forked from dgraph-io/dgraph
-
Notifications
You must be signed in to change notification settings - Fork 0
/
bp128.go
408 lines (366 loc) · 11.8 KB
/
bp128.go
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
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
// Package bp128 implements SIMD-BP128 integer encoding and decoding.
// It requires an x86_64/AMD64 CPU that supports SSE2 instructions.
//
// For more details on SIMD-BP128 algorithm see "Decoding billions of
// integers per second through vectorization" by Daniel Lemire, Leonid
// Boytsov, and Nathan Kurz at http://arxiv.org/pdf/1209.2137
//
// For the original C++ implementation visit
// https://github.com/lemire/SIMDCompressionAndIntersection.
package bp128
import (
"encoding/binary"
"math"
"sort"
"github.com/dgraph-io/dgraph/x"
)
const (
// BlockSize is the number of integers per block. Each
// block address must be aligned at 16-byte boundaries.
BlockSize = 256
intSize = 64
bitVarint = 0x80
)
var (
maxBits func(*uint64, *uint64) uint8
fpack []func(*uint64, *byte, *uint64)
funpack []func(*byte, *uint64, *uint64)
)
func init() {
if BlockSize == 128 {
fpack = fdpack128
maxBits = maxBits128
funpack = fdunpack128
} else if BlockSize == 256 {
fpack = fdpack256
maxBits = maxBits256
funpack = fdunpack256
} else {
x.Fatalf("Unknown block size")
}
}
var fdpack128 = []func(in *uint64, out *byte, seed *uint64){
dpack128_0, dpack128_1, dpack128_2, dpack128_3, dpack128_4, dpack128_5,
dpack128_6, dpack128_7, dpack128_8, dpack128_9, dpack128_10, dpack128_11,
dpack128_12, dpack128_13, dpack128_14, dpack128_15, dpack128_16, dpack128_17,
dpack128_18, dpack128_19, dpack128_20, dpack128_21, dpack128_22, dpack128_23,
dpack128_24, dpack128_25, dpack128_26, dpack128_27, dpack128_28, dpack128_29,
dpack128_30, dpack128_31, dpack128_32, dpack128_33, dpack128_34, dpack128_35,
dpack128_36, dpack128_37, dpack128_38, dpack128_39, dpack128_40, dpack128_41,
dpack128_42, dpack128_43, dpack128_44, dpack128_45, dpack128_46, dpack128_47,
dpack128_48, dpack128_49, dpack128_50, dpack128_51, dpack128_52, dpack128_53,
dpack128_54, dpack128_55, dpack128_56, dpack128_57, dpack128_58, dpack128_59,
dpack128_60, dpack128_61, dpack128_62, dpack128_63, dpack128_64,
}
var fdunpack128 = []func(in *byte, out *uint64, seed *uint64){
dunpack128_0, dunpack128_1, dunpack128_2, dunpack128_3, dunpack128_4,
dunpack128_5, dunpack128_6, dunpack128_7, dunpack128_8, dunpack128_9,
dunpack128_10, dunpack128_11, dunpack128_12, dunpack128_13, dunpack128_14,
dunpack128_15, dunpack128_16, dunpack128_17, dunpack128_18, dunpack128_19,
dunpack128_20, dunpack128_21, dunpack128_22, dunpack128_23, dunpack128_24,
dunpack128_25, dunpack128_26, dunpack128_27, dunpack128_28, dunpack128_29,
dunpack128_30, dunpack128_31, dunpack128_32, dunpack128_33, dunpack128_34,
dunpack128_35, dunpack128_36, dunpack128_37, dunpack128_38, dunpack128_39,
dunpack128_40, dunpack128_41, dunpack128_42, dunpack128_43, dunpack128_44,
dunpack128_45, dunpack128_46, dunpack128_47, dunpack128_48, dunpack128_49,
dunpack128_50, dunpack128_51, dunpack128_52, dunpack128_53, dunpack128_54,
dunpack128_55, dunpack128_56, dunpack128_57, dunpack128_58, dunpack128_59,
dunpack128_60, dunpack128_61, dunpack128_62, dunpack128_63, dunpack128_64,
}
var fdpack256 = []func(in *uint64, out *byte, seed *uint64){
dpack256_0, dpack256_1, dpack256_2, dpack256_3, dpack256_4, dpack256_5,
dpack256_6, dpack256_7, dpack256_8, dpack256_9, dpack256_10, dpack256_11,
dpack256_12, dpack256_13, dpack256_14, dpack256_15, dpack256_16, dpack256_17,
dpack256_18, dpack256_19, dpack256_20, dpack256_21, dpack256_22, dpack256_23,
dpack256_24, dpack256_25, dpack256_26, dpack256_27, dpack256_28, dpack256_29,
dpack256_30, dpack256_31, dpack256_32, dpack256_33, dpack256_34, dpack256_35,
dpack256_36, dpack256_37, dpack256_38, dpack256_39, dpack256_40, dpack256_41,
dpack256_42, dpack256_43, dpack256_44, dpack256_45, dpack256_46, dpack256_47,
dpack256_48, dpack256_49, dpack256_50, dpack256_51, dpack256_52, dpack256_53,
dpack256_54, dpack256_55, dpack256_56, dpack256_57, dpack256_58, dpack256_59,
dpack256_60, dpack256_61, dpack256_62, dpack256_63, dpack256_64,
}
var fdunpack256 = []func(in *byte, out *uint64, seed *uint64){
dunpack256_0, dunpack256_1, dunpack256_2, dunpack256_3, dunpack256_4,
dunpack256_5, dunpack256_6, dunpack256_7, dunpack256_8, dunpack256_9,
dunpack256_10, dunpack256_11, dunpack256_12, dunpack256_13, dunpack256_14,
dunpack256_15, dunpack256_16, dunpack256_17, dunpack256_18, dunpack256_19,
dunpack256_20, dunpack256_21, dunpack256_22, dunpack256_23, dunpack256_24,
dunpack256_25, dunpack256_26, dunpack256_27, dunpack256_28, dunpack256_29,
dunpack256_30, dunpack256_31, dunpack256_32, dunpack256_33, dunpack256_34,
dunpack256_35, dunpack256_36, dunpack256_37, dunpack256_38, dunpack256_39,
dunpack256_40, dunpack256_41, dunpack256_42, dunpack256_43, dunpack256_44,
dunpack256_45, dunpack256_46, dunpack256_47, dunpack256_48, dunpack256_49,
dunpack256_50, dunpack256_51, dunpack256_52, dunpack256_53, dunpack256_54,
dunpack256_55, dunpack256_56, dunpack256_57, dunpack256_58, dunpack256_59,
dunpack256_60, dunpack256_61, dunpack256_62, dunpack256_63, dunpack256_64,
}
type BPackEncoder struct {
data x.BytesBuffer
metadata x.BytesBuffer
length int
// Used to store seed of last block
lastSeed []uint64
// Offset into data
offset int
}
func (bp *BPackEncoder) PackAppend(in []uint64) {
if len(in) == 0 {
return
}
if len(bp.lastSeed) == 0 && len(in) >= 2 {
bp.lastSeed = make([]uint64, 2)
bp.lastSeed[0] = in[0]
bp.lastSeed[1] = in[1]
} else if len(bp.lastSeed) == 0 && len(in) == 1 {
// We won't use seed value for varint, writing it in metadata
// to have uniform length for metadata
bp.lastSeed = make([]uint64, 2)
}
bp.length += len(in)
b := bp.metadata.Slice(20)
binary.BigEndian.PutUint64(b[0:8], bp.lastSeed[0])
binary.BigEndian.PutUint64(b[8:16], bp.lastSeed[1])
binary.BigEndian.PutUint32(b[16:20], uint32(bp.offset))
// This should be the last block
if len(in) < BlockSize {
b = bp.data.Slice(1 + 10*len(in))
b[0] = 0 | bitVarint
off := 1
for _, num := range in {
off += binary.PutUvarint(b[off:], num)
}
bp.data.TruncateBy(len(b) - off)
return
}
bs := maxBits(&in[0], &bp.lastSeed[0])
nBytes := int(bs)*BlockSize/8 + 1
b = bp.data.Slice(nBytes)
b[0] = bs
if bs > 0 {
fpack[bs](&in[0], &b[1], &bp.lastSeed[0])
}
bp.offset += nBytes
}
func (bp *BPackEncoder) WriteTo(in []byte) {
x.AssertTruef(bp.length > 0, "cannot pack zero length posting list")
binary.BigEndian.PutUint32(in[:4], uint32(bp.length))
if bp.length < BlockSize {
// If number of integers are less all are stored as varint
// and without metadata.
bp.data.CopyTo(in[4:])
return
}
offset := bp.metadata.CopyTo(in[4:])
bp.data.CopyTo(in[4+offset:])
}
func (bp *BPackEncoder) Size() int {
if bp.length == 0 {
return 0
}
return 4 + bp.data.Length() + bp.metadata.Length()
}
func (bp *BPackEncoder) Length() int {
return bp.length
}
func NumIntegers(data []byte) int {
if len(data) == 0 {
return 0
}
return int(binary.BigEndian.Uint32(data[0:4]))
}
type BPackIterator struct {
data []byte
metadata []byte
length int
in_offset int
count int
valid bool
lastSeed []uint64
// Byte slice which would be reused for decompression
buf []uint64
// out is the slice ready to be read by the user, would
// point to some offset in buf
out []uint64
}
func numBlocks(len int) int {
if len < BlockSize {
return 0
}
if len%BlockSize == 0 {
return len / BlockSize
}
return len/BlockSize + 1
}
func (pi *BPackIterator) Init(data []byte, afterUid uint64) {
if len(data) == 0 {
return
}
pi.length = int(binary.BigEndian.Uint32(data[0:4]))
nBlocks := numBlocks(pi.length)
pi.data = data[4+nBlocks*20:]
pi.metadata = data[4 : 4+nBlocks*20]
pi.out = make([]uint64, BlockSize, BlockSize)
pi.buf = pi.out
pi.lastSeed = make([]uint64, 2)
pi.valid = true
if afterUid > 0 {
pi.search(afterUid, nBlocks)
uidx := sort.Search(len(pi.out), func(idx int) bool {
return afterUid < pi.out[idx]
})
pi.out = pi.out[uidx:]
return
}
if len(pi.metadata) > 0 {
pi.lastSeed[0] = binary.BigEndian.Uint64(pi.metadata[0:8])
pi.lastSeed[1] = binary.BigEndian.Uint64(pi.metadata[8:16])
}
pi.Next()
return
}
func (pi *BPackIterator) search(afterUid uint64, numBlocks int) {
if len(pi.metadata) == 0 {
pi.Next()
return
}
// Search in metadata whose seed[1] > afterUid
idx := sort.Search(numBlocks, func(idx int) bool {
i := idx * 20
return afterUid < binary.BigEndian.Uint64(pi.metadata[i+8:i+16])
})
// seed is stored for previous block, so search there. If not found
// then search in last block.
if idx >= numBlocks {
idx = numBlocks - 1
} else if idx > 0 {
idx -= 1
}
pi.count = idx * BlockSize
i := idx * 20
pi.in_offset = int(binary.BigEndian.Uint32(pi.metadata[i+16 : i+20]))
pi.lastSeed[0] = binary.BigEndian.Uint64(pi.metadata[i : i+8])
pi.lastSeed[1] = binary.BigEndian.Uint64(pi.metadata[i+8 : i+16])
pi.Next()
}
func (pi *BPackIterator) AfterUid(uid uint64) (found bool) {
// Current uncompressed block doesn't have uid, search for appropriate
// block, uncompress it and store it in pi.out
if len(pi.out) > 0 && pi.out[len(pi.out)-1] < uid {
nBlocks := numBlocks(pi.length)
pi.search(uid-1, nBlocks)
}
// Search for uid in the current block
uidx := sort.Search(len(pi.out), func(idx int) bool {
return pi.out[idx] >= uid
})
if uidx < len(pi.out) && pi.out[uidx] == uid {
found = true
uidx++
}
// Expose slice whose startId > uid to the user
if uidx < len(pi.out) {
pi.out = pi.out[uidx:]
return
}
pi.Next()
return
}
func (pi *BPackIterator) Valid() bool {
return pi.valid
}
func (pi *BPackIterator) Length() int {
return pi.length
}
// Returns the startIndex
func (pi *BPackIterator) StartIdx() int {
return pi.count - len(pi.out)
}
func (pi *BPackIterator) Uids() []uint64 {
return pi.out
}
func (pi *BPackIterator) Next() {
if pi.count >= pi.length {
pi.valid = false
pi.out = pi.buf[:0]
return
}
sz := uint8(pi.data[pi.in_offset])
pi.in_offset++
if sz&bitVarint != 0 {
//varint is the last block and has less than blockSize integers
pi.out = pi.buf[:0]
for pi.count < pi.length {
i, n := binary.Uvarint(pi.data[pi.in_offset:])
pi.out = append(pi.out, i)
pi.in_offset += n
pi.count++
}
return
}
pi.out = pi.buf[:BlockSize]
funpack[sz](&pi.data[pi.in_offset], &pi.out[0], &pi.lastSeed[0])
pi.in_offset += (int(sz) * BlockSize) / 8
pi.count += BlockSize
}
func (pi *BPackIterator) SkipNext() {
if pi.count >= pi.length {
pi.valid = false
pi.out = pi.buf[:0]
return
}
// Find the bit size of the block
sz := uint8(pi.data[pi.in_offset])
// If it's varint block,(The last one)
if sz&bitVarint != 0 {
pi.in_offset = len(pi.data)
pi.count = pi.length
return
}
// Calculate size of the block based on bitsize
pi.in_offset += (int(sz)*BlockSize)/8 + 1
pi.count += BlockSize
// Update seed
i := (pi.count / BlockSize) * 20
pi.lastSeed[0] = binary.BigEndian.Uint64(pi.metadata[i : i+8])
pi.lastSeed[1] = binary.BigEndian.Uint64(pi.metadata[i+8 : i+16])
}
func (pi *BPackIterator) MaxIntInBlock() uint64 {
nBlocks := numBlocks(pi.length)
currBlock := pi.count / BlockSize
// We find max value through seed value stored in next meta block, so
// if it's a last block, we don't know the max so we return maxuint64
if currBlock >= nBlocks-1 {
return math.MaxUint64
}
// MaxInt in current block can be found by seed value of next block
midx := (currBlock + 1) * 20
return binary.BigEndian.Uint64(pi.metadata[midx+8 : midx+16])
}
func DeltaUnpack(in []byte, out []uint64) {
var bi BPackIterator
bi.Init(in, 0)
offset := 0
x.AssertTrue(len(out) == bi.Length())
for bi.Valid() {
uids := bi.Uids()
// Benchmarks would be slower due to this copy
copy(out[offset:], uids)
offset += len(uids)
bi.Next()
}
}
func DeltaPack(in []uint64) []byte {
var bp BPackEncoder
offset := 0
for offset+BlockSize <= len(in) {
bp.PackAppend(in[offset : offset+BlockSize])
offset += BlockSize
}
if offset < len(in) {
bp.PackAppend(in[offset:])
}
x := make([]byte, bp.Size())
bp.WriteTo(x)
return x
}