1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
|
// Copyright Vespa.ai. Licensed under the terms of the Apache 2.0 license. See LICENSE in the project root.
#include "zc4_posting_writer.h"
#include <vespa/searchlib/index/docidandfeatures.h>
#include <vespa/searchlib/index/postinglistcounts.h>
#include <cassert>
using search::index::DocIdAndFeatures;
using search::index::PostingListCounts;
using search::index::PostingListParams;
namespace search::diskindex {
template <bool bigEndian>
Zc4PostingWriter<bigEndian>::Zc4PostingWriter(PostingListCounts &counts)
: Zc4PostingWriterBase(counts),
_encode_context(),
_encode_features(nullptr)
{
_encode_context.setWriteContext(&_writeContext);
_writeContext.setEncodeContext(&_encode_context);
}
template <bool bigEndian>
Zc4PostingWriter<bigEndian>::~Zc4PostingWriter() = default;
template <bool bigEndian>
void
Zc4PostingWriter<bigEndian>::reset_chunk()
{
_docIds.clear();
if (_encode_features != nullptr) {
_encode_features->setupWrite(_featureWriteContext);
_featureOffset = 0;
}
}
template <bool bigEndian>
void
Zc4PostingWriter<bigEndian>::flush_word_with_skip(bool hasMore)
{
assert(_docIds.size() >= _minSkipDocs || !_counts._segments.empty());
if (_encode_features != nullptr) {
_encode_features->flush();
}
EncodeContext &e = _encode_context;
uint32_t numDocs = _docIds.size();
e.encodeExpGolomb(numDocs - 1, K_VALUE_ZCPOSTING_NUMDOCS);
if (numDocs >= _minChunkDocs) {
e.writeBits((hasMore ? 1 : 0), 1);
}
calc_skip_info(_encode_features != nullptr);
uint32_t docIdsSize = _zcDocIds.size();
uint32_t l1SkipSize = _l1Skip.size();
uint32_t l2SkipSize = _l2Skip.size();
uint32_t l3SkipSize = _l3Skip.size();
uint32_t l4SkipSize = _l4Skip.size();
e.encodeExpGolomb(docIdsSize - 1, K_VALUE_ZCPOSTING_DOCIDSSIZE);
e.encodeExpGolomb(l1SkipSize, K_VALUE_ZCPOSTING_L1SKIPSIZE);
if (l1SkipSize != 0) {
e.encodeExpGolomb(l2SkipSize, K_VALUE_ZCPOSTING_L2SKIPSIZE);
if (l2SkipSize != 0) {
e.encodeExpGolomb(l3SkipSize, K_VALUE_ZCPOSTING_L3SKIPSIZE);
if (l3SkipSize != 0) {
e.encodeExpGolomb(l4SkipSize, K_VALUE_ZCPOSTING_L4SKIPSIZE);
}
}
}
if (_encode_features != nullptr) {
e.encodeExpGolomb(_featureOffset, K_VALUE_ZCPOSTING_FEATURESSIZE);
}
// Encode last document id in chunk or word.
if (_dynamicK) {
uint32_t docIdK = e.calcDocIdK((_counts._segments.empty() &&
!hasMore) ?
numDocs : 1,
_docIdLimit);
e.encodeExpGolomb(_docIdLimit - 1 - _docIds.back()._doc_id,
docIdK);
} else {
e.encodeExpGolomb(_docIdLimit - 1 - _docIds.back()._doc_id,
K_VALUE_ZCPOSTING_LASTDOCID);
}
e.smallAlign(8); // Byte align
uint8_t *docIds = _zcDocIds._mallocStart;
e.writeBits(reinterpret_cast<const uint64_t *>(docIds),
0,
docIdsSize * 8);
if (l1SkipSize > 0) {
uint8_t *l1Skip = _l1Skip._mallocStart;
e.writeBits(reinterpret_cast<const uint64_t *>(l1Skip),
0,
l1SkipSize * 8);
}
if (l2SkipSize > 0) {
uint8_t *l2Skip = _l2Skip._mallocStart;
e.writeBits(reinterpret_cast<const uint64_t *>(l2Skip),
0,
l2SkipSize * 8);
}
if (l3SkipSize > 0) {
uint8_t *l3Skip = _l3Skip._mallocStart;
e.writeBits(reinterpret_cast<const uint64_t *>(l3Skip),
0,
l3SkipSize * 8);
}
if (l4SkipSize > 0) {
uint8_t *l4Skip = _l4Skip._mallocStart;
e.writeBits(reinterpret_cast<const uint64_t *>(l4Skip),
0,
l4SkipSize * 8);
}
// Write features
e.writeBits(_featureWriteContext.getComprBuf(), 0, _featureOffset);
_counts._numDocs += numDocs;
if (hasMore || !_counts._segments.empty()) {
uint64_t writePos = e.getWriteOffset();
PostingListCounts::Segment seg;
seg._bitLength = writePos - (_writePos + _counts._bitLength);
seg._numDocs = numDocs;
seg._lastDoc = _docIds.back()._doc_id;
_counts._segments.push_back(seg);
_counts._bitLength += seg._bitLength;
}
// reset tables in preparation for next word or next chunk
clear_skip_info();
reset_chunk();
}
template <bool bigEndian>
void
Zc4PostingWriter<bigEndian>::write_docid_and_features(const DocIdAndFeatures &features)
{
if (__builtin_expect(_docIds.size() >= _minChunkDocs, false)) {
flush_word_with_skip(true);
}
if (_encode_features != nullptr) {
_encode_features->writeFeatures(features);
uint64_t writeOffset = _encode_features->getWriteOffset();
uint64_t featureSize = writeOffset - _featureOffset;
assert(static_cast<uint32_t>(featureSize) == featureSize);
_docIds.emplace_back(features.doc_id(), features.field_length(), features.num_occs(),
static_cast<uint32_t>(featureSize));
_featureOffset = writeOffset;
} else {
_docIds.emplace_back(features.doc_id(), features.field_length(), features.num_occs(), 0);
}
}
template <bool bigEndian>
void
Zc4PostingWriter<bigEndian>::flush_word_no_skip()
{
// Too few document ids for skip info.
assert(_docIds.size() < _minSkipDocs && _counts._segments.empty());
if (_encode_features != nullptr) {
_encode_features->flush();
}
EncodeContext &e = _encode_context;
uint32_t numDocs = _docIds.size();
e.encodeExpGolomb(numDocs - 1, K_VALUE_ZCPOSTING_NUMDOCS);
uint32_t docIdK = _dynamicK ? e.calcDocIdK(numDocs, _docIdLimit) : K_VALUE_ZCPOSTING_DELTA_DOCID;
uint32_t baseDocId = 1;
const uint64_t *features = _featureWriteContext.getComprBuf();
uint64_t featureOffset = 0;
for (const auto& elem : _docIds) {
uint32_t docId = elem._doc_id;
uint32_t featureSize = elem._features_size;
e.encodeExpGolomb(docId - baseDocId, docIdK);
baseDocId = docId + 1;
if (_encode_interleaved_features) {
assert(elem._field_length > 0);
e.encodeExpGolomb(elem._field_length - 1, K_VALUE_ZCPOSTING_FIELD_LENGTH);
assert(elem._num_occs > 0);
e.encodeExpGolomb(elem._num_occs - 1, K_VALUE_ZCPOSTING_NUM_OCCS);
}
if (featureSize != 0) {
e.writeBits(features + (featureOffset >> 6),
featureOffset & 63,
featureSize);
featureOffset += featureSize;
}
}
_counts._numDocs += numDocs;
reset_chunk();
}
template <bool bigEndian>
void
Zc4PostingWriter<bigEndian>::flush_word()
{
if (__builtin_expect(_docIds.size() >= _minSkipDocs ||
!_counts._segments.empty(), false)) {
// Use skip information if enough documents or chunking has happened
flush_word_with_skip(false);
_numWords++;
} else if (_docIds.size() > 0) {
flush_word_no_skip();
_numWords++;
}
EncodeContext &e = _encode_context;
uint64_t writePos = e.getWriteOffset();
_counts._bitLength = writePos - _writePos;
_writePos = writePos;
}
template <bool bigEndian>
void
Zc4PostingWriter<bigEndian>::set_encode_features(EncodeContext *encode_features)
{
_encode_features = encode_features;
if (_encode_features != nullptr) {
_encode_features->setWriteContext(&_featureWriteContext);
_encode_features->setupWrite(_featureWriteContext);
}
_featureWriteContext.setEncodeContext(_encode_features);
_featureOffset = 0;
}
template <bool bigEndian>
void
Zc4PostingWriter<bigEndian>::on_open()
{
_numWords = 0;
_writePos = _encode_context.getWriteOffset(); // Position after file header
}
template <bool bigEndian>
void
Zc4PostingWriter<bigEndian>::on_close()
{
_encode_context.pad_for_memory_map_and_flush();
}
template class Zc4PostingWriter<false>;
template class Zc4PostingWriter<true>;
}
|
