1// Protocol Buffers - Google's data interchange format 2// Copyright 2014 Google Inc. All rights reserved. 3// https://developers.google.com/protocol-buffers/ 4// 5// Redistribution and use in source and binary forms, with or without 6// modification, are permitted provided that the following conditions are 7// met: 8// 9// * Redistributions of source code must retain the above copyright 10// notice, this list of conditions and the following disclaimer. 11// * Redistributions in binary form must reproduce the above 12// copyright notice, this list of conditions and the following disclaimer 13// in the documentation and/or other materials provided with the 14// distribution. 15// * Neither the name of Google Inc. nor the names of its 16// contributors may be used to endorse or promote products derived from 17// this software without specific prior written permission. 18// 19// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS 20// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT 21// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR 22// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT 23// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, 24// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT 25// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, 26// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY 27// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT 28// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE 29// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 30 31// Fast memory copying and comparison routines. 32// strings::fastmemcmp_inlined() replaces memcmp() 33// strings::memcpy_inlined() replaces memcpy() 34// strings::memeq(a, b, n) replaces memcmp(a, b, n) == 0 35// 36// strings::*_inlined() routines are inline versions of the 37// routines exported by this module. Sometimes using the inlined 38// versions is faster. Measure before using the inlined versions. 39// 40// Performance measurement: 41// strings::fastmemcmp_inlined 42// Analysis: memcmp, fastmemcmp_inlined, fastmemcmp 43// 2012-01-30 44 45#ifndef GOOGLE_PROTOBUF_STUBS_FASTMEM_H_ 46#define GOOGLE_PROTOBUF_STUBS_FASTMEM_H_ 47 48#include <stddef.h> 49#include <stdio.h> 50#include <string.h> 51 52#include <google/protobuf/stubs/common.h> 53 54namespace google { 55namespace protobuf { 56namespace internal { 57 58// Return true if the n bytes at a equal the n bytes at b. 59// The regions are allowed to overlap. 60// 61// The performance is similar to the performance memcmp(), but faster for 62// moderately-sized inputs, or inputs that share a common prefix and differ 63// somewhere in their last 8 bytes. Further optimizations can be added later 64// if it makes sense to do so.:w 65inline bool memeq(const char* a, const char* b, size_t n) { 66 size_t n_rounded_down = n & ~static_cast<size_t>(7); 67 if (GOOGLE_PREDICT_FALSE(n_rounded_down == 0)) { // n <= 7 68 return memcmp(a, b, n) == 0; 69 } 70 // n >= 8 71 uint64 u = GOOGLE_UNALIGNED_LOAD64(a) ^ GOOGLE_UNALIGNED_LOAD64(b); 72 uint64 v = GOOGLE_UNALIGNED_LOAD64(a + n - 8) ^ GOOGLE_UNALIGNED_LOAD64(b + n - 8); 73 if ((u | v) != 0) { // The first or last 8 bytes differ. 74 return false; 75 } 76 a += 8; 77 b += 8; 78 n = n_rounded_down - 8; 79 if (n > 128) { 80 // As of 2012, memcmp on x86-64 uses a big unrolled loop with SSE2 81 // instructions, and while we could try to do something faster, it 82 // doesn't seem worth pursuing. 83 return memcmp(a, b, n) == 0; 84 } 85 for (; n >= 16; n -= 16) { 86 uint64 x = GOOGLE_UNALIGNED_LOAD64(a) ^ GOOGLE_UNALIGNED_LOAD64(b); 87 uint64 y = GOOGLE_UNALIGNED_LOAD64(a + 8) ^ GOOGLE_UNALIGNED_LOAD64(b + 8); 88 if ((x | y) != 0) { 89 return false; 90 } 91 a += 16; 92 b += 16; 93 } 94 // n must be 0 or 8 now because it was a multiple of 8 at the top of the loop. 95 return n == 0 || GOOGLE_UNALIGNED_LOAD64(a) == GOOGLE_UNALIGNED_LOAD64(b); 96} 97 98inline int fastmemcmp_inlined(const char *a, const char *b, size_t n) { 99 if (n >= 64) { 100 return memcmp(a, b, n); 101 } 102 const char* a_limit = a + n; 103 while (a + sizeof(uint64) <= a_limit && 104 GOOGLE_UNALIGNED_LOAD64(a) == GOOGLE_UNALIGNED_LOAD64(b)) { 105 a += sizeof(uint64); 106 b += sizeof(uint64); 107 } 108 if (a + sizeof(uint32) <= a_limit && 109 GOOGLE_UNALIGNED_LOAD32(a) == GOOGLE_UNALIGNED_LOAD32(b)) { 110 a += sizeof(uint32); 111 b += sizeof(uint32); 112 } 113 while (a < a_limit) { 114 int d = static_cast<uint32>(*a++) - static_cast<uint32>(*b++); 115 if (d) return d; 116 } 117 return 0; 118} 119 120// The standard memcpy operation is slow for variable small sizes. 121// This implementation inlines the optimal realization for sizes 1 to 16. 122// To avoid code bloat don't use it in case of not performance-critical spots, 123// nor when you don't expect very frequent values of size <= 16. 124inline void memcpy_inlined(char *dst, const char *src, size_t size) { 125 // Compiler inlines code with minimal amount of data movement when third 126 // parameter of memcpy is a constant. 127 switch (size) { 128 case 1: memcpy(dst, src, 1); break; 129 case 2: memcpy(dst, src, 2); break; 130 case 3: memcpy(dst, src, 3); break; 131 case 4: memcpy(dst, src, 4); break; 132 case 5: memcpy(dst, src, 5); break; 133 case 6: memcpy(dst, src, 6); break; 134 case 7: memcpy(dst, src, 7); break; 135 case 8: memcpy(dst, src, 8); break; 136 case 9: memcpy(dst, src, 9); break; 137 case 10: memcpy(dst, src, 10); break; 138 case 11: memcpy(dst, src, 11); break; 139 case 12: memcpy(dst, src, 12); break; 140 case 13: memcpy(dst, src, 13); break; 141 case 14: memcpy(dst, src, 14); break; 142 case 15: memcpy(dst, src, 15); break; 143 case 16: memcpy(dst, src, 16); break; 144 default: memcpy(dst, src, size); break; 145 } 146} 147 148} // namespace internal 149} // namespace protobuf 150} // namespace google 151 152#endif // GOOGLE_PROTOBUF_STUBS_FASTMEM_H_ 153