| 274 | } |
| 275 | |
| 276 | uint32_t Extend(uint32_t crc, const char* data, size_t n) { |
| 277 | static bool accelerate = CanAccelerateCRC32C(); |
| 278 | if (accelerate) { |
| 279 | return port::AcceleratedCRC32C(crc, data, n); |
| 280 | } |
| 281 | |
| 282 | const uint8_t* p = reinterpret_cast<const uint8_t*>(data); |
| 283 | const uint8_t* e = p + n; |
| 284 | uint32_t l = crc ^ kCRC32Xor; |
| 285 | |
| 286 | // Process one byte at a time. |
| 287 | #define STEP1 \ |
| 288 | do { \ |
| 289 | int c = (l & 0xff) ^ *p++; \ |
| 290 | l = kByteExtensionTable[c] ^ (l >> 8); \ |
| 291 | } while (0) |
| 292 | |
| 293 | // Process one of the 4 strides of 4-byte data. |
| 294 | #define STEP4(s) \ |
| 295 | do { \ |
| 296 | crc##s = ReadUint32LE(p + s * 4) ^ kStrideExtensionTable3[crc##s & 0xff] ^ \ |
| 297 | kStrideExtensionTable2[(crc##s >> 8) & 0xff] ^ \ |
| 298 | kStrideExtensionTable1[(crc##s >> 16) & 0xff] ^ \ |
| 299 | kStrideExtensionTable0[crc##s >> 24]; \ |
| 300 | } while (0) |
| 301 | |
| 302 | // Process a 16-byte swath of 4 strides, each of which has 4 bytes of data. |
| 303 | #define STEP16 \ |
| 304 | do { \ |
| 305 | STEP4(0); \ |
| 306 | STEP4(1); \ |
| 307 | STEP4(2); \ |
| 308 | STEP4(3); \ |
| 309 | p += 16; \ |
| 310 | } while (0) |
| 311 | |
| 312 | // Process 4 bytes that were already loaded into a word. |
| 313 | #define STEP4W(w) \ |
| 314 | do { \ |
| 315 | w ^= l; \ |
| 316 | for (size_t i = 0; i < 4; ++i) { \ |
| 317 | w = (w >> 8) ^ kByteExtensionTable[w & 0xff]; \ |
| 318 | } \ |
| 319 | l = w; \ |
| 320 | } while (0) |
| 321 | |
| 322 | // Point x at first 4-byte aligned byte in the buffer. This might be past the |
| 323 | // end of the buffer. |
| 324 | const uint8_t* x = RoundUp<4>(p); |
| 325 | if (x <= e) { |
| 326 | // Process bytes p is 4-byte aligned. |
| 327 | while (p != x) { |
| 328 | STEP1; |
| 329 | } |
| 330 | } |
| 331 | |
| 332 | if ((e - p) >= 16) { |
| 333 | // Load a 16-byte swath into the stride partial results. |