            pageMaxOrdinal = Math.max(ordinal, pageMaxOrdinal);
            pageMinOrdinal = Math.min(ordinal, pageMinOrdinal);
        }
        if (pageIndex == pageSize - 1) {
            // end of page, we now know enough to estimate memory usage
            pagedSingleValuesSize += getPageMemoryUsage(values, acceptableOverheadRatio, pageSize, pageMinOrdinal, pageMaxOrdinal);

            pageMinOrdinal = Long.MAX_VALUE;
            pageMaxOrdinal = Long.MIN_VALUE;
        }
    }

    if (pageIndex > 0) {
        // last page estimation
        pageIndex++;
        pagedSingleValuesSize += getPageMemoryUsage(values, acceptableOverheadRatio, pageSize, pageMinOrdinal, pageMaxOrdinal);
    }

    if (ordinalsSize < singleValuesSize) {
        if (ordinalsSize < pagedSingleValuesSize) {
            format = CommonSettings.MemoryStorageFormat.ORDINALS;
        } else {
            format = CommonSettings.MemoryStorageFormat.PAGED;
        }
    } else {
        if (pagedSingleValuesSize < singleValuesSize) {
            format = CommonSettings.MemoryStorageFormat.PAGED;
        } else {
            format = CommonSettings.MemoryStorageFormat.PACKED;
        }
    }
    return format;
}
 

@Override
protected Accountable createValue(LeafReader reader, CacheKey key)
    throws IOException {

  final int maxDoc = reader.maxDoc();

  Terms terms = reader.terms(key.field);

  final float acceptableOverheadRatio = ((Float) key.custom).floatValue();

  final PagedBytes bytes = new PagedBytes(15);

  int startTermsBPV;

  // TODO: use Uninvert?
  if (terms != null) {
    // Try for coarse estimate for number of bits; this
    // should be an underestimate most of the time, which
    // is fine -- GrowableWriter will reallocate as needed
    long numUniqueTerms = terms.size();
    if (numUniqueTerms != -1L) {
      if (numUniqueTerms > maxDoc) {
        throw new IllegalStateException("Type mismatch: " + key.field + " was indexed with multiple values per document, use SORTED_SET instead");
      }

      startTermsBPV = PackedInts.bitsRequired(numUniqueTerms);
    } else {
      startTermsBPV = 1;
    }
  } else {
    startTermsBPV = 1;
  }

  PackedLongValues.Builder termOrdToBytesOffset = PackedLongValues.monotonicBuilder(PackedInts.COMPACT);
  final GrowableWriter docToTermOrd = new GrowableWriter(startTermsBPV, maxDoc, acceptableOverheadRatio);

  int termOrd = 0;

  // TODO: use Uninvert?

  if (terms != null) {
    final TermsEnum termsEnum = terms.iterator();
    PostingsEnum docs = null;

    while(true) {
      final BytesRef term = termsEnum.next();
      if (term == null) {
        break;
      }
      if (termOrd >= maxDoc) {
        throw new IllegalStateException("Type mismatch: " + key.field + " was indexed with multiple values per document, use SORTED_SET instead");
      }

      termOrdToBytesOffset.add(bytes.copyUsingLengthPrefix(term));
      docs = termsEnum.postings(docs, PostingsEnum.NONE);
      while (true) {
        final int docID = docs.nextDoc();
        if (docID == DocIdSetIterator.NO_MORE_DOCS) {
          break;
        }
        // Store 1+ ord into packed bits
        docToTermOrd.set(docID, 1+termOrd);
      }
      termOrd++;
    }
  }

  // maybe an int-only impl?
  return new SortedDocValuesImpl(bytes.freeze(true), termOrdToBytesOffset.build(), docToTermOrd.getMutable(), termOrd);
}
 

@Override
protected Accountable createValue(LeafReader reader, CacheKey key)
    throws IOException {

  // TODO: would be nice to first check if DocTermsIndex
  // was already cached for this field and then return
  // that instead, to avoid insanity

  final int maxDoc = reader.maxDoc();
  Terms terms = reader.terms(key.field);

  final float acceptableOverheadRatio = ((Float) key.custom).floatValue();

  final int termCountHardLimit = maxDoc;

  // Holds the actual term data, expanded.
  final PagedBytes bytes = new PagedBytes(15);

  int startBPV;

  if (terms != null) {
    // Try for coarse estimate for number of bits; this
    // should be an underestimate most of the time, which
    // is fine -- GrowableWriter will reallocate as needed
    long numUniqueTerms = terms.size();
    if (numUniqueTerms != -1L) {
      if (numUniqueTerms > termCountHardLimit) {
        numUniqueTerms = termCountHardLimit;
      }
      startBPV = PackedInts.bitsRequired(numUniqueTerms*4);
    } else {
      startBPV = 1;
    }
  } else {
    startBPV = 1;
  }

  final GrowableWriter docToOffset = new GrowableWriter(startBPV, maxDoc, acceptableOverheadRatio);
  
  // pointer==0 means not set
  bytes.copyUsingLengthPrefix(new BytesRef());

  if (terms != null) {
    int termCount = 0;
    final TermsEnum termsEnum = terms.iterator();
    PostingsEnum docs = null;
    while(true) {
      if (termCount++ == termCountHardLimit) {
        // app is misusing the API (there is more than
        // one term per doc); in this case we make best
        // effort to load what we can (see LUCENE-2142)
        break;
      }

      final BytesRef term = termsEnum.next();
      if (term == null) {
        break;
      }
      final long pointer = bytes.copyUsingLengthPrefix(term);
      docs = termsEnum.postings(docs, PostingsEnum.NONE);
      while (true) {
        final int docID = docs.nextDoc();
        if (docID == DocIdSetIterator.NO_MORE_DOCS) {
          break;
        }
        docToOffset.set(docID, pointer);
      }
    }
  }

  final PackedInts.Reader offsetReader = docToOffset.getMutable();
  Bits docsWithField = new Bits() {
    @Override
    public boolean get(int index) {
      return offsetReader.get(index) != 0;
    }

    @Override
    public int length() {
      return maxDoc;
    }
  };

  wrapper.setDocsWithField(reader, key.field, docsWithField, null);
  // maybe an int-only impl?
  return new BinaryDocValuesImpl(bytes.freeze(true), offsetReader, docsWithField);
}