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
|
package flate
import "fmt"
// fastGen maintains the table for matches,
// and the previous byte block for level 2.
// This is the generic implementation.
type fastEncL2 struct {
fastGen
table [bTableSize]tableEntry
}
// EncodeL2 uses a similar algorithm to level 1, but is capable
// of matching across blocks giving better compression at a small slowdown.
func (e *fastEncL2) Encode(dst *tokens, src []byte) {
const (
inputMargin = 12 - 1
minNonLiteralBlockSize = 1 + 1 + inputMargin
)
if debugDecode && 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 {
// When should we start skipping if we haven't found matches in a long while.
const skipLog = 5
const doEvery = 2
nextS := s
var candidate tableEntry
for {
nextHash := hash4u(cv, bTableBits)
s = nextS
nextS = s + doEvery + (s-nextEmit)>>skipLog
if nextS > sLimit {
goto emitRemainder
}
candidate = e.table[nextHash]
now := load6432(src, nextS)
e.table[nextHash] = tableEntry{offset: s + e.cur, val: cv}
nextHash = hash4u(uint32(now), bTableBits)
offset := s - (candidate.offset - e.cur)
if offset < maxMatchOffset && cv == candidate.val {
e.table[nextHash] = tableEntry{offset: nextS + e.cur, val: uint32(now)}
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, val: cv}
offset = s - (candidate.offset - e.cur)
if offset < maxMatchOffset && cv == candidate.val {
break
}
cv = uint32(now)
}
// 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.
// Call emitCopy, and then see if another emitCopy could be our next
// move. Repeat until we find no match for the input immediately after
// what was consumed by the last emitCopy call.
//
// If we exit this loop normally then we need to call emitLiteral next,
// though we don't yet know how big the literal will be. We handle that
// by proceeding to the next iteration of the main loop. We also can
// exit this loop via goto if we get close to exhausting the input.
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
l := e.matchlenLong(s+4, t+4, src) + 4
// Extend backwards
for t > 0 && s > nextEmit && src[t-1] == src[s-1] {
s--
t--
l++
}
if nextEmit < s {
emitLiteral(dst, src[nextEmit:s])
}
dst.AddMatchLong(l, uint32(s-t-baseMatchOffset))
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[hash4u(cv, bTableBits)] = tableEntry{offset: s + e.cur, val: cv}
}
goto emitRemainder
}
// Store every second hash in-between, but offset by 1.
for i := s - l + 2; i < s-5; i += 7 {
x := load6432(src, int32(i))
nextHash := hash4u(uint32(x), bTableBits)
e.table[nextHash] = tableEntry{offset: e.cur + i, val: uint32(x)}
// Skip one
x >>= 16
nextHash = hash4u(uint32(x), bTableBits)
e.table[nextHash] = tableEntry{offset: e.cur + i + 2, val: uint32(x)}
// Skip one
x >>= 16
nextHash = hash4u(uint32(x), bTableBits)
e.table[nextHash] = tableEntry{offset: e.cur + i + 4, val: uint32(x)}
}
// We could immediately start working at s now, but to improve
// compression we first update the hash table at s-2 to 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 := hash4u(uint32(x), bTableBits)
prevHash2 := hash4u(uint32(x>>8), bTableBits)
e.table[prevHash] = tableEntry{offset: o, val: uint32(x)}
e.table[prevHash2] = tableEntry{offset: o + 1, val: uint32(x >> 8)}
currHash := hash4u(uint32(x>>16), bTableBits)
candidate = e.table[currHash]
e.table[currHash] = tableEntry{offset: o + 2, val: uint32(x >> 16)}
offset := s - (candidate.offset - e.cur)
if offset > maxMatchOffset || uint32(x>>16) != candidate.val {
cv = uint32(x >> 24)
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:])
}
}
|
