Facebook sync (#504)

* Facebook sync

* Update swig wrappers.

* Fix comment.

AutoTune.cpp

@@ -6,9 +6,11 @@
  * LICENSE file in the root directory of this source tree.
  */
 
-/* Copyright 2004-present Facebook. All Rights Reserved.
-   implementation of Hyper-parameter auto-tuning
-*/
+// -*- c++ -*-
+
+/*
+ * implementation of Hyper-parameter auto-tuning
+ */
 
 #include "AutoTune.h"
 
@@ -25,7 +27,8 @@
 #include "MetaIndexes.h"
 #include "IndexScalarQuantizer.h"
 #include "IndexHNSW.h"
-
+#include "IndexBinaryFlat.h"
+#include "IndexBinaryIVF.h"
 
 namespace faiss {
 
@@ -734,11 +737,8 @@ Index *index_factory (int d, const char *description_in, MetricType metric)
             vt_1 = new OPQMatrix (d, opq_M);
         } else if (stok == "L2norm") {
             vt_1 = new NormalizationTransform (d, 2.0);
-        // coarse quantizers
-        } else if (!coarse_quantizer &&
-                   sscanf (tok, "IVF%d_HNSW%d", &ncentroids, &M) == 2) {
-            FAISS_THROW_IF_NOT (metric == METRIC_L2);
-            coarse_quantizer_1 = new IndexHNSWFlat (d, M);
+
+
         } else if (!coarse_quantizer &&
                    sscanf (tok, "IVF%d", &ncentroids) == 1) {
             if (metric == METRIC_L2) {
@@ -755,11 +755,14 @@ Index *index_factory (int d, const char *description_in, MetricType metric)
             add_idmap = true;
 
             // IVFs
-        } else if (!index && stok == "Flat") {
+        } else if (!index && (stok == "Flat" || stok == "FlatDedup")) {
             if (coarse_quantizer) {
                 // if there was an IVF in front, then it is an IVFFlat
-                IndexIVF *index_ivf = new IndexIVFFlat (
-                    coarse_quantizer, d, ncentroids, metric);
+                IndexIVF *index_ivf = stok == "Flat" ?
+                    new IndexIVFFlat (
+                          coarse_quantizer, d, ncentroids, metric) :
+                    new IndexIVFFlatDedup (
+                          coarse_quantizer, d, ncentroids, metric);
                 index_ivf->quantizer_trains_alone =
                     get_trains_alone (coarse_quantizer);
                 index_ivf->cp.spherical = metric == METRIC_INNER_PRODUCT;
@@ -767,12 +770,16 @@ Index *index_factory (int d, const char *description_in, MetricType metric)
                 index_ivf->own_fields = true;
                 index_1 = index_ivf;
             } else {
+                FAISS_THROW_IF_NOT_MSG (stok != "FlatDedup",
+                                        "dedup supported only for IVFFlat");
                 index_1 = new IndexFlat (d, metric);
             }
-        } else if (!index && (stok == "SQ8" || stok == "SQ4")) {
+        } else if (!index && (stok == "SQ8" || stok == "SQ4" ||
+                              stok == "SQfp16")) {
             ScalarQuantizer::QuantizerType qt =
                 stok == "SQ8" ? ScalarQuantizer::QT_8bit :
                 stok == "SQ4" ? ScalarQuantizer::QT_4bit :
+                stok == "SQfp16" ? ScalarQuantizer::QT_fp16 :
                 ScalarQuantizer::QT_4bit;
             if (coarse_quantizer) {
                 IndexIVFScalarQuantizer *index_ivf =
@@ -905,7 +912,27 @@ Index *index_factory (int d, const char *description_in, MetricType metric)
     return index;
 }
 
+IndexBinary *index_binary_factory(int d, const char *description)
+{
+    IndexBinary *index = nullptr;
+
+    int ncentroids = -1;
 
+    if (sscanf(description, "BIVF%d", &ncentroids) == 1) {
+        IndexBinaryIVF *index_ivf = new IndexBinaryIVF(
+            new IndexBinaryFlat(d), d, ncentroids
+        );
+        index_ivf->own_fields = true;
+        index = index_ivf;
+    } else if (std::string(description) == "BFlat") {
+        index = new IndexBinaryFlat(d);
+    } else {
+        FAISS_THROW_IF_NOT_FMT(index, "descrption %s did not generate an index",
+                               description);
+    }
+
+    return index;
+}
 
 
-}; // namespace faiss
+} // namespace faiss

AutoTune.h

@@ -6,7 +6,6 @@
  * LICENSE file in the root directory of this source tree.
  */
 
-// Copyright 2004-present Facebook. All Rights Reserved.
 // -*- c++ -*-
 
 #ifndef FAISS_AUTO_TUNE_H
@@ -15,6 +14,7 @@
 #include <vector>
 
 #include "Index.h"
+#include "IndexBinary.h"
 
 namespace faiss {
 
@@ -203,6 +203,8 @@ struct ParameterSpace {
 Index *index_factory (int d, const char *description,
                       MetricType metric = METRIC_L2);
 
+IndexBinary *index_binary_factory (int d, const char *description);
+
 
 
 } // namespace faiss

AuxIndexStructures.cpp

@@ -6,11 +6,12 @@
  * LICENSE file in the root directory of this source tree.
  */
 
-// Copyright 2004-present Facebook. All Rights Reserved
 // -*- c++ -*-
 
 #include "AuxIndexStructures.h"
 
+#include "FaissAssert.h"
+
 #include <cstring>
 
 namespace faiss {
@@ -208,9 +209,44 @@ bool IDSelectorBatch::is_member (idx_t i) const
 }
 
 
+/***********************************************************************
+ * IO functions
+ ***********************************************************************/
+
 
+int IOReader::fileno ()
+{
+    FAISS_THROW_MSG ("IOReader does not support memory mapping");
+}
 
+int IOWriter::fileno ()
+{
+    FAISS_THROW_MSG ("IOWriter does not support memory mapping");
+}
 
 
+size_t VectorIOWriter::operator()(
+                const void *ptr, size_t size, size_t nitems)
+{
+    size_t o = data.size();
+    data.resize(o + size * nitems);
+    memcpy (&data[o], ptr, size * nitems);
+    return nitems;
+}
 
+size_t VectorIOReader::operator()(
+                  void *ptr, size_t size, size_t nitems)
+{
+    if (rp >= data.size()) return 0;
+    size_t nremain = (data.size() - rp) / size;
+    if (nremain < nitems) nitems = nremain;
+    memcpy (ptr, &data[rp], size * nitems);
+    rp += size * nitems;
+    return nitems;
 }
+
+
+
+
+
+} // namespace faiss

AuxIndexStructures.h

@@ -6,14 +6,16 @@
  * LICENSE file in the root directory of this source tree.
  */
 
-// Copyright 2004-present Facebook. All Rights Reserved
 // -*- c++ -*-
+
 // Auxiliary index structures, that are used in indexes but that can
 // be forward-declared
 
 #ifndef FAISS_AUX_INDEX_STRUCTURES_H
 #define FAISS_AUX_INDEX_STRUCTURES_H
 
+#include <stdint.h>
+
 #include <vector>
 #include <unordered_set>
 
@@ -157,10 +159,8 @@ struct RangeSearchPartialResult: BufferList {
     /// begin a new result
     QueryResult & new_result (idx_t qno);
 
-
     void finalize ();
 
-
     /// called by range_search before do_allocation
     void set_lims ();
 
@@ -169,6 +169,46 @@ struct RangeSearchPartialResult: BufferList {
 
 };
 
+/***********************************************************
+ * Abstract I/O objects
+ ***********************************************************/
+
+
+struct IOReader {
+    // fread
+    virtual size_t operator()(
+         void *ptr, size_t size, size_t nitems) = 0;
+
+    // return a file number that can be memory-mapped
+    virtual int fileno ();
+
+    virtual ~IOReader() {}
+};
+
+struct IOWriter {
+    // fwrite
+    virtual size_t operator()(
+         const void *ptr, size_t size, size_t nitems) = 0;
+
+    // return a file number that can be memory-mapped
+    virtual int fileno ();
+
+    virtual ~IOWriter() {}
+};
+
+
+struct VectorIOReader:IOReader {
+    const std::vector<uint8_t> data;
+    size_t rp = 0;
+    size_t operator()(void *ptr, size_t size, size_t nitems) override;
+};
+
+struct VectorIOWriter:IOWriter {
+    std::vector<uint8_t> data;
+    size_t operator()(const void *ptr, size_t size, size_t nitems) override;
+};
+
+
 
 }; // namespace faiss
 

Clustering.cpp

@@ -6,14 +6,11 @@
  * LICENSE file in the root directory of this source tree.
  */
 
-/* Copyright 2004-present Facebook. All Rights Reserved.
-   kmeans clustering routines
-*/
+// -*- c++ -*-
 
 #include "Clustering.h"
 
 
-
 #include <cmath>
 #include <cstdio>
 #include <cstring>

Clustering.h

@@ -6,7 +6,6 @@
  * LICENSE file in the root directory of this source tree.
  */
 
-// Copyright 2004-present Facebook. All Rights Reserved
 // -*- c++ -*-
 
 #ifndef FAISS_CLUSTERING_H

FaissAssert.h

@@ -6,7 +6,7 @@
  * LICENSE file in the root directory of this source tree.
  */
 
-// Copyright 2004-present Facebook. All Rights Reserved.
+// -*- c++ -*-
 
 #ifndef FAISS_ASSERT_INCLUDED
 #define FAISS_ASSERT_INCLUDED

FaissException.cpp

@@ -6,7 +6,7 @@
  * LICENSE file in the root directory of this source tree.
  */
 
-// Copyright 2004-present Facebook. All Rights Reserved.
+// -*- c++ -*-
 
 #include "FaissException.h"
 

FaissException.h

@@ -6,7 +6,7 @@
  * LICENSE file in the root directory of this source tree.
  */
 
-// Copyright 2004-present Facebook. All Rights Reserved.
+// -*- c++ -*-
 
 #ifndef FAISS_EXCEPTION_INCLUDED
 #define FAISS_EXCEPTION_INCLUDED

Heap.cpp

@@ -6,7 +6,8 @@
  * LICENSE file in the root directory of this source tree.
  */
 
-/* Copyright 2004-present Facebook. All Rights Reserved. */
+// -*- c++ -*-
+
 /* Function for soft heap */
 
 #include "Heap.h"
@@ -15,13 +16,6 @@
 namespace faiss {
 
 
-
-
-
-
-
-
-
 template <typename C>
 void HeapArray<C>::heapify ()
 {
@@ -126,7 +120,4 @@ template class HeapArray<CMin <int, long> >;
 template class HeapArray<CMax <int, long> >;
 
 
-
-
-
 }  // END namespace fasis

Heap.h

@@ -6,8 +6,9 @@
  * LICENSE file in the root directory of this source tree.
  */
 
-/* Copyright 2004-present Facebook. All Rights Reserved.
- *
+// -*- c++ -*-
+
+/*
  * C++ support for heaps. The set of functions is tailored for
  * efficient similarity search.
  *
@@ -31,7 +32,6 @@
 #include <limits>
 
 
-
 namespace faiss {
 
 /*******************************************************************
@@ -490,8 +490,6 @@ void indirect_heap_push (size_t k,
 }
 
 
-
-
 } // namespace faiss
 
 #endif  /* FAISS_Heap_h */

Index.cpp

@@ -6,13 +6,14 @@
  * LICENSE file in the root directory of this source tree.
  */
 
-// Copyright 2004-present Facebook. All Rights Reserved
+// -*- c++ -*-
 
 #include "IndexFlat.h"
 #include "FaissAssert.h"
 
 #include <cstring>
 
+
 namespace faiss {
 
 Index::~Index ()

Index.h

@@ -6,7 +6,6 @@
  * LICENSE file in the root directory of this source tree.
  */
 
-// Copyright 2004-present Facebook. All Rights Reserved
 // -*- c++ -*-
 
 #ifndef FAISS_INDEX_H

IndexBinary.cpp

+/**
+ * Copyright (c) 2015-present, Facebook, Inc.
+ * All rights reserved.
+ *
+ * This source code is licensed under the BSD+Patents license found in the
+ * LICENSE file in the root directory of this source tree.
+ */
+
+// -*- c++ -*-
+
+#include "IndexBinary.h"
+#include "FaissAssert.h"
+
+#include <cstring>
+
+namespace faiss {
+
+IndexBinary::~IndexBinary() {}
+
+void IndexBinary::train(idx_t, const uint8_t *) {
+  // Does nothing by default.
+}
+
+void IndexBinary::range_search(idx_t, const uint8_t *, int,
+                               RangeSearchResult *) const {
+  FAISS_THROW_MSG("range search not implemented");
+}
+
+void IndexBinary::assign(idx_t n, const uint8_t *x, idx_t *labels, idx_t k) {
+  int *distances = new int[n * k];
+  ScopeDeleter<int> del(distances);
+  search(n, x, k, distances, labels);
+}
+
+void IndexBinary::add_with_ids(idx_t, const uint8_t *, const long *) {
+  FAISS_THROW_MSG("add_with_ids not implemented for this type of index");
+}
+
+long IndexBinary::remove_ids(const IDSelector&) {
+  FAISS_THROW_MSG("remove_ids not implemented for this type of index");
+  return -1;
+}
+
+void IndexBinary::reconstruct(idx_t, uint8_t *) const {
+  FAISS_THROW_MSG("reconstruct not implemented for this type of index");
+}
+
+void IndexBinary::reconstruct_n(idx_t i0, idx_t ni, uint8_t *recons) const {
+  for (idx_t i = 0; i < ni; i++) {
+    reconstruct(i0 + i, recons + i * d);
+  }
+}
+
+void IndexBinary::search_and_reconstruct(idx_t n, const uint8_t *x, idx_t k,
+                                         int32_t *distances, idx_t *labels,
+                                         uint8_t *recons) const {
+  search(n, x, k, distances, labels);
+  for (idx_t i = 0; i < n; ++i) {
+    for (idx_t j = 0; j < k; ++j) {
+      idx_t ij = i * k + j;
+      idx_t key = labels[ij];
+      uint8_t *reconstructed = recons + ij * d;
+      if (key < 0) {
+        // Fill with NaNs
+        memset(reconstructed, -1, sizeof(*reconstructed) * d);
+      } else {
+        reconstruct(key, reconstructed);
+      }
+    }
+  }
+}
+
+void IndexBinary::display() const {
+  printf("Index: %s  -> %ld elements\n", typeid (*this).name(), ntotal);
+}
+
+
+}  // namespace faiss

IndexBinary.h

+/**
+ * Copyright (c) 2015-present, Facebook, Inc.
+ * All rights reserved.
+ *
+ * This source code is licensed under the BSD+Patents license found in the
+ * LICENSE file in the root directory of this source tree.
+ */
+
+// -*- c++ -*-
+
+#ifndef FAISS_INDEX_BINARY_H
+#define FAISS_INDEX_BINARY_H
+
+#include <cstdio>
+#include <typeinfo>
+#include <string>
+#include <sstream>
+
+#include "FaissAssert.h"
+#include "Index.h"
+
+
+namespace faiss {
+
+
+/// Forward declarations see AuxIndexStructures.h
+struct IDSelector;
+struct RangeSearchResult;
+
+/** Abstract structure for a binary index.
+ *
+ * Supports adding vertices and searching them.
+ *
+ * All queries are symmetric because there is no distinction between codes and
+ * vectors.
+ */
+struct IndexBinary {
+  typedef long idx_t;    ///< all indices are this type
+
+  int d;                 ///< vector dimension
+  int code_size;   ///< number of bytes per vector ( = d / 8 )
+  idx_t ntotal;          ///< total nb of indexed vectors
+  bool verbose;          ///< verbosity level
+
+  /// set if the Index does not require training, or if training is done already
+  bool is_trained;
+
+  /// type of metric this index uses for search
+  MetricType metric_type;
+
+  explicit IndexBinary(idx_t d = 0, MetricType metric = METRIC_L2)
+      : d(d),
+        code_size(d / 8),
+        ntotal(0),
+        verbose(false),
+        is_trained(true),
+        metric_type(metric) {
+        FAISS_THROW_IF_NOT(d % 8 == 0);
+      }
+
+  virtual ~IndexBinary();
+
+
+  /** Perform training on a representative set of vectors.
+   *
+   * @param n      nb of training vectors
+   * @param x      training vecors, size n * d / 8
+   */
+  virtual void train(idx_t n, const uint8_t *x);
+
+  /** Add n vectors of dimension d to the index.
+   *
+   * Vectors are implicitly assigned labels ntotal .. ntotal + n - 1
+   * @param x      input matrix, size n * d / 8
+   */
+  virtual void add(idx_t n, const uint8_t *x) = 0;
+
+  /** Same as add, but stores xids instead of sequential ids.
+   *
+   * The default implementation fails with an assertion, as it is
+   * not supported by all indexes.
+   *
+   * @param xids if non-null, ids to store for the vectors (size n)
+   */
+  virtual void add_with_ids(idx_t n, const uint8_t *x, const long *xids);
+
+  /** Query n vectors of dimension d to the index.
+   *
+   * return at most k vectors. If there are not enough results for a
+   * query, the result array is padded with -1s.
+   *
+   * @param x           input vectors to search, size n * d / 8
+   * @param labels      output labels of the NNs, size n*k
+   * @param distances   output pairwise distances, size n*k
+   */
+  virtual void search(idx_t n, const uint8_t *x, idx_t k,
+                      int32_t *distances, idx_t *labels) const = 0;
+
+  /** Query n vectors of dimension d to the index.
+   *
+   * return all vectors with distance < radius. Note that many
+   * indexes do not implement the range_search (only the k-NN search
+   * is mandatory).
+   *
+   * @param x           input vectors to search, size n * d / 8
+   * @param radius      search radius
+   * @param result      result table
+   */
+  virtual void range_search(idx_t n, const uint8_t *x, int radius,
+                            RangeSearchResult *result) const;
+
+  /** Return the indexes of the k vectors closest to the query x.
+   *
+   * This function is identical to search but only returns labels of neighbors.
+   * @param x           input vectors to search, size n * d / 8
+   * @param labels      output labels of the NNs, size n*k
+   */
+  void assign(idx_t n, const uint8_t *x, idx_t *labels, idx_t k = 1);
+
+  /// Removes all elements from the database.
+  virtual void reset() = 0;
+
+  /** Removes IDs from the index. Not supported by all indexes.
+   */
+  virtual long remove_ids(const IDSelector& sel);
+
+  /** Reconstruct a stored vector.
+   *
+   * This function may not be defined for some indexes.
+   * @param key         id of the vector to reconstruct
+   * @param recons      reconstucted vector (size d / 8)
+   */
+  virtual void reconstruct(idx_t key, uint8_t *recons) const;
+
+
+  /** Reconstruct vectors i0 to i0 + ni - 1.
+   *
+   * This function may not be defined for some indexes.
+   * @param recons      reconstucted vectors (size ni * d / 8)
+   */
+  virtual void reconstruct_n(idx_t i0, idx_t ni, uint8_t *recons) const;
+
+  /** Similar to search, but also reconstructs the stored vectors (or an
+   * approximation in the case of lossy coding) for the search results.
+   *
+   * If there are not enough results for a query, the resulting array
+   * is padded with -1s.
+   *
+   * @param recons      reconstructed vectors size (n, k, d)
+   **/
+  virtual void search_and_reconstruct(idx_t n, const uint8_t *x, idx_t k,
+                                      int32_t *distances, idx_t *labels,
+                                      uint8_t *recons) const;
+
+  /** Display the actual class name and some more info. */
+  void display() const;
+};
+
+
+}  // namespace faiss
+
+#endif  // FAISS_INDEX_BINARY_H

IndexBinaryFlat.cpp

+/**
+ * Copyright (c) 2015-present, Facebook, Inc.
+ * All rights reserved.
+ *
+ * This source code is licensed under the BSD+Patents license found in the
+ * LICENSE file in the root directory of this source tree.
+ */
+
+// -*- c++ -*-
+
+#include "IndexBinaryFlat.h"
+
+#include <cstring>
+#include "hamming.h"
+#include "utils.h"
+#include "Heap.h"
+#include "FaissAssert.h"
+#include "AuxIndexStructures.h"
+
+namespace faiss {
+
+IndexBinaryFlat::IndexBinaryFlat(idx_t d)
+    : IndexBinary(d) {}
+
+void IndexBinaryFlat::add(idx_t n, const uint8_t *x) {
+  xb.insert(xb.end(), x, x + n * code_size);
+  ntotal += n;
+}
+
+void IndexBinaryFlat::reset() {
+  xb.clear();
+  ntotal = 0;
+}
+
+void IndexBinaryFlat::search(idx_t n, const uint8_t *x, idx_t k,
+                             int32_t *distances, idx_t *labels) const {
+  const size_t block_size = query_batch_size;
+  for (size_t s = 0; s < n; s += block_size) {
+    idx_t nn = block_size;
+    if (s + block_size > n) {
+      nn = n - s;
+    }
+
+    if (use_heap) {
+      // We see the distances and labels as heaps.
+      int_maxheap_array_t res = {
+        size_t(nn), size_t(k), labels + s * k, distances + s * k
+      };
+
+      hammings_knn_hc(&res, x + s * code_size, xb.data(), ntotal, code_size,
+                      /* ordered = */ true);
+    } else {
+      hammings_knn_mc(x + s * code_size, xb.data(), nn, ntotal, k, code_size,
+                      distances, labels);
+    }
+  }
+}
+
+long IndexBinaryFlat::remove_ids(const IDSelector& sel) {
+  idx_t j = 0;
+  for (idx_t i = 0; i < ntotal; i++) {
+    if (sel.is_member(i)) {
+      // should be removed
+    } else {
+      if (i > j) {
+        memmove(&xb[code_size * j], &xb[code_size * i], sizeof(xb[0]) * code_size);
+      }
+      j++;
+    }
+  }
+  long nremove = ntotal - j;
+  if (nremove > 0) {
+    ntotal = j;
+    xb.resize(ntotal * code_size);
+  }
+  return nremove;
+}
+
+void IndexBinaryFlat::reconstruct(idx_t key, uint8_t *recons) const {
+  memcpy(recons, &(xb[code_size * key]), sizeof(*recons) * code_size);
+}
+
+
+}  // namespace faiss

IndexBinaryFlat.h

+/**
+ * Copyright (c) 2015-present, Facebook, Inc.
+ * All rights reserved.
+ *
+ * This source code is licensed under the BSD+Patents license found in the
+ * LICENSE file in the root directory of this source tree.
+ */
+
+// -*- c++ -*-
+
+#ifndef INDEX_BINARY_FLAT_H
+#define INDEX_BINARY_FLAT_H
+
+#include <vector>
+
+#include "IndexBinary.h"
+
+namespace faiss {
+
+
+/** Index that stores the full vectors and performs exhaustive search. */
+struct IndexBinaryFlat : IndexBinary {
+  /// database vectors, size ntotal * d / 8
+  std::vector<uint8_t> xb;
+
+  /** Select between using a heap or counting to select the k smallest values
+   * when scanning inverted lists.
+   */
+  bool use_heap = true;
+
+  size_t query_batch_size = 32;
+
+  explicit IndexBinaryFlat(idx_t d);
+
+  void add(idx_t n, const uint8_t *x) override;
+
+  void reset() override;
+
+  void search(idx_t n, const uint8_t *x, idx_t k,
+              int32_t *distances, idx_t *labels) const override;
+
+  void reconstruct(idx_t key, uint8_t *recons) const override;
+
+  /** Remove some ids. Note that because of the indexing structure,
+   * the semantics of this operation are different from the usual ones:
+   * the new ids are shifted. */
+  long remove_ids(const IDSelector& sel) override;
+
+  IndexBinaryFlat() {}
+};
+
+
+}  // namespace faiss
+
+#endif  // INDEX_BINARY_FLAT_H

IndexBinaryIVF.h

+/**
+ * Copyright (c) 2015-present, Facebook, Inc.
+ * All rights reserved.
+ *
+ * This source code is licensed under the BSD+Patents license found in the
+ * LICENSE file in the root directory of this source tree.
+ */
+
+// -*- c++ -*-
+
+#ifndef FAISS_INDEX_BINARY_IVF_H
+#define FAISS_INDEX_BINARY_IVF_H
+
+
+#include <vector>
+
+#include "IndexBinary.h"
+#include "IndexIVF.h"
+#include "Clustering.h"
+#include "Heap.h"
+
+
+namespace faiss {
+
+
+/** Index based on a inverted file (IVF)
+ *
+ * In the inverted file, the quantizer (an IndexBinary instance) provides a
+ * quantization index for each vector to be added. The quantization
+ * index maps to a list (aka inverted list or posting list), where the
+ * id of the vector is stored.
+ *
+ * The inverted list object is required only after trainng. If none is
+ * set externally, an ArrayInvertedLists is used automatically.
+ *
+ * At search time, the vector to be searched is also quantized, and
+ * only the list corresponding to the quantization index is
+ * searched. This speeds up the search by making it
+ * non-exhaustive. This can be relaxed using multi-probe search: a few
+ * (nprobe) quantization indices are selected and several inverted
+ * lists are visited.
+ */
+struct IndexBinaryIVF : IndexBinary {
+    /// Acess to the actual data
+    InvertedLists *invlists;
+    bool own_invlists;
+
+    size_t nprobe;            ///< number of probes at query time
+    size_t max_codes;         ///< max nb of codes to visit to do a query
+
+    /** Select between using a heap or counting to select the k smallest values
+     * when scanning inverted lists.
+     */
+    bool use_heap = true;
+
+    /// map for direct access to the elements. Enables reconstruct().
+    bool maintain_direct_map;
+    std::vector<long> direct_map;
+
+    IndexBinary *quantizer;   ///< quantizer that maps vectors to inverted lists
+    size_t nlist;             ///< number of possible key values
+
+    /**
+     * = 0: use the quantizer as index in a kmeans training
+     * = 1: just pass on the training set to the train() of the quantizer
+     * = 2: kmeans training on a flat index + add the centroids to the quantizer
+     */
+    bool own_fields;          ///< whether object owns the quantizer
+
+    ClusteringParameters cp; ///< to override default clustering params
+
+    /// Trains the quantizer and calls train_residual to train sub-quantizers
+    void train_q1(size_t n, const uint8_t *x, bool verbose);
+
+    /** The Inverted file takes a quantizer (an IndexBinary) on input,
+     * which implements the function mapping a vector to a list
+     * identifier. The pointer is borrowed: the quantizer should not
+     * be deleted while the IndexBinaryIVF is in use.
+     */
+    IndexBinaryIVF(IndexBinary *quantizer, size_t d, size_t nlist);
+
+    IndexBinaryIVF();
+
+    ~IndexBinaryIVF() override;
+
+    void reset() override;
+
+    /// Trains the quantizer and calls train_residual to train sub-quantizers
+    void train(idx_t n, const uint8_t *x) override;
+
+    /// Quantizes x and calls add_with_key
+    void add(idx_t n, const uint8_t *x) override;
+
+    void add_with_ids(idx_t n, const uint8_t *x, const long *xids) override;
+
+    /// same as add_with_ids, with precomputed coarse quantizer
+    void add_core (idx_t n, const uint8_t * x, const long *xids,
+                   const long *precomputed_idx);
+
+    /** Search a set of vectors, that are pre-quantized by the IVF
+     *  quantizer. Fill in the corresponding heaps with the query
+     *  results. search() calls this.
+     *
+     * @param n      nb of vectors to query
+     * @param x      query vectors, size nx * d
+     * @param assign coarse quantization indices, size nx * nprobe
+     * @param centroid_dis
+     *               distances to coarse centroids, size nx * nprobe
+     * @param distance
+     *               output distances, size n * k
+     * @param labels output labels, size n * k
+     * @param store_pairs store inv list index + inv list offset
+     *                     instead in upper/lower 32 bit of result,
+     *                     instead of ids (used for reranking).
+     * @param params used to override the object's search parameters
+     */
+    void search_preassigned(idx_t n, const uint8_t *x, idx_t k,
+                            const idx_t *assign,
+                            const int32_t *centroid_dis,
+                            int32_t *distances, idx_t *labels,
+                            bool store_pairs,
+                            const IVFSearchParameters *params=nullptr
+                            ) const;
+
+    /** assign the vectors, then call search_preassign */
+    virtual void search(idx_t n, const uint8_t *x, idx_t k,
+                        int32_t *distances, idx_t *labels) const override;
+
+    void reconstruct(idx_t key, uint8_t *recons) const override;
+
+    /** Reconstruct a subset of the indexed vectors.
+     *
+     * Overrides default implementation to bypass reconstruct() which requires
+     * direct_map to be maintained.
+     *
+     * @param i0     first vector to reconstruct
+     * @param ni     nb of vectors to reconstruct
+     * @param recons output array of reconstructed vectors, size ni * d / 8
+     */
+    void reconstruct_n(idx_t i0, idx_t ni, uint8_t *recons) const override;
+
+    /** Similar to search, but also reconstructs the stored vectors (or an
+     * approximation in the case of lossy coding) for the search results.
+     *
+     * Overrides default implementation to avoid having to maintain direct_map
+     * and instead fetch the code offsets through the `store_pairs` flag in
+     * search_preassigned().
+     *
+     * @param recons      reconstructed vectors size (n, k, d / 8)
+     */
+    void search_and_reconstruct(idx_t n, const uint8_t *x, idx_t k,
+                                int32_t *distances, idx_t *labels,
+                                uint8_t *recons) const override;
+
+    /** Reconstruct a vector given the location in terms of (inv list index +
+     * inv list offset) instead of the id.
+     *
+     * Useful for reconstructing when the direct_map is not maintained and
+     * the inv list offset is computed by search_preassigned() with
+     * `store_pairs` set.
+     */
+    virtual void reconstruct_from_offset(long list_no, long offset,
+                                         uint8_t* recons) const;
+
+
+    /// Dataset manipulation functions
+
+    long remove_ids(const IDSelector& sel) override;
+
+    /** moves the entries from another dataset to self. On output,
+     * other is empty. add_id is added to all moved ids (for
+     * sequential ids, this would be this->ntotal */
+    virtual void merge_from(IndexBinaryIVF& other, idx_t add_id);
+
+    size_t get_list_size(size_t list_no) const
+    { return invlists->list_size(list_no); }
+
+    /** intialize a direct map
+     *
+     * @param new_maintain_direct_map    if true, create a direct map,
+     *                                   else clear it
+     */
+    void make_direct_map(bool new_maintain_direct_map=true);
+
+    /// 1= perfectly balanced, >1: imbalanced
+    double imbalance_factor() const;
+
+    /// display some stats about the inverted lists
+    void print_stats() const;
+
+    void replace_invlists(InvertedLists *il, bool own=false);
+};
+
+
+}  // namespace faiss
+
+#endif  // FAISS_INDEX_BINARY_IVF_H

IndexFlat.cpp

@@ -6,7 +6,7 @@
  * LICENSE file in the root directory of this source tree.
  */
 
-// Copyright 2004-present Facebook. All Rights Reserved
+// -*- c++ -*-
 
 #include "IndexFlat.h"
 
@@ -18,6 +18,7 @@
 
 #include "AuxIndexStructures.h"
 
+
 namespace faiss {
 
 IndexFlat::IndexFlat (idx_t d, MetricType metric):

IndexFlat.h

@@ -6,7 +6,6 @@
  * LICENSE file in the root directory of this source tree.
  */
 
-// Copyright 2004-present Facebook. All Rights Reserved
 // -*- c++ -*-
 
 #ifndef INDEX_FLAT_H
@@ -24,7 +23,7 @@ struct IndexFlat: Index {
     /// database vectors, size ntotal * d
     std::vector<float> xb;
 
-    explicit IndexFlat (idx_t d, MetricType metric = METRIC_INNER_PRODUCT);
+    explicit IndexFlat (idx_t d, MetricType metric = METRIC_L2);
 
     void add(idx_t n, const float* x) override;
 

IndexHNSW.cpp

@@ -6,6 +6,8 @@
  * LICENSE file in the root directory of this source tree.
  */
 
+// -*- c++ -*-
+
 #include "IndexHNSW.h"
 
 
@@ -1694,19 +1696,17 @@ struct PQDis: HNSW::DistanceComputer {
         return accu;
     }
 
-
-    PQDis(const IndexPQ & storage, const float *q = nullptr):
-        pq(storage.pq)
-    {
-        precomputed_table.resize(pq.M * pq.ksub);
-        nb = storage.ntotal;
-        d = storage.d;
-        codes = storage.codes.data();
-        code_size = pq.code_size;
-        FAISS_ASSERT(pq.ksub == 256);
-        FAISS_ASSERT(pq.sdc_table.size() == pq.ksub * pq.ksub * pq.M);
-        sdc = pq.sdc_table.data();
-        ndis = 0;
+    PQDis(const IndexPQ& storage, const float* /*q*/ = nullptr)
+        : pq(storage.pq) {
+      precomputed_table.resize(pq.M * pq.ksub);
+      nb = storage.ntotal;
+      d = storage.d;
+      codes = storage.codes.data();
+      code_size = pq.code_size;
+      FAISS_ASSERT(pq.ksub == 256);
+      FAISS_ASSERT(pq.sdc_table.size() == pq.ksub * pq.ksub * pq.M);
+      sdc = pq.sdc_table.data();
+      ndis = 0;
     }
 
     void set_query(const float *x) override {
@@ -1771,15 +1771,13 @@ struct SQDis: HNSW::DistanceComputer {
         return dc->compute_code_distance (codei, codej);
     }
 
-
-    SQDis(const IndexScalarQuantizer & storage, const float *q = nullptr):
-        sq(storage.sq)
-    {
-        nb = storage.ntotal;
-        d = storage.d;
-        codes = storage.codes.data();
-        code_size = sq.code_size;
-        dc = sq.get_distance_computer();
+    SQDis(const IndexScalarQuantizer& storage, const float* /*q*/ = nullptr)
+        : sq(storage.sq) {
+      nb = storage.ntotal;
+      d = storage.d;
+      codes = storage.codes.data();
+      code_size = sq.code_size;
+      dc = sq.get_distance_computer();
     }
 
     void set_query(const float *x) override {

IndexHNSW.h

@@ -6,6 +6,8 @@
  * LICENSE file in the root directory of this source tree.
  */
 
+// -*- c++ -*-
+
 #pragma once
 
 #include <vector>
@@ -320,5 +322,4 @@ struct IndexHNSW2Level: IndexHNSW {
 };
 
 
-
-};
+}

IndexIVF.cpp

@@ -6,9 +6,7 @@
  * LICENSE file in the root directory of this source tree.
  */
 
-/* Copyright 2004-present Facebook. All Rights Reserved.
-   Inverted list structure.
-*/
+// -*- c++ -*-
 
 #include "IndexIVF.h"
 
@@ -367,9 +365,10 @@ void IndexIVF::search_and_reconstruct (idx_t n, const float *x, idx_t k,
     }
 }
 
-void IndexIVF::reconstruct_from_offset (long list_no, long offset,
-                                        float* recons) const
-{
+void IndexIVF::reconstruct_from_offset(
+    long /*list_no*/,
+    long /*offset*/,
+    float* /*recons*/) const {
   FAISS_THROW_MSG ("reconstruct_from_offset not implemented");
 }
 
@@ -607,5 +606,4 @@ void IndexIVFStats::reset()
 IndexIVFStats indexIVF_stats;
 
 
-
 } // namespace faiss

IndexIVF.h

@@ -6,7 +6,6 @@
  * LICENSE file in the root directory of this source tree.
  */
 
-// Copyright 2004-present Facebook. All Rights Reserved.
 // -*- c++ -*-
 
 #ifndef FAISS_INDEX_IVF_H
@@ -143,6 +142,12 @@ struct ArrayInvertedLists: InvertedLists {
 };
 
 
+struct IVFSearchParameters {
+    size_t nprobe;            ///< number of probes at query time
+    size_t max_codes;         ///< max nb of codes to visit to do a query
+    virtual ~IVFSearchParameters () {}
+};
+
 /** Index based on a inverted file (IVF)
  *
  * In the inverted file, the quantizer (an Index instance) provides a
@@ -213,12 +218,15 @@ struct IndexIVF: Index, Level1Quantizer {
      * @param store_pairs store inv list index + inv list offset
      *                     instead in upper/lower 32 bit of result,
      *                     instead of ids (used for reranking).
+     * @param params used to override the object's search parameters
      */
     virtual void search_preassigned (idx_t n, const float *x, idx_t k,
                                      const idx_t *assign,
                                      const float *centroid_dis,
                                      float *distances, idx_t *labels,
-                                     bool store_pairs) const = 0;
+                                     bool store_pairs,
+                                     const IVFSearchParameters *params=nullptr
+                                     ) const = 0;
 
     /** assign the vectors, then call search_preassign */
     virtual void search (idx_t n, const float *x, idx_t k,
@@ -317,11 +325,7 @@ struct IndexIVFStats {
 extern IndexIVFStats indexIVF_stats;
 
 
-
 } // namespace faiss
 
 
-
-
-
 #endif

IndexIVFFlat.h

@@ -11,9 +11,9 @@
 #ifndef FAISS_INDEX_IVF_FLAT_H
 #define FAISS_INDEX_IVF_FLAT_H
 
-#include "IndexIVF.h"
-
+#include <unordered_map>
 
+#include "IndexIVF.h"
 
 
 namespace faiss {
@@ -39,7 +39,9 @@ struct IndexIVFFlat: IndexIVF {
                              const idx_t *assign,
                              const float *centroid_dis,
                              float *distances, idx_t *labels,
-                             bool store_pairs) const override;
+                             bool store_pairs,
+                             const IVFSearchParameters *params=nullptr
+                             ) const override;
 
     void range_search(
         idx_t n,
@@ -55,7 +57,7 @@ struct IndexIVFFlat: IndexIVF {
      * @param idx    vector indices to update, size nv
      * @param v      vectors of new values, size nv*d
      */
-    void update_vectors (int nv, idx_t *idx, const float *v);
+    virtual void update_vectors (int nv, idx_t *idx, const float *v);
 
     void reconstruct_from_offset (long list_no, long offset,
                                   float* recons) const override;
@@ -64,6 +66,55 @@ struct IndexIVFFlat: IndexIVF {
 };
 
 
+struct IndexIVFFlatDedup: IndexIVFFlat {
+
+    /** Maps ids stored in the index to the ids of vectors that are
+     *  the same. When a vector is unique, it does not appear in the
+     *  instances map */
+    std::unordered_multimap <idx_t, idx_t> instances;
+
+    IndexIVFFlatDedup (
+            Index * quantizer, size_t d, size_t nlist_,
+            MetricType = METRIC_L2);
+
+    /// also dedups the training set
+    void train(idx_t n, const float* x) override;
+
+    /// implemented for all IndexIVF* classes
+    void add_with_ids(idx_t n, const float* x, const long* xids) override;
+
+    void search_preassigned (idx_t n, const float *x, idx_t k,
+                             const idx_t *assign,
+                             const float *centroid_dis,
+                             float *distances, idx_t *labels,
+                             bool store_pairs,
+                             const IVFSearchParameters *params=nullptr
+                             ) const override;
+
+    long remove_ids(const IDSelector& sel) override;
+
+    /// not implemented
+    void range_search(
+        idx_t n,
+        const float* x,
+        float radius,
+        RangeSearchResult* result) const override;
+
+    /// not implemented
+    void update_vectors (int nv, idx_t *idx, const float *v) override;
+
+
+    /// not implemented
+    void reconstruct_from_offset (long list_no, long offset,
+                                  float* recons) const override;
+
+    IndexIVFFlatDedup () {}
+
+
+};
+
+
+
 } // namespace faiss
 
 #endif

IndexIVFPQ.cpp

@@ -6,9 +6,7 @@
  * LICENSE file in the root directory of this source tree.
  */
 
-/* Copyright 2004-present Facebook. All Rights Reserved.
-   Inverted list structure.
-*/
+// -*- c++ -*-
 
 #include "IndexIVFPQ.h"
 
@@ -33,9 +31,6 @@
 namespace faiss {
 
 
-
-
-
 /*****************************************
  * IndexIVFPQ implementation
  ******************************************/
@@ -452,6 +447,7 @@ struct QueryTables {
      *****************************************************/
 
     const IndexIVFPQ & ivfpq;
+    const IVFSearchParameters *params;
 
     // copied from IndexIVFPQ for easier access
     int d;
@@ -459,6 +455,7 @@ struct QueryTables {
     MetricType metric_type;
     bool by_residual;
     int use_precomputed_table;
+    int polysemous_ht;
 
     // pre-allocated data buffers
     float * sim_table, * sim_table_2;
@@ -470,7 +467,8 @@ struct QueryTables {
     // for table pointers
     std::vector<const float *> sim_table_ptrs;
 
-    explicit QueryTables (const IndexIVFPQ & ivfpq):
+    explicit QueryTables (const IndexIVFPQ & ivfpq,
+                          const IVFSearchParameters *params):
         ivfpq(ivfpq),
         d(ivfpq.d),
         pq (ivfpq.pq),
@@ -485,7 +483,12 @@ struct QueryTables {
         decoded_vec = residual_vec + d;
 
         // for polysemous
-        if (ivfpq.polysemous_ht != 0)  {
+        polysemous_ht = ivfpq.polysemous_ht;
+        if (auto ivfpq_params =
+            dynamic_cast<const IVFPQSearchParameters *>(params)) {
+            polysemous_ht = ivfpq_params->polysemous_ht;
+        }
+        if (polysemous_ht != 0)  {
             q_code.resize (pq.code_size);
         }
         init_list_cycles = 0;
@@ -506,7 +509,7 @@ struct QueryTables {
             init_query_IP ();
         else
             init_query_L2 ();
-        if (!by_residual && ivfpq.polysemous_ht != 0)
+        if (!by_residual && polysemous_ht != 0)
             pq.compute_code (qi, q_code.data());
     }
 
@@ -579,7 +582,7 @@ struct QueryTables {
         // decoded_vec = centroid
         float dis0 = -fvec_inner_product (qi, decoded_vec, d);
 
-        if (ivfpq.polysemous_ht) {
+        if (polysemous_ht) {
             for (int i = 0; i < d; i++) {
                 residual_vec [i] = qi[i] - decoded_vec[i];
             }
@@ -629,7 +632,7 @@ struct QueryTables {
                 const float *pc = &ivfpq.precomputed_table
                     [(ki * pq.M + cm * Mf) * pq.ksub];
 
-                if (ivfpq.polysemous_ht == 0) {
+                if (polysemous_ht == 0) {
 
                     // sum up with query-specific table
                     fvec_madd (Mf * pq.ksub,
@@ -694,7 +697,7 @@ struct QueryTables {
           FAISS_THROW_MSG ("need precomputed tables");
         }
 
-        if (ivfpq.polysemous_ht) {
+        if (polysemous_ht) {
             FAISS_THROW_MSG ("not implemented");
             // Not clear that it makes sense to implemente this,
             // because it costs M * ksub, which is what we wanted to
@@ -720,8 +723,9 @@ struct InvertedListScanner: QueryTables {
     const IDType * list_ids;
     size_t list_size;
 
-    explicit InvertedListScanner (const IndexIVFPQ & ivfpq):
-        QueryTables (ivfpq)
+    explicit InvertedListScanner (const IndexIVFPQ & ivfpq,
+                                  const IVFSearchParameters *params):
+        QueryTables (ivfpq, params)
     {
         FAISS_THROW_IF_NOT (pq.byte_per_idx == 1);
         n_hamming_pass = 0;
@@ -931,27 +935,30 @@ void IndexIVFPQ::search_preassigned (idx_t nx, const float *qx, idx_t k,
                                      const idx_t *keys,
                                      const float *coarse_dis,
                                      float *distances, idx_t *labels,
-                                     bool store_pairs) const
+                                     bool store_pairs,
+                                     const IVFSearchParameters *params
+                                     ) const
 {
     float_maxheap_array_t res = {
         size_t(nx), size_t(k),
         labels, distances
     };
 
+    long local_nprobe = params ? params->nprobe : nprobe;
+    long local_max_codes = params ? params->max_codes : max_codes;
+
 #pragma omp parallel
     {
-        InvertedListScanner<long> qt (*this);
-        size_t stats_nlist = 0;
-        size_t stats_ncode = 0;
+        InvertedListScanner<long> qt (*this, params);
+        size_t stats_nlist = 0, stats_ncode = 0;
         uint64_t init_query_cycles = 0;
-        uint64_t scan_cycles = 0;
-        uint64_t heap_cycles = 0;
+        uint64_t scan_cycles = 0, heap_cycles = 0;
 
 #pragma omp  for
         for (size_t i = 0; i < nx; i++) {
             const float *qi = qx + i * d;
-            const long * keysi = keys + i * nprobe;
-            const float *coarse_dis_i = coarse_dis + i * nprobe;
+            const long * keysi = keys + i * local_nprobe;
+            const float *coarse_dis_i = coarse_dis + i * local_nprobe;
             float * heap_sim = res.get_val (i);
             long * heap_ids = res.get_ids (i);
 
@@ -966,7 +973,7 @@ void IndexIVFPQ::search_preassigned (idx_t nx, const float *qx, idx_t k,
 
             size_t nscan = 0;
 
-            for (size_t ik = 0; ik < nprobe; ik++) {
+            for (size_t ik = 0; ik < local_nprobe; ik++) {
                 long key = keysi[ik];  /* select the list  */
                 if (key < 0) {
                     // not enough centroids for multiprobe
@@ -994,7 +1001,7 @@ void IndexIVFPQ::search_preassigned (idx_t nx, const float *qx, idx_t k,
                 }
                 scan_cycles += TOC;
 
-                if (max_codes && nscan >= max_codes) break;
+                if (local_max_codes && nscan >= local_max_codes) break;
             }
             stats_ncode += nscan;
             TIC;
@@ -1160,7 +1167,9 @@ void IndexIVFPQR::search_preassigned (idx_t n, const float *x, idx_t k,
                                       const idx_t *idx,
                                       const float *L1_dis,
                                       float *distances, idx_t *labels,
-                                      bool store_pairs) const
+                                      bool store_pairs,
+                                      const IVFSearchParameters *params
+                                      ) const
 {
     uint64_t t0;
     TIC;
@@ -1171,9 +1180,10 @@ void IndexIVFPQR::search_preassigned (idx_t n, const float *x, idx_t k,
         float *coarse_distances = new float [k_coarse * n];
         ScopeDeleter<float> del(coarse_distances);
 
-        IndexIVFPQ::search_preassigned (n, x, k_coarse,
-                            idx, L1_dis, coarse_distances, coarse_labels,
-                            true);
+        IndexIVFPQ::search_preassigned (
+                   n, x, k_coarse,
+                   idx, L1_dis, coarse_distances, coarse_labels,
+                   true, params);
     }
 
 
@@ -1392,13 +1402,12 @@ void Index2Layer::add(idx_t n, const float* x)
 }
 
 void Index2Layer::search(
-        idx_t n,
-        const float* x,
-        idx_t k,
-        float* distances,
-        idx_t* labels) const
-{
-    FAISS_THROW_MSG ("not implemented");
+    idx_t /*n*/,
+    const float* /*x*/,
+    idx_t /*k*/,
+    float* /*distances*/,
+    idx_t* /*labels*/) const {
+  FAISS_THROW_MSG("not implemented");
 }
 
 
@@ -1456,10 +1465,4 @@ void Index2Layer::reset()
 }
 
 
-
-
-
-
-
-
 } // namespace faiss

IndexIVFPQ.h

@@ -6,7 +6,6 @@
  * LICENSE file in the root directory of this source tree.
  */
 
-// Copyright 2004-present Facebook. All Rights Reserved.
 // -*- c++ -*-
 
 #ifndef FAISS_INDEX_IVFPQ_H
@@ -21,6 +20,11 @@
 
 namespace faiss {
 
+struct IVFPQSearchParameters: IVFSearchParameters {
+    size_t scan_table_threshold;   ///< use table computation or on-the-fly?
+    int polysemous_ht;             ///< Hamming thresh for polysemous filtering
+    ~IVFPQSearchParameters () {}
+};
 
 
 /** Inverted file with Product Quantizer encoding. Each residual
@@ -103,7 +107,9 @@ struct IndexIVFPQ: IndexIVF {
                              const idx_t *assign,
                              const float *centroid_dis,
                              float *distances, idx_t *labels,
-                             bool store_pairs) const override;
+                             bool store_pairs,
+                             const IVFSearchParameters *params=nullptr
+                             ) const override;
 
     /// build precomputed table
     void precompute_table ();
@@ -181,7 +187,9 @@ struct IndexIVFPQR: IndexIVFPQ {
                              const idx_t *assign,
                              const float *centroid_dis,
                              float *distances, idx_t *labels,
-                             bool store_pairs) const override;
+                             bool store_pairs,
+                             const IVFSearchParameters *params=nullptr
+                             ) const override;
 
     IndexIVFPQR();
 };
@@ -242,11 +250,7 @@ struct Index2Layer: Index {
 };
 
 
-
 } // namespace faiss
 
 
-
-
-
 #endif

IndexLSH.cpp

@@ -6,7 +6,7 @@
  * LICENSE file in the root directory of this source tree.
  */
 
-// Copyright 2004-present Facebook. All Rights Reserved.
+// -*- c++ -*-
 
 #include "IndexLSH.h"
 
@@ -19,6 +19,7 @@
 #include "hamming.h"
 #include "FaissAssert.h"
 
+
 namespace faiss {
 
 /***************************************************************
@@ -146,8 +147,8 @@ void IndexLSH::search (
 
     int_maxheap_array_t res = { size_t(n), size_t(k), labels, idistances};
 
-    hammings_knn (&res, qcodes, codes.data(),
-                  ntotal, bytes_per_vec, true);
+    hammings_knn_hc (&res, qcodes, codes.data(),
+                     ntotal, bytes_per_vec, true);
 
 
     // convert distances to floats
@@ -176,5 +177,4 @@ void IndexLSH::reset() {
 }
 
 
-
 } // namespace faiss

IndexLSH.h

@@ -6,7 +6,6 @@
  * LICENSE file in the root directory of this source tree.
  */
 
-// Copyright 2004-present Facebook. All Rights Reserved.
 // -*- c++ -*-
 
 #ifndef INDEX_LSH_H
@@ -73,12 +72,7 @@ struct IndexLSH:Index {
 };
 
 
-
 }
 
 
-
-
-
-
 #endif

IndexPQ.cpp

@@ -6,9 +6,7 @@
  * LICENSE file in the root directory of this source tree.
  */
 
-/* Copyright 2004-present Facebook. All Rights Reserved.
-   Index based on product quantiztion.
-*/
+// -*- c++ -*-
 
 #include "IndexPQ.h"
 
@@ -203,13 +201,13 @@ void IndexPQ::search (idx_t n, const float *x, idx_t k,
 
             if (search_type == ST_HE) {
 
-                hammings_knn (&res, q_codes, codes.data(),
-                              ntotal, pq.code_size, true);
+                hammings_knn_hc (&res, q_codes, codes.data(),
+                                 ntotal, pq.code_size, true);
 
             } else if (search_type == ST_generalized_HE) {
 
-                generalized_hammings_knn (&res, q_codes, codes.data(),
-                                          ntotal, pq.code_size, true);
+                generalized_hammings_knn_hc (&res, q_codes, codes.data(),
+                                             ntotal, pq.code_size, true);
             }
 
             // convert distances to floats
@@ -1092,5 +1090,4 @@ void MultiIndexQuantizer2::search(
 }
 
 
-
-} // END namespace faiss
+} // namespace faiss

IndexPQ.h

@@ -6,7 +6,6 @@
  * LICENSE file in the root directory of this source tree.
  */
 
-// Copyright 2004-present Facebook. All Rights Reserved.
 // -*- c++ -*-
 
 #ifndef FAISS_INDEX_PQ_H
@@ -63,6 +62,8 @@ struct IndexPQ: Index {
 
     void reconstruct(idx_t key, float* recons) const override;
 
+    long remove_ids(const IDSelector& sel) override;
+
     /******************************************************
      * Polysemous codes implementation
      ******************************************************/
@@ -83,11 +84,6 @@ struct IndexPQ: Index {
 
     Search_type_t search_type;
 
-    /** remove some ids. NB that Because of the structure of the
-     * indexing structre, the semantics of this operation are
-     * different from the usual ones: the new ids are shifted */
-    long remove_ids(const IDSelector& sel) override;
-
     // just encode the sign of the components, instead of using the PQ encoder
     // used only for the queries
     bool encode_signs;
@@ -189,5 +185,4 @@ struct MultiIndexQuantizer2: MultiIndexQuantizer {
 } // namespace faiss
 
 
-
 #endif

IndexScalarQuantizer.h

@@ -6,6 +6,8 @@
  * LICENSE file in the root directory of this source tree.
  */
 
+// -*- c++ -*-
+
 #ifndef FAISS_INDEX_SCALAR_QUANTIZER_H
 #define FAISS_INDEX_SCALAR_QUANTIZER_H
 
@@ -34,6 +36,7 @@ struct ScalarQuantizer {
         QT_4bit,             ///< 4 bits per component
         QT_8bit_uniform,     ///< same, shared range for all dimensions
         QT_4bit_uniform,
+        QT_fp16,
     };
 
     QuantizerType qtype;
@@ -160,7 +163,9 @@ struct IndexIVFScalarQuantizer: IndexIVF {
                              const idx_t *assign,
                              const float *centroid_dis,
                              float *distances, idx_t *labels,
-                             bool store_pairs) const override;
+                             bool store_pairs,
+                             const IVFSearchParameters *params=nullptr
+                             ) const override;
 
     void reconstruct_from_offset (long list_no, long offset,
                                   float* recons) const override;

MetaIndexes.cpp

@@ -6,7 +6,6 @@
  * LICENSE file in the root directory of this source tree.
  */
 
-// Copyright 2004-present Facebook. All Rights Reserved
 // -*- c++ -*-
 
 #include "MetaIndexes.h"
@@ -729,8 +728,4 @@ IndexSplitVectors::~IndexSplitVectors ()
 }
 
 
-
-
-
-
-}; // namespace faiss
+} // namespace faiss

MetaIndexes.h

@@ -6,7 +6,6 @@
  * LICENSE file in the root directory of this source tree.
  */
 
-// Copyright 2004-present Facebook. All Rights Reserved.
 // -*- c++ -*-
 
 #ifndef META_INDEXES_H
@@ -170,8 +169,7 @@ struct IndexSplitVectors: Index {
 };
 
 
-
-}
+} // namespace faiss
 
 
 #endif

OnDiskInvertedLists.cpp

@@ -6,6 +6,7 @@
  * LICENSE file in the root directory of this source tree.
  */
 
+// -*- c++ -*-
 
 #include "OnDiskInvertedLists.h"
 
@@ -223,17 +224,7 @@ int OnDiskInvertedLists::OngoingPrefetch::global_cs = 0;
 
 void OnDiskInvertedLists::prefetch_lists (const long *list_nos, int n) const
 {
-    if (use_madvise) {
-        for (int i = 0; i < n; ++i) {
-            const auto list_no = list_nos[i];
-            const auto size = list_size(list_no);
-
-            madvise((void *) get_ids(list_no), size * sizeof(Index::idx_t), MADV_WILLNEED);
-            madvise((void *) get_codes(list_no), size * code_size, MADV_WILLNEED);
-        }
-    } else {
-        pf->prefetch_lists (list_nos, n);
-    }
+    pf->prefetch_lists (list_nos, n);
 }
 
 
@@ -337,7 +328,6 @@ OnDiskInvertedLists::OnDiskInvertedLists (
     totsize (0),
     ptr (nullptr),
     read_only (false),
-    use_madvise (false),
     locks (new LockLevels ()),
     pf (new OngoingPrefetch (this))
 {
@@ -351,7 +341,6 @@ OnDiskInvertedLists::OnDiskInvertedLists ():
     totsize (0),
     ptr (nullptr),
     read_only (false),
-    use_madvise (false),
     locks (new LockLevels ()),
     pf (new OngoingPrefetch (this))
 {
@@ -382,11 +371,19 @@ size_t OnDiskInvertedLists::list_size(size_t list_no) const
 
 const uint8_t * OnDiskInvertedLists::get_codes (size_t list_no) const
 {
+    if (lists[list_no].offset == INVALID_OFFSET) {
+        return nullptr;
+    }
+
     return ptr + lists[list_no].offset;
 }
 
 const Index::idx_t * OnDiskInvertedLists::get_ids (size_t list_no) const
 {
+    if (lists[list_no].offset == INVALID_OFFSET) {
+        return nullptr;
+    }
+
     return (const idx_t*)(ptr + lists[list_no].offset +
                           code_size * lists[list_no].capacity);
 }
@@ -606,8 +603,4 @@ size_t OnDiskInvertedLists::merge_from (const InvertedLists **ils, int n_il)
 }
 
 
-
-
-
-
 } // namespace faiss

OnDiskInvertedLists.h

@@ -6,7 +6,7 @@
  * LICENSE file in the root directory of this source tree.
  */
 
-
+// -*- c++ -*-
 
 #ifndef FAISS_ON_DISK_INVERTED_LISTS_H
 #define FAISS_ON_DISK_INVERTED_LISTS_H
@@ -73,8 +73,7 @@ struct OnDiskInvertedLists: InvertedLists {
     size_t totsize;
     uint8_t *ptr; // mmap base pointer
     bool read_only;  /// are inverted lists mapped read-only
-    bool use_madvise; // use madvice for prefetching
-    
+
     OnDiskInvertedLists (size_t nlist, size_t code_size,
                          const char *filename);
 
@@ -118,8 +117,6 @@ struct OnDiskInvertedLists: InvertedLists {
 };
 
 
-
-
 } // namespace faiss
 
 #endif

PolysemousTraining.cpp

@@ -6,6 +6,8 @@
  * LICENSE file in the root directory of this source tree.
  */
 
+// -*- c++ -*-
+
 #include "PolysemousTraining.h"
 
 #include <cstdlib>
@@ -946,5 +948,4 @@ void PolysemousTraining::optimize_pq_for_hamming (ProductQuantizer &pq,
 }
 
 
-
 } // namespace faiss

PolysemousTraining.h

@@ -6,7 +6,6 @@
  * LICENSE file in the root directory of this source tree.
  */
 
-// Copyright 2004-present Facebook. All Rights Reserved.
 // -*- c++ -*-
 
 #ifndef FAISS_POLYSEMOUS_TRAINING_INCLUDED
@@ -19,8 +18,6 @@
 namespace faiss {
 
 
-
-
 /// parameters used for the simulated annealing method
 struct SimulatedAnnealingParameters {
 
@@ -156,7 +153,6 @@ struct PolysemousTraining: SimulatedAnnealingParameters {
 };
 
 
-
 } // namespace faiss
 
 

ProductQuantizer.cpp

@@ -6,9 +6,7 @@
  * LICENSE file in the root directory of this source tree.
  */
 
-/* Copyright 2004-present Facebook. All Rights Reserved.
-   Index based on product quantiztion.
-*/
+// -*- c++ -*-
 
 #include "ProductQuantizer.h"
 
@@ -40,8 +38,6 @@ int sgemm_ (const char *transa, const char *transb, FINTEGER *m, FINTEGER *
 namespace faiss {
 
 
-
-
 /* compute an estimator using look-up tables for typical values of M */
 template <typename CT, class C>
 void pq_estimators_from_tables_Mmul4 (int M, const CT * codes,
@@ -661,8 +657,4 @@ void ProductQuantizer::search_sdc (const uint8_t * qcodes,
 }
 
 
-
-
-
-
 } // namespace faiss

ProductQuantizer.h

@@ -6,7 +6,6 @@
  * LICENSE file in the root directory of this source tree.
  */
 
-// Copyright 2004-present Facebook. All Rights Reserved.
 // -*- c++ -*-
 
 #ifndef FAISS_PRODUCT_QUANTIZER_H
@@ -171,7 +170,6 @@ struct ProductQuantizer {
 };
 
 
-
 } // namespace faiss
 
 

VectorTransform.cpp

@@ -6,7 +6,6 @@
  * LICENSE file in the root directory of this source tree.
  */
 
-// Copyright 2004-present Facebook. All Rights Reserved
 // -*- c++ -*-
 
 #include "VectorTransform.h"

VectorTransform.h

@@ -6,7 +6,6 @@
  * LICENSE file in the root directory of this source tree.
  */
 
-// Copyright 2004-present Facebook. All Rights Reserved.
 // -*- c++ -*-
 
 #ifndef FAISS_VECTOR_TRANSFORM_H
@@ -304,9 +303,7 @@ struct IndexPreTransform: Index {
 };
 
 
-
 } // namespace faiss
 
 
-
 #endif

benchs/bench_gpu_1bn.py

@@ -552,10 +552,7 @@ def compute_populated_index(preproc):
     t0 = time.time()
 
     for i in range(ngpu):
-        if ngpu == 1:
-            index_src = faiss.index_gpu_to_cpu(gpu_index)
-        else:
-            index_src = faiss.index_gpu_to_cpu(gpu_index.at(i))
+        index_src = faiss.index_gpu_to_cpu(gpu_index.at(i))
         print "  index %d size %d" % (i, index_src.ntotal)
         index_src.copy_subset_to(indexall, 0, 0, nb)
 

benchs/bench_vector_ops.py

+# Copyright (c) 2015-present, Facebook, Inc.
+# All rights reserved.
+#
+# This source code is licensed under the BSD+Patents license found in the
+# LICENSE file in the root directory of this source tree.
+
+#! /usr/bin/env python2
+
+import numpy as np
+import faiss
+import time
+
+xd = 100
+yd = 1000000
+
+np.random.seed(1234)
+
+faiss.omp_set_num_threads(1)
+
+print 'xd=%d yd=%d' % (xd, yd)
+
+print 'Running inner products test..'
+for d in 3, 4, 12, 36, 64:
+
+    x = faiss.rand(xd * d).reshape(xd, d)
+    y = faiss.rand(yd * d).reshape(yd, d)
+
+    distances = np.empty((xd, yd), dtype='float32')
+
+    t0 = time.time()
+    for i in xrange(xd):
+        faiss.fvec_inner_products_ny(faiss.swig_ptr(distances[i]),
+                                     faiss.swig_ptr(x[i]),
+                                     faiss.swig_ptr(y),
+                                     d, yd)
+    t1 = time.time()
+
+    # sparse verification
+    ntry = 100
+    num, denom = 0, 0
+    for t in range(ntry):
+        xi = np.random.randint(xd)
+        yi = np.random.randint(yd)
+        num += abs(distances[xi, yi] - np.dot(x[xi], y[yi]))
+        denom += abs(distances[xi, yi])
+
+    print 'd=%d t=%.3f s diff=%g' % (d, t1 - t0, num / denom)
+
+
+print 'Running L2sqr test..'
+for d in 3, 4, 12, 36, 64:
+
+    x = faiss.rand(xd * d).reshape(xd, d)
+    y = faiss.rand(yd * d).reshape(yd, d)
+
+    distances = np.empty((xd, yd), dtype='float32')
+
+    t0 = time.time()
+    for i in xrange(xd):
+        faiss.fvec_L2sqr_ny(faiss.swig_ptr(distances[i]),
+                            faiss.swig_ptr(x[i]),
+                            faiss.swig_ptr(y),
+                            d, yd)
+    t1 = time.time()
+
+    # sparse verification
+    ntry = 100
+    num, denom = 0, 0
+    for t in range(ntry):
+        xi = np.random.randint(xd)
+        yi = np.random.randint(yd)
+        num += abs(distances[xi, yi] - np.sum((x[xi] - y[yi]) ** 2))
+        denom += abs(distances[xi, yi])
+
+    print 'd=%d t=%.3f s diff=%g' % (d, t1 - t0, num / denom)

benchs/link_and_code/README.md

 
 
+
 README for the link & code implementation
 =========================================
 
@@ -10,14 +11,12 @@ Link & code is an indexing method that combines HNSW indexing with
 compression and exploits the neighborhood structure of the similarity
 graph to improve the reconstruction. It is described in
 
-```
 @inproceedings{link_and_code,
    author = {Matthijs Douze and Alexandre Sablayrolles and Herv\'e J\'egou},
    title = {Link and code: Fast indexing with graphs and compact regression codes},
    booktitle = {CVPR},
    year = {2018}
 }
-```
 
 ArXiV [here](https://arxiv.org/abs/1804.09996)
 
@@ -30,7 +29,7 @@ The test runs with 3 files:
 
 - `datasets.py`: code to load the datasets. The example code runs on the
   deep1b and bigann datasets. See the [toplevel README](../README.md)
-  on how to download them. They should be put in a directory, edit
+  on how to downlod them. They should be put in a directory, edit
   datasets.py to set the path.
 
 - `neighbor_codec.py`: this is where the representation is trained.
@@ -62,7 +61,7 @@ Set `bdir` to a scratch directory.
 
 Explanation of the flags:
 
-- `--db deep100M`: dataset to process
+- `--db deep1M`: dataset to process
 
 - `--M0 6`: number of links on the base level (L6)
 
@@ -135,7 +134,7 @@ python bench_link_and_code.py \
    --beta_nsq 1 --beta_k 1  \
    --beta_centroids $bdir/deep1M_PQ40_M6_nsq0.npy \
    --neigh_recons_codes $bdir/deep1M_PQ36_M6_nsq0_codes.npy \
-   --k_reorder 0,5 --efSearch 16,64,256,1024
+   --k_reorder 0 --efSearch 16,64,256,1024
 ```
 
 The arguments are similar to the previous table. Note that nsq = 0 is

benchs/link_and_code/bench_link_and_code.py

+# Copyright (c) 2015-present, Facebook, Inc.
+# All rights reserved.
+#
+# This source code is licensed under the BSD+Patents license found in the
+# LICENSE file in the root directory of this source tree.
+
 #!/usr/bin/env python2
 
 import os

benchs/link_and_code/datasets.py

+# Copyright (c) 2015-present, Facebook, Inc.
+# All rights reserved.
+#
+# This source code is licensed under the BSD+Patents license found in the
+# LICENSE file in the root directory of this source tree.
+
 #! /usr/bin/env python2
 
 """

benchs/link_and_code/neighbor_codec.py

+# Copyright (c) 2015-present, Facebook, Inc.
+# All rights reserved.
+#
+# This source code is licensed under the BSD+Patents license found in the
+# LICENSE file in the root directory of this source tree.
+
 #! /usr/bin/env python2
 
 """

demos/demo_imi_flat.cpp

@@ -6,7 +6,6 @@
  * LICENSE file in the root directory of this source tree.
  */
 
-// Copyright 2004-present Facebook. All Rights Reserved
 
 
 #include <cmath>

demos/demo_imi_pq.cpp

@@ -6,7 +6,6 @@
  * LICENSE file in the root directory of this source tree.
  */
 
-// Copyright 2004-present Facebook. All Rights Reserved
 
 
 #include <cmath>

demos/demo_ivfpq_indexing.cpp

@@ -6,7 +6,6 @@
  * LICENSE file in the root directory of this source tree.
  */
 
-// Copyright 2004-present Facebook. All Rights Reserved
 
 
 #include <cmath>

demos/demo_sift1M.cpp

@@ -6,7 +6,6 @@
  * LICENSE file in the root directory of this source tree.
  */
 
- // Copyright 2004-present Facebook. All Rights Reserved
 
 
 #include <cmath>

gpu/GpuAutoTune.cpp

@@ -6,7 +6,6 @@
  * LICENSE file in the root directory of this source tree.
  */
 
-// Copyright 2004-present Facebook. All Rights Reserved.
 #include "GpuAutoTune.h"
 #include <typeinfo>
 

gpu/GpuAutoTune.h

@@ -6,7 +6,6 @@
  * LICENSE file in the root directory of this source tree.
  */
 
-// Copyright 2004-present Facebook. All Rights Reserved.
 #pragma once
 
 #include "../Index.h"

gpu/GpuClonerOptions.cpp

@@ -6,7 +6,6 @@
  * LICENSE file in the root directory of this source tree.
  */
 
-// Copyright 2004-present Facebook. All Rights Reserved.
 #include "GpuClonerOptions.h"
 
 namespace faiss { namespace gpu {

gpu/GpuClonerOptions.h

@@ -6,7 +6,6 @@
  * LICENSE file in the root directory of this source tree.
  */
 
-// Copyright 2004-present Facebook. All Rights Reserved.
 #pragma once
 
 #include "GpuIndicesOptions.h"

gpu/GpuFaissAssert.h

@@ -6,7 +6,6 @@
  * LICENSE file in the root directory of this source tree.
  */
 
-// Copyright 2004-present Facebook. All Rights Reserved.
 
 #ifndef GPU_FAISS_ASSERT_INCLUDED
 #define GPU_FAISS_ASSERT_INCLUDED

gpu/GpuIndex.cu

@@ -6,7 +6,6 @@
  * LICENSE file in the root directory of this source tree.
  */
 
-// Copyright 2004-present Facebook. All Rights Reserved.
 
 #include "GpuIndex.h"
 #include "../FaissAssert.h"

gpu/GpuIndex.h

@@ -6,7 +6,6 @@
  * LICENSE file in the root directory of this source tree.
  */
 
-// Copyright 2004-present Facebook. All Rights Reserved.
 
 #pragma once
 

gpu/GpuIndexFlat.cu

@@ -6,7 +6,6 @@
  * LICENSE file in the root directory of this source tree.
  */
 
-// Copyright 2004-present Facebook. All Rights Reserved.
 
 #include "GpuIndexFlat.h"
 #include "../IndexFlat.h"

gpu/GpuIndexFlat.h

@@ -6,7 +6,6 @@
  * LICENSE file in the root directory of this source tree.
  */
 
-// Copyright 2004-present Facebook. All Rights Reserved.
 
 #pragma once
 

gpu/GpuIndexIVF.cu

@@ -6,7 +6,6 @@
  * LICENSE file in the root directory of this source tree.
  */
 
-// Copyright 2004-present Facebook. All Rights Reserved.
 
 #include "GpuIndexIVF.h"
 #include "../FaissAssert.h"

gpu/GpuIndexIVF.h

@@ -6,7 +6,6 @@
  * LICENSE file in the root directory of this source tree.
  */
 
-// Copyright 2004-present Facebook. All Rights Reserved.
 
 #pragma once
 

gpu/GpuIndexIVFFlat.cu

@@ -6,7 +6,6 @@
  * LICENSE file in the root directory of this source tree.
  */
 
-// Copyright 2004-present Facebook. All Rights Reserved.
 
 #include "GpuIndexIVFFlat.h"
 #include "../IndexFlat.h"

gpu/GpuIndexIVFFlat.h

@@ -6,7 +6,6 @@
  * LICENSE file in the root directory of this source tree.
  */
 
-// Copyright 2004-present Facebook. All Rights Reserved.
 
 #pragma once
 

gpu/GpuIndexIVFPQ.cu

@@ -6,7 +6,6 @@
  * LICENSE file in the root directory of this source tree.
  */
 
-// Copyright 2004-present Facebook. All Rights Reserved.
 
 #include "GpuIndexIVFPQ.h"
 #include "../IndexFlat.h"

gpu/GpuIndexIVFPQ.h

@@ -6,7 +6,6 @@
  * LICENSE file in the root directory of this source tree.
  */
 
-// Copyright 2004-present Facebook. All Rights Reserved.
 
 #pragma once
 

gpu/GpuIndicesOptions.h

@@ -6,7 +6,6 @@
  * LICENSE file in the root directory of this source tree.
  */
 
-// Copyright 2004-present Facebook. All Rights Reserved.
 
 #pragma once
 

gpu/GpuResources.cpp

@@ -6,7 +6,6 @@
  * LICENSE file in the root directory of this source tree.
  */
 
-// Copyright 2004-present Facebook. All Rights Reserved.
 
 #include "GpuResources.h"
 #include "utils/DeviceUtils.h"

gpu/GpuResources.h

@@ -6,7 +6,6 @@
  * LICENSE file in the root directory of this source tree.
  */
 
-// Copyright 2004-present Facebook. All Rights Reserved.
 
 #pragma once
 

gpu/IndexProxy.cpp

@@ -6,7 +6,6 @@
  * LICENSE file in the root directory of this source tree.
  */
 
-// Copyright 2004-present Facebook. All Rights Reserved.
 
 #include "IndexProxy.h"
 #include "../FaissAssert.h"

gpu/IndexProxy.h

@@ -6,7 +6,6 @@
  * LICENSE file in the root directory of this source tree.
  */
 
-// Copyright 2004-present Facebook. All Rights Reserved.
 
 #pragma once
 

gpu/StandardGpuResources.cpp

@@ -6,7 +6,6 @@
  * LICENSE file in the root directory of this source tree.
  */
 
-// Copyright 2004-present Facebook. All Rights Reserved.
 
 #include "StandardGpuResources.h"
 #include "../FaissAssert.h"

gpu/StandardGpuResources.h

@@ -6,7 +6,6 @@
  * LICENSE file in the root directory of this source tree.
  */
 
-// Copyright 2004-present Facebook. All Rights Reserved.
 
 #pragma once
 

gpu/impl/BroadcastSum.cu

@@ -6,7 +6,6 @@
  * LICENSE file in the root directory of this source tree.
  */
 
-// Copyright 2004-present Facebook. All Rights Reserved.
 
 #include <algorithm>
 #include "../../FaissAssert.h"

gpu/impl/BroadcastSum.cuh

@@ -6,7 +6,6 @@
  * LICENSE file in the root directory of this source tree.
  */
 
-// Copyright 2004-present Facebook. All Rights Reserved.
 
 #pragma once
 

gpu/impl/Distance.cu

@@ -6,7 +6,6 @@
  * LICENSE file in the root directory of this source tree.
  */
 
-// Copyright 2004-present Facebook. All Rights Reserved.
 
 #include "Distance.cuh"
 #include "BroadcastSum.cuh"

gpu/impl/Distance.cuh

@@ -6,7 +6,6 @@
  * LICENSE file in the root directory of this source tree.
  */
 
-// Copyright 2004-present Facebook. All Rights Reserved.
 
 #pragma once
 

gpu/impl/FlatIndex.cu

@@ -6,7 +6,6 @@
  * LICENSE file in the root directory of this source tree.
  */
 
-// Copyright 2004-present Facebook. All Rights Reserved.
 
 #include "FlatIndex.cuh"
 #include "Distance.cuh"

gpu/impl/FlatIndex.cuh

@@ -6,7 +6,6 @@
  * LICENSE file in the root directory of this source tree.
  */
 
-// Copyright 2004-present Facebook. All Rights Reserved.
 
 #pragma once
 

gpu/impl/IVFBase.cu

@@ -6,7 +6,6 @@
  * LICENSE file in the root directory of this source tree.
  */
 
-// Copyright 2004-present Facebook. All Rights Reserved.
 
 #include "IVFBase.cuh"
 #include "../GpuResources.h"

gpu/impl/IVFBase.cuh

@@ -6,7 +6,6 @@
  * LICENSE file in the root directory of this source tree.
  */
 
-// Copyright 2004-present Facebook. All Rights Reserved.
 
 #pragma once
 

gpu/impl/IVFFlat.cu

@@ -6,7 +6,6 @@
  * LICENSE file in the root directory of this source tree.
  */
 
-// Copyright 2004-present Facebook. All Rights Reserved.
 
 #include "IVFFlat.cuh"
 #include "../GpuResources.h"

gpu/impl/IVFFlat.cuh

@@ -6,7 +6,6 @@
  * LICENSE file in the root directory of this source tree.
  */
 
-// Copyright 2004-present Facebook. All Rights Reserved.
 
 #pragma once
 

gpu/impl/IVFFlatScan.cu

@@ -6,7 +6,6 @@
  * LICENSE file in the root directory of this source tree.
  */
 
-// Copyright 2004-present Facebook. All Rights Reserved.
 
 #include "IVFFlatScan.cuh"
 #include "../GpuResources.h"

gpu/impl/IVFFlatScan.cuh

@@ -6,7 +6,6 @@
  * LICENSE file in the root directory of this source tree.
  */
 
-// Copyright 2004-present Facebook. All Rights Reserved.
 
 #pragma once
 

gpu/impl/IVFPQ.cu

@@ -6,7 +6,6 @@
  * LICENSE file in the root directory of this source tree.
  */
 
-// Copyright 2004-present Facebook. All Rights Reserved.
 
 #include "IVFPQ.cuh"
 #include "../GpuResources.h"

gpu/impl/IVFPQ.cuh

@@ -6,7 +6,6 @@
  * LICENSE file in the root directory of this source tree.
  */
 
-// Copyright 2004-present Facebook. All Rights Reserved.
 
 #pragma once
 

gpu/impl/IVFUtils.cu

@@ -6,7 +6,6 @@
  * LICENSE file in the root directory of this source tree.
  */
 
-// Copyright 2004-present Facebook. All Rights Reserved.
 
 #include "IVFUtils.cuh"
 #include "../utils/DeviceUtils.h"

gpu/impl/IVFUtils.cuh

@@ -6,7 +6,6 @@
  * LICENSE file in the root directory of this source tree.
  */
 
-// Copyright 2004-present Facebook. All Rights Reserved.
 
 #pragma once
 

gpu/impl/IVFUtilsSelect1.cu

@@ -6,7 +6,6 @@
  * LICENSE file in the root directory of this source tree.
  */
 
-// Copyright 2004-present Facebook. All Rights Reserved.
 
 #include "IVFUtils.cuh"
 #include "../utils/DeviceUtils.h"

gpu/impl/IVFUtilsSelect2.cu

@@ -6,7 +6,6 @@
  * LICENSE file in the root directory of this source tree.
  */
 
-// Copyright 2004-present Facebook. All Rights Reserved.
 
 #include "IVFUtils.cuh"
 #include "../utils/DeviceUtils.h"

gpu/impl/InvertedListAppend.cu

@@ -6,7 +6,6 @@
  * LICENSE file in the root directory of this source tree.
  */
 
-// Copyright 2004-present Facebook. All Rights Reserved.
 
 #include "InvertedListAppend.cuh"
 #include "../../FaissAssert.h"

gpu/impl/InvertedListAppend.cuh

@@ -6,7 +6,6 @@
  * LICENSE file in the root directory of this source tree.
  */
 
-// Copyright 2004-present Facebook. All Rights Reserved.
 
 #pragma once
 

gpu/impl/L2Norm.cu

@@ -6,7 +6,6 @@
  * LICENSE file in the root directory of this source tree.
  */
 
-// Copyright 2004-present Facebook. All Rights Reserved.
 
 #include "L2Norm.cuh"
 #include "../../FaissAssert.h"

gpu/impl/L2Norm.cuh

@@ -6,7 +6,6 @@
  * LICENSE file in the root directory of this source tree.
  */
 
-// Copyright 2004-present Facebook. All Rights Reserved.
 
 #pragma once
 

gpu/impl/L2Select.cu

@@ -6,7 +6,6 @@
  * LICENSE file in the root directory of this source tree.
  */
 
-// Copyright 2004-present Facebook. All Rights Reserved.
 
 #include "L2Select.cuh"
 #include "../../FaissAssert.h"