github.com/Filosottile/go@v0.0.0-20170906193555-dbed9972d994/src/bytes/bytes_amd64.go (about)

     1  // Copyright 2016 The 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  package bytes
     6  
     7  import "internal/cpu"
     8  
     9  //go:noescape
    10  
    11  // indexShortStr returns the index of the first instance of c in s, or -1 if c is not present in s.
    12  // indexShortStr requires 2 <= len(c) <= shortStringLen
    13  func indexShortStr(s, c []byte) int  // ../runtime/asm_amd64.s
    14  func countByte(s []byte, c byte) int // ../runtime/asm_amd64.s
    15  
    16  var shortStringLen int
    17  
    18  func init() {
    19  	if cpu.X86.HasAVX2 {
    20  		shortStringLen = 63
    21  	} else {
    22  		shortStringLen = 31
    23  	}
    24  }
    25  
    26  // Index returns the index of the first instance of sep in s, or -1 if sep is not present in s.
    27  func Index(s, sep []byte) int {
    28  	n := len(sep)
    29  	switch {
    30  	case n == 0:
    31  		return 0
    32  	case n == 1:
    33  		return IndexByte(s, sep[0])
    34  	case n == len(s):
    35  		if Equal(sep, s) {
    36  			return 0
    37  		}
    38  		return -1
    39  	case n > len(s):
    40  		return -1
    41  	case n <= shortStringLen:
    42  		// Use brute force when s and sep both are small
    43  		if len(s) <= 64 {
    44  			return indexShortStr(s, sep)
    45  		}
    46  		c := sep[0]
    47  		i := 0
    48  		t := s[:len(s)-n+1]
    49  		fails := 0
    50  		for i < len(t) {
    51  			if t[i] != c {
    52  				// IndexByte skips 16/32 bytes per iteration,
    53  				// so it's faster than indexShortStr.
    54  				o := IndexByte(t[i:], c)
    55  				if o < 0 {
    56  					return -1
    57  				}
    58  				i += o
    59  			}
    60  			if Equal(s[i:i+n], sep) {
    61  				return i
    62  			}
    63  			fails++
    64  			i++
    65  			// Switch to indexShortStr when IndexByte produces too many false positives.
    66  			// Too many means more that 1 error per 8 characters.
    67  			// Allow some errors in the beginning.
    68  			if fails > (i+16)/8 {
    69  				r := indexShortStr(s[i:], sep)
    70  				if r >= 0 {
    71  					return r + i
    72  				}
    73  				return -1
    74  			}
    75  		}
    76  		return -1
    77  	}
    78  	// Rabin-Karp search
    79  	hashsep, pow := hashStr(sep)
    80  	var h uint32
    81  	for i := 0; i < n; i++ {
    82  		h = h*primeRK + uint32(s[i])
    83  	}
    84  	if h == hashsep && Equal(s[:n], sep) {
    85  		return 0
    86  	}
    87  	for i := n; i < len(s); {
    88  		h *= primeRK
    89  		h += uint32(s[i])
    90  		h -= pow * uint32(s[i-n])
    91  		i++
    92  		if h == hashsep && Equal(s[i-n:i], sep) {
    93  			return i - n
    94  		}
    95  	}
    96  	return -1
    97  }
    98  
    99  // Count counts the number of non-overlapping instances of sep in s.
   100  // If sep is an empty slice, Count returns 1 + the number of Unicode code points in s.
   101  func Count(s, sep []byte) int {
   102  	if len(sep) == 1 && cpu.X86.HasPOPCNT {
   103  		return countByte(s, sep[0])
   104  	}
   105  	return countGeneric(s, sep)
   106  }
   107  
   108  // primeRK is the prime base used in Rabin-Karp algorithm.
   109  const primeRK = 16777619
   110  
   111  // hashStr returns the hash and the appropriate multiplicative
   112  // factor for use in Rabin-Karp algorithm.
   113  func hashStr(sep []byte) (uint32, uint32) {
   114  	hash := uint32(0)
   115  	for i := 0; i < len(sep); i++ {
   116  		hash = hash*primeRK + uint32(sep[i])
   117  	}
   118  	var pow, sq uint32 = 1, primeRK
   119  	for i := len(sep); i > 0; i >>= 1 {
   120  		if i&1 != 0 {
   121  			pow *= sq
   122  		}
   123  		sq *= sq
   124  	}
   125  	return hash, pow
   126  }