1 |
747 |
jeremybenn |
// Copyright 2009 The Go Authors. All rights reserved.
|
2 |
|
|
// Use of this source code is governed by a BSD-style
|
3 |
|
|
// license that can be found in the LICENSE file.
|
4 |
|
|
|
5 |
|
|
package flate
|
6 |
|
|
|
7 |
|
|
import (
|
8 |
|
|
"math"
|
9 |
|
|
"sort"
|
10 |
|
|
)
|
11 |
|
|
|
12 |
|
|
type huffmanEncoder struct {
|
13 |
|
|
codeBits []uint8
|
14 |
|
|
code []uint16
|
15 |
|
|
}
|
16 |
|
|
|
17 |
|
|
type literalNode struct {
|
18 |
|
|
literal uint16
|
19 |
|
|
freq int32
|
20 |
|
|
}
|
21 |
|
|
|
22 |
|
|
type chain struct {
|
23 |
|
|
// The sum of the leaves in this tree
|
24 |
|
|
freq int32
|
25 |
|
|
|
26 |
|
|
// The number of literals to the left of this item at this level
|
27 |
|
|
leafCount int32
|
28 |
|
|
|
29 |
|
|
// The right child of this chain in the previous level.
|
30 |
|
|
up *chain
|
31 |
|
|
}
|
32 |
|
|
|
33 |
|
|
type levelInfo struct {
|
34 |
|
|
// Our level. for better printing
|
35 |
|
|
level int32
|
36 |
|
|
|
37 |
|
|
// The most recent chain generated for this level
|
38 |
|
|
lastChain *chain
|
39 |
|
|
|
40 |
|
|
// The frequency of the next character to add to this level
|
41 |
|
|
nextCharFreq int32
|
42 |
|
|
|
43 |
|
|
// The frequency of the next pair (from level below) to add to this level.
|
44 |
|
|
// Only valid if the "needed" value of the next lower level is 0.
|
45 |
|
|
nextPairFreq int32
|
46 |
|
|
|
47 |
|
|
// The number of chains remaining to generate for this level before moving
|
48 |
|
|
// up to the next level
|
49 |
|
|
needed int32
|
50 |
|
|
|
51 |
|
|
// The levelInfo for level+1
|
52 |
|
|
up *levelInfo
|
53 |
|
|
|
54 |
|
|
// The levelInfo for level-1
|
55 |
|
|
down *levelInfo
|
56 |
|
|
}
|
57 |
|
|
|
58 |
|
|
func maxNode() literalNode { return literalNode{math.MaxUint16, math.MaxInt32} }
|
59 |
|
|
|
60 |
|
|
func newHuffmanEncoder(size int) *huffmanEncoder {
|
61 |
|
|
return &huffmanEncoder{make([]uint8, size), make([]uint16, size)}
|
62 |
|
|
}
|
63 |
|
|
|
64 |
|
|
// Generates a HuffmanCode corresponding to the fixed literal table
|
65 |
|
|
func generateFixedLiteralEncoding() *huffmanEncoder {
|
66 |
|
|
h := newHuffmanEncoder(maxLit)
|
67 |
|
|
codeBits := h.codeBits
|
68 |
|
|
code := h.code
|
69 |
|
|
var ch uint16
|
70 |
|
|
for ch = 0; ch < maxLit; ch++ {
|
71 |
|
|
var bits uint16
|
72 |
|
|
var size uint8
|
73 |
|
|
switch {
|
74 |
|
|
case ch < 144:
|
75 |
|
|
// size 8, 000110000 .. 10111111
|
76 |
|
|
bits = ch + 48
|
77 |
|
|
size = 8
|
78 |
|
|
break
|
79 |
|
|
case ch < 256:
|
80 |
|
|
// size 9, 110010000 .. 111111111
|
81 |
|
|
bits = ch + 400 - 144
|
82 |
|
|
size = 9
|
83 |
|
|
break
|
84 |
|
|
case ch < 280:
|
85 |
|
|
// size 7, 0000000 .. 0010111
|
86 |
|
|
bits = ch - 256
|
87 |
|
|
size = 7
|
88 |
|
|
break
|
89 |
|
|
default:
|
90 |
|
|
// size 8, 11000000 .. 11000111
|
91 |
|
|
bits = ch + 192 - 280
|
92 |
|
|
size = 8
|
93 |
|
|
}
|
94 |
|
|
codeBits[ch] = size
|
95 |
|
|
code[ch] = reverseBits(bits, size)
|
96 |
|
|
}
|
97 |
|
|
return h
|
98 |
|
|
}
|
99 |
|
|
|
100 |
|
|
func generateFixedOffsetEncoding() *huffmanEncoder {
|
101 |
|
|
h := newHuffmanEncoder(30)
|
102 |
|
|
codeBits := h.codeBits
|
103 |
|
|
code := h.code
|
104 |
|
|
for ch := uint16(0); ch < 30; ch++ {
|
105 |
|
|
codeBits[ch] = 5
|
106 |
|
|
code[ch] = reverseBits(ch, 5)
|
107 |
|
|
}
|
108 |
|
|
return h
|
109 |
|
|
}
|
110 |
|
|
|
111 |
|
|
var fixedLiteralEncoding *huffmanEncoder = generateFixedLiteralEncoding()
|
112 |
|
|
var fixedOffsetEncoding *huffmanEncoder = generateFixedOffsetEncoding()
|
113 |
|
|
|
114 |
|
|
func (h *huffmanEncoder) bitLength(freq []int32) int64 {
|
115 |
|
|
var total int64
|
116 |
|
|
for i, f := range freq {
|
117 |
|
|
if f != 0 {
|
118 |
|
|
total += int64(f) * int64(h.codeBits[i])
|
119 |
|
|
}
|
120 |
|
|
}
|
121 |
|
|
return total
|
122 |
|
|
}
|
123 |
|
|
|
124 |
|
|
// Return the number of literals assigned to each bit size in the Huffman encoding
|
125 |
|
|
//
|
126 |
|
|
// This method is only called when list.length >= 3
|
127 |
|
|
// The cases of 0, 1, and 2 literals are handled by special case code.
|
128 |
|
|
//
|
129 |
|
|
// list An array of the literals with non-zero frequencies
|
130 |
|
|
// and their associated frequencies. The array is in order of increasing
|
131 |
|
|
// frequency, and has as its last element a special element with frequency
|
132 |
|
|
// MaxInt32
|
133 |
|
|
// maxBits The maximum number of bits that should be used to encode any literal.
|
134 |
|
|
// return An integer array in which array[i] indicates the number of literals
|
135 |
|
|
// that should be encoded in i bits.
|
136 |
|
|
func (h *huffmanEncoder) bitCounts(list []literalNode, maxBits int32) []int32 {
|
137 |
|
|
n := int32(len(list))
|
138 |
|
|
list = list[0 : n+1]
|
139 |
|
|
list[n] = maxNode()
|
140 |
|
|
|
141 |
|
|
// The tree can't have greater depth than n - 1, no matter what. This
|
142 |
|
|
// saves a little bit of work in some small cases
|
143 |
|
|
if maxBits > n-1 {
|
144 |
|
|
maxBits = n - 1
|
145 |
|
|
}
|
146 |
|
|
|
147 |
|
|
// Create information about each of the levels.
|
148 |
|
|
// A bogus "Level 0" whose sole purpose is so that
|
149 |
|
|
// level1.prev.needed==0. This makes level1.nextPairFreq
|
150 |
|
|
// be a legitimate value that never gets chosen.
|
151 |
|
|
top := &levelInfo{needed: 0}
|
152 |
|
|
chain2 := &chain{list[1].freq, 2, new(chain)}
|
153 |
|
|
for level := int32(1); level <= maxBits; level++ {
|
154 |
|
|
// For every level, the first two items are the first two characters.
|
155 |
|
|
// We initialize the levels as if we had already figured this out.
|
156 |
|
|
top = &levelInfo{
|
157 |
|
|
level: level,
|
158 |
|
|
lastChain: chain2,
|
159 |
|
|
nextCharFreq: list[2].freq,
|
160 |
|
|
nextPairFreq: list[0].freq + list[1].freq,
|
161 |
|
|
down: top,
|
162 |
|
|
}
|
163 |
|
|
top.down.up = top
|
164 |
|
|
if level == 1 {
|
165 |
|
|
top.nextPairFreq = math.MaxInt32
|
166 |
|
|
}
|
167 |
|
|
}
|
168 |
|
|
|
169 |
|
|
// We need a total of 2*n - 2 items at top level and have already generated 2.
|
170 |
|
|
top.needed = 2*n - 4
|
171 |
|
|
|
172 |
|
|
l := top
|
173 |
|
|
for {
|
174 |
|
|
if l.nextPairFreq == math.MaxInt32 && l.nextCharFreq == math.MaxInt32 {
|
175 |
|
|
// We've run out of both leafs and pairs.
|
176 |
|
|
// End all calculations for this level.
|
177 |
|
|
// To m sure we never come back to this level or any lower level,
|
178 |
|
|
// set nextPairFreq impossibly large.
|
179 |
|
|
l.lastChain = nil
|
180 |
|
|
l.needed = 0
|
181 |
|
|
l = l.up
|
182 |
|
|
l.nextPairFreq = math.MaxInt32
|
183 |
|
|
continue
|
184 |
|
|
}
|
185 |
|
|
|
186 |
|
|
prevFreq := l.lastChain.freq
|
187 |
|
|
if l.nextCharFreq < l.nextPairFreq {
|
188 |
|
|
// The next item on this row is a leaf node.
|
189 |
|
|
n := l.lastChain.leafCount + 1
|
190 |
|
|
l.lastChain = &chain{l.nextCharFreq, n, l.lastChain.up}
|
191 |
|
|
l.nextCharFreq = list[n].freq
|
192 |
|
|
} else {
|
193 |
|
|
// The next item on this row is a pair from the previous row.
|
194 |
|
|
// nextPairFreq isn't valid until we generate two
|
195 |
|
|
// more values in the level below
|
196 |
|
|
l.lastChain = &chain{l.nextPairFreq, l.lastChain.leafCount, l.down.lastChain}
|
197 |
|
|
l.down.needed = 2
|
198 |
|
|
}
|
199 |
|
|
|
200 |
|
|
if l.needed--; l.needed == 0 {
|
201 |
|
|
// We've done everything we need to do for this level.
|
202 |
|
|
// Continue calculating one level up. Fill in nextPairFreq
|
203 |
|
|
// of that level with the sum of the two nodes we've just calculated on
|
204 |
|
|
// this level.
|
205 |
|
|
up := l.up
|
206 |
|
|
if up == nil {
|
207 |
|
|
// All done!
|
208 |
|
|
break
|
209 |
|
|
}
|
210 |
|
|
up.nextPairFreq = prevFreq + l.lastChain.freq
|
211 |
|
|
l = up
|
212 |
|
|
} else {
|
213 |
|
|
// If we stole from below, move down temporarily to replenish it.
|
214 |
|
|
for l.down.needed > 0 {
|
215 |
|
|
l = l.down
|
216 |
|
|
}
|
217 |
|
|
}
|
218 |
|
|
}
|
219 |
|
|
|
220 |
|
|
// Somethings is wrong if at the end, the top level is null or hasn't used
|
221 |
|
|
// all of the leaves.
|
222 |
|
|
if top.lastChain.leafCount != n {
|
223 |
|
|
panic("top.lastChain.leafCount != n")
|
224 |
|
|
}
|
225 |
|
|
|
226 |
|
|
bitCount := make([]int32, maxBits+1)
|
227 |
|
|
bits := 1
|
228 |
|
|
for chain := top.lastChain; chain.up != nil; chain = chain.up {
|
229 |
|
|
// chain.leafCount gives the number of literals requiring at least "bits"
|
230 |
|
|
// bits to encode.
|
231 |
|
|
bitCount[bits] = chain.leafCount - chain.up.leafCount
|
232 |
|
|
bits++
|
233 |
|
|
}
|
234 |
|
|
return bitCount
|
235 |
|
|
}
|
236 |
|
|
|
237 |
|
|
// Look at the leaves and assign them a bit count and an encoding as specified
|
238 |
|
|
// in RFC 1951 3.2.2
|
239 |
|
|
func (h *huffmanEncoder) assignEncodingAndSize(bitCount []int32, list []literalNode) {
|
240 |
|
|
code := uint16(0)
|
241 |
|
|
for n, bits := range bitCount {
|
242 |
|
|
code <<= 1
|
243 |
|
|
if n == 0 || bits == 0 {
|
244 |
|
|
continue
|
245 |
|
|
}
|
246 |
|
|
// The literals list[len(list)-bits] .. list[len(list)-bits]
|
247 |
|
|
// are encoded using "bits" bits, and get the values
|
248 |
|
|
// code, code + 1, .... The code values are
|
249 |
|
|
// assigned in literal order (not frequency order).
|
250 |
|
|
chunk := list[len(list)-int(bits):]
|
251 |
|
|
sortByLiteral(chunk)
|
252 |
|
|
for _, node := range chunk {
|
253 |
|
|
h.codeBits[node.literal] = uint8(n)
|
254 |
|
|
h.code[node.literal] = reverseBits(code, uint8(n))
|
255 |
|
|
code++
|
256 |
|
|
}
|
257 |
|
|
list = list[0 : len(list)-int(bits)]
|
258 |
|
|
}
|
259 |
|
|
}
|
260 |
|
|
|
261 |
|
|
// Update this Huffman Code object to be the minimum code for the specified frequency count.
|
262 |
|
|
//
|
263 |
|
|
// freq An array of frequencies, in which frequency[i] gives the frequency of literal i.
|
264 |
|
|
// maxBits The maximum number of bits to use for any literal.
|
265 |
|
|
func (h *huffmanEncoder) generate(freq []int32, maxBits int32) {
|
266 |
|
|
list := make([]literalNode, len(freq)+1)
|
267 |
|
|
// Number of non-zero literals
|
268 |
|
|
count := 0
|
269 |
|
|
// Set list to be the set of all non-zero literals and their frequencies
|
270 |
|
|
for i, f := range freq {
|
271 |
|
|
if f != 0 {
|
272 |
|
|
list[count] = literalNode{uint16(i), f}
|
273 |
|
|
count++
|
274 |
|
|
} else {
|
275 |
|
|
h.codeBits[i] = 0
|
276 |
|
|
}
|
277 |
|
|
}
|
278 |
|
|
// If freq[] is shorter than codeBits[], fill rest of codeBits[] with zeros
|
279 |
|
|
h.codeBits = h.codeBits[0:len(freq)]
|
280 |
|
|
list = list[0:count]
|
281 |
|
|
if count <= 2 {
|
282 |
|
|
// Handle the small cases here, because they are awkward for the general case code. With
|
283 |
|
|
// two or fewer literals, everything has bit length 1.
|
284 |
|
|
for i, node := range list {
|
285 |
|
|
// "list" is in order of increasing literal value.
|
286 |
|
|
h.codeBits[node.literal] = 1
|
287 |
|
|
h.code[node.literal] = uint16(i)
|
288 |
|
|
}
|
289 |
|
|
return
|
290 |
|
|
}
|
291 |
|
|
sortByFreq(list)
|
292 |
|
|
|
293 |
|
|
// Get the number of literals for each bit count
|
294 |
|
|
bitCount := h.bitCounts(list, maxBits)
|
295 |
|
|
// And do the assignment
|
296 |
|
|
h.assignEncodingAndSize(bitCount, list)
|
297 |
|
|
}
|
298 |
|
|
|
299 |
|
|
type literalNodeSorter struct {
|
300 |
|
|
a []literalNode
|
301 |
|
|
less func(i, j int) bool
|
302 |
|
|
}
|
303 |
|
|
|
304 |
|
|
func (s literalNodeSorter) Len() int { return len(s.a) }
|
305 |
|
|
|
306 |
|
|
func (s literalNodeSorter) Less(i, j int) bool {
|
307 |
|
|
return s.less(i, j)
|
308 |
|
|
}
|
309 |
|
|
|
310 |
|
|
func (s literalNodeSorter) Swap(i, j int) { s.a[i], s.a[j] = s.a[j], s.a[i] }
|
311 |
|
|
|
312 |
|
|
func sortByFreq(a []literalNode) {
|
313 |
|
|
s := &literalNodeSorter{a, func(i, j int) bool {
|
314 |
|
|
if a[i].freq == a[j].freq {
|
315 |
|
|
return a[i].literal < a[j].literal
|
316 |
|
|
}
|
317 |
|
|
return a[i].freq < a[j].freq
|
318 |
|
|
}}
|
319 |
|
|
sort.Sort(s)
|
320 |
|
|
}
|
321 |
|
|
|
322 |
|
|
func sortByLiteral(a []literalNode) {
|
323 |
|
|
s := &literalNodeSorter{a, func(i, j int) bool { return a[i].literal < a[j].literal }}
|
324 |
|
|
sort.Sort(s)
|
325 |
|
|
}
|