File IndexIVFFastScan.h

namespace faiss

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)

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.

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.

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.

Definition of inverted lists + a few common classes that implement the interface.

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.

In this file are the implementations of extra metrics beyond L2 and inner product

Implements a few neural net layers, mainly to support QINCo

Defines a few objects that apply transformations to a set of vectors Often these are pre-processing steps.

Variables

FAISS_API IVFFastScanStats IVFFastScan_stats

struct IndexIVFFastScan : public faiss::IndexIVF

#include <IndexIVFFastScan.h>

Fast scan version of IVFPQ and IVFAQ. Works for 4-bit PQ/AQ for now.

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

Implementations (implem): 0: auto-select implementation (default) 1: orig’s search, re-implemented 2: orig’s search, re-ordered by invlist 10: optimizer int16 search, collect results in heap, no qbs 11: idem, collect results in reservoir 12: optimizer int16 search, collect results in heap, uses qbs 13: idem, collect results in reservoir 14: internally multithreaded implem over nq * nprobe 15: same with reservoir

For range search, only 10 and 12 are supported. add 100 to the implem to force single-thread scanning (the coarse quantizer may still use multiple threads).

Subclassed by faiss::IndexIVFAdditiveQuantizerFastScan, faiss::IndexIVFPQFastScan

Public Functions

IndexIVFFastScan(Index *quantizer, size_t d, size_t nlist, size_t code_size, MetricType metric = METRIC_L2)

IndexIVFFastScan()

void init_fastscan(size_t M, size_t nbits, size_t nlist, MetricType metric, int bbs)

void init_code_packer()

~IndexIVFFastScan() override

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

default implementation that calls encode_vectors

virtual bool lookup_table_is_3d() const = 0

virtual void compute_LUT(size_t n, const float *x, const CoarseQuantized &cq, AlignedTable<float> &dis_tables, AlignedTable<float> &biases) const = 0

void compute_LUT_uint8(size_t n, const float *x, const CoarseQuantized &cq, AlignedTable<uint8_t> &dis_tables, AlignedTable<uint16_t> &biases, float *normalizers) const

virtual void search(idx_t n, const float *x, idx_t k, float *distances, idx_t *labels, const SearchParameters *params = nullptr) const override

assign the vectors, then call search_preassign

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 IVFSearchParameters *params = nullptr, IndexIVFStats *stats = nullptr) const override

search a set of vectors, that are pre-quantized by the IVF quantizer. Fill in the corresponding heaps with the query results. The default implementation uses InvertedListScanners to do the search.

Parameters:

• n – nb of vectors to query

• x – query vectors, size nx * d

• assign – coarse quantization indices, size nx * nprobe

• centroid_dis – distances to coarse centroids, size nx * nprobe

• distance – output distances, size n * k

• labels – output labels, size n * k

• store_pairs – store inv list index + inv list offset instead in upper/lower 32 bit of result, instead of ids (used for reranking).

• params – used to override the object’s search parameters

• stats – search stats to be updated (can be null)

virtual void range_search(idx_t n, const float *x, float radius, RangeSearchResult *result, const SearchParameters *params = nullptr) const override

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).

Parameters:

• n – number of vectors

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

• radius – search radius

• result – result table

void search_dispatch_implem(idx_t n, const float *x, idx_t k, float *distances, idx_t *labels, const CoarseQuantized &cq, const NormTableScaler *scaler, const IVFSearchParameters *params = nullptr) const

void range_search_dispatch_implem(idx_t n, const float *x, float radius, RangeSearchResult &rres, const CoarseQuantized &cq_in, const NormTableScaler *scaler, const IVFSearchParameters *params = nullptr) const

template<class C>
void search_implem_1(idx_t n, const float *x, idx_t k, float *distances, idx_t *labels, const CoarseQuantized &cq, const NormTableScaler *scaler, const IVFSearchParameters *params = nullptr) const

template<class C>
void search_implem_2(idx_t n, const float *x, idx_t k, float *distances, idx_t *labels, const CoarseQuantized &cq, const NormTableScaler *scaler, const IVFSearchParameters *params = nullptr) const

void search_implem_10(idx_t n, const float *x, SIMDResultHandlerToFloat &handler, const CoarseQuantized &cq, size_t *ndis_out, size_t *nlist_out, const NormTableScaler *scaler, const IVFSearchParameters *params = nullptr) const

void search_implem_12(idx_t n, const float *x, SIMDResultHandlerToFloat &handler, const CoarseQuantized &cq, size_t *ndis_out, size_t *nlist_out, const NormTableScaler *scaler, const IVFSearchParameters *params = nullptr) const

void search_implem_14(idx_t n, const float *x, idx_t k, float *distances, idx_t *labels, const CoarseQuantized &cq, int impl, const NormTableScaler *scaler, const IVFSearchParameters *params = nullptr) const

virtual void reconstruct_from_offset(int64_t list_no, int64_t offset, float *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 CodePacker *get_CodePacker() const override

void reconstruct_orig_invlists()

Public Members

int bbs

size_t M

size_t nbits

size_t ksub

size_t M2

int implem = 0

int skip = 0

int qbs = 0

size_t qbs2 = 0

InvertedLists *orig_invlists = nullptr

orig’s inverted lists (for debugging)

struct CoarseQuantized

Public Members

size_t nprobe

const float *dis = nullptr

const idx_t *ids = nullptr

struct IVFFastScanStats

Public Functions

inline double Mcy_at(int i)

inline double Mcy_reservoir_at(int i)

inline IVFFastScanStats()

inline void reset()

Public Members

uint64_t times[10]

uint64_t t_compute_distance_tables

uint64_t t_round

uint64_t t_copy_pack

uint64_t t_scan

uint64_t t_to_flat

uint64_t reservoir_times[4]

double t_aq_encode

double t_aq_norm_encode