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
|
package flate
import (
"encoding/binary"
"fmt"
"math/bits"
)
// fastGen maintains the table for matches,
// and the previous byte block for level 2.
// This is the generic implementation.
type fastEncL1 struct {
fastGen
table [tableSize]tableEntry
}
// EncodeL1 uses a similar algorithm to level 1
func (e *fastEncL1) Encode(dst *tokens, src []byte) {
const (
inputMargin = 12 - 1
minNonLiteralBlockSize = 1 + 1 + inputMargin
)
if debugDeflate && e.cur < 0 {
panic(fmt.Sprint("e.cur < 0: ", e.cur))
}
// Protect against e.cur wraparound.
for e.cur >= bufferReset {
if len(e.hist) == 0 {
for i := range e.table[:] {
e.table[i] = tableEntry{}
}
e.cur = maxMatchOffset
break
}
// Shift down everything in the table that isn't already too far away.
minOff := e.cur + int32(len(e.hist)) - maxMatchOffset
for i := range e.table[:] {
v := e.table[i].offset
if v <= minOff {
v = 0
} else {
v = v - e.cur + maxMatchOffset
}
e.table[i].offset = v
}
e.cur = maxMatchOffset
}
s := e.addBlock(src)
// This check isn't in the Snappy implementation, but there, the caller
// instead of the callee handles this case.
if len(src) < minNonLiteralBlockSize {
// We do not fill the token table.
// This will be picked up by caller.
dst.n = uint16(len(src))
return
}
// Override src
src = e.hist
nextEmit := s
// sLimit is when to stop looking for offset/length copies. The inputMargin
// lets us use a fast path for emitLiteral in the main loop, while we are
// looking for copies.
sLimit := int32(len(src) - inputMargin)
// nextEmit is where in src the next emitLiteral should start from.
cv := load3232(src, s)
for {
const skipLog = 5
const doEvery = 2
nextS := s
var candidate tableEntry
for {
nextHash := hash(cv)
candidate = e.table[nextHash]
nextS = s + doEvery + (s-nextEmit)>>skipLog
if nextS > sLimit {
goto emitRemainder
}
now := load6432(src, nextS)
e.table[nextHash] = tableEntry{offset: s + e.cur}
nextHash = hash(uint32(now))
offset := s - (candidate.offset - e.cur)
if offset < maxMatchOffset && cv == load3232(src, candidate.offset-e.cur) {
e.table[nextHash] = tableEntry{offset: nextS + e.cur}
break
}
// Do one right away...
cv = uint32(now)
s = nextS
nextS++
candidate = e.table[nextHash]
now >>= 8
e.table[nextHash] = tableEntry{offset: s + e.cur}
offset = s - (candidate.offset - e.cur)
if offset < maxMatchOffset && cv == load3232(src, candidate.offset-e.cur) {
e.table[nextHash] = tableEntry{offset: nextS + e.cur}
break
}
cv = uint32(now)
s = nextS
}
// A 4-byte match has been found. We'll later see if more than 4 bytes
// match. But, prior to the match, src[nextEmit:s] are unmatched. Emit
// them as literal bytes.
for {
// Invariant: we have a 4-byte match at s, and no need to emit any
// literal bytes prior to s.
// Extend the 4-byte match as long as possible.
t := candidate.offset - e.cur
var l = int32(4)
if false {
l = e.matchlenLong(s+4, t+4, src) + 4
} else {
// inlined:
a := src[s+4:]
b := src[t+4:]
for len(a) >= 8 {
if diff := binary.LittleEndian.Uint64(a) ^ binary.LittleEndian.Uint64(b); diff != 0 {
l += int32(bits.TrailingZeros64(diff) >> 3)
break
}
l += 8
a = a[8:]
b = b[8:]
}
if len(a) < 8 {
b = b[:len(a)]
for i := range a {
if a[i] != b[i] {
break
}
l++
}
}
}
// Extend backwards
for t > 0 && s > nextEmit && src[t-1] == src[s-1] {
s--
t--
l++
}
if nextEmit < s {
if false {
emitLiteral(dst, src[nextEmit:s])
} else {
for _, v := range src[nextEmit:s] {
dst.tokens[dst.n] = token(v)
dst.litHist[v]++
dst.n++
}
}
}
// Save the match found
if false {
dst.AddMatchLong(l, uint32(s-t-baseMatchOffset))
} else {
// Inlined...
xoffset := uint32(s - t - baseMatchOffset)
xlength := l
oc := offsetCode(xoffset)
xoffset |= oc << 16
for xlength > 0 {
xl := xlength
if xl > 258 {
if xl > 258+baseMatchLength {
xl = 258
} else {
xl = 258 - baseMatchLength
}
}
xlength -= xl
xl -= baseMatchLength
dst.extraHist[lengthCodes1[uint8(xl)]]++
dst.offHist[oc]++
dst.tokens[dst.n] = token(matchType | uint32(xl)<<lengthShift | xoffset)
dst.n++
}
}
s += l
nextEmit = s
if nextS >= s {
s = nextS + 1
}
if s >= sLimit {
// Index first pair after match end.
if int(s+l+4) < len(src) {
cv := load3232(src, s)
e.table[hash(cv)] = tableEntry{offset: s + e.cur}
}
goto emitRemainder
}
// We could immediately start working at s now, but to improve
// compression we first update the hash table at s-2 and at s. If
// another emitCopy is not our next move, also calculate nextHash
// at s+1. At least on GOARCH=amd64, these three hash calculations
// are faster as one load64 call (with some shifts) instead of
// three load32 calls.
x := load6432(src, s-2)
o := e.cur + s - 2
prevHash := hash(uint32(x))
e.table[prevHash] = tableEntry{offset: o}
x >>= 16
currHash := hash(uint32(x))
candidate = e.table[currHash]
e.table[currHash] = tableEntry{offset: o + 2}
offset := s - (candidate.offset - e.cur)
if offset > maxMatchOffset || uint32(x) != load3232(src, candidate.offset-e.cur) {
cv = uint32(x >> 8)
s++
break
}
}
}
emitRemainder:
if int(nextEmit) < len(src) {
// If nothing was added, don't encode literals.
if dst.n == 0 {
return
}
emitLiteral(dst, src[nextEmit:])
}
}
|
