faiss 0.1.5 → 0.2.2
Sign up to get free protection for your applications and to get access to all the features.
- checksums.yaml +4 -4
- data/CHANGELOG.md +24 -0
- data/README.md +12 -0
- data/ext/faiss/ext.cpp +1 -1
- data/ext/faiss/extconf.rb +6 -2
- data/ext/faiss/index.cpp +114 -43
- data/ext/faiss/index_binary.cpp +24 -30
- data/ext/faiss/kmeans.cpp +20 -16
- data/ext/faiss/numo.hpp +867 -0
- data/ext/faiss/pca_matrix.cpp +13 -14
- data/ext/faiss/product_quantizer.cpp +23 -24
- data/ext/faiss/utils.cpp +10 -37
- data/ext/faiss/utils.h +2 -13
- data/lib/faiss.rb +0 -5
- data/lib/faiss/version.rb +1 -1
- data/vendor/faiss/faiss/AutoTune.cpp +292 -291
- data/vendor/faiss/faiss/AutoTune.h +55 -56
- data/vendor/faiss/faiss/Clustering.cpp +334 -195
- data/vendor/faiss/faiss/Clustering.h +88 -35
- data/vendor/faiss/faiss/IVFlib.cpp +171 -195
- data/vendor/faiss/faiss/IVFlib.h +48 -51
- data/vendor/faiss/faiss/Index.cpp +85 -103
- data/vendor/faiss/faiss/Index.h +54 -48
- data/vendor/faiss/faiss/Index2Layer.cpp +139 -164
- data/vendor/faiss/faiss/Index2Layer.h +22 -22
- data/vendor/faiss/faiss/IndexBinary.cpp +45 -37
- data/vendor/faiss/faiss/IndexBinary.h +140 -132
- data/vendor/faiss/faiss/IndexBinaryFlat.cpp +73 -53
- data/vendor/faiss/faiss/IndexBinaryFlat.h +29 -24
- data/vendor/faiss/faiss/IndexBinaryFromFloat.cpp +46 -43
- data/vendor/faiss/faiss/IndexBinaryFromFloat.h +16 -15
- data/vendor/faiss/faiss/IndexBinaryHNSW.cpp +215 -232
- data/vendor/faiss/faiss/IndexBinaryHNSW.h +25 -24
- data/vendor/faiss/faiss/IndexBinaryHash.cpp +182 -177
- data/vendor/faiss/faiss/IndexBinaryHash.h +41 -34
- data/vendor/faiss/faiss/IndexBinaryIVF.cpp +489 -461
- data/vendor/faiss/faiss/IndexBinaryIVF.h +97 -68
- data/vendor/faiss/faiss/IndexFlat.cpp +116 -147
- data/vendor/faiss/faiss/IndexFlat.h +35 -46
- data/vendor/faiss/faiss/IndexHNSW.cpp +372 -348
- data/vendor/faiss/faiss/IndexHNSW.h +57 -41
- data/vendor/faiss/faiss/IndexIVF.cpp +474 -454
- data/vendor/faiss/faiss/IndexIVF.h +146 -113
- data/vendor/faiss/faiss/IndexIVFFlat.cpp +248 -250
- data/vendor/faiss/faiss/IndexIVFFlat.h +48 -51
- data/vendor/faiss/faiss/IndexIVFPQ.cpp +457 -516
- data/vendor/faiss/faiss/IndexIVFPQ.h +74 -66
- data/vendor/faiss/faiss/IndexIVFPQFastScan.cpp +406 -372
- data/vendor/faiss/faiss/IndexIVFPQFastScan.h +82 -57
- data/vendor/faiss/faiss/IndexIVFPQR.cpp +104 -102
- data/vendor/faiss/faiss/IndexIVFPQR.h +33 -28
- data/vendor/faiss/faiss/IndexIVFSpectralHash.cpp +125 -133
- data/vendor/faiss/faiss/IndexIVFSpectralHash.h +19 -21
- data/vendor/faiss/faiss/IndexLSH.cpp +75 -96
- data/vendor/faiss/faiss/IndexLSH.h +21 -26
- data/vendor/faiss/faiss/IndexLattice.cpp +42 -56
- data/vendor/faiss/faiss/IndexLattice.h +11 -16
- data/vendor/faiss/faiss/IndexNNDescent.cpp +231 -0
- data/vendor/faiss/faiss/IndexNNDescent.h +72 -0
- data/vendor/faiss/faiss/IndexNSG.cpp +303 -0
- data/vendor/faiss/faiss/IndexNSG.h +85 -0
- data/vendor/faiss/faiss/IndexPQ.cpp +405 -464
- data/vendor/faiss/faiss/IndexPQ.h +64 -67
- data/vendor/faiss/faiss/IndexPQFastScan.cpp +143 -170
- data/vendor/faiss/faiss/IndexPQFastScan.h +46 -32
- data/vendor/faiss/faiss/IndexPreTransform.cpp +120 -150
- data/vendor/faiss/faiss/IndexPreTransform.h +33 -36
- data/vendor/faiss/faiss/IndexRefine.cpp +115 -131
- data/vendor/faiss/faiss/IndexRefine.h +22 -23
- data/vendor/faiss/faiss/IndexReplicas.cpp +147 -153
- data/vendor/faiss/faiss/IndexReplicas.h +62 -56
- data/vendor/faiss/faiss/IndexResidual.cpp +291 -0
- data/vendor/faiss/faiss/IndexResidual.h +152 -0
- data/vendor/faiss/faiss/IndexScalarQuantizer.cpp +120 -155
- data/vendor/faiss/faiss/IndexScalarQuantizer.h +41 -45
- data/vendor/faiss/faiss/IndexShards.cpp +256 -240
- data/vendor/faiss/faiss/IndexShards.h +85 -73
- data/vendor/faiss/faiss/MatrixStats.cpp +112 -97
- data/vendor/faiss/faiss/MatrixStats.h +7 -10
- data/vendor/faiss/faiss/MetaIndexes.cpp +135 -157
- data/vendor/faiss/faiss/MetaIndexes.h +40 -34
- data/vendor/faiss/faiss/MetricType.h +7 -7
- data/vendor/faiss/faiss/VectorTransform.cpp +652 -474
- data/vendor/faiss/faiss/VectorTransform.h +61 -89
- data/vendor/faiss/faiss/clone_index.cpp +77 -73
- data/vendor/faiss/faiss/clone_index.h +4 -9
- data/vendor/faiss/faiss/gpu/GpuAutoTune.cpp +33 -38
- data/vendor/faiss/faiss/gpu/GpuAutoTune.h +11 -9
- data/vendor/faiss/faiss/gpu/GpuCloner.cpp +197 -170
- data/vendor/faiss/faiss/gpu/GpuCloner.h +53 -35
- data/vendor/faiss/faiss/gpu/GpuClonerOptions.cpp +12 -14
- data/vendor/faiss/faiss/gpu/GpuClonerOptions.h +27 -25
- data/vendor/faiss/faiss/gpu/GpuDistance.h +116 -112
- data/vendor/faiss/faiss/gpu/GpuFaissAssert.h +1 -2
- data/vendor/faiss/faiss/gpu/GpuIndex.h +134 -137
- data/vendor/faiss/faiss/gpu/GpuIndexBinaryFlat.h +76 -73
- data/vendor/faiss/faiss/gpu/GpuIndexFlat.h +173 -162
- data/vendor/faiss/faiss/gpu/GpuIndexIVF.h +67 -64
- data/vendor/faiss/faiss/gpu/GpuIndexIVFFlat.h +89 -86
- data/vendor/faiss/faiss/gpu/GpuIndexIVFPQ.h +150 -141
- data/vendor/faiss/faiss/gpu/GpuIndexIVFScalarQuantizer.h +101 -103
- data/vendor/faiss/faiss/gpu/GpuIndicesOptions.h +17 -16
- data/vendor/faiss/faiss/gpu/GpuResources.cpp +116 -128
- data/vendor/faiss/faiss/gpu/GpuResources.h +182 -186
- data/vendor/faiss/faiss/gpu/StandardGpuResources.cpp +433 -422
- data/vendor/faiss/faiss/gpu/StandardGpuResources.h +131 -130
- data/vendor/faiss/faiss/gpu/impl/InterleavedCodes.cpp +468 -456
- data/vendor/faiss/faiss/gpu/impl/InterleavedCodes.h +25 -19
- data/vendor/faiss/faiss/gpu/impl/RemapIndices.cpp +22 -20
- data/vendor/faiss/faiss/gpu/impl/RemapIndices.h +9 -8
- data/vendor/faiss/faiss/gpu/perf/IndexWrapper-inl.h +39 -44
- data/vendor/faiss/faiss/gpu/perf/IndexWrapper.h +16 -14
- data/vendor/faiss/faiss/gpu/perf/PerfClustering.cpp +77 -71
- data/vendor/faiss/faiss/gpu/perf/PerfIVFPQAdd.cpp +109 -88
- data/vendor/faiss/faiss/gpu/perf/WriteIndex.cpp +75 -64
- data/vendor/faiss/faiss/gpu/test/TestCodePacking.cpp +230 -215
- data/vendor/faiss/faiss/gpu/test/TestGpuIndexBinaryFlat.cpp +80 -86
- data/vendor/faiss/faiss/gpu/test/TestGpuIndexFlat.cpp +284 -277
- data/vendor/faiss/faiss/gpu/test/TestGpuIndexIVFFlat.cpp +416 -416
- data/vendor/faiss/faiss/gpu/test/TestGpuIndexIVFPQ.cpp +611 -517
- data/vendor/faiss/faiss/gpu/test/TestGpuIndexIVFScalarQuantizer.cpp +166 -164
- data/vendor/faiss/faiss/gpu/test/TestGpuMemoryException.cpp +61 -53
- data/vendor/faiss/faiss/gpu/test/TestUtils.cpp +274 -238
- data/vendor/faiss/faiss/gpu/test/TestUtils.h +73 -57
- data/vendor/faiss/faiss/gpu/test/demo_ivfpq_indexing_gpu.cpp +47 -50
- data/vendor/faiss/faiss/gpu/utils/DeviceUtils.h +79 -72
- data/vendor/faiss/faiss/gpu/utils/StackDeviceMemory.cpp +140 -146
- data/vendor/faiss/faiss/gpu/utils/StackDeviceMemory.h +69 -71
- data/vendor/faiss/faiss/gpu/utils/StaticUtils.h +21 -16
- data/vendor/faiss/faiss/gpu/utils/Timer.cpp +25 -29
- data/vendor/faiss/faiss/gpu/utils/Timer.h +30 -29
- data/vendor/faiss/faiss/impl/AdditiveQuantizer.cpp +270 -0
- data/vendor/faiss/faiss/impl/AdditiveQuantizer.h +115 -0
- data/vendor/faiss/faiss/impl/AuxIndexStructures.cpp +90 -120
- data/vendor/faiss/faiss/impl/AuxIndexStructures.h +81 -65
- data/vendor/faiss/faiss/impl/FaissAssert.h +73 -58
- data/vendor/faiss/faiss/impl/FaissException.cpp +56 -48
- data/vendor/faiss/faiss/impl/FaissException.h +41 -29
- data/vendor/faiss/faiss/impl/HNSW.cpp +595 -611
- data/vendor/faiss/faiss/impl/HNSW.h +179 -200
- data/vendor/faiss/faiss/impl/LocalSearchQuantizer.cpp +672 -0
- data/vendor/faiss/faiss/impl/LocalSearchQuantizer.h +172 -0
- data/vendor/faiss/faiss/impl/NNDescent.cpp +487 -0
- data/vendor/faiss/faiss/impl/NNDescent.h +154 -0
- data/vendor/faiss/faiss/impl/NSG.cpp +682 -0
- data/vendor/faiss/faiss/impl/NSG.h +199 -0
- data/vendor/faiss/faiss/impl/PolysemousTraining.cpp +484 -454
- data/vendor/faiss/faiss/impl/PolysemousTraining.h +52 -55
- data/vendor/faiss/faiss/impl/ProductQuantizer-inl.h +26 -47
- data/vendor/faiss/faiss/impl/ProductQuantizer.cpp +469 -459
- data/vendor/faiss/faiss/impl/ProductQuantizer.h +76 -87
- data/vendor/faiss/faiss/impl/ResidualQuantizer.cpp +448 -0
- data/vendor/faiss/faiss/impl/ResidualQuantizer.h +130 -0
- data/vendor/faiss/faiss/impl/ResultHandler.h +96 -132
- data/vendor/faiss/faiss/impl/ScalarQuantizer.cpp +648 -701
- data/vendor/faiss/faiss/impl/ScalarQuantizer.h +48 -46
- data/vendor/faiss/faiss/impl/ThreadedIndex-inl.h +129 -131
- data/vendor/faiss/faiss/impl/ThreadedIndex.h +61 -55
- data/vendor/faiss/faiss/impl/index_read.cpp +547 -479
- data/vendor/faiss/faiss/impl/index_write.cpp +497 -407
- data/vendor/faiss/faiss/impl/io.cpp +75 -94
- data/vendor/faiss/faiss/impl/io.h +31 -41
- data/vendor/faiss/faiss/impl/io_macros.h +40 -29
- data/vendor/faiss/faiss/impl/lattice_Zn.cpp +137 -186
- data/vendor/faiss/faiss/impl/lattice_Zn.h +40 -51
- data/vendor/faiss/faiss/impl/platform_macros.h +29 -8
- data/vendor/faiss/faiss/impl/pq4_fast_scan.cpp +77 -124
- data/vendor/faiss/faiss/impl/pq4_fast_scan.h +39 -48
- data/vendor/faiss/faiss/impl/pq4_fast_scan_search_1.cpp +41 -52
- data/vendor/faiss/faiss/impl/pq4_fast_scan_search_qbs.cpp +80 -117
- data/vendor/faiss/faiss/impl/simd_result_handlers.h +109 -137
- data/vendor/faiss/faiss/index_factory.cpp +269 -218
- data/vendor/faiss/faiss/index_factory.h +6 -7
- data/vendor/faiss/faiss/index_io.h +23 -26
- data/vendor/faiss/faiss/invlists/BlockInvertedLists.cpp +67 -75
- data/vendor/faiss/faiss/invlists/BlockInvertedLists.h +22 -24
- data/vendor/faiss/faiss/invlists/DirectMap.cpp +96 -112
- data/vendor/faiss/faiss/invlists/DirectMap.h +29 -33
- data/vendor/faiss/faiss/invlists/InvertedLists.cpp +307 -364
- data/vendor/faiss/faiss/invlists/InvertedLists.h +151 -151
- data/vendor/faiss/faiss/invlists/InvertedListsIOHook.cpp +29 -34
- data/vendor/faiss/faiss/invlists/InvertedListsIOHook.h +17 -18
- data/vendor/faiss/faiss/invlists/OnDiskInvertedLists.cpp +257 -293
- data/vendor/faiss/faiss/invlists/OnDiskInvertedLists.h +50 -45
- data/vendor/faiss/faiss/python/python_callbacks.cpp +23 -26
- data/vendor/faiss/faiss/python/python_callbacks.h +9 -16
- data/vendor/faiss/faiss/utils/AlignedTable.h +79 -44
- data/vendor/faiss/faiss/utils/Heap.cpp +40 -48
- data/vendor/faiss/faiss/utils/Heap.h +186 -209
- data/vendor/faiss/faiss/utils/WorkerThread.cpp +67 -76
- data/vendor/faiss/faiss/utils/WorkerThread.h +32 -33
- data/vendor/faiss/faiss/utils/distances.cpp +301 -310
- data/vendor/faiss/faiss/utils/distances.h +133 -118
- data/vendor/faiss/faiss/utils/distances_simd.cpp +456 -516
- data/vendor/faiss/faiss/utils/extra_distances-inl.h +117 -0
- data/vendor/faiss/faiss/utils/extra_distances.cpp +113 -232
- data/vendor/faiss/faiss/utils/extra_distances.h +30 -29
- data/vendor/faiss/faiss/utils/hamming-inl.h +260 -209
- data/vendor/faiss/faiss/utils/hamming.cpp +375 -469
- data/vendor/faiss/faiss/utils/hamming.h +62 -85
- data/vendor/faiss/faiss/utils/ordered_key_value.h +16 -18
- data/vendor/faiss/faiss/utils/partitioning.cpp +393 -318
- data/vendor/faiss/faiss/utils/partitioning.h +26 -21
- data/vendor/faiss/faiss/utils/quantize_lut.cpp +78 -66
- data/vendor/faiss/faiss/utils/quantize_lut.h +22 -20
- data/vendor/faiss/faiss/utils/random.cpp +39 -63
- data/vendor/faiss/faiss/utils/random.h +13 -16
- data/vendor/faiss/faiss/utils/simdlib.h +4 -2
- data/vendor/faiss/faiss/utils/simdlib_avx2.h +88 -85
- data/vendor/faiss/faiss/utils/simdlib_emulated.h +226 -165
- data/vendor/faiss/faiss/utils/simdlib_neon.h +832 -0
- data/vendor/faiss/faiss/utils/utils.cpp +304 -287
- data/vendor/faiss/faiss/utils/utils.h +53 -48
- metadata +24 -10
- data/lib/faiss/index.rb +0 -20
- data/lib/faiss/index_binary.rb +0 -20
- data/lib/faiss/kmeans.rb +0 -15
- data/lib/faiss/pca_matrix.rb +0 -15
- data/lib/faiss/product_quantizer.rb +0 -22
@@ -15,9 +15,8 @@
|
|
15
15
|
* the interface.
|
16
16
|
*/
|
17
17
|
|
18
|
-
#include <vector>
|
19
18
|
#include <faiss/Index.h>
|
20
|
-
|
19
|
+
#include <vector>
|
21
20
|
|
22
21
|
namespace faiss {
|
23
22
|
|
@@ -31,10 +30,10 @@ namespace faiss {
|
|
31
30
|
struct InvertedLists {
|
32
31
|
typedef Index::idx_t idx_t;
|
33
32
|
|
34
|
-
size_t nlist;
|
35
|
-
size_t code_size;
|
33
|
+
size_t nlist; ///< number of possible key values
|
34
|
+
size_t code_size; ///< code size per vector in bytes
|
36
35
|
|
37
|
-
InvertedLists
|
36
|
+
InvertedLists(size_t nlist, size_t code_size);
|
38
37
|
|
39
38
|
/// used for BlockInvertedLists, where the codes are packed into groups
|
40
39
|
/// and the individual code size is meaningless
|
@@ -51,70 +50,77 @@ struct InvertedLists {
|
|
51
50
|
*
|
52
51
|
* @return codes size list_size * code_size
|
53
52
|
*/
|
54
|
-
virtual const uint8_t
|
53
|
+
virtual const uint8_t* get_codes(size_t list_no) const = 0;
|
55
54
|
|
56
55
|
/** get the ids for an inverted list
|
57
56
|
* must be released by release_ids
|
58
57
|
*
|
59
58
|
* @return ids size list_size
|
60
59
|
*/
|
61
|
-
virtual const idx_t
|
60
|
+
virtual const idx_t* get_ids(size_t list_no) const = 0;
|
62
61
|
|
63
62
|
/// release codes returned by get_codes (default implementation is nop
|
64
|
-
virtual void release_codes
|
63
|
+
virtual void release_codes(size_t list_no, const uint8_t* codes) const;
|
65
64
|
|
66
65
|
/// release ids returned by get_ids
|
67
|
-
virtual void release_ids
|
66
|
+
virtual void release_ids(size_t list_no, const idx_t* ids) const;
|
68
67
|
|
69
68
|
/// @return a single id in an inverted list
|
70
|
-
virtual idx_t get_single_id
|
69
|
+
virtual idx_t get_single_id(size_t list_no, size_t offset) const;
|
71
70
|
|
72
71
|
/// @return a single code in an inverted list
|
73
72
|
/// (should be deallocated with release_codes)
|
74
|
-
virtual const uint8_t
|
75
|
-
size_t list_no, size_t offset) const;
|
73
|
+
virtual const uint8_t* get_single_code(size_t list_no, size_t offset) const;
|
76
74
|
|
77
75
|
/// prepare the following lists (default does nothing)
|
78
76
|
/// a list can be -1 hence the signed long
|
79
|
-
virtual void prefetch_lists
|
77
|
+
virtual void prefetch_lists(const idx_t* list_nos, int nlist) const;
|
80
78
|
|
81
79
|
/*************************
|
82
80
|
* writing functions */
|
83
81
|
|
84
82
|
/// add one entry to an inverted list
|
85
|
-
virtual size_t add_entry
|
86
|
-
const uint8_t *code);
|
83
|
+
virtual size_t add_entry(size_t list_no, idx_t theid, const uint8_t* code);
|
87
84
|
|
88
|
-
virtual size_t add_entries
|
89
|
-
|
90
|
-
|
85
|
+
virtual size_t add_entries(
|
86
|
+
size_t list_no,
|
87
|
+
size_t n_entry,
|
88
|
+
const idx_t* ids,
|
89
|
+
const uint8_t* code) = 0;
|
91
90
|
|
92
|
-
virtual void update_entry
|
93
|
-
|
91
|
+
virtual void update_entry(
|
92
|
+
size_t list_no,
|
93
|
+
size_t offset,
|
94
|
+
idx_t id,
|
95
|
+
const uint8_t* code);
|
94
96
|
|
95
|
-
virtual void update_entries
|
96
|
-
|
97
|
+
virtual void update_entries(
|
98
|
+
size_t list_no,
|
99
|
+
size_t offset,
|
100
|
+
size_t n_entry,
|
101
|
+
const idx_t* ids,
|
102
|
+
const uint8_t* code) = 0;
|
97
103
|
|
98
|
-
virtual void resize
|
104
|
+
virtual void resize(size_t list_no, size_t new_size) = 0;
|
99
105
|
|
100
|
-
virtual void reset
|
106
|
+
virtual void reset();
|
101
107
|
|
102
108
|
/// move all entries from oivf (empty on output)
|
103
|
-
void merge_from
|
109
|
+
void merge_from(InvertedLists* oivf, size_t add_id);
|
104
110
|
|
105
|
-
virtual ~InvertedLists
|
111
|
+
virtual ~InvertedLists();
|
106
112
|
|
107
113
|
/*************************
|
108
114
|
* statistics */
|
109
115
|
|
110
116
|
/// 1= perfectly balanced, >1: imbalanced
|
111
|
-
double imbalance_factor
|
117
|
+
double imbalance_factor() const;
|
112
118
|
|
113
119
|
/// display some stats about the inverted lists
|
114
|
-
void print_stats
|
120
|
+
void print_stats() const;
|
115
121
|
|
116
122
|
/// sum up list sizes
|
117
|
-
size_t compute_ntotal
|
123
|
+
size_t compute_ntotal() const;
|
118
124
|
|
119
125
|
/**************************************
|
120
126
|
* Scoped inverted lists (for automatic deallocation)
|
@@ -138,71 +144,76 @@ struct InvertedLists {
|
|
138
144
|
*/
|
139
145
|
|
140
146
|
struct ScopedIds {
|
141
|
-
const InvertedLists
|
142
|
-
const idx_t
|
147
|
+
const InvertedLists* il;
|
148
|
+
const idx_t* ids;
|
143
149
|
size_t list_no;
|
144
150
|
|
145
|
-
ScopedIds
|
146
|
-
|
147
|
-
{}
|
151
|
+
ScopedIds(const InvertedLists* il, size_t list_no)
|
152
|
+
: il(il), ids(il->get_ids(list_no)), list_no(list_no) {}
|
148
153
|
|
149
|
-
const idx_t
|
154
|
+
const idx_t* get() {
|
155
|
+
return ids;
|
156
|
+
}
|
150
157
|
|
151
|
-
idx_t operator
|
158
|
+
idx_t operator[](size_t i) const {
|
152
159
|
return ids[i];
|
153
160
|
}
|
154
161
|
|
155
|
-
~ScopedIds
|
156
|
-
il->release_ids
|
162
|
+
~ScopedIds() {
|
163
|
+
il->release_ids(list_no, ids);
|
157
164
|
}
|
158
165
|
};
|
159
166
|
|
160
167
|
struct ScopedCodes {
|
161
|
-
const InvertedLists
|
162
|
-
const uint8_t
|
168
|
+
const InvertedLists* il;
|
169
|
+
const uint8_t* codes;
|
163
170
|
size_t list_no;
|
164
171
|
|
165
|
-
ScopedCodes
|
166
|
-
|
167
|
-
{}
|
172
|
+
ScopedCodes(const InvertedLists* il, size_t list_no)
|
173
|
+
: il(il), codes(il->get_codes(list_no)), list_no(list_no) {}
|
168
174
|
|
169
|
-
ScopedCodes
|
170
|
-
|
171
|
-
|
172
|
-
|
175
|
+
ScopedCodes(const InvertedLists* il, size_t list_no, size_t offset)
|
176
|
+
: il(il),
|
177
|
+
codes(il->get_single_code(list_no, offset)),
|
178
|
+
list_no(list_no) {}
|
173
179
|
|
174
|
-
const uint8_t
|
180
|
+
const uint8_t* get() {
|
181
|
+
return codes;
|
182
|
+
}
|
175
183
|
|
176
|
-
~ScopedCodes
|
177
|
-
il->release_codes
|
184
|
+
~ScopedCodes() {
|
185
|
+
il->release_codes(list_no, codes);
|
178
186
|
}
|
179
187
|
};
|
180
|
-
|
181
|
-
|
182
188
|
};
|
183
189
|
|
184
|
-
|
185
190
|
/// simple (default) implementation as an array of inverted lists
|
186
|
-
struct ArrayInvertedLists: InvertedLists {
|
187
|
-
std::vector
|
188
|
-
std::vector
|
191
|
+
struct ArrayInvertedLists : InvertedLists {
|
192
|
+
std::vector<std::vector<uint8_t>> codes; // binary codes, size nlist
|
193
|
+
std::vector<std::vector<idx_t>> ids; ///< Inverted lists for indexes
|
189
194
|
|
190
|
-
ArrayInvertedLists
|
195
|
+
ArrayInvertedLists(size_t nlist, size_t code_size);
|
191
196
|
|
192
197
|
size_t list_size(size_t list_no) const override;
|
193
|
-
const uint8_t
|
194
|
-
const idx_t
|
198
|
+
const uint8_t* get_codes(size_t list_no) const override;
|
199
|
+
const idx_t* get_ids(size_t list_no) const override;
|
195
200
|
|
196
|
-
size_t add_entries
|
197
|
-
|
198
|
-
|
201
|
+
size_t add_entries(
|
202
|
+
size_t list_no,
|
203
|
+
size_t n_entry,
|
204
|
+
const idx_t* ids,
|
205
|
+
const uint8_t* code) override;
|
199
206
|
|
200
|
-
void update_entries
|
201
|
-
|
207
|
+
void update_entries(
|
208
|
+
size_t list_no,
|
209
|
+
size_t offset,
|
210
|
+
size_t n_entry,
|
211
|
+
const idx_t* ids,
|
212
|
+
const uint8_t* code) override;
|
202
213
|
|
203
|
-
void resize
|
214
|
+
void resize(size_t list_no, size_t new_size) override;
|
204
215
|
|
205
|
-
|
216
|
+
~ArrayInvertedLists() override;
|
206
217
|
};
|
207
218
|
|
208
219
|
/*****************************************************************
|
@@ -213,154 +224,143 @@ struct ArrayInvertedLists: InvertedLists {
|
|
213
224
|
*****************************************************************/
|
214
225
|
|
215
226
|
/// invlists that fail for all write functions
|
216
|
-
struct ReadOnlyInvertedLists: InvertedLists {
|
217
|
-
|
218
|
-
|
219
|
-
|
220
|
-
|
221
|
-
|
222
|
-
|
223
|
-
|
224
|
-
|
225
|
-
|
226
|
-
|
227
|
-
|
228
|
-
|
229
|
-
|
227
|
+
struct ReadOnlyInvertedLists : InvertedLists {
|
228
|
+
ReadOnlyInvertedLists(size_t nlist, size_t code_size)
|
229
|
+
: InvertedLists(nlist, code_size) {}
|
230
|
+
|
231
|
+
size_t add_entries(
|
232
|
+
size_t list_no,
|
233
|
+
size_t n_entry,
|
234
|
+
const idx_t* ids,
|
235
|
+
const uint8_t* code) override;
|
236
|
+
|
237
|
+
void update_entries(
|
238
|
+
size_t list_no,
|
239
|
+
size_t offset,
|
240
|
+
size_t n_entry,
|
241
|
+
const idx_t* ids,
|
242
|
+
const uint8_t* code) override;
|
243
|
+
|
244
|
+
void resize(size_t list_no, size_t new_size) override;
|
230
245
|
};
|
231
246
|
|
232
|
-
|
233
247
|
/// Horizontal stack of inverted lists
|
234
|
-
struct HStackInvertedLists: ReadOnlyInvertedLists {
|
235
|
-
|
236
|
-
std::vector<const InvertedLists *>ils;
|
248
|
+
struct HStackInvertedLists : ReadOnlyInvertedLists {
|
249
|
+
std::vector<const InvertedLists*> ils;
|
237
250
|
|
238
251
|
/// build InvertedLists by concatenating nil of them
|
239
|
-
HStackInvertedLists
|
252
|
+
HStackInvertedLists(int nil, const InvertedLists** ils);
|
240
253
|
|
241
254
|
size_t list_size(size_t list_no) const override;
|
242
|
-
const uint8_t
|
243
|
-
const idx_t
|
244
|
-
|
245
|
-
void prefetch_lists (const idx_t *list_nos, int nlist) const override;
|
255
|
+
const uint8_t* get_codes(size_t list_no) const override;
|
256
|
+
const idx_t* get_ids(size_t list_no) const override;
|
246
257
|
|
247
|
-
void
|
248
|
-
void release_ids (size_t list_no, const idx_t *ids) const override;
|
258
|
+
void prefetch_lists(const idx_t* list_nos, int nlist) const override;
|
249
259
|
|
250
|
-
|
260
|
+
void release_codes(size_t list_no, const uint8_t* codes) const override;
|
261
|
+
void release_ids(size_t list_no, const idx_t* ids) const override;
|
251
262
|
|
252
|
-
|
253
|
-
size_t list_no, size_t offset) const override;
|
263
|
+
idx_t get_single_id(size_t list_no, size_t offset) const override;
|
254
264
|
|
265
|
+
const uint8_t* get_single_code(size_t list_no, size_t offset)
|
266
|
+
const override;
|
255
267
|
};
|
256
268
|
|
257
269
|
using ConcatenatedInvertedLists = HStackInvertedLists;
|
258
270
|
|
259
|
-
|
260
271
|
/// vertical slice of indexes in another InvertedLists
|
261
|
-
struct SliceInvertedLists: ReadOnlyInvertedLists {
|
262
|
-
const InvertedLists
|
272
|
+
struct SliceInvertedLists : ReadOnlyInvertedLists {
|
273
|
+
const InvertedLists* il;
|
263
274
|
idx_t i0, i1;
|
264
275
|
|
265
|
-
SliceInvertedLists(const InvertedLists
|
276
|
+
SliceInvertedLists(const InvertedLists* il, idx_t i0, idx_t i1);
|
266
277
|
|
267
278
|
size_t list_size(size_t list_no) const override;
|
268
|
-
const uint8_t
|
269
|
-
const idx_t
|
279
|
+
const uint8_t* get_codes(size_t list_no) const override;
|
280
|
+
const idx_t* get_ids(size_t list_no) const override;
|
270
281
|
|
271
|
-
void release_codes
|
272
|
-
void release_ids
|
282
|
+
void release_codes(size_t list_no, const uint8_t* codes) const override;
|
283
|
+
void release_ids(size_t list_no, const idx_t* ids) const override;
|
273
284
|
|
274
|
-
idx_t get_single_id
|
285
|
+
idx_t get_single_id(size_t list_no, size_t offset) const override;
|
275
286
|
|
276
|
-
const uint8_t
|
277
|
-
|
287
|
+
const uint8_t* get_single_code(size_t list_no, size_t offset)
|
288
|
+
const override;
|
278
289
|
|
279
|
-
void prefetch_lists
|
290
|
+
void prefetch_lists(const idx_t* list_nos, int nlist) const override;
|
280
291
|
};
|
281
292
|
|
282
|
-
|
283
|
-
|
284
|
-
std::vector<const InvertedLists *>ils;
|
293
|
+
struct VStackInvertedLists : ReadOnlyInvertedLists {
|
294
|
+
std::vector<const InvertedLists*> ils;
|
285
295
|
std::vector<idx_t> cumsz;
|
286
296
|
|
287
297
|
/// build InvertedLists by concatenating nil of them
|
288
|
-
VStackInvertedLists
|
298
|
+
VStackInvertedLists(int nil, const InvertedLists** ils);
|
289
299
|
|
290
300
|
size_t list_size(size_t list_no) const override;
|
291
|
-
const uint8_t
|
292
|
-
const idx_t
|
301
|
+
const uint8_t* get_codes(size_t list_no) const override;
|
302
|
+
const idx_t* get_ids(size_t list_no) const override;
|
293
303
|
|
294
|
-
void release_codes
|
295
|
-
void release_ids
|
304
|
+
void release_codes(size_t list_no, const uint8_t* codes) const override;
|
305
|
+
void release_ids(size_t list_no, const idx_t* ids) const override;
|
296
306
|
|
297
|
-
idx_t get_single_id
|
307
|
+
idx_t get_single_id(size_t list_no, size_t offset) const override;
|
298
308
|
|
299
|
-
const uint8_t
|
300
|
-
|
301
|
-
|
302
|
-
void prefetch_lists (const idx_t *list_nos, int nlist) const override;
|
309
|
+
const uint8_t* get_single_code(size_t list_no, size_t offset)
|
310
|
+
const override;
|
303
311
|
|
312
|
+
void prefetch_lists(const idx_t* list_nos, int nlist) const override;
|
304
313
|
};
|
305
314
|
|
306
|
-
|
307
315
|
/** use the first inverted lists if they are non-empty otherwise use the second
|
308
316
|
*
|
309
317
|
* This is useful if il1 has a few inverted lists that are too long,
|
310
318
|
* and that il0 has replacement lists for those, with empty lists for
|
311
319
|
* the others. */
|
312
|
-
struct MaskedInvertedLists: ReadOnlyInvertedLists {
|
313
|
-
|
314
|
-
const InvertedLists
|
315
|
-
const InvertedLists *il1;
|
320
|
+
struct MaskedInvertedLists : ReadOnlyInvertedLists {
|
321
|
+
const InvertedLists* il0;
|
322
|
+
const InvertedLists* il1;
|
316
323
|
|
317
|
-
MaskedInvertedLists
|
318
|
-
const InvertedLists *il1);
|
324
|
+
MaskedInvertedLists(const InvertedLists* il0, const InvertedLists* il1);
|
319
325
|
|
320
326
|
size_t list_size(size_t list_no) const override;
|
321
|
-
const uint8_t
|
322
|
-
const idx_t
|
327
|
+
const uint8_t* get_codes(size_t list_no) const override;
|
328
|
+
const idx_t* get_ids(size_t list_no) const override;
|
323
329
|
|
324
|
-
void release_codes
|
325
|
-
void release_ids
|
330
|
+
void release_codes(size_t list_no, const uint8_t* codes) const override;
|
331
|
+
void release_ids(size_t list_no, const idx_t* ids) const override;
|
326
332
|
|
327
|
-
idx_t get_single_id
|
333
|
+
idx_t get_single_id(size_t list_no, size_t offset) const override;
|
328
334
|
|
329
|
-
const uint8_t
|
330
|
-
|
331
|
-
|
332
|
-
void prefetch_lists (const idx_t *list_nos, int nlist) const override;
|
335
|
+
const uint8_t* get_single_code(size_t list_no, size_t offset)
|
336
|
+
const override;
|
333
337
|
|
338
|
+
void prefetch_lists(const idx_t* list_nos, int nlist) const override;
|
334
339
|
};
|
335
340
|
|
336
|
-
|
337
341
|
/** if the inverted list in il is smaller than maxsize then return it,
|
338
342
|
* otherwise return an empty invlist */
|
339
|
-
struct StopWordsInvertedLists: ReadOnlyInvertedLists {
|
340
|
-
|
341
|
-
const InvertedLists *il0;
|
343
|
+
struct StopWordsInvertedLists : ReadOnlyInvertedLists {
|
344
|
+
const InvertedLists* il0;
|
342
345
|
size_t maxsize;
|
343
346
|
|
344
|
-
StopWordsInvertedLists
|
347
|
+
StopWordsInvertedLists(const InvertedLists* il, size_t maxsize);
|
345
348
|
|
346
349
|
size_t list_size(size_t list_no) const override;
|
347
|
-
const uint8_t
|
348
|
-
const idx_t
|
350
|
+
const uint8_t* get_codes(size_t list_no) const override;
|
351
|
+
const idx_t* get_ids(size_t list_no) const override;
|
349
352
|
|
350
|
-
void release_codes
|
351
|
-
void release_ids
|
353
|
+
void release_codes(size_t list_no, const uint8_t* codes) const override;
|
354
|
+
void release_ids(size_t list_no, const idx_t* ids) const override;
|
352
355
|
|
353
|
-
idx_t get_single_id
|
356
|
+
idx_t get_single_id(size_t list_no, size_t offset) const override;
|
354
357
|
|
355
|
-
const uint8_t
|
356
|
-
|
357
|
-
|
358
|
-
void prefetch_lists (const idx_t *list_nos, int nlist) const override;
|
358
|
+
const uint8_t* get_single_code(size_t list_no, size_t offset)
|
359
|
+
const override;
|
359
360
|
|
361
|
+
void prefetch_lists(const idx_t* list_nos, int nlist) const override;
|
360
362
|
};
|
361
363
|
|
362
|
-
|
363
364
|
} // namespace faiss
|
364
365
|
|
365
|
-
|
366
366
|
#endif
|