nux-1.14.0
|
00001 /* 00002 * Copyright 2010 Inalogic® Inc. 00003 * 00004 * This program is free software: you can redistribute it and/or modify it 00005 * under the terms of the GNU Lesser General Public License, as 00006 * published by the Free Software Foundation; either version 2.1 or 3.0 00007 * of the License. 00008 * 00009 * This program is distributed in the hope that it will be useful, but 00010 * WITHOUT ANY WARRANTY; without even the implied warranties of 00011 * MERCHANTABILITY, SATISFACTORY QUALITY or FITNESS FOR A PARTICULAR 00012 * PURPOSE. See the applicable version of the GNU Lesser General Public 00013 * License for more details. 00014 * 00015 * You should have received a copy of both the GNU Lesser General Public 00016 * License along with this program. If not, see <http://www.gnu.org/licenses/> 00017 * 00018 * Authored by: Jay Taoko <jaytaoko@inalogic.com> 00019 * 00020 */ 00021 00022 00023 #include "NuxCore.h" 00024 #include "CRC32.h" 00025 00026 00027 namespace nux 00028 { 00029 // The constants here are for the CRC-32 generator 00030 // polynomial, as defined in the Microsoft 00031 // Systems Journal, March 1995, pp. 107-108 00032 00033 const unsigned int CRCTable [256] = 00034 { 00035 0x00000000, 0x77073096, 0xEE0E612C, 0x990951BA, 00036 0x076DC419, 0x706AF48F, 0xE963A535, 0x9E6495A3, 00037 0x0EDB8832, 0x79DCB8A4, 0xE0D5E91E, 0x97D2D988, 00038 0x09B64C2B, 0x7EB17CBD, 0xE7B82D07, 0x90BF1D91, 00039 0x1DB71064, 0x6AB020F2, 0xF3B97148, 0x84BE41DE, 00040 0x1ADAD47D, 0x6DDDE4EB, 0xF4D4B551, 0x83D385C7, 00041 0x136C9856, 0x646BA8C0, 0xFD62F97A, 0x8A65C9EC, 00042 0x14015C4F, 0x63066CD9, 0xFA0F3D63, 0x8D080DF5, 00043 0x3B6E20C8, 0x4C69105E, 0xD56041E4, 0xA2677172, 00044 0x3C03E4D1, 0x4B04D447, 0xD20D85FD, 0xA50AB56B, 00045 0x35B5A8FA, 0x42B2986C, 0xDBBBC9D6, 0xACBCF940, 00046 0x32D86CE3, 0x45DF5C75, 0xDCD60DCF, 0xABD13D59, 00047 0x26D930AC, 0x51DE003A, 0xC8D75180, 0xBFD06116, 00048 0x21B4F4B5, 0x56B3C423, 0xCFBA9599, 0xB8BDA50F, 00049 0x2802B89E, 0x5F058808, 0xC60CD9B2, 0xB10BE924, 00050 0x2F6F7C87, 0x58684C11, 0xC1611DAB, 0xB6662D3D, 00051 00052 0x76DC4190, 0x01DB7106, 0x98D220BC, 0xEFD5102A, 00053 0x71B18589, 0x06B6B51F, 0x9FBFE4A5, 0xE8B8D433, 00054 0x7807C9A2, 0x0F00F934, 0x9609A88E, 0xE10E9818, 00055 0x7F6A0DBB, 0x086D3D2D, 0x91646C97, 0xE6635C01, 00056 0x6B6B51F4, 0x1C6C6162, 0x856530D8, 0xF262004E, 00057 0x6C0695ED, 0x1B01A57B, 0x8208F4C1, 0xF50FC457, 00058 0x65B0D9C6, 0x12B7E950, 0x8BBEB8EA, 0xFCB9887C, 00059 0x62DD1DDF, 0x15DA2D49, 0x8CD37CF3, 0xFBD44C65, 00060 0x4DB26158, 0x3AB551CE, 0xA3BC0074, 0xD4BB30E2, 00061 0x4ADFA541, 0x3DD895D7, 0xA4D1C46D, 0xD3D6F4FB, 00062 0x4369E96A, 0x346ED9FC, 0xAD678846, 0xDA60B8D0, 00063 0x44042D73, 0x33031DE5, 0xAA0A4C5F, 0xDD0D7CC9, 00064 0x5005713C, 0x270241AA, 0xBE0B1010, 0xC90C2086, 00065 0x5768B525, 0x206F85B3, 0xB966D409, 0xCE61E49F, 00066 0x5EDEF90E, 0x29D9C998, 0xB0D09822, 0xC7D7A8B4, 00067 0x59B33D17, 0x2EB40D81, 0xB7BD5C3B, 0xC0BA6CAD, 00068 00069 0xEDB88320, 0x9ABFB3B6, 0x03B6E20C, 0x74B1D29A, 00070 0xEAD54739, 0x9DD277AF, 0x04DB2615, 0x73DC1683, 00071 0xE3630B12, 0x94643B84, 0x0D6D6A3E, 0x7A6A5AA8, 00072 0xE40ECF0B, 0x9309FF9D, 0x0A00AE27, 0x7D079EB1, 00073 0xF00F9344, 0x8708A3D2, 0x1E01F268, 0x6906C2FE, 00074 0xF762575D, 0x806567CB, 0x196C3671, 0x6E6B06E7, 00075 0xFED41B76, 0x89D32BE0, 0x10DA7A5A, 0x67DD4ACC, 00076 0xF9B9DF6F, 0x8EBEEFF9, 0x17B7BE43, 0x60B08ED5, 00077 0xD6D6A3E8, 0xA1D1937E, 0x38D8C2C4, 0x4FDFF252, 00078 0xD1BB67F1, 0xA6BC5767, 0x3FB506DD, 0x48B2364B, 00079 0xD80D2BDA, 0xAF0A1B4C, 0x36034AF6, 0x41047A60, 00080 0xDF60EFC3, 0xA867DF55, 0x316E8EEF, 0x4669BE79, 00081 0xCB61B38C, 0xBC66831A, 0x256FD2A0, 0x5268E236, 00082 0xCC0C7795, 0xBB0B4703, 0x220216B9, 0x5505262F, 00083 0xC5BA3BBE, 0xB2BD0B28, 0x2BB45A92, 0x5CB36A04, 00084 0xC2D7FFA7, 0xB5D0CF31, 0x2CD99E8B, 0x5BDEAE1D, 00085 00086 0x9B64C2B0, 0xEC63F226, 0x756AA39C, 0x026D930A, 00087 0x9C0906A9, 0xEB0E363F, 0x72076785, 0x05005713, 00088 0x95BF4A82, 0xE2B87A14, 0x7BB12BAE, 0x0CB61B38, 00089 0x92D28E9B, 0xE5D5BE0D, 0x7CDCEFB7, 0x0BDBDF21, 00090 0x86D3D2D4, 0xF1D4E242, 0x68DDB3F8, 0x1FDA836E, 00091 0x81BE16CD, 0xF6B9265B, 0x6FB077E1, 0x18B74777, 00092 0x88085AE6, 0xFF0F6A70, 0x66063BCA, 0x11010B5C, 00093 0x8F659EFF, 0xF862AE69, 0x616BFFD3, 0x166CCF45, 00094 0xA00AE278, 0xD70DD2EE, 0x4E048354, 0x3903B3C2, 00095 0xA7672661, 0xD06016F7, 0x4969474D, 0x3E6E77DB, 00096 0xAED16A4A, 0xD9D65ADC, 0x40DF0B66, 0x37D83BF0, 00097 0xA9BCAE53, 0xDEBB9EC5, 0x47B2CF7F, 0x30B5FFE9, 00098 0xBDBDF21C, 0xCABAC28A, 0x53B39330, 0x24B4A3A6, 00099 0xBAD03605, 0xCDD70693, 0x54DE5729, 0x23D967BF, 00100 0xB3667A2E, 0xC4614AB8, 0x5D681B02, 0x2A6F2B94, 00101 0xB40BBE37, 0xC30C8EA1, 0x5A05DF1B, 0x2D02EF8D 00102 }; 00103 00104 CRC32::CRC32() 00105 { 00106 Initialize(); 00107 } 00108 00109 void CRC32::Initialize (void) 00110 { 00111 Memset (&CRCTable, 0, sizeof (CRCTable) ); 00112 00113 // 256 values representing ASCII character codes. 00114 for (int iCodes = 0; iCodes <= 0xFF; iCodes++) 00115 { 00116 CRCTable[iCodes] = Reflect (iCodes, 8) << 24; 00117 00118 for (int iPos = 0; iPos < 8; iPos++) 00119 { 00120 CRCTable[iCodes] = (CRCTable[iCodes] << 1) ^ (CRCTable[iCodes] & (1 << 31) ? CRC32_POLYNOMIAL : 0); 00121 } 00122 00123 CRCTable[iCodes] = Reflect (CRCTable[iCodes], 32); 00124 } 00125 } 00126 00128 // Reflection is a requirement for the official CRC-32 standard. 00129 // You can create CRCs without it, but they won't conform to the standard. 00130 t_u32 CRC32::Reflect (t_u32 ulReflect, char cChar) 00131 { 00132 t_u32 ulValue = 0; 00133 00134 // Swap bit 0 for bit 7 bit 1 For bit 6, etc.... 00135 for (int iPos = 1; iPos < (cChar + 1); iPos++) 00136 { 00137 if (ulReflect & 1) 00138 ulValue |= 1 << (cChar - iPos); 00139 00140 ulReflect >>= 1; 00141 } 00142 00143 return ulValue; 00144 } 00145 00146 t_u32 CRC32::FileCRC (const char *sFileName) 00147 { 00148 t_u32 ulCRC = 0xffffffff; 00149 00150 FILE *fSource = NULL; 00151 char sBuf[CRC32BUFSZ]; 00152 t_u32 iBytesRead = 0; 00153 00154 #ifdef WIN32_SECURE 00155 00156 if (FOPEN_S (&fSource, sFileName, "rb") != 0) 00157 #else 00158 if (FOPEN_S (fSource, sFileName, "rb") != 0) 00159 #endif 00160 { 00161 return 0xffffffff; 00162 } 00163 00164 do 00165 { 00166 iBytesRead = (t_u32) fread (sBuf, sizeof (char), CRC32BUFSZ, fSource); 00167 PartialCRC (&ulCRC, sBuf, iBytesRead); 00168 } 00169 while (iBytesRead == CRC32BUFSZ); 00170 00171 fclose (fSource); 00172 00173 return (ulCRC ^ 0xffffffff); 00174 } 00175 00176 // This function uses the CRCTable lookup table to generate a CRC for sData 00177 t_u32 CRC32::FullCRC (const char *sData, t_u32 ulLength) 00178 { 00179 t_u32 ulCRC = 0xffffffff; 00180 PartialCRC (&ulCRC, sData, ulLength); 00181 return ulCRC ^ 0xffffffff; 00182 } 00183 00184 // Perform the algorithm on each character 00185 // in the string, using the lookup table values. 00186 void CRC32::PartialCRC (t_u32 *ulInCRC, const char *sData, t_u32 ulLength) 00187 { 00188 while (ulLength--) 00189 { 00190 *ulInCRC = (*ulInCRC >> 8) ^ CRCTable[ (*ulInCRC & 0xFF) ^ *sData++]; 00191 } 00192 } 00193 00194 00195 00196 /* 00197 * A brief CRC tutorial. 00198 * 00199 * A CRC is a long-division remainder. You add the CRC to the message, 00200 * and the whole thing (message+CRC) is a multiple of the given 00201 * CRC polynomial. To check the CRC, you can either check that the 00202 * CRC matches the recomputed value, *or* you can check that the 00203 * remainder computed on the message+CRC is 0. This latter approach 00204 * is used by a lot of hardware implementations, and is why so many 00205 * protocols put the end-of-frame flag after the CRC. 00206 * 00207 * It's actually the same long division you learned in school, except that 00208 * - We're working in binary, so the digits are only 0 and 1, and 00209 * - When dividing polynomials, there are no carries. Rather than add and 00210 * subtract, we just xor. Thus, we tend to get a bit sloppy about 00211 * the difference between adding and subtracting. 00212 * 00213 * A 32-bit CRC polynomial is actually 33 bits long. But since it's 00214 * 33 bits long, bit 32 is always going to be set, so usually the CRC 00215 * is written in hex with the most significant bit omitted. (If you're 00216 * familiar with the IEEE 754 floating-point format, it's the same idea.) 00217 * 00218 * Note that a CRC is computed over a string of *bits*, so you have 00219 * to decide on the endianness of the bits within each byte. To get 00220 * the best error-detecting properties, this should correspond to the 00221 * order they're actually sent. For example, standard RS-232 serial is 00222 * little-endian; the most significant bit (sometimes used for parity) 00223 * is sent last. And when appending a CRC word to a message, you should 00224 * do it in the right order, matching the endianness. 00225 * 00226 * Just like with ordinary division, the remainder is always smaller than 00227 * the divisor (the CRC polynomial) you're dividing by. Each step of the 00228 * division, you take one more digit (bit) of the dividend and append it 00229 * to the current remainder. Then you figure out the appropriate multiple 00230 * of the divisor to subtract to being the remainder back into range. 00231 * In binary, it's easy - it has to be either 0 or 1, and to make the 00232 * XOR cancel, it's just a copy of bit 32 of the remainder. 00233 * 00234 * When computing a CRC, we don't care about the quotient, so we can 00235 * throw the quotient bit away, but subtract the appropriate multiple of 00236 * the polynomial from the remainder and we're back to where we started, 00237 * ready to process the next bit. 00238 * 00239 * A big-endian CRC written this way would be coded like: 00240 * for (i = 0; i < input_bits; i++) { 00241 * multiple = remainder & 0x80000000 ? CRCPOLY : 0; 00242 * remainder = (remainder << 1 | next_input_bit()) ^ multiple; 00243 * } 00244 * Notice how, to get at bit 32 of the shifted remainder, we look 00245 * at bit 31 of the remainder *before* shifting it. 00246 * 00247 * But also notice how the next_input_bit() bits we're shifting into 00248 * the remainder don't actually affect any decision-making until 00249 * 32 bits later. Thus, the first 32 cycles of this are pretty boring. 00250 * Also, to add the CRC to a message, we need a 32-bit-long hole for it at 00251 * the end, so we have to add 32 extra cycles shifting in zeros at the 00252 * end of every message, 00253 * 00254 * So the standard trick is to rearrage merging in the next_input_bit() 00255 * until the moment it's needed. Then the first 32 cycles can be precomputed, 00256 * and merging in the final 32 zero bits to make room for the CRC can be 00257 * skipped entirely. 00258 * This changes the code to: 00259 * for (i = 0; i < input_bits; i++) { 00260 * remainder ^= next_input_bit() << 31; 00261 * multiple = (remainder & 0x80000000) ? CRCPOLY : 0; 00262 * remainder = (remainder << 1) ^ multiple; 00263 * } 00264 * With this optimization, the little-endian code is simpler: 00265 * for (i = 0; i < input_bits; i++) { 00266 * remainder ^= next_input_bit(); 00267 * multiple = (remainder & 1) ? CRCPOLY : 0; 00268 * remainder = (remainder >> 1) ^ multiple; 00269 * } 00270 * 00271 * Note that the other details of endianness have been hidden in CRCPOLY 00272 * (which must be bit-reversed) and next_input_bit(). 00273 * 00274 * However, as long as next_input_bit is returning the bits in a sensible 00275 * order, we can actually do the merging 8 or more bits at a time rather 00276 * than one bit at a time: 00277 * for (i = 0; i < input_bytes; i++) { 00278 * remainder ^= next_input_byte() << 24; 00279 * for (j = 0; j < 8; j++) { 00280 * multiple = (remainder & 0x80000000) ? CRCPOLY : 0; 00281 * remainder = (remainder << 1) ^ multiple; 00282 * } 00283 * } 00284 * Or in little-endian: 00285 * for (i = 0; i < input_bytes; i++) { 00286 * remainder ^= next_input_byte(); 00287 * for (j = 0; j < 8; j++) { 00288 * multiple = (remainder & 1) ? CRCPOLY : 0; 00289 * remainder = (remainder << 1) ^ multiple; 00290 * } 00291 * } 00292 * If the input is a multiple of 32 bits, you can even XOR in a 32-bit 00293 * word at a time and increase the inner loop count to 32. 00294 * 00295 * You can also mix and match the two loop styles, for example doing the 00296 * bulk of a message byte-at-a-time and adding bit-at-a-time processing 00297 * for any fractional bytes at the end. 00298 * 00299 * The only remaining optimization is to the byte-at-a-time table method. 00300 * Here, rather than just shifting one bit of the remainder to decide 00301 * in the correct multiple to subtract, we can shift a byte at a time. 00302 * This produces a 40-bit (rather than a 33-bit) intermediate remainder, 00303 * but again the multiple of the polynomial to subtract depends only on 00304 * the high bits, the high 8 bits in this case. 00305 * 00306 * The multile we need in that case is the low 32 bits of a 40-bit 00307 * value whose high 8 bits are given, and which is a multiple of the 00308 * generator polynomial. This is simply the CRC-32 of the given 00309 * one-byte message. 00310 * 00311 * Two more details: normally, appending zero bits to a message which 00312 * is already a multiple of a polynomial produces a larger multiple of that 00313 * polynomial. To enable a CRC to detect this condition, it's common to 00314 * invert the CRC before appending it. This makes the remainder of the 00315 * message+crc come out not as zero, but some fixed non-zero value. 00316 * 00317 * The same problem applies to zero bits prepended to the message, and 00318 * a similar solution is used. Instead of starting with a remainder of 00319 * 0, an initial remainder of all ones is used. As long as you start 00320 * the same way on decoding, it doesn't make a difference. 00321 */ 00322 00323 }