...

Source file src/github.com/golang/snappy/encode_other.go

Documentation: github.com/golang/snappy

     1  // Copyright 2016 The Snappy-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  // +build !amd64,!arm64 appengine !gc noasm
     6  
     7  package snappy
     8  
     9  func load32(b []byte, i int) uint32 {
    10  	b = b[i : i+4 : len(b)] // Help the compiler eliminate bounds checks on the next line.
    11  	return uint32(b[0]) | uint32(b[1])<<8 | uint32(b[2])<<16 | uint32(b[3])<<24
    12  }
    13  
    14  func load64(b []byte, i int) uint64 {
    15  	b = b[i : i+8 : len(b)] // Help the compiler eliminate bounds checks on the next line.
    16  	return uint64(b[0]) | uint64(b[1])<<8 | uint64(b[2])<<16 | uint64(b[3])<<24 |
    17  		uint64(b[4])<<32 | uint64(b[5])<<40 | uint64(b[6])<<48 | uint64(b[7])<<56
    18  }
    19  
    20  // emitLiteral writes a literal chunk and returns the number of bytes written.
    21  //
    22  // It assumes that:
    23  //	dst is long enough to hold the encoded bytes
    24  //	1 <= len(lit) && len(lit) <= 65536
    25  func emitLiteral(dst, lit []byte) int {
    26  	i, n := 0, uint(len(lit)-1)
    27  	switch {
    28  	case n < 60:
    29  		dst[0] = uint8(n)<<2 | tagLiteral
    30  		i = 1
    31  	case n < 1<<8:
    32  		dst[0] = 60<<2 | tagLiteral
    33  		dst[1] = uint8(n)
    34  		i = 2
    35  	default:
    36  		dst[0] = 61<<2 | tagLiteral
    37  		dst[1] = uint8(n)
    38  		dst[2] = uint8(n >> 8)
    39  		i = 3
    40  	}
    41  	return i + copy(dst[i:], lit)
    42  }
    43  
    44  // emitCopy writes a copy chunk and returns the number of bytes written.
    45  //
    46  // It assumes that:
    47  //	dst is long enough to hold the encoded bytes
    48  //	1 <= offset && offset <= 65535
    49  //	4 <= length && length <= 65535
    50  func emitCopy(dst []byte, offset, length int) int {
    51  	i := 0
    52  	// The maximum length for a single tagCopy1 or tagCopy2 op is 64 bytes. The
    53  	// threshold for this loop is a little higher (at 68 = 64 + 4), and the
    54  	// length emitted down below is is a little lower (at 60 = 64 - 4), because
    55  	// it's shorter to encode a length 67 copy as a length 60 tagCopy2 followed
    56  	// by a length 7 tagCopy1 (which encodes as 3+2 bytes) than to encode it as
    57  	// a length 64 tagCopy2 followed by a length 3 tagCopy2 (which encodes as
    58  	// 3+3 bytes). The magic 4 in the 64±4 is because the minimum length for a
    59  	// tagCopy1 op is 4 bytes, which is why a length 3 copy has to be an
    60  	// encodes-as-3-bytes tagCopy2 instead of an encodes-as-2-bytes tagCopy1.
    61  	for length >= 68 {
    62  		// Emit a length 64 copy, encoded as 3 bytes.
    63  		dst[i+0] = 63<<2 | tagCopy2
    64  		dst[i+1] = uint8(offset)
    65  		dst[i+2] = uint8(offset >> 8)
    66  		i += 3
    67  		length -= 64
    68  	}
    69  	if length > 64 {
    70  		// Emit a length 60 copy, encoded as 3 bytes.
    71  		dst[i+0] = 59<<2 | tagCopy2
    72  		dst[i+1] = uint8(offset)
    73  		dst[i+2] = uint8(offset >> 8)
    74  		i += 3
    75  		length -= 60
    76  	}
    77  	if length >= 12 || offset >= 2048 {
    78  		// Emit the remaining copy, encoded as 3 bytes.
    79  		dst[i+0] = uint8(length-1)<<2 | tagCopy2
    80  		dst[i+1] = uint8(offset)
    81  		dst[i+2] = uint8(offset >> 8)
    82  		return i + 3
    83  	}
    84  	// Emit the remaining copy, encoded as 2 bytes.
    85  	dst[i+0] = uint8(offset>>8)<<5 | uint8(length-4)<<2 | tagCopy1
    86  	dst[i+1] = uint8(offset)
    87  	return i + 2
    88  }
    89  
    90  // extendMatch returns the largest k such that k <= len(src) and that
    91  // src[i:i+k-j] and src[j:k] have the same contents.
    92  //
    93  // It assumes that:
    94  //	0 <= i && i < j && j <= len(src)
    95  func extendMatch(src []byte, i, j int) int {
    96  	for ; j < len(src) && src[i] == src[j]; i, j = i+1, j+1 {
    97  	}
    98  	return j
    99  }
   100  
   101  func hash(u, shift uint32) uint32 {
   102  	return (u * 0x1e35a7bd) >> shift
   103  }
   104  
   105  // encodeBlock encodes a non-empty src to a guaranteed-large-enough dst. It
   106  // assumes that the varint-encoded length of the decompressed bytes has already
   107  // been written.
   108  //
   109  // It also assumes that:
   110  //	len(dst) >= MaxEncodedLen(len(src)) &&
   111  // 	minNonLiteralBlockSize <= len(src) && len(src) <= maxBlockSize
   112  func encodeBlock(dst, src []byte) (d int) {
   113  	// Initialize the hash table. Its size ranges from 1<<8 to 1<<14 inclusive.
   114  	// The table element type is uint16, as s < sLimit and sLimit < len(src)
   115  	// and len(src) <= maxBlockSize and maxBlockSize == 65536.
   116  	const (
   117  		maxTableSize = 1 << 14
   118  		// tableMask is redundant, but helps the compiler eliminate bounds
   119  		// checks.
   120  		tableMask = maxTableSize - 1
   121  	)
   122  	shift := uint32(32 - 8)
   123  	for tableSize := 1 << 8; tableSize < maxTableSize && tableSize < len(src); tableSize *= 2 {
   124  		shift--
   125  	}
   126  	// In Go, all array elements are zero-initialized, so there is no advantage
   127  	// to a smaller tableSize per se. However, it matches the C++ algorithm,
   128  	// and in the asm versions of this code, we can get away with zeroing only
   129  	// the first tableSize elements.
   130  	var table [maxTableSize]uint16
   131  
   132  	// sLimit is when to stop looking for offset/length copies. The inputMargin
   133  	// lets us use a fast path for emitLiteral in the main loop, while we are
   134  	// looking for copies.
   135  	sLimit := len(src) - inputMargin
   136  
   137  	// nextEmit is where in src the next emitLiteral should start from.
   138  	nextEmit := 0
   139  
   140  	// The encoded form must start with a literal, as there are no previous
   141  	// bytes to copy, so we start looking for hash matches at s == 1.
   142  	s := 1
   143  	nextHash := hash(load32(src, s), shift)
   144  
   145  	for {
   146  		// Copied from the C++ snappy implementation:
   147  		//
   148  		// Heuristic match skipping: If 32 bytes are scanned with no matches
   149  		// found, start looking only at every other byte. If 32 more bytes are
   150  		// scanned (or skipped), look at every third byte, etc.. When a match
   151  		// is found, immediately go back to looking at every byte. This is a
   152  		// small loss (~5% performance, ~0.1% density) for compressible data
   153  		// due to more bookkeeping, but for non-compressible data (such as
   154  		// JPEG) it's a huge win since the compressor quickly "realizes" the
   155  		// data is incompressible and doesn't bother looking for matches
   156  		// everywhere.
   157  		//
   158  		// The "skip" variable keeps track of how many bytes there are since
   159  		// the last match; dividing it by 32 (ie. right-shifting by five) gives
   160  		// the number of bytes to move ahead for each iteration.
   161  		skip := 32
   162  
   163  		nextS := s
   164  		candidate := 0
   165  		for {
   166  			s = nextS
   167  			bytesBetweenHashLookups := skip >> 5
   168  			nextS = s + bytesBetweenHashLookups
   169  			skip += bytesBetweenHashLookups
   170  			if nextS > sLimit {
   171  				goto emitRemainder
   172  			}
   173  			candidate = int(table[nextHash&tableMask])
   174  			table[nextHash&tableMask] = uint16(s)
   175  			nextHash = hash(load32(src, nextS), shift)
   176  			if load32(src, s) == load32(src, candidate) {
   177  				break
   178  			}
   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  		d += emitLiteral(dst[d:], src[nextEmit:s])
   185  
   186  		// Call emitCopy, and then see if another emitCopy could be our next
   187  		// move. Repeat until we find no match for the input immediately after
   188  		// what was consumed by the last emitCopy call.
   189  		//
   190  		// If we exit this loop normally then we need to call emitLiteral next,
   191  		// though we don't yet know how big the literal will be. We handle that
   192  		// by proceeding to the next iteration of the main loop. We also can
   193  		// exit this loop via goto if we get close to exhausting the input.
   194  		for {
   195  			// Invariant: we have a 4-byte match at s, and no need to emit any
   196  			// literal bytes prior to s.
   197  			base := s
   198  
   199  			// Extend the 4-byte match as long as possible.
   200  			//
   201  			// This is an inlined version of:
   202  			//	s = extendMatch(src, candidate+4, s+4)
   203  			s += 4
   204  			for i := candidate + 4; s < len(src) && src[i] == src[s]; i, s = i+1, s+1 {
   205  			}
   206  
   207  			d += emitCopy(dst[d:], base-candidate, s-base)
   208  			nextEmit = s
   209  			if s >= sLimit {
   210  				goto emitRemainder
   211  			}
   212  
   213  			// We could immediately start working at s now, but to improve
   214  			// compression we first update the hash table at s-1 and at s. If
   215  			// another emitCopy is not our next move, also calculate nextHash
   216  			// at s+1. At least on GOARCH=amd64, these three hash calculations
   217  			// are faster as one load64 call (with some shifts) instead of
   218  			// three load32 calls.
   219  			x := load64(src, s-1)
   220  			prevHash := hash(uint32(x>>0), shift)
   221  			table[prevHash&tableMask] = uint16(s - 1)
   222  			currHash := hash(uint32(x>>8), shift)
   223  			candidate = int(table[currHash&tableMask])
   224  			table[currHash&tableMask] = uint16(s)
   225  			if uint32(x>>8) != load32(src, candidate) {
   226  				nextHash = hash(uint32(x>>16), shift)
   227  				s++
   228  				break
   229  			}
   230  		}
   231  	}
   232  
   233  emitRemainder:
   234  	if nextEmit < len(src) {
   235  		d += emitLiteral(dst[d:], src[nextEmit:])
   236  	}
   237  	return d
   238  }
   239  

View as plain text