aboutsummaryrefslogtreecommitdiff
path: root/vendor/github.com/klauspost/compress/flate/stateless.go
blob: 544162a4318fafb32d66b0cefb9656e6a8437a7b (plain)
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
package flate

import (
	"io"
	"math"
	"sync"
)

const (
	maxStatelessBlock = math.MaxInt16
	// dictionary will be taken from maxStatelessBlock, so limit it.
	maxStatelessDict = 8 << 10

	slTableBits  = 13
	slTableSize  = 1 << slTableBits
	slTableShift = 32 - slTableBits
)

type statelessWriter struct {
	dst    io.Writer
	closed bool
}

func (s *statelessWriter) Close() error {
	if s.closed {
		return nil
	}
	s.closed = true
	// Emit EOF block
	return StatelessDeflate(s.dst, nil, true, nil)
}

func (s *statelessWriter) Write(p []byte) (n int, err error) {
	err = StatelessDeflate(s.dst, p, false, nil)
	if err != nil {
		return 0, err
	}
	return len(p), nil
}

func (s *statelessWriter) Reset(w io.Writer) {
	s.dst = w
	s.closed = false
}

// NewStatelessWriter will do compression but without maintaining any state
// between Write calls.
// There will be no memory kept between Write calls,
// but compression and speed will be suboptimal.
// Because of this, the size of actual Write calls will affect output size.
func NewStatelessWriter(dst io.Writer) io.WriteCloser {
	return &statelessWriter{dst: dst}
}

// bitWriterPool contains bit writers that can be reused.
var bitWriterPool = sync.Pool{
	New: func() interface{} {
		return newHuffmanBitWriter(nil)
	},
}

// StatelessDeflate allows to compress directly to a Writer without retaining state.
// When returning everything will be flushed.
// Up to 8KB of an optional dictionary can be given which is presumed to presumed to precede the block.
// Longer dictionaries will be truncated and will still produce valid output.
// Sending nil dictionary is perfectly fine.
func StatelessDeflate(out io.Writer, in []byte, eof bool, dict []byte) error {
	var dst tokens
	bw := bitWriterPool.Get().(*huffmanBitWriter)
	bw.reset(out)
	defer func() {
		// don't keep a reference to our output
		bw.reset(nil)
		bitWriterPool.Put(bw)
	}()
	if eof && len(in) == 0 {
		// Just write an EOF block.
		// Could be faster...
		bw.writeStoredHeader(0, true)
		bw.flush()
		return bw.err
	}

	// Truncate dict
	if len(dict) > maxStatelessDict {
		dict = dict[len(dict)-maxStatelessDict:]
	}

	for len(in) > 0 {
		todo := in
		if len(todo) > maxStatelessBlock-len(dict) {
			todo = todo[:maxStatelessBlock-len(dict)]
		}
		in = in[len(todo):]
		uncompressed := todo
		if len(dict) > 0 {
			// combine dict and source
			bufLen := len(todo) + len(dict)
			combined := make([]byte, bufLen)
			copy(combined, dict)
			copy(combined[len(dict):], todo)
			todo = combined
		}
		// Compress
		statelessEnc(&dst, todo, int16(len(dict)))
		isEof := eof && len(in) == 0

		if dst.n == 0 {
			bw.writeStoredHeader(len(uncompressed), isEof)
			if bw.err != nil {
				return bw.err
			}
			bw.writeBytes(uncompressed)
		} else if int(dst.n) > len(uncompressed)-len(uncompressed)>>4 {
			// If we removed less than 1/16th, huffman compress the block.
			bw.writeBlockHuff(isEof, uncompressed, len(in) == 0)
		} else {
			bw.writeBlockDynamic(&dst, isEof, uncompressed, len(in) == 0)
		}
		if len(in) > 0 {
			// Retain a dict if we have more
			dict = todo[len(todo)-maxStatelessDict:]
			dst.Reset()
		}
		if bw.err != nil {
			return bw.err
		}
	}
	if !eof {
		// Align, only a stored block can do that.
		bw.writeStoredHeader(0, false)
	}
	bw.flush()
	return bw.err
}

func hashSL(u uint32) uint32 {
	return (u * 0x1e35a7bd) >> slTableShift
}

func load3216(b []byte, i int16) uint32 {
	// Help the compiler eliminate bounds checks on the read so it can be done in a single read.
	b = b[i:]
	b = b[:4]
	return uint32(b[0]) | uint32(b[1])<<8 | uint32(b[2])<<16 | uint32(b[3])<<24
}

func load6416(b []byte, i int16) uint64 {
	// Help the compiler eliminate bounds checks on the read so it can be done in a single read.
	b = b[i:]
	b = b[:8]
	return uint64(b[0]) | uint64(b[1])<<8 | uint64(b[2])<<16 | uint64(b[3])<<24 |
		uint64(b[4])<<32 | uint64(b[5])<<40 | uint64(b[6])<<48 | uint64(b[7])<<56
}

func statelessEnc(dst *tokens, src []byte, startAt int16) {
	const (
		inputMargin            = 12 - 1
		minNonLiteralBlockSize = 1 + 1 + inputMargin
	)

	type tableEntry struct {
		offset int16
	}

	var table [slTableSize]tableEntry

	// This check isn't in the Snappy implementation, but there, the caller
	// instead of the callee handles this case.
	if len(src)-int(startAt) < minNonLiteralBlockSize {
		// We do not fill the token table.
		// This will be picked up by caller.
		dst.n = 0
		return
	}
	// Index until startAt
	if startAt > 0 {
		cv := load3232(src, 0)
		for i := int16(0); i < startAt; i++ {
			table[hashSL(cv)] = tableEntry{offset: i}
			cv = (cv >> 8) | (uint32(src[i+4]) << 24)
		}
	}

	s := startAt + 1
	nextEmit := startAt
	// 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 := int16(len(src) - inputMargin)

	// nextEmit is where in src the next emitLiteral should start from.
	cv := load3216(src, s)

	for {
		const skipLog = 5
		const doEvery = 2

		nextS := s
		var candidate tableEntry
		for {
			nextHash := hashSL(cv)
			candidate = table[nextHash]
			nextS = s + doEvery + (s-nextEmit)>>skipLog
			if nextS > sLimit || nextS <= 0 {
				goto emitRemainder
			}

			now := load6416(src, nextS)
			table[nextHash] = tableEntry{offset: s}
			nextHash = hashSL(uint32(now))

			if cv == load3216(src, candidate.offset) {
				table[nextHash] = tableEntry{offset: nextS}
				break
			}

			// Do one right away...
			cv = uint32(now)
			s = nextS
			nextS++
			candidate = table[nextHash]
			now >>= 8
			table[nextHash] = tableEntry{offset: s}

			if cv == load3216(src, candidate.offset) {
				table[nextHash] = tableEntry{offset: nextS}
				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
			l := int16(matchLen(src[s+4:], src[t+4:]) + 4)

			// 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
			dst.AddMatchLong(int32(l), uint32(s-t-baseMatchOffset))
			s += l
			nextEmit = s
			if nextS >= s {
				s = nextS + 1
			}
			if s >= sLimit {
				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 := load6416(src, s-2)
			o := s - 2
			prevHash := hashSL(uint32(x))
			table[prevHash] = tableEntry{offset: o}
			x >>= 16
			currHash := hashSL(uint32(x))
			candidate = table[currHash]
			table[currHash] = tableEntry{offset: o + 2}

			if uint32(x) != load3216(src, candidate.offset) {
				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:])
	}
}