summaryrefslogtreecommitdiff
path: root/vendor/github.com/klauspost/compress/zstd/seqdec.go
blob: cef69e35b52f58ca7954fd467c923ae9504ed92e (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
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
// Copyright 2019+ Klaus Post. All rights reserved.
// License information can be found in the LICENSE file.
// Based on work by Yann Collet, released under BSD License.

package zstd

import (
	"errors"
	"fmt"
	"io"
)

type seq struct {
	litLen   uint32
	matchLen uint32
	offset   uint32

	// Codes are stored here for the encoder
	// so they only have to be looked up once.
	llCode, mlCode, ofCode uint8
}

func (s seq) String() string {
	if s.offset <= 3 {
		if s.offset == 0 {
			return fmt.Sprint("litLen:", s.litLen, ", matchLen:", s.matchLen+zstdMinMatch, ", offset: INVALID (0)")
		}
		return fmt.Sprint("litLen:", s.litLen, ", matchLen:", s.matchLen+zstdMinMatch, ", offset:", s.offset, " (repeat)")
	}
	return fmt.Sprint("litLen:", s.litLen, ", matchLen:", s.matchLen+zstdMinMatch, ", offset:", s.offset-3, " (new)")
}

type seqCompMode uint8

const (
	compModePredefined seqCompMode = iota
	compModeRLE
	compModeFSE
	compModeRepeat
)

type sequenceDec struct {
	// decoder keeps track of the current state and updates it from the bitstream.
	fse    *fseDecoder
	state  fseState
	repeat bool
}

// init the state of the decoder with input from stream.
func (s *sequenceDec) init(br *bitReader) error {
	if s.fse == nil {
		return errors.New("sequence decoder not defined")
	}
	s.state.init(br, s.fse.actualTableLog, s.fse.dt[:1<<s.fse.actualTableLog])
	return nil
}

// sequenceDecs contains all 3 sequence decoders and their state.
type sequenceDecs struct {
	litLengths   sequenceDec
	offsets      sequenceDec
	matchLengths sequenceDec
	prevOffset   [3]int
	hist         []byte
	literals     []byte
	out          []byte
	maxBits      uint8
}

// initialize all 3 decoders from the stream input.
func (s *sequenceDecs) initialize(br *bitReader, hist *history, literals, out []byte) error {
	if err := s.litLengths.init(br); err != nil {
		return errors.New("litLengths:" + err.Error())
	}
	if err := s.offsets.init(br); err != nil {
		return errors.New("offsets:" + err.Error())
	}
	if err := s.matchLengths.init(br); err != nil {
		return errors.New("matchLengths:" + err.Error())
	}
	s.literals = literals
	s.hist = hist.b
	s.prevOffset = hist.recentOffsets
	s.maxBits = s.litLengths.fse.maxBits + s.offsets.fse.maxBits + s.matchLengths.fse.maxBits
	s.out = out
	return nil
}

// decode sequences from the stream with the provided history.
func (s *sequenceDecs) decode(seqs int, br *bitReader, hist []byte) error {
	startSize := len(s.out)
	for i := seqs - 1; i >= 0; i-- {
		if br.overread() {
			printf("reading sequence %d, exceeded available data\n", seqs-i)
			return io.ErrUnexpectedEOF
		}
		var litLen, matchOff, matchLen int
		if br.off > 4+((maxOffsetBits+16+16)>>3) {
			litLen, matchOff, matchLen = s.nextFast(br)
			br.fillFast()
		} else {
			litLen, matchOff, matchLen = s.next(br)
			br.fill()
		}

		if debugSequences {
			println("Seq", seqs-i-1, "Litlen:", litLen, "matchOff:", matchOff, "(abs) matchLen:", matchLen)
		}

		if litLen > len(s.literals) {
			return fmt.Errorf("unexpected literal count, want %d bytes, but only %d is available", litLen, len(s.literals))
		}
		size := litLen + matchLen + len(s.out)
		if size-startSize > maxBlockSize {
			return fmt.Errorf("output (%d) bigger than max block size", size)
		}
		if size > cap(s.out) {
			// Not enough size, will be extremely rarely triggered,
			// but could be if destination slice is too small for sync operations.
			// We add maxBlockSize to the capacity.
			s.out = append(s.out, make([]byte, maxBlockSize)...)
			s.out = s.out[:len(s.out)-maxBlockSize]
		}
		if matchLen > maxMatchLen {
			return fmt.Errorf("match len (%d) bigger than max allowed length", matchLen)
		}
		if matchOff > len(s.out)+len(hist)+litLen {
			return fmt.Errorf("match offset (%d) bigger than current history (%d)", matchOff, len(s.out)+len(hist)+litLen)
		}
		if matchOff == 0 && matchLen > 0 {
			return fmt.Errorf("zero matchoff and matchlen > 0")
		}

		s.out = append(s.out, s.literals[:litLen]...)
		s.literals = s.literals[litLen:]
		out := s.out

		// Copy from history.
		// TODO: Blocks without history could be made to ignore this completely.
		if v := matchOff - len(s.out); v > 0 {
			// v is the start position in history from end.
			start := len(s.hist) - v
			if matchLen > v {
				// Some goes into current block.
				// Copy remainder of history
				out = append(out, s.hist[start:]...)
				matchOff -= v
				matchLen -= v
			} else {
				out = append(out, s.hist[start:start+matchLen]...)
				matchLen = 0
			}
		}
		// We must be in current buffer now
		if matchLen > 0 {
			start := len(s.out) - matchOff
			if matchLen <= len(s.out)-start {
				// No overlap
				out = append(out, s.out[start:start+matchLen]...)
			} else {
				// Overlapping copy
				// Extend destination slice and copy one byte at the time.
				out = out[:len(out)+matchLen]
				src := out[start : start+matchLen]
				// Destination is the space we just added.
				dst := out[len(out)-matchLen:]
				dst = dst[:len(src)]
				for i := range src {
					dst[i] = src[i]
				}
			}
		}
		s.out = out
		if i == 0 {
			// This is the last sequence, so we shouldn't update state.
			break
		}
		if true {
			// Manually inlined, ~ 5-20% faster
			// Update all 3 states at once. Approx 20% faster.
			a, b, c := s.litLengths.state.state, s.matchLengths.state.state, s.offsets.state.state

			nBits := a.nbBits + b.nbBits + c.nbBits
			if nBits == 0 {
				s.litLengths.state.state = s.litLengths.state.dt[a.newState]
				s.matchLengths.state.state = s.matchLengths.state.dt[b.newState]
				s.offsets.state.state = s.offsets.state.dt[c.newState]
			} else {
				bits := br.getBitsFast(nBits)
				lowBits := uint16(bits >> ((c.nbBits + b.nbBits) & 31))
				s.litLengths.state.state = s.litLengths.state.dt[a.newState+lowBits]

				lowBits = uint16(bits >> (c.nbBits & 31))
				lowBits &= bitMask[b.nbBits&15]
				s.matchLengths.state.state = s.matchLengths.state.dt[b.newState+lowBits]

				lowBits = uint16(bits) & bitMask[c.nbBits&15]
				s.offsets.state.state = s.offsets.state.dt[c.newState+lowBits]
			}
		} else {
			s.updateAlt(br)
		}
	}

	// Add final literals
	s.out = append(s.out, s.literals...)
	return nil
}

// update states, at least 27 bits must be available.
func (s *sequenceDecs) update(br *bitReader) {
	// Max 8 bits
	s.litLengths.state.next(br)
	// Max 9 bits
	s.matchLengths.state.next(br)
	// Max 8 bits
	s.offsets.state.next(br)
}

var bitMask [16]uint16

func init() {
	for i := range bitMask[:] {
		bitMask[i] = uint16((1 << uint(i)) - 1)
	}
}

// update states, at least 27 bits must be available.
func (s *sequenceDecs) updateAlt(br *bitReader) {
	// Update all 3 states at once. Approx 20% faster.
	a, b, c := s.litLengths.state.state, s.matchLengths.state.state, s.offsets.state.state

	nBits := a.nbBits + b.nbBits + c.nbBits
	if nBits == 0 {
		s.litLengths.state.state = s.litLengths.state.dt[a.newState]
		s.matchLengths.state.state = s.matchLengths.state.dt[b.newState]
		s.offsets.state.state = s.offsets.state.dt[c.newState]
		return
	}
	bits := br.getBitsFast(nBits)
	lowBits := uint16(bits >> ((c.nbBits + b.nbBits) & 31))
	s.litLengths.state.state = s.litLengths.state.dt[a.newState+lowBits]

	lowBits = uint16(bits >> (c.nbBits & 31))
	lowBits &= bitMask[b.nbBits&15]
	s.matchLengths.state.state = s.matchLengths.state.dt[b.newState+lowBits]

	lowBits = uint16(bits) & bitMask[c.nbBits&15]
	s.offsets.state.state = s.offsets.state.dt[c.newState+lowBits]
}

// nextFast will return new states when there are at least 4 unused bytes left on the stream when done.
func (s *sequenceDecs) nextFast(br *bitReader) (ll, mo, ml int) {
	// Final will not read from stream.
	ll, llB := s.litLengths.state.final()
	ml, mlB := s.matchLengths.state.final()
	mo, moB := s.offsets.state.final()

	// extra bits are stored in reverse order.
	br.fillFast()
	if s.maxBits <= 32 {
		mo += br.getBits(moB)
		ml += br.getBits(mlB)
		ll += br.getBits(llB)
	} else {
		mo += br.getBits(moB)
		br.fillFast()
		// matchlength+literal length, max 32 bits
		ml += br.getBits(mlB)
		ll += br.getBits(llB)
	}

	// mo = s.adjustOffset(mo, ll, moB)
	// Inlined for rather big speedup
	if moB > 1 {
		s.prevOffset[2] = s.prevOffset[1]
		s.prevOffset[1] = s.prevOffset[0]
		s.prevOffset[0] = mo
		return
	}

	if ll == 0 {
		// There is an exception though, when current sequence's literals_length = 0.
		// In this case, repeated offsets are shifted by one, so an offset_value of 1 means Repeated_Offset2,
		// an offset_value of 2 means Repeated_Offset3, and an offset_value of 3 means Repeated_Offset1 - 1_byte.
		mo++
	}

	if mo == 0 {
		mo = s.prevOffset[0]
		return
	}
	var temp int
	if mo == 3 {
		temp = s.prevOffset[0] - 1
	} else {
		temp = s.prevOffset[mo]
	}

	if temp == 0 {
		// 0 is not valid; input is corrupted; force offset to 1
		println("temp was 0")
		temp = 1
	}

	if mo != 1 {
		s.prevOffset[2] = s.prevOffset[1]
	}
	s.prevOffset[1] = s.prevOffset[0]
	s.prevOffset[0] = temp
	mo = temp
	return
}

func (s *sequenceDecs) next(br *bitReader) (ll, mo, ml int) {
	// Final will not read from stream.
	ll, llB := s.litLengths.state.final()
	ml, mlB := s.matchLengths.state.final()
	mo, moB := s.offsets.state.final()

	// extra bits are stored in reverse order.
	br.fill()
	if s.maxBits <= 32 {
		mo += br.getBits(moB)
		ml += br.getBits(mlB)
		ll += br.getBits(llB)
	} else {
		mo += br.getBits(moB)
		br.fill()
		// matchlength+literal length, max 32 bits
		ml += br.getBits(mlB)
		ll += br.getBits(llB)

	}
	mo = s.adjustOffset(mo, ll, moB)
	return
}

func (s *sequenceDecs) adjustOffset(offset, litLen int, offsetB uint8) int {
	if offsetB > 1 {
		s.prevOffset[2] = s.prevOffset[1]
		s.prevOffset[1] = s.prevOffset[0]
		s.prevOffset[0] = offset
		return offset
	}

	if litLen == 0 {
		// There is an exception though, when current sequence's literals_length = 0.
		// In this case, repeated offsets are shifted by one, so an offset_value of 1 means Repeated_Offset2,
		// an offset_value of 2 means Repeated_Offset3, and an offset_value of 3 means Repeated_Offset1 - 1_byte.
		offset++
	}

	if offset == 0 {
		return s.prevOffset[0]
	}
	var temp int
	if offset == 3 {
		temp = s.prevOffset[0] - 1
	} else {
		temp = s.prevOffset[offset]
	}

	if temp == 0 {
		// 0 is not valid; input is corrupted; force offset to 1
		println("temp was 0")
		temp = 1
	}

	if offset != 1 {
		s.prevOffset[2] = s.prevOffset[1]
	}
	s.prevOffset[1] = s.prevOffset[0]
	s.prevOffset[0] = temp
	return temp
}

// mergeHistory will merge history.
func (s *sequenceDecs) mergeHistory(hist *sequenceDecs) (*sequenceDecs, error) {
	for i := uint(0); i < 3; i++ {
		var sNew, sHist *sequenceDec
		switch i {
		default:
			// same as "case 0":
			sNew = &s.litLengths
			sHist = &hist.litLengths
		case 1:
			sNew = &s.offsets
			sHist = &hist.offsets
		case 2:
			sNew = &s.matchLengths
			sHist = &hist.matchLengths
		}
		if sNew.repeat {
			if sHist.fse == nil {
				return nil, fmt.Errorf("sequence stream %d, repeat requested, but no history", i)
			}
			continue
		}
		if sNew.fse == nil {
			return nil, fmt.Errorf("sequence stream %d, no fse found", i)
		}
		if sHist.fse != nil && !sHist.fse.preDefined {
			fseDecoderPool.Put(sHist.fse)
		}
		sHist.fse = sNew.fse
	}
	return hist, nil
}