| 1 | package flate
|
|---|
| 2 |
|
|---|
| 3 | import "fmt"
|
|---|
| 4 |
|
|---|
| 5 | type fastEncL5 struct {
|
|---|
| 6 | fastGen
|
|---|
| 7 | table [tableSize]tableEntry
|
|---|
| 8 | bTable [tableSize]tableEntryPrev
|
|---|
| 9 | }
|
|---|
| 10 |
|
|---|
| 11 | func (e *fastEncL5) Encode(dst *tokens, src []byte) {
|
|---|
| 12 | const (
|
|---|
| 13 | inputMargin = 12 - 1
|
|---|
| 14 | minNonLiteralBlockSize = 1 + 1 + inputMargin
|
|---|
| 15 | )
|
|---|
| 16 | if debugDeflate && e.cur < 0 {
|
|---|
| 17 | panic(fmt.Sprint("e.cur < 0: ", e.cur))
|
|---|
| 18 | }
|
|---|
| 19 |
|
|---|
| 20 | // Protect against e.cur wraparound.
|
|---|
| 21 | for e.cur >= bufferReset {
|
|---|
| 22 | if len(e.hist) == 0 {
|
|---|
| 23 | for i := range e.table[:] {
|
|---|
| 24 | e.table[i] = tableEntry{}
|
|---|
| 25 | }
|
|---|
| 26 | for i := range e.bTable[:] {
|
|---|
| 27 | e.bTable[i] = tableEntryPrev{}
|
|---|
| 28 | }
|
|---|
| 29 | e.cur = maxMatchOffset
|
|---|
| 30 | break
|
|---|
| 31 | }
|
|---|
| 32 | // Shift down everything in the table that isn't already too far away.
|
|---|
| 33 | minOff := e.cur + int32(len(e.hist)) - maxMatchOffset
|
|---|
| 34 | for i := range e.table[:] {
|
|---|
| 35 | v := e.table[i].offset
|
|---|
| 36 | if v <= minOff {
|
|---|
| 37 | v = 0
|
|---|
| 38 | } else {
|
|---|
| 39 | v = v - e.cur + maxMatchOffset
|
|---|
| 40 | }
|
|---|
| 41 | e.table[i].offset = v
|
|---|
| 42 | }
|
|---|
| 43 | for i := range e.bTable[:] {
|
|---|
| 44 | v := e.bTable[i]
|
|---|
| 45 | if v.Cur.offset <= minOff {
|
|---|
| 46 | v.Cur.offset = 0
|
|---|
| 47 | v.Prev.offset = 0
|
|---|
| 48 | } else {
|
|---|
| 49 | v.Cur.offset = v.Cur.offset - e.cur + maxMatchOffset
|
|---|
| 50 | if v.Prev.offset <= minOff {
|
|---|
| 51 | v.Prev.offset = 0
|
|---|
| 52 | } else {
|
|---|
| 53 | v.Prev.offset = v.Prev.offset - e.cur + maxMatchOffset
|
|---|
| 54 | }
|
|---|
| 55 | }
|
|---|
| 56 | e.bTable[i] = v
|
|---|
| 57 | }
|
|---|
| 58 | e.cur = maxMatchOffset
|
|---|
| 59 | }
|
|---|
| 60 |
|
|---|
| 61 | s := e.addBlock(src)
|
|---|
| 62 |
|
|---|
| 63 | // This check isn't in the Snappy implementation, but there, the caller
|
|---|
| 64 | // instead of the callee handles this case.
|
|---|
| 65 | if len(src) < minNonLiteralBlockSize {
|
|---|
| 66 | // We do not fill the token table.
|
|---|
| 67 | // This will be picked up by caller.
|
|---|
| 68 | dst.n = uint16(len(src))
|
|---|
| 69 | return
|
|---|
| 70 | }
|
|---|
| 71 |
|
|---|
| 72 | // Override src
|
|---|
| 73 | src = e.hist
|
|---|
| 74 | nextEmit := s
|
|---|
| 75 |
|
|---|
| 76 | // sLimit is when to stop looking for offset/length copies. The inputMargin
|
|---|
| 77 | // lets us use a fast path for emitLiteral in the main loop, while we are
|
|---|
| 78 | // looking for copies.
|
|---|
| 79 | sLimit := int32(len(src) - inputMargin)
|
|---|
| 80 |
|
|---|
| 81 | // nextEmit is where in src the next emitLiteral should start from.
|
|---|
| 82 | cv := load6432(src, s)
|
|---|
| 83 | for {
|
|---|
| 84 | const skipLog = 6
|
|---|
| 85 | const doEvery = 1
|
|---|
| 86 |
|
|---|
| 87 | nextS := s
|
|---|
| 88 | var l int32
|
|---|
| 89 | var t int32
|
|---|
| 90 | for {
|
|---|
| 91 | nextHashS := hash4x64(cv, tableBits)
|
|---|
| 92 | nextHashL := hash7(cv, tableBits)
|
|---|
| 93 |
|
|---|
| 94 | s = nextS
|
|---|
| 95 | nextS = s + doEvery + (s-nextEmit)>>skipLog
|
|---|
| 96 | if nextS > sLimit {
|
|---|
| 97 | goto emitRemainder
|
|---|
| 98 | }
|
|---|
| 99 | // Fetch a short+long candidate
|
|---|
| 100 | sCandidate := e.table[nextHashS]
|
|---|
| 101 | lCandidate := e.bTable[nextHashL]
|
|---|
| 102 | next := load6432(src, nextS)
|
|---|
| 103 | entry := tableEntry{offset: s + e.cur}
|
|---|
| 104 | e.table[nextHashS] = entry
|
|---|
| 105 | eLong := &e.bTable[nextHashL]
|
|---|
| 106 | eLong.Cur, eLong.Prev = entry, eLong.Cur
|
|---|
| 107 |
|
|---|
| 108 | nextHashS = hash4x64(next, tableBits)
|
|---|
| 109 | nextHashL = hash7(next, tableBits)
|
|---|
| 110 |
|
|---|
| 111 | t = lCandidate.Cur.offset - e.cur
|
|---|
| 112 | if s-t < maxMatchOffset {
|
|---|
| 113 | if uint32(cv) == load3232(src, lCandidate.Cur.offset-e.cur) {
|
|---|
| 114 | // Store the next match
|
|---|
| 115 | e.table[nextHashS] = tableEntry{offset: nextS + e.cur}
|
|---|
| 116 | eLong := &e.bTable[nextHashL]
|
|---|
| 117 | eLong.Cur, eLong.Prev = tableEntry{offset: nextS + e.cur}, eLong.Cur
|
|---|
| 118 |
|
|---|
| 119 | t2 := lCandidate.Prev.offset - e.cur
|
|---|
| 120 | if s-t2 < maxMatchOffset && uint32(cv) == load3232(src, lCandidate.Prev.offset-e.cur) {
|
|---|
| 121 | l = e.matchlen(s+4, t+4, src) + 4
|
|---|
| 122 | ml1 := e.matchlen(s+4, t2+4, src) + 4
|
|---|
| 123 | if ml1 > l {
|
|---|
| 124 | t = t2
|
|---|
| 125 | l = ml1
|
|---|
| 126 | break
|
|---|
| 127 | }
|
|---|
| 128 | }
|
|---|
| 129 | break
|
|---|
| 130 | }
|
|---|
| 131 | t = lCandidate.Prev.offset - e.cur
|
|---|
| 132 | if s-t < maxMatchOffset && uint32(cv) == load3232(src, lCandidate.Prev.offset-e.cur) {
|
|---|
| 133 | // Store the next match
|
|---|
| 134 | e.table[nextHashS] = tableEntry{offset: nextS + e.cur}
|
|---|
| 135 | eLong := &e.bTable[nextHashL]
|
|---|
| 136 | eLong.Cur, eLong.Prev = tableEntry{offset: nextS + e.cur}, eLong.Cur
|
|---|
| 137 | break
|
|---|
| 138 | }
|
|---|
| 139 | }
|
|---|
| 140 |
|
|---|
| 141 | t = sCandidate.offset - e.cur
|
|---|
| 142 | if s-t < maxMatchOffset && uint32(cv) == load3232(src, sCandidate.offset-e.cur) {
|
|---|
| 143 | // Found a 4 match...
|
|---|
| 144 | l = e.matchlen(s+4, t+4, src) + 4
|
|---|
| 145 | lCandidate = e.bTable[nextHashL]
|
|---|
| 146 | // Store the next match
|
|---|
| 147 |
|
|---|
| 148 | e.table[nextHashS] = tableEntry{offset: nextS + e.cur}
|
|---|
| 149 | eLong := &e.bTable[nextHashL]
|
|---|
| 150 | eLong.Cur, eLong.Prev = tableEntry{offset: nextS + e.cur}, eLong.Cur
|
|---|
| 151 |
|
|---|
| 152 | // If the next long is a candidate, use that...
|
|---|
| 153 | t2 := lCandidate.Cur.offset - e.cur
|
|---|
| 154 | if nextS-t2 < maxMatchOffset {
|
|---|
| 155 | if load3232(src, lCandidate.Cur.offset-e.cur) == uint32(next) {
|
|---|
| 156 | ml := e.matchlen(nextS+4, t2+4, src) + 4
|
|---|
| 157 | if ml > l {
|
|---|
| 158 | t = t2
|
|---|
| 159 | s = nextS
|
|---|
| 160 | l = ml
|
|---|
| 161 | break
|
|---|
| 162 | }
|
|---|
| 163 | }
|
|---|
| 164 | // If the previous long is a candidate, use that...
|
|---|
| 165 | t2 = lCandidate.Prev.offset - e.cur
|
|---|
| 166 | if nextS-t2 < maxMatchOffset && load3232(src, lCandidate.Prev.offset-e.cur) == uint32(next) {
|
|---|
| 167 | ml := e.matchlen(nextS+4, t2+4, src) + 4
|
|---|
| 168 | if ml > l {
|
|---|
| 169 | t = t2
|
|---|
| 170 | s = nextS
|
|---|
| 171 | l = ml
|
|---|
| 172 | break
|
|---|
| 173 | }
|
|---|
| 174 | }
|
|---|
| 175 | }
|
|---|
| 176 | break
|
|---|
| 177 | }
|
|---|
| 178 | cv = next
|
|---|
| 179 | }
|
|---|
| 180 |
|
|---|
| 181 | // A 4-byte match has been found. We'll later see if more than 4 bytes
|
|---|
| 182 | // match. But, prior to the match, src[nextEmit:s] are unmatched. Emit
|
|---|
| 183 | // them as literal bytes.
|
|---|
| 184 |
|
|---|
| 185 | if l == 0 {
|
|---|
| 186 | // Extend the 4-byte match as long as possible.
|
|---|
| 187 | l = e.matchlenLong(s+4, t+4, src) + 4
|
|---|
| 188 | } else if l == maxMatchLength {
|
|---|
| 189 | l += e.matchlenLong(s+l, t+l, src)
|
|---|
| 190 | }
|
|---|
| 191 |
|
|---|
| 192 | // Try to locate a better match by checking the end of best match...
|
|---|
| 193 | if sAt := s + l; l < 30 && sAt < sLimit {
|
|---|
| 194 | eLong := e.bTable[hash7(load6432(src, sAt), tableBits)].Cur.offset
|
|---|
| 195 | // Test current
|
|---|
| 196 | t2 := eLong - e.cur - l
|
|---|
| 197 | off := s - t2
|
|---|
| 198 | if t2 >= 0 && off < maxMatchOffset && off > 0 {
|
|---|
| 199 | if l2 := e.matchlenLong(s, t2, src); l2 > l {
|
|---|
| 200 | t = t2
|
|---|
| 201 | l = l2
|
|---|
| 202 | }
|
|---|
| 203 | }
|
|---|
| 204 | }
|
|---|
| 205 |
|
|---|
| 206 | // Extend backwards
|
|---|
| 207 | for t > 0 && s > nextEmit && src[t-1] == src[s-1] {
|
|---|
| 208 | s--
|
|---|
| 209 | t--
|
|---|
| 210 | l++
|
|---|
| 211 | }
|
|---|
| 212 | if nextEmit < s {
|
|---|
| 213 | if false {
|
|---|
| 214 | emitLiteral(dst, src[nextEmit:s])
|
|---|
| 215 | } else {
|
|---|
| 216 | for _, v := range src[nextEmit:s] {
|
|---|
| 217 | dst.tokens[dst.n] = token(v)
|
|---|
| 218 | dst.litHist[v]++
|
|---|
| 219 | dst.n++
|
|---|
| 220 | }
|
|---|
| 221 | }
|
|---|
| 222 | }
|
|---|
| 223 | if debugDeflate {
|
|---|
| 224 | if t >= s {
|
|---|
| 225 | panic(fmt.Sprintln("s-t", s, t))
|
|---|
| 226 | }
|
|---|
| 227 | if (s - t) > maxMatchOffset {
|
|---|
| 228 | panic(fmt.Sprintln("mmo", s-t))
|
|---|
| 229 | }
|
|---|
| 230 | if l < baseMatchLength {
|
|---|
| 231 | panic("bml")
|
|---|
| 232 | }
|
|---|
| 233 | }
|
|---|
| 234 |
|
|---|
| 235 | dst.AddMatchLong(l, uint32(s-t-baseMatchOffset))
|
|---|
| 236 | s += l
|
|---|
| 237 | nextEmit = s
|
|---|
| 238 | if nextS >= s {
|
|---|
| 239 | s = nextS + 1
|
|---|
| 240 | }
|
|---|
| 241 |
|
|---|
| 242 | if s >= sLimit {
|
|---|
| 243 | goto emitRemainder
|
|---|
| 244 | }
|
|---|
| 245 |
|
|---|
| 246 | // Store every 3rd hash in-between.
|
|---|
| 247 | if true {
|
|---|
| 248 | const hashEvery = 3
|
|---|
| 249 | i := s - l + 1
|
|---|
| 250 | if i < s-1 {
|
|---|
| 251 | cv := load6432(src, i)
|
|---|
| 252 | t := tableEntry{offset: i + e.cur}
|
|---|
| 253 | e.table[hash4x64(cv, tableBits)] = t
|
|---|
| 254 | eLong := &e.bTable[hash7(cv, tableBits)]
|
|---|
| 255 | eLong.Cur, eLong.Prev = t, eLong.Cur
|
|---|
| 256 |
|
|---|
| 257 | // Do an long at i+1
|
|---|
| 258 | cv >>= 8
|
|---|
| 259 | t = tableEntry{offset: t.offset + 1}
|
|---|
| 260 | eLong = &e.bTable[hash7(cv, tableBits)]
|
|---|
| 261 | eLong.Cur, eLong.Prev = t, eLong.Cur
|
|---|
| 262 |
|
|---|
| 263 | // We only have enough bits for a short entry at i+2
|
|---|
| 264 | cv >>= 8
|
|---|
| 265 | t = tableEntry{offset: t.offset + 1}
|
|---|
| 266 | e.table[hash4x64(cv, tableBits)] = t
|
|---|
| 267 |
|
|---|
| 268 | // Skip one - otherwise we risk hitting 's'
|
|---|
| 269 | i += 4
|
|---|
| 270 | for ; i < s-1; i += hashEvery {
|
|---|
| 271 | cv := load6432(src, i)
|
|---|
| 272 | t := tableEntry{offset: i + e.cur}
|
|---|
| 273 | t2 := tableEntry{offset: t.offset + 1}
|
|---|
| 274 | eLong := &e.bTable[hash7(cv, tableBits)]
|
|---|
| 275 | eLong.Cur, eLong.Prev = t, eLong.Cur
|
|---|
| 276 | e.table[hash4u(uint32(cv>>8), tableBits)] = t2
|
|---|
| 277 | }
|
|---|
| 278 | }
|
|---|
| 279 | }
|
|---|
| 280 |
|
|---|
| 281 | // We could immediately start working at s now, but to improve
|
|---|
| 282 | // compression we first update the hash table at s-1 and at s.
|
|---|
| 283 | x := load6432(src, s-1)
|
|---|
| 284 | o := e.cur + s - 1
|
|---|
| 285 | prevHashS := hash4x64(x, tableBits)
|
|---|
| 286 | prevHashL := hash7(x, tableBits)
|
|---|
| 287 | e.table[prevHashS] = tableEntry{offset: o}
|
|---|
| 288 | eLong := &e.bTable[prevHashL]
|
|---|
| 289 | eLong.Cur, eLong.Prev = tableEntry{offset: o}, eLong.Cur
|
|---|
| 290 | cv = x >> 8
|
|---|
| 291 | }
|
|---|
| 292 |
|
|---|
| 293 | emitRemainder:
|
|---|
| 294 | if int(nextEmit) < len(src) {
|
|---|
| 295 | // If nothing was added, don't encode literals.
|
|---|
| 296 | if dst.n == 0 {
|
|---|
| 297 | return
|
|---|
| 298 | }
|
|---|
| 299 |
|
|---|
| 300 | emitLiteral(dst, src[nextEmit:])
|
|---|
| 301 | }
|
|---|
| 302 | }
|
|---|
