io.airlift.compress.snappy.SnappyRawCompressor Maven / Gradle / Ivy
/*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/
package io.airlift.compress.snappy;
import java.util.Arrays;
import static io.airlift.compress.snappy.SnappyConstants.COPY_1_BYTE_OFFSET;
import static io.airlift.compress.snappy.SnappyConstants.COPY_2_BYTE_OFFSET;
import static io.airlift.compress.snappy.SnappyConstants.SIZE_OF_INT;
import static io.airlift.compress.snappy.SnappyConstants.SIZE_OF_LONG;
import static io.airlift.compress.snappy.SnappyConstants.SIZE_OF_SHORT;
import static io.airlift.compress.snappy.UnsafeUtil.UNSAFE;
public final class SnappyRawCompressor
{
// The size of a compression block. Note that many parts of the compression
// code assumes that BLOCK_SIZE <= 65536; in particular, the hash table
// can only store 16-bit offsets, and EmitCopy() also assumes the offset
// is 65535 bytes or less. Note also that if you change this, it will
// affect the framing format (see framing_format.txt).
//
// Note that there might be older data around that is compressed with larger
// block sizes, so the decompression code should not rely on the
// non-existence of long back-references.
private static final int BLOCK_LOG = 16;
private static final int BLOCK_SIZE = 1 << BLOCK_LOG;
private static final int INPUT_MARGIN_BYTES = 15;
private static final int MAX_HASH_TABLE_BITS = 14;
public static final int MAX_HASH_TABLE_SIZE = 1 << MAX_HASH_TABLE_BITS;
private SnappyRawCompressor() {}
public static int maxCompressedLength(int sourceLength)
{
// Compressed data can be defined as:
// compressed := item* literal*
// item := literal* copy
//
// The trailing literal sequence has a space blowup of at most 62/60
// since a literal of length 60 needs one tag byte + one extra byte
// for length information.
//
// Item blowup is trickier to measure. Suppose the "copy" op copies
// 4 bytes of data. Because of a special check in the encoding code,
// we produce a 4-byte copy only if the offset is < 65536. Therefore
// the copy op takes 3 bytes to encode, and this type of item leads
// to at most the 62/60 blowup for representing literals.
//
// Suppose the "copy" op copies 5 bytes of data. If the offset is big
// enough, it will take 5 bytes to encode the copy op. Therefore the
// worst case here is a one-byte literal followed by a five-byte copy.
// I.e., 6 bytes of input turn into 7 bytes of "compressed" data.
//
// This last factor dominates the blowup, so the final estimate is:
return 32 + sourceLength + sourceLength / 6;
}
public static int compress(
final Object inputBase,
final long inputAddress,
final long inputLimit,
final Object outputBase,
final long outputAddress,
final long outputLimit,
final short[] table)
{
// The compression code assumes output is larger than the max compression size (with 32 bytes of
// extra padding), and does not check bounds for writing to output.
int maxCompressedLength = maxCompressedLength((int) (inputLimit - inputAddress));
if (outputLimit - outputAddress < maxCompressedLength) {
throw new IllegalArgumentException("Output buffer must be at least " + maxCompressedLength + " bytes");
}
// First write the uncompressed size to the output as a variable length int
long output = writeUncompressedLength(outputBase, outputAddress, (int) (inputLimit - inputAddress));
for (long blockAddress = inputAddress; blockAddress < inputLimit; blockAddress += BLOCK_SIZE) {
final long blockLimit = Math.min(inputLimit, blockAddress + BLOCK_SIZE);
long input = blockAddress;
assert blockLimit - blockAddress <= BLOCK_SIZE;
int blockHashTableSize = getHashTableSize((int) (blockLimit - blockAddress));
Arrays.fill(table, 0, blockHashTableSize, (short) 0);
// todo given that hashTableSize is required to be a power of 2, this is overly complex
final int shift = 32 - log2Floor(blockHashTableSize);
assert (blockHashTableSize & (blockHashTableSize - 1)) == 0 : "table must be power of two";
assert 0xFFFFFFFF >>> shift == blockHashTableSize - 1;
// Bytes in [nextEmitAddress, input) will be emitted as literal bytes. Or
// [nextEmitAddress, inputLimit) after the main loop.
long nextEmitAddress = input;
final long fastInputLimit = blockLimit - INPUT_MARGIN_BYTES;
while (input <= fastInputLimit) {
assert nextEmitAddress <= input;
// The body of this loop emits a literal once and then emits a copy one
// or more times. (The exception is that when we're close to exhausting
// the input we exit and emit a literal.)
//
// In the first iteration of this loop we're just starting, so
// there's nothing to copy, so we must emit a literal once. And we
// only start a new iteration when the current iteration has determined
// that a literal will precede the next copy (if any).
//
// Step 1: Scan forward in the input looking for a 4-byte-long match.
// If we get close to exhausting the input exit and emit a final literal.
//
// Heuristic match skipping: If 32 bytes are scanned with no matches
// found, start looking only at every other byte. If 32 more bytes are
// scanned, look at every third byte, etc.. When a match is found,
// immediately go back to looking at every byte. This is a small loss
// (~5% performance, ~0.1% density) for compressible data due to more
// bookkeeping, but for non-compressible data (such as JPEG) it's a huge
// win since the compressor quickly "realizes" the data is incompressible
// and doesn't bother looking for matches everywhere.
//
// The "skip" variable keeps track of how many bytes there are since the
// last match; dividing it by 32 (ie. right-shifting by five) gives the
// number of bytes to move ahead for each iteration.
int skip = 32;
long candidateIndex = 0;
for (input += 1; input + (skip >>> 5) <= fastInputLimit; input += ((skip++) >>> 5)) {
// hash the 4 bytes starting at the input pointer
int currentInt = UNSAFE.getInt(inputBase, input);
int hash = hashBytes(currentInt, shift);
// get the position of a 4 bytes sequence with the same hash
candidateIndex = blockAddress + (table[hash] & 0xFFFF);
assert candidateIndex >= 0;
assert candidateIndex < input;
// update the hash to point to the current position
table[hash] = (short) (input - blockAddress);
// if the 4 byte sequence a the candidate index matches the sequence at the
// current position, proceed to the next phase
if (currentInt == UNSAFE.getInt(inputBase, candidateIndex)) {
break;
}
}
if (input + (skip >>> 5) > fastInputLimit) {
break;
}
// Step 2: A 4-byte match has been found. We'll later see if more
// than 4 bytes match. But, prior to the match, input
// bytes [nextEmit, ip) are unmatched. Emit them as "literal bytes."
assert nextEmitAddress + 16 <= blockLimit;
int literalLength = (int) (input - nextEmitAddress);
output = emitLiteralLength(outputBase, output, literalLength);
// Fast copy can use 8 extra bytes of input and output, which is safe because:
// - The input will always have INPUT_MARGIN_BYTES = 15 extra available bytes
// - The output will always have 32 spare bytes (see MaxCompressedLength).
output = fastCopy(inputBase, nextEmitAddress, outputBase, output, literalLength);
// Step 3: Call EmitCopy, and then see if another EmitCopy could
// be our next move. Repeat until we find no match for the
// input immediately after what was consumed by the last EmitCopy call.
//
// If we exit this loop normally then we need to call EmitLiteral next,
// though we don't yet know how big the literal will be. We handle that
// by proceeding to the next iteration of the main loop. We also can exit
// this loop via goto if we get close to exhausting the input.
int inputBytes;
do {
// We have a 4-byte match at input, and no need to emit any
// "literal bytes" prior to input.
assert (blockLimit >= input + SIZE_OF_INT);
// determine match length
int matched = count(inputBase, input + SIZE_OF_INT, candidateIndex + SIZE_OF_INT, blockLimit);
matched += SIZE_OF_INT;
// Emit the copy operation for this chunk
output = emitCopy(outputBase, output, input, candidateIndex, matched);
input += matched;
// are we done?
if (input >= fastInputLimit) {
break;
}
// We could immediately start working at input now, but to improve
// compression we first update table[Hash(ip - 1, ...)].
long longValue = UNSAFE.getLong(inputBase, input - 1);
int prevInt = (int) longValue;
inputBytes = (int) (longValue >>> 8);
// add hash starting with previous byte
int prevHash = hashBytes(prevInt, shift);
table[prevHash] = (short) (input - blockAddress - 1);
// update hash of current byte
int curHash = hashBytes(inputBytes, shift);
candidateIndex = blockAddress + (table[curHash] & 0xFFFF);
table[curHash] = (short) (input - blockAddress);
} while (inputBytes == UNSAFE.getInt(inputBase, candidateIndex));
nextEmitAddress = input;
}
// Emit the remaining bytes as a literal
if (nextEmitAddress < blockLimit) {
int literalLength = (int) (blockLimit - nextEmitAddress);
output = emitLiteralLength(outputBase, output, literalLength);
UNSAFE.copyMemory(inputBase, nextEmitAddress, outputBase, output, literalLength);
output += literalLength;
}
}
return (int) (output - outputAddress);
}
private static int count(Object inputBase, final long start, long matchStart, long matchLimit)
{
long current = start;
// first, compare long at a time
while (current < matchLimit - (SIZE_OF_LONG - 1)) {
long diff = UNSAFE.getLong(inputBase, matchStart) ^ UNSAFE.getLong(inputBase, current);
if (diff != 0) {
current += Long.numberOfTrailingZeros(diff) >> 3;
return (int) (current - start);
}
current += SIZE_OF_LONG;
matchStart += SIZE_OF_LONG;
}
if (current < matchLimit - (SIZE_OF_INT - 1) && UNSAFE.getInt(inputBase, matchStart) == UNSAFE.getInt(inputBase, current)) {
current += SIZE_OF_INT;
matchStart += SIZE_OF_INT;
}
if (current < matchLimit - (SIZE_OF_SHORT - 1) && UNSAFE.getShort(inputBase, matchStart) == UNSAFE.getShort(inputBase, current)) {
current += SIZE_OF_SHORT;
matchStart += SIZE_OF_SHORT;
}
if (current < matchLimit && UNSAFE.getByte(inputBase, matchStart) == UNSAFE.getByte(inputBase, current)) {
++current;
}
return (int) (current - start);
}
private static long emitLiteralLength(Object outputBase, long output, int literalLength)
{
int n = literalLength - 1; // Zero-length literals are disallowed
if (n < 60) {
// Size fits in tag byte
UNSAFE.putByte(outputBase, output++, (byte) (n << 2));
}
else {
int bytes;
if (n < (1 << 8)) {
UNSAFE.putByte(outputBase, output++, (byte) (59 + 1 << 2));
bytes = 1;
}
else if (n < (1 << 16)) {
UNSAFE.putByte(outputBase, output++, (byte) (59 + 2 << 2));
bytes = 2;
}
else if (n < (1 << 24)) {
UNSAFE.putByte(outputBase, output++, (byte) (59 + 3 << 2));
bytes = 3;
}
else {
UNSAFE.putByte(outputBase, output++, (byte) (59 + 4 << 2));
bytes = 4;
}
// System is assumed to be little endian, so low bytes will be zero for the smaller numbers
UNSAFE.putInt(outputBase, output, n);
output += bytes;
}
return output;
}
private static long fastCopy(final Object inputBase, long input, final Object outputBase, long output, final int literalLength)
{
final long outputLimit = output + literalLength;
do {
UNSAFE.putLong(outputBase, output, UNSAFE.getLong(inputBase, input));
input += SIZE_OF_LONG;
output += SIZE_OF_LONG;
}
while (output < outputLimit);
return outputLimit;
}
private static long emitCopy(Object outputBase, long output, long input, long matchIndex, int matchLength)
{
long offset = input - matchIndex;
// Emit 64 byte copies but make sure to keep at least four bytes reserved
while (matchLength >= 68) {
UNSAFE.putByte(outputBase, output++, (byte) (COPY_2_BYTE_OFFSET + ((64 - 1) << 2)));
UNSAFE.putShort(outputBase, output, (short) offset);
output += SIZE_OF_SHORT;
matchLength -= 64;
}
// Emit an extra 60 byte copy if have too much data to fit in one copy
// length < 68
if (matchLength > 64) {
UNSAFE.putByte(outputBase, output++, (byte) (COPY_2_BYTE_OFFSET + ((60 - 1) << 2)));
UNSAFE.putShort(outputBase, output, (short) offset);
output += SIZE_OF_SHORT;
matchLength -= 60;
}
// Emit remainder
if ((matchLength < 12) && (offset < 2048)) {
int lenMinus4 = matchLength - 4;
UNSAFE.putByte(outputBase, output++, (byte) (COPY_1_BYTE_OFFSET + ((lenMinus4) << 2) + ((offset >>> 8) << 5)));
UNSAFE.putByte(outputBase, output++, (byte) (offset));
}
else {
UNSAFE.putByte(outputBase, output++, (byte) (COPY_2_BYTE_OFFSET + ((matchLength - 1) << 2)));
UNSAFE.putShort(outputBase, output, (short) offset);
output += SIZE_OF_SHORT;
}
return output;
}
private static int getHashTableSize(int inputSize)
{
// Use smaller hash table when input.size() is smaller, since we
// fill the table, incurring O(hash table size) overhead for
// compression, and if the input is short, we won't need that
// many hash table entries anyway.
assert (MAX_HASH_TABLE_SIZE >= 256);
// smallest power of 2 larger than inputSize
int target = Integer.highestOneBit(inputSize - 1) << 1;
// keep it between MIN_TABLE_SIZE and MAX_TABLE_SIZE
return Math.max(Math.min(target, MAX_HASH_TABLE_SIZE), 256);
}
// Any hash function will produce a valid compressed stream, but a good
// hash function reduces the number of collisions and thus yields better
// compression for compressible input, and more speed for incompressible
// input. Of course, it doesn't hurt if the hash function is reasonably fast
// either, as it gets called a lot.
private static int hashBytes(int value, int shift)
{
return (value * 0x1e35a7bd) >>> shift;
}
private static int log2Floor(int n)
{
return n == 0 ? -1 : 31 ^ Integer.numberOfLeadingZeros(n);
}
private static final int HIGH_BIT_MASK = 0x80;
/**
* Writes the uncompressed length as variable length integer.
*/
private static long writeUncompressedLength(Object outputBase, long outputAddress, int uncompressedLength)
{
if (uncompressedLength < (1 << 7) && uncompressedLength >= 0) {
UNSAFE.putByte(outputBase, outputAddress++, (byte) (uncompressedLength));
}
else if (uncompressedLength < (1 << 14) && uncompressedLength > 0) {
UNSAFE.putByte(outputBase, outputAddress++, (byte) (uncompressedLength | HIGH_BIT_MASK));
UNSAFE.putByte(outputBase, outputAddress++, (byte) (uncompressedLength >>> 7));
}
else if (uncompressedLength < (1 << 21) && uncompressedLength > 0) {
UNSAFE.putByte(outputBase, outputAddress++, (byte) (uncompressedLength | HIGH_BIT_MASK));
UNSAFE.putByte(outputBase, outputAddress++, (byte) ((uncompressedLength >>> 7) | HIGH_BIT_MASK));
UNSAFE.putByte(outputBase, outputAddress++, (byte) (uncompressedLength >>> 14));
}
else if (uncompressedLength < (1 << 28) && uncompressedLength > 0) {
UNSAFE.putByte(outputBase, outputAddress++, (byte) (uncompressedLength | HIGH_BIT_MASK));
UNSAFE.putByte(outputBase, outputAddress++, (byte) ((uncompressedLength >>> 7) | HIGH_BIT_MASK));
UNSAFE.putByte(outputBase, outputAddress++, (byte) ((uncompressedLength >>> 14) | HIGH_BIT_MASK));
UNSAFE.putByte(outputBase, outputAddress++, (byte) (uncompressedLength >>> 21));
}
else {
UNSAFE.putByte(outputBase, outputAddress++, (byte) (uncompressedLength | HIGH_BIT_MASK));
UNSAFE.putByte(outputBase, outputAddress++, (byte) ((uncompressedLength >>> 7) | HIGH_BIT_MASK));
UNSAFE.putByte(outputBase, outputAddress++, (byte) ((uncompressedLength >>> 14) | HIGH_BIT_MASK));
UNSAFE.putByte(outputBase, outputAddress++, (byte) ((uncompressedLength >>> 21) | HIGH_BIT_MASK));
UNSAFE.putByte(outputBase, outputAddress++, (byte) (uncompressedLength >>> 28));
}
return outputAddress;
}
}
© 2015 - 2024 Weber Informatics LLC | Privacy Policy