src.org.tukaani.xz.lzma.LZMAEncoder Maven / Gradle / Ivy
/*
* LZMAEncoder
*
* Authors: Lasse Collin
* Igor Pavlov LZMA2_UNCOMPRESSED_LIMIT.
*
* A compressed LZMA2 chunk can hold 2 MiB of uncompressed data.
* A single LZMA symbol may indicate up to MATCH_LEN_MAX bytes
* of data, so the LZMA2 chunk is considered full when there is
* less space than MATCH_LEN_MAX bytes.
*/
private static final int LZMA2_UNCOMPRESSED_LIMIT
= (2 << 20) - MATCH_LEN_MAX;
/**
* LZMA2 chunk is considered full when its compressed size exceeds
* LZMA2_COMPRESSED_LIMIT.
*
* The maximum compressed size of a LZMA2 chunk is 64 KiB.
* A single LZMA symbol might use 20 bytes of space even though
* it usually takes just one byte or so. Two more bytes are needed
* for LZMA2 uncompressed chunks (see LZMA2OutputStream.writeChunk).
* Leave a little safety margin and use 26 bytes.
*/
private static final int LZMA2_COMPRESSED_LIMIT = (64 << 10) - 26;
private static final int DIST_PRICE_UPDATE_INTERVAL = FULL_DISTANCES;
private static final int ALIGN_PRICE_UPDATE_INTERVAL = ALIGN_SIZE;
private final RangeEncoder rc;
final LZEncoder lz;
final LiteralEncoder literalEncoder;
final LengthEncoder matchLenEncoder;
final LengthEncoder repLenEncoder;
final int niceLen;
private int distPriceCount = 0;
private int alignPriceCount = 0;
private final int distSlotPricesSize;
private final int[][] distSlotPrices;
private final int[][] fullDistPrices
= new int[DIST_STATES][FULL_DISTANCES];
private final int[] alignPrices = new int[ALIGN_SIZE];
int back = 0;
int readAhead = -1;
private int uncompressedSize = 0;
public static int getMemoryUsage(int mode, int dictSize,
int extraSizeBefore, int mf) {
int m = 80;
switch (mode) {
case MODE_FAST:
m += LZMAEncoderFast.getMemoryUsage(
dictSize, extraSizeBefore, mf);
break;
case MODE_NORMAL:
m += LZMAEncoderNormal.getMemoryUsage(
dictSize, extraSizeBefore, mf);
break;
default:
throw new IllegalArgumentException();
}
return m;
}
public static LZMAEncoder getInstance(
RangeEncoder rc, int lc, int lp, int pb, int mode,
int dictSize, int extraSizeBefore,
int niceLen, int mf, int depthLimit) {
switch (mode) {
case MODE_FAST:
return new LZMAEncoderFast(rc, lc, lp, pb,
dictSize, extraSizeBefore,
niceLen, mf, depthLimit);
case MODE_NORMAL:
return new LZMAEncoderNormal(rc, lc, lp, pb,
dictSize, extraSizeBefore,
niceLen, mf, depthLimit);
}
throw new IllegalArgumentException();
}
/**
* Gets an integer [0, 63] matching the highest two bits of an integer.
* This is like bit scan reverse (BSR) on x86 except that this also
* cares about the second highest bit.
*/
public static int getDistSlot(int dist) {
if (dist <= DIST_MODEL_START)
return dist;
int n = dist;
int i = 31;
if ((n & 0xFFFF0000) == 0) {
n <<= 16;
i = 15;
}
if ((n & 0xFF000000) == 0) {
n <<= 8;
i -= 8;
}
if ((n & 0xF0000000) == 0) {
n <<= 4;
i -= 4;
}
if ((n & 0xC0000000) == 0) {
n <<= 2;
i -= 2;
}
if ((n & 0x80000000) == 0)
--i;
return (i << 1) + ((dist >>> (i - 1)) & 1);
}
/**
* Gets the next LZMA symbol.
*
* There are three types of symbols: literal (a single byte),
* repeated match, and normal match. The symbol is indicated
* by the return value and by the variable back.
*
* Literal: back == -1 and return value is 1.
* The literal itself needs to be read from lz separately.
*
* Repeated match: back is in the range [0, 3] and
* the return value is the length of the repeated match.
*
* Normal match: back - REPS (back - 4)
* is the distance of the match and the return value is the length
* of the match.
*/
abstract int getNextSymbol();
LZMAEncoder(RangeEncoder rc, LZEncoder lz,
int lc, int lp, int pb, int dictSize, int niceLen) {
super(pb);
this.rc = rc;
this.lz = lz;
this.niceLen = niceLen;
literalEncoder = new LiteralEncoder(lc, lp);
matchLenEncoder = new LengthEncoder(pb, niceLen);
repLenEncoder = new LengthEncoder(pb, niceLen);
distSlotPricesSize = getDistSlot(dictSize - 1) + 1;
distSlotPrices = new int[DIST_STATES][distSlotPricesSize];
reset();
}
public LZEncoder getLZEncoder() {
return lz;
}
public void reset() {
super.reset();
literalEncoder.reset();
matchLenEncoder.reset();
repLenEncoder.reset();
distPriceCount = 0;
alignPriceCount = 0;
uncompressedSize += readAhead + 1;
readAhead = -1;
}
public int getUncompressedSize() {
return uncompressedSize;
}
public void resetUncompressedSize() {
uncompressedSize = 0;
}
/**
* Compresses for LZMA2.
*
* @return true if the LZMA2 chunk became full, false otherwise
*/
public boolean encodeForLZMA2() {
if (!lz.isStarted() && !encodeInit())
return false;
while (uncompressedSize <= LZMA2_UNCOMPRESSED_LIMIT
&& rc.getPendingSize() <= LZMA2_COMPRESSED_LIMIT)
if (!encodeSymbol())
return false;
return true;
}
private boolean encodeInit() {
assert readAhead == -1;
if (!lz.hasEnoughData(0))
return false;
// The first symbol must be a literal unless using
// a preset dictionary. This code isn't run if using
// a preset dictionary.
skip(1);
rc.encodeBit(isMatch[state.get()], 0, 0);
literalEncoder.encodeInit();
--readAhead;
assert readAhead == -1;
++uncompressedSize;
assert uncompressedSize == 1;
return true;
}
private boolean encodeSymbol() {
if (!lz.hasEnoughData(readAhead + 1))
return false;
int len = getNextSymbol();
assert readAhead >= 0;
int posState = (lz.getPos() - readAhead) & posMask;
if (back == -1) {
// Literal i.e. eight-bit byte
assert len == 1;
rc.encodeBit(isMatch[state.get()], posState, 0);
literalEncoder.encode();
} else {
// Some type of match
rc.encodeBit(isMatch[state.get()], posState, 1);
if (back < REPS) {
// Repeated match i.e. the same distance
// has been used earlier.
assert lz.getMatchLen(-readAhead, reps[back], len) == len;
rc.encodeBit(isRep, state.get(), 1);
encodeRepMatch(back, len, posState);
} else {
// Normal match
assert lz.getMatchLen(-readAhead, back - REPS, len) == len;
rc.encodeBit(isRep, state.get(), 0);
encodeMatch(back - REPS, len, posState);
}
}
readAhead -= len;
uncompressedSize += len;
return true;
}
private void encodeMatch(int dist, int len, int posState) {
state.updateMatch();
matchLenEncoder.encode(len, posState);
int distSlot = getDistSlot(dist);
rc.encodeBitTree(distSlots[getDistState(len)], distSlot);
if (distSlot >= DIST_MODEL_START) {
int footerBits = (distSlot >>> 1) - 1;
int base = (2 | (distSlot & 1)) << footerBits;
int distReduced = dist - base;
if (distSlot < DIST_MODEL_END) {
rc.encodeReverseBitTree(
distSpecial[distSlot - DIST_MODEL_START],
distReduced);
} else {
rc.encodeDirectBits(distReduced >>> ALIGN_BITS,
footerBits - ALIGN_BITS);
rc.encodeReverseBitTree(distAlign, distReduced & ALIGN_MASK);
--alignPriceCount;
}
}
reps[3] = reps[2];
reps[2] = reps[1];
reps[1] = reps[0];
reps[0] = dist;
--distPriceCount;
}
private void encodeRepMatch(int rep, int len, int posState) {
if (rep == 0) {
rc.encodeBit(isRep0, state.get(), 0);
rc.encodeBit(isRep0Long[state.get()], posState, len == 1 ? 0 : 1);
} else {
int dist = reps[rep];
rc.encodeBit(isRep0, state.get(), 1);
if (rep == 1) {
rc.encodeBit(isRep1, state.get(), 0);
} else {
rc.encodeBit(isRep1, state.get(), 1);
rc.encodeBit(isRep2, state.get(), rep - 2);
if (rep == 3)
reps[3] = reps[2];
reps[2] = reps[1];
}
reps[1] = reps[0];
reps[0] = dist;
}
if (len == 1) {
state.updateShortRep();
} else {
repLenEncoder.encode(len, posState);
state.updateLongRep();
}
}
Matches getMatches() {
++readAhead;
Matches matches = lz.getMatches();
assert lz.verifyMatches(matches);
return matches;
}
void skip(int len) {
readAhead += len;
lz.skip(len);
}
int getAnyMatchPrice(State state, int posState) {
return RangeEncoder.getBitPrice(isMatch[state.get()][posState], 1);
}
int getNormalMatchPrice(int anyMatchPrice, State state) {
return anyMatchPrice
+ RangeEncoder.getBitPrice(isRep[state.get()], 0);
}
int getAnyRepPrice(int anyMatchPrice, State state) {
return anyMatchPrice
+ RangeEncoder.getBitPrice(isRep[state.get()], 1);
}
int getShortRepPrice(int anyRepPrice, State state, int posState) {
return anyRepPrice
+ RangeEncoder.getBitPrice(isRep0[state.get()], 0)
+ RangeEncoder.getBitPrice(isRep0Long[state.get()][posState],
0);
}
int getLongRepPrice(int anyRepPrice, int rep, State state, int posState) {
int price = anyRepPrice;
if (rep == 0) {
price += RangeEncoder.getBitPrice(isRep0[state.get()], 0)
+ RangeEncoder.getBitPrice(
isRep0Long[state.get()][posState], 1);
} else {
price += RangeEncoder.getBitPrice(isRep0[state.get()], 1);
if (rep == 1)
price += RangeEncoder.getBitPrice(isRep1[state.get()], 0);
else
price += RangeEncoder.getBitPrice(isRep1[state.get()], 1)
+ RangeEncoder.getBitPrice(isRep2[state.get()],
rep - 2);
}
return price;
}
int getLongRepAndLenPrice(int rep, int len, State state, int posState) {
int anyMatchPrice = getAnyMatchPrice(state, posState);
int anyRepPrice = getAnyRepPrice(anyMatchPrice, state);
int longRepPrice = getLongRepPrice(anyRepPrice, rep, state, posState);
return longRepPrice + repLenEncoder.getPrice(len, posState);
}
int getMatchAndLenPrice(int normalMatchPrice,
int dist, int len, int posState) {
int price = normalMatchPrice
+ matchLenEncoder.getPrice(len, posState);
int distState = getDistState(len);
if (dist < FULL_DISTANCES) {
price += fullDistPrices[distState][dist];
} else {
// Note that distSlotPrices includes also
// the price of direct bits.
int distSlot = getDistSlot(dist);
price += distSlotPrices[distState][distSlot]
+ alignPrices[dist & ALIGN_MASK];
}
return price;
}
private void updateDistPrices() {
distPriceCount = DIST_PRICE_UPDATE_INTERVAL;
for (int distState = 0; distState < DIST_STATES; ++distState) {
for (int distSlot = 0; distSlot < distSlotPricesSize; ++distSlot)
distSlotPrices[distState][distSlot]
= RangeEncoder.getBitTreePrice(
distSlots[distState], distSlot);
for (int distSlot = DIST_MODEL_END; distSlot < distSlotPricesSize;
++distSlot) {
int count = (distSlot >>> 1) - 1 - ALIGN_BITS;
distSlotPrices[distState][distSlot]
+= RangeEncoder.getDirectBitsPrice(count);
}
for (int dist = 0; dist < DIST_MODEL_START; ++dist)
fullDistPrices[distState][dist]
= distSlotPrices[distState][dist];
}
int dist = DIST_MODEL_START;
for (int distSlot = DIST_MODEL_START; distSlot < DIST_MODEL_END;
++distSlot) {
int footerBits = (distSlot >>> 1) - 1;
int base = (2 | (distSlot & 1)) << footerBits;
int limit = distSpecial[distSlot - DIST_MODEL_START].length;
for (int i = 0; i < limit; ++i) {
int distReduced = dist - base;
int price = RangeEncoder.getReverseBitTreePrice(
distSpecial[distSlot - DIST_MODEL_START],
distReduced);
for (int distState = 0; distState < DIST_STATES; ++distState)
fullDistPrices[distState][dist]
= distSlotPrices[distState][distSlot] + price;
++dist;
}
}
assert dist == FULL_DISTANCES;
}
private void updateAlignPrices() {
alignPriceCount = ALIGN_PRICE_UPDATE_INTERVAL;
for (int i = 0; i < ALIGN_SIZE; ++i)
alignPrices[i] = RangeEncoder.getReverseBitTreePrice(distAlign,
i);
}
/**
* Updates the lookup tables used for calculating match distance
* and length prices. The updating is skipped for performance reasons
* if the tables haven't changed much since the previous update.
*/
void updatePrices() {
if (distPriceCount <= 0)
updateDistPrices();
if (alignPriceCount <= 0)
updateAlignPrices();
matchLenEncoder.updatePrices();
repLenEncoder.updatePrices();
}
class LiteralEncoder extends LiteralCoder {
LiteralSubencoder[] subencoders;
LiteralEncoder(int lc, int lp) {
super(lc, lp);
subencoders = new LiteralSubencoder[1 << (lc + lp)];
for (int i = 0; i < subencoders.length; ++i)
subencoders[i] = new LiteralSubencoder();
}
void reset() {
for (int i = 0; i < subencoders.length; ++i)
subencoders[i].reset();
}
void encodeInit() {
// When encoding the first byte of the stream, there is
// no previous byte in the dictionary so the encode function
// wouldn't work.
assert readAhead >= 0;
subencoders[0].encode();
}
void encode() {
assert readAhead >= 0;
int i = getSubcoderIndex(lz.getByte(1 + readAhead),
lz.getPos() - readAhead);
subencoders[i].encode();
}
int getPrice(int curByte, int matchByte,
int prevByte, int pos, State state) {
int price = RangeEncoder.getBitPrice(
isMatch[state.get()][pos & posMask], 0);
int i = getSubcoderIndex(prevByte, pos);
price += state.isLiteral()
? subencoders[i].getNormalPrice(curByte)
: subencoders[i].getMatchedPrice(curByte, matchByte);
return price;
}
private class LiteralSubencoder extends LiteralSubcoder {
void encode() {
int symbol = lz.getByte(readAhead) | 0x100;
if (state.isLiteral()) {
int subencoderIndex;
int bit;
do {
subencoderIndex = symbol >>> 8;
bit = (symbol >>> 7) & 1;
rc.encodeBit(probs, subencoderIndex, bit);
symbol <<= 1;
} while (symbol < 0x10000);
} else {
int matchByte = lz.getByte(reps[0] + 1 + readAhead);
int offset = 0x100;
int subencoderIndex;
int matchBit;
int bit;
do {
matchByte <<= 1;
matchBit = matchByte & offset;
subencoderIndex = offset + matchBit + (symbol >>> 8);
bit = (symbol >>> 7) & 1;
rc.encodeBit(probs, subencoderIndex, bit);
symbol <<= 1;
offset &= ~(matchByte ^ symbol);
} while (symbol < 0x10000);
}
state.updateLiteral();
}
int getNormalPrice(int symbol) {
int price = 0;
int subencoderIndex;
int bit;
symbol |= 0x100;
do {
subencoderIndex = symbol >>> 8;
bit = (symbol >>> 7) & 1;
price += RangeEncoder.getBitPrice(probs[subencoderIndex],
bit);
symbol <<= 1;
} while (symbol < (0x100 << 8));
return price;
}
int getMatchedPrice(int symbol, int matchByte) {
int price = 0;
int offset = 0x100;
int subencoderIndex;
int matchBit;
int bit;
symbol |= 0x100;
do {
matchByte <<= 1;
matchBit = matchByte & offset;
subencoderIndex = offset + matchBit + (symbol >>> 8);
bit = (symbol >>> 7) & 1;
price += RangeEncoder.getBitPrice(probs[subencoderIndex],
bit);
symbol <<= 1;
offset &= ~(matchByte ^ symbol);
} while (symbol < (0x100 << 8));
return price;
}
}
}
class LengthEncoder extends LengthCoder {
/**
* The prices are updated after at least
* PRICE_UPDATE_INTERVAL many lengths
* have been encoded with the same posState.
*/
private static final int PRICE_UPDATE_INTERVAL = 32; // FIXME?
private final int[] counters;
private final int[][] prices;
LengthEncoder(int pb, int niceLen) {
int posStates = 1 << pb;
counters = new int[posStates];
// Always allocate at least LOW_SYMBOLS + MID_SYMBOLS because
// it makes updatePrices slightly simpler. The prices aren't
// usually needed anyway if niceLen < 18.
int lenSymbols = Math.max(niceLen - MATCH_LEN_MIN + 1,
LOW_SYMBOLS + MID_SYMBOLS);
prices = new int[posStates][lenSymbols];
}
void reset() {
super.reset();
// Reset counters to zero to force price update before
// the prices are needed.
for (int i = 0; i < counters.length; ++i)
counters[i] = 0;
}
void encode(int len, int posState) {
len -= MATCH_LEN_MIN;
if (len < LOW_SYMBOLS) {
rc.encodeBit(choice, 0, 0);
rc.encodeBitTree(low[posState], len);
} else {
rc.encodeBit(choice, 0, 1);
len -= LOW_SYMBOLS;
if (len < MID_SYMBOLS) {
rc.encodeBit(choice, 1, 0);
rc.encodeBitTree(mid[posState], len);
} else {
rc.encodeBit(choice, 1, 1);
rc.encodeBitTree(high, len - MID_SYMBOLS);
}
}
--counters[posState];
}
int getPrice(int len, int posState) {
return prices[posState][len - MATCH_LEN_MIN];
}
void updatePrices() {
for (int posState = 0; posState < counters.length; ++posState) {
if (counters[posState] <= 0) {
counters[posState] = PRICE_UPDATE_INTERVAL;
updatePrices(posState);
}
}
}
private void updatePrices(int posState) {
int choice0Price = RangeEncoder.getBitPrice(choice[0], 0);
int i = 0;
for (; i < LOW_SYMBOLS; ++i)
prices[posState][i] = choice0Price
+ RangeEncoder.getBitTreePrice(low[posState], i);
choice0Price = RangeEncoder.getBitPrice(choice[0], 1);
int choice1Price = RangeEncoder.getBitPrice(choice[1], 0);
for (; i < LOW_SYMBOLS + MID_SYMBOLS; ++i)
prices[posState][i] = choice0Price + choice1Price
+ RangeEncoder.getBitTreePrice(mid[posState],
i - LOW_SYMBOLS);
choice1Price = RangeEncoder.getBitPrice(choice[1], 1);
for (; i < prices[posState].length; ++i)
prices[posState][i] = choice0Price + choice1Price
+ RangeEncoder.getBitTreePrice(high, i - LOW_SYMBOLS
- MID_SYMBOLS);
}
}
}