aboutsummaryrefslogtreecommitdiff
path: root/vendor/github.com/klauspost/compress/zstd/fse_decoder.go
blob: 9efe34feb34815ad118ef6ebeb51121f39d91e25 (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
// 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"
)

const (
	tablelogAbsoluteMax = 9
)

const (
	/*!MEMORY_USAGE :
	 *  Memory usage formula : N->2^N Bytes (examples : 10 -> 1KB; 12 -> 4KB ; 16 -> 64KB; 20 -> 1MB; etc.)
	 *  Increasing memory usage improves compression ratio
	 *  Reduced memory usage can improve speed, due to cache effect
	 *  Recommended max value is 14, for 16KB, which nicely fits into Intel x86 L1 cache */
	maxMemoryUsage = 11

	maxTableLog    = maxMemoryUsage - 2
	maxTablesize   = 1 << maxTableLog
	maxTableMask   = (1 << maxTableLog) - 1
	minTablelog    = 5
	maxSymbolValue = 255
)

// fseDecoder provides temporary storage for compression and decompression.
type fseDecoder struct {
	dt             [maxTablesize]decSymbol // Decompression table.
	symbolLen      uint16                  // Length of active part of the symbol table.
	actualTableLog uint8                   // Selected tablelog.
	maxBits        uint8                   // Maximum number of additional bits

	// used for table creation to avoid allocations.
	stateTable [256]uint16
	norm       [maxSymbolValue + 1]int16
	preDefined bool
}

// tableStep returns the next table index.
func tableStep(tableSize uint32) uint32 {
	return (tableSize >> 1) + (tableSize >> 3) + 3
}

// readNCount will read the symbol distribution so decoding tables can be constructed.
func (s *fseDecoder) readNCount(b *byteReader, maxSymbol uint16) error {
	var (
		charnum   uint16
		previous0 bool
	)
	if b.remain() < 4 {
		return errors.New("input too small")
	}
	bitStream := b.Uint32()
	nbBits := uint((bitStream & 0xF) + minTablelog) // extract tableLog
	if nbBits > tablelogAbsoluteMax {
		println("Invalid tablelog:", nbBits)
		return errors.New("tableLog too large")
	}
	bitStream >>= 4
	bitCount := uint(4)

	s.actualTableLog = uint8(nbBits)
	remaining := int32((1 << nbBits) + 1)
	threshold := int32(1 << nbBits)
	gotTotal := int32(0)
	nbBits++

	for remaining > 1 && charnum <= maxSymbol {
		if previous0 {
			//println("prev0")
			n0 := charnum
			for (bitStream & 0xFFFF) == 0xFFFF {
				//println("24 x 0")
				n0 += 24
				if r := b.remain(); r > 5 {
					b.advance(2)
					bitStream = b.Uint32() >> bitCount
				} else {
					// end of bit stream
					bitStream >>= 16
					bitCount += 16
				}
			}
			//printf("bitstream: %d, 0b%b", bitStream&3, bitStream)
			for (bitStream & 3) == 3 {
				n0 += 3
				bitStream >>= 2
				bitCount += 2
			}
			n0 += uint16(bitStream & 3)
			bitCount += 2

			if n0 > maxSymbolValue {
				return errors.New("maxSymbolValue too small")
			}
			//println("inserting ", n0-charnum, "zeroes from idx", charnum, "ending before", n0)
			for charnum < n0 {
				s.norm[uint8(charnum)] = 0
				charnum++
			}

			if r := b.remain(); r >= 7 || r+int(bitCount>>3) >= 4 {
				b.advance(bitCount >> 3)
				bitCount &= 7
				bitStream = b.Uint32() >> bitCount
			} else {
				bitStream >>= 2
			}
		}

		max := (2*threshold - 1) - remaining
		var count int32

		if int32(bitStream)&(threshold-1) < max {
			count = int32(bitStream) & (threshold - 1)
			if debug && nbBits < 1 {
				panic("nbBits underflow")
			}
			bitCount += nbBits - 1
		} else {
			count = int32(bitStream) & (2*threshold - 1)
			if count >= threshold {
				count -= max
			}
			bitCount += nbBits
		}

		// extra accuracy
		count--
		if count < 0 {
			// -1 means +1
			remaining += count
			gotTotal -= count
		} else {
			remaining -= count
			gotTotal += count
		}
		s.norm[charnum&0xff] = int16(count)
		charnum++
		previous0 = count == 0
		for remaining < threshold {
			nbBits--
			threshold >>= 1
		}

		//println("b.off:", b.off, "len:", len(b.b), "bc:", bitCount, "remain:", b.remain())
		if r := b.remain(); r >= 7 || r+int(bitCount>>3) >= 4 {
			b.advance(bitCount >> 3)
			bitCount &= 7
		} else {
			bitCount -= (uint)(8 * (len(b.b) - 4 - b.off))
			b.off = len(b.b) - 4
			//println("b.off:", b.off, "len:", len(b.b), "bc:", bitCount, "iend", iend)
		}
		bitStream = b.Uint32() >> (bitCount & 31)
		//printf("bitstream is now: 0b%b", bitStream)
	}
	s.symbolLen = charnum
	if s.symbolLen <= 1 {
		return fmt.Errorf("symbolLen (%d) too small", s.symbolLen)
	}
	if s.symbolLen > maxSymbolValue+1 {
		return fmt.Errorf("symbolLen (%d) too big", s.symbolLen)
	}
	if remaining != 1 {
		return fmt.Errorf("corruption detected (remaining %d != 1)", remaining)
	}
	if bitCount > 32 {
		return fmt.Errorf("corruption detected (bitCount %d > 32)", bitCount)
	}
	if gotTotal != 1<<s.actualTableLog {
		return fmt.Errorf("corruption detected (total %d != %d)", gotTotal, 1<<s.actualTableLog)
	}
	b.advance((bitCount + 7) >> 3)
	// println(s.norm[:s.symbolLen], s.symbolLen)
	return s.buildDtable()
}

// decSymbol contains information about a state entry,
// Including the state offset base, the output symbol and
// the number of bits to read for the low part of the destination state.
// Using a composite uint64 is faster than a struct with separate members.
type decSymbol uint64

func newDecSymbol(nbits, addBits uint8, newState uint16, baseline uint32) decSymbol {
	return decSymbol(nbits) | (decSymbol(addBits) << 8) | (decSymbol(newState) << 16) | (decSymbol(baseline) << 32)
}

func (d decSymbol) nbBits() uint8 {
	return uint8(d)
}

func (d decSymbol) addBits() uint8 {
	return uint8(d >> 8)
}

func (d decSymbol) newState() uint16 {
	return uint16(d >> 16)
}

func (d decSymbol) baseline() uint32 {
	return uint32(d >> 32)
}

func (d decSymbol) baselineInt() int {
	return int(d >> 32)
}

func (d *decSymbol) set(nbits, addBits uint8, newState uint16, baseline uint32) {
	*d = decSymbol(nbits) | (decSymbol(addBits) << 8) | (decSymbol(newState) << 16) | (decSymbol(baseline) << 32)
}

func (d *decSymbol) setNBits(nBits uint8) {
	const mask = 0xffffffffffffff00
	*d = (*d & mask) | decSymbol(nBits)
}

func (d *decSymbol) setAddBits(addBits uint8) {
	const mask = 0xffffffffffff00ff
	*d = (*d & mask) | (decSymbol(addBits) << 8)
}

func (d *decSymbol) setNewState(state uint16) {
	const mask = 0xffffffff0000ffff
	*d = (*d & mask) | decSymbol(state)<<16
}

func (d *decSymbol) setBaseline(baseline uint32) {
	const mask = 0xffffffff
	*d = (*d & mask) | decSymbol(baseline)<<32
}

func (d *decSymbol) setExt(addBits uint8, baseline uint32) {
	const mask = 0xffff00ff
	*d = (*d & mask) | (decSymbol(addBits) << 8) | (decSymbol(baseline) << 32)
}

// decSymbolValue returns the transformed decSymbol for the given symbol.
func decSymbolValue(symb uint8, t []baseOffset) (decSymbol, error) {
	if int(symb) >= len(t) {
		return 0, fmt.Errorf("rle symbol %d >= max %d", symb, len(t))
	}
	lu := t[symb]
	return newDecSymbol(0, lu.addBits, 0, lu.baseLine), nil
}

// setRLE will set the decoder til RLE mode.
func (s *fseDecoder) setRLE(symbol decSymbol) {
	s.actualTableLog = 0
	s.maxBits = symbol.addBits()
	s.dt[0] = symbol
}

// buildDtable will build the decoding table.
func (s *fseDecoder) buildDtable() error {
	tableSize := uint32(1 << s.actualTableLog)
	highThreshold := tableSize - 1
	symbolNext := s.stateTable[:256]

	// Init, lay down lowprob symbols
	{
		for i, v := range s.norm[:s.symbolLen] {
			if v == -1 {
				s.dt[highThreshold].setAddBits(uint8(i))
				highThreshold--
				symbolNext[i] = 1
			} else {
				symbolNext[i] = uint16(v)
			}
		}
	}
	// Spread symbols
	{
		tableMask := tableSize - 1
		step := tableStep(tableSize)
		position := uint32(0)
		for ss, v := range s.norm[:s.symbolLen] {
			for i := 0; i < int(v); i++ {
				s.dt[position].setAddBits(uint8(ss))
				position = (position + step) & tableMask
				for position > highThreshold {
					// lowprob area
					position = (position + step) & tableMask
				}
			}
		}
		if position != 0 {
			// position must reach all cells once, otherwise normalizedCounter is incorrect
			return errors.New("corrupted input (position != 0)")
		}
	}

	// Build Decoding table
	{
		tableSize := uint16(1 << s.actualTableLog)
		for u, v := range s.dt[:tableSize] {
			symbol := v.addBits()
			nextState := symbolNext[symbol]
			symbolNext[symbol] = nextState + 1
			nBits := s.actualTableLog - byte(highBits(uint32(nextState)))
			s.dt[u&maxTableMask].setNBits(nBits)
			newState := (nextState << nBits) - tableSize
			if newState > tableSize {
				return fmt.Errorf("newState (%d) outside table size (%d)", newState, tableSize)
			}
			if newState == uint16(u) && nBits == 0 {
				// Seems weird that this is possible with nbits > 0.
				return fmt.Errorf("newState (%d) == oldState (%d) and no bits", newState, u)
			}
			s.dt[u&maxTableMask].setNewState(newState)
		}
	}
	return nil
}

// transform will transform the decoder table into a table usable for
// decoding without having to apply the transformation while decoding.
// The state will contain the base value and the number of bits to read.
func (s *fseDecoder) transform(t []baseOffset) error {
	tableSize := uint16(1 << s.actualTableLog)
	s.maxBits = 0
	for i, v := range s.dt[:tableSize] {
		add := v.addBits()
		if int(add) >= len(t) {
			return fmt.Errorf("invalid decoding table entry %d, symbol %d >= max (%d)", i, v.addBits(), len(t))
		}
		lu := t[add]
		if lu.addBits > s.maxBits {
			s.maxBits = lu.addBits
		}
		v.setExt(lu.addBits, lu.baseLine)
		s.dt[i] = v
	}
	return nil
}

type fseState struct {
	dt    []decSymbol
	state decSymbol
}

// Initialize and decodeAsync first state and symbol.
func (s *fseState) init(br *bitReader, tableLog uint8, dt []decSymbol) {
	s.dt = dt
	br.fill()
	s.state = dt[br.getBits(tableLog)]
}

// next returns the current symbol and sets the next state.
// At least tablelog bits must be available in the bit reader.
func (s *fseState) next(br *bitReader) {
	lowBits := uint16(br.getBits(s.state.nbBits()))
	s.state = s.dt[s.state.newState()+lowBits]
}

// finished returns true if all bits have been read from the bitstream
// and the next state would require reading bits from the input.
func (s *fseState) finished(br *bitReader) bool {
	return br.finished() && s.state.nbBits() > 0
}

// final returns the current state symbol without decoding the next.
func (s *fseState) final() (int, uint8) {
	return s.state.baselineInt(), s.state.addBits()
}

// final returns the current state symbol without decoding the next.
func (s decSymbol) final() (int, uint8) {
	return s.baselineInt(), s.addBits()
}

// nextFast returns the next symbol and sets the next state.
// This can only be used if no symbols are 0 bits.
// At least tablelog bits must be available in the bit reader.
func (s *fseState) nextFast(br *bitReader) (uint32, uint8) {
	lowBits := uint16(br.getBitsFast(s.state.nbBits()))
	s.state = s.dt[s.state.newState()+lowBits]
	return s.state.baseline(), s.state.addBits()
}