...
Run Format

Source file src/compress/flate/deflatefast.go

Documentation: compress/flate

  // Copyright 2016 The Go Authors. All rights reserved.
  // Use of this source code is governed by a BSD-style
  // license that can be found in the LICENSE file.
  
  package flate
  
  // This encoding algorithm, which prioritizes speed over output size, is
  // based on Snappy's LZ77-style encoder: github.com/golang/snappy
  
  const (
  	tableBits  = 14             // Bits used in the table.
  	tableSize  = 1 << tableBits // Size of the table.
  	tableMask  = tableSize - 1  // Mask for table indices. Redundant, but can eliminate bounds checks.
  	tableShift = 32 - tableBits // Right-shift to get the tableBits most significant bits of a uint32.
  )
  
  func load32(b []byte, i int32) uint32 {
  	b = b[i : i+4 : len(b)] // Help the compiler eliminate bounds checks on the next line.
  	return uint32(b[0]) | uint32(b[1])<<8 | uint32(b[2])<<16 | uint32(b[3])<<24
  }
  
  func load64(b []byte, i int32) uint64 {
  	b = b[i : i+8 : len(b)] // Help the compiler eliminate bounds checks on the next line.
  	return uint64(b[0]) | uint64(b[1])<<8 | uint64(b[2])<<16 | uint64(b[3])<<24 |
  		uint64(b[4])<<32 | uint64(b[5])<<40 | uint64(b[6])<<48 | uint64(b[7])<<56
  }
  
  func hash(u uint32) uint32 {
  	return (u * 0x1e35a7bd) >> tableShift
  }
  
  // These constants are defined by the Snappy implementation so that its
  // assembly implementation can fast-path some 16-bytes-at-a-time copies. They
  // aren't necessary in the pure Go implementation, as we don't use those same
  // optimizations, but using the same thresholds doesn't really hurt.
  const (
  	inputMargin            = 16 - 1
  	minNonLiteralBlockSize = 1 + 1 + inputMargin
  )
  
  type tableEntry struct {
  	val    uint32 // Value at destination
  	offset int32
  }
  
  // deflateFast maintains the table for matches,
  // and the previous byte block for cross block matching.
  type deflateFast struct {
  	table [tableSize]tableEntry
  	prev  []byte // Previous block, zero length if unknown.
  	cur   int32  // Current match offset.
  }
  
  func newDeflateFast() *deflateFast {
  	return &deflateFast{cur: maxStoreBlockSize, prev: make([]byte, 0, maxStoreBlockSize)}
  }
  
  // encode encodes a block given in src and appends tokens
  // to dst and returns the result.
  func (e *deflateFast) encode(dst []token, src []byte) []token {
  	// Ensure that e.cur doesn't wrap.
  	if e.cur > 1<<30 {
  		e.resetAll()
  	}
  
  	// This check isn't in the Snappy implementation, but there, the caller
  	// instead of the callee handles this case.
  	if len(src) < minNonLiteralBlockSize {
  		e.cur += maxStoreBlockSize
  		e.prev = e.prev[:0]
  		return emitLiteral(dst, src)
  	}
  
  	// sLimit is when to stop looking for offset/length copies. The inputMargin
  	// lets us use a fast path for emitLiteral in the main loop, while we are
  	// looking for copies.
  	sLimit := int32(len(src) - inputMargin)
  
  	// nextEmit is where in src the next emitLiteral should start from.
  	nextEmit := int32(0)
  	s := int32(0)
  	cv := load32(src, s)
  	nextHash := hash(cv)
  
  	for {
  		// Copied from the C++ snappy implementation:
  		//
  		// 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 (or skipped), 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.
  		skip := int32(32)
  
  		nextS := s
  		var candidate tableEntry
  		for {
  			s = nextS
  			bytesBetweenHashLookups := skip >> 5
  			nextS = s + bytesBetweenHashLookups
  			skip += bytesBetweenHashLookups
  			if nextS > sLimit {
  				goto emitRemainder
  			}
  			candidate = e.table[nextHash&tableMask]
  			now := load32(src, nextS)
  			e.table[nextHash&tableMask] = tableEntry{offset: s + e.cur, val: cv}
  			nextHash = hash(now)
  
  			offset := s - (candidate.offset - e.cur)
  			if offset > maxMatchOffset || cv != candidate.val {
  				// Out of range or not matched.
  				cv = now
  				continue
  			}
  			break
  		}
  
  		// A 4-byte match has been found. We'll later see if more than 4 bytes
  		// match. But, prior to the match, src[nextEmit:s] are unmatched. Emit
  		// them as literal bytes.
  		dst = emitLiteral(dst, src[nextEmit:s])
  
  		// 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.
  		for {
  			// Invariant: we have a 4-byte match at s, and no need to emit any
  			// literal bytes prior to s.
  
  			// Extend the 4-byte match as long as possible.
  			//
  			s += 4
  			t := candidate.offset - e.cur + 4
  			l := e.matchLen(s, t, src)
  
  			// matchToken is flate's equivalent of Snappy's emitCopy. (length,offset)
  			dst = append(dst, matchToken(uint32(l+4-baseMatchLength), uint32(s-t-baseMatchOffset)))
  			s += l
  			nextEmit = s
  			if s >= sLimit {
  				goto emitRemainder
  			}
  
  			// We could immediately start working at s now, but to improve
  			// compression we first update the hash table at s-1 and at s. If
  			// another emitCopy is not our next move, also calculate nextHash
  			// at s+1. At least on GOARCH=amd64, these three hash calculations
  			// are faster as one load64 call (with some shifts) instead of
  			// three load32 calls.
  			x := load64(src, s-1)
  			prevHash := hash(uint32(x))
  			e.table[prevHash&tableMask] = tableEntry{offset: e.cur + s - 1, val: uint32(x)}
  			x >>= 8
  			currHash := hash(uint32(x))
  			candidate = e.table[currHash&tableMask]
  			e.table[currHash&tableMask] = tableEntry{offset: e.cur + s, val: uint32(x)}
  
  			offset := s - (candidate.offset - e.cur)
  			if offset > maxMatchOffset || uint32(x) != candidate.val {
  				cv = uint32(x >> 8)
  				nextHash = hash(cv)
  				s++
  				break
  			}
  		}
  	}
  
  emitRemainder:
  	if int(nextEmit) < len(src) {
  		dst = emitLiteral(dst, src[nextEmit:])
  	}
  	e.cur += int32(len(src))
  	e.prev = e.prev[:len(src)]
  	copy(e.prev, src)
  	return dst
  }
  
  func emitLiteral(dst []token, lit []byte) []token {
  	for _, v := range lit {
  		dst = append(dst, literalToken(uint32(v)))
  	}
  	return dst
  }
  
  // matchLen returns the match length between src[s:] and src[t:].
  // t can be negative to indicate the match is starting in e.prev.
  // We assume that src[s-4:s] and src[t-4:t] already match.
  func (e *deflateFast) matchLen(s, t int32, src []byte) int32 {
  	s1 := int(s) + maxMatchLength - 4
  	if s1 > len(src) {
  		s1 = len(src)
  	}
  
  	// If we are inside the current block
  	if t >= 0 {
  		b := src[t:]
  		a := src[s:s1]
  		b = b[:len(a)]
  		// Extend the match to be as long as possible.
  		for i := range a {
  			if a[i] != b[i] {
  				return int32(i)
  			}
  		}
  		return int32(len(a))
  	}
  
  	// We found a match in the previous block.
  	tp := int32(len(e.prev)) + t
  	if tp < 0 {
  		return 0
  	}
  
  	// Extend the match to be as long as possible.
  	a := src[s:s1]
  	b := e.prev[tp:]
  	if len(b) > len(a) {
  		b = b[:len(a)]
  	}
  	a = a[:len(b)]
  	for i := range b {
  		if a[i] != b[i] {
  			return int32(i)
  		}
  	}
  
  	// If we reached our limit, we matched everything we are
  	// allowed to in the previous block and we return.
  	n := int32(len(b))
  	if int(s+n) == s1 {
  		return n
  	}
  
  	// Continue looking for more matches in the current block.
  	a = src[s+n : s1]
  	b = src[:len(a)]
  	for i := range a {
  		if a[i] != b[i] {
  			return int32(i) + n
  		}
  	}
  	return int32(len(a)) + n
  }
  
  // Reset resets the encoding history.
  // This ensures that no matches are made to the previous block.
  func (e *deflateFast) reset() {
  	e.prev = e.prev[:0]
  	// Bump the offset, so all matches will fail distance check.
  	e.cur += maxMatchOffset
  
  	// Protect against e.cur wraparound.
  	if e.cur > 1<<30 {
  		e.resetAll()
  	}
  }
  
  // resetAll resets the deflateFast struct and is only called in rare
  // situations to prevent integer overflow. It manually resets each field
  // to avoid causing large stack growth.
  //
  // See https://golang.org/issue/18636.
  func (e *deflateFast) resetAll() {
  	// This is equivalent to:
  	//	*e = deflateFast{cur: maxStoreBlockSize, prev: e.prev[:0]}
  	e.cur = maxStoreBlockSize
  	e.prev = e.prev[:0]
  	for i := range e.table {
  		e.table[i] = tableEntry{}
  	}
  }
  

View as plain text