File IndexAdditiveQuantizerFastScan.h

namespace faiss

Copyright (c) Facebook, Inc. and its affiliates.

This source code is licensed under the MIT license found in the LICENSE file in the root directory of this source tree.

Copyright (c) Facebook, Inc. and its affiliates.

This source code is licensed under the MIT license found in the LICENSE file in the root directory of this source tree. Implementation of k-means clustering with many variants.

Copyright (c) Facebook, Inc. and its affiliates.

This source code is licensed under the MIT license found in the LICENSE file in the root directory of this source tree. IDSelector is intended to define a subset of vectors to handle (for removal or as subset to search)

Copyright (c) Facebook, Inc. and its affiliates.

This source code is licensed under the MIT license found in the LICENSE file in the root directory of this source tree. PQ4 SIMD packing and accumulation functions

The basic kernel accumulates nq query vectors with bbs = nb * 2 * 16 vectors and produces an output matrix for that. It is interesting for nq * nb <= 4, otherwise register spilling becomes too large.

The implementation of these functions is spread over 3 cpp files to reduce parallel compile times. Templates are instantiated explicitly.

Copyright (c) Facebook, Inc. and its affiliates.

This source code is licensed under the MIT license found in the LICENSE file in the root directory of this source tree. This file contains callbacks for kernels that compute distances.

Throughout the library, vectors are provided as float * pointers. Most algorithms can be optimized when several vectors are processed (added/searched) together in a batch. In this case, they are passed in as a matrix. When n vectors of size d are provided as float * x, component j of vector i is

x[ i * d + j ]

where 0 <= i < n and 0 <= j < d. In other words, matrices are always compact. When specifying the size of the matrix, we call it an n*d matrix, which implies a row-major storage.

Copyright (c) Facebook, Inc. and its affiliates.

This source code is licensed under the MIT license found in the LICENSE file in the root directory of this source tree. I/O functions can read/write to a filename, a file handle or to an object that abstracts the medium.

The read functions return objects that should be deallocated with delete. All references within these objectes are owned by the object.

Copyright (c) Facebook, Inc. and its affiliates.

This source code is licensed under the MIT license found in the LICENSE file in the root directory of this source tree. Definition of inverted lists + a few common classes that implement the interface.

Copyright (c) Facebook, Inc. and its affiliates.

This source code is licensed under the MIT license found in the LICENSE file in the root directory of this source tree. Since IVF (inverted file) indexes are of so much use for large-scale use cases, we group a few functions related to them in this small library. Most functions work both on IndexIVFs and IndexIVFs embedded within an IndexPreTransform.

Copyright (c) Facebook, Inc. and its affiliates.

This source code is licensed under the MIT license found in the LICENSE file in the root directory of this source tree. In this file are the implementations of extra metrics beyond L2 and inner product

Copyright (c) Facebook, Inc. and its affiliates.

This source code is licensed under the MIT license found in the LICENSE file in the root directory of this source tree. Implements a few neural net layers, mainly to support QINCo

Copyright (c) Facebook, Inc. and its affiliates.

This source code is licensed under the MIT license found in the LICENSE file in the root directory of this source tree. Defines a few objects that apply transformations to a set of vectors Often these are pre-processing steps.

struct IndexAdditiveQuantizerFastScan : public faiss::IndexFastScan
#include <IndexAdditiveQuantizerFastScan.h>

Fast scan version of IndexAQ. Works for 4-bit AQ for now.

The codes are not stored sequentially but grouped in blocks of size bbs. This makes it possible to compute distances quickly with SIMD instructions.

Implementations: 12: blocked loop with internal loop on Q with qbs 13: same with reservoir accumulator to store results 14: no qbs with heap accumulator 15: no qbs with reservoir accumulator

Subclassed by faiss::IndexLocalSearchQuantizerFastScan, faiss::IndexProductLocalSearchQuantizerFastScan, faiss::IndexProductResidualQuantizerFastScan, faiss::IndexResidualQuantizerFastScan

Public Types

using Search_type_t = AdditiveQuantizer::Search_type_t

Public Functions

explicit IndexAdditiveQuantizerFastScan(AdditiveQuantizer *aq, MetricType metric = METRIC_L2, int bbs = 32)
void init(AdditiveQuantizer *aq, MetricType metric = METRIC_L2, int bbs = 32)
IndexAdditiveQuantizerFastScan()
~IndexAdditiveQuantizerFastScan() override
explicit IndexAdditiveQuantizerFastScan(const IndexAdditiveQuantizer &orig, int bbs = 32)

build from an existing IndexAQ

virtual void train(idx_t n, const float *x) override

Perform training on a representative set of vectors

Parameters:
  • n – nb of training vectors

  • x – training vecors, size n * d

void estimate_norm_scale(idx_t n, const float *x)
virtual void compute_codes(uint8_t *codes, idx_t n, const float *x) const override
virtual void compute_float_LUT(float *lut, idx_t n, const float *x) const override
virtual void search(idx_t n, const float *x, idx_t k, float *distances, idx_t *labels, const SearchParameters *params = nullptr) const override

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.

Parameters:
  • n – number of vectors

  • x – input vectors to search, size n * d

  • k – number of extracted vectors

  • distances – output pairwise distances, size n*k

  • labels – output labels of the NNs, size n*k

virtual void sa_decode(idx_t n, const uint8_t *bytes, float *x) const override

Decode a set of vectors.

NOTE: The codes in the IndexAdditiveQuantizerFastScan object are non- contiguous. But this method requires a contiguous representation.

Parameters:
  • n – number of vectors

  • bytes – input encoded vectors, size n * code_size

  • x – output vectors, size n * d

Public Members

AdditiveQuantizer *aq
bool rescale_norm = true
int norm_scale = 1
size_t max_train_points = 0
struct IndexResidualQuantizerFastScan : public faiss::IndexAdditiveQuantizerFastScan
#include <IndexAdditiveQuantizerFastScan.h>

Index based on a residual quantizer. Stored vectors are approximated by residual quantization codes. Can also be used as a codec

Public Types

using Search_type_t = AdditiveQuantizer::Search_type_t

Public Functions

IndexResidualQuantizerFastScan(int d, size_t M, size_t nbits, MetricType metric = METRIC_L2, Search_type_t search_type = AdditiveQuantizer::ST_norm_rq2x4, int bbs = 32)

Constructor.

Parameters:
  • d – dimensionality of the input vectors

  • M – number of subquantizers

  • nbits – number of bit per subvector index

  • d – dimensionality of the input vectors

  • M – number of subquantizers

  • nbits – number of bit per subvector index

  • metric – metric type

  • search_type – AQ search type

IndexResidualQuantizerFastScan()
void init(AdditiveQuantizer *aq, MetricType metric = METRIC_L2, int bbs = 32)
virtual void train(idx_t n, const float *x) override

Perform training on a representative set of vectors

Parameters:
  • n – nb of training vectors

  • x – training vecors, size n * d

void estimate_norm_scale(idx_t n, const float *x)
virtual void compute_codes(uint8_t *codes, idx_t n, const float *x) const override
virtual void compute_float_LUT(float *lut, idx_t n, const float *x) const override
virtual void search(idx_t n, const float *x, idx_t k, float *distances, idx_t *labels, const SearchParameters *params = nullptr) const override

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.

Parameters:
  • n – number of vectors

  • x – input vectors to search, size n * d

  • k – number of extracted vectors

  • distances – output pairwise distances, size n*k

  • labels – output labels of the NNs, size n*k

virtual void sa_decode(idx_t n, const uint8_t *bytes, float *x) const override

Decode a set of vectors.

NOTE: The codes in the IndexAdditiveQuantizerFastScan object are non- contiguous. But this method requires a contiguous representation.

Parameters:
  • n – number of vectors

  • bytes – input encoded vectors, size n * code_size

  • x – output vectors, size n * d

Public Members

ResidualQuantizer rq

The residual quantizer used to encode the vectors.

AdditiveQuantizer *aq
bool rescale_norm = true
int norm_scale = 1
size_t max_train_points = 0
struct IndexLocalSearchQuantizerFastScan : public faiss::IndexAdditiveQuantizerFastScan
#include <IndexAdditiveQuantizerFastScan.h>

Index based on a local search quantizer. Stored vectors are approximated by local search quantization codes. Can also be used as a codec

Public Types

using Search_type_t = AdditiveQuantizer::Search_type_t

Public Functions

IndexLocalSearchQuantizerFastScan(int d, size_t M, size_t nbits, MetricType metric = METRIC_L2, Search_type_t search_type = AdditiveQuantizer::ST_norm_lsq2x4, int bbs = 32)

Constructor.

Parameters:
  • d – dimensionality of the input vectors

  • M – number of subquantizers

  • nbits – number of bit per subvector index

  • d – dimensionality of the input vectors

  • M – number of subquantizers

  • nbits – number of bit per subvector index

  • metric – metric type

  • search_type – AQ search type

IndexLocalSearchQuantizerFastScan()
void init(AdditiveQuantizer *aq, MetricType metric = METRIC_L2, int bbs = 32)
virtual void train(idx_t n, const float *x) override

Perform training on a representative set of vectors

Parameters:
  • n – nb of training vectors

  • x – training vecors, size n * d

void estimate_norm_scale(idx_t n, const float *x)
virtual void compute_codes(uint8_t *codes, idx_t n, const float *x) const override
virtual void compute_float_LUT(float *lut, idx_t n, const float *x) const override
virtual void search(idx_t n, const float *x, idx_t k, float *distances, idx_t *labels, const SearchParameters *params = nullptr) const override

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.

Parameters:
  • n – number of vectors

  • x – input vectors to search, size n * d

  • k – number of extracted vectors

  • distances – output pairwise distances, size n*k

  • labels – output labels of the NNs, size n*k

virtual void sa_decode(idx_t n, const uint8_t *bytes, float *x) const override

Decode a set of vectors.

NOTE: The codes in the IndexAdditiveQuantizerFastScan object are non- contiguous. But this method requires a contiguous representation.

Parameters:
  • n – number of vectors

  • bytes – input encoded vectors, size n * code_size

  • x – output vectors, size n * d

Public Members

LocalSearchQuantizer lsq
AdditiveQuantizer *aq
bool rescale_norm = true
int norm_scale = 1
size_t max_train_points = 0
struct IndexProductResidualQuantizerFastScan : public faiss::IndexAdditiveQuantizerFastScan
#include <IndexAdditiveQuantizerFastScan.h>

Index based on a product residual quantizer. Stored vectors are approximated by product residual quantization codes. Can also be used as a codec

Public Types

using Search_type_t = AdditiveQuantizer::Search_type_t

Public Functions

IndexProductResidualQuantizerFastScan(int d, size_t nsplits, size_t Msub, size_t nbits, MetricType metric = METRIC_L2, Search_type_t search_type = AdditiveQuantizer::ST_norm_rq2x4, int bbs = 32)

Constructor.

Parameters:
  • d – dimensionality of the input vectors

  • nsplits – number of residual quantizers

  • Msub – number of subquantizers per RQ

  • nbits – number of bit per subvector index

  • d – dimensionality of the input vectors

  • nsplits – number of residual quantizers

  • Msub – number of subquantizers per RQ

  • nbits – number of bit per subvector index

  • metric – metric type

  • search_type – AQ search type

IndexProductResidualQuantizerFastScan()
void init(AdditiveQuantizer *aq, MetricType metric = METRIC_L2, int bbs = 32)
virtual void train(idx_t n, const float *x) override

Perform training on a representative set of vectors

Parameters:
  • n – nb of training vectors

  • x – training vecors, size n * d

void estimate_norm_scale(idx_t n, const float *x)
virtual void compute_codes(uint8_t *codes, idx_t n, const float *x) const override
virtual void compute_float_LUT(float *lut, idx_t n, const float *x) const override
virtual void search(idx_t n, const float *x, idx_t k, float *distances, idx_t *labels, const SearchParameters *params = nullptr) const override

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.

Parameters:
  • n – number of vectors

  • x – input vectors to search, size n * d

  • k – number of extracted vectors

  • distances – output pairwise distances, size n*k

  • labels – output labels of the NNs, size n*k

virtual void sa_decode(idx_t n, const uint8_t *bytes, float *x) const override

Decode a set of vectors.

NOTE: The codes in the IndexAdditiveQuantizerFastScan object are non- contiguous. But this method requires a contiguous representation.

Parameters:
  • n – number of vectors

  • bytes – input encoded vectors, size n * code_size

  • x – output vectors, size n * d

Public Members

ProductResidualQuantizer prq

The product residual quantizer used to encode the vectors.

AdditiveQuantizer *aq
bool rescale_norm = true
int norm_scale = 1
size_t max_train_points = 0
struct IndexProductLocalSearchQuantizerFastScan : public faiss::IndexAdditiveQuantizerFastScan
#include <IndexAdditiveQuantizerFastScan.h>

Index based on a product local search quantizer. Stored vectors are approximated by product local search quantization codes. Can also be used as a codec

Public Types

using Search_type_t = AdditiveQuantizer::Search_type_t

Public Functions

IndexProductLocalSearchQuantizerFastScan(int d, size_t nsplits, size_t Msub, size_t nbits, MetricType metric = METRIC_L2, Search_type_t search_type = AdditiveQuantizer::ST_norm_rq2x4, int bbs = 32)

Constructor.

Parameters:
  • d – dimensionality of the input vectors

  • nsplits – number of local search quantizers

  • Msub – number of subquantizers per LSQ

  • nbits – number of bit per subvector index

  • d – dimensionality of the input vectors

  • nsplits – number of local search quantizers

  • Msub – number of subquantizers per LSQ

  • nbits – number of bit per subvector index

  • metric – metric type

  • search_type – AQ search type

IndexProductLocalSearchQuantizerFastScan()
void init(AdditiveQuantizer *aq, MetricType metric = METRIC_L2, int bbs = 32)
virtual void train(idx_t n, const float *x) override

Perform training on a representative set of vectors

Parameters:
  • n – nb of training vectors

  • x – training vecors, size n * d

void estimate_norm_scale(idx_t n, const float *x)
virtual void compute_codes(uint8_t *codes, idx_t n, const float *x) const override
virtual void compute_float_LUT(float *lut, idx_t n, const float *x) const override
virtual void search(idx_t n, const float *x, idx_t k, float *distances, idx_t *labels, const SearchParameters *params = nullptr) const override

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.

Parameters:
  • n – number of vectors

  • x – input vectors to search, size n * d

  • k – number of extracted vectors

  • distances – output pairwise distances, size n*k

  • labels – output labels of the NNs, size n*k

virtual void sa_decode(idx_t n, const uint8_t *bytes, float *x) const override

Decode a set of vectors.

NOTE: The codes in the IndexAdditiveQuantizerFastScan object are non- contiguous. But this method requires a contiguous representation.

Parameters:
  • n – number of vectors

  • bytes – input encoded vectors, size n * code_size

  • x – output vectors, size n * d

Public Members

ProductLocalSearchQuantizer plsq

The product local search quantizer used to encode the vectors.

AdditiveQuantizer *aq
bool rescale_norm = true
int norm_scale = 1
size_t max_train_points = 0