Repository navigation
Expand file tree
/
Copy pathblock.go
More file actions
413 lines (365 loc) · 10.4 KB
/
Copy pathblock.go
File metadata and controls
413 lines (365 loc) · 10.4 KB
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
409
410
411
412
413
package lz4
import (
"encoding/binary"
"math/bits"
"sync"
)
// blockHash hashes the lower 6 bytes into a value < htSize.
func blockHash(x uint64) uint32 {
const prime6bytes = 227718039650203
return uint32(((x << (64 - 48)) * prime6bytes) >> (64 - hashLog))
}
// CompressBlockBound returns the maximum size of a given buffer of size n, when not compressible.
func CompressBlockBound(n int) int {
return n + n/255 + 16
}
// UncompressBlock uncompresses the source buffer into the destination one,
// and returns the uncompressed size.
//
// The destination buffer must be sized appropriately.
//
// An error is returned if the source data is invalid or the destination buffer is too small.
func UncompressBlock(src, dst []byte) (int, error) {
if len(src) == 0 {
return 0, nil
}
if di := decodeBlock(dst, src); di >= 0 {
return di, nil
}
return 0, ErrInvalidSourceShortBuffer
}
// CompressBlock compresses the source buffer into the destination one.
// This is the fast version of LZ4 compression and also the default one.
//
// The argument hashTable is scratch space for a hash table used by the
// compressor. If provided, it should have length at least 1<<16. If it is
// shorter (or nil), CompressBlock allocates its own hash table.
//
// The size of the compressed data is returned.
//
// If the destination buffer size is lower than CompressBlockBound and
// the compressed size is 0 and no error, then the data is incompressible.
//
// An error is returned if the destination buffer is too small.
func CompressBlock(src, dst []byte, hashTable []int) (_ int, err error) {
defer recoverBlock(&err)
// Return 0, nil only if the destination buffer size is < CompressBlockBound.
isNotCompressible := len(dst) < CompressBlockBound(len(src))
// adaptSkipLog sets how quickly the compressor begins skipping blocks when data is incompressible.
// This significantly speeds up incompressible data and usually has very small impact on compression.
// bytes to skip = 1 + (bytes since last match >> adaptSkipLog)
const adaptSkipLog = 7
if len(hashTable) < htSize {
htIface := htPool.Get()
defer htPool.Put(htIface)
hashTable = (*(htIface).(*[htSize]int))[:]
}
// Prove to the compiler the table has at least htSize elements.
// The compiler can see that "uint32() >> hashShift" cannot be out of bounds.
hashTable = hashTable[:htSize]
// si: Current position of the search.
// anchor: Position of the current literals.
var si, di, anchor int
sn := len(src) - mfLimit
if sn <= 0 {
goto lastLiterals
}
// Fast scan strategy: the hash table only stores the last 4 bytes sequences.
for si < sn {
// Hash the next 6 bytes (sequence)...
match := binary.LittleEndian.Uint64(src[si:])
h := blockHash(match)
h2 := blockHash(match >> 8)
// We check a match at s, s+1 and s+2 and pick the first one we get.
// Checking 3 only requires us to load the source one.
ref := hashTable[h]
ref2 := hashTable[h2]
hashTable[h] = si
hashTable[h2] = si + 1
offset := si - ref
// If offset <= 0 we got an old entry in the hash table.
if offset <= 0 || offset >= winSize || // Out of window.
uint32(match) != binary.LittleEndian.Uint32(src[ref:]) { // Hash collision on different matches.
// No match. Start calculating another hash.
// The processor can usually do this out-of-order.
h = blockHash(match >> 16)
ref = hashTable[h]
// Check the second match at si+1
si += 1
offset = si - ref2
if offset <= 0 || offset >= winSize ||
uint32(match>>8) != binary.LittleEndian.Uint32(src[ref2:]) {
// No match. Check the third match at si+2
si += 1
offset = si - ref
hashTable[h] = si
if offset <= 0 || offset >= winSize ||
uint32(match>>16) != binary.LittleEndian.Uint32(src[ref:]) {
// Skip one extra byte (at si+3) before we check 3 matches again.
si += 2 + (si-anchor)>>adaptSkipLog
continue
}
}
}
// Match found.
lLen := si - anchor // Literal length.
// We already matched 4 bytes.
mLen := 4
// Extend backwards if we can, reducing literals.
tOff := si - offset - 1
for lLen > 0 && tOff >= 0 && src[si-1] == src[tOff] {
si--
tOff--
lLen--
mLen++
}
// Add the match length, so we continue search at the end.
// Use mLen to store the offset base.
si, mLen = si+mLen, si+minMatch
// Find the longest match by looking by batches of 8 bytes.
for si+8 < sn {
x := binary.LittleEndian.Uint64(src[si:]) ^ binary.LittleEndian.Uint64(src[si-offset:])
if x == 0 {
si += 8
} else {
// Stop is first non-zero byte.
si += bits.TrailingZeros64(x) >> 3
break
}
}
mLen = si - mLen
if mLen < 0xF {
dst[di] = byte(mLen)
} else {
dst[di] = 0xF
}
// Encode literals length.
if lLen < 0xF {
dst[di] |= byte(lLen << 4)
} else {
dst[di] |= 0xF0
di++
l := lLen - 0xF
for ; l >= 0xFF; l -= 0xFF {
dst[di] = 0xFF
di++
}
dst[di] = byte(l)
}
di++
// Literals.
copy(dst[di:di+lLen], src[anchor:anchor+lLen])
di += lLen + 2
anchor = si
// Encode offset.
_ = dst[di] // Bound check elimination.
dst[di-2], dst[di-1] = byte(offset), byte(offset>>8)
// Encode match length part 2.
if mLen >= 0xF {
for mLen -= 0xF; mLen >= 0xFF; mLen -= 0xFF {
dst[di] = 0xFF
di++
}
dst[di] = byte(mLen)
di++
}
// Check if we can load next values.
if si >= sn {
break
}
// Hash match end-2
h = blockHash(binary.LittleEndian.Uint64(src[si-2:]))
hashTable[h] = si - 2
}
lastLiterals:
if isNotCompressible && anchor == 0 {
// Incompressible.
return 0, nil
}
// Last literals.
lLen := len(src) - anchor
if lLen < 0xF {
dst[di] = byte(lLen << 4)
} else {
dst[di] = 0xF0
di++
for lLen -= 0xF; lLen >= 0xFF; lLen -= 0xFF {
dst[di] = 0xFF
di++
}
dst[di] = byte(lLen)
}
di++
// Write the last literals.
if isNotCompressible && di >= anchor {
// Incompressible.
return 0, nil
}
di += copy(dst[di:di+len(src)-anchor], src[anchor:])
return di, nil
}
// Pool of hash tables for CompressBlock.
var htPool = sync.Pool{
New: func() interface{} {
return new([htSize]int)
},
}
// blockHash hashes 4 bytes into a value < winSize.
func blockHashHC(x uint32) uint32 {
const hasher uint32 = 2654435761 // Knuth multiplicative hash.
return x * hasher >> (32 - winSizeLog)
}
// CompressBlockHC compresses the source buffer src into the destination dst
// with max search depth (use 0 or negative value for no max).
//
// CompressBlockHC compression ratio is better than CompressBlock but it is also slower.
//
// The size of the compressed data is returned.
//
// If the destination buffer size is lower than CompressBlockBound and
// the compressed size is 0 and no error, then the data is incompressible.
//
// An error is returned if the destination buffer is too small.
func CompressBlockHC(src, dst []byte, depth int) (_ int, err error) {
defer recoverBlock(&err)
// Return 0, nil only if the destination buffer size is < CompressBlockBound.
isNotCompressible := len(dst) < CompressBlockBound(len(src))
// adaptSkipLog sets how quickly the compressor begins skipping blocks when data is incompressible.
// This significantly speeds up incompressible data and usually has very small impact on compression.
// bytes to skip = 1 + (bytes since last match >> adaptSkipLog)
const adaptSkipLog = 7
var si, di, anchor int
// hashTable: stores the last position found for a given hash
// chainTable: stores previous positions for a given hash
var hashTable, chainTable [winSize]int
if depth <= 0 {
depth = winSize
}
sn := len(src) - mfLimit
if sn <= 0 {
goto lastLiterals
}
for si < sn {
// Hash the next 4 bytes (sequence).
match := binary.LittleEndian.Uint32(src[si:])
h := blockHashHC(match)
// Follow the chain until out of window and give the longest match.
mLen := 0
offset := 0
for next, try := hashTable[h], depth; try > 0 && next > 0 && si-next < winSize; next = chainTable[next&winMask] {
// The first (mLen==0) or next byte (mLen>=minMatch) at current match length
// must match to improve on the match length.
if src[next+mLen] != src[si+mLen] {
continue
}
ml := 0
// Compare the current position with a previous with the same hash.
for ml < sn-si {
x := binary.LittleEndian.Uint64(src[next+ml:]) ^ binary.LittleEndian.Uint64(src[si+ml:])
if x == 0 {
ml += 8
} else {
// Stop is first non-zero byte.
ml += bits.TrailingZeros64(x) >> 3
break
}
}
if ml < minMatch || ml <= mLen {
// Match too small (<minMath) or smaller than the current match.
continue
}
// Found a longer match, keep its position and length.
mLen = ml
offset = si - next
// Try another previous position with the same hash.
try--
}
chainTable[si&winMask] = hashTable[h]
hashTable[h] = si
// No match found.
if mLen == 0 {
si += 1 + (si-anchor)>>adaptSkipLog
continue
}
// Match found.
// Update hash/chain tables with overlapping bytes:
// si already hashed, add everything from si+1 up to the match length.
winStart := si + 1
if ws := si + mLen - winSize; ws > winStart {
winStart = ws
}
for si, ml := winStart, si+mLen; si < ml; {
match >>= 8
match |= uint32(src[si+3]) << 24
h := blockHashHC(match)
chainTable[si&winMask] = hashTable[h]
hashTable[h] = si
si++
}
lLen := si - anchor
si += mLen
mLen -= minMatch // Match length does not include minMatch.
if mLen < 0xF {
dst[di] = byte(mLen)
} else {
dst[di] = 0xF
}
// Encode literals length.
if lLen < 0xF {
dst[di] |= byte(lLen << 4)
} else {
dst[di] |= 0xF0
di++
l := lLen - 0xF
for ; l >= 0xFF; l -= 0xFF {
dst[di] = 0xFF
di++
}
dst[di] = byte(l)
}
di++
// Literals.
copy(dst[di:di+lLen], src[anchor:anchor+lLen])
di += lLen
anchor = si
// Encode offset.
di += 2
dst[di-2], dst[di-1] = byte(offset), byte(offset>>8)
// Encode match length part 2.
if mLen >= 0xF {
for mLen -= 0xF; mLen >= 0xFF; mLen -= 0xFF {
dst[di] = 0xFF
di++
}
dst[di] = byte(mLen)
di++
}
}
if isNotCompressible && anchor == 0 {
// Incompressible.
return 0, nil
}
// Last literals.
lastLiterals:
lLen := len(src) - anchor
if lLen < 0xF {
dst[di] = byte(lLen << 4)
} else {
dst[di] = 0xF0
di++
lLen -= 0xF
for ; lLen >= 0xFF; lLen -= 0xFF {
dst[di] = 0xFF
di++
}
dst[di] = byte(lLen)
}
di++
// Write the last literals.
if isNotCompressible && di >= anchor {
// Incompressible.
return 0, nil
}
di += copy(dst[di:di+len(src)-anchor], src[anchor:])
return di, nil
}